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 1955c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 1965c6dcc57SGao Xiang struct inode *const inode; 1975c6dcc57SGao Xiang struct erofs_map_blocks map; 1985c6dcc57SGao 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) { \ 2165c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 2175c6dcc57SGao 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 2226f39d1e1SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 2231825c8d7SGao Xiang enum z_erofs_cache_alloctype type, 224eaa9172aSGao Xiang struct page **pagepool) 22547e4937aSGao Xiang { 2266f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 2275c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 22847e4937aSGao Xiang bool standalone = true; 2296f39d1e1SGao Xiang /* 2306f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 2316f39d1e1SGao Xiang * can be used if low memory otherwise. 2326f39d1e1SGao Xiang */ 2331825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 2341825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 23581382f5fSGao Xiang struct page **pages; 23681382f5fSGao Xiang pgoff_t index; 23747e4937aSGao Xiang 2385c6dcc57SGao Xiang if (fe->mode < COLLECT_PRIMARY_FOLLOWED) 23947e4937aSGao Xiang return; 24047e4937aSGao Xiang 24181382f5fSGao Xiang pages = pcl->compressed_pages; 24281382f5fSGao Xiang index = pcl->obj.index; 24381382f5fSGao Xiang for (; index < pcl->obj.index + pcl->pclusterpages; ++index, ++pages) { 24447e4937aSGao Xiang struct page *page; 24547e4937aSGao Xiang compressed_page_t t; 2461825c8d7SGao Xiang struct page *newpage = NULL; 24747e4937aSGao Xiang 24847e4937aSGao Xiang /* the compressed page was loaded before */ 24947e4937aSGao Xiang if (READ_ONCE(*pages)) 25047e4937aSGao Xiang continue; 25147e4937aSGao Xiang 25247e4937aSGao Xiang page = find_get_page(mc, index); 25347e4937aSGao Xiang 25447e4937aSGao Xiang if (page) { 25547e4937aSGao Xiang t = tag_compressed_page_justfound(page); 2560b964600SGao Xiang } else { 2570b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 2580b964600SGao Xiang standalone = false; 2590b964600SGao Xiang switch (type) { 2600b964600SGao Xiang case TRYALLOC: 2611825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 2621825c8d7SGao Xiang if (!newpage) 26347e4937aSGao Xiang continue; 2640b964600SGao Xiang set_page_private(newpage, 2650b964600SGao Xiang Z_EROFS_PREALLOCATED_PAGE); 2660b964600SGao Xiang t = tag_compressed_page_justfound(newpage); 2670b964600SGao Xiang break; 2680b964600SGao Xiang default: /* DONTALLOC */ 2690b964600SGao Xiang continue; 2700b964600SGao Xiang } 27147e4937aSGao Xiang } 27247e4937aSGao Xiang 27347e4937aSGao Xiang if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t))) 27447e4937aSGao Xiang continue; 27547e4937aSGao Xiang 276eaa9172aSGao Xiang if (page) 27747e4937aSGao Xiang put_page(page); 278eaa9172aSGao Xiang else if (newpage) 279eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 28047e4937aSGao Xiang } 28147e4937aSGao Xiang 2820b964600SGao Xiang /* 2830b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 2840b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 2850b964600SGao Xiang */ 2860b964600SGao Xiang if (standalone) 2875c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 28847e4937aSGao Xiang } 28947e4937aSGao Xiang 29047e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 29147e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 29247e4937aSGao Xiang struct erofs_workgroup *grp) 29347e4937aSGao Xiang { 29447e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 29547e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 29647e4937aSGao Xiang int i; 29747e4937aSGao Xiang 298cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 29947e4937aSGao Xiang /* 30047e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 30147e4937aSGao Xiang * therefore no need to worry about available decompression users. 30247e4937aSGao Xiang */ 3039f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 30447e4937aSGao Xiang struct page *page = pcl->compressed_pages[i]; 30547e4937aSGao Xiang 30647e4937aSGao Xiang if (!page) 30747e4937aSGao Xiang continue; 30847e4937aSGao Xiang 30947e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 31047e4937aSGao Xiang if (!trylock_page(page)) 31147e4937aSGao Xiang return -EBUSY; 31247e4937aSGao Xiang 313f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 31447e4937aSGao Xiang continue; 31547e4937aSGao Xiang 31647e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 31747e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 3186aaa7b06SGao Xiang detach_page_private(page); 31947e4937aSGao Xiang unlock_page(page); 32047e4937aSGao Xiang } 32147e4937aSGao Xiang return 0; 32247e4937aSGao Xiang } 32347e4937aSGao Xiang 324d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 32547e4937aSGao Xiang { 32647e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 32747e4937aSGao Xiang int ret = 0; /* 0 - busy */ 32847e4937aSGao Xiang 32947e4937aSGao Xiang if (erofs_workgroup_try_to_freeze(&pcl->obj, 1)) { 33047e4937aSGao Xiang unsigned int i; 33147e4937aSGao Xiang 332cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 3339f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 33447e4937aSGao Xiang if (pcl->compressed_pages[i] == page) { 33547e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 33647e4937aSGao Xiang ret = 1; 33747e4937aSGao Xiang break; 33847e4937aSGao Xiang } 33947e4937aSGao Xiang } 34047e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 34147e4937aSGao Xiang 3426aaa7b06SGao Xiang if (ret) 3436aaa7b06SGao Xiang detach_page_private(page); 34447e4937aSGao Xiang } 34547e4937aSGao Xiang return ret; 34647e4937aSGao Xiang } 34747e4937aSGao Xiang 34847e4937aSGao Xiang /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 3495c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 35047e4937aSGao Xiang struct page *page) 35147e4937aSGao Xiang { 3525c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 35347e4937aSGao Xiang 3545c6dcc57SGao Xiang while (fe->icpage_ptr > pcl->compressed_pages) 3555c6dcc57SGao Xiang if (!cmpxchg(--fe->icpage_ptr, NULL, page)) 35647e4937aSGao Xiang return true; 35747e4937aSGao Xiang return false; 35847e4937aSGao Xiang } 35947e4937aSGao Xiang 36047e4937aSGao Xiang /* callers must be with collection lock held */ 3615c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 36286432a6dSGao Xiang struct page *page, enum z_erofs_page_type type, 36386432a6dSGao Xiang bool pvec_safereuse) 36447e4937aSGao Xiang { 36547e4937aSGao Xiang int ret; 36647e4937aSGao Xiang 36747e4937aSGao Xiang /* give priority for inplaceio */ 3685c6dcc57SGao Xiang if (fe->mode >= COLLECT_PRIMARY && 36947e4937aSGao Xiang type == Z_EROFS_PAGE_TYPE_EXCLUSIVE && 3705c6dcc57SGao Xiang z_erofs_try_inplace_io(fe, page)) 37147e4937aSGao Xiang return 0; 37247e4937aSGao Xiang 3735c6dcc57SGao Xiang ret = z_erofs_pagevec_enqueue(&fe->vector, page, type, 37486432a6dSGao Xiang pvec_safereuse); 3755c6dcc57SGao Xiang fe->cl->vcnt += (unsigned int)ret; 37647e4937aSGao Xiang return ret ? 0 : -EAGAIN; 37747e4937aSGao Xiang } 37847e4937aSGao Xiang 3795c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 38047e4937aSGao Xiang { 3815c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 3825c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 38347e4937aSGao Xiang 384473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 385473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 386473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 38747e4937aSGao Xiang *owned_head = &pcl->next; 388473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 3895c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_FOLLOWED; 390473e15b0SGao Xiang return; 391473e15b0SGao Xiang } 392473e15b0SGao Xiang 39347e4937aSGao Xiang /* 394473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 395473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 39647e4937aSGao Xiang */ 39747e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 398473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 39947e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 4005c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_HOOKED; 4015c6dcc57SGao Xiang f->tailpcl = NULL; 402473e15b0SGao Xiang return; 40347e4937aSGao Xiang } 404473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 4055c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY; 40647e4937aSGao Xiang } 40747e4937aSGao Xiang 4085c6dcc57SGao Xiang static int z_erofs_lookup_collection(struct z_erofs_decompress_frontend *fe, 40947e4937aSGao Xiang struct inode *inode, 41047e4937aSGao Xiang struct erofs_map_blocks *map) 41147e4937aSGao Xiang { 4125c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 41347e4937aSGao Xiang struct z_erofs_collection *cl; 41447e4937aSGao Xiang unsigned int length; 41547e4937aSGao Xiang 41664094a04SGao Xiang /* to avoid unexpected loop formed by corrupted images */ 4175c6dcc57SGao Xiang if (fe->owned_head == &pcl->next || pcl == fe->tailpcl) { 41847e4937aSGao Xiang DBG_BUGON(1); 4199e579fc1SGao Xiang return -EFSCORRUPTED; 42047e4937aSGao Xiang } 42147e4937aSGao Xiang 42247e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 4238d8a09b0SGao Xiang if (cl->pageofs != (map->m_la & ~PAGE_MASK)) { 42447e4937aSGao Xiang DBG_BUGON(1); 4259e579fc1SGao Xiang return -EFSCORRUPTED; 42647e4937aSGao Xiang } 42747e4937aSGao Xiang 42847e4937aSGao Xiang length = READ_ONCE(pcl->length); 42947e4937aSGao Xiang if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 43047e4937aSGao Xiang if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 43147e4937aSGao Xiang DBG_BUGON(1); 4329e579fc1SGao Xiang return -EFSCORRUPTED; 43347e4937aSGao Xiang } 43447e4937aSGao Xiang } else { 43547e4937aSGao Xiang unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 43647e4937aSGao Xiang 43747e4937aSGao Xiang if (map->m_flags & EROFS_MAP_FULL_MAPPED) 43847e4937aSGao Xiang llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 43947e4937aSGao Xiang 44047e4937aSGao Xiang while (llen > length && 44147e4937aSGao Xiang length != cmpxchg_relaxed(&pcl->length, length, llen)) { 44247e4937aSGao Xiang cpu_relax(); 44347e4937aSGao Xiang length = READ_ONCE(pcl->length); 44447e4937aSGao Xiang } 44547e4937aSGao Xiang } 44647e4937aSGao Xiang mutex_lock(&cl->lock); 44747e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 4485c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 4495c6dcc57SGao Xiang fe->tailpcl = pcl; 450473e15b0SGao Xiang 4515c6dcc57SGao Xiang z_erofs_try_to_claim_pcluster(fe); 4525c6dcc57SGao Xiang fe->cl = cl; 4539e579fc1SGao Xiang return 0; 45447e4937aSGao Xiang } 45547e4937aSGao Xiang 4565c6dcc57SGao Xiang static int z_erofs_register_collection(struct z_erofs_decompress_frontend *fe, 45747e4937aSGao Xiang struct inode *inode, 45847e4937aSGao Xiang struct erofs_map_blocks *map) 45947e4937aSGao Xiang { 460cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 46147e4937aSGao Xiang struct z_erofs_pcluster *pcl; 46247e4937aSGao Xiang struct z_erofs_collection *cl; 46364094a04SGao Xiang struct erofs_workgroup *grp; 46447e4937aSGao Xiang int err; 46547e4937aSGao Xiang 4668f899262SGao Xiang if (!(map->m_flags & EROFS_MAP_ENCODED)) { 4678f899262SGao Xiang DBG_BUGON(1); 4688f899262SGao Xiang return -EFSCORRUPTED; 4698f899262SGao Xiang } 4708f899262SGao Xiang 4719f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 472cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 473cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 4749f6cc76eSGao Xiang if (IS_ERR(pcl)) 4759f6cc76eSGao Xiang return PTR_ERR(pcl); 47647e4937aSGao Xiang 47764094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 4788f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 47947e4937aSGao Xiang pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 48047e4937aSGao Xiang (map->m_flags & EROFS_MAP_FULL_MAPPED ? 48147e4937aSGao Xiang Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 48247e4937aSGao Xiang 48347e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 4845c6dcc57SGao Xiang pcl->next = fe->owned_head; 4855c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED; 48647e4937aSGao Xiang 48747e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 48847e4937aSGao Xiang cl->pageofs = map->m_la & ~PAGE_MASK; 48947e4937aSGao Xiang 49047e4937aSGao Xiang /* 49147e4937aSGao Xiang * lock all primary followed works before visible to others 49247e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 49347e4937aSGao Xiang */ 4949f6cc76eSGao Xiang mutex_init(&cl->lock); 49564094a04SGao Xiang DBG_BUGON(!mutex_trylock(&cl->lock)); 49647e4937aSGao Xiang 497cecf864dSYue Hu if (ztailpacking) { 498cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 499cecf864dSYue Hu pcl->pageofs_in = erofs_blkoff(map->m_pa); 500cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 501cecf864dSYue Hu } else { 502cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 503cecf864dSYue Hu 50464094a04SGao Xiang grp = erofs_insert_workgroup(inode->i_sb, &pcl->obj); 50564094a04SGao Xiang if (IS_ERR(grp)) { 50664094a04SGao Xiang err = PTR_ERR(grp); 50764094a04SGao Xiang goto err_out; 50864094a04SGao Xiang } 50964094a04SGao Xiang 51064094a04SGao Xiang if (grp != &pcl->obj) { 5115c6dcc57SGao Xiang fe->pcl = container_of(grp, 512cecf864dSYue Hu struct z_erofs_pcluster, obj); 51364094a04SGao Xiang err = -EEXIST; 51464094a04SGao Xiang goto err_out; 51547e4937aSGao Xiang } 516cecf864dSYue Hu } 51747e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 5185c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 5195c6dcc57SGao Xiang fe->tailpcl = pcl; 5205c6dcc57SGao Xiang fe->owned_head = &pcl->next; 5215c6dcc57SGao Xiang fe->pcl = pcl; 5225c6dcc57SGao Xiang fe->cl = cl; 5239e579fc1SGao Xiang return 0; 52464094a04SGao Xiang 52564094a04SGao Xiang err_out: 52664094a04SGao Xiang mutex_unlock(&cl->lock); 5279f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 52864094a04SGao Xiang return err; 52947e4937aSGao Xiang } 53047e4937aSGao Xiang 5315c6dcc57SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe, 53247e4937aSGao Xiang struct inode *inode, 53347e4937aSGao Xiang struct erofs_map_blocks *map) 53447e4937aSGao Xiang { 53564094a04SGao Xiang struct erofs_workgroup *grp; 5369e579fc1SGao Xiang int ret; 53747e4937aSGao Xiang 5385c6dcc57SGao Xiang DBG_BUGON(fe->cl); 53947e4937aSGao Xiang 54047e4937aSGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous collection */ 5415c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 5425c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 54347e4937aSGao Xiang 544cecf864dSYue Hu if (map->m_flags & EROFS_MAP_META) { 545cecf864dSYue Hu if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 54647e4937aSGao Xiang DBG_BUGON(1); 547cecf864dSYue Hu return -EFSCORRUPTED; 548cecf864dSYue Hu } 549cecf864dSYue Hu goto tailpacking; 55047e4937aSGao Xiang } 55147e4937aSGao Xiang 55264094a04SGao Xiang grp = erofs_find_workgroup(inode->i_sb, map->m_pa >> PAGE_SHIFT); 55364094a04SGao Xiang if (grp) { 5545c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 55564094a04SGao Xiang } else { 556cecf864dSYue Hu tailpacking: 5575c6dcc57SGao Xiang ret = z_erofs_register_collection(fe, inode, map); 55864094a04SGao Xiang if (!ret) 55964094a04SGao Xiang goto out; 56064094a04SGao Xiang if (ret != -EEXIST) 5619e579fc1SGao Xiang return ret; 56264094a04SGao Xiang } 56347e4937aSGao Xiang 5645c6dcc57SGao Xiang ret = z_erofs_lookup_collection(fe, inode, map); 56564094a04SGao Xiang if (ret) { 5665c6dcc57SGao Xiang erofs_workgroup_put(&fe->pcl->obj); 56764094a04SGao Xiang return ret; 56864094a04SGao Xiang } 56964094a04SGao Xiang 57064094a04SGao Xiang out: 5715c6dcc57SGao Xiang z_erofs_pagevec_ctor_init(&fe->vector, Z_EROFS_NR_INLINE_PAGEVECS, 5725c6dcc57SGao Xiang fe->cl->pagevec, fe->cl->vcnt); 57381382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 5745c6dcc57SGao Xiang fe->icpage_ptr = fe->pcl->compressed_pages + 5755c6dcc57SGao Xiang z_erofs_pclusterpages(fe->pcl); 57647e4937aSGao Xiang return 0; 57747e4937aSGao Xiang } 57847e4937aSGao Xiang 57947e4937aSGao Xiang /* 58047e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 58147e4937aSGao Xiang * only after a RCU grace period. 58247e4937aSGao Xiang */ 58347e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 58447e4937aSGao Xiang { 58547e4937aSGao Xiang struct z_erofs_collection *const cl = 58647e4937aSGao Xiang container_of(head, struct z_erofs_collection, rcu); 58747e4937aSGao Xiang 5889f6cc76eSGao Xiang z_erofs_free_pcluster(container_of(cl, struct z_erofs_pcluster, 58947e4937aSGao Xiang primary_collection)); 59047e4937aSGao Xiang } 59147e4937aSGao Xiang 59247e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 59347e4937aSGao Xiang { 59447e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 59547e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 59647e4937aSGao Xiang struct z_erofs_collection *const cl = z_erofs_primarycollection(pcl); 59747e4937aSGao Xiang 59847e4937aSGao Xiang call_rcu(&cl->rcu, z_erofs_rcu_callback); 59947e4937aSGao Xiang } 60047e4937aSGao Xiang 60147e4937aSGao Xiang static void z_erofs_collection_put(struct z_erofs_collection *cl) 60247e4937aSGao Xiang { 60347e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 60447e4937aSGao Xiang container_of(cl, struct z_erofs_pcluster, primary_collection); 60547e4937aSGao Xiang 60647e4937aSGao Xiang erofs_workgroup_put(&pcl->obj); 60747e4937aSGao Xiang } 60847e4937aSGao Xiang 6095c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 61047e4937aSGao Xiang { 6115c6dcc57SGao Xiang struct z_erofs_collection *cl = fe->cl; 61247e4937aSGao Xiang 61347e4937aSGao Xiang if (!cl) 61447e4937aSGao Xiang return false; 61547e4937aSGao Xiang 6165c6dcc57SGao Xiang z_erofs_pagevec_ctor_exit(&fe->vector, false); 61747e4937aSGao Xiang mutex_unlock(&cl->lock); 61847e4937aSGao Xiang 61947e4937aSGao Xiang /* 62047e4937aSGao Xiang * if all pending pages are added, don't hold its reference 62147e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 62247e4937aSGao Xiang */ 6235c6dcc57SGao Xiang if (fe->mode < COLLECT_PRIMARY_FOLLOWED_NOINPLACE) 62447e4937aSGao Xiang z_erofs_collection_put(cl); 62547e4937aSGao Xiang 6265c6dcc57SGao Xiang fe->cl = NULL; 62747e4937aSGao Xiang return true; 62847e4937aSGao Xiang } 62947e4937aSGao Xiang 63047e4937aSGao Xiang static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe, 63147e4937aSGao Xiang unsigned int cachestrategy, 63247e4937aSGao Xiang erofs_off_t la) 63347e4937aSGao Xiang { 63447e4937aSGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 63547e4937aSGao Xiang return false; 63647e4937aSGao Xiang 63747e4937aSGao Xiang if (fe->backmost) 63847e4937aSGao Xiang return true; 63947e4937aSGao Xiang 64047e4937aSGao Xiang return cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 64147e4937aSGao Xiang la < fe->headoffset; 64247e4937aSGao Xiang } 64347e4937aSGao Xiang 64447e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 645eaa9172aSGao Xiang struct page *page, struct page **pagepool) 64647e4937aSGao Xiang { 64747e4937aSGao Xiang struct inode *const inode = fe->inode; 648bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 64947e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 65047e4937aSGao Xiang const loff_t offset = page_offset(page); 651dc76ea8cSGao Xiang bool tight = true; 65247e4937aSGao Xiang 65347e4937aSGao Xiang enum z_erofs_cache_alloctype cache_strategy; 65447e4937aSGao Xiang enum z_erofs_page_type page_type; 65547e4937aSGao Xiang unsigned int cur, end, spiltted, index; 65647e4937aSGao Xiang int err = 0; 65747e4937aSGao Xiang 65847e4937aSGao Xiang /* register locked file pages as online pages in pack */ 65947e4937aSGao Xiang z_erofs_onlinepage_init(page); 66047e4937aSGao Xiang 66147e4937aSGao Xiang spiltted = 0; 66247e4937aSGao Xiang end = PAGE_SIZE; 66347e4937aSGao Xiang repeat: 66447e4937aSGao Xiang cur = end - 1; 66547e4937aSGao Xiang 66647e4937aSGao Xiang /* lucky, within the range of the current map_blocks */ 66747e4937aSGao Xiang if (offset + cur >= map->m_la && 66847e4937aSGao Xiang offset + cur < map->m_la + map->m_llen) { 66947e4937aSGao Xiang /* didn't get a valid collection previously (very rare) */ 6705c6dcc57SGao Xiang if (!fe->cl) 67147e4937aSGao Xiang goto restart_now; 67247e4937aSGao Xiang goto hitted; 67347e4937aSGao Xiang } 67447e4937aSGao Xiang 67547e4937aSGao Xiang /* go ahead the next map_blocks */ 6764f761fa2SGao Xiang erofs_dbg("%s: [out-of-range] pos %llu", __func__, offset + cur); 67747e4937aSGao Xiang 6785c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 67947e4937aSGao Xiang fe->backmost = false; 68047e4937aSGao Xiang 68147e4937aSGao Xiang map->m_la = offset + cur; 68247e4937aSGao Xiang map->m_llen = 0; 68347e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 6848d8a09b0SGao Xiang if (err) 68547e4937aSGao Xiang goto err_out; 68647e4937aSGao Xiang 68747e4937aSGao Xiang restart_now: 6888d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) 68947e4937aSGao Xiang goto hitted; 69047e4937aSGao Xiang 6915c6dcc57SGao Xiang err = z_erofs_collector_begin(fe, inode, map); 6928d8a09b0SGao Xiang if (err) 69347e4937aSGao Xiang goto err_out; 69447e4937aSGao Xiang 6955c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 69609c54379SGao Xiang void *mp; 697cecf864dSYue Hu 69809c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 69909c54379SGao Xiang erofs_blknr(map->m_pa), EROFS_NO_KMAP); 70009c54379SGao Xiang if (IS_ERR(mp)) { 70109c54379SGao Xiang err = PTR_ERR(mp); 702cecf864dSYue Hu erofs_err(inode->i_sb, 703cecf864dSYue Hu "failed to get inline page, err %d", err); 704cecf864dSYue Hu goto err_out; 705cecf864dSYue Hu } 70609c54379SGao Xiang get_page(fe->map.buf.page); 7075c6dcc57SGao Xiang WRITE_ONCE(fe->pcl->compressed_pages[0], fe->map.buf.page); 7085c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 709cecf864dSYue Hu } else { 7106f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 711cecf864dSYue Hu if (should_alloc_managed_pages(fe, sbi->opt.cache_strategy, 712cecf864dSYue Hu map->m_la)) 7131825c8d7SGao Xiang cache_strategy = TRYALLOC; 71447e4937aSGao Xiang else 71547e4937aSGao Xiang cache_strategy = DONTALLOC; 71647e4937aSGao Xiang 7176f39d1e1SGao Xiang z_erofs_bind_cache(fe, cache_strategy, pagepool); 718cecf864dSYue Hu } 71947e4937aSGao Xiang hitted: 720dc76ea8cSGao Xiang /* 721dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 722dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 723dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 724dc76ea8cSGao Xiang * for inplace I/O or pagevec (should be processed in strict order.) 725dc76ea8cSGao Xiang */ 7265c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_HOOKED && 7275c6dcc57SGao Xiang fe->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 728dc76ea8cSGao Xiang 72947e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 7308d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 73147e4937aSGao Xiang zero_user_segment(page, cur, end); 73247e4937aSGao Xiang goto next_part; 73347e4937aSGao Xiang } 73447e4937aSGao Xiang 73547e4937aSGao Xiang /* let's derive page type */ 73647e4937aSGao Xiang page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 73747e4937aSGao Xiang (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 73847e4937aSGao Xiang (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 73947e4937aSGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 74047e4937aSGao Xiang 74147e4937aSGao Xiang if (cur) 7425c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_FOLLOWED); 74347e4937aSGao Xiang 74447e4937aSGao Xiang retry: 7455c6dcc57SGao Xiang err = z_erofs_attach_page(fe, page, page_type, 7465c6dcc57SGao Xiang fe->mode >= COLLECT_PRIMARY_FOLLOWED); 7476aaa7b06SGao Xiang /* should allocate an additional short-lived page for pagevec */ 74847e4937aSGao Xiang if (err == -EAGAIN) { 74947e4937aSGao Xiang struct page *const newpage = 750e3f78d5eSChao Yu alloc_page(GFP_NOFS | __GFP_NOFAIL); 75147e4937aSGao Xiang 7526aaa7b06SGao Xiang set_page_private(newpage, Z_EROFS_SHORTLIVED_PAGE); 7535c6dcc57SGao Xiang err = z_erofs_attach_page(fe, newpage, 75486432a6dSGao Xiang Z_EROFS_PAGE_TYPE_EXCLUSIVE, true); 7558d8a09b0SGao Xiang if (!err) 75647e4937aSGao Xiang goto retry; 75747e4937aSGao Xiang } 75847e4937aSGao Xiang 7598d8a09b0SGao Xiang if (err) 76047e4937aSGao Xiang goto err_out; 76147e4937aSGao Xiang 76247e4937aSGao Xiang index = page->index - (map->m_la >> PAGE_SHIFT); 76347e4937aSGao Xiang 76447e4937aSGao Xiang z_erofs_onlinepage_fixup(page, index, true); 76547e4937aSGao Xiang 76647e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 76747e4937aSGao Xiang ++spiltted; 76847e4937aSGao Xiang /* also update nr_pages */ 7695c6dcc57SGao Xiang fe->cl->nr_pages = max_t(pgoff_t, fe->cl->nr_pages, index + 1); 77047e4937aSGao Xiang next_part: 77147e4937aSGao Xiang /* can be used for verification */ 77247e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 77347e4937aSGao Xiang 77447e4937aSGao Xiang end = cur; 77547e4937aSGao Xiang if (end > 0) 77647e4937aSGao Xiang goto repeat; 77747e4937aSGao Xiang 77847e4937aSGao Xiang out: 77947e4937aSGao Xiang z_erofs_onlinepage_endio(page); 78047e4937aSGao Xiang 7814f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 78247e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 78347e4937aSGao Xiang return err; 78447e4937aSGao Xiang 78547e4937aSGao Xiang /* if some error occurred while processing this page */ 78647e4937aSGao Xiang err_out: 78747e4937aSGao Xiang SetPageError(page); 78847e4937aSGao Xiang goto out; 78947e4937aSGao Xiang } 79047e4937aSGao Xiang 79140452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 79240452ffcSHuang Jianan unsigned int readahead_pages) 79340452ffcSHuang Jianan { 794*a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 79540452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 79640452ffcSHuang Jianan !readahead_pages) 79740452ffcSHuang Jianan return true; 79840452ffcSHuang Jianan 79940452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 80040452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 80140452ffcSHuang Jianan return true; 80240452ffcSHuang Jianan 80340452ffcSHuang Jianan return false; 80440452ffcSHuang Jianan } 80540452ffcSHuang Jianan 8066aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 8076aaa7b06SGao Xiang { 8086aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 8096aaa7b06SGao Xiang } 8106aaa7b06SGao Xiang 81147e4937aSGao Xiang static int z_erofs_decompress_pcluster(struct super_block *sb, 81247e4937aSGao Xiang struct z_erofs_pcluster *pcl, 813eaa9172aSGao Xiang struct page **pagepool) 81447e4937aSGao Xiang { 81547e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 816cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 81747e4937aSGao Xiang struct z_erofs_pagevec_ctor ctor; 8189f6cc76eSGao Xiang unsigned int i, inputsize, outputsize, llen, nr_pages; 81947e4937aSGao Xiang struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 82047e4937aSGao Xiang struct page **pages, **compressed_pages, *page; 82147e4937aSGao Xiang 82247e4937aSGao Xiang enum z_erofs_page_type page_type; 82347e4937aSGao Xiang bool overlapped, partial; 82447e4937aSGao Xiang struct z_erofs_collection *cl; 82547e4937aSGao Xiang int err; 82647e4937aSGao Xiang 82747e4937aSGao Xiang might_sleep(); 82847e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 82947e4937aSGao Xiang DBG_BUGON(!READ_ONCE(cl->nr_pages)); 83047e4937aSGao Xiang 83147e4937aSGao Xiang mutex_lock(&cl->lock); 83247e4937aSGao Xiang nr_pages = cl->nr_pages; 83347e4937aSGao Xiang 8348d8a09b0SGao Xiang if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 83547e4937aSGao Xiang pages = pages_onstack; 83647e4937aSGao Xiang } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 83747e4937aSGao Xiang mutex_trylock(&z_pagemap_global_lock)) { 83847e4937aSGao Xiang pages = z_pagemap_global; 83947e4937aSGao Xiang } else { 84047e4937aSGao Xiang gfp_t gfp_flags = GFP_KERNEL; 84147e4937aSGao Xiang 84247e4937aSGao Xiang if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 84347e4937aSGao Xiang gfp_flags |= __GFP_NOFAIL; 84447e4937aSGao Xiang 84547e4937aSGao Xiang pages = kvmalloc_array(nr_pages, sizeof(struct page *), 84647e4937aSGao Xiang gfp_flags); 84747e4937aSGao Xiang 84847e4937aSGao Xiang /* fallback to global pagemap for the lowmem scenario */ 8498d8a09b0SGao Xiang if (!pages) { 85047e4937aSGao Xiang mutex_lock(&z_pagemap_global_lock); 85147e4937aSGao Xiang pages = z_pagemap_global; 85247e4937aSGao Xiang } 85347e4937aSGao Xiang } 85447e4937aSGao Xiang 85547e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) 85647e4937aSGao Xiang pages[i] = NULL; 85747e4937aSGao Xiang 85847e4937aSGao Xiang err = 0; 85947e4937aSGao Xiang z_erofs_pagevec_ctor_init(&ctor, Z_EROFS_NR_INLINE_PAGEVECS, 86047e4937aSGao Xiang cl->pagevec, 0); 86147e4937aSGao Xiang 86247e4937aSGao Xiang for (i = 0; i < cl->vcnt; ++i) { 86347e4937aSGao Xiang unsigned int pagenr; 86447e4937aSGao Xiang 86547e4937aSGao Xiang page = z_erofs_pagevec_dequeue(&ctor, &page_type); 86647e4937aSGao Xiang 86747e4937aSGao Xiang /* all pages in pagevec ought to be valid */ 86847e4937aSGao Xiang DBG_BUGON(!page); 8696aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 87047e4937aSGao Xiang 8716aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 87247e4937aSGao Xiang continue; 87347e4937aSGao Xiang 87447e4937aSGao Xiang if (page_type == Z_EROFS_VLE_PAGE_TYPE_HEAD) 87547e4937aSGao Xiang pagenr = 0; 87647e4937aSGao Xiang else 87747e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 87847e4937aSGao Xiang 87947e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 88047e4937aSGao Xiang 88147e4937aSGao Xiang /* 88247e4937aSGao Xiang * currently EROFS doesn't support multiref(dedup), 88347e4937aSGao Xiang * so here erroring out one multiref page. 88447e4937aSGao Xiang */ 8858d8a09b0SGao Xiang if (pages[pagenr]) { 88647e4937aSGao Xiang DBG_BUGON(1); 88747e4937aSGao Xiang SetPageError(pages[pagenr]); 88847e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 88947e4937aSGao Xiang err = -EFSCORRUPTED; 89047e4937aSGao Xiang } 89147e4937aSGao Xiang pages[pagenr] = page; 89247e4937aSGao Xiang } 89347e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&ctor, true); 89447e4937aSGao Xiang 89547e4937aSGao Xiang overlapped = false; 89647e4937aSGao Xiang compressed_pages = pcl->compressed_pages; 89747e4937aSGao Xiang 898cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 89947e4937aSGao Xiang unsigned int pagenr; 90047e4937aSGao Xiang 90147e4937aSGao Xiang page = compressed_pages[i]; 90247e4937aSGao Xiang /* all compressed pages ought to be valid */ 90347e4937aSGao Xiang DBG_BUGON(!page); 90447e4937aSGao Xiang 905cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 906cecf864dSYue Hu if (!PageUptodate(page)) 907cecf864dSYue Hu err = -EIO; 908cecf864dSYue Hu continue; 909cecf864dSYue Hu } 910cecf864dSYue Hu 911cecf864dSYue Hu DBG_BUGON(z_erofs_page_is_invalidated(page)); 9126aaa7b06SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 91347e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) { 9148d8a09b0SGao Xiang if (!PageUptodate(page)) 91547e4937aSGao Xiang err = -EIO; 91647e4937aSGao Xiang continue; 91747e4937aSGao Xiang } 91847e4937aSGao Xiang 91947e4937aSGao Xiang /* 92047e4937aSGao Xiang * only if non-head page can be selected 92147e4937aSGao Xiang * for inplace decompression 92247e4937aSGao Xiang */ 92347e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 92447e4937aSGao Xiang 92547e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 9268d8a09b0SGao Xiang if (pages[pagenr]) { 92747e4937aSGao Xiang DBG_BUGON(1); 92847e4937aSGao Xiang SetPageError(pages[pagenr]); 92947e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 93047e4937aSGao Xiang err = -EFSCORRUPTED; 93147e4937aSGao Xiang } 93247e4937aSGao Xiang pages[pagenr] = page; 93347e4937aSGao Xiang 93447e4937aSGao Xiang overlapped = true; 93547e4937aSGao Xiang } 93647e4937aSGao Xiang 9376aaa7b06SGao Xiang /* PG_error needs checking for all non-managed pages */ 9388d8a09b0SGao Xiang if (PageError(page)) { 93947e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 94047e4937aSGao Xiang err = -EIO; 94147e4937aSGao Xiang } 94247e4937aSGao Xiang } 94347e4937aSGao Xiang 9448d8a09b0SGao Xiang if (err) 94547e4937aSGao Xiang goto out; 94647e4937aSGao Xiang 94747e4937aSGao Xiang llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 94847e4937aSGao Xiang if (nr_pages << PAGE_SHIFT >= cl->pageofs + llen) { 94947e4937aSGao Xiang outputsize = llen; 95047e4937aSGao Xiang partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 95147e4937aSGao Xiang } else { 95247e4937aSGao Xiang outputsize = (nr_pages << PAGE_SHIFT) - cl->pageofs; 95347e4937aSGao Xiang partial = true; 95447e4937aSGao Xiang } 95547e4937aSGao Xiang 956cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 957cecf864dSYue Hu inputsize = pcl->tailpacking_size; 958cecf864dSYue Hu else 959cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 960cecf864dSYue Hu 96147e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 96247e4937aSGao Xiang .sb = sb, 96347e4937aSGao Xiang .in = compressed_pages, 96447e4937aSGao Xiang .out = pages, 965cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 96647e4937aSGao Xiang .pageofs_out = cl->pageofs, 9679f6cc76eSGao Xiang .inputsize = inputsize, 96847e4937aSGao Xiang .outputsize = outputsize, 96947e4937aSGao Xiang .alg = pcl->algorithmformat, 97047e4937aSGao Xiang .inplace_io = overlapped, 97147e4937aSGao Xiang .partial_decoding = partial 97247e4937aSGao Xiang }, pagepool); 97347e4937aSGao Xiang 97447e4937aSGao Xiang out: 975cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 976cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 977cecf864dSYue Hu page = compressed_pages[0]; 978cecf864dSYue Hu WRITE_ONCE(compressed_pages[0], NULL); 979cecf864dSYue Hu put_page(page); 980cecf864dSYue Hu } else { 981cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 98247e4937aSGao Xiang page = compressed_pages[i]; 98347e4937aSGao Xiang 98447e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 98547e4937aSGao Xiang continue; 98647e4937aSGao Xiang 9876aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 9886aaa7b06SGao Xiang (void)z_erofs_put_shortlivedpage(pagepool, page); 98947e4937aSGao Xiang WRITE_ONCE(compressed_pages[i], NULL); 99047e4937aSGao Xiang } 991cecf864dSYue Hu } 99247e4937aSGao Xiang 99347e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) { 99447e4937aSGao Xiang page = pages[i]; 99547e4937aSGao Xiang if (!page) 99647e4937aSGao Xiang continue; 99747e4937aSGao Xiang 9986aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 99947e4937aSGao Xiang 10006aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10016aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 100247e4937aSGao Xiang continue; 100347e4937aSGao Xiang 10048d8a09b0SGao Xiang if (err < 0) 100547e4937aSGao Xiang SetPageError(page); 100647e4937aSGao Xiang 100747e4937aSGao Xiang z_erofs_onlinepage_endio(page); 100847e4937aSGao Xiang } 100947e4937aSGao Xiang 101047e4937aSGao Xiang if (pages == z_pagemap_global) 101147e4937aSGao Xiang mutex_unlock(&z_pagemap_global_lock); 10128d8a09b0SGao Xiang else if (pages != pages_onstack) 101347e4937aSGao Xiang kvfree(pages); 101447e4937aSGao Xiang 101547e4937aSGao Xiang cl->nr_pages = 0; 101647e4937aSGao Xiang cl->vcnt = 0; 101747e4937aSGao Xiang 101847e4937aSGao Xiang /* all cl locks MUST be taken before the following line */ 101947e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 102047e4937aSGao Xiang 102147e4937aSGao Xiang /* all cl locks SHOULD be released right now */ 102247e4937aSGao Xiang mutex_unlock(&cl->lock); 102347e4937aSGao Xiang 102447e4937aSGao Xiang z_erofs_collection_put(cl); 102547e4937aSGao Xiang return err; 102647e4937aSGao Xiang } 102747e4937aSGao Xiang 10280c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1029eaa9172aSGao Xiang struct page **pagepool) 103047e4937aSGao Xiang { 103147e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 103247e4937aSGao Xiang 103347e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 103447e4937aSGao Xiang struct z_erofs_pcluster *pcl; 103547e4937aSGao Xiang 103647e4937aSGao Xiang /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 103747e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 103847e4937aSGao Xiang 103947e4937aSGao Xiang /* no possible that 'owned' equals NULL */ 104047e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 104147e4937aSGao Xiang 104247e4937aSGao Xiang pcl = container_of(owned, struct z_erofs_pcluster, next); 104347e4937aSGao Xiang owned = READ_ONCE(pcl->next); 104447e4937aSGao Xiang 1045a4b1fab1SGao Xiang z_erofs_decompress_pcluster(io->sb, pcl, pagepool); 104647e4937aSGao Xiang } 104747e4937aSGao Xiang } 104847e4937aSGao Xiang 10490c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 105047e4937aSGao Xiang { 1051a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1052a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1053eaa9172aSGao Xiang struct page *pagepool = NULL; 105447e4937aSGao Xiang 1055a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 10560c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 105747e4937aSGao Xiang 1058eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1059a4b1fab1SGao Xiang kvfree(bgq); 106047e4937aSGao Xiang } 106147e4937aSGao Xiang 10627865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 10637865827cSGao Xiang bool sync, int bios) 10647865827cSGao Xiang { 10657865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 10667865827cSGao Xiang 10677865827cSGao Xiang /* wake up the caller thread for sync decompression */ 10687865827cSGao Xiang if (sync) { 10697865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 107060b30050SHongyu Jin complete(&io->u.done); 107160b30050SHongyu Jin 10727865827cSGao Xiang return; 10737865827cSGao Xiang } 10747865827cSGao Xiang 10757865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 10767865827cSGao Xiang return; 10777865827cSGao Xiang /* Use workqueue and sync decompression for atomic contexts only */ 10787865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 10797865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 10807865827cSGao Xiang /* enable sync decompression for readahead */ 10817865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 10827865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 10837865827cSGao Xiang return; 10847865827cSGao Xiang } 10857865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 10867865827cSGao Xiang } 10877865827cSGao Xiang 108847e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 108947e4937aSGao Xiang unsigned int nr, 1090eaa9172aSGao Xiang struct page **pagepool, 10919f2731d6SGao Xiang struct address_space *mc) 109247e4937aSGao Xiang { 109347e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 10949f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 109547e4937aSGao Xiang bool tocache = false; 109647e4937aSGao Xiang 109747e4937aSGao Xiang struct address_space *mapping; 109847e4937aSGao Xiang struct page *oldpage, *page; 109947e4937aSGao Xiang 110047e4937aSGao Xiang compressed_page_t t; 110147e4937aSGao Xiang int justfound; 110247e4937aSGao Xiang 110347e4937aSGao Xiang repeat: 110447e4937aSGao Xiang page = READ_ONCE(pcl->compressed_pages[nr]); 110547e4937aSGao Xiang oldpage = page; 110647e4937aSGao Xiang 110747e4937aSGao Xiang if (!page) 110847e4937aSGao Xiang goto out_allocpage; 110947e4937aSGao Xiang 111047e4937aSGao Xiang /* process the target tagged pointer */ 111147e4937aSGao Xiang t = tagptr_init(compressed_page_t, page); 111247e4937aSGao Xiang justfound = tagptr_unfold_tags(t); 111347e4937aSGao Xiang page = tagptr_unfold_ptr(t); 111447e4937aSGao Xiang 11151825c8d7SGao Xiang /* 11161825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 11171825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 11181825c8d7SGao Xiang */ 11191825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 11201825c8d7SGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 11211825c8d7SGao Xiang set_page_private(page, 0); 11221825c8d7SGao Xiang tocache = true; 11231825c8d7SGao Xiang goto out_tocache; 11241825c8d7SGao Xiang } 112547e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 112647e4937aSGao Xiang 112747e4937aSGao Xiang /* 11286aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 112947e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 113047e4937aSGao Xiang */ 113147e4937aSGao Xiang if (mapping && mapping != mc) 113247e4937aSGao Xiang /* ought to be unmanaged pages */ 113347e4937aSGao Xiang goto out; 113447e4937aSGao Xiang 11356aaa7b06SGao Xiang /* directly return for shortlived page as well */ 11366aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 11376aaa7b06SGao Xiang goto out; 11386aaa7b06SGao Xiang 113947e4937aSGao Xiang lock_page(page); 114047e4937aSGao Xiang 114147e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 114247e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 114347e4937aSGao Xiang 114447e4937aSGao Xiang /* the page is still in manage cache */ 114547e4937aSGao Xiang if (page->mapping == mc) { 114647e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 114747e4937aSGao Xiang 114847e4937aSGao Xiang ClearPageError(page); 114947e4937aSGao Xiang if (!PagePrivate(page)) { 115047e4937aSGao Xiang /* 115147e4937aSGao Xiang * impossible to be !PagePrivate(page) for 115247e4937aSGao Xiang * the current restriction as well if 115347e4937aSGao Xiang * the page is already in compressed_pages[]. 115447e4937aSGao Xiang */ 115547e4937aSGao Xiang DBG_BUGON(!justfound); 115647e4937aSGao Xiang 115747e4937aSGao Xiang justfound = 0; 115847e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 115947e4937aSGao Xiang SetPagePrivate(page); 116047e4937aSGao Xiang } 116147e4937aSGao Xiang 116247e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 116347e4937aSGao Xiang if (PageUptodate(page)) { 116447e4937aSGao Xiang unlock_page(page); 116547e4937aSGao Xiang page = NULL; 116647e4937aSGao Xiang } 116747e4937aSGao Xiang goto out; 116847e4937aSGao Xiang } 116947e4937aSGao Xiang 117047e4937aSGao Xiang /* 117147e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 117247e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 117347e4937aSGao Xiang */ 117447e4937aSGao Xiang DBG_BUGON(page->mapping); 117547e4937aSGao Xiang DBG_BUGON(!justfound); 117647e4937aSGao Xiang 117747e4937aSGao Xiang tocache = true; 117847e4937aSGao Xiang unlock_page(page); 117947e4937aSGao Xiang put_page(page); 118047e4937aSGao Xiang out_allocpage: 11815ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 11825ddcee1fSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_pages[nr], oldpage, page)) { 1183eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 11845ddcee1fSGao Xiang cond_resched(); 11855ddcee1fSGao Xiang goto repeat; 11865ddcee1fSGao Xiang } 11871825c8d7SGao Xiang out_tocache: 1188bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1189bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1190bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1191bf225074SGao Xiang goto out; 1192a30573b3SGao Xiang } 1193bf225074SGao Xiang attach_page_private(page, pcl); 1194bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1195bf225074SGao Xiang put_page(page); 1196bf225074SGao Xiang 119747e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 119847e4937aSGao Xiang return page; 119947e4937aSGao Xiang } 120047e4937aSGao Xiang 1201a4b1fab1SGao Xiang static struct z_erofs_decompressqueue * 1202a4b1fab1SGao Xiang jobqueue_init(struct super_block *sb, 1203a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 120447e4937aSGao Xiang { 1205a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 120647e4937aSGao Xiang 1207a4b1fab1SGao Xiang if (fg && !*fg) { 1208a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1209a4b1fab1SGao Xiang if (!q) { 1210a4b1fab1SGao Xiang *fg = true; 1211a4b1fab1SGao Xiang goto fg_out; 121247e4937aSGao Xiang } 12130c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1214a4b1fab1SGao Xiang } else { 1215a4b1fab1SGao Xiang fg_out: 1216a4b1fab1SGao Xiang q = fgq; 121760b30050SHongyu Jin init_completion(&fgq->u.done); 1218a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 1219a4b1fab1SGao Xiang } 1220a4b1fab1SGao Xiang q->sb = sb; 1221a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1222a4b1fab1SGao Xiang return q; 122347e4937aSGao Xiang } 122447e4937aSGao Xiang 122547e4937aSGao Xiang /* define decompression jobqueue types */ 122647e4937aSGao Xiang enum { 122747e4937aSGao Xiang JQ_BYPASS, 122847e4937aSGao Xiang JQ_SUBMIT, 122947e4937aSGao Xiang NR_JOBQUEUES, 123047e4937aSGao Xiang }; 123147e4937aSGao Xiang 123247e4937aSGao Xiang static void *jobqueueset_init(struct super_block *sb, 1233a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[], 1234a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 123547e4937aSGao Xiang { 123647e4937aSGao Xiang /* 123747e4937aSGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 123847e4937aSGao Xiang * no need to read from device for all pclusters in this queue. 123947e4937aSGao Xiang */ 1240a4b1fab1SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1241a4b1fab1SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 124247e4937aSGao Xiang 1243a4b1fab1SGao Xiang return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 124447e4937aSGao Xiang } 124547e4937aSGao Xiang 124647e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 124747e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 124847e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 124947e4937aSGao Xiang { 125047e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 125147e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 125247e4937aSGao Xiang 125347e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 125447e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 125547e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 125647e4937aSGao Xiang 125747e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 125847e4937aSGao Xiang 125947e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 126047e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 126147e4937aSGao Xiang 126247e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 126347e4937aSGao Xiang } 126447e4937aSGao Xiang 12657865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 12667865827cSGao Xiang { 12677865827cSGao Xiang tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 12687865827cSGao Xiang struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 12697865827cSGao Xiang blk_status_t err = bio->bi_status; 12707865827cSGao Xiang struct bio_vec *bvec; 12717865827cSGao Xiang struct bvec_iter_all iter_all; 12727865827cSGao Xiang 12737865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 12747865827cSGao Xiang struct page *page = bvec->bv_page; 12757865827cSGao Xiang 12767865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 12777865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 12787865827cSGao Xiang 12797865827cSGao Xiang if (err) 12807865827cSGao Xiang SetPageError(page); 12817865827cSGao Xiang 12827865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 12837865827cSGao Xiang if (!err) 12847865827cSGao Xiang SetPageUptodate(page); 12857865827cSGao Xiang unlock_page(page); 12867865827cSGao Xiang } 12877865827cSGao Xiang } 12887865827cSGao Xiang z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 12897865827cSGao Xiang bio_put(bio); 12907865827cSGao Xiang } 12917865827cSGao Xiang 12921e4a2955SGao Xiang static void z_erofs_submit_queue(struct super_block *sb, 12936ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1294eaa9172aSGao Xiang struct page **pagepool, 1295a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1296a4b1fab1SGao Xiang bool *force_fg) 129747e4937aSGao Xiang { 1298bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 129947e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1300a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 130147e4937aSGao Xiang void *bi_private; 13025c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1303dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 13043f649ab7SKees Cook pgoff_t last_index; 1305dfeab2e9SGao Xiang struct block_device *last_bdev; 13061e4a2955SGao Xiang unsigned int nr_bios = 0; 13071e4a2955SGao Xiang struct bio *bio = NULL; 130847e4937aSGao Xiang 1309a4b1fab1SGao Xiang bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1310a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1311a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 131247e4937aSGao Xiang 131347e4937aSGao Xiang /* by default, all need io submission */ 131447e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 131547e4937aSGao Xiang 131647e4937aSGao Xiang do { 1317dfeab2e9SGao Xiang struct erofs_map_dev mdev; 131847e4937aSGao Xiang struct z_erofs_pcluster *pcl; 13191e4a2955SGao Xiang pgoff_t cur, end; 13201e4a2955SGao Xiang unsigned int i = 0; 13211e4a2955SGao Xiang bool bypass = true; 132247e4937aSGao Xiang 132347e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 132447e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 132547e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 132647e4937aSGao Xiang 132747e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 132847e4937aSGao Xiang 1329cecf864dSYue Hu /* close the main owned chain at first */ 1330cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1331cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1332cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1333cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1334cecf864dSYue Hu continue; 1335cecf864dSYue Hu } 1336cecf864dSYue Hu 1337dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1338dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1339dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1340dfeab2e9SGao Xiang }; 1341dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1342dfeab2e9SGao Xiang 1343dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 13449f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 134547e4937aSGao Xiang 13461e4a2955SGao Xiang do { 13471e4a2955SGao Xiang struct page *page; 134847e4937aSGao Xiang 13491e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 13509f2731d6SGao Xiang MNGD_MAPPING(sbi)); 13511e4a2955SGao Xiang if (!page) 13521e4a2955SGao Xiang continue; 135347e4937aSGao Xiang 1354dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1355dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 135647e4937aSGao Xiang submit_bio_retry: 135794e4e153SGao Xiang submit_bio(bio); 135847e4937aSGao Xiang bio = NULL; 135947e4937aSGao Xiang } 136047e4937aSGao Xiang 136147e4937aSGao Xiang if (!bio) { 136207888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 136307888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 13640c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1365dfeab2e9SGao Xiang 1366dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 13671e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1368a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1369a5c0b780SGao Xiang bio->bi_private = bi_private; 13706ea5aad3SGao Xiang if (f->readahead) 13716ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 137247e4937aSGao Xiang ++nr_bios; 137347e4937aSGao Xiang } 137447e4937aSGao Xiang 13756c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 137647e4937aSGao Xiang goto submit_bio_retry; 137747e4937aSGao Xiang 13781e4a2955SGao Xiang last_index = cur; 13791e4a2955SGao Xiang bypass = false; 13801e4a2955SGao Xiang } while (++cur < end); 138147e4937aSGao Xiang 13821e4a2955SGao Xiang if (!bypass) 138347e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 138447e4937aSGao Xiang else 138547e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 138647e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 138747e4937aSGao Xiang 138847e4937aSGao Xiang if (bio) 138994e4e153SGao Xiang submit_bio(bio); 139047e4937aSGao Xiang 1391587a67b7SGao Xiang /* 1392587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1393587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1394587a67b7SGao Xiang */ 1395587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1396587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 13971e4a2955SGao Xiang return; 1398587a67b7SGao Xiang } 1399a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 140047e4937aSGao Xiang } 140147e4937aSGao Xiang 14020c638f70SGao Xiang static void z_erofs_runqueue(struct super_block *sb, 14036ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1404eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 140547e4937aSGao Xiang { 1406a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 140747e4937aSGao Xiang 14085c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 140947e4937aSGao Xiang return; 14106ea5aad3SGao Xiang z_erofs_submit_queue(sb, f, pagepool, io, &force_fg); 141147e4937aSGao Xiang 14120c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 14130c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 141447e4937aSGao Xiang 141547e4937aSGao Xiang if (!force_fg) 141647e4937aSGao Xiang return; 141747e4937aSGao Xiang 141847e4937aSGao Xiang /* wait until all bios are completed */ 141960b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 142047e4937aSGao Xiang 14210c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 14220c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 142347e4937aSGao Xiang } 142447e4937aSGao Xiang 142538629291SGao Xiang /* 142638629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 142738629291SGao Xiang * approximate readmore strategies as a start. 142838629291SGao Xiang */ 142938629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 143038629291SGao Xiang struct readahead_control *rac, 143138629291SGao Xiang erofs_off_t end, 1432eaa9172aSGao Xiang struct page **pagepool, 143338629291SGao Xiang bool backmost) 143438629291SGao Xiang { 143538629291SGao Xiang struct inode *inode = f->inode; 143638629291SGao Xiang struct erofs_map_blocks *map = &f->map; 143738629291SGao Xiang erofs_off_t cur; 143838629291SGao Xiang int err; 143938629291SGao Xiang 144038629291SGao Xiang if (backmost) { 144138629291SGao Xiang map->m_la = end; 1442622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1443622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 144438629291SGao Xiang if (err) 144538629291SGao Xiang return; 144638629291SGao Xiang 144738629291SGao Xiang /* expend ra for the trailing edge if readahead */ 144838629291SGao Xiang if (rac) { 144938629291SGao Xiang loff_t newstart = readahead_pos(rac); 145038629291SGao Xiang 145138629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 145238629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 145338629291SGao Xiang return; 145438629291SGao Xiang } 145538629291SGao Xiang end = round_up(end, PAGE_SIZE); 145638629291SGao Xiang } else { 145738629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 145838629291SGao Xiang 145938629291SGao Xiang if (!map->m_llen) 146038629291SGao Xiang return; 146138629291SGao Xiang } 146238629291SGao Xiang 146338629291SGao Xiang cur = map->m_la + map->m_llen - 1; 146438629291SGao Xiang while (cur >= end) { 146538629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 146638629291SGao Xiang struct page *page; 146738629291SGao Xiang 146838629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 146938629291SGao Xiang if (!page) 147038629291SGao Xiang goto skip; 147138629291SGao Xiang 147238629291SGao Xiang if (PageUptodate(page)) { 147338629291SGao Xiang unlock_page(page); 147438629291SGao Xiang put_page(page); 147538629291SGao Xiang goto skip; 147638629291SGao Xiang } 147738629291SGao Xiang 147838629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 147938629291SGao Xiang if (err) 148038629291SGao Xiang erofs_err(inode->i_sb, 148138629291SGao Xiang "readmore error at page %lu @ nid %llu", 148238629291SGao Xiang index, EROFS_I(inode)->nid); 148338629291SGao Xiang put_page(page); 148438629291SGao Xiang skip: 148538629291SGao Xiang if (cur < PAGE_SIZE) 148638629291SGao Xiang break; 148738629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 148838629291SGao Xiang } 148938629291SGao Xiang } 149038629291SGao Xiang 1491*a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 149247e4937aSGao Xiang { 1493*a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 149447e4937aSGao Xiang struct inode *const inode = page->mapping->host; 149540452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 149647e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1497eaa9172aSGao Xiang struct page *pagepool = NULL; 149847e4937aSGao Xiang int err; 149947e4937aSGao Xiang 150047e4937aSGao Xiang trace_erofs_readpage(page, false); 150147e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 150247e4937aSGao Xiang 150338629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 150438629291SGao Xiang &pagepool, true); 15051825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 150638629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 150738629291SGao Xiang 15085c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 150947e4937aSGao Xiang 151047e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 151140452ffcSHuang Jianan z_erofs_runqueue(inode->i_sb, &f, &pagepool, 151240452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 151347e4937aSGao Xiang 151447e4937aSGao Xiang if (err) 15154f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 151647e4937aSGao Xiang 151709c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1518eaa9172aSGao Xiang erofs_release_pages(&pagepool); 151947e4937aSGao Xiang return err; 152047e4937aSGao Xiang } 152147e4937aSGao Xiang 15220615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 152347e4937aSGao Xiang { 15240615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 152547e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 152647e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1527eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 152838629291SGao Xiang unsigned int nr_pages; 152947e4937aSGao Xiang 15306ea5aad3SGao Xiang f.readahead = true; 15310615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 153247e4937aSGao Xiang 153338629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 153438629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 153538629291SGao Xiang nr_pages = readahead_count(rac); 153638629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 153738629291SGao Xiang 15380615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 153947e4937aSGao Xiang set_page_private(page, (unsigned long)head); 154047e4937aSGao Xiang head = page; 154147e4937aSGao Xiang } 154247e4937aSGao Xiang 154347e4937aSGao Xiang while (head) { 154447e4937aSGao Xiang struct page *page = head; 154547e4937aSGao Xiang int err; 154647e4937aSGao Xiang 154747e4937aSGao Xiang /* traversal in reverse order */ 154847e4937aSGao Xiang head = (void *)page_private(page); 154947e4937aSGao Xiang 15501825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1551a5876e24SGao Xiang if (err) 15524f761fa2SGao Xiang erofs_err(inode->i_sb, 15534f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 15544f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 155547e4937aSGao Xiang put_page(page); 155647e4937aSGao Xiang } 155738629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 15585c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 155947e4937aSGao Xiang 156038629291SGao Xiang z_erofs_runqueue(inode->i_sb, &f, &pagepool, 156140452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 156209c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1563eaa9172aSGao Xiang erofs_release_pages(&pagepool); 156447e4937aSGao Xiang } 156547e4937aSGao Xiang 15660c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1567*a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 15680615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 156947e4937aSGao Xiang }; 1570