147e4937aSGao Xiang // SPDX-License-Identifier: GPL-2.0-only 247e4937aSGao Xiang /* 347e4937aSGao Xiang * Copyright (C) 2018 HUAWEI, Inc. 4592e7cd0SAlexander A. Klimov * https://www.huawei.com/ 547e4937aSGao Xiang */ 647e4937aSGao Xiang #include "zdata.h" 747e4937aSGao Xiang #include "compress.h" 847e4937aSGao Xiang #include <linux/prefetch.h> 947e4937aSGao Xiang 1047e4937aSGao Xiang #include <trace/events/erofs.h> 1147e4937aSGao Xiang 1247e4937aSGao Xiang /* 139f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 149f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 159f6cc76eSGao Xiang */ 169f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 179f6cc76eSGao Xiang struct kmem_cache *slab; 189f6cc76eSGao Xiang unsigned int maxpages; 199f6cc76eSGao Xiang char name[48]; 209f6cc76eSGao Xiang }; 219f6cc76eSGao Xiang 229f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 239f6cc76eSGao Xiang 249f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 259f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 269f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 279f6cc76eSGao Xiang }; 289f6cc76eSGao Xiang 299f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 309f6cc76eSGao Xiang { 319f6cc76eSGao Xiang int i; 329f6cc76eSGao Xiang 339f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 349f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 359f6cc76eSGao Xiang continue; 369f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 379f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 389f6cc76eSGao Xiang } 399f6cc76eSGao Xiang } 409f6cc76eSGao Xiang 419f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 429f6cc76eSGao Xiang { 439f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 449f6cc76eSGao Xiang struct z_erofs_pcluster *a; 459f6cc76eSGao Xiang unsigned int size; 469f6cc76eSGao Xiang 479f6cc76eSGao Xiang for (pcs = pcluster_pool; 489f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 499f6cc76eSGao Xiang size = struct_size(a, compressed_pages, pcs->maxpages); 509f6cc76eSGao Xiang 519f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 529f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 539f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 549f6cc76eSGao Xiang if (pcs->slab) 559f6cc76eSGao Xiang continue; 569f6cc76eSGao Xiang 579f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 589f6cc76eSGao Xiang return -ENOMEM; 599f6cc76eSGao Xiang } 609f6cc76eSGao Xiang return 0; 619f6cc76eSGao Xiang } 629f6cc76eSGao Xiang 639f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 649f6cc76eSGao Xiang { 659f6cc76eSGao Xiang int i; 669f6cc76eSGao Xiang 679f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 689f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 699f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 709f6cc76eSGao Xiang 719f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 729f6cc76eSGao Xiang continue; 739f6cc76eSGao Xiang 749f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 759f6cc76eSGao Xiang if (!pcl) 769f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 779f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 789f6cc76eSGao Xiang return pcl; 799f6cc76eSGao Xiang } 809f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 819f6cc76eSGao Xiang } 829f6cc76eSGao Xiang 839f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 849f6cc76eSGao Xiang { 85cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 869f6cc76eSGao Xiang int i; 879f6cc76eSGao Xiang 889f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 899f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 909f6cc76eSGao Xiang 91cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 929f6cc76eSGao Xiang continue; 939f6cc76eSGao Xiang 949f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 959f6cc76eSGao Xiang return; 969f6cc76eSGao Xiang } 979f6cc76eSGao Xiang DBG_BUGON(1); 989f6cc76eSGao Xiang } 999f6cc76eSGao Xiang 10047e4937aSGao Xiang /* how to allocate cached pages for a pcluster */ 10147e4937aSGao Xiang enum z_erofs_cache_alloctype { 10247e4937aSGao Xiang DONTALLOC, /* don't allocate any cached pages */ 1031825c8d7SGao Xiang /* 1041825c8d7SGao Xiang * try to use cached I/O if page allocation succeeds or fallback 1051825c8d7SGao Xiang * to in-place I/O instead to avoid any direct reclaim. 1061825c8d7SGao Xiang */ 1071825c8d7SGao Xiang TRYALLOC, 10847e4937aSGao Xiang }; 10947e4937aSGao Xiang 11047e4937aSGao Xiang /* 11147e4937aSGao Xiang * tagged pointer with 1-bit tag for all compressed pages 11247e4937aSGao Xiang * tag 0 - the page is just found with an extra page reference 11347e4937aSGao Xiang */ 11447e4937aSGao Xiang typedef tagptr1_t compressed_page_t; 11547e4937aSGao Xiang 11647e4937aSGao Xiang #define tag_compressed_page_justfound(page) \ 11747e4937aSGao Xiang tagptr_fold(compressed_page_t, page, 1) 11847e4937aSGao Xiang 11947e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 12047e4937aSGao Xiang 12147e4937aSGao Xiang void z_erofs_exit_zip_subsystem(void) 12247e4937aSGao Xiang { 12347e4937aSGao Xiang destroy_workqueue(z_erofs_workqueue); 1249f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 12547e4937aSGao Xiang } 12647e4937aSGao Xiang 12799634bf3SGao Xiang static inline int z_erofs_init_workqueue(void) 12847e4937aSGao Xiang { 12947e4937aSGao Xiang const unsigned int onlinecpus = num_possible_cpus(); 13047e4937aSGao Xiang 13147e4937aSGao Xiang /* 13247e4937aSGao Xiang * no need to spawn too many threads, limiting threads could minimum 13347e4937aSGao Xiang * scheduling overhead, perhaps per-CPU threads should be better? 13447e4937aSGao Xiang */ 1350e62ea33SGao Xiang z_erofs_workqueue = alloc_workqueue("erofs_unzipd", 1360e62ea33SGao Xiang WQ_UNBOUND | WQ_HIGHPRI, 13747e4937aSGao Xiang onlinecpus + onlinecpus / 4); 13847e4937aSGao Xiang return z_erofs_workqueue ? 0 : -ENOMEM; 13947e4937aSGao Xiang } 14047e4937aSGao Xiang 14147e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 14247e4937aSGao Xiang { 1439f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 14447e4937aSGao Xiang 1459f6cc76eSGao Xiang if (err) 1469f6cc76eSGao Xiang return err; 1479f6cc76eSGao Xiang err = z_erofs_init_workqueue(); 1489f6cc76eSGao Xiang if (err) 1499f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 1509f6cc76eSGao Xiang return err; 15147e4937aSGao Xiang } 15247e4937aSGao Xiang 15347e4937aSGao Xiang enum z_erofs_collectmode { 15447e4937aSGao Xiang COLLECT_SECONDARY, 15547e4937aSGao Xiang COLLECT_PRIMARY, 15647e4937aSGao Xiang /* 15747e4937aSGao Xiang * The current collection was the tail of an exist chain, in addition 15847e4937aSGao Xiang * that the previous processed chained collections are all decided to 15947e4937aSGao Xiang * be hooked up to it. 16047e4937aSGao Xiang * A new chain will be created for the remaining collections which are 16147e4937aSGao Xiang * not processed yet, therefore different from COLLECT_PRIMARY_FOLLOWED, 16247e4937aSGao Xiang * the next collection cannot reuse the whole page safely in 16347e4937aSGao Xiang * the following scenario: 16447e4937aSGao Xiang * ________________________________________________________________ 16547e4937aSGao Xiang * | tail (partial) page | head (partial) page | 16647e4937aSGao Xiang * | (belongs to the next cl) | (belongs to the current cl) | 16747e4937aSGao Xiang * |_______PRIMARY_FOLLOWED_______|________PRIMARY_HOOKED___________| 16847e4937aSGao Xiang */ 16947e4937aSGao Xiang COLLECT_PRIMARY_HOOKED, 1700b964600SGao Xiang /* 1710b964600SGao Xiang * a weak form of COLLECT_PRIMARY_FOLLOWED, the difference is that it 1720b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 1730b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 1740b964600SGao Xiang * pagevec) since it can be directly decoded without I/O submission. 1750b964600SGao Xiang */ 17647e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED_NOINPLACE, 17747e4937aSGao Xiang /* 17847e4937aSGao Xiang * The current collection has been linked with the owned chain, and 17947e4937aSGao Xiang * could also be linked with the remaining collections, which means 18047e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 18147e4937aSGao Xiang * the current collection can safely use the whole page (since 18247e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 18347e4937aSGao Xiang * illustrated below: 18447e4937aSGao Xiang * ________________________________________________________________ 18547e4937aSGao Xiang * | tail (partial) page | head (partial) page | 18647e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 18747e4937aSGao Xiang * | PRIMARY_FOLLOWED or | | 18847e4937aSGao Xiang * |_____PRIMARY_HOOKED___|____________PRIMARY_FOLLOWED____________| 18947e4937aSGao Xiang * 19047e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 19147e4937aSGao Xiang */ 19247e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED, 19347e4937aSGao Xiang }; 19447e4937aSGao Xiang 195*5c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 196*5c6dcc57SGao Xiang struct inode *const inode; 197*5c6dcc57SGao Xiang struct erofs_map_blocks map; 198*5c6dcc57SGao Xiang 19947e4937aSGao Xiang struct z_erofs_pagevec_ctor vector; 20047e4937aSGao Xiang 20147e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 20247e4937aSGao Xiang struct z_erofs_collection *cl; 20381382f5fSGao Xiang /* a pointer used to pick up inplace I/O pages */ 20481382f5fSGao Xiang struct page **icpage_ptr; 20547e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 20647e4937aSGao Xiang 20747e4937aSGao Xiang enum z_erofs_collectmode mode; 20847e4937aSGao Xiang 2096ea5aad3SGao Xiang bool readahead; 21047e4937aSGao Xiang /* used for applying cache strategy on the fly */ 21147e4937aSGao Xiang bool backmost; 21247e4937aSGao Xiang erofs_off_t headoffset; 21347e4937aSGao Xiang }; 21447e4937aSGao Xiang 21547e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 216*5c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 217*5c6dcc57SGao Xiang .mode = COLLECT_PRIMARY_FOLLOWED } 21847e4937aSGao Xiang 21947e4937aSGao Xiang static struct page *z_pagemap_global[Z_EROFS_VMAP_GLOBAL_PAGES]; 22047e4937aSGao Xiang static DEFINE_MUTEX(z_pagemap_global_lock); 22147e4937aSGao Xiang 222*5c6dcc57SGao Xiang static void preload_compressed_pages(struct z_erofs_decompress_frontend *fe, 22347e4937aSGao Xiang struct address_space *mc, 2241825c8d7SGao Xiang enum z_erofs_cache_alloctype type, 225eaa9172aSGao Xiang struct page **pagepool) 22647e4937aSGao Xiang { 227*5c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 22847e4937aSGao Xiang bool standalone = true; 2291825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 2301825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 23181382f5fSGao Xiang struct page **pages; 23281382f5fSGao Xiang pgoff_t index; 23347e4937aSGao Xiang 234*5c6dcc57SGao Xiang if (fe->mode < COLLECT_PRIMARY_FOLLOWED) 23547e4937aSGao Xiang return; 23647e4937aSGao Xiang 23781382f5fSGao Xiang pages = pcl->compressed_pages; 23881382f5fSGao Xiang index = pcl->obj.index; 23981382f5fSGao Xiang for (; index < pcl->obj.index + pcl->pclusterpages; ++index, ++pages) { 24047e4937aSGao Xiang struct page *page; 24147e4937aSGao Xiang compressed_page_t t; 2421825c8d7SGao Xiang struct page *newpage = NULL; 24347e4937aSGao Xiang 24447e4937aSGao Xiang /* the compressed page was loaded before */ 24547e4937aSGao Xiang if (READ_ONCE(*pages)) 24647e4937aSGao Xiang continue; 24747e4937aSGao Xiang 24847e4937aSGao Xiang page = find_get_page(mc, index); 24947e4937aSGao Xiang 25047e4937aSGao Xiang if (page) { 25147e4937aSGao Xiang t = tag_compressed_page_justfound(page); 2520b964600SGao Xiang } else { 2530b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 2540b964600SGao Xiang standalone = false; 2550b964600SGao Xiang switch (type) { 2560b964600SGao Xiang case TRYALLOC: 2571825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 2581825c8d7SGao Xiang if (!newpage) 25947e4937aSGao Xiang continue; 2600b964600SGao Xiang set_page_private(newpage, 2610b964600SGao Xiang Z_EROFS_PREALLOCATED_PAGE); 2620b964600SGao Xiang t = tag_compressed_page_justfound(newpage); 2630b964600SGao Xiang break; 2640b964600SGao Xiang default: /* DONTALLOC */ 2650b964600SGao Xiang continue; 2660b964600SGao Xiang } 26747e4937aSGao Xiang } 26847e4937aSGao Xiang 26947e4937aSGao Xiang if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t))) 27047e4937aSGao Xiang continue; 27147e4937aSGao Xiang 272eaa9172aSGao Xiang if (page) 27347e4937aSGao Xiang put_page(page); 274eaa9172aSGao Xiang else if (newpage) 275eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 27647e4937aSGao Xiang } 27747e4937aSGao Xiang 2780b964600SGao Xiang /* 2790b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 2800b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 2810b964600SGao Xiang */ 2820b964600SGao Xiang if (standalone) 283*5c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 28447e4937aSGao Xiang } 28547e4937aSGao Xiang 28647e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 28747e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 28847e4937aSGao Xiang struct erofs_workgroup *grp) 28947e4937aSGao Xiang { 29047e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 29147e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 29247e4937aSGao Xiang int i; 29347e4937aSGao Xiang 294cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 29547e4937aSGao Xiang /* 29647e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 29747e4937aSGao Xiang * therefore no need to worry about available decompression users. 29847e4937aSGao Xiang */ 2999f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 30047e4937aSGao Xiang struct page *page = pcl->compressed_pages[i]; 30147e4937aSGao Xiang 30247e4937aSGao Xiang if (!page) 30347e4937aSGao Xiang continue; 30447e4937aSGao Xiang 30547e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 30647e4937aSGao Xiang if (!trylock_page(page)) 30747e4937aSGao Xiang return -EBUSY; 30847e4937aSGao Xiang 309f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 31047e4937aSGao Xiang continue; 31147e4937aSGao Xiang 31247e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 31347e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 3146aaa7b06SGao Xiang detach_page_private(page); 31547e4937aSGao Xiang unlock_page(page); 31647e4937aSGao Xiang } 31747e4937aSGao Xiang return 0; 31847e4937aSGao Xiang } 31947e4937aSGao Xiang 320d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 32147e4937aSGao Xiang { 32247e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 32347e4937aSGao Xiang int ret = 0; /* 0 - busy */ 32447e4937aSGao Xiang 32547e4937aSGao Xiang if (erofs_workgroup_try_to_freeze(&pcl->obj, 1)) { 32647e4937aSGao Xiang unsigned int i; 32747e4937aSGao Xiang 328cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 3299f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 33047e4937aSGao Xiang if (pcl->compressed_pages[i] == page) { 33147e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 33247e4937aSGao Xiang ret = 1; 33347e4937aSGao Xiang break; 33447e4937aSGao Xiang } 33547e4937aSGao Xiang } 33647e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 33747e4937aSGao Xiang 3386aaa7b06SGao Xiang if (ret) 3396aaa7b06SGao Xiang detach_page_private(page); 34047e4937aSGao Xiang } 34147e4937aSGao Xiang return ret; 34247e4937aSGao Xiang } 34347e4937aSGao Xiang 34447e4937aSGao Xiang /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 345*5c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 34647e4937aSGao Xiang struct page *page) 34747e4937aSGao Xiang { 348*5c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 34947e4937aSGao Xiang 350*5c6dcc57SGao Xiang while (fe->icpage_ptr > pcl->compressed_pages) 351*5c6dcc57SGao Xiang if (!cmpxchg(--fe->icpage_ptr, NULL, page)) 35247e4937aSGao Xiang return true; 35347e4937aSGao Xiang return false; 35447e4937aSGao Xiang } 35547e4937aSGao Xiang 35647e4937aSGao Xiang /* callers must be with collection lock held */ 357*5c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 35886432a6dSGao Xiang struct page *page, enum z_erofs_page_type type, 35986432a6dSGao Xiang bool pvec_safereuse) 36047e4937aSGao Xiang { 36147e4937aSGao Xiang int ret; 36247e4937aSGao Xiang 36347e4937aSGao Xiang /* give priority for inplaceio */ 364*5c6dcc57SGao Xiang if (fe->mode >= COLLECT_PRIMARY && 36547e4937aSGao Xiang type == Z_EROFS_PAGE_TYPE_EXCLUSIVE && 366*5c6dcc57SGao Xiang z_erofs_try_inplace_io(fe, page)) 36747e4937aSGao Xiang return 0; 36847e4937aSGao Xiang 369*5c6dcc57SGao Xiang ret = z_erofs_pagevec_enqueue(&fe->vector, page, type, 37086432a6dSGao Xiang pvec_safereuse); 371*5c6dcc57SGao Xiang fe->cl->vcnt += (unsigned int)ret; 37247e4937aSGao Xiang return ret ? 0 : -EAGAIN; 37347e4937aSGao Xiang } 37447e4937aSGao Xiang 375*5c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 37647e4937aSGao Xiang { 377*5c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 378*5c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 37947e4937aSGao Xiang 380473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 381473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 382473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 38347e4937aSGao Xiang *owned_head = &pcl->next; 384473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 385*5c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_FOLLOWED; 386473e15b0SGao Xiang return; 387473e15b0SGao Xiang } 388473e15b0SGao Xiang 38947e4937aSGao Xiang /* 390473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 391473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 39247e4937aSGao Xiang */ 39347e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 394473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 39547e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 396*5c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_HOOKED; 397*5c6dcc57SGao Xiang f->tailpcl = NULL; 398473e15b0SGao Xiang return; 39947e4937aSGao Xiang } 400473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 401*5c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY; 40247e4937aSGao Xiang } 40347e4937aSGao Xiang 404*5c6dcc57SGao Xiang static int z_erofs_lookup_collection(struct z_erofs_decompress_frontend *fe, 40547e4937aSGao Xiang struct inode *inode, 40647e4937aSGao Xiang struct erofs_map_blocks *map) 40747e4937aSGao Xiang { 408*5c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 40947e4937aSGao Xiang struct z_erofs_collection *cl; 41047e4937aSGao Xiang unsigned int length; 41147e4937aSGao Xiang 41264094a04SGao Xiang /* to avoid unexpected loop formed by corrupted images */ 413*5c6dcc57SGao Xiang if (fe->owned_head == &pcl->next || pcl == fe->tailpcl) { 41447e4937aSGao Xiang DBG_BUGON(1); 4159e579fc1SGao Xiang return -EFSCORRUPTED; 41647e4937aSGao Xiang } 41747e4937aSGao Xiang 41847e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 4198d8a09b0SGao Xiang if (cl->pageofs != (map->m_la & ~PAGE_MASK)) { 42047e4937aSGao Xiang DBG_BUGON(1); 4219e579fc1SGao Xiang return -EFSCORRUPTED; 42247e4937aSGao Xiang } 42347e4937aSGao Xiang 42447e4937aSGao Xiang length = READ_ONCE(pcl->length); 42547e4937aSGao Xiang if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 42647e4937aSGao Xiang if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 42747e4937aSGao Xiang DBG_BUGON(1); 4289e579fc1SGao Xiang return -EFSCORRUPTED; 42947e4937aSGao Xiang } 43047e4937aSGao Xiang } else { 43147e4937aSGao Xiang unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 43247e4937aSGao Xiang 43347e4937aSGao Xiang if (map->m_flags & EROFS_MAP_FULL_MAPPED) 43447e4937aSGao Xiang llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 43547e4937aSGao Xiang 43647e4937aSGao Xiang while (llen > length && 43747e4937aSGao Xiang length != cmpxchg_relaxed(&pcl->length, length, llen)) { 43847e4937aSGao Xiang cpu_relax(); 43947e4937aSGao Xiang length = READ_ONCE(pcl->length); 44047e4937aSGao Xiang } 44147e4937aSGao Xiang } 44247e4937aSGao Xiang mutex_lock(&cl->lock); 44347e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 444*5c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 445*5c6dcc57SGao Xiang fe->tailpcl = pcl; 446473e15b0SGao Xiang 447*5c6dcc57SGao Xiang z_erofs_try_to_claim_pcluster(fe); 448*5c6dcc57SGao Xiang fe->cl = cl; 4499e579fc1SGao Xiang return 0; 45047e4937aSGao Xiang } 45147e4937aSGao Xiang 452*5c6dcc57SGao Xiang static int z_erofs_register_collection(struct z_erofs_decompress_frontend *fe, 45347e4937aSGao Xiang struct inode *inode, 45447e4937aSGao Xiang struct erofs_map_blocks *map) 45547e4937aSGao Xiang { 456cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 45747e4937aSGao Xiang struct z_erofs_pcluster *pcl; 45847e4937aSGao Xiang struct z_erofs_collection *cl; 45964094a04SGao Xiang struct erofs_workgroup *grp; 46047e4937aSGao Xiang int err; 46147e4937aSGao Xiang 4628f899262SGao Xiang if (!(map->m_flags & EROFS_MAP_ENCODED)) { 4638f899262SGao Xiang DBG_BUGON(1); 4648f899262SGao Xiang return -EFSCORRUPTED; 4658f899262SGao Xiang } 4668f899262SGao Xiang 4679f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 468cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 469cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 4709f6cc76eSGao Xiang if (IS_ERR(pcl)) 4719f6cc76eSGao Xiang return PTR_ERR(pcl); 47247e4937aSGao Xiang 47364094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 4748f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 47547e4937aSGao Xiang pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 47647e4937aSGao Xiang (map->m_flags & EROFS_MAP_FULL_MAPPED ? 47747e4937aSGao Xiang Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 47847e4937aSGao Xiang 47947e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 480*5c6dcc57SGao Xiang pcl->next = fe->owned_head; 481*5c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED; 48247e4937aSGao Xiang 48347e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 48447e4937aSGao Xiang cl->pageofs = map->m_la & ~PAGE_MASK; 48547e4937aSGao Xiang 48647e4937aSGao Xiang /* 48747e4937aSGao Xiang * lock all primary followed works before visible to others 48847e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 48947e4937aSGao Xiang */ 4909f6cc76eSGao Xiang mutex_init(&cl->lock); 49164094a04SGao Xiang DBG_BUGON(!mutex_trylock(&cl->lock)); 49247e4937aSGao Xiang 493cecf864dSYue Hu if (ztailpacking) { 494cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 495cecf864dSYue Hu pcl->pageofs_in = erofs_blkoff(map->m_pa); 496cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 497cecf864dSYue Hu } else { 498cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 499cecf864dSYue Hu 50064094a04SGao Xiang grp = erofs_insert_workgroup(inode->i_sb, &pcl->obj); 50164094a04SGao Xiang if (IS_ERR(grp)) { 50264094a04SGao Xiang err = PTR_ERR(grp); 50364094a04SGao Xiang goto err_out; 50464094a04SGao Xiang } 50564094a04SGao Xiang 50664094a04SGao Xiang if (grp != &pcl->obj) { 507*5c6dcc57SGao Xiang fe->pcl = container_of(grp, 508cecf864dSYue Hu struct z_erofs_pcluster, obj); 50964094a04SGao Xiang err = -EEXIST; 51064094a04SGao Xiang goto err_out; 51147e4937aSGao Xiang } 512cecf864dSYue Hu } 51347e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 514*5c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 515*5c6dcc57SGao Xiang fe->tailpcl = pcl; 516*5c6dcc57SGao Xiang fe->owned_head = &pcl->next; 517*5c6dcc57SGao Xiang fe->pcl = pcl; 518*5c6dcc57SGao Xiang fe->cl = cl; 5199e579fc1SGao Xiang return 0; 52064094a04SGao Xiang 52164094a04SGao Xiang err_out: 52264094a04SGao Xiang mutex_unlock(&cl->lock); 5239f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 52464094a04SGao Xiang return err; 52547e4937aSGao Xiang } 52647e4937aSGao Xiang 527*5c6dcc57SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe, 52847e4937aSGao Xiang struct inode *inode, 52947e4937aSGao Xiang struct erofs_map_blocks *map) 53047e4937aSGao Xiang { 53164094a04SGao Xiang struct erofs_workgroup *grp; 5329e579fc1SGao Xiang int ret; 53347e4937aSGao Xiang 534*5c6dcc57SGao Xiang DBG_BUGON(fe->cl); 53547e4937aSGao Xiang 53647e4937aSGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous collection */ 537*5c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 538*5c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 53947e4937aSGao Xiang 540cecf864dSYue Hu if (map->m_flags & EROFS_MAP_META) { 541cecf864dSYue Hu if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 54247e4937aSGao Xiang DBG_BUGON(1); 543cecf864dSYue Hu return -EFSCORRUPTED; 544cecf864dSYue Hu } 545cecf864dSYue Hu goto tailpacking; 54647e4937aSGao Xiang } 54747e4937aSGao Xiang 54864094a04SGao Xiang grp = erofs_find_workgroup(inode->i_sb, map->m_pa >> PAGE_SHIFT); 54964094a04SGao Xiang if (grp) { 550*5c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 55164094a04SGao Xiang } else { 552cecf864dSYue Hu tailpacking: 553*5c6dcc57SGao Xiang ret = z_erofs_register_collection(fe, inode, map); 55464094a04SGao Xiang if (!ret) 55564094a04SGao Xiang goto out; 55664094a04SGao Xiang if (ret != -EEXIST) 5579e579fc1SGao Xiang return ret; 55864094a04SGao Xiang } 55947e4937aSGao Xiang 560*5c6dcc57SGao Xiang ret = z_erofs_lookup_collection(fe, inode, map); 56164094a04SGao Xiang if (ret) { 562*5c6dcc57SGao Xiang erofs_workgroup_put(&fe->pcl->obj); 56364094a04SGao Xiang return ret; 56464094a04SGao Xiang } 56564094a04SGao Xiang 56664094a04SGao Xiang out: 567*5c6dcc57SGao Xiang z_erofs_pagevec_ctor_init(&fe->vector, Z_EROFS_NR_INLINE_PAGEVECS, 568*5c6dcc57SGao Xiang fe->cl->pagevec, fe->cl->vcnt); 56981382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 570*5c6dcc57SGao Xiang fe->icpage_ptr = fe->pcl->compressed_pages + 571*5c6dcc57SGao Xiang z_erofs_pclusterpages(fe->pcl); 57247e4937aSGao Xiang return 0; 57347e4937aSGao Xiang } 57447e4937aSGao Xiang 57547e4937aSGao Xiang /* 57647e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 57747e4937aSGao Xiang * only after a RCU grace period. 57847e4937aSGao Xiang */ 57947e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 58047e4937aSGao Xiang { 58147e4937aSGao Xiang struct z_erofs_collection *const cl = 58247e4937aSGao Xiang container_of(head, struct z_erofs_collection, rcu); 58347e4937aSGao Xiang 5849f6cc76eSGao Xiang z_erofs_free_pcluster(container_of(cl, struct z_erofs_pcluster, 58547e4937aSGao Xiang primary_collection)); 58647e4937aSGao Xiang } 58747e4937aSGao Xiang 58847e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 58947e4937aSGao Xiang { 59047e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 59147e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 59247e4937aSGao Xiang struct z_erofs_collection *const cl = z_erofs_primarycollection(pcl); 59347e4937aSGao Xiang 59447e4937aSGao Xiang call_rcu(&cl->rcu, z_erofs_rcu_callback); 59547e4937aSGao Xiang } 59647e4937aSGao Xiang 59747e4937aSGao Xiang static void z_erofs_collection_put(struct z_erofs_collection *cl) 59847e4937aSGao Xiang { 59947e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 60047e4937aSGao Xiang container_of(cl, struct z_erofs_pcluster, primary_collection); 60147e4937aSGao Xiang 60247e4937aSGao Xiang erofs_workgroup_put(&pcl->obj); 60347e4937aSGao Xiang } 60447e4937aSGao Xiang 605*5c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 60647e4937aSGao Xiang { 607*5c6dcc57SGao Xiang struct z_erofs_collection *cl = fe->cl; 60847e4937aSGao Xiang 60947e4937aSGao Xiang if (!cl) 61047e4937aSGao Xiang return false; 61147e4937aSGao Xiang 612*5c6dcc57SGao Xiang z_erofs_pagevec_ctor_exit(&fe->vector, false); 61347e4937aSGao Xiang mutex_unlock(&cl->lock); 61447e4937aSGao Xiang 61547e4937aSGao Xiang /* 61647e4937aSGao Xiang * if all pending pages are added, don't hold its reference 61747e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 61847e4937aSGao Xiang */ 619*5c6dcc57SGao Xiang if (fe->mode < COLLECT_PRIMARY_FOLLOWED_NOINPLACE) 62047e4937aSGao Xiang z_erofs_collection_put(cl); 62147e4937aSGao Xiang 622*5c6dcc57SGao Xiang fe->cl = NULL; 62347e4937aSGao Xiang return true; 62447e4937aSGao Xiang } 62547e4937aSGao Xiang 62647e4937aSGao Xiang static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe, 62747e4937aSGao Xiang unsigned int cachestrategy, 62847e4937aSGao Xiang erofs_off_t la) 62947e4937aSGao Xiang { 63047e4937aSGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 63147e4937aSGao Xiang return false; 63247e4937aSGao Xiang 63347e4937aSGao Xiang if (fe->backmost) 63447e4937aSGao Xiang return true; 63547e4937aSGao Xiang 63647e4937aSGao Xiang return cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 63747e4937aSGao Xiang la < fe->headoffset; 63847e4937aSGao Xiang } 63947e4937aSGao Xiang 64047e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 641eaa9172aSGao Xiang struct page *page, struct page **pagepool) 64247e4937aSGao Xiang { 64347e4937aSGao Xiang struct inode *const inode = fe->inode; 644bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 64547e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 64647e4937aSGao Xiang const loff_t offset = page_offset(page); 647dc76ea8cSGao Xiang bool tight = true; 64847e4937aSGao Xiang 64947e4937aSGao Xiang enum z_erofs_cache_alloctype cache_strategy; 65047e4937aSGao Xiang enum z_erofs_page_type page_type; 65147e4937aSGao Xiang unsigned int cur, end, spiltted, index; 65247e4937aSGao Xiang int err = 0; 65347e4937aSGao Xiang 65447e4937aSGao Xiang /* register locked file pages as online pages in pack */ 65547e4937aSGao Xiang z_erofs_onlinepage_init(page); 65647e4937aSGao Xiang 65747e4937aSGao Xiang spiltted = 0; 65847e4937aSGao Xiang end = PAGE_SIZE; 65947e4937aSGao Xiang repeat: 66047e4937aSGao Xiang cur = end - 1; 66147e4937aSGao Xiang 66247e4937aSGao Xiang /* lucky, within the range of the current map_blocks */ 66347e4937aSGao Xiang if (offset + cur >= map->m_la && 66447e4937aSGao Xiang offset + cur < map->m_la + map->m_llen) { 66547e4937aSGao Xiang /* didn't get a valid collection previously (very rare) */ 666*5c6dcc57SGao Xiang if (!fe->cl) 66747e4937aSGao Xiang goto restart_now; 66847e4937aSGao Xiang goto hitted; 66947e4937aSGao Xiang } 67047e4937aSGao Xiang 67147e4937aSGao Xiang /* go ahead the next map_blocks */ 6724f761fa2SGao Xiang erofs_dbg("%s: [out-of-range] pos %llu", __func__, offset + cur); 67347e4937aSGao Xiang 674*5c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 67547e4937aSGao Xiang fe->backmost = false; 67647e4937aSGao Xiang 67747e4937aSGao Xiang map->m_la = offset + cur; 67847e4937aSGao Xiang map->m_llen = 0; 67947e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 6808d8a09b0SGao Xiang if (err) 68147e4937aSGao Xiang goto err_out; 68247e4937aSGao Xiang 68347e4937aSGao Xiang restart_now: 6848d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) 68547e4937aSGao Xiang goto hitted; 68647e4937aSGao Xiang 687*5c6dcc57SGao Xiang err = z_erofs_collector_begin(fe, inode, map); 6888d8a09b0SGao Xiang if (err) 68947e4937aSGao Xiang goto err_out; 69047e4937aSGao Xiang 691*5c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 69209c54379SGao Xiang void *mp; 693cecf864dSYue Hu 69409c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 69509c54379SGao Xiang erofs_blknr(map->m_pa), EROFS_NO_KMAP); 69609c54379SGao Xiang if (IS_ERR(mp)) { 69709c54379SGao Xiang err = PTR_ERR(mp); 698cecf864dSYue Hu erofs_err(inode->i_sb, 699cecf864dSYue Hu "failed to get inline page, err %d", err); 700cecf864dSYue Hu goto err_out; 701cecf864dSYue Hu } 70209c54379SGao Xiang get_page(fe->map.buf.page); 703*5c6dcc57SGao Xiang WRITE_ONCE(fe->pcl->compressed_pages[0], fe->map.buf.page); 704*5c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 705cecf864dSYue Hu } else { 706cecf864dSYue Hu /* preload all compressed pages (can change mode if needed) */ 707cecf864dSYue Hu if (should_alloc_managed_pages(fe, sbi->opt.cache_strategy, 708cecf864dSYue Hu map->m_la)) 7091825c8d7SGao Xiang cache_strategy = TRYALLOC; 71047e4937aSGao Xiang else 71147e4937aSGao Xiang cache_strategy = DONTALLOC; 71247e4937aSGao Xiang 713*5c6dcc57SGao Xiang preload_compressed_pages(fe, MNGD_MAPPING(sbi), 7141825c8d7SGao Xiang cache_strategy, pagepool); 715cecf864dSYue Hu } 71647e4937aSGao Xiang 71747e4937aSGao Xiang hitted: 718dc76ea8cSGao Xiang /* 719dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 720dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 721dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 722dc76ea8cSGao Xiang * for inplace I/O or pagevec (should be processed in strict order.) 723dc76ea8cSGao Xiang */ 724*5c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_HOOKED && 725*5c6dcc57SGao Xiang fe->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 726dc76ea8cSGao Xiang 72747e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 7288d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 72947e4937aSGao Xiang zero_user_segment(page, cur, end); 73047e4937aSGao Xiang goto next_part; 73147e4937aSGao Xiang } 73247e4937aSGao Xiang 73347e4937aSGao Xiang /* let's derive page type */ 73447e4937aSGao Xiang page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 73547e4937aSGao Xiang (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 73647e4937aSGao Xiang (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 73747e4937aSGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 73847e4937aSGao Xiang 73947e4937aSGao Xiang if (cur) 740*5c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_FOLLOWED); 74147e4937aSGao Xiang 74247e4937aSGao Xiang retry: 743*5c6dcc57SGao Xiang err = z_erofs_attach_page(fe, page, page_type, 744*5c6dcc57SGao Xiang fe->mode >= COLLECT_PRIMARY_FOLLOWED); 7456aaa7b06SGao Xiang /* should allocate an additional short-lived page for pagevec */ 74647e4937aSGao Xiang if (err == -EAGAIN) { 74747e4937aSGao Xiang struct page *const newpage = 748e3f78d5eSChao Yu alloc_page(GFP_NOFS | __GFP_NOFAIL); 74947e4937aSGao Xiang 7506aaa7b06SGao Xiang set_page_private(newpage, Z_EROFS_SHORTLIVED_PAGE); 751*5c6dcc57SGao Xiang err = z_erofs_attach_page(fe, newpage, 75286432a6dSGao Xiang Z_EROFS_PAGE_TYPE_EXCLUSIVE, true); 7538d8a09b0SGao Xiang if (!err) 75447e4937aSGao Xiang goto retry; 75547e4937aSGao Xiang } 75647e4937aSGao Xiang 7578d8a09b0SGao Xiang if (err) 75847e4937aSGao Xiang goto err_out; 75947e4937aSGao Xiang 76047e4937aSGao Xiang index = page->index - (map->m_la >> PAGE_SHIFT); 76147e4937aSGao Xiang 76247e4937aSGao Xiang z_erofs_onlinepage_fixup(page, index, true); 76347e4937aSGao Xiang 76447e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 76547e4937aSGao Xiang ++spiltted; 76647e4937aSGao Xiang /* also update nr_pages */ 767*5c6dcc57SGao Xiang fe->cl->nr_pages = max_t(pgoff_t, fe->cl->nr_pages, index + 1); 76847e4937aSGao Xiang next_part: 76947e4937aSGao Xiang /* can be used for verification */ 77047e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 77147e4937aSGao Xiang 77247e4937aSGao Xiang end = cur; 77347e4937aSGao Xiang if (end > 0) 77447e4937aSGao Xiang goto repeat; 77547e4937aSGao Xiang 77647e4937aSGao Xiang out: 77747e4937aSGao Xiang z_erofs_onlinepage_endio(page); 77847e4937aSGao Xiang 7794f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 78047e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 78147e4937aSGao Xiang return err; 78247e4937aSGao Xiang 78347e4937aSGao Xiang /* if some error occurred while processing this page */ 78447e4937aSGao Xiang err_out: 78547e4937aSGao Xiang SetPageError(page); 78647e4937aSGao Xiang goto out; 78747e4937aSGao Xiang } 78847e4937aSGao Xiang 78940452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 79040452ffcSHuang Jianan unsigned int readahead_pages) 79140452ffcSHuang Jianan { 79240452ffcSHuang Jianan /* auto: enable for readpage, disable for readahead */ 79340452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 79440452ffcSHuang Jianan !readahead_pages) 79540452ffcSHuang Jianan return true; 79640452ffcSHuang Jianan 79740452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 79840452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 79940452ffcSHuang Jianan return true; 80040452ffcSHuang Jianan 80140452ffcSHuang Jianan return false; 80240452ffcSHuang Jianan } 80340452ffcSHuang Jianan 8046aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 8056aaa7b06SGao Xiang { 8066aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 8076aaa7b06SGao Xiang } 8086aaa7b06SGao Xiang 80947e4937aSGao Xiang static int z_erofs_decompress_pcluster(struct super_block *sb, 81047e4937aSGao Xiang struct z_erofs_pcluster *pcl, 811eaa9172aSGao Xiang struct page **pagepool) 81247e4937aSGao Xiang { 81347e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 814cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 81547e4937aSGao Xiang struct z_erofs_pagevec_ctor ctor; 8169f6cc76eSGao Xiang unsigned int i, inputsize, outputsize, llen, nr_pages; 81747e4937aSGao Xiang struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 81847e4937aSGao Xiang struct page **pages, **compressed_pages, *page; 81947e4937aSGao Xiang 82047e4937aSGao Xiang enum z_erofs_page_type page_type; 82147e4937aSGao Xiang bool overlapped, partial; 82247e4937aSGao Xiang struct z_erofs_collection *cl; 82347e4937aSGao Xiang int err; 82447e4937aSGao Xiang 82547e4937aSGao Xiang might_sleep(); 82647e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 82747e4937aSGao Xiang DBG_BUGON(!READ_ONCE(cl->nr_pages)); 82847e4937aSGao Xiang 82947e4937aSGao Xiang mutex_lock(&cl->lock); 83047e4937aSGao Xiang nr_pages = cl->nr_pages; 83147e4937aSGao Xiang 8328d8a09b0SGao Xiang if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 83347e4937aSGao Xiang pages = pages_onstack; 83447e4937aSGao Xiang } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 83547e4937aSGao Xiang mutex_trylock(&z_pagemap_global_lock)) { 83647e4937aSGao Xiang pages = z_pagemap_global; 83747e4937aSGao Xiang } else { 83847e4937aSGao Xiang gfp_t gfp_flags = GFP_KERNEL; 83947e4937aSGao Xiang 84047e4937aSGao Xiang if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 84147e4937aSGao Xiang gfp_flags |= __GFP_NOFAIL; 84247e4937aSGao Xiang 84347e4937aSGao Xiang pages = kvmalloc_array(nr_pages, sizeof(struct page *), 84447e4937aSGao Xiang gfp_flags); 84547e4937aSGao Xiang 84647e4937aSGao Xiang /* fallback to global pagemap for the lowmem scenario */ 8478d8a09b0SGao Xiang if (!pages) { 84847e4937aSGao Xiang mutex_lock(&z_pagemap_global_lock); 84947e4937aSGao Xiang pages = z_pagemap_global; 85047e4937aSGao Xiang } 85147e4937aSGao Xiang } 85247e4937aSGao Xiang 85347e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) 85447e4937aSGao Xiang pages[i] = NULL; 85547e4937aSGao Xiang 85647e4937aSGao Xiang err = 0; 85747e4937aSGao Xiang z_erofs_pagevec_ctor_init(&ctor, Z_EROFS_NR_INLINE_PAGEVECS, 85847e4937aSGao Xiang cl->pagevec, 0); 85947e4937aSGao Xiang 86047e4937aSGao Xiang for (i = 0; i < cl->vcnt; ++i) { 86147e4937aSGao Xiang unsigned int pagenr; 86247e4937aSGao Xiang 86347e4937aSGao Xiang page = z_erofs_pagevec_dequeue(&ctor, &page_type); 86447e4937aSGao Xiang 86547e4937aSGao Xiang /* all pages in pagevec ought to be valid */ 86647e4937aSGao Xiang DBG_BUGON(!page); 8676aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 86847e4937aSGao Xiang 8696aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 87047e4937aSGao Xiang continue; 87147e4937aSGao Xiang 87247e4937aSGao Xiang if (page_type == Z_EROFS_VLE_PAGE_TYPE_HEAD) 87347e4937aSGao Xiang pagenr = 0; 87447e4937aSGao Xiang else 87547e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 87647e4937aSGao Xiang 87747e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 87847e4937aSGao Xiang 87947e4937aSGao Xiang /* 88047e4937aSGao Xiang * currently EROFS doesn't support multiref(dedup), 88147e4937aSGao Xiang * so here erroring out one multiref page. 88247e4937aSGao Xiang */ 8838d8a09b0SGao Xiang if (pages[pagenr]) { 88447e4937aSGao Xiang DBG_BUGON(1); 88547e4937aSGao Xiang SetPageError(pages[pagenr]); 88647e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 88747e4937aSGao Xiang err = -EFSCORRUPTED; 88847e4937aSGao Xiang } 88947e4937aSGao Xiang pages[pagenr] = page; 89047e4937aSGao Xiang } 89147e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&ctor, true); 89247e4937aSGao Xiang 89347e4937aSGao Xiang overlapped = false; 89447e4937aSGao Xiang compressed_pages = pcl->compressed_pages; 89547e4937aSGao Xiang 896cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 89747e4937aSGao Xiang unsigned int pagenr; 89847e4937aSGao Xiang 89947e4937aSGao Xiang page = compressed_pages[i]; 90047e4937aSGao Xiang /* all compressed pages ought to be valid */ 90147e4937aSGao Xiang DBG_BUGON(!page); 90247e4937aSGao Xiang 903cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 904cecf864dSYue Hu if (!PageUptodate(page)) 905cecf864dSYue Hu err = -EIO; 906cecf864dSYue Hu continue; 907cecf864dSYue Hu } 908cecf864dSYue Hu 909cecf864dSYue Hu DBG_BUGON(z_erofs_page_is_invalidated(page)); 9106aaa7b06SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 91147e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) { 9128d8a09b0SGao Xiang if (!PageUptodate(page)) 91347e4937aSGao Xiang err = -EIO; 91447e4937aSGao Xiang continue; 91547e4937aSGao Xiang } 91647e4937aSGao Xiang 91747e4937aSGao Xiang /* 91847e4937aSGao Xiang * only if non-head page can be selected 91947e4937aSGao Xiang * for inplace decompression 92047e4937aSGao Xiang */ 92147e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 92247e4937aSGao Xiang 92347e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 9248d8a09b0SGao Xiang if (pages[pagenr]) { 92547e4937aSGao Xiang DBG_BUGON(1); 92647e4937aSGao Xiang SetPageError(pages[pagenr]); 92747e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 92847e4937aSGao Xiang err = -EFSCORRUPTED; 92947e4937aSGao Xiang } 93047e4937aSGao Xiang pages[pagenr] = page; 93147e4937aSGao Xiang 93247e4937aSGao Xiang overlapped = true; 93347e4937aSGao Xiang } 93447e4937aSGao Xiang 9356aaa7b06SGao Xiang /* PG_error needs checking for all non-managed pages */ 9368d8a09b0SGao Xiang if (PageError(page)) { 93747e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 93847e4937aSGao Xiang err = -EIO; 93947e4937aSGao Xiang } 94047e4937aSGao Xiang } 94147e4937aSGao Xiang 9428d8a09b0SGao Xiang if (err) 94347e4937aSGao Xiang goto out; 94447e4937aSGao Xiang 94547e4937aSGao Xiang llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 94647e4937aSGao Xiang if (nr_pages << PAGE_SHIFT >= cl->pageofs + llen) { 94747e4937aSGao Xiang outputsize = llen; 94847e4937aSGao Xiang partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 94947e4937aSGao Xiang } else { 95047e4937aSGao Xiang outputsize = (nr_pages << PAGE_SHIFT) - cl->pageofs; 95147e4937aSGao Xiang partial = true; 95247e4937aSGao Xiang } 95347e4937aSGao Xiang 954cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 955cecf864dSYue Hu inputsize = pcl->tailpacking_size; 956cecf864dSYue Hu else 957cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 958cecf864dSYue Hu 95947e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 96047e4937aSGao Xiang .sb = sb, 96147e4937aSGao Xiang .in = compressed_pages, 96247e4937aSGao Xiang .out = pages, 963cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 96447e4937aSGao Xiang .pageofs_out = cl->pageofs, 9659f6cc76eSGao Xiang .inputsize = inputsize, 96647e4937aSGao Xiang .outputsize = outputsize, 96747e4937aSGao Xiang .alg = pcl->algorithmformat, 96847e4937aSGao Xiang .inplace_io = overlapped, 96947e4937aSGao Xiang .partial_decoding = partial 97047e4937aSGao Xiang }, pagepool); 97147e4937aSGao Xiang 97247e4937aSGao Xiang out: 973cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 974cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 975cecf864dSYue Hu page = compressed_pages[0]; 976cecf864dSYue Hu WRITE_ONCE(compressed_pages[0], NULL); 977cecf864dSYue Hu put_page(page); 978cecf864dSYue Hu } else { 979cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 98047e4937aSGao Xiang page = compressed_pages[i]; 98147e4937aSGao Xiang 98247e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 98347e4937aSGao Xiang continue; 98447e4937aSGao Xiang 9856aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 9866aaa7b06SGao Xiang (void)z_erofs_put_shortlivedpage(pagepool, page); 98747e4937aSGao Xiang WRITE_ONCE(compressed_pages[i], NULL); 98847e4937aSGao Xiang } 989cecf864dSYue Hu } 99047e4937aSGao Xiang 99147e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) { 99247e4937aSGao Xiang page = pages[i]; 99347e4937aSGao Xiang if (!page) 99447e4937aSGao Xiang continue; 99547e4937aSGao Xiang 9966aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 99747e4937aSGao Xiang 9986aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 9996aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 100047e4937aSGao Xiang continue; 100147e4937aSGao Xiang 10028d8a09b0SGao Xiang if (err < 0) 100347e4937aSGao Xiang SetPageError(page); 100447e4937aSGao Xiang 100547e4937aSGao Xiang z_erofs_onlinepage_endio(page); 100647e4937aSGao Xiang } 100747e4937aSGao Xiang 100847e4937aSGao Xiang if (pages == z_pagemap_global) 100947e4937aSGao Xiang mutex_unlock(&z_pagemap_global_lock); 10108d8a09b0SGao Xiang else if (pages != pages_onstack) 101147e4937aSGao Xiang kvfree(pages); 101247e4937aSGao Xiang 101347e4937aSGao Xiang cl->nr_pages = 0; 101447e4937aSGao Xiang cl->vcnt = 0; 101547e4937aSGao Xiang 101647e4937aSGao Xiang /* all cl locks MUST be taken before the following line */ 101747e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 101847e4937aSGao Xiang 101947e4937aSGao Xiang /* all cl locks SHOULD be released right now */ 102047e4937aSGao Xiang mutex_unlock(&cl->lock); 102147e4937aSGao Xiang 102247e4937aSGao Xiang z_erofs_collection_put(cl); 102347e4937aSGao Xiang return err; 102447e4937aSGao Xiang } 102547e4937aSGao Xiang 10260c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1027eaa9172aSGao Xiang struct page **pagepool) 102847e4937aSGao Xiang { 102947e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 103047e4937aSGao Xiang 103147e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 103247e4937aSGao Xiang struct z_erofs_pcluster *pcl; 103347e4937aSGao Xiang 103447e4937aSGao Xiang /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 103547e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 103647e4937aSGao Xiang 103747e4937aSGao Xiang /* no possible that 'owned' equals NULL */ 103847e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 103947e4937aSGao Xiang 104047e4937aSGao Xiang pcl = container_of(owned, struct z_erofs_pcluster, next); 104147e4937aSGao Xiang owned = READ_ONCE(pcl->next); 104247e4937aSGao Xiang 1043a4b1fab1SGao Xiang z_erofs_decompress_pcluster(io->sb, pcl, pagepool); 104447e4937aSGao Xiang } 104547e4937aSGao Xiang } 104647e4937aSGao Xiang 10470c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 104847e4937aSGao Xiang { 1049a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1050a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1051eaa9172aSGao Xiang struct page *pagepool = NULL; 105247e4937aSGao Xiang 1053a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 10540c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 105547e4937aSGao Xiang 1056eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1057a4b1fab1SGao Xiang kvfree(bgq); 105847e4937aSGao Xiang } 105947e4937aSGao Xiang 10607865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 10617865827cSGao Xiang bool sync, int bios) 10627865827cSGao Xiang { 10637865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 10647865827cSGao Xiang 10657865827cSGao Xiang /* wake up the caller thread for sync decompression */ 10667865827cSGao Xiang if (sync) { 10677865827cSGao Xiang unsigned long flags; 10687865827cSGao Xiang 10697865827cSGao Xiang spin_lock_irqsave(&io->u.wait.lock, flags); 10707865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 10717865827cSGao Xiang wake_up_locked(&io->u.wait); 10727865827cSGao Xiang spin_unlock_irqrestore(&io->u.wait.lock, flags); 10737865827cSGao Xiang return; 10747865827cSGao Xiang } 10757865827cSGao Xiang 10767865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 10777865827cSGao Xiang return; 10787865827cSGao Xiang /* Use workqueue and sync decompression for atomic contexts only */ 10797865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 10807865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 10817865827cSGao Xiang /* enable sync decompression for readahead */ 10827865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 10837865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 10847865827cSGao Xiang return; 10857865827cSGao Xiang } 10867865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 10877865827cSGao Xiang } 10887865827cSGao Xiang 108947e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 109047e4937aSGao Xiang unsigned int nr, 1091eaa9172aSGao Xiang struct page **pagepool, 109247e4937aSGao Xiang struct address_space *mc, 109347e4937aSGao Xiang gfp_t gfp) 109447e4937aSGao Xiang { 109547e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 109647e4937aSGao Xiang bool tocache = false; 109747e4937aSGao Xiang 109847e4937aSGao Xiang struct address_space *mapping; 109947e4937aSGao Xiang struct page *oldpage, *page; 110047e4937aSGao Xiang 110147e4937aSGao Xiang compressed_page_t t; 110247e4937aSGao Xiang int justfound; 110347e4937aSGao Xiang 110447e4937aSGao Xiang repeat: 110547e4937aSGao Xiang page = READ_ONCE(pcl->compressed_pages[nr]); 110647e4937aSGao Xiang oldpage = page; 110747e4937aSGao Xiang 110847e4937aSGao Xiang if (!page) 110947e4937aSGao Xiang goto out_allocpage; 111047e4937aSGao Xiang 111147e4937aSGao Xiang /* process the target tagged pointer */ 111247e4937aSGao Xiang t = tagptr_init(compressed_page_t, page); 111347e4937aSGao Xiang justfound = tagptr_unfold_tags(t); 111447e4937aSGao Xiang page = tagptr_unfold_ptr(t); 111547e4937aSGao Xiang 11161825c8d7SGao Xiang /* 11171825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 11181825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 11191825c8d7SGao Xiang */ 11201825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 11211825c8d7SGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 11221825c8d7SGao Xiang set_page_private(page, 0); 11231825c8d7SGao Xiang tocache = true; 11241825c8d7SGao Xiang goto out_tocache; 11251825c8d7SGao Xiang } 112647e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 112747e4937aSGao Xiang 112847e4937aSGao Xiang /* 11296aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 113047e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 113147e4937aSGao Xiang */ 113247e4937aSGao Xiang if (mapping && mapping != mc) 113347e4937aSGao Xiang /* ought to be unmanaged pages */ 113447e4937aSGao Xiang goto out; 113547e4937aSGao Xiang 11366aaa7b06SGao Xiang /* directly return for shortlived page as well */ 11376aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 11386aaa7b06SGao Xiang goto out; 11396aaa7b06SGao Xiang 114047e4937aSGao Xiang lock_page(page); 114147e4937aSGao Xiang 114247e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 114347e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 114447e4937aSGao Xiang 114547e4937aSGao Xiang /* the page is still in manage cache */ 114647e4937aSGao Xiang if (page->mapping == mc) { 114747e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 114847e4937aSGao Xiang 114947e4937aSGao Xiang ClearPageError(page); 115047e4937aSGao Xiang if (!PagePrivate(page)) { 115147e4937aSGao Xiang /* 115247e4937aSGao Xiang * impossible to be !PagePrivate(page) for 115347e4937aSGao Xiang * the current restriction as well if 115447e4937aSGao Xiang * the page is already in compressed_pages[]. 115547e4937aSGao Xiang */ 115647e4937aSGao Xiang DBG_BUGON(!justfound); 115747e4937aSGao Xiang 115847e4937aSGao Xiang justfound = 0; 115947e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 116047e4937aSGao Xiang SetPagePrivate(page); 116147e4937aSGao Xiang } 116247e4937aSGao Xiang 116347e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 116447e4937aSGao Xiang if (PageUptodate(page)) { 116547e4937aSGao Xiang unlock_page(page); 116647e4937aSGao Xiang page = NULL; 116747e4937aSGao Xiang } 116847e4937aSGao Xiang goto out; 116947e4937aSGao Xiang } 117047e4937aSGao Xiang 117147e4937aSGao Xiang /* 117247e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 117347e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 117447e4937aSGao Xiang */ 117547e4937aSGao Xiang DBG_BUGON(page->mapping); 117647e4937aSGao Xiang DBG_BUGON(!justfound); 117747e4937aSGao Xiang 117847e4937aSGao Xiang tocache = true; 117947e4937aSGao Xiang unlock_page(page); 118047e4937aSGao Xiang put_page(page); 118147e4937aSGao Xiang out_allocpage: 11825ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 11835ddcee1fSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_pages[nr], oldpage, page)) { 1184eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 11855ddcee1fSGao Xiang cond_resched(); 11865ddcee1fSGao Xiang goto repeat; 11875ddcee1fSGao Xiang } 11881825c8d7SGao Xiang out_tocache: 1189bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1190bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1191bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1192bf225074SGao Xiang goto out; 1193a30573b3SGao Xiang } 1194bf225074SGao Xiang attach_page_private(page, pcl); 1195bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1196bf225074SGao Xiang put_page(page); 1197bf225074SGao Xiang 119847e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 119947e4937aSGao Xiang return page; 120047e4937aSGao Xiang } 120147e4937aSGao Xiang 1202a4b1fab1SGao Xiang static struct z_erofs_decompressqueue * 1203a4b1fab1SGao Xiang jobqueue_init(struct super_block *sb, 1204a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 120547e4937aSGao Xiang { 1206a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 120747e4937aSGao Xiang 1208a4b1fab1SGao Xiang if (fg && !*fg) { 1209a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1210a4b1fab1SGao Xiang if (!q) { 1211a4b1fab1SGao Xiang *fg = true; 1212a4b1fab1SGao Xiang goto fg_out; 121347e4937aSGao Xiang } 12140c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1215a4b1fab1SGao Xiang } else { 1216a4b1fab1SGao Xiang fg_out: 1217a4b1fab1SGao Xiang q = fgq; 1218a4b1fab1SGao Xiang init_waitqueue_head(&fgq->u.wait); 1219a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 1220a4b1fab1SGao Xiang } 1221a4b1fab1SGao Xiang q->sb = sb; 1222a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1223a4b1fab1SGao Xiang return q; 122447e4937aSGao Xiang } 122547e4937aSGao Xiang 122647e4937aSGao Xiang /* define decompression jobqueue types */ 122747e4937aSGao Xiang enum { 122847e4937aSGao Xiang JQ_BYPASS, 122947e4937aSGao Xiang JQ_SUBMIT, 123047e4937aSGao Xiang NR_JOBQUEUES, 123147e4937aSGao Xiang }; 123247e4937aSGao Xiang 123347e4937aSGao Xiang static void *jobqueueset_init(struct super_block *sb, 1234a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[], 1235a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 123647e4937aSGao Xiang { 123747e4937aSGao Xiang /* 123847e4937aSGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 123947e4937aSGao Xiang * no need to read from device for all pclusters in this queue. 124047e4937aSGao Xiang */ 1241a4b1fab1SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1242a4b1fab1SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 124347e4937aSGao Xiang 1244a4b1fab1SGao Xiang return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 124547e4937aSGao Xiang } 124647e4937aSGao Xiang 124747e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 124847e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 124947e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 125047e4937aSGao Xiang { 125147e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 125247e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 125347e4937aSGao Xiang 125447e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 125547e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 125647e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 125747e4937aSGao Xiang 125847e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 125947e4937aSGao Xiang 126047e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 126147e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 126247e4937aSGao Xiang 126347e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 126447e4937aSGao Xiang } 126547e4937aSGao Xiang 12667865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 12677865827cSGao Xiang { 12687865827cSGao Xiang tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 12697865827cSGao Xiang struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 12707865827cSGao Xiang blk_status_t err = bio->bi_status; 12717865827cSGao Xiang struct bio_vec *bvec; 12727865827cSGao Xiang struct bvec_iter_all iter_all; 12737865827cSGao Xiang 12747865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 12757865827cSGao Xiang struct page *page = bvec->bv_page; 12767865827cSGao Xiang 12777865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 12787865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 12797865827cSGao Xiang 12807865827cSGao Xiang if (err) 12817865827cSGao Xiang SetPageError(page); 12827865827cSGao Xiang 12837865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 12847865827cSGao Xiang if (!err) 12857865827cSGao Xiang SetPageUptodate(page); 12867865827cSGao Xiang unlock_page(page); 12877865827cSGao Xiang } 12887865827cSGao Xiang } 12897865827cSGao Xiang z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 12907865827cSGao Xiang bio_put(bio); 12917865827cSGao Xiang } 12927865827cSGao Xiang 12931e4a2955SGao Xiang static void z_erofs_submit_queue(struct super_block *sb, 12946ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1295eaa9172aSGao Xiang struct page **pagepool, 1296a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1297a4b1fab1SGao Xiang bool *force_fg) 129847e4937aSGao Xiang { 1299bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 130047e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1301a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 130247e4937aSGao Xiang void *bi_private; 1303*5c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1304dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 13053f649ab7SKees Cook pgoff_t last_index; 1306dfeab2e9SGao Xiang struct block_device *last_bdev; 13071e4a2955SGao Xiang unsigned int nr_bios = 0; 13081e4a2955SGao Xiang struct bio *bio = NULL; 130947e4937aSGao Xiang 1310a4b1fab1SGao Xiang bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1311a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1312a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 131347e4937aSGao Xiang 131447e4937aSGao Xiang /* by default, all need io submission */ 131547e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 131647e4937aSGao Xiang 131747e4937aSGao Xiang do { 1318dfeab2e9SGao Xiang struct erofs_map_dev mdev; 131947e4937aSGao Xiang struct z_erofs_pcluster *pcl; 13201e4a2955SGao Xiang pgoff_t cur, end; 13211e4a2955SGao Xiang unsigned int i = 0; 13221e4a2955SGao Xiang bool bypass = true; 132347e4937aSGao Xiang 132447e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 132547e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 132647e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 132747e4937aSGao Xiang 132847e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 132947e4937aSGao Xiang 1330cecf864dSYue Hu /* close the main owned chain at first */ 1331cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1332cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1333cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1334cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1335cecf864dSYue Hu continue; 1336cecf864dSYue Hu } 1337cecf864dSYue Hu 1338dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1339dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1340dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1341dfeab2e9SGao Xiang }; 1342dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1343dfeab2e9SGao Xiang 1344dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 13459f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 134647e4937aSGao Xiang 13471e4a2955SGao Xiang do { 13481e4a2955SGao Xiang struct page *page; 134947e4937aSGao Xiang 13501e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 135147e4937aSGao Xiang MNGD_MAPPING(sbi), 135247e4937aSGao Xiang GFP_NOFS); 13531e4a2955SGao Xiang if (!page) 13541e4a2955SGao Xiang continue; 135547e4937aSGao Xiang 1356dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1357dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 135847e4937aSGao Xiang submit_bio_retry: 135994e4e153SGao Xiang submit_bio(bio); 136047e4937aSGao Xiang bio = NULL; 136147e4937aSGao Xiang } 136247e4937aSGao Xiang 136347e4937aSGao Xiang if (!bio) { 1364a8affc03SChristoph Hellwig bio = bio_alloc(GFP_NOIO, BIO_MAX_VECS); 13650c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1366dfeab2e9SGao Xiang 1367dfeab2e9SGao Xiang bio_set_dev(bio, mdev.m_bdev); 1368dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 13691e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1370a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1371a5c0b780SGao Xiang bio->bi_private = bi_private; 137294e4e153SGao Xiang bio->bi_opf = REQ_OP_READ; 13736ea5aad3SGao Xiang if (f->readahead) 13746ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 137547e4937aSGao Xiang ++nr_bios; 137647e4937aSGao Xiang } 137747e4937aSGao Xiang 13786c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 137947e4937aSGao Xiang goto submit_bio_retry; 138047e4937aSGao Xiang 13811e4a2955SGao Xiang last_index = cur; 13821e4a2955SGao Xiang bypass = false; 13831e4a2955SGao Xiang } while (++cur < end); 138447e4937aSGao Xiang 13851e4a2955SGao Xiang if (!bypass) 138647e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 138747e4937aSGao Xiang else 138847e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 138947e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 139047e4937aSGao Xiang 139147e4937aSGao Xiang if (bio) 139294e4e153SGao Xiang submit_bio(bio); 139347e4937aSGao Xiang 1394587a67b7SGao Xiang /* 1395587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1396587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1397587a67b7SGao Xiang */ 1398587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1399587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 14001e4a2955SGao Xiang return; 1401587a67b7SGao Xiang } 1402a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 140347e4937aSGao Xiang } 140447e4937aSGao Xiang 14050c638f70SGao Xiang static void z_erofs_runqueue(struct super_block *sb, 14066ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1407eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 140847e4937aSGao Xiang { 1409a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 141047e4937aSGao Xiang 1411*5c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 141247e4937aSGao Xiang return; 14136ea5aad3SGao Xiang z_erofs_submit_queue(sb, f, pagepool, io, &force_fg); 141447e4937aSGao Xiang 14150c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 14160c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 141747e4937aSGao Xiang 141847e4937aSGao Xiang if (!force_fg) 141947e4937aSGao Xiang return; 142047e4937aSGao Xiang 142147e4937aSGao Xiang /* wait until all bios are completed */ 1422a93f8c36SGao Xiang io_wait_event(io[JQ_SUBMIT].u.wait, 142347e4937aSGao Xiang !atomic_read(&io[JQ_SUBMIT].pending_bios)); 142447e4937aSGao Xiang 14250c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 14260c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 142747e4937aSGao Xiang } 142847e4937aSGao Xiang 142938629291SGao Xiang /* 143038629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 143138629291SGao Xiang * approximate readmore strategies as a start. 143238629291SGao Xiang */ 143338629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 143438629291SGao Xiang struct readahead_control *rac, 143538629291SGao Xiang erofs_off_t end, 1436eaa9172aSGao Xiang struct page **pagepool, 143738629291SGao Xiang bool backmost) 143838629291SGao Xiang { 143938629291SGao Xiang struct inode *inode = f->inode; 144038629291SGao Xiang struct erofs_map_blocks *map = &f->map; 144138629291SGao Xiang erofs_off_t cur; 144238629291SGao Xiang int err; 144338629291SGao Xiang 144438629291SGao Xiang if (backmost) { 144538629291SGao Xiang map->m_la = end; 1446622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1447622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 144838629291SGao Xiang if (err) 144938629291SGao Xiang return; 145038629291SGao Xiang 145138629291SGao Xiang /* expend ra for the trailing edge if readahead */ 145238629291SGao Xiang if (rac) { 145338629291SGao Xiang loff_t newstart = readahead_pos(rac); 145438629291SGao Xiang 145538629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 145638629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 145738629291SGao Xiang return; 145838629291SGao Xiang } 145938629291SGao Xiang end = round_up(end, PAGE_SIZE); 146038629291SGao Xiang } else { 146138629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 146238629291SGao Xiang 146338629291SGao Xiang if (!map->m_llen) 146438629291SGao Xiang return; 146538629291SGao Xiang } 146638629291SGao Xiang 146738629291SGao Xiang cur = map->m_la + map->m_llen - 1; 146838629291SGao Xiang while (cur >= end) { 146938629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 147038629291SGao Xiang struct page *page; 147138629291SGao Xiang 147238629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 147338629291SGao Xiang if (!page) 147438629291SGao Xiang goto skip; 147538629291SGao Xiang 147638629291SGao Xiang if (PageUptodate(page)) { 147738629291SGao Xiang unlock_page(page); 147838629291SGao Xiang put_page(page); 147938629291SGao Xiang goto skip; 148038629291SGao Xiang } 148138629291SGao Xiang 148238629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 148338629291SGao Xiang if (err) 148438629291SGao Xiang erofs_err(inode->i_sb, 148538629291SGao Xiang "readmore error at page %lu @ nid %llu", 148638629291SGao Xiang index, EROFS_I(inode)->nid); 148738629291SGao Xiang put_page(page); 148838629291SGao Xiang skip: 148938629291SGao Xiang if (cur < PAGE_SIZE) 149038629291SGao Xiang break; 149138629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 149238629291SGao Xiang } 149338629291SGao Xiang } 149438629291SGao Xiang 14950c638f70SGao Xiang static int z_erofs_readpage(struct file *file, struct page *page) 149647e4937aSGao Xiang { 149747e4937aSGao Xiang struct inode *const inode = page->mapping->host; 149840452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 149947e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1500eaa9172aSGao Xiang struct page *pagepool = NULL; 150147e4937aSGao Xiang int err; 150247e4937aSGao Xiang 150347e4937aSGao Xiang trace_erofs_readpage(page, false); 150447e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 150547e4937aSGao Xiang 150638629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 150738629291SGao Xiang &pagepool, true); 15081825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 150938629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 151038629291SGao Xiang 1511*5c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 151247e4937aSGao Xiang 151347e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 151440452ffcSHuang Jianan z_erofs_runqueue(inode->i_sb, &f, &pagepool, 151540452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 151647e4937aSGao Xiang 151747e4937aSGao Xiang if (err) 15184f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 151947e4937aSGao Xiang 152009c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1521eaa9172aSGao Xiang erofs_release_pages(&pagepool); 152247e4937aSGao Xiang return err; 152347e4937aSGao Xiang } 152447e4937aSGao Xiang 15250615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 152647e4937aSGao Xiang { 15270615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 152847e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 152947e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1530eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 153138629291SGao Xiang unsigned int nr_pages; 153247e4937aSGao Xiang 15336ea5aad3SGao Xiang f.readahead = true; 15340615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 153547e4937aSGao Xiang 153638629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 153738629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 153838629291SGao Xiang nr_pages = readahead_count(rac); 153938629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 154038629291SGao Xiang 15410615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 154247e4937aSGao Xiang set_page_private(page, (unsigned long)head); 154347e4937aSGao Xiang head = page; 154447e4937aSGao Xiang } 154547e4937aSGao Xiang 154647e4937aSGao Xiang while (head) { 154747e4937aSGao Xiang struct page *page = head; 154847e4937aSGao Xiang int err; 154947e4937aSGao Xiang 155047e4937aSGao Xiang /* traversal in reverse order */ 155147e4937aSGao Xiang head = (void *)page_private(page); 155247e4937aSGao Xiang 15531825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1554a5876e24SGao Xiang if (err) 15554f761fa2SGao Xiang erofs_err(inode->i_sb, 15564f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 15574f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 155847e4937aSGao Xiang put_page(page); 155947e4937aSGao Xiang } 156038629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 1561*5c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 156247e4937aSGao Xiang 156338629291SGao Xiang z_erofs_runqueue(inode->i_sb, &f, &pagepool, 156440452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 156509c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1566eaa9172aSGao Xiang erofs_release_pages(&pagepool); 156747e4937aSGao Xiang } 156847e4937aSGao Xiang 15690c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 15700c638f70SGao Xiang .readpage = z_erofs_readpage, 15710615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 157247e4937aSGao Xiang }; 1573