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 { 859f6cc76eSGao Xiang int i; 869f6cc76eSGao Xiang 879f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 889f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 899f6cc76eSGao Xiang 909f6cc76eSGao Xiang if (pcl->pclusterpages > pcs->maxpages) 919f6cc76eSGao Xiang continue; 929f6cc76eSGao Xiang 939f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 949f6cc76eSGao Xiang return; 959f6cc76eSGao Xiang } 969f6cc76eSGao Xiang DBG_BUGON(1); 979f6cc76eSGao Xiang } 989f6cc76eSGao Xiang 999f6cc76eSGao Xiang /* 10047e4937aSGao Xiang * a compressed_pages[] placeholder in order to avoid 10147e4937aSGao Xiang * being filled with file pages for in-place decompression. 10247e4937aSGao Xiang */ 10347e4937aSGao Xiang #define PAGE_UNALLOCATED ((void *)0x5F0E4B1D) 10447e4937aSGao Xiang 10547e4937aSGao Xiang /* how to allocate cached pages for a pcluster */ 10647e4937aSGao Xiang enum z_erofs_cache_alloctype { 10747e4937aSGao Xiang DONTALLOC, /* don't allocate any cached pages */ 10847e4937aSGao Xiang DELAYEDALLOC, /* delayed allocation (at the time of submitting io) */ 1091825c8d7SGao Xiang /* 1101825c8d7SGao Xiang * try to use cached I/O if page allocation succeeds or fallback 1111825c8d7SGao Xiang * to in-place I/O instead to avoid any direct reclaim. 1121825c8d7SGao Xiang */ 1131825c8d7SGao Xiang TRYALLOC, 11447e4937aSGao Xiang }; 11547e4937aSGao Xiang 11647e4937aSGao Xiang /* 11747e4937aSGao Xiang * tagged pointer with 1-bit tag for all compressed pages 11847e4937aSGao Xiang * tag 0 - the page is just found with an extra page reference 11947e4937aSGao Xiang */ 12047e4937aSGao Xiang typedef tagptr1_t compressed_page_t; 12147e4937aSGao Xiang 12247e4937aSGao Xiang #define tag_compressed_page_justfound(page) \ 12347e4937aSGao Xiang tagptr_fold(compressed_page_t, page, 1) 12447e4937aSGao Xiang 12547e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 12647e4937aSGao Xiang 12747e4937aSGao Xiang void z_erofs_exit_zip_subsystem(void) 12847e4937aSGao Xiang { 12947e4937aSGao Xiang destroy_workqueue(z_erofs_workqueue); 1309f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 13147e4937aSGao Xiang } 13247e4937aSGao Xiang 13399634bf3SGao Xiang static inline int z_erofs_init_workqueue(void) 13447e4937aSGao Xiang { 13547e4937aSGao Xiang const unsigned int onlinecpus = num_possible_cpus(); 13647e4937aSGao Xiang 13747e4937aSGao Xiang /* 13847e4937aSGao Xiang * no need to spawn too many threads, limiting threads could minimum 13947e4937aSGao Xiang * scheduling overhead, perhaps per-CPU threads should be better? 14047e4937aSGao Xiang */ 1410e62ea33SGao Xiang z_erofs_workqueue = alloc_workqueue("erofs_unzipd", 1420e62ea33SGao Xiang WQ_UNBOUND | WQ_HIGHPRI, 14347e4937aSGao Xiang onlinecpus + onlinecpus / 4); 14447e4937aSGao Xiang return z_erofs_workqueue ? 0 : -ENOMEM; 14547e4937aSGao Xiang } 14647e4937aSGao Xiang 14747e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 14847e4937aSGao Xiang { 1499f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 15047e4937aSGao Xiang 1519f6cc76eSGao Xiang if (err) 1529f6cc76eSGao Xiang return err; 1539f6cc76eSGao Xiang err = z_erofs_init_workqueue(); 1549f6cc76eSGao Xiang if (err) 1559f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 1569f6cc76eSGao Xiang return err; 15747e4937aSGao Xiang } 15847e4937aSGao Xiang 15947e4937aSGao Xiang enum z_erofs_collectmode { 16047e4937aSGao Xiang COLLECT_SECONDARY, 16147e4937aSGao Xiang COLLECT_PRIMARY, 16247e4937aSGao Xiang /* 16347e4937aSGao Xiang * The current collection was the tail of an exist chain, in addition 16447e4937aSGao Xiang * that the previous processed chained collections are all decided to 16547e4937aSGao Xiang * be hooked up to it. 16647e4937aSGao Xiang * A new chain will be created for the remaining collections which are 16747e4937aSGao Xiang * not processed yet, therefore different from COLLECT_PRIMARY_FOLLOWED, 16847e4937aSGao Xiang * the next collection cannot reuse the whole page safely in 16947e4937aSGao Xiang * the following scenario: 17047e4937aSGao Xiang * ________________________________________________________________ 17147e4937aSGao Xiang * | tail (partial) page | head (partial) page | 17247e4937aSGao Xiang * | (belongs to the next cl) | (belongs to the current cl) | 17347e4937aSGao Xiang * |_______PRIMARY_FOLLOWED_______|________PRIMARY_HOOKED___________| 17447e4937aSGao Xiang */ 17547e4937aSGao Xiang COLLECT_PRIMARY_HOOKED, 1760b964600SGao Xiang /* 1770b964600SGao Xiang * a weak form of COLLECT_PRIMARY_FOLLOWED, the difference is that it 1780b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 1790b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 1800b964600SGao Xiang * pagevec) since it can be directly decoded without I/O submission. 1810b964600SGao Xiang */ 18247e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED_NOINPLACE, 18347e4937aSGao Xiang /* 18447e4937aSGao Xiang * The current collection has been linked with the owned chain, and 18547e4937aSGao Xiang * could also be linked with the remaining collections, which means 18647e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 18747e4937aSGao Xiang * the current collection can safely use the whole page (since 18847e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 18947e4937aSGao Xiang * illustrated below: 19047e4937aSGao Xiang * ________________________________________________________________ 19147e4937aSGao Xiang * | tail (partial) page | head (partial) page | 19247e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 19347e4937aSGao Xiang * | PRIMARY_FOLLOWED or | | 19447e4937aSGao Xiang * |_____PRIMARY_HOOKED___|____________PRIMARY_FOLLOWED____________| 19547e4937aSGao Xiang * 19647e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 19747e4937aSGao Xiang */ 19847e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED, 19947e4937aSGao Xiang }; 20047e4937aSGao Xiang 20147e4937aSGao Xiang struct z_erofs_collector { 20247e4937aSGao Xiang struct z_erofs_pagevec_ctor vector; 20347e4937aSGao Xiang 20447e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 20547e4937aSGao Xiang struct z_erofs_collection *cl; 20681382f5fSGao Xiang /* a pointer used to pick up inplace I/O pages */ 20781382f5fSGao Xiang struct page **icpage_ptr; 20847e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 20947e4937aSGao Xiang 21047e4937aSGao Xiang enum z_erofs_collectmode mode; 21147e4937aSGao Xiang }; 21247e4937aSGao Xiang 21347e4937aSGao Xiang struct z_erofs_decompress_frontend { 21447e4937aSGao Xiang struct inode *const inode; 21547e4937aSGao Xiang 21647e4937aSGao Xiang struct z_erofs_collector clt; 21747e4937aSGao Xiang struct erofs_map_blocks map; 21847e4937aSGao Xiang 2196ea5aad3SGao Xiang bool readahead; 22047e4937aSGao Xiang /* used for applying cache strategy on the fly */ 22147e4937aSGao Xiang bool backmost; 22247e4937aSGao Xiang erofs_off_t headoffset; 22347e4937aSGao Xiang }; 22447e4937aSGao Xiang 22547e4937aSGao Xiang #define COLLECTOR_INIT() { \ 22647e4937aSGao Xiang .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 22747e4937aSGao Xiang .mode = COLLECT_PRIMARY_FOLLOWED } 22847e4937aSGao Xiang 22947e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 23047e4937aSGao Xiang .inode = __i, .clt = COLLECTOR_INIT(), \ 23147e4937aSGao Xiang .backmost = true, } 23247e4937aSGao Xiang 23347e4937aSGao Xiang static struct page *z_pagemap_global[Z_EROFS_VMAP_GLOBAL_PAGES]; 23447e4937aSGao Xiang static DEFINE_MUTEX(z_pagemap_global_lock); 23547e4937aSGao Xiang 23647e4937aSGao Xiang static void preload_compressed_pages(struct z_erofs_collector *clt, 23747e4937aSGao Xiang struct address_space *mc, 2381825c8d7SGao Xiang enum z_erofs_cache_alloctype type, 239*eaa9172aSGao Xiang struct page **pagepool) 24047e4937aSGao Xiang { 24181382f5fSGao Xiang struct z_erofs_pcluster *pcl = clt->pcl; 24247e4937aSGao Xiang bool standalone = true; 2431825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 2441825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 24581382f5fSGao Xiang struct page **pages; 24681382f5fSGao Xiang pgoff_t index; 24747e4937aSGao Xiang 24847e4937aSGao Xiang if (clt->mode < COLLECT_PRIMARY_FOLLOWED) 24947e4937aSGao Xiang return; 25047e4937aSGao Xiang 25181382f5fSGao Xiang pages = pcl->compressed_pages; 25281382f5fSGao Xiang index = pcl->obj.index; 25381382f5fSGao Xiang for (; index < pcl->obj.index + pcl->pclusterpages; ++index, ++pages) { 25447e4937aSGao Xiang struct page *page; 25547e4937aSGao Xiang compressed_page_t t; 2561825c8d7SGao Xiang struct page *newpage = NULL; 25747e4937aSGao Xiang 25847e4937aSGao Xiang /* the compressed page was loaded before */ 25947e4937aSGao Xiang if (READ_ONCE(*pages)) 26047e4937aSGao Xiang continue; 26147e4937aSGao Xiang 26247e4937aSGao Xiang page = find_get_page(mc, index); 26347e4937aSGao Xiang 26447e4937aSGao Xiang if (page) { 26547e4937aSGao Xiang t = tag_compressed_page_justfound(page); 2660b964600SGao Xiang } else { 2670b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 2680b964600SGao Xiang standalone = false; 2690b964600SGao Xiang switch (type) { 2700b964600SGao Xiang case DELAYEDALLOC: 2710b964600SGao Xiang t = tagptr_init(compressed_page_t, 2720b964600SGao Xiang PAGE_UNALLOCATED); 2730b964600SGao Xiang break; 2740b964600SGao Xiang case TRYALLOC: 2751825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 2761825c8d7SGao Xiang if (!newpage) 27747e4937aSGao Xiang continue; 2780b964600SGao Xiang set_page_private(newpage, 2790b964600SGao Xiang Z_EROFS_PREALLOCATED_PAGE); 2800b964600SGao Xiang t = tag_compressed_page_justfound(newpage); 2810b964600SGao Xiang break; 2820b964600SGao Xiang default: /* DONTALLOC */ 2830b964600SGao Xiang continue; 2840b964600SGao Xiang } 28547e4937aSGao Xiang } 28647e4937aSGao Xiang 28747e4937aSGao Xiang if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t))) 28847e4937aSGao Xiang continue; 28947e4937aSGao Xiang 290*eaa9172aSGao Xiang if (page) 29147e4937aSGao Xiang put_page(page); 292*eaa9172aSGao Xiang else if (newpage) 293*eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 29447e4937aSGao Xiang } 29547e4937aSGao Xiang 2960b964600SGao Xiang /* 2970b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 2980b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 2990b964600SGao Xiang */ 3000b964600SGao Xiang if (standalone) 30147e4937aSGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 30247e4937aSGao Xiang } 30347e4937aSGao Xiang 30447e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 30547e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 30647e4937aSGao Xiang struct erofs_workgroup *grp) 30747e4937aSGao Xiang { 30847e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 30947e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 31047e4937aSGao Xiang int i; 31147e4937aSGao Xiang 31247e4937aSGao Xiang /* 31347e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 31447e4937aSGao Xiang * therefore no need to worry about available decompression users. 31547e4937aSGao Xiang */ 3169f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 31747e4937aSGao Xiang struct page *page = pcl->compressed_pages[i]; 31847e4937aSGao Xiang 31947e4937aSGao Xiang if (!page) 32047e4937aSGao Xiang continue; 32147e4937aSGao Xiang 32247e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 32347e4937aSGao Xiang if (!trylock_page(page)) 32447e4937aSGao Xiang return -EBUSY; 32547e4937aSGao Xiang 326f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 32747e4937aSGao Xiang continue; 32847e4937aSGao Xiang 32947e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 33047e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 3316aaa7b06SGao Xiang detach_page_private(page); 33247e4937aSGao Xiang unlock_page(page); 33347e4937aSGao Xiang } 33447e4937aSGao Xiang return 0; 33547e4937aSGao Xiang } 33647e4937aSGao Xiang 337d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 33847e4937aSGao Xiang { 33947e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 34047e4937aSGao Xiang int ret = 0; /* 0 - busy */ 34147e4937aSGao Xiang 34247e4937aSGao Xiang if (erofs_workgroup_try_to_freeze(&pcl->obj, 1)) { 34347e4937aSGao Xiang unsigned int i; 34447e4937aSGao Xiang 3459f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 34647e4937aSGao Xiang if (pcl->compressed_pages[i] == page) { 34747e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 34847e4937aSGao Xiang ret = 1; 34947e4937aSGao Xiang break; 35047e4937aSGao Xiang } 35147e4937aSGao Xiang } 35247e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 35347e4937aSGao Xiang 3546aaa7b06SGao Xiang if (ret) 3556aaa7b06SGao Xiang detach_page_private(page); 35647e4937aSGao Xiang } 35747e4937aSGao Xiang return ret; 35847e4937aSGao Xiang } 35947e4937aSGao Xiang 36047e4937aSGao Xiang /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 36181382f5fSGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_collector *clt, 36247e4937aSGao Xiang struct page *page) 36347e4937aSGao Xiang { 36447e4937aSGao Xiang struct z_erofs_pcluster *const pcl = clt->pcl; 36547e4937aSGao Xiang 36681382f5fSGao Xiang while (clt->icpage_ptr > pcl->compressed_pages) 36781382f5fSGao Xiang if (!cmpxchg(--clt->icpage_ptr, NULL, page)) 36847e4937aSGao Xiang return true; 36947e4937aSGao Xiang return false; 37047e4937aSGao Xiang } 37147e4937aSGao Xiang 37247e4937aSGao Xiang /* callers must be with collection lock held */ 37347e4937aSGao Xiang static int z_erofs_attach_page(struct z_erofs_collector *clt, 37447e4937aSGao Xiang struct page *page, 37547e4937aSGao Xiang enum z_erofs_page_type type) 37647e4937aSGao Xiang { 37747e4937aSGao Xiang int ret; 37847e4937aSGao Xiang 37947e4937aSGao Xiang /* give priority for inplaceio */ 38047e4937aSGao Xiang if (clt->mode >= COLLECT_PRIMARY && 38147e4937aSGao Xiang type == Z_EROFS_PAGE_TYPE_EXCLUSIVE && 38299634bf3SGao Xiang z_erofs_try_inplace_io(clt, page)) 38347e4937aSGao Xiang return 0; 38447e4937aSGao Xiang 3857dea3de7SYue Hu ret = z_erofs_pagevec_enqueue(&clt->vector, page, type); 38647e4937aSGao Xiang clt->cl->vcnt += (unsigned int)ret; 38747e4937aSGao Xiang 38847e4937aSGao Xiang return ret ? 0 : -EAGAIN; 38947e4937aSGao Xiang } 39047e4937aSGao Xiang 391473e15b0SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_collector *clt) 39247e4937aSGao Xiang { 393473e15b0SGao Xiang struct z_erofs_pcluster *pcl = clt->pcl; 394473e15b0SGao Xiang z_erofs_next_pcluster_t *owned_head = &clt->owned_head; 39547e4937aSGao Xiang 396473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 397473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 398473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 39947e4937aSGao Xiang *owned_head = &pcl->next; 400473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 401473e15b0SGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED; 402473e15b0SGao Xiang return; 403473e15b0SGao Xiang } 404473e15b0SGao Xiang 40547e4937aSGao Xiang /* 406473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 407473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 40847e4937aSGao Xiang */ 40947e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 410473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 41147e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 412473e15b0SGao Xiang clt->mode = COLLECT_PRIMARY_HOOKED; 413473e15b0SGao Xiang clt->tailpcl = NULL; 414473e15b0SGao Xiang return; 41547e4937aSGao Xiang } 416473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 417473e15b0SGao Xiang clt->mode = COLLECT_PRIMARY; 41847e4937aSGao Xiang } 41947e4937aSGao Xiang 4209e579fc1SGao Xiang static int z_erofs_lookup_collection(struct z_erofs_collector *clt, 42147e4937aSGao Xiang struct inode *inode, 42247e4937aSGao Xiang struct erofs_map_blocks *map) 42347e4937aSGao Xiang { 42464094a04SGao Xiang struct z_erofs_pcluster *pcl = clt->pcl; 42547e4937aSGao Xiang struct z_erofs_collection *cl; 42647e4937aSGao Xiang unsigned int length; 42747e4937aSGao Xiang 42864094a04SGao Xiang /* to avoid unexpected loop formed by corrupted images */ 42947e4937aSGao Xiang if (clt->owned_head == &pcl->next || pcl == clt->tailpcl) { 43047e4937aSGao Xiang DBG_BUGON(1); 4319e579fc1SGao Xiang return -EFSCORRUPTED; 43247e4937aSGao Xiang } 43347e4937aSGao Xiang 43447e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 4358d8a09b0SGao Xiang if (cl->pageofs != (map->m_la & ~PAGE_MASK)) { 43647e4937aSGao Xiang DBG_BUGON(1); 4379e579fc1SGao Xiang return -EFSCORRUPTED; 43847e4937aSGao Xiang } 43947e4937aSGao Xiang 44047e4937aSGao Xiang length = READ_ONCE(pcl->length); 44147e4937aSGao Xiang if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 44247e4937aSGao Xiang if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 44347e4937aSGao Xiang DBG_BUGON(1); 4449e579fc1SGao Xiang return -EFSCORRUPTED; 44547e4937aSGao Xiang } 44647e4937aSGao Xiang } else { 44747e4937aSGao Xiang unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 44847e4937aSGao Xiang 44947e4937aSGao Xiang if (map->m_flags & EROFS_MAP_FULL_MAPPED) 45047e4937aSGao Xiang llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 45147e4937aSGao Xiang 45247e4937aSGao Xiang while (llen > length && 45347e4937aSGao Xiang length != cmpxchg_relaxed(&pcl->length, length, llen)) { 45447e4937aSGao Xiang cpu_relax(); 45547e4937aSGao Xiang length = READ_ONCE(pcl->length); 45647e4937aSGao Xiang } 45747e4937aSGao Xiang } 45847e4937aSGao Xiang mutex_lock(&cl->lock); 45947e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 46047e4937aSGao Xiang if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 46147e4937aSGao Xiang clt->tailpcl = pcl; 462473e15b0SGao Xiang 463473e15b0SGao Xiang z_erofs_try_to_claim_pcluster(clt); 46447e4937aSGao Xiang clt->cl = cl; 4659e579fc1SGao Xiang return 0; 46647e4937aSGao Xiang } 46747e4937aSGao Xiang 4689e579fc1SGao Xiang static int z_erofs_register_collection(struct z_erofs_collector *clt, 46947e4937aSGao Xiang struct inode *inode, 47047e4937aSGao Xiang struct erofs_map_blocks *map) 47147e4937aSGao Xiang { 47247e4937aSGao Xiang struct z_erofs_pcluster *pcl; 47347e4937aSGao Xiang struct z_erofs_collection *cl; 47464094a04SGao Xiang struct erofs_workgroup *grp; 47547e4937aSGao Xiang int err; 47647e4937aSGao Xiang 4778f899262SGao Xiang if (!(map->m_flags & EROFS_MAP_ENCODED)) { 4788f899262SGao Xiang DBG_BUGON(1); 4798f899262SGao Xiang return -EFSCORRUPTED; 4808f899262SGao Xiang } 4818f899262SGao Xiang 4829f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 4839f6cc76eSGao Xiang pcl = z_erofs_alloc_pcluster(map->m_plen >> PAGE_SHIFT); 4849f6cc76eSGao Xiang if (IS_ERR(pcl)) 4859f6cc76eSGao Xiang return PTR_ERR(pcl); 48647e4937aSGao Xiang 48764094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 48847e4937aSGao Xiang pcl->obj.index = map->m_pa >> PAGE_SHIFT; 4898f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 49047e4937aSGao Xiang pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 49147e4937aSGao Xiang (map->m_flags & EROFS_MAP_FULL_MAPPED ? 49247e4937aSGao Xiang Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 49347e4937aSGao Xiang 49447e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 49547e4937aSGao Xiang pcl->next = clt->owned_head; 49647e4937aSGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED; 49747e4937aSGao Xiang 49847e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 49947e4937aSGao Xiang cl->pageofs = map->m_la & ~PAGE_MASK; 50047e4937aSGao Xiang 50147e4937aSGao Xiang /* 50247e4937aSGao Xiang * lock all primary followed works before visible to others 50347e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 50447e4937aSGao Xiang */ 5059f6cc76eSGao Xiang mutex_init(&cl->lock); 50664094a04SGao Xiang DBG_BUGON(!mutex_trylock(&cl->lock)); 50747e4937aSGao Xiang 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) { 51564094a04SGao Xiang clt->pcl = container_of(grp, struct z_erofs_pcluster, obj); 51664094a04SGao Xiang err = -EEXIST; 51764094a04SGao Xiang goto err_out; 51847e4937aSGao Xiang } 51947e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 52047e4937aSGao Xiang if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 52147e4937aSGao Xiang clt->tailpcl = pcl; 52247e4937aSGao Xiang clt->owned_head = &pcl->next; 52347e4937aSGao Xiang clt->pcl = pcl; 52447e4937aSGao Xiang clt->cl = cl; 5259e579fc1SGao Xiang return 0; 52664094a04SGao Xiang 52764094a04SGao Xiang err_out: 52864094a04SGao Xiang mutex_unlock(&cl->lock); 5299f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 53064094a04SGao Xiang return err; 53147e4937aSGao Xiang } 53247e4937aSGao Xiang 53347e4937aSGao Xiang static int z_erofs_collector_begin(struct z_erofs_collector *clt, 53447e4937aSGao Xiang struct inode *inode, 53547e4937aSGao Xiang struct erofs_map_blocks *map) 53647e4937aSGao Xiang { 53764094a04SGao Xiang struct erofs_workgroup *grp; 5389e579fc1SGao Xiang int ret; 53947e4937aSGao Xiang 54047e4937aSGao Xiang DBG_BUGON(clt->cl); 54147e4937aSGao Xiang 54247e4937aSGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous collection */ 54347e4937aSGao Xiang DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_NIL); 54447e4937aSGao Xiang DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 54547e4937aSGao Xiang 54647e4937aSGao Xiang if (!PAGE_ALIGNED(map->m_pa)) { 54747e4937aSGao Xiang DBG_BUGON(1); 54847e4937aSGao Xiang return -EINVAL; 54947e4937aSGao Xiang } 55047e4937aSGao Xiang 55164094a04SGao Xiang grp = erofs_find_workgroup(inode->i_sb, map->m_pa >> PAGE_SHIFT); 55264094a04SGao Xiang if (grp) { 55364094a04SGao Xiang clt->pcl = container_of(grp, struct z_erofs_pcluster, obj); 55464094a04SGao Xiang } else { 5559e579fc1SGao Xiang ret = z_erofs_register_collection(clt, inode, map); 55647e4937aSGao Xiang 55764094a04SGao Xiang if (!ret) 55864094a04SGao Xiang goto out; 55964094a04SGao Xiang if (ret != -EEXIST) 5609e579fc1SGao Xiang return ret; 56164094a04SGao Xiang } 56247e4937aSGao Xiang 56364094a04SGao Xiang ret = z_erofs_lookup_collection(clt, inode, map); 56464094a04SGao Xiang if (ret) { 56564094a04SGao Xiang erofs_workgroup_put(&clt->pcl->obj); 56664094a04SGao Xiang return ret; 56764094a04SGao Xiang } 56864094a04SGao Xiang 56964094a04SGao Xiang out: 57047e4937aSGao Xiang z_erofs_pagevec_ctor_init(&clt->vector, Z_EROFS_NR_INLINE_PAGEVECS, 5719e579fc1SGao Xiang clt->cl->pagevec, clt->cl->vcnt); 57247e4937aSGao Xiang 57381382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 57481382f5fSGao Xiang clt->icpage_ptr = clt->pcl->compressed_pages + clt->pcl->pclusterpages; 57547e4937aSGao Xiang return 0; 57647e4937aSGao Xiang } 57747e4937aSGao Xiang 57847e4937aSGao Xiang /* 57947e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 58047e4937aSGao Xiang * only after a RCU grace period. 58147e4937aSGao Xiang */ 58247e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 58347e4937aSGao Xiang { 58447e4937aSGao Xiang struct z_erofs_collection *const cl = 58547e4937aSGao Xiang container_of(head, struct z_erofs_collection, rcu); 58647e4937aSGao Xiang 5879f6cc76eSGao Xiang z_erofs_free_pcluster(container_of(cl, struct z_erofs_pcluster, 58847e4937aSGao Xiang primary_collection)); 58947e4937aSGao Xiang } 59047e4937aSGao Xiang 59147e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 59247e4937aSGao Xiang { 59347e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 59447e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 59547e4937aSGao Xiang struct z_erofs_collection *const cl = z_erofs_primarycollection(pcl); 59647e4937aSGao Xiang 59747e4937aSGao Xiang call_rcu(&cl->rcu, z_erofs_rcu_callback); 59847e4937aSGao Xiang } 59947e4937aSGao Xiang 60047e4937aSGao Xiang static void z_erofs_collection_put(struct z_erofs_collection *cl) 60147e4937aSGao Xiang { 60247e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 60347e4937aSGao Xiang container_of(cl, struct z_erofs_pcluster, primary_collection); 60447e4937aSGao Xiang 60547e4937aSGao Xiang erofs_workgroup_put(&pcl->obj); 60647e4937aSGao Xiang } 60747e4937aSGao Xiang 60847e4937aSGao Xiang static bool z_erofs_collector_end(struct z_erofs_collector *clt) 60947e4937aSGao Xiang { 61047e4937aSGao Xiang struct z_erofs_collection *cl = clt->cl; 61147e4937aSGao Xiang 61247e4937aSGao Xiang if (!cl) 61347e4937aSGao Xiang return false; 61447e4937aSGao Xiang 61547e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&clt->vector, false); 61647e4937aSGao Xiang mutex_unlock(&cl->lock); 61747e4937aSGao Xiang 61847e4937aSGao Xiang /* 61947e4937aSGao Xiang * if all pending pages are added, don't hold its reference 62047e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 62147e4937aSGao Xiang */ 62247e4937aSGao Xiang if (clt->mode < COLLECT_PRIMARY_FOLLOWED_NOINPLACE) 62347e4937aSGao Xiang z_erofs_collection_put(cl); 62447e4937aSGao Xiang 62547e4937aSGao Xiang clt->cl = NULL; 62647e4937aSGao Xiang return true; 62747e4937aSGao Xiang } 62847e4937aSGao Xiang 62947e4937aSGao Xiang static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe, 63047e4937aSGao Xiang unsigned int cachestrategy, 63147e4937aSGao Xiang erofs_off_t la) 63247e4937aSGao Xiang { 63347e4937aSGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 63447e4937aSGao Xiang return false; 63547e4937aSGao Xiang 63647e4937aSGao Xiang if (fe->backmost) 63747e4937aSGao Xiang return true; 63847e4937aSGao Xiang 63947e4937aSGao Xiang return cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 64047e4937aSGao Xiang la < fe->headoffset; 64147e4937aSGao Xiang } 64247e4937aSGao Xiang 64347e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 644*eaa9172aSGao Xiang struct page *page, struct page **pagepool) 64547e4937aSGao Xiang { 64647e4937aSGao Xiang struct inode *const inode = fe->inode; 647bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 64847e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 64947e4937aSGao Xiang struct z_erofs_collector *const clt = &fe->clt; 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) */ 67047e4937aSGao Xiang if (!clt->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 67847e4937aSGao Xiang if (z_erofs_collector_end(clt)) 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 69147e4937aSGao Xiang err = z_erofs_collector_begin(clt, inode, map); 6928d8a09b0SGao Xiang if (err) 69347e4937aSGao Xiang goto err_out; 69447e4937aSGao Xiang 69547e4937aSGao Xiang /* preload all compressed pages (maybe downgrade role if necessary) */ 696e6242465SGao Xiang if (should_alloc_managed_pages(fe, sbi->opt.cache_strategy, map->m_la)) 6971825c8d7SGao Xiang cache_strategy = TRYALLOC; 69847e4937aSGao Xiang else 69947e4937aSGao Xiang cache_strategy = DONTALLOC; 70047e4937aSGao Xiang 7011825c8d7SGao Xiang preload_compressed_pages(clt, MNGD_MAPPING(sbi), 7021825c8d7SGao Xiang cache_strategy, pagepool); 70347e4937aSGao Xiang 70447e4937aSGao Xiang hitted: 705dc76ea8cSGao Xiang /* 706dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 707dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 708dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 709dc76ea8cSGao Xiang * for inplace I/O or pagevec (should be processed in strict order.) 710dc76ea8cSGao Xiang */ 711dc76ea8cSGao Xiang tight &= (clt->mode >= COLLECT_PRIMARY_HOOKED && 712dc76ea8cSGao Xiang clt->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 713dc76ea8cSGao Xiang 71447e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 7158d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 71647e4937aSGao Xiang zero_user_segment(page, cur, end); 71747e4937aSGao Xiang goto next_part; 71847e4937aSGao Xiang } 71947e4937aSGao Xiang 72047e4937aSGao Xiang /* let's derive page type */ 72147e4937aSGao Xiang page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 72247e4937aSGao Xiang (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 72347e4937aSGao Xiang (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 72447e4937aSGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 72547e4937aSGao Xiang 72647e4937aSGao Xiang if (cur) 72747e4937aSGao Xiang tight &= (clt->mode >= COLLECT_PRIMARY_FOLLOWED); 72847e4937aSGao Xiang 72947e4937aSGao Xiang retry: 73047e4937aSGao Xiang err = z_erofs_attach_page(clt, page, page_type); 7316aaa7b06SGao Xiang /* should allocate an additional short-lived page for pagevec */ 73247e4937aSGao Xiang if (err == -EAGAIN) { 73347e4937aSGao Xiang struct page *const newpage = 734e3f78d5eSChao Yu alloc_page(GFP_NOFS | __GFP_NOFAIL); 73547e4937aSGao Xiang 7366aaa7b06SGao Xiang set_page_private(newpage, Z_EROFS_SHORTLIVED_PAGE); 73747e4937aSGao Xiang err = z_erofs_attach_page(clt, newpage, 73847e4937aSGao Xiang Z_EROFS_PAGE_TYPE_EXCLUSIVE); 7398d8a09b0SGao Xiang if (!err) 74047e4937aSGao Xiang goto retry; 74147e4937aSGao Xiang } 74247e4937aSGao Xiang 7438d8a09b0SGao Xiang if (err) 74447e4937aSGao Xiang goto err_out; 74547e4937aSGao Xiang 74647e4937aSGao Xiang index = page->index - (map->m_la >> PAGE_SHIFT); 74747e4937aSGao Xiang 74847e4937aSGao Xiang z_erofs_onlinepage_fixup(page, index, true); 74947e4937aSGao Xiang 75047e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 75147e4937aSGao Xiang ++spiltted; 75247e4937aSGao Xiang /* also update nr_pages */ 75347e4937aSGao Xiang clt->cl->nr_pages = max_t(pgoff_t, clt->cl->nr_pages, index + 1); 75447e4937aSGao Xiang next_part: 75547e4937aSGao Xiang /* can be used for verification */ 75647e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 75747e4937aSGao Xiang 75847e4937aSGao Xiang end = cur; 75947e4937aSGao Xiang if (end > 0) 76047e4937aSGao Xiang goto repeat; 76147e4937aSGao Xiang 76247e4937aSGao Xiang out: 76347e4937aSGao Xiang z_erofs_onlinepage_endio(page); 76447e4937aSGao Xiang 7654f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 76647e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 76747e4937aSGao Xiang return err; 76847e4937aSGao Xiang 76947e4937aSGao Xiang /* if some error occurred while processing this page */ 77047e4937aSGao Xiang err_out: 77147e4937aSGao Xiang SetPageError(page); 77247e4937aSGao Xiang goto out; 77347e4937aSGao Xiang } 77447e4937aSGao Xiang 775648f2de0SHuang Jianan static void z_erofs_decompressqueue_work(struct work_struct *work); 776a4b1fab1SGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 777a4b1fab1SGao Xiang bool sync, int bios) 77847e4937aSGao Xiang { 77930048cdaSHuang Jianan struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 78030048cdaSHuang Jianan 781a4b1fab1SGao Xiang /* wake up the caller thread for sync decompression */ 782a4b1fab1SGao Xiang if (sync) { 78347e4937aSGao Xiang unsigned long flags; 78447e4937aSGao Xiang 78547e4937aSGao Xiang spin_lock_irqsave(&io->u.wait.lock, flags); 78647e4937aSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 78747e4937aSGao Xiang wake_up_locked(&io->u.wait); 78847e4937aSGao Xiang spin_unlock_irqrestore(&io->u.wait.lock, flags); 78947e4937aSGao Xiang return; 79047e4937aSGao Xiang } 79147e4937aSGao Xiang 792648f2de0SHuang Jianan if (atomic_add_return(bios, &io->pending_bios)) 793648f2de0SHuang Jianan return; 79430048cdaSHuang Jianan /* Use workqueue and sync decompression for atomic contexts only */ 795648f2de0SHuang Jianan if (in_atomic() || irqs_disabled()) { 79647e4937aSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 797e6242465SGao Xiang sbi->opt.readahead_sync_decompress = true; 798648f2de0SHuang Jianan return; 799648f2de0SHuang Jianan } 800648f2de0SHuang Jianan z_erofs_decompressqueue_work(&io->u.work); 80147e4937aSGao Xiang } 80247e4937aSGao Xiang 8036aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 8046aaa7b06SGao Xiang { 8056aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 8066aaa7b06SGao Xiang } 8076aaa7b06SGao Xiang 8080c638f70SGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 80947e4937aSGao Xiang { 810a4b1fab1SGao Xiang tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 811a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 81247e4937aSGao Xiang blk_status_t err = bio->bi_status; 81347e4937aSGao Xiang struct bio_vec *bvec; 81447e4937aSGao Xiang struct bvec_iter_all iter_all; 81547e4937aSGao Xiang 81647e4937aSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 81747e4937aSGao Xiang struct page *page = bvec->bv_page; 81847e4937aSGao Xiang 81947e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 8206aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 82147e4937aSGao Xiang 8228d8a09b0SGao Xiang if (err) 82347e4937aSGao Xiang SetPageError(page); 82447e4937aSGao Xiang 825a4b1fab1SGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 826a4b1fab1SGao Xiang if (!err) 827a4b1fab1SGao Xiang SetPageUptodate(page); 82847e4937aSGao Xiang unlock_page(page); 82947e4937aSGao Xiang } 830a4b1fab1SGao Xiang } 831a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 83247e4937aSGao Xiang bio_put(bio); 83347e4937aSGao Xiang } 83447e4937aSGao Xiang 83547e4937aSGao Xiang static int z_erofs_decompress_pcluster(struct super_block *sb, 83647e4937aSGao Xiang struct z_erofs_pcluster *pcl, 837*eaa9172aSGao Xiang struct page **pagepool) 83847e4937aSGao Xiang { 83947e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 84047e4937aSGao Xiang struct z_erofs_pagevec_ctor ctor; 8419f6cc76eSGao Xiang unsigned int i, inputsize, outputsize, llen, nr_pages; 84247e4937aSGao Xiang struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 84347e4937aSGao Xiang struct page **pages, **compressed_pages, *page; 84447e4937aSGao Xiang 84547e4937aSGao Xiang enum z_erofs_page_type page_type; 84647e4937aSGao Xiang bool overlapped, partial; 84747e4937aSGao Xiang struct z_erofs_collection *cl; 84847e4937aSGao Xiang int err; 84947e4937aSGao Xiang 85047e4937aSGao Xiang might_sleep(); 85147e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 85247e4937aSGao Xiang DBG_BUGON(!READ_ONCE(cl->nr_pages)); 85347e4937aSGao Xiang 85447e4937aSGao Xiang mutex_lock(&cl->lock); 85547e4937aSGao Xiang nr_pages = cl->nr_pages; 85647e4937aSGao Xiang 8578d8a09b0SGao Xiang if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 85847e4937aSGao Xiang pages = pages_onstack; 85947e4937aSGao Xiang } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 86047e4937aSGao Xiang mutex_trylock(&z_pagemap_global_lock)) { 86147e4937aSGao Xiang pages = z_pagemap_global; 86247e4937aSGao Xiang } else { 86347e4937aSGao Xiang gfp_t gfp_flags = GFP_KERNEL; 86447e4937aSGao Xiang 86547e4937aSGao Xiang if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 86647e4937aSGao Xiang gfp_flags |= __GFP_NOFAIL; 86747e4937aSGao Xiang 86847e4937aSGao Xiang pages = kvmalloc_array(nr_pages, sizeof(struct page *), 86947e4937aSGao Xiang gfp_flags); 87047e4937aSGao Xiang 87147e4937aSGao Xiang /* fallback to global pagemap for the lowmem scenario */ 8728d8a09b0SGao Xiang if (!pages) { 87347e4937aSGao Xiang mutex_lock(&z_pagemap_global_lock); 87447e4937aSGao Xiang pages = z_pagemap_global; 87547e4937aSGao Xiang } 87647e4937aSGao Xiang } 87747e4937aSGao Xiang 87847e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) 87947e4937aSGao Xiang pages[i] = NULL; 88047e4937aSGao Xiang 88147e4937aSGao Xiang err = 0; 88247e4937aSGao Xiang z_erofs_pagevec_ctor_init(&ctor, Z_EROFS_NR_INLINE_PAGEVECS, 88347e4937aSGao Xiang cl->pagevec, 0); 88447e4937aSGao Xiang 88547e4937aSGao Xiang for (i = 0; i < cl->vcnt; ++i) { 88647e4937aSGao Xiang unsigned int pagenr; 88747e4937aSGao Xiang 88847e4937aSGao Xiang page = z_erofs_pagevec_dequeue(&ctor, &page_type); 88947e4937aSGao Xiang 89047e4937aSGao Xiang /* all pages in pagevec ought to be valid */ 89147e4937aSGao Xiang DBG_BUGON(!page); 8926aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 89347e4937aSGao Xiang 8946aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 89547e4937aSGao Xiang continue; 89647e4937aSGao Xiang 89747e4937aSGao Xiang if (page_type == Z_EROFS_VLE_PAGE_TYPE_HEAD) 89847e4937aSGao Xiang pagenr = 0; 89947e4937aSGao Xiang else 90047e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 90147e4937aSGao Xiang 90247e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 90347e4937aSGao Xiang 90447e4937aSGao Xiang /* 90547e4937aSGao Xiang * currently EROFS doesn't support multiref(dedup), 90647e4937aSGao Xiang * so here erroring out one multiref page. 90747e4937aSGao Xiang */ 9088d8a09b0SGao Xiang if (pages[pagenr]) { 90947e4937aSGao Xiang DBG_BUGON(1); 91047e4937aSGao Xiang SetPageError(pages[pagenr]); 91147e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 91247e4937aSGao Xiang err = -EFSCORRUPTED; 91347e4937aSGao Xiang } 91447e4937aSGao Xiang pages[pagenr] = page; 91547e4937aSGao Xiang } 91647e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&ctor, true); 91747e4937aSGao Xiang 91847e4937aSGao Xiang overlapped = false; 91947e4937aSGao Xiang compressed_pages = pcl->compressed_pages; 92047e4937aSGao Xiang 9219f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 92247e4937aSGao Xiang unsigned int pagenr; 92347e4937aSGao Xiang 92447e4937aSGao Xiang page = compressed_pages[i]; 92547e4937aSGao Xiang 92647e4937aSGao Xiang /* all compressed pages ought to be valid */ 92747e4937aSGao Xiang DBG_BUGON(!page); 9286aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 92947e4937aSGao Xiang 9306aaa7b06SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 93147e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) { 9328d8a09b0SGao Xiang if (!PageUptodate(page)) 93347e4937aSGao Xiang err = -EIO; 93447e4937aSGao Xiang continue; 93547e4937aSGao Xiang } 93647e4937aSGao Xiang 93747e4937aSGao Xiang /* 93847e4937aSGao Xiang * only if non-head page can be selected 93947e4937aSGao Xiang * for inplace decompression 94047e4937aSGao Xiang */ 94147e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 94247e4937aSGao Xiang 94347e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 9448d8a09b0SGao Xiang if (pages[pagenr]) { 94547e4937aSGao Xiang DBG_BUGON(1); 94647e4937aSGao Xiang SetPageError(pages[pagenr]); 94747e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 94847e4937aSGao Xiang err = -EFSCORRUPTED; 94947e4937aSGao Xiang } 95047e4937aSGao Xiang pages[pagenr] = page; 95147e4937aSGao Xiang 95247e4937aSGao Xiang overlapped = true; 95347e4937aSGao Xiang } 95447e4937aSGao Xiang 9556aaa7b06SGao Xiang /* PG_error needs checking for all non-managed pages */ 9568d8a09b0SGao Xiang if (PageError(page)) { 95747e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 95847e4937aSGao Xiang err = -EIO; 95947e4937aSGao Xiang } 96047e4937aSGao Xiang } 96147e4937aSGao Xiang 9628d8a09b0SGao Xiang if (err) 96347e4937aSGao Xiang goto out; 96447e4937aSGao Xiang 96547e4937aSGao Xiang llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 96647e4937aSGao Xiang if (nr_pages << PAGE_SHIFT >= cl->pageofs + llen) { 96747e4937aSGao Xiang outputsize = llen; 96847e4937aSGao Xiang partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 96947e4937aSGao Xiang } else { 97047e4937aSGao Xiang outputsize = (nr_pages << PAGE_SHIFT) - cl->pageofs; 97147e4937aSGao Xiang partial = true; 97247e4937aSGao Xiang } 97347e4937aSGao Xiang 9749f6cc76eSGao Xiang inputsize = pcl->pclusterpages * PAGE_SIZE; 97547e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 97647e4937aSGao Xiang .sb = sb, 97747e4937aSGao Xiang .in = compressed_pages, 97847e4937aSGao Xiang .out = pages, 97947e4937aSGao Xiang .pageofs_out = cl->pageofs, 9809f6cc76eSGao Xiang .inputsize = inputsize, 98147e4937aSGao Xiang .outputsize = outputsize, 98247e4937aSGao Xiang .alg = pcl->algorithmformat, 98347e4937aSGao Xiang .inplace_io = overlapped, 98447e4937aSGao Xiang .partial_decoding = partial 98547e4937aSGao Xiang }, pagepool); 98647e4937aSGao Xiang 98747e4937aSGao Xiang out: 988fe6adcceSRuiqi Gong /* must handle all compressed pages before ending pages */ 9899f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 99047e4937aSGao Xiang page = compressed_pages[i]; 99147e4937aSGao Xiang 99247e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 99347e4937aSGao Xiang continue; 99447e4937aSGao Xiang 9956aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 9966aaa7b06SGao Xiang (void)z_erofs_put_shortlivedpage(pagepool, page); 99747e4937aSGao Xiang 99847e4937aSGao Xiang WRITE_ONCE(compressed_pages[i], NULL); 99947e4937aSGao Xiang } 100047e4937aSGao Xiang 100147e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) { 100247e4937aSGao Xiang page = pages[i]; 100347e4937aSGao Xiang if (!page) 100447e4937aSGao Xiang continue; 100547e4937aSGao Xiang 10066aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 100747e4937aSGao Xiang 10086aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10096aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 101047e4937aSGao Xiang continue; 101147e4937aSGao Xiang 10128d8a09b0SGao Xiang if (err < 0) 101347e4937aSGao Xiang SetPageError(page); 101447e4937aSGao Xiang 101547e4937aSGao Xiang z_erofs_onlinepage_endio(page); 101647e4937aSGao Xiang } 101747e4937aSGao Xiang 101847e4937aSGao Xiang if (pages == z_pagemap_global) 101947e4937aSGao Xiang mutex_unlock(&z_pagemap_global_lock); 10208d8a09b0SGao Xiang else if (pages != pages_onstack) 102147e4937aSGao Xiang kvfree(pages); 102247e4937aSGao Xiang 102347e4937aSGao Xiang cl->nr_pages = 0; 102447e4937aSGao Xiang cl->vcnt = 0; 102547e4937aSGao Xiang 102647e4937aSGao Xiang /* all cl locks MUST be taken before the following line */ 102747e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 102847e4937aSGao Xiang 102947e4937aSGao Xiang /* all cl locks SHOULD be released right now */ 103047e4937aSGao Xiang mutex_unlock(&cl->lock); 103147e4937aSGao Xiang 103247e4937aSGao Xiang z_erofs_collection_put(cl); 103347e4937aSGao Xiang return err; 103447e4937aSGao Xiang } 103547e4937aSGao Xiang 10360c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1037*eaa9172aSGao Xiang struct page **pagepool) 103847e4937aSGao Xiang { 103947e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 104047e4937aSGao Xiang 104147e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 104247e4937aSGao Xiang struct z_erofs_pcluster *pcl; 104347e4937aSGao Xiang 104447e4937aSGao Xiang /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 104547e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 104647e4937aSGao Xiang 104747e4937aSGao Xiang /* no possible that 'owned' equals NULL */ 104847e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 104947e4937aSGao Xiang 105047e4937aSGao Xiang pcl = container_of(owned, struct z_erofs_pcluster, next); 105147e4937aSGao Xiang owned = READ_ONCE(pcl->next); 105247e4937aSGao Xiang 1053a4b1fab1SGao Xiang z_erofs_decompress_pcluster(io->sb, pcl, pagepool); 105447e4937aSGao Xiang } 105547e4937aSGao Xiang } 105647e4937aSGao Xiang 10570c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 105847e4937aSGao Xiang { 1059a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1060a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1061*eaa9172aSGao Xiang struct page *pagepool = NULL; 106247e4937aSGao Xiang 1063a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 10640c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 106547e4937aSGao Xiang 1066*eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1067a4b1fab1SGao Xiang kvfree(bgq); 106847e4937aSGao Xiang } 106947e4937aSGao Xiang 107047e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 107147e4937aSGao Xiang unsigned int nr, 1072*eaa9172aSGao Xiang struct page **pagepool, 107347e4937aSGao Xiang struct address_space *mc, 107447e4937aSGao Xiang gfp_t gfp) 107547e4937aSGao Xiang { 107647e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 107747e4937aSGao Xiang bool tocache = false; 107847e4937aSGao Xiang 107947e4937aSGao Xiang struct address_space *mapping; 108047e4937aSGao Xiang struct page *oldpage, *page; 108147e4937aSGao Xiang 108247e4937aSGao Xiang compressed_page_t t; 108347e4937aSGao Xiang int justfound; 108447e4937aSGao Xiang 108547e4937aSGao Xiang repeat: 108647e4937aSGao Xiang page = READ_ONCE(pcl->compressed_pages[nr]); 108747e4937aSGao Xiang oldpage = page; 108847e4937aSGao Xiang 108947e4937aSGao Xiang if (!page) 109047e4937aSGao Xiang goto out_allocpage; 109147e4937aSGao Xiang 109247e4937aSGao Xiang /* 109347e4937aSGao Xiang * the cached page has not been allocated and 109447e4937aSGao Xiang * an placeholder is out there, prepare it now. 109547e4937aSGao Xiang */ 1096bda17a45SGao Xiang if (page == PAGE_UNALLOCATED) { 109747e4937aSGao Xiang tocache = true; 109847e4937aSGao Xiang goto out_allocpage; 109947e4937aSGao Xiang } 110047e4937aSGao Xiang 110147e4937aSGao Xiang /* process the target tagged pointer */ 110247e4937aSGao Xiang t = tagptr_init(compressed_page_t, page); 110347e4937aSGao Xiang justfound = tagptr_unfold_tags(t); 110447e4937aSGao Xiang page = tagptr_unfold_ptr(t); 110547e4937aSGao Xiang 11061825c8d7SGao Xiang /* 11071825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 11081825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 11091825c8d7SGao Xiang */ 11101825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 11111825c8d7SGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 11121825c8d7SGao Xiang set_page_private(page, 0); 11131825c8d7SGao Xiang tocache = true; 11141825c8d7SGao Xiang goto out_tocache; 11151825c8d7SGao Xiang } 111647e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 111747e4937aSGao Xiang 111847e4937aSGao Xiang /* 11196aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 112047e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 112147e4937aSGao Xiang */ 112247e4937aSGao Xiang if (mapping && mapping != mc) 112347e4937aSGao Xiang /* ought to be unmanaged pages */ 112447e4937aSGao Xiang goto out; 112547e4937aSGao Xiang 11266aaa7b06SGao Xiang /* directly return for shortlived page as well */ 11276aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 11286aaa7b06SGao Xiang goto out; 11296aaa7b06SGao Xiang 113047e4937aSGao Xiang lock_page(page); 113147e4937aSGao Xiang 113247e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 113347e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 113447e4937aSGao Xiang 113547e4937aSGao Xiang /* the page is still in manage cache */ 113647e4937aSGao Xiang if (page->mapping == mc) { 113747e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 113847e4937aSGao Xiang 113947e4937aSGao Xiang ClearPageError(page); 114047e4937aSGao Xiang if (!PagePrivate(page)) { 114147e4937aSGao Xiang /* 114247e4937aSGao Xiang * impossible to be !PagePrivate(page) for 114347e4937aSGao Xiang * the current restriction as well if 114447e4937aSGao Xiang * the page is already in compressed_pages[]. 114547e4937aSGao Xiang */ 114647e4937aSGao Xiang DBG_BUGON(!justfound); 114747e4937aSGao Xiang 114847e4937aSGao Xiang justfound = 0; 114947e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 115047e4937aSGao Xiang SetPagePrivate(page); 115147e4937aSGao Xiang } 115247e4937aSGao Xiang 115347e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 115447e4937aSGao Xiang if (PageUptodate(page)) { 115547e4937aSGao Xiang unlock_page(page); 115647e4937aSGao Xiang page = NULL; 115747e4937aSGao Xiang } 115847e4937aSGao Xiang goto out; 115947e4937aSGao Xiang } 116047e4937aSGao Xiang 116147e4937aSGao Xiang /* 116247e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 116347e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 116447e4937aSGao Xiang */ 116547e4937aSGao Xiang DBG_BUGON(page->mapping); 116647e4937aSGao Xiang DBG_BUGON(!justfound); 116747e4937aSGao Xiang 116847e4937aSGao Xiang tocache = true; 116947e4937aSGao Xiang unlock_page(page); 117047e4937aSGao Xiang put_page(page); 117147e4937aSGao Xiang out_allocpage: 11725ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 11735ddcee1fSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_pages[nr], oldpage, page)) { 1174*eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 11755ddcee1fSGao Xiang cond_resched(); 11765ddcee1fSGao Xiang goto repeat; 11775ddcee1fSGao Xiang } 11781825c8d7SGao Xiang out_tocache: 1179bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1180bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1181bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1182bf225074SGao Xiang goto out; 1183a30573b3SGao Xiang } 1184bf225074SGao Xiang attach_page_private(page, pcl); 1185bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1186bf225074SGao Xiang put_page(page); 1187bf225074SGao Xiang 118847e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 118947e4937aSGao Xiang return page; 119047e4937aSGao Xiang } 119147e4937aSGao Xiang 1192a4b1fab1SGao Xiang static struct z_erofs_decompressqueue * 1193a4b1fab1SGao Xiang jobqueue_init(struct super_block *sb, 1194a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 119547e4937aSGao Xiang { 1196a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 119747e4937aSGao Xiang 1198a4b1fab1SGao Xiang if (fg && !*fg) { 1199a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1200a4b1fab1SGao Xiang if (!q) { 1201a4b1fab1SGao Xiang *fg = true; 1202a4b1fab1SGao Xiang goto fg_out; 120347e4937aSGao Xiang } 12040c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1205a4b1fab1SGao Xiang } else { 1206a4b1fab1SGao Xiang fg_out: 1207a4b1fab1SGao Xiang q = fgq; 1208a4b1fab1SGao Xiang init_waitqueue_head(&fgq->u.wait); 1209a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 1210a4b1fab1SGao Xiang } 1211a4b1fab1SGao Xiang q->sb = sb; 1212a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1213a4b1fab1SGao Xiang return q; 121447e4937aSGao Xiang } 121547e4937aSGao Xiang 121647e4937aSGao Xiang /* define decompression jobqueue types */ 121747e4937aSGao Xiang enum { 121847e4937aSGao Xiang JQ_BYPASS, 121947e4937aSGao Xiang JQ_SUBMIT, 122047e4937aSGao Xiang NR_JOBQUEUES, 122147e4937aSGao Xiang }; 122247e4937aSGao Xiang 122347e4937aSGao Xiang static void *jobqueueset_init(struct super_block *sb, 1224a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[], 1225a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 122647e4937aSGao Xiang { 122747e4937aSGao Xiang /* 122847e4937aSGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 122947e4937aSGao Xiang * no need to read from device for all pclusters in this queue. 123047e4937aSGao Xiang */ 1231a4b1fab1SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1232a4b1fab1SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 123347e4937aSGao Xiang 1234a4b1fab1SGao Xiang return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 123547e4937aSGao Xiang } 123647e4937aSGao Xiang 123747e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 123847e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 123947e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 124047e4937aSGao Xiang { 124147e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 124247e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 124347e4937aSGao Xiang 124447e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 124547e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 124647e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 124747e4937aSGao Xiang 124847e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 124947e4937aSGao Xiang 125047e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 125147e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 125247e4937aSGao Xiang 125347e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 125447e4937aSGao Xiang } 125547e4937aSGao Xiang 12561e4a2955SGao Xiang static void z_erofs_submit_queue(struct super_block *sb, 12576ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1258*eaa9172aSGao Xiang struct page **pagepool, 1259a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1260a4b1fab1SGao Xiang bool *force_fg) 126147e4937aSGao Xiang { 1262bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 126347e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1264a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 126547e4937aSGao Xiang void *bi_private; 12666ea5aad3SGao Xiang z_erofs_next_pcluster_t owned_head = f->clt.owned_head; 1267dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 12683f649ab7SKees Cook pgoff_t last_index; 1269dfeab2e9SGao Xiang struct block_device *last_bdev; 12701e4a2955SGao Xiang unsigned int nr_bios = 0; 12711e4a2955SGao Xiang struct bio *bio = NULL; 127247e4937aSGao Xiang 1273a4b1fab1SGao Xiang bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1274a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1275a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 127647e4937aSGao Xiang 127747e4937aSGao Xiang /* by default, all need io submission */ 127847e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 127947e4937aSGao Xiang 128047e4937aSGao Xiang do { 1281dfeab2e9SGao Xiang struct erofs_map_dev mdev; 128247e4937aSGao Xiang struct z_erofs_pcluster *pcl; 12831e4a2955SGao Xiang pgoff_t cur, end; 12841e4a2955SGao Xiang unsigned int i = 0; 12851e4a2955SGao Xiang bool bypass = true; 128647e4937aSGao Xiang 128747e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 128847e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 128947e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 129047e4937aSGao Xiang 129147e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 129247e4937aSGao Xiang 1293dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1294dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1295dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1296dfeab2e9SGao Xiang }; 1297dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1298dfeab2e9SGao Xiang 1299dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 13009f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 130147e4937aSGao Xiang 130247e4937aSGao Xiang /* close the main owned chain at first */ 130347e4937aSGao Xiang owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 130447e4937aSGao Xiang Z_EROFS_PCLUSTER_TAIL_CLOSED); 130547e4937aSGao Xiang 13061e4a2955SGao Xiang do { 13071e4a2955SGao Xiang struct page *page; 130847e4937aSGao Xiang 13091e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 131047e4937aSGao Xiang MNGD_MAPPING(sbi), 131147e4937aSGao Xiang GFP_NOFS); 13121e4a2955SGao Xiang if (!page) 13131e4a2955SGao Xiang continue; 131447e4937aSGao Xiang 1315dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1316dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 131747e4937aSGao Xiang submit_bio_retry: 131894e4e153SGao Xiang submit_bio(bio); 131947e4937aSGao Xiang bio = NULL; 132047e4937aSGao Xiang } 132147e4937aSGao Xiang 132247e4937aSGao Xiang if (!bio) { 1323a8affc03SChristoph Hellwig bio = bio_alloc(GFP_NOIO, BIO_MAX_VECS); 13240c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1325dfeab2e9SGao Xiang 1326dfeab2e9SGao Xiang bio_set_dev(bio, mdev.m_bdev); 1327dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 13281e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1329a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1330a5c0b780SGao Xiang bio->bi_private = bi_private; 133194e4e153SGao Xiang bio->bi_opf = REQ_OP_READ; 13326ea5aad3SGao Xiang if (f->readahead) 13336ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 133447e4937aSGao Xiang ++nr_bios; 133547e4937aSGao Xiang } 133647e4937aSGao Xiang 13376c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 133847e4937aSGao Xiang goto submit_bio_retry; 133947e4937aSGao Xiang 13401e4a2955SGao Xiang last_index = cur; 13411e4a2955SGao Xiang bypass = false; 13421e4a2955SGao Xiang } while (++cur < end); 134347e4937aSGao Xiang 13441e4a2955SGao Xiang if (!bypass) 134547e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 134647e4937aSGao Xiang else 134747e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 134847e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 134947e4937aSGao Xiang 135047e4937aSGao Xiang if (bio) 135194e4e153SGao Xiang submit_bio(bio); 135247e4937aSGao Xiang 1353587a67b7SGao Xiang /* 1354587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1355587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1356587a67b7SGao Xiang */ 1357587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1358587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 13591e4a2955SGao Xiang return; 1360587a67b7SGao Xiang } 1361a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 136247e4937aSGao Xiang } 136347e4937aSGao Xiang 13640c638f70SGao Xiang static void z_erofs_runqueue(struct super_block *sb, 13656ea5aad3SGao Xiang struct z_erofs_decompress_frontend *f, 1366*eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 136747e4937aSGao Xiang { 1368a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 136947e4937aSGao Xiang 13706ea5aad3SGao Xiang if (f->clt.owned_head == Z_EROFS_PCLUSTER_TAIL) 137147e4937aSGao Xiang return; 13726ea5aad3SGao Xiang z_erofs_submit_queue(sb, f, pagepool, io, &force_fg); 137347e4937aSGao Xiang 13740c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 13750c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 137647e4937aSGao Xiang 137747e4937aSGao Xiang if (!force_fg) 137847e4937aSGao Xiang return; 137947e4937aSGao Xiang 138047e4937aSGao Xiang /* wait until all bios are completed */ 1381a93f8c36SGao Xiang io_wait_event(io[JQ_SUBMIT].u.wait, 138247e4937aSGao Xiang !atomic_read(&io[JQ_SUBMIT].pending_bios)); 138347e4937aSGao Xiang 13840c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 13850c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 138647e4937aSGao Xiang } 138747e4937aSGao Xiang 138838629291SGao Xiang /* 138938629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 139038629291SGao Xiang * approximate readmore strategies as a start. 139138629291SGao Xiang */ 139238629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 139338629291SGao Xiang struct readahead_control *rac, 139438629291SGao Xiang erofs_off_t end, 1395*eaa9172aSGao Xiang struct page **pagepool, 139638629291SGao Xiang bool backmost) 139738629291SGao Xiang { 139838629291SGao Xiang struct inode *inode = f->inode; 139938629291SGao Xiang struct erofs_map_blocks *map = &f->map; 140038629291SGao Xiang erofs_off_t cur; 140138629291SGao Xiang int err; 140238629291SGao Xiang 140338629291SGao Xiang if (backmost) { 140438629291SGao Xiang map->m_la = end; 1405622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1406622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 140738629291SGao Xiang if (err) 140838629291SGao Xiang return; 140938629291SGao Xiang 141038629291SGao Xiang /* expend ra for the trailing edge if readahead */ 141138629291SGao Xiang if (rac) { 141238629291SGao Xiang loff_t newstart = readahead_pos(rac); 141338629291SGao Xiang 141438629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 141538629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 141638629291SGao Xiang return; 141738629291SGao Xiang } 141838629291SGao Xiang end = round_up(end, PAGE_SIZE); 141938629291SGao Xiang } else { 142038629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 142138629291SGao Xiang 142238629291SGao Xiang if (!map->m_llen) 142338629291SGao Xiang return; 142438629291SGao Xiang } 142538629291SGao Xiang 142638629291SGao Xiang cur = map->m_la + map->m_llen - 1; 142738629291SGao Xiang while (cur >= end) { 142838629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 142938629291SGao Xiang struct page *page; 143038629291SGao Xiang 143138629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 143238629291SGao Xiang if (!page) 143338629291SGao Xiang goto skip; 143438629291SGao Xiang 143538629291SGao Xiang if (PageUptodate(page)) { 143638629291SGao Xiang unlock_page(page); 143738629291SGao Xiang put_page(page); 143838629291SGao Xiang goto skip; 143938629291SGao Xiang } 144038629291SGao Xiang 144138629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 144238629291SGao Xiang if (err) 144338629291SGao Xiang erofs_err(inode->i_sb, 144438629291SGao Xiang "readmore error at page %lu @ nid %llu", 144538629291SGao Xiang index, EROFS_I(inode)->nid); 144638629291SGao Xiang put_page(page); 144738629291SGao Xiang skip: 144838629291SGao Xiang if (cur < PAGE_SIZE) 144938629291SGao Xiang break; 145038629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 145138629291SGao Xiang } 145238629291SGao Xiang } 145338629291SGao Xiang 14540c638f70SGao Xiang static int z_erofs_readpage(struct file *file, struct page *page) 145547e4937aSGao Xiang { 145647e4937aSGao Xiang struct inode *const inode = page->mapping->host; 145747e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1458*eaa9172aSGao Xiang struct page *pagepool = NULL; 145947e4937aSGao Xiang int err; 146047e4937aSGao Xiang 146147e4937aSGao Xiang trace_erofs_readpage(page, false); 146247e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 146347e4937aSGao Xiang 146438629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 146538629291SGao Xiang &pagepool, true); 14661825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 146738629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 146838629291SGao Xiang 146947e4937aSGao Xiang (void)z_erofs_collector_end(&f.clt); 147047e4937aSGao Xiang 147147e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 14726ea5aad3SGao Xiang z_erofs_runqueue(inode->i_sb, &f, &pagepool, true); 147347e4937aSGao Xiang 147447e4937aSGao Xiang if (err) 14754f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 147647e4937aSGao Xiang 147747e4937aSGao Xiang if (f.map.mpage) 147847e4937aSGao Xiang put_page(f.map.mpage); 147947e4937aSGao Xiang 1480*eaa9172aSGao Xiang erofs_release_pages(&pagepool); 148147e4937aSGao Xiang return err; 148247e4937aSGao Xiang } 148347e4937aSGao Xiang 14840615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 148547e4937aSGao Xiang { 14860615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 148747e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 148847e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1489*eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 149038629291SGao Xiang unsigned int nr_pages; 149147e4937aSGao Xiang 14926ea5aad3SGao Xiang f.readahead = true; 14930615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 149447e4937aSGao Xiang 149538629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 149638629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 149738629291SGao Xiang nr_pages = readahead_count(rac); 149838629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 149938629291SGao Xiang 15000615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 150147e4937aSGao Xiang set_page_private(page, (unsigned long)head); 150247e4937aSGao Xiang head = page; 150347e4937aSGao Xiang } 150447e4937aSGao Xiang 150547e4937aSGao Xiang while (head) { 150647e4937aSGao Xiang struct page *page = head; 150747e4937aSGao Xiang int err; 150847e4937aSGao Xiang 150947e4937aSGao Xiang /* traversal in reverse order */ 151047e4937aSGao Xiang head = (void *)page_private(page); 151147e4937aSGao Xiang 15121825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1513a5876e24SGao Xiang if (err) 15144f761fa2SGao Xiang erofs_err(inode->i_sb, 15154f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 15164f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 151747e4937aSGao Xiang put_page(page); 151847e4937aSGao Xiang } 151938629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 152047e4937aSGao Xiang (void)z_erofs_collector_end(&f.clt); 152147e4937aSGao Xiang 152238629291SGao Xiang z_erofs_runqueue(inode->i_sb, &f, &pagepool, 152338629291SGao Xiang sbi->opt.readahead_sync_decompress && 152438629291SGao Xiang nr_pages <= sbi->opt.max_sync_decompress_pages); 152547e4937aSGao Xiang if (f.map.mpage) 152647e4937aSGao Xiang put_page(f.map.mpage); 1527*eaa9172aSGao Xiang erofs_release_pages(&pagepool); 152847e4937aSGao Xiang } 152947e4937aSGao Xiang 15300c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 15310c638f70SGao Xiang .readpage = z_erofs_readpage, 15320615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 153347e4937aSGao Xiang }; 1534