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 19547e4937aSGao Xiang struct z_erofs_collector { 19647e4937aSGao Xiang struct z_erofs_pagevec_ctor vector; 19747e4937aSGao Xiang 19847e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 19947e4937aSGao Xiang struct z_erofs_collection *cl; 20081382f5fSGao Xiang /* a pointer used to pick up inplace I/O pages */ 20181382f5fSGao Xiang struct page **icpage_ptr; 20247e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 20347e4937aSGao Xiang 20447e4937aSGao Xiang enum z_erofs_collectmode mode; 20547e4937aSGao Xiang }; 20647e4937aSGao Xiang 20747e4937aSGao Xiang struct z_erofs_decompress_frontend { 20847e4937aSGao Xiang struct inode *const inode; 20947e4937aSGao Xiang 21047e4937aSGao Xiang struct z_erofs_collector clt; 21147e4937aSGao Xiang struct erofs_map_blocks map; 21247e4937aSGao Xiang 2136ea5aad3SGao Xiang bool readahead; 21447e4937aSGao Xiang /* used for applying cache strategy on the fly */ 21547e4937aSGao Xiang bool backmost; 21647e4937aSGao Xiang erofs_off_t headoffset; 21747e4937aSGao Xiang }; 21847e4937aSGao Xiang 21947e4937aSGao Xiang #define COLLECTOR_INIT() { \ 22047e4937aSGao Xiang .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 22147e4937aSGao Xiang .mode = COLLECT_PRIMARY_FOLLOWED } 22247e4937aSGao Xiang 22347e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 22447e4937aSGao Xiang .inode = __i, .clt = COLLECTOR_INIT(), \ 22547e4937aSGao Xiang .backmost = true, } 22647e4937aSGao Xiang 22747e4937aSGao Xiang static struct page *z_pagemap_global[Z_EROFS_VMAP_GLOBAL_PAGES]; 22847e4937aSGao Xiang static DEFINE_MUTEX(z_pagemap_global_lock); 22947e4937aSGao Xiang 23047e4937aSGao Xiang static void preload_compressed_pages(struct z_erofs_collector *clt, 23147e4937aSGao Xiang struct address_space *mc, 2321825c8d7SGao Xiang enum z_erofs_cache_alloctype type, 233eaa9172aSGao Xiang struct page **pagepool) 23447e4937aSGao Xiang { 23581382f5fSGao Xiang struct z_erofs_pcluster *pcl = clt->pcl; 23647e4937aSGao Xiang bool standalone = true; 2371825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 2381825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 23981382f5fSGao Xiang struct page **pages; 24081382f5fSGao Xiang pgoff_t index; 24147e4937aSGao Xiang 24247e4937aSGao Xiang if (clt->mode < COLLECT_PRIMARY_FOLLOWED) 24347e4937aSGao Xiang return; 24447e4937aSGao Xiang 24581382f5fSGao Xiang pages = pcl->compressed_pages; 24681382f5fSGao Xiang index = pcl->obj.index; 24781382f5fSGao Xiang for (; index < pcl->obj.index + pcl->pclusterpages; ++index, ++pages) { 24847e4937aSGao Xiang struct page *page; 24947e4937aSGao Xiang compressed_page_t t; 2501825c8d7SGao Xiang struct page *newpage = NULL; 25147e4937aSGao Xiang 25247e4937aSGao Xiang /* the compressed page was loaded before */ 25347e4937aSGao Xiang if (READ_ONCE(*pages)) 25447e4937aSGao Xiang continue; 25547e4937aSGao Xiang 25647e4937aSGao Xiang page = find_get_page(mc, index); 25747e4937aSGao Xiang 25847e4937aSGao Xiang if (page) { 25947e4937aSGao Xiang t = tag_compressed_page_justfound(page); 2600b964600SGao Xiang } else { 2610b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 2620b964600SGao Xiang standalone = false; 2630b964600SGao Xiang switch (type) { 2640b964600SGao Xiang case TRYALLOC: 2651825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 2661825c8d7SGao Xiang if (!newpage) 26747e4937aSGao Xiang continue; 2680b964600SGao Xiang set_page_private(newpage, 2690b964600SGao Xiang Z_EROFS_PREALLOCATED_PAGE); 2700b964600SGao Xiang t = tag_compressed_page_justfound(newpage); 2710b964600SGao Xiang break; 2720b964600SGao Xiang default: /* DONTALLOC */ 2730b964600SGao Xiang continue; 2740b964600SGao Xiang } 27547e4937aSGao Xiang } 27647e4937aSGao Xiang 27747e4937aSGao Xiang if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t))) 27847e4937aSGao Xiang continue; 27947e4937aSGao Xiang 280eaa9172aSGao Xiang if (page) 28147e4937aSGao Xiang put_page(page); 282eaa9172aSGao Xiang else if (newpage) 283eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 28447e4937aSGao Xiang } 28547e4937aSGao Xiang 2860b964600SGao Xiang /* 2870b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 2880b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 2890b964600SGao Xiang */ 2900b964600SGao Xiang if (standalone) 29147e4937aSGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 29247e4937aSGao Xiang } 29347e4937aSGao Xiang 29447e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 29547e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 29647e4937aSGao Xiang struct erofs_workgroup *grp) 29747e4937aSGao Xiang { 29847e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 29947e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 30047e4937aSGao Xiang int i; 30147e4937aSGao Xiang 302cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 30347e4937aSGao Xiang /* 30447e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 30547e4937aSGao Xiang * therefore no need to worry about available decompression users. 30647e4937aSGao Xiang */ 3079f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 30847e4937aSGao Xiang struct page *page = pcl->compressed_pages[i]; 30947e4937aSGao Xiang 31047e4937aSGao Xiang if (!page) 31147e4937aSGao Xiang continue; 31247e4937aSGao Xiang 31347e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 31447e4937aSGao Xiang if (!trylock_page(page)) 31547e4937aSGao Xiang return -EBUSY; 31647e4937aSGao Xiang 317f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 31847e4937aSGao Xiang continue; 31947e4937aSGao Xiang 32047e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 32147e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 3226aaa7b06SGao Xiang detach_page_private(page); 32347e4937aSGao Xiang unlock_page(page); 32447e4937aSGao Xiang } 32547e4937aSGao Xiang return 0; 32647e4937aSGao Xiang } 32747e4937aSGao Xiang 328d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 32947e4937aSGao Xiang { 33047e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 33147e4937aSGao Xiang int ret = 0; /* 0 - busy */ 33247e4937aSGao Xiang 33347e4937aSGao Xiang if (erofs_workgroup_try_to_freeze(&pcl->obj, 1)) { 33447e4937aSGao Xiang unsigned int i; 33547e4937aSGao Xiang 336cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 3379f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 33847e4937aSGao Xiang if (pcl->compressed_pages[i] == page) { 33947e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 34047e4937aSGao Xiang ret = 1; 34147e4937aSGao Xiang break; 34247e4937aSGao Xiang } 34347e4937aSGao Xiang } 34447e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 34547e4937aSGao Xiang 3466aaa7b06SGao Xiang if (ret) 3476aaa7b06SGao Xiang detach_page_private(page); 34847e4937aSGao Xiang } 34947e4937aSGao Xiang return ret; 35047e4937aSGao Xiang } 35147e4937aSGao Xiang 35247e4937aSGao Xiang /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 35381382f5fSGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_collector *clt, 35447e4937aSGao Xiang struct page *page) 35547e4937aSGao Xiang { 35647e4937aSGao Xiang struct z_erofs_pcluster *const pcl = clt->pcl; 35747e4937aSGao Xiang 35881382f5fSGao Xiang while (clt->icpage_ptr > pcl->compressed_pages) 35981382f5fSGao Xiang if (!cmpxchg(--clt->icpage_ptr, NULL, page)) 36047e4937aSGao Xiang return true; 36147e4937aSGao Xiang return false; 36247e4937aSGao Xiang } 36347e4937aSGao Xiang 36447e4937aSGao Xiang /* callers must be with collection lock held */ 36547e4937aSGao Xiang static int z_erofs_attach_page(struct z_erofs_collector *clt, 36686432a6dSGao Xiang struct page *page, enum z_erofs_page_type type, 36786432a6dSGao Xiang bool pvec_safereuse) 36847e4937aSGao Xiang { 36947e4937aSGao Xiang int ret; 37047e4937aSGao Xiang 37147e4937aSGao Xiang /* give priority for inplaceio */ 37247e4937aSGao Xiang if (clt->mode >= COLLECT_PRIMARY && 37347e4937aSGao Xiang type == Z_EROFS_PAGE_TYPE_EXCLUSIVE && 37499634bf3SGao Xiang z_erofs_try_inplace_io(clt, page)) 37547e4937aSGao Xiang return 0; 37647e4937aSGao Xiang 37786432a6dSGao Xiang ret = z_erofs_pagevec_enqueue(&clt->vector, page, type, 37886432a6dSGao Xiang pvec_safereuse); 37947e4937aSGao Xiang clt->cl->vcnt += (unsigned int)ret; 38047e4937aSGao Xiang return ret ? 0 : -EAGAIN; 38147e4937aSGao Xiang } 38247e4937aSGao Xiang 383473e15b0SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_collector *clt) 38447e4937aSGao Xiang { 385473e15b0SGao Xiang struct z_erofs_pcluster *pcl = clt->pcl; 386473e15b0SGao Xiang z_erofs_next_pcluster_t *owned_head = &clt->owned_head; 38747e4937aSGao Xiang 388473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 389473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 390473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 39147e4937aSGao Xiang *owned_head = &pcl->next; 392473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 393473e15b0SGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED; 394473e15b0SGao Xiang return; 395473e15b0SGao Xiang } 396473e15b0SGao Xiang 39747e4937aSGao Xiang /* 398473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 399473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 40047e4937aSGao Xiang */ 40147e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 402473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 40347e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 404473e15b0SGao Xiang clt->mode = COLLECT_PRIMARY_HOOKED; 405473e15b0SGao Xiang clt->tailpcl = NULL; 406473e15b0SGao Xiang return; 40747e4937aSGao Xiang } 408473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 409473e15b0SGao Xiang clt->mode = COLLECT_PRIMARY; 41047e4937aSGao Xiang } 41147e4937aSGao Xiang 4129e579fc1SGao Xiang static int z_erofs_lookup_collection(struct z_erofs_collector *clt, 41347e4937aSGao Xiang struct inode *inode, 41447e4937aSGao Xiang struct erofs_map_blocks *map) 41547e4937aSGao Xiang { 41664094a04SGao Xiang struct z_erofs_pcluster *pcl = clt->pcl; 41747e4937aSGao Xiang struct z_erofs_collection *cl; 41847e4937aSGao Xiang unsigned int length; 41947e4937aSGao Xiang 42064094a04SGao Xiang /* to avoid unexpected loop formed by corrupted images */ 42147e4937aSGao Xiang if (clt->owned_head == &pcl->next || pcl == clt->tailpcl) { 42247e4937aSGao Xiang DBG_BUGON(1); 4239e579fc1SGao Xiang return -EFSCORRUPTED; 42447e4937aSGao Xiang } 42547e4937aSGao Xiang 42647e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 4278d8a09b0SGao Xiang if (cl->pageofs != (map->m_la & ~PAGE_MASK)) { 42847e4937aSGao Xiang DBG_BUGON(1); 4299e579fc1SGao Xiang return -EFSCORRUPTED; 43047e4937aSGao Xiang } 43147e4937aSGao Xiang 43247e4937aSGao Xiang length = READ_ONCE(pcl->length); 43347e4937aSGao Xiang if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 43447e4937aSGao Xiang if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 43547e4937aSGao Xiang DBG_BUGON(1); 4369e579fc1SGao Xiang return -EFSCORRUPTED; 43747e4937aSGao Xiang } 43847e4937aSGao Xiang } else { 43947e4937aSGao Xiang unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 44047e4937aSGao Xiang 44147e4937aSGao Xiang if (map->m_flags & EROFS_MAP_FULL_MAPPED) 44247e4937aSGao Xiang llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 44347e4937aSGao Xiang 44447e4937aSGao Xiang while (llen > length && 44547e4937aSGao Xiang length != cmpxchg_relaxed(&pcl->length, length, llen)) { 44647e4937aSGao Xiang cpu_relax(); 44747e4937aSGao Xiang length = READ_ONCE(pcl->length); 44847e4937aSGao Xiang } 44947e4937aSGao Xiang } 45047e4937aSGao Xiang mutex_lock(&cl->lock); 45147e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 45247e4937aSGao Xiang if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 45347e4937aSGao Xiang clt->tailpcl = pcl; 454473e15b0SGao Xiang 455473e15b0SGao Xiang z_erofs_try_to_claim_pcluster(clt); 45647e4937aSGao Xiang clt->cl = cl; 4579e579fc1SGao Xiang return 0; 45847e4937aSGao Xiang } 45947e4937aSGao Xiang 4609e579fc1SGao Xiang static int z_erofs_register_collection(struct z_erofs_collector *clt, 46147e4937aSGao Xiang struct inode *inode, 46247e4937aSGao Xiang struct erofs_map_blocks *map) 46347e4937aSGao Xiang { 464cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 46547e4937aSGao Xiang struct z_erofs_pcluster *pcl; 46647e4937aSGao Xiang struct z_erofs_collection *cl; 46764094a04SGao Xiang struct erofs_workgroup *grp; 46847e4937aSGao Xiang int err; 46947e4937aSGao Xiang 4708f899262SGao Xiang if (!(map->m_flags & EROFS_MAP_ENCODED)) { 4718f899262SGao Xiang DBG_BUGON(1); 4728f899262SGao Xiang return -EFSCORRUPTED; 4738f899262SGao Xiang } 4748f899262SGao Xiang 4759f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 476cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 477cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 4789f6cc76eSGao Xiang if (IS_ERR(pcl)) 4799f6cc76eSGao Xiang return PTR_ERR(pcl); 48047e4937aSGao Xiang 48164094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 4828f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 48347e4937aSGao Xiang pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 48447e4937aSGao Xiang (map->m_flags & EROFS_MAP_FULL_MAPPED ? 48547e4937aSGao Xiang Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 48647e4937aSGao Xiang 48747e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 48847e4937aSGao Xiang pcl->next = clt->owned_head; 48947e4937aSGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED; 49047e4937aSGao Xiang 49147e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 49247e4937aSGao Xiang cl->pageofs = map->m_la & ~PAGE_MASK; 49347e4937aSGao Xiang 49447e4937aSGao Xiang /* 49547e4937aSGao Xiang * lock all primary followed works before visible to others 49647e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 49747e4937aSGao Xiang */ 4989f6cc76eSGao Xiang mutex_init(&cl->lock); 49964094a04SGao Xiang DBG_BUGON(!mutex_trylock(&cl->lock)); 50047e4937aSGao Xiang 501cecf864dSYue Hu if (ztailpacking) { 502cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 503cecf864dSYue Hu pcl->pageofs_in = erofs_blkoff(map->m_pa); 504cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 505cecf864dSYue Hu } else { 506cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 507cecf864dSYue Hu 50864094a04SGao Xiang grp = erofs_insert_workgroup(inode->i_sb, &pcl->obj); 50964094a04SGao Xiang if (IS_ERR(grp)) { 51064094a04SGao Xiang err = PTR_ERR(grp); 51164094a04SGao Xiang goto err_out; 51264094a04SGao Xiang } 51364094a04SGao Xiang 51464094a04SGao Xiang if (grp != &pcl->obj) { 515cecf864dSYue Hu clt->pcl = container_of(grp, 516cecf864dSYue Hu struct z_erofs_pcluster, obj); 51764094a04SGao Xiang err = -EEXIST; 51864094a04SGao Xiang goto err_out; 51947e4937aSGao Xiang } 520cecf864dSYue Hu } 52147e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 52247e4937aSGao Xiang if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 52347e4937aSGao Xiang clt->tailpcl = pcl; 52447e4937aSGao Xiang clt->owned_head = &pcl->next; 52547e4937aSGao Xiang clt->pcl = pcl; 52647e4937aSGao Xiang clt->cl = cl; 5279e579fc1SGao Xiang return 0; 52864094a04SGao Xiang 52964094a04SGao Xiang err_out: 53064094a04SGao Xiang mutex_unlock(&cl->lock); 5319f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 53264094a04SGao Xiang return err; 53347e4937aSGao Xiang } 53447e4937aSGao Xiang 53547e4937aSGao Xiang static int z_erofs_collector_begin(struct z_erofs_collector *clt, 53647e4937aSGao Xiang struct inode *inode, 53747e4937aSGao Xiang struct erofs_map_blocks *map) 53847e4937aSGao Xiang { 53964094a04SGao Xiang struct erofs_workgroup *grp; 5409e579fc1SGao Xiang int ret; 54147e4937aSGao Xiang 54247e4937aSGao Xiang DBG_BUGON(clt->cl); 54347e4937aSGao Xiang 54447e4937aSGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous collection */ 54547e4937aSGao Xiang DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_NIL); 54647e4937aSGao Xiang DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 54747e4937aSGao Xiang 548cecf864dSYue Hu if (map->m_flags & EROFS_MAP_META) { 549cecf864dSYue Hu if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 55047e4937aSGao Xiang DBG_BUGON(1); 551cecf864dSYue Hu return -EFSCORRUPTED; 552cecf864dSYue Hu } 553cecf864dSYue Hu goto tailpacking; 55447e4937aSGao Xiang } 55547e4937aSGao Xiang 55664094a04SGao Xiang grp = erofs_find_workgroup(inode->i_sb, map->m_pa >> PAGE_SHIFT); 55764094a04SGao Xiang if (grp) { 55864094a04SGao Xiang clt->pcl = container_of(grp, struct z_erofs_pcluster, obj); 55964094a04SGao Xiang } else { 560cecf864dSYue Hu tailpacking: 5619e579fc1SGao Xiang ret = z_erofs_register_collection(clt, inode, map); 56264094a04SGao Xiang if (!ret) 56364094a04SGao Xiang goto out; 56464094a04SGao Xiang if (ret != -EEXIST) 5659e579fc1SGao Xiang return ret; 56664094a04SGao Xiang } 56747e4937aSGao Xiang 56864094a04SGao Xiang ret = z_erofs_lookup_collection(clt, inode, map); 56964094a04SGao Xiang if (ret) { 57064094a04SGao Xiang erofs_workgroup_put(&clt->pcl->obj); 57164094a04SGao Xiang return ret; 57264094a04SGao Xiang } 57364094a04SGao Xiang 57464094a04SGao Xiang out: 57547e4937aSGao Xiang z_erofs_pagevec_ctor_init(&clt->vector, Z_EROFS_NR_INLINE_PAGEVECS, 5769e579fc1SGao Xiang clt->cl->pagevec, clt->cl->vcnt); 57781382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 578cecf864dSYue Hu clt->icpage_ptr = clt->pcl->compressed_pages + 579cecf864dSYue Hu z_erofs_pclusterpages(clt->pcl); 58047e4937aSGao Xiang return 0; 58147e4937aSGao Xiang } 58247e4937aSGao Xiang 58347e4937aSGao Xiang /* 58447e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 58547e4937aSGao Xiang * only after a RCU grace period. 58647e4937aSGao Xiang */ 58747e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 58847e4937aSGao Xiang { 58947e4937aSGao Xiang struct z_erofs_collection *const cl = 59047e4937aSGao Xiang container_of(head, struct z_erofs_collection, rcu); 59147e4937aSGao Xiang 5929f6cc76eSGao Xiang z_erofs_free_pcluster(container_of(cl, struct z_erofs_pcluster, 59347e4937aSGao Xiang primary_collection)); 59447e4937aSGao Xiang } 59547e4937aSGao Xiang 59647e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 59747e4937aSGao Xiang { 59847e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 59947e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 60047e4937aSGao Xiang struct z_erofs_collection *const cl = z_erofs_primarycollection(pcl); 60147e4937aSGao Xiang 60247e4937aSGao Xiang call_rcu(&cl->rcu, z_erofs_rcu_callback); 60347e4937aSGao Xiang } 60447e4937aSGao Xiang 60547e4937aSGao Xiang static void z_erofs_collection_put(struct z_erofs_collection *cl) 60647e4937aSGao Xiang { 60747e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 60847e4937aSGao Xiang container_of(cl, struct z_erofs_pcluster, primary_collection); 60947e4937aSGao Xiang 61047e4937aSGao Xiang erofs_workgroup_put(&pcl->obj); 61147e4937aSGao Xiang } 61247e4937aSGao Xiang 61347e4937aSGao Xiang static bool z_erofs_collector_end(struct z_erofs_collector *clt) 61447e4937aSGao Xiang { 61547e4937aSGao Xiang struct z_erofs_collection *cl = clt->cl; 61647e4937aSGao Xiang 61747e4937aSGao Xiang if (!cl) 61847e4937aSGao Xiang return false; 61947e4937aSGao Xiang 62047e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&clt->vector, false); 62147e4937aSGao Xiang mutex_unlock(&cl->lock); 62247e4937aSGao Xiang 62347e4937aSGao Xiang /* 62447e4937aSGao Xiang * if all pending pages are added, don't hold its reference 62547e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 62647e4937aSGao Xiang */ 62747e4937aSGao Xiang if (clt->mode < COLLECT_PRIMARY_FOLLOWED_NOINPLACE) 62847e4937aSGao Xiang z_erofs_collection_put(cl); 62947e4937aSGao Xiang 63047e4937aSGao Xiang clt->cl = NULL; 63147e4937aSGao Xiang return true; 63247e4937aSGao Xiang } 63347e4937aSGao Xiang 63447e4937aSGao Xiang static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe, 63547e4937aSGao Xiang unsigned int cachestrategy, 63647e4937aSGao Xiang erofs_off_t la) 63747e4937aSGao Xiang { 63847e4937aSGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 63947e4937aSGao Xiang return false; 64047e4937aSGao Xiang 64147e4937aSGao Xiang if (fe->backmost) 64247e4937aSGao Xiang return true; 64347e4937aSGao Xiang 64447e4937aSGao Xiang return cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 64547e4937aSGao Xiang la < fe->headoffset; 64647e4937aSGao Xiang } 64747e4937aSGao Xiang 64847e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 649eaa9172aSGao Xiang struct page *page, struct page **pagepool) 65047e4937aSGao Xiang { 65147e4937aSGao Xiang struct inode *const inode = fe->inode; 652bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 65347e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 65447e4937aSGao Xiang struct z_erofs_collector *const clt = &fe->clt; 65547e4937aSGao Xiang const loff_t offset = page_offset(page); 656dc76ea8cSGao Xiang bool tight = true; 65747e4937aSGao Xiang 65847e4937aSGao Xiang enum z_erofs_cache_alloctype cache_strategy; 65947e4937aSGao Xiang enum z_erofs_page_type page_type; 66047e4937aSGao Xiang unsigned int cur, end, spiltted, index; 66147e4937aSGao Xiang int err = 0; 66247e4937aSGao Xiang 66347e4937aSGao Xiang /* register locked file pages as online pages in pack */ 66447e4937aSGao Xiang z_erofs_onlinepage_init(page); 66547e4937aSGao Xiang 66647e4937aSGao Xiang spiltted = 0; 66747e4937aSGao Xiang end = PAGE_SIZE; 66847e4937aSGao Xiang repeat: 66947e4937aSGao Xiang cur = end - 1; 67047e4937aSGao Xiang 67147e4937aSGao Xiang /* lucky, within the range of the current map_blocks */ 67247e4937aSGao Xiang if (offset + cur >= map->m_la && 67347e4937aSGao Xiang offset + cur < map->m_la + map->m_llen) { 67447e4937aSGao Xiang /* didn't get a valid collection previously (very rare) */ 67547e4937aSGao Xiang if (!clt->cl) 67647e4937aSGao Xiang goto restart_now; 67747e4937aSGao Xiang goto hitted; 67847e4937aSGao Xiang } 67947e4937aSGao Xiang 68047e4937aSGao Xiang /* go ahead the next map_blocks */ 6814f761fa2SGao Xiang erofs_dbg("%s: [out-of-range] pos %llu", __func__, offset + cur); 68247e4937aSGao Xiang 68347e4937aSGao Xiang if (z_erofs_collector_end(clt)) 68447e4937aSGao Xiang fe->backmost = false; 68547e4937aSGao Xiang 68647e4937aSGao Xiang map->m_la = offset + cur; 68747e4937aSGao Xiang map->m_llen = 0; 68847e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 6898d8a09b0SGao Xiang if (err) 69047e4937aSGao Xiang goto err_out; 69147e4937aSGao Xiang 69247e4937aSGao Xiang restart_now: 6938d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) 69447e4937aSGao Xiang goto hitted; 69547e4937aSGao Xiang 69647e4937aSGao Xiang err = z_erofs_collector_begin(clt, inode, map); 6978d8a09b0SGao Xiang if (err) 69847e4937aSGao Xiang goto err_out; 69947e4937aSGao Xiang 700cecf864dSYue Hu if (z_erofs_is_inline_pcluster(clt->pcl)) { 701*09c54379SGao Xiang void *mp; 702cecf864dSYue Hu 703*09c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 704*09c54379SGao Xiang erofs_blknr(map->m_pa), EROFS_NO_KMAP); 705*09c54379SGao Xiang if (IS_ERR(mp)) { 706*09c54379SGao Xiang err = PTR_ERR(mp); 707cecf864dSYue Hu erofs_err(inode->i_sb, 708cecf864dSYue Hu "failed to get inline page, err %d", err); 709cecf864dSYue Hu goto err_out; 710cecf864dSYue Hu } 711*09c54379SGao Xiang get_page(fe->map.buf.page); 712*09c54379SGao Xiang WRITE_ONCE(clt->pcl->compressed_pages[0], fe->map.buf.page); 713cecf864dSYue Hu clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 714cecf864dSYue Hu } else { 715cecf864dSYue Hu /* preload all compressed pages (can change mode if needed) */ 716cecf864dSYue Hu if (should_alloc_managed_pages(fe, sbi->opt.cache_strategy, 717cecf864dSYue Hu map->m_la)) 7181825c8d7SGao Xiang cache_strategy = TRYALLOC; 71947e4937aSGao Xiang else 72047e4937aSGao Xiang cache_strategy = DONTALLOC; 72147e4937aSGao Xiang 7221825c8d7SGao Xiang preload_compressed_pages(clt, MNGD_MAPPING(sbi), 7231825c8d7SGao Xiang cache_strategy, pagepool); 724cecf864dSYue Hu } 72547e4937aSGao Xiang 72647e4937aSGao Xiang hitted: 727dc76ea8cSGao Xiang /* 728dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 729dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 730dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 731dc76ea8cSGao Xiang * for inplace I/O or pagevec (should be processed in strict order.) 732dc76ea8cSGao Xiang */ 733dc76ea8cSGao Xiang tight &= (clt->mode >= COLLECT_PRIMARY_HOOKED && 734dc76ea8cSGao Xiang clt->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 735dc76ea8cSGao Xiang 73647e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 7378d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 73847e4937aSGao Xiang zero_user_segment(page, cur, end); 73947e4937aSGao Xiang goto next_part; 74047e4937aSGao Xiang } 74147e4937aSGao Xiang 74247e4937aSGao Xiang /* let's derive page type */ 74347e4937aSGao Xiang page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 74447e4937aSGao Xiang (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 74547e4937aSGao Xiang (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 74647e4937aSGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 74747e4937aSGao Xiang 74847e4937aSGao Xiang if (cur) 74947e4937aSGao Xiang tight &= (clt->mode >= COLLECT_PRIMARY_FOLLOWED); 75047e4937aSGao Xiang 75147e4937aSGao Xiang retry: 75286432a6dSGao Xiang err = z_erofs_attach_page(clt, page, page_type, 75386432a6dSGao Xiang clt->mode >= COLLECT_PRIMARY_FOLLOWED); 7546aaa7b06SGao Xiang /* should allocate an additional short-lived page for pagevec */ 75547e4937aSGao Xiang if (err == -EAGAIN) { 75647e4937aSGao Xiang struct page *const newpage = 757e3f78d5eSChao Yu alloc_page(GFP_NOFS | __GFP_NOFAIL); 75847e4937aSGao Xiang 7596aaa7b06SGao Xiang set_page_private(newpage, Z_EROFS_SHORTLIVED_PAGE); 76047e4937aSGao Xiang err = z_erofs_attach_page(clt, newpage, 76186432a6dSGao Xiang Z_EROFS_PAGE_TYPE_EXCLUSIVE, true); 7628d8a09b0SGao Xiang if (!err) 76347e4937aSGao Xiang goto retry; 76447e4937aSGao Xiang } 76547e4937aSGao Xiang 7668d8a09b0SGao Xiang if (err) 76747e4937aSGao Xiang goto err_out; 76847e4937aSGao Xiang 76947e4937aSGao Xiang index = page->index - (map->m_la >> PAGE_SHIFT); 77047e4937aSGao Xiang 77147e4937aSGao Xiang z_erofs_onlinepage_fixup(page, index, true); 77247e4937aSGao Xiang 77347e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 77447e4937aSGao Xiang ++spiltted; 77547e4937aSGao Xiang /* also update nr_pages */ 77647e4937aSGao Xiang clt->cl->nr_pages = max_t(pgoff_t, clt->cl->nr_pages, index + 1); 77747e4937aSGao Xiang next_part: 77847e4937aSGao Xiang /* can be used for verification */ 77947e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 78047e4937aSGao Xiang 78147e4937aSGao Xiang end = cur; 78247e4937aSGao Xiang if (end > 0) 78347e4937aSGao Xiang goto repeat; 78447e4937aSGao Xiang 78547e4937aSGao Xiang out: 78647e4937aSGao Xiang z_erofs_onlinepage_endio(page); 78747e4937aSGao Xiang 7884f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 78947e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 79047e4937aSGao Xiang return err; 79147e4937aSGao Xiang 79247e4937aSGao Xiang /* if some error occurred while processing this page */ 79347e4937aSGao Xiang err_out: 79447e4937aSGao Xiang SetPageError(page); 79547e4937aSGao Xiang goto out; 79647e4937aSGao Xiang } 79747e4937aSGao Xiang 79840452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 79940452ffcSHuang Jianan unsigned int readahead_pages) 80040452ffcSHuang Jianan { 80140452ffcSHuang Jianan /* auto: enable for readpage, disable for readahead */ 80240452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 80340452ffcSHuang Jianan !readahead_pages) 80440452ffcSHuang Jianan return true; 80540452ffcSHuang Jianan 80640452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 80740452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 80840452ffcSHuang Jianan return true; 80940452ffcSHuang Jianan 81040452ffcSHuang Jianan return false; 81140452ffcSHuang Jianan } 81240452ffcSHuang Jianan 813648f2de0SHuang Jianan static void z_erofs_decompressqueue_work(struct work_struct *work); 814a4b1fab1SGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 815a4b1fab1SGao Xiang bool sync, int bios) 81647e4937aSGao Xiang { 81730048cdaSHuang Jianan struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 81830048cdaSHuang Jianan 819a4b1fab1SGao Xiang /* wake up the caller thread for sync decompression */ 820a4b1fab1SGao Xiang if (sync) { 82147e4937aSGao Xiang unsigned long flags; 82247e4937aSGao Xiang 82347e4937aSGao Xiang spin_lock_irqsave(&io->u.wait.lock, flags); 82447e4937aSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 82547e4937aSGao Xiang wake_up_locked(&io->u.wait); 82647e4937aSGao Xiang spin_unlock_irqrestore(&io->u.wait.lock, flags); 82747e4937aSGao Xiang return; 82847e4937aSGao Xiang } 82947e4937aSGao Xiang 830648f2de0SHuang Jianan if (atomic_add_return(bios, &io->pending_bios)) 831648f2de0SHuang Jianan return; 83230048cdaSHuang Jianan /* Use workqueue and sync decompression for atomic contexts only */ 833648f2de0SHuang Jianan if (in_atomic() || irqs_disabled()) { 83447e4937aSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 83540452ffcSHuang Jianan /* enable sync decompression for readahead */ 83640452ffcSHuang Jianan if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 83740452ffcSHuang Jianan sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 838648f2de0SHuang Jianan return; 839648f2de0SHuang Jianan } 840648f2de0SHuang Jianan z_erofs_decompressqueue_work(&io->u.work); 84147e4937aSGao Xiang } 84247e4937aSGao Xiang 8436aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 8446aaa7b06SGao Xiang { 8456aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 8466aaa7b06SGao Xiang } 8476aaa7b06SGao Xiang 8480c638f70SGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 84947e4937aSGao Xiang { 850a4b1fab1SGao Xiang tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 851a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 85247e4937aSGao Xiang blk_status_t err = bio->bi_status; 85347e4937aSGao Xiang struct bio_vec *bvec; 85447e4937aSGao Xiang struct bvec_iter_all iter_all; 85547e4937aSGao Xiang 85647e4937aSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 85747e4937aSGao Xiang struct page *page = bvec->bv_page; 85847e4937aSGao Xiang 85947e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 8606aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 86147e4937aSGao Xiang 8628d8a09b0SGao Xiang if (err) 86347e4937aSGao Xiang SetPageError(page); 86447e4937aSGao Xiang 865a4b1fab1SGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 866a4b1fab1SGao Xiang if (!err) 867a4b1fab1SGao Xiang SetPageUptodate(page); 86847e4937aSGao Xiang unlock_page(page); 86947e4937aSGao Xiang } 870a4b1fab1SGao Xiang } 871a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 87247e4937aSGao Xiang bio_put(bio); 87347e4937aSGao Xiang } 87447e4937aSGao Xiang 87547e4937aSGao Xiang static int z_erofs_decompress_pcluster(struct super_block *sb, 87647e4937aSGao Xiang struct z_erofs_pcluster *pcl, 877eaa9172aSGao Xiang struct page **pagepool) 87847e4937aSGao Xiang { 87947e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 880cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 88147e4937aSGao Xiang struct z_erofs_pagevec_ctor ctor; 8829f6cc76eSGao Xiang unsigned int i, inputsize, outputsize, llen, nr_pages; 88347e4937aSGao Xiang struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 88447e4937aSGao Xiang struct page **pages, **compressed_pages, *page; 88547e4937aSGao Xiang 88647e4937aSGao Xiang enum z_erofs_page_type page_type; 88747e4937aSGao Xiang bool overlapped, partial; 88847e4937aSGao Xiang struct z_erofs_collection *cl; 88947e4937aSGao Xiang int err; 89047e4937aSGao Xiang 89147e4937aSGao Xiang might_sleep(); 89247e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 89347e4937aSGao Xiang DBG_BUGON(!READ_ONCE(cl->nr_pages)); 89447e4937aSGao Xiang 89547e4937aSGao Xiang mutex_lock(&cl->lock); 89647e4937aSGao Xiang nr_pages = cl->nr_pages; 89747e4937aSGao Xiang 8988d8a09b0SGao Xiang if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 89947e4937aSGao Xiang pages = pages_onstack; 90047e4937aSGao Xiang } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 90147e4937aSGao Xiang mutex_trylock(&z_pagemap_global_lock)) { 90247e4937aSGao Xiang pages = z_pagemap_global; 90347e4937aSGao Xiang } else { 90447e4937aSGao Xiang gfp_t gfp_flags = GFP_KERNEL; 90547e4937aSGao Xiang 90647e4937aSGao Xiang if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 90747e4937aSGao Xiang gfp_flags |= __GFP_NOFAIL; 90847e4937aSGao Xiang 90947e4937aSGao Xiang pages = kvmalloc_array(nr_pages, sizeof(struct page *), 91047e4937aSGao Xiang gfp_flags); 91147e4937aSGao Xiang 91247e4937aSGao Xiang /* fallback to global pagemap for the lowmem scenario */ 9138d8a09b0SGao Xiang if (!pages) { 91447e4937aSGao Xiang mutex_lock(&z_pagemap_global_lock); 91547e4937aSGao Xiang pages = z_pagemap_global; 91647e4937aSGao Xiang } 91747e4937aSGao Xiang } 91847e4937aSGao Xiang 91947e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) 92047e4937aSGao Xiang pages[i] = NULL; 92147e4937aSGao Xiang 92247e4937aSGao Xiang err = 0; 92347e4937aSGao Xiang z_erofs_pagevec_ctor_init(&ctor, Z_EROFS_NR_INLINE_PAGEVECS, 92447e4937aSGao Xiang cl->pagevec, 0); 92547e4937aSGao Xiang 92647e4937aSGao Xiang for (i = 0; i < cl->vcnt; ++i) { 92747e4937aSGao Xiang unsigned int pagenr; 92847e4937aSGao Xiang 92947e4937aSGao Xiang page = z_erofs_pagevec_dequeue(&ctor, &page_type); 93047e4937aSGao Xiang 93147e4937aSGao Xiang /* all pages in pagevec ought to be valid */ 93247e4937aSGao Xiang DBG_BUGON(!page); 9336aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 93447e4937aSGao Xiang 9356aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 93647e4937aSGao Xiang continue; 93747e4937aSGao Xiang 93847e4937aSGao Xiang if (page_type == Z_EROFS_VLE_PAGE_TYPE_HEAD) 93947e4937aSGao Xiang pagenr = 0; 94047e4937aSGao Xiang else 94147e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 94247e4937aSGao Xiang 94347e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 94447e4937aSGao Xiang 94547e4937aSGao Xiang /* 94647e4937aSGao Xiang * currently EROFS doesn't support multiref(dedup), 94747e4937aSGao Xiang * so here erroring out one multiref page. 94847e4937aSGao Xiang */ 9498d8a09b0SGao Xiang if (pages[pagenr]) { 95047e4937aSGao Xiang DBG_BUGON(1); 95147e4937aSGao Xiang SetPageError(pages[pagenr]); 95247e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 95347e4937aSGao Xiang err = -EFSCORRUPTED; 95447e4937aSGao Xiang } 95547e4937aSGao Xiang pages[pagenr] = page; 95647e4937aSGao Xiang } 95747e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&ctor, true); 95847e4937aSGao Xiang 95947e4937aSGao Xiang overlapped = false; 96047e4937aSGao Xiang compressed_pages = pcl->compressed_pages; 96147e4937aSGao Xiang 962cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 96347e4937aSGao Xiang unsigned int pagenr; 96447e4937aSGao Xiang 96547e4937aSGao Xiang page = compressed_pages[i]; 96647e4937aSGao Xiang /* all compressed pages ought to be valid */ 96747e4937aSGao Xiang DBG_BUGON(!page); 96847e4937aSGao Xiang 969cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 970cecf864dSYue Hu if (!PageUptodate(page)) 971cecf864dSYue Hu err = -EIO; 972cecf864dSYue Hu continue; 973cecf864dSYue Hu } 974cecf864dSYue Hu 975cecf864dSYue Hu DBG_BUGON(z_erofs_page_is_invalidated(page)); 9766aaa7b06SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 97747e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) { 9788d8a09b0SGao Xiang if (!PageUptodate(page)) 97947e4937aSGao Xiang err = -EIO; 98047e4937aSGao Xiang continue; 98147e4937aSGao Xiang } 98247e4937aSGao Xiang 98347e4937aSGao Xiang /* 98447e4937aSGao Xiang * only if non-head page can be selected 98547e4937aSGao Xiang * for inplace decompression 98647e4937aSGao Xiang */ 98747e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 98847e4937aSGao Xiang 98947e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 9908d8a09b0SGao Xiang if (pages[pagenr]) { 99147e4937aSGao Xiang DBG_BUGON(1); 99247e4937aSGao Xiang SetPageError(pages[pagenr]); 99347e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 99447e4937aSGao Xiang err = -EFSCORRUPTED; 99547e4937aSGao Xiang } 99647e4937aSGao Xiang pages[pagenr] = page; 99747e4937aSGao Xiang 99847e4937aSGao Xiang overlapped = true; 99947e4937aSGao Xiang } 100047e4937aSGao Xiang 10016aaa7b06SGao Xiang /* PG_error needs checking for all non-managed pages */ 10028d8a09b0SGao Xiang if (PageError(page)) { 100347e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 100447e4937aSGao Xiang err = -EIO; 100547e4937aSGao Xiang } 100647e4937aSGao Xiang } 100747e4937aSGao Xiang 10088d8a09b0SGao Xiang if (err) 100947e4937aSGao Xiang goto out; 101047e4937aSGao Xiang 101147e4937aSGao Xiang llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 101247e4937aSGao Xiang if (nr_pages << PAGE_SHIFT >= cl->pageofs + llen) { 101347e4937aSGao Xiang outputsize = llen; 101447e4937aSGao Xiang partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 101547e4937aSGao Xiang } else { 101647e4937aSGao Xiang outputsize = (nr_pages << PAGE_SHIFT) - cl->pageofs; 101747e4937aSGao Xiang partial = true; 101847e4937aSGao Xiang } 101947e4937aSGao Xiang 1020cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1021cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1022cecf864dSYue Hu else 1023cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1024cecf864dSYue Hu 102547e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 102647e4937aSGao Xiang .sb = sb, 102747e4937aSGao Xiang .in = compressed_pages, 102847e4937aSGao Xiang .out = pages, 1029cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 103047e4937aSGao Xiang .pageofs_out = cl->pageofs, 10319f6cc76eSGao Xiang .inputsize = inputsize, 103247e4937aSGao Xiang .outputsize = outputsize, 103347e4937aSGao Xiang .alg = pcl->algorithmformat, 103447e4937aSGao Xiang .inplace_io = overlapped, 103547e4937aSGao Xiang .partial_decoding = partial 103647e4937aSGao Xiang }, pagepool); 103747e4937aSGao Xiang 103847e4937aSGao Xiang out: 1039cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1040cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1041cecf864dSYue Hu page = compressed_pages[0]; 1042cecf864dSYue Hu WRITE_ONCE(compressed_pages[0], NULL); 1043cecf864dSYue Hu put_page(page); 1044cecf864dSYue Hu } else { 1045cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 104647e4937aSGao Xiang page = compressed_pages[i]; 104747e4937aSGao Xiang 104847e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 104947e4937aSGao Xiang continue; 105047e4937aSGao Xiang 10516aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10526aaa7b06SGao Xiang (void)z_erofs_put_shortlivedpage(pagepool, page); 105347e4937aSGao Xiang WRITE_ONCE(compressed_pages[i], NULL); 105447e4937aSGao Xiang } 1055cecf864dSYue Hu } 105647e4937aSGao Xiang 105747e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) { 105847e4937aSGao Xiang page = pages[i]; 105947e4937aSGao Xiang if (!page) 106047e4937aSGao Xiang continue; 106147e4937aSGao Xiang 10626aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 106347e4937aSGao Xiang 10646aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10656aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 106647e4937aSGao Xiang continue; 106747e4937aSGao Xiang 10688d8a09b0SGao Xiang if (err < 0) 106947e4937aSGao Xiang SetPageError(page); 107047e4937aSGao Xiang 107147e4937aSGao Xiang z_erofs_onlinepage_endio(page); 107247e4937aSGao Xiang } 107347e4937aSGao Xiang 107447e4937aSGao Xiang if (pages == z_pagemap_global) 107547e4937aSGao Xiang mutex_unlock(&z_pagemap_global_lock); 10768d8a09b0SGao Xiang else if (pages != pages_onstack) 107747e4937aSGao Xiang kvfree(pages); 107847e4937aSGao Xiang 107947e4937aSGao Xiang cl->nr_pages = 0; 108047e4937aSGao Xiang cl->vcnt = 0; 108147e4937aSGao Xiang 108247e4937aSGao Xiang /* all cl locks MUST be taken before the following line */ 108347e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 108447e4937aSGao Xiang 108547e4937aSGao Xiang /* all cl locks SHOULD be released right now */ 108647e4937aSGao Xiang mutex_unlock(&cl->lock); 108747e4937aSGao Xiang 108847e4937aSGao Xiang z_erofs_collection_put(cl); 108947e4937aSGao Xiang return err; 109047e4937aSGao Xiang } 109147e4937aSGao Xiang 10920c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1093eaa9172aSGao Xiang struct page **pagepool) 109447e4937aSGao Xiang { 109547e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 109647e4937aSGao Xiang 109747e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 109847e4937aSGao Xiang struct z_erofs_pcluster *pcl; 109947e4937aSGao Xiang 110047e4937aSGao Xiang /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 110147e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 110247e4937aSGao Xiang 110347e4937aSGao Xiang /* no possible that 'owned' equals NULL */ 110447e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 110547e4937aSGao Xiang 110647e4937aSGao Xiang pcl = container_of(owned, struct z_erofs_pcluster, next); 110747e4937aSGao Xiang owned = READ_ONCE(pcl->next); 110847e4937aSGao Xiang 1109a4b1fab1SGao Xiang z_erofs_decompress_pcluster(io->sb, pcl, pagepool); 111047e4937aSGao Xiang } 111147e4937aSGao Xiang } 111247e4937aSGao Xiang 11130c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 111447e4937aSGao Xiang { 1115a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1116a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1117eaa9172aSGao Xiang struct page *pagepool = NULL; 111847e4937aSGao Xiang 1119a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 11200c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 112147e4937aSGao Xiang 1122eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1123a4b1fab1SGao Xiang kvfree(bgq); 112447e4937aSGao Xiang } 112547e4937aSGao Xiang 112647e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 112747e4937aSGao Xiang unsigned int nr, 1128eaa9172aSGao Xiang struct page **pagepool, 112947e4937aSGao Xiang struct address_space *mc, 113047e4937aSGao Xiang gfp_t gfp) 113147e4937aSGao Xiang { 113247e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 113347e4937aSGao Xiang bool tocache = false; 113447e4937aSGao Xiang 113547e4937aSGao Xiang struct address_space *mapping; 113647e4937aSGao Xiang struct page *oldpage, *page; 113747e4937aSGao Xiang 113847e4937aSGao Xiang compressed_page_t t; 113947e4937aSGao Xiang int justfound; 114047e4937aSGao Xiang 114147e4937aSGao Xiang repeat: 114247e4937aSGao Xiang page = READ_ONCE(pcl->compressed_pages[nr]); 114347e4937aSGao Xiang oldpage = page; 114447e4937aSGao Xiang 114547e4937aSGao Xiang if (!page) 114647e4937aSGao Xiang goto out_allocpage; 114747e4937aSGao Xiang 114847e4937aSGao Xiang /* process the target tagged pointer */ 114947e4937aSGao Xiang t = tagptr_init(compressed_page_t, page); 115047e4937aSGao Xiang justfound = tagptr_unfold_tags(t); 115147e4937aSGao Xiang page = tagptr_unfold_ptr(t); 115247e4937aSGao Xiang 11531825c8d7SGao Xiang /* 11541825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 11551825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 11561825c8d7SGao Xiang */ 11571825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 11581825c8d7SGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 11591825c8d7SGao Xiang set_page_private(page, 0); 11601825c8d7SGao Xiang tocache = true; 11611825c8d7SGao Xiang goto out_tocache; 11621825c8d7SGao Xiang } 116347e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 116447e4937aSGao Xiang 116547e4937aSGao Xiang /* 11666aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 116747e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 116847e4937aSGao Xiang */ 116947e4937aSGao Xiang if (mapping && mapping != mc) 117047e4937aSGao Xiang /* ought to be unmanaged pages */ 117147e4937aSGao Xiang goto out; 117247e4937aSGao Xiang 11736aaa7b06SGao Xiang /* directly return for shortlived page as well */ 11746aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 11756aaa7b06SGao Xiang goto out; 11766aaa7b06SGao Xiang 117747e4937aSGao Xiang lock_page(page); 117847e4937aSGao Xiang 117947e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 118047e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 118147e4937aSGao Xiang 118247e4937aSGao Xiang /* the page is still in manage cache */ 118347e4937aSGao Xiang if (page->mapping == mc) { 118447e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 118547e4937aSGao Xiang 118647e4937aSGao Xiang ClearPageError(page); 118747e4937aSGao Xiang if (!PagePrivate(page)) { 118847e4937aSGao Xiang /* 118947e4937aSGao Xiang * impossible to be !PagePrivate(page) for 119047e4937aSGao Xiang * the current restriction as well if 119147e4937aSGao Xiang * the page is already in compressed_pages[]. 119247e4937aSGao Xiang */ 119347e4937aSGao Xiang DBG_BUGON(!justfound); 119447e4937aSGao Xiang 119547e4937aSGao Xiang justfound = 0; 119647e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 119747e4937aSGao Xiang SetPagePrivate(page); 119847e4937aSGao Xiang } 119947e4937aSGao Xiang 120047e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 120147e4937aSGao Xiang if (PageUptodate(page)) { 120247e4937aSGao Xiang unlock_page(page); 120347e4937aSGao Xiang page = NULL; 120447e4937aSGao Xiang } 120547e4937aSGao Xiang goto out; 120647e4937aSGao Xiang } 120747e4937aSGao Xiang 120847e4937aSGao Xiang /* 120947e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 121047e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 121147e4937aSGao Xiang */ 121247e4937aSGao Xiang DBG_BUGON(page->mapping); 121347e4937aSGao Xiang DBG_BUGON(!justfound); 121447e4937aSGao Xiang 121547e4937aSGao Xiang tocache = true; 121647e4937aSGao Xiang unlock_page(page); 121747e4937aSGao Xiang put_page(page); 121847e4937aSGao Xiang out_allocpage: 12195ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 12205ddcee1fSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_pages[nr], oldpage, page)) { 1221eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 12225ddcee1fSGao Xiang cond_resched(); 12235ddcee1fSGao Xiang goto repeat; 12245ddcee1fSGao Xiang } 12251825c8d7SGao Xiang out_tocache: 1226bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1227bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1228bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1229bf225074SGao Xiang goto out; 1230a30573b3SGao Xiang } 1231bf225074SGao Xiang attach_page_private(page, pcl); 1232bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1233bf225074SGao Xiang put_page(page); 1234bf225074SGao Xiang 123547e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 123647e4937aSGao Xiang return page; 123747e4937aSGao Xiang } 123847e4937aSGao Xiang 1239a4b1fab1SGao Xiang static struct z_erofs_decompressqueue * 1240a4b1fab1SGao Xiang jobqueue_init(struct super_block *sb, 1241a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 124247e4937aSGao Xiang { 1243a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 124447e4937aSGao Xiang 1245a4b1fab1SGao Xiang if (fg && !*fg) { 1246a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1247a4b1fab1SGao Xiang if (!q) { 1248a4b1fab1SGao Xiang *fg = true; 1249a4b1fab1SGao Xiang goto fg_out; 125047e4937aSGao Xiang } 12510c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1252a4b1fab1SGao Xiang } else { 1253a4b1fab1SGao Xiang fg_out: 1254a4b1fab1SGao Xiang q = fgq; 1255a4b1fab1SGao Xiang init_waitqueue_head(&fgq->u.wait); 1256a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 1257a4b1fab1SGao Xiang } 1258a4b1fab1SGao Xiang q->sb = sb; 1259a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1260a4b1fab1SGao Xiang return q; 126147e4937aSGao Xiang } 126247e4937aSGao Xiang 126347e4937aSGao Xiang /* define decompression jobqueue types */ 126447e4937aSGao Xiang enum { 126547e4937aSGao Xiang JQ_BYPASS, 126647e4937aSGao Xiang JQ_SUBMIT, 126747e4937aSGao Xiang NR_JOBQUEUES, 126847e4937aSGao Xiang }; 126947e4937aSGao Xiang 127047e4937aSGao Xiang static void *jobqueueset_init(struct super_block *sb, 1271a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[], 1272a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 127347e4937aSGao Xiang { 127447e4937aSGao Xiang /* 127547e4937aSGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 127647e4937aSGao Xiang * no need to read from device for all pclusters in this queue. 127747e4937aSGao Xiang */ 1278a4b1fab1SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1279a4b1fab1SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 128047e4937aSGao Xiang 1281a4b1fab1SGao Xiang return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 128247e4937aSGao Xiang } 128347e4937aSGao Xiang 128447e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 128547e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 128647e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 128747e4937aSGao Xiang { 128847e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 128947e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 129047e4937aSGao Xiang 129147e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 129247e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 129347e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 129447e4937aSGao Xiang 129547e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 129647e4937aSGao Xiang 129747e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 129847e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 129947e4937aSGao Xiang 130047e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 130147e4937aSGao Xiang } 130247e4937aSGao Xiang 13031e4a2955SGao Xiang static void z_erofs_submit_queue(struct super_block *sb, 13046ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1305eaa9172aSGao Xiang struct page **pagepool, 1306a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1307a4b1fab1SGao Xiang bool *force_fg) 130847e4937aSGao Xiang { 1309bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 131047e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1311a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 131247e4937aSGao Xiang void *bi_private; 13136ea5aad3SGao Xiang z_erofs_next_pcluster_t owned_head = f->clt.owned_head; 1314dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 13153f649ab7SKees Cook pgoff_t last_index; 1316dfeab2e9SGao Xiang struct block_device *last_bdev; 13171e4a2955SGao Xiang unsigned int nr_bios = 0; 13181e4a2955SGao Xiang struct bio *bio = NULL; 131947e4937aSGao Xiang 1320a4b1fab1SGao Xiang bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1321a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1322a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 132347e4937aSGao Xiang 132447e4937aSGao Xiang /* by default, all need io submission */ 132547e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 132647e4937aSGao Xiang 132747e4937aSGao Xiang do { 1328dfeab2e9SGao Xiang struct erofs_map_dev mdev; 132947e4937aSGao Xiang struct z_erofs_pcluster *pcl; 13301e4a2955SGao Xiang pgoff_t cur, end; 13311e4a2955SGao Xiang unsigned int i = 0; 13321e4a2955SGao Xiang bool bypass = true; 133347e4937aSGao Xiang 133447e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 133547e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 133647e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 133747e4937aSGao Xiang 133847e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 133947e4937aSGao Xiang 1340cecf864dSYue Hu /* close the main owned chain at first */ 1341cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1342cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1343cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1344cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1345cecf864dSYue Hu continue; 1346cecf864dSYue Hu } 1347cecf864dSYue Hu 1348dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1349dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1350dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1351dfeab2e9SGao Xiang }; 1352dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1353dfeab2e9SGao Xiang 1354dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 13559f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 135647e4937aSGao Xiang 13571e4a2955SGao Xiang do { 13581e4a2955SGao Xiang struct page *page; 135947e4937aSGao Xiang 13601e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 136147e4937aSGao Xiang MNGD_MAPPING(sbi), 136247e4937aSGao Xiang GFP_NOFS); 13631e4a2955SGao Xiang if (!page) 13641e4a2955SGao Xiang continue; 136547e4937aSGao Xiang 1366dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1367dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 136847e4937aSGao Xiang submit_bio_retry: 136994e4e153SGao Xiang submit_bio(bio); 137047e4937aSGao Xiang bio = NULL; 137147e4937aSGao Xiang } 137247e4937aSGao Xiang 137347e4937aSGao Xiang if (!bio) { 1374a8affc03SChristoph Hellwig bio = bio_alloc(GFP_NOIO, BIO_MAX_VECS); 13750c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1376dfeab2e9SGao Xiang 1377dfeab2e9SGao Xiang bio_set_dev(bio, mdev.m_bdev); 1378dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 13791e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1380a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1381a5c0b780SGao Xiang bio->bi_private = bi_private; 138294e4e153SGao Xiang bio->bi_opf = REQ_OP_READ; 13836ea5aad3SGao Xiang if (f->readahead) 13846ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 138547e4937aSGao Xiang ++nr_bios; 138647e4937aSGao Xiang } 138747e4937aSGao Xiang 13886c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 138947e4937aSGao Xiang goto submit_bio_retry; 139047e4937aSGao Xiang 13911e4a2955SGao Xiang last_index = cur; 13921e4a2955SGao Xiang bypass = false; 13931e4a2955SGao Xiang } while (++cur < end); 139447e4937aSGao Xiang 13951e4a2955SGao Xiang if (!bypass) 139647e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 139747e4937aSGao Xiang else 139847e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 139947e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 140047e4937aSGao Xiang 140147e4937aSGao Xiang if (bio) 140294e4e153SGao Xiang submit_bio(bio); 140347e4937aSGao Xiang 1404587a67b7SGao Xiang /* 1405587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1406587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1407587a67b7SGao Xiang */ 1408587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1409587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 14101e4a2955SGao Xiang return; 1411587a67b7SGao Xiang } 1412a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 141347e4937aSGao Xiang } 141447e4937aSGao Xiang 14150c638f70SGao Xiang static void z_erofs_runqueue(struct super_block *sb, 14166ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1417eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 141847e4937aSGao Xiang { 1419a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 142047e4937aSGao Xiang 14216ea5aad3SGao Xiang if (f->clt.owned_head == Z_EROFS_PCLUSTER_TAIL) 142247e4937aSGao Xiang return; 14236ea5aad3SGao Xiang z_erofs_submit_queue(sb, f, pagepool, io, &force_fg); 142447e4937aSGao Xiang 14250c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 14260c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 142747e4937aSGao Xiang 142847e4937aSGao Xiang if (!force_fg) 142947e4937aSGao Xiang return; 143047e4937aSGao Xiang 143147e4937aSGao Xiang /* wait until all bios are completed */ 1432a93f8c36SGao Xiang io_wait_event(io[JQ_SUBMIT].u.wait, 143347e4937aSGao Xiang !atomic_read(&io[JQ_SUBMIT].pending_bios)); 143447e4937aSGao Xiang 14350c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 14360c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 143747e4937aSGao Xiang } 143847e4937aSGao Xiang 143938629291SGao Xiang /* 144038629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 144138629291SGao Xiang * approximate readmore strategies as a start. 144238629291SGao Xiang */ 144338629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 144438629291SGao Xiang struct readahead_control *rac, 144538629291SGao Xiang erofs_off_t end, 1446eaa9172aSGao Xiang struct page **pagepool, 144738629291SGao Xiang bool backmost) 144838629291SGao Xiang { 144938629291SGao Xiang struct inode *inode = f->inode; 145038629291SGao Xiang struct erofs_map_blocks *map = &f->map; 145138629291SGao Xiang erofs_off_t cur; 145238629291SGao Xiang int err; 145338629291SGao Xiang 145438629291SGao Xiang if (backmost) { 145538629291SGao Xiang map->m_la = end; 1456622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1457622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 145838629291SGao Xiang if (err) 145938629291SGao Xiang return; 146038629291SGao Xiang 146138629291SGao Xiang /* expend ra for the trailing edge if readahead */ 146238629291SGao Xiang if (rac) { 146338629291SGao Xiang loff_t newstart = readahead_pos(rac); 146438629291SGao Xiang 146538629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 146638629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 146738629291SGao Xiang return; 146838629291SGao Xiang } 146938629291SGao Xiang end = round_up(end, PAGE_SIZE); 147038629291SGao Xiang } else { 147138629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 147238629291SGao Xiang 147338629291SGao Xiang if (!map->m_llen) 147438629291SGao Xiang return; 147538629291SGao Xiang } 147638629291SGao Xiang 147738629291SGao Xiang cur = map->m_la + map->m_llen - 1; 147838629291SGao Xiang while (cur >= end) { 147938629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 148038629291SGao Xiang struct page *page; 148138629291SGao Xiang 148238629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 148338629291SGao Xiang if (!page) 148438629291SGao Xiang goto skip; 148538629291SGao Xiang 148638629291SGao Xiang if (PageUptodate(page)) { 148738629291SGao Xiang unlock_page(page); 148838629291SGao Xiang put_page(page); 148938629291SGao Xiang goto skip; 149038629291SGao Xiang } 149138629291SGao Xiang 149238629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 149338629291SGao Xiang if (err) 149438629291SGao Xiang erofs_err(inode->i_sb, 149538629291SGao Xiang "readmore error at page %lu @ nid %llu", 149638629291SGao Xiang index, EROFS_I(inode)->nid); 149738629291SGao Xiang put_page(page); 149838629291SGao Xiang skip: 149938629291SGao Xiang if (cur < PAGE_SIZE) 150038629291SGao Xiang break; 150138629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 150238629291SGao Xiang } 150338629291SGao Xiang } 150438629291SGao Xiang 15050c638f70SGao Xiang static int z_erofs_readpage(struct file *file, struct page *page) 150647e4937aSGao Xiang { 150747e4937aSGao Xiang struct inode *const inode = page->mapping->host; 150840452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 150947e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1510eaa9172aSGao Xiang struct page *pagepool = NULL; 151147e4937aSGao Xiang int err; 151247e4937aSGao Xiang 151347e4937aSGao Xiang trace_erofs_readpage(page, false); 151447e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 151547e4937aSGao Xiang 151638629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 151738629291SGao Xiang &pagepool, true); 15181825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 151938629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 152038629291SGao Xiang 152147e4937aSGao Xiang (void)z_erofs_collector_end(&f.clt); 152247e4937aSGao Xiang 152347e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 152440452ffcSHuang Jianan z_erofs_runqueue(inode->i_sb, &f, &pagepool, 152540452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 152647e4937aSGao Xiang 152747e4937aSGao Xiang if (err) 15284f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 152947e4937aSGao Xiang 1530*09c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1531eaa9172aSGao Xiang erofs_release_pages(&pagepool); 153247e4937aSGao Xiang return err; 153347e4937aSGao Xiang } 153447e4937aSGao Xiang 15350615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 153647e4937aSGao Xiang { 15370615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 153847e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 153947e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1540eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 154138629291SGao Xiang unsigned int nr_pages; 154247e4937aSGao Xiang 15436ea5aad3SGao Xiang f.readahead = true; 15440615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 154547e4937aSGao Xiang 154638629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 154738629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 154838629291SGao Xiang nr_pages = readahead_count(rac); 154938629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 155038629291SGao Xiang 15510615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 155247e4937aSGao Xiang set_page_private(page, (unsigned long)head); 155347e4937aSGao Xiang head = page; 155447e4937aSGao Xiang } 155547e4937aSGao Xiang 155647e4937aSGao Xiang while (head) { 155747e4937aSGao Xiang struct page *page = head; 155847e4937aSGao Xiang int err; 155947e4937aSGao Xiang 156047e4937aSGao Xiang /* traversal in reverse order */ 156147e4937aSGao Xiang head = (void *)page_private(page); 156247e4937aSGao Xiang 15631825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1564a5876e24SGao Xiang if (err) 15654f761fa2SGao Xiang erofs_err(inode->i_sb, 15664f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 15674f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 156847e4937aSGao Xiang put_page(page); 156947e4937aSGao Xiang } 157038629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 157147e4937aSGao Xiang (void)z_erofs_collector_end(&f.clt); 157247e4937aSGao Xiang 157338629291SGao Xiang z_erofs_runqueue(inode->i_sb, &f, &pagepool, 157440452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 1575*09c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1576eaa9172aSGao Xiang erofs_release_pages(&pagepool); 157747e4937aSGao Xiang } 157847e4937aSGao Xiang 15790c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 15800c638f70SGao Xiang .readpage = z_erofs_readpage, 15810615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 158247e4937aSGao Xiang }; 1583