147e4937aSGao Xiang // SPDX-License-Identifier: GPL-2.0-only 247e4937aSGao Xiang /* 347e4937aSGao Xiang * Copyright (C) 2018 HUAWEI, Inc. 447e4937aSGao Xiang * http://www.huawei.com/ 547e4937aSGao Xiang * Created by Gao Xiang <gaoxiang25@huawei.com> 647e4937aSGao Xiang */ 747e4937aSGao Xiang #include "zdata.h" 847e4937aSGao Xiang #include "compress.h" 947e4937aSGao Xiang #include <linux/prefetch.h> 1047e4937aSGao Xiang 1147e4937aSGao Xiang #include <trace/events/erofs.h> 1247e4937aSGao Xiang 1347e4937aSGao Xiang /* 1447e4937aSGao Xiang * a compressed_pages[] placeholder in order to avoid 1547e4937aSGao Xiang * being filled with file pages for in-place decompression. 1647e4937aSGao Xiang */ 1747e4937aSGao Xiang #define PAGE_UNALLOCATED ((void *)0x5F0E4B1D) 1847e4937aSGao Xiang 1947e4937aSGao Xiang /* how to allocate cached pages for a pcluster */ 2047e4937aSGao Xiang enum z_erofs_cache_alloctype { 2147e4937aSGao Xiang DONTALLOC, /* don't allocate any cached pages */ 2247e4937aSGao Xiang DELAYEDALLOC, /* delayed allocation (at the time of submitting io) */ 2347e4937aSGao Xiang }; 2447e4937aSGao Xiang 2547e4937aSGao Xiang /* 2647e4937aSGao Xiang * tagged pointer with 1-bit tag for all compressed pages 2747e4937aSGao Xiang * tag 0 - the page is just found with an extra page reference 2847e4937aSGao Xiang */ 2947e4937aSGao Xiang typedef tagptr1_t compressed_page_t; 3047e4937aSGao Xiang 3147e4937aSGao Xiang #define tag_compressed_page_justfound(page) \ 3247e4937aSGao Xiang tagptr_fold(compressed_page_t, page, 1) 3347e4937aSGao Xiang 3447e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 3547e4937aSGao Xiang static struct kmem_cache *pcluster_cachep __read_mostly; 3647e4937aSGao Xiang 3747e4937aSGao Xiang void z_erofs_exit_zip_subsystem(void) 3847e4937aSGao Xiang { 3947e4937aSGao Xiang destroy_workqueue(z_erofs_workqueue); 4047e4937aSGao Xiang kmem_cache_destroy(pcluster_cachep); 4147e4937aSGao Xiang } 4247e4937aSGao Xiang 4399634bf3SGao Xiang static inline int z_erofs_init_workqueue(void) 4447e4937aSGao Xiang { 4547e4937aSGao Xiang const unsigned int onlinecpus = num_possible_cpus(); 4647e4937aSGao Xiang const unsigned int flags = WQ_UNBOUND | WQ_HIGHPRI | WQ_CPU_INTENSIVE; 4747e4937aSGao Xiang 4847e4937aSGao Xiang /* 4947e4937aSGao Xiang * no need to spawn too many threads, limiting threads could minimum 5047e4937aSGao Xiang * scheduling overhead, perhaps per-CPU threads should be better? 5147e4937aSGao Xiang */ 5247e4937aSGao Xiang z_erofs_workqueue = alloc_workqueue("erofs_unzipd", flags, 5347e4937aSGao Xiang onlinecpus + onlinecpus / 4); 5447e4937aSGao Xiang return z_erofs_workqueue ? 0 : -ENOMEM; 5547e4937aSGao Xiang } 5647e4937aSGao Xiang 5799634bf3SGao Xiang static void z_erofs_pcluster_init_once(void *ptr) 5847e4937aSGao Xiang { 5947e4937aSGao Xiang struct z_erofs_pcluster *pcl = ptr; 6047e4937aSGao Xiang struct z_erofs_collection *cl = z_erofs_primarycollection(pcl); 6147e4937aSGao Xiang unsigned int i; 6247e4937aSGao Xiang 6347e4937aSGao Xiang mutex_init(&cl->lock); 6447e4937aSGao Xiang cl->nr_pages = 0; 6547e4937aSGao Xiang cl->vcnt = 0; 6647e4937aSGao Xiang for (i = 0; i < Z_EROFS_CLUSTER_MAX_PAGES; ++i) 6747e4937aSGao Xiang pcl->compressed_pages[i] = NULL; 6847e4937aSGao Xiang } 6947e4937aSGao Xiang 7099634bf3SGao Xiang static void z_erofs_pcluster_init_always(struct z_erofs_pcluster *pcl) 7147e4937aSGao Xiang { 7247e4937aSGao Xiang struct z_erofs_collection *cl = z_erofs_primarycollection(pcl); 7347e4937aSGao Xiang 7447e4937aSGao Xiang atomic_set(&pcl->obj.refcount, 1); 7547e4937aSGao Xiang 7647e4937aSGao Xiang DBG_BUGON(cl->nr_pages); 7747e4937aSGao Xiang DBG_BUGON(cl->vcnt); 7847e4937aSGao Xiang } 7947e4937aSGao Xiang 8047e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 8147e4937aSGao Xiang { 8247e4937aSGao Xiang pcluster_cachep = kmem_cache_create("erofs_compress", 8347e4937aSGao Xiang Z_EROFS_WORKGROUP_SIZE, 0, 8499634bf3SGao Xiang SLAB_RECLAIM_ACCOUNT, 8599634bf3SGao Xiang z_erofs_pcluster_init_once); 8647e4937aSGao Xiang if (pcluster_cachep) { 8799634bf3SGao Xiang if (!z_erofs_init_workqueue()) 8847e4937aSGao Xiang return 0; 8947e4937aSGao Xiang 9047e4937aSGao Xiang kmem_cache_destroy(pcluster_cachep); 9147e4937aSGao Xiang } 9247e4937aSGao Xiang return -ENOMEM; 9347e4937aSGao Xiang } 9447e4937aSGao Xiang 9547e4937aSGao Xiang enum z_erofs_collectmode { 9647e4937aSGao Xiang COLLECT_SECONDARY, 9747e4937aSGao Xiang COLLECT_PRIMARY, 9847e4937aSGao Xiang /* 9947e4937aSGao Xiang * The current collection was the tail of an exist chain, in addition 10047e4937aSGao Xiang * that the previous processed chained collections are all decided to 10147e4937aSGao Xiang * be hooked up to it. 10247e4937aSGao Xiang * A new chain will be created for the remaining collections which are 10347e4937aSGao Xiang * not processed yet, therefore different from COLLECT_PRIMARY_FOLLOWED, 10447e4937aSGao Xiang * the next collection cannot reuse the whole page safely in 10547e4937aSGao Xiang * the following scenario: 10647e4937aSGao Xiang * ________________________________________________________________ 10747e4937aSGao Xiang * | tail (partial) page | head (partial) page | 10847e4937aSGao Xiang * | (belongs to the next cl) | (belongs to the current cl) | 10947e4937aSGao Xiang * |_______PRIMARY_FOLLOWED_______|________PRIMARY_HOOKED___________| 11047e4937aSGao Xiang */ 11147e4937aSGao Xiang COLLECT_PRIMARY_HOOKED, 11247e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED_NOINPLACE, 11347e4937aSGao Xiang /* 11447e4937aSGao Xiang * The current collection has been linked with the owned chain, and 11547e4937aSGao Xiang * could also be linked with the remaining collections, which means 11647e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 11747e4937aSGao Xiang * the current collection can safely use the whole page (since 11847e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 11947e4937aSGao Xiang * illustrated below: 12047e4937aSGao Xiang * ________________________________________________________________ 12147e4937aSGao Xiang * | tail (partial) page | head (partial) page | 12247e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 12347e4937aSGao Xiang * | PRIMARY_FOLLOWED or | | 12447e4937aSGao Xiang * |_____PRIMARY_HOOKED___|____________PRIMARY_FOLLOWED____________| 12547e4937aSGao Xiang * 12647e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 12747e4937aSGao Xiang */ 12847e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED, 12947e4937aSGao Xiang }; 13047e4937aSGao Xiang 13147e4937aSGao Xiang struct z_erofs_collector { 13247e4937aSGao Xiang struct z_erofs_pagevec_ctor vector; 13347e4937aSGao Xiang 13447e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 13547e4937aSGao Xiang struct z_erofs_collection *cl; 13647e4937aSGao Xiang struct page **compressedpages; 13747e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 13847e4937aSGao Xiang 13947e4937aSGao Xiang enum z_erofs_collectmode mode; 14047e4937aSGao Xiang }; 14147e4937aSGao Xiang 14247e4937aSGao Xiang struct z_erofs_decompress_frontend { 14347e4937aSGao Xiang struct inode *const inode; 14447e4937aSGao Xiang 14547e4937aSGao Xiang struct z_erofs_collector clt; 14647e4937aSGao Xiang struct erofs_map_blocks map; 14747e4937aSGao Xiang 14847e4937aSGao Xiang /* used for applying cache strategy on the fly */ 14947e4937aSGao Xiang bool backmost; 15047e4937aSGao Xiang erofs_off_t headoffset; 15147e4937aSGao Xiang }; 15247e4937aSGao Xiang 15347e4937aSGao Xiang #define COLLECTOR_INIT() { \ 15447e4937aSGao Xiang .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 15547e4937aSGao Xiang .mode = COLLECT_PRIMARY_FOLLOWED } 15647e4937aSGao Xiang 15747e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 15847e4937aSGao Xiang .inode = __i, .clt = COLLECTOR_INIT(), \ 15947e4937aSGao Xiang .backmost = true, } 16047e4937aSGao Xiang 16147e4937aSGao Xiang static struct page *z_pagemap_global[Z_EROFS_VMAP_GLOBAL_PAGES]; 16247e4937aSGao Xiang static DEFINE_MUTEX(z_pagemap_global_lock); 16347e4937aSGao Xiang 16447e4937aSGao Xiang static void preload_compressed_pages(struct z_erofs_collector *clt, 16547e4937aSGao Xiang struct address_space *mc, 16647e4937aSGao Xiang enum z_erofs_cache_alloctype type, 16747e4937aSGao Xiang struct list_head *pagepool) 16847e4937aSGao Xiang { 16947e4937aSGao Xiang const struct z_erofs_pcluster *pcl = clt->pcl; 17047e4937aSGao Xiang const unsigned int clusterpages = BIT(pcl->clusterbits); 17147e4937aSGao Xiang struct page **pages = clt->compressedpages; 17247e4937aSGao Xiang pgoff_t index = pcl->obj.index + (pages - pcl->compressed_pages); 17347e4937aSGao Xiang bool standalone = true; 17447e4937aSGao Xiang 17547e4937aSGao Xiang if (clt->mode < COLLECT_PRIMARY_FOLLOWED) 17647e4937aSGao Xiang return; 17747e4937aSGao Xiang 17847e4937aSGao Xiang for (; pages < pcl->compressed_pages + clusterpages; ++pages) { 17947e4937aSGao Xiang struct page *page; 18047e4937aSGao Xiang compressed_page_t t; 18147e4937aSGao Xiang 18247e4937aSGao Xiang /* the compressed page was loaded before */ 18347e4937aSGao Xiang if (READ_ONCE(*pages)) 18447e4937aSGao Xiang continue; 18547e4937aSGao Xiang 18647e4937aSGao Xiang page = find_get_page(mc, index); 18747e4937aSGao Xiang 18847e4937aSGao Xiang if (page) { 18947e4937aSGao Xiang t = tag_compressed_page_justfound(page); 19047e4937aSGao Xiang } else if (type == DELAYEDALLOC) { 19147e4937aSGao Xiang t = tagptr_init(compressed_page_t, PAGE_UNALLOCATED); 19247e4937aSGao Xiang } else { /* DONTALLOC */ 19347e4937aSGao Xiang if (standalone) 19447e4937aSGao Xiang clt->compressedpages = pages; 19547e4937aSGao Xiang standalone = false; 19647e4937aSGao Xiang continue; 19747e4937aSGao Xiang } 19847e4937aSGao Xiang 19947e4937aSGao Xiang if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t))) 20047e4937aSGao Xiang continue; 20147e4937aSGao Xiang 20247e4937aSGao Xiang if (page) 20347e4937aSGao Xiang put_page(page); 20447e4937aSGao Xiang } 20547e4937aSGao Xiang 20647e4937aSGao Xiang if (standalone) /* downgrade to PRIMARY_FOLLOWED_NOINPLACE */ 20747e4937aSGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 20847e4937aSGao Xiang } 20947e4937aSGao Xiang 21047e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 21147e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 21247e4937aSGao Xiang struct erofs_workgroup *grp) 21347e4937aSGao Xiang { 21447e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 21547e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 21647e4937aSGao Xiang struct address_space *const mapping = MNGD_MAPPING(sbi); 21747e4937aSGao Xiang const unsigned int clusterpages = BIT(pcl->clusterbits); 21847e4937aSGao Xiang int i; 21947e4937aSGao Xiang 22047e4937aSGao Xiang /* 22147e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 22247e4937aSGao Xiang * therefore no need to worry about available decompression users. 22347e4937aSGao Xiang */ 22447e4937aSGao Xiang for (i = 0; i < clusterpages; ++i) { 22547e4937aSGao Xiang struct page *page = pcl->compressed_pages[i]; 22647e4937aSGao Xiang 22747e4937aSGao Xiang if (!page) 22847e4937aSGao Xiang continue; 22947e4937aSGao Xiang 23047e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 23147e4937aSGao Xiang if (!trylock_page(page)) 23247e4937aSGao Xiang return -EBUSY; 23347e4937aSGao Xiang 2348d8a09b0SGao Xiang if (page->mapping != mapping) 23547e4937aSGao Xiang continue; 23647e4937aSGao Xiang 23747e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 23847e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 23947e4937aSGao Xiang set_page_private(page, 0); 24047e4937aSGao Xiang ClearPagePrivate(page); 24147e4937aSGao Xiang 24247e4937aSGao Xiang unlock_page(page); 24347e4937aSGao Xiang put_page(page); 24447e4937aSGao Xiang } 24547e4937aSGao Xiang return 0; 24647e4937aSGao Xiang } 24747e4937aSGao Xiang 24847e4937aSGao Xiang int erofs_try_to_free_cached_page(struct address_space *mapping, 24947e4937aSGao Xiang struct page *page) 25047e4937aSGao Xiang { 25147e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 25247e4937aSGao Xiang const unsigned int clusterpages = BIT(pcl->clusterbits); 25347e4937aSGao Xiang int ret = 0; /* 0 - busy */ 25447e4937aSGao Xiang 25547e4937aSGao Xiang if (erofs_workgroup_try_to_freeze(&pcl->obj, 1)) { 25647e4937aSGao Xiang unsigned int i; 25747e4937aSGao Xiang 25847e4937aSGao Xiang for (i = 0; i < clusterpages; ++i) { 25947e4937aSGao Xiang if (pcl->compressed_pages[i] == page) { 26047e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 26147e4937aSGao Xiang ret = 1; 26247e4937aSGao Xiang break; 26347e4937aSGao Xiang } 26447e4937aSGao Xiang } 26547e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 26647e4937aSGao Xiang 26747e4937aSGao Xiang if (ret) { 26847e4937aSGao Xiang ClearPagePrivate(page); 26947e4937aSGao Xiang put_page(page); 27047e4937aSGao Xiang } 27147e4937aSGao Xiang } 27247e4937aSGao Xiang return ret; 27347e4937aSGao Xiang } 27447e4937aSGao Xiang 27547e4937aSGao Xiang /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 27699634bf3SGao Xiang static inline bool z_erofs_try_inplace_io(struct z_erofs_collector *clt, 27747e4937aSGao Xiang struct page *page) 27847e4937aSGao Xiang { 27947e4937aSGao Xiang struct z_erofs_pcluster *const pcl = clt->pcl; 28047e4937aSGao Xiang const unsigned int clusterpages = BIT(pcl->clusterbits); 28147e4937aSGao Xiang 28247e4937aSGao Xiang while (clt->compressedpages < pcl->compressed_pages + clusterpages) { 28347e4937aSGao Xiang if (!cmpxchg(clt->compressedpages++, NULL, page)) 28447e4937aSGao Xiang return true; 28547e4937aSGao Xiang } 28647e4937aSGao Xiang return false; 28747e4937aSGao Xiang } 28847e4937aSGao Xiang 28947e4937aSGao Xiang /* callers must be with collection lock held */ 29047e4937aSGao Xiang static int z_erofs_attach_page(struct z_erofs_collector *clt, 29147e4937aSGao Xiang struct page *page, 29247e4937aSGao Xiang enum z_erofs_page_type type) 29347e4937aSGao Xiang { 29447e4937aSGao Xiang int ret; 29547e4937aSGao Xiang bool occupied; 29647e4937aSGao Xiang 29747e4937aSGao Xiang /* give priority for inplaceio */ 29847e4937aSGao Xiang if (clt->mode >= COLLECT_PRIMARY && 29947e4937aSGao Xiang type == Z_EROFS_PAGE_TYPE_EXCLUSIVE && 30099634bf3SGao Xiang z_erofs_try_inplace_io(clt, page)) 30147e4937aSGao Xiang return 0; 30247e4937aSGao Xiang 30347e4937aSGao Xiang ret = z_erofs_pagevec_enqueue(&clt->vector, 30447e4937aSGao Xiang page, type, &occupied); 30547e4937aSGao Xiang clt->cl->vcnt += (unsigned int)ret; 30647e4937aSGao Xiang 30747e4937aSGao Xiang return ret ? 0 : -EAGAIN; 30847e4937aSGao Xiang } 30947e4937aSGao Xiang 31047e4937aSGao Xiang static enum z_erofs_collectmode 31147e4937aSGao Xiang try_to_claim_pcluster(struct z_erofs_pcluster *pcl, 31247e4937aSGao Xiang z_erofs_next_pcluster_t *owned_head) 31347e4937aSGao Xiang { 31447e4937aSGao Xiang /* let's claim these following types of pclusters */ 31547e4937aSGao Xiang retry: 31647e4937aSGao Xiang if (pcl->next == Z_EROFS_PCLUSTER_NIL) { 31747e4937aSGao Xiang /* type 1, nil pcluster */ 31847e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 31947e4937aSGao Xiang *owned_head) != Z_EROFS_PCLUSTER_NIL) 32047e4937aSGao Xiang goto retry; 32147e4937aSGao Xiang 32247e4937aSGao Xiang *owned_head = &pcl->next; 32347e4937aSGao Xiang /* lucky, I am the followee :) */ 32447e4937aSGao Xiang return COLLECT_PRIMARY_FOLLOWED; 32547e4937aSGao Xiang } else if (pcl->next == Z_EROFS_PCLUSTER_TAIL) { 32647e4937aSGao Xiang /* 32747e4937aSGao Xiang * type 2, link to the end of a existing open chain, 32847e4937aSGao Xiang * be careful that its submission itself is governed 32947e4937aSGao Xiang * by the original owned chain. 33047e4937aSGao Xiang */ 33147e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 33247e4937aSGao Xiang *owned_head) != Z_EROFS_PCLUSTER_TAIL) 33347e4937aSGao Xiang goto retry; 33447e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 33547e4937aSGao Xiang return COLLECT_PRIMARY_HOOKED; 33647e4937aSGao Xiang } 33747e4937aSGao Xiang return COLLECT_PRIMARY; /* :( better luck next time */ 33847e4937aSGao Xiang } 33947e4937aSGao Xiang 3409e579fc1SGao Xiang static int z_erofs_lookup_collection(struct z_erofs_collector *clt, 34147e4937aSGao Xiang struct inode *inode, 34247e4937aSGao Xiang struct erofs_map_blocks *map) 34347e4937aSGao Xiang { 34447e4937aSGao Xiang struct erofs_workgroup *grp; 34547e4937aSGao Xiang struct z_erofs_pcluster *pcl; 34647e4937aSGao Xiang struct z_erofs_collection *cl; 34747e4937aSGao Xiang unsigned int length; 34847e4937aSGao Xiang 349997626d8SVladimir Zapolskiy grp = erofs_find_workgroup(inode->i_sb, map->m_pa >> PAGE_SHIFT); 35047e4937aSGao Xiang if (!grp) 3519e579fc1SGao Xiang return -ENOENT; 35247e4937aSGao Xiang 35347e4937aSGao Xiang pcl = container_of(grp, struct z_erofs_pcluster, obj); 35447e4937aSGao Xiang if (clt->owned_head == &pcl->next || pcl == clt->tailpcl) { 35547e4937aSGao Xiang DBG_BUGON(1); 35647e4937aSGao Xiang erofs_workgroup_put(grp); 3579e579fc1SGao Xiang return -EFSCORRUPTED; 35847e4937aSGao Xiang } 35947e4937aSGao Xiang 36047e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 3618d8a09b0SGao Xiang if (cl->pageofs != (map->m_la & ~PAGE_MASK)) { 36247e4937aSGao Xiang DBG_BUGON(1); 36347e4937aSGao Xiang erofs_workgroup_put(grp); 3649e579fc1SGao Xiang return -EFSCORRUPTED; 36547e4937aSGao Xiang } 36647e4937aSGao Xiang 36747e4937aSGao Xiang length = READ_ONCE(pcl->length); 36847e4937aSGao Xiang if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 36947e4937aSGao Xiang if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 37047e4937aSGao Xiang DBG_BUGON(1); 37147e4937aSGao Xiang erofs_workgroup_put(grp); 3729e579fc1SGao Xiang return -EFSCORRUPTED; 37347e4937aSGao Xiang } 37447e4937aSGao Xiang } else { 37547e4937aSGao Xiang unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 37647e4937aSGao Xiang 37747e4937aSGao Xiang if (map->m_flags & EROFS_MAP_FULL_MAPPED) 37847e4937aSGao Xiang llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 37947e4937aSGao Xiang 38047e4937aSGao Xiang while (llen > length && 38147e4937aSGao Xiang length != cmpxchg_relaxed(&pcl->length, length, llen)) { 38247e4937aSGao Xiang cpu_relax(); 38347e4937aSGao Xiang length = READ_ONCE(pcl->length); 38447e4937aSGao Xiang } 38547e4937aSGao Xiang } 38647e4937aSGao Xiang mutex_lock(&cl->lock); 38747e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 38847e4937aSGao Xiang if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 38947e4937aSGao Xiang clt->tailpcl = pcl; 39047e4937aSGao Xiang clt->mode = try_to_claim_pcluster(pcl, &clt->owned_head); 39147e4937aSGao Xiang /* clean tailpcl if the current owned_head is Z_EROFS_PCLUSTER_TAIL */ 39247e4937aSGao Xiang if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 39347e4937aSGao Xiang clt->tailpcl = NULL; 39447e4937aSGao Xiang clt->pcl = pcl; 39547e4937aSGao Xiang clt->cl = cl; 3969e579fc1SGao Xiang return 0; 39747e4937aSGao Xiang } 39847e4937aSGao Xiang 3999e579fc1SGao Xiang static int z_erofs_register_collection(struct z_erofs_collector *clt, 40047e4937aSGao Xiang struct inode *inode, 40147e4937aSGao Xiang struct erofs_map_blocks *map) 40247e4937aSGao Xiang { 40347e4937aSGao Xiang struct z_erofs_pcluster *pcl; 40447e4937aSGao Xiang struct z_erofs_collection *cl; 40547e4937aSGao Xiang int err; 40647e4937aSGao Xiang 40747e4937aSGao Xiang /* no available workgroup, let's allocate one */ 40847e4937aSGao Xiang pcl = kmem_cache_alloc(pcluster_cachep, GFP_NOFS); 4098d8a09b0SGao Xiang if (!pcl) 4109e579fc1SGao Xiang return -ENOMEM; 41147e4937aSGao Xiang 41299634bf3SGao Xiang z_erofs_pcluster_init_always(pcl); 41347e4937aSGao Xiang pcl->obj.index = map->m_pa >> PAGE_SHIFT; 41447e4937aSGao Xiang 41547e4937aSGao Xiang pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 41647e4937aSGao Xiang (map->m_flags & EROFS_MAP_FULL_MAPPED ? 41747e4937aSGao Xiang Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 41847e4937aSGao Xiang 41947e4937aSGao Xiang if (map->m_flags & EROFS_MAP_ZIPPED) 42047e4937aSGao Xiang pcl->algorithmformat = Z_EROFS_COMPRESSION_LZ4; 42147e4937aSGao Xiang else 42247e4937aSGao Xiang pcl->algorithmformat = Z_EROFS_COMPRESSION_SHIFTED; 42347e4937aSGao Xiang 424a5876e24SGao Xiang pcl->clusterbits = EROFS_I(inode)->z_physical_clusterbits[0]; 42547e4937aSGao Xiang pcl->clusterbits -= PAGE_SHIFT; 42647e4937aSGao Xiang 42747e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 42847e4937aSGao Xiang pcl->next = clt->owned_head; 42947e4937aSGao Xiang clt->mode = COLLECT_PRIMARY_FOLLOWED; 43047e4937aSGao Xiang 43147e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 43247e4937aSGao Xiang cl->pageofs = map->m_la & ~PAGE_MASK; 43347e4937aSGao Xiang 43447e4937aSGao Xiang /* 43547e4937aSGao Xiang * lock all primary followed works before visible to others 43647e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 43747e4937aSGao Xiang */ 43847e4937aSGao Xiang mutex_trylock(&cl->lock); 43947e4937aSGao Xiang 440e5e9a432SVladimir Zapolskiy err = erofs_register_workgroup(inode->i_sb, &pcl->obj); 44147e4937aSGao Xiang if (err) { 44247e4937aSGao Xiang mutex_unlock(&cl->lock); 44347e4937aSGao Xiang kmem_cache_free(pcluster_cachep, pcl); 4449e579fc1SGao Xiang return -EAGAIN; 44547e4937aSGao Xiang } 44647e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 44747e4937aSGao Xiang if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 44847e4937aSGao Xiang clt->tailpcl = pcl; 44947e4937aSGao Xiang clt->owned_head = &pcl->next; 45047e4937aSGao Xiang clt->pcl = pcl; 45147e4937aSGao Xiang clt->cl = cl; 4529e579fc1SGao Xiang return 0; 45347e4937aSGao Xiang } 45447e4937aSGao Xiang 45547e4937aSGao Xiang static int z_erofs_collector_begin(struct z_erofs_collector *clt, 45647e4937aSGao Xiang struct inode *inode, 45747e4937aSGao Xiang struct erofs_map_blocks *map) 45847e4937aSGao Xiang { 4599e579fc1SGao Xiang int ret; 46047e4937aSGao Xiang 46147e4937aSGao Xiang DBG_BUGON(clt->cl); 46247e4937aSGao Xiang 46347e4937aSGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous collection */ 46447e4937aSGao Xiang DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_NIL); 46547e4937aSGao Xiang DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 46647e4937aSGao Xiang 46747e4937aSGao Xiang if (!PAGE_ALIGNED(map->m_pa)) { 46847e4937aSGao Xiang DBG_BUGON(1); 46947e4937aSGao Xiang return -EINVAL; 47047e4937aSGao Xiang } 47147e4937aSGao Xiang 47247e4937aSGao Xiang repeat: 4739e579fc1SGao Xiang ret = z_erofs_lookup_collection(clt, inode, map); 4749e579fc1SGao Xiang if (ret == -ENOENT) { 4759e579fc1SGao Xiang ret = z_erofs_register_collection(clt, inode, map); 47647e4937aSGao Xiang 4779e579fc1SGao Xiang /* someone registered at the same time, give another try */ 4789e579fc1SGao Xiang if (ret == -EAGAIN) { 4799e579fc1SGao Xiang cond_resched(); 48047e4937aSGao Xiang goto repeat; 48147e4937aSGao Xiang } 4829e579fc1SGao Xiang } 48347e4937aSGao Xiang 4849e579fc1SGao Xiang if (ret) 4859e579fc1SGao Xiang return ret; 48647e4937aSGao Xiang 48747e4937aSGao Xiang z_erofs_pagevec_ctor_init(&clt->vector, Z_EROFS_NR_INLINE_PAGEVECS, 4889e579fc1SGao Xiang clt->cl->pagevec, clt->cl->vcnt); 48947e4937aSGao Xiang 49047e4937aSGao Xiang clt->compressedpages = clt->pcl->compressed_pages; 49147e4937aSGao Xiang if (clt->mode <= COLLECT_PRIMARY) /* cannot do in-place I/O */ 49247e4937aSGao Xiang clt->compressedpages += Z_EROFS_CLUSTER_MAX_PAGES; 49347e4937aSGao Xiang return 0; 49447e4937aSGao Xiang } 49547e4937aSGao Xiang 49647e4937aSGao Xiang /* 49747e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 49847e4937aSGao Xiang * only after a RCU grace period. 49947e4937aSGao Xiang */ 50047e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 50147e4937aSGao Xiang { 50247e4937aSGao Xiang struct z_erofs_collection *const cl = 50347e4937aSGao Xiang container_of(head, struct z_erofs_collection, rcu); 50447e4937aSGao Xiang 50547e4937aSGao Xiang kmem_cache_free(pcluster_cachep, 50647e4937aSGao Xiang container_of(cl, struct z_erofs_pcluster, 50747e4937aSGao Xiang primary_collection)); 50847e4937aSGao Xiang } 50947e4937aSGao Xiang 51047e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 51147e4937aSGao Xiang { 51247e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 51347e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 51447e4937aSGao Xiang struct z_erofs_collection *const cl = z_erofs_primarycollection(pcl); 51547e4937aSGao Xiang 51647e4937aSGao Xiang call_rcu(&cl->rcu, z_erofs_rcu_callback); 51747e4937aSGao Xiang } 51847e4937aSGao Xiang 51947e4937aSGao Xiang static void z_erofs_collection_put(struct z_erofs_collection *cl) 52047e4937aSGao Xiang { 52147e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 52247e4937aSGao Xiang container_of(cl, struct z_erofs_pcluster, primary_collection); 52347e4937aSGao Xiang 52447e4937aSGao Xiang erofs_workgroup_put(&pcl->obj); 52547e4937aSGao Xiang } 52647e4937aSGao Xiang 52747e4937aSGao Xiang static bool z_erofs_collector_end(struct z_erofs_collector *clt) 52847e4937aSGao Xiang { 52947e4937aSGao Xiang struct z_erofs_collection *cl = clt->cl; 53047e4937aSGao Xiang 53147e4937aSGao Xiang if (!cl) 53247e4937aSGao Xiang return false; 53347e4937aSGao Xiang 53447e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&clt->vector, false); 53547e4937aSGao Xiang mutex_unlock(&cl->lock); 53647e4937aSGao Xiang 53747e4937aSGao Xiang /* 53847e4937aSGao Xiang * if all pending pages are added, don't hold its reference 53947e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 54047e4937aSGao Xiang */ 54147e4937aSGao Xiang if (clt->mode < COLLECT_PRIMARY_FOLLOWED_NOINPLACE) 54247e4937aSGao Xiang z_erofs_collection_put(cl); 54347e4937aSGao Xiang 54447e4937aSGao Xiang clt->cl = NULL; 54547e4937aSGao Xiang return true; 54647e4937aSGao Xiang } 54747e4937aSGao Xiang 54847e4937aSGao Xiang static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe, 54947e4937aSGao Xiang unsigned int cachestrategy, 55047e4937aSGao Xiang erofs_off_t la) 55147e4937aSGao Xiang { 55247e4937aSGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 55347e4937aSGao Xiang return false; 55447e4937aSGao Xiang 55547e4937aSGao Xiang if (fe->backmost) 55647e4937aSGao Xiang return true; 55747e4937aSGao Xiang 55847e4937aSGao Xiang return cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 55947e4937aSGao Xiang la < fe->headoffset; 56047e4937aSGao Xiang } 56147e4937aSGao Xiang 56247e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 56347e4937aSGao Xiang struct page *page, 56447e4937aSGao Xiang struct list_head *pagepool) 56547e4937aSGao Xiang { 56647e4937aSGao Xiang struct inode *const inode = fe->inode; 567bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 56847e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 56947e4937aSGao Xiang struct z_erofs_collector *const clt = &fe->clt; 57047e4937aSGao Xiang const loff_t offset = page_offset(page); 571dc76ea8cSGao Xiang bool tight = true; 57247e4937aSGao Xiang 57347e4937aSGao Xiang enum z_erofs_cache_alloctype cache_strategy; 57447e4937aSGao Xiang enum z_erofs_page_type page_type; 57547e4937aSGao Xiang unsigned int cur, end, spiltted, index; 57647e4937aSGao Xiang int err = 0; 57747e4937aSGao Xiang 57847e4937aSGao Xiang /* register locked file pages as online pages in pack */ 57947e4937aSGao Xiang z_erofs_onlinepage_init(page); 58047e4937aSGao Xiang 58147e4937aSGao Xiang spiltted = 0; 58247e4937aSGao Xiang end = PAGE_SIZE; 58347e4937aSGao Xiang repeat: 58447e4937aSGao Xiang cur = end - 1; 58547e4937aSGao Xiang 58647e4937aSGao Xiang /* lucky, within the range of the current map_blocks */ 58747e4937aSGao Xiang if (offset + cur >= map->m_la && 58847e4937aSGao Xiang offset + cur < map->m_la + map->m_llen) { 58947e4937aSGao Xiang /* didn't get a valid collection previously (very rare) */ 59047e4937aSGao Xiang if (!clt->cl) 59147e4937aSGao Xiang goto restart_now; 59247e4937aSGao Xiang goto hitted; 59347e4937aSGao Xiang } 59447e4937aSGao Xiang 59547e4937aSGao Xiang /* go ahead the next map_blocks */ 5964f761fa2SGao Xiang erofs_dbg("%s: [out-of-range] pos %llu", __func__, offset + cur); 59747e4937aSGao Xiang 59847e4937aSGao Xiang if (z_erofs_collector_end(clt)) 59947e4937aSGao Xiang fe->backmost = false; 60047e4937aSGao Xiang 60147e4937aSGao Xiang map->m_la = offset + cur; 60247e4937aSGao Xiang map->m_llen = 0; 60347e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 6048d8a09b0SGao Xiang if (err) 60547e4937aSGao Xiang goto err_out; 60647e4937aSGao Xiang 60747e4937aSGao Xiang restart_now: 6088d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) 60947e4937aSGao Xiang goto hitted; 61047e4937aSGao Xiang 61147e4937aSGao Xiang err = z_erofs_collector_begin(clt, inode, map); 6128d8a09b0SGao Xiang if (err) 61347e4937aSGao Xiang goto err_out; 61447e4937aSGao Xiang 61547e4937aSGao Xiang /* preload all compressed pages (maybe downgrade role if necessary) */ 61647e4937aSGao Xiang if (should_alloc_managed_pages(fe, sbi->cache_strategy, map->m_la)) 61747e4937aSGao Xiang cache_strategy = DELAYEDALLOC; 61847e4937aSGao Xiang else 61947e4937aSGao Xiang cache_strategy = DONTALLOC; 62047e4937aSGao Xiang 62147e4937aSGao Xiang preload_compressed_pages(clt, MNGD_MAPPING(sbi), 62247e4937aSGao Xiang cache_strategy, pagepool); 62347e4937aSGao Xiang 62447e4937aSGao Xiang hitted: 625dc76ea8cSGao Xiang /* 626dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 627dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 628dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 629dc76ea8cSGao Xiang * for inplace I/O or pagevec (should be processed in strict order.) 630dc76ea8cSGao Xiang */ 631dc76ea8cSGao Xiang tight &= (clt->mode >= COLLECT_PRIMARY_HOOKED && 632dc76ea8cSGao Xiang clt->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 633dc76ea8cSGao Xiang 63447e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 6358d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 63647e4937aSGao Xiang zero_user_segment(page, cur, end); 63747e4937aSGao Xiang goto next_part; 63847e4937aSGao Xiang } 63947e4937aSGao Xiang 64047e4937aSGao Xiang /* let's derive page type */ 64147e4937aSGao Xiang page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 64247e4937aSGao Xiang (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 64347e4937aSGao Xiang (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 64447e4937aSGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 64547e4937aSGao Xiang 64647e4937aSGao Xiang if (cur) 64747e4937aSGao Xiang tight &= (clt->mode >= COLLECT_PRIMARY_FOLLOWED); 64847e4937aSGao Xiang 64947e4937aSGao Xiang retry: 65047e4937aSGao Xiang err = z_erofs_attach_page(clt, page, page_type); 65147e4937aSGao Xiang /* should allocate an additional staging page for pagevec */ 65247e4937aSGao Xiang if (err == -EAGAIN) { 65347e4937aSGao Xiang struct page *const newpage = 6545ddcee1fSGao Xiang erofs_allocpage(pagepool, GFP_NOFS | __GFP_NOFAIL); 65547e4937aSGao Xiang 6565ddcee1fSGao Xiang newpage->mapping = Z_EROFS_MAPPING_STAGING; 65747e4937aSGao Xiang err = z_erofs_attach_page(clt, newpage, 65847e4937aSGao Xiang Z_EROFS_PAGE_TYPE_EXCLUSIVE); 6598d8a09b0SGao Xiang if (!err) 66047e4937aSGao Xiang goto retry; 66147e4937aSGao Xiang } 66247e4937aSGao Xiang 6638d8a09b0SGao Xiang if (err) 66447e4937aSGao Xiang goto err_out; 66547e4937aSGao Xiang 66647e4937aSGao Xiang index = page->index - (map->m_la >> PAGE_SHIFT); 66747e4937aSGao Xiang 66847e4937aSGao Xiang z_erofs_onlinepage_fixup(page, index, true); 66947e4937aSGao Xiang 67047e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 67147e4937aSGao Xiang ++spiltted; 67247e4937aSGao Xiang /* also update nr_pages */ 67347e4937aSGao Xiang clt->cl->nr_pages = max_t(pgoff_t, clt->cl->nr_pages, index + 1); 67447e4937aSGao Xiang next_part: 67547e4937aSGao Xiang /* can be used for verification */ 67647e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 67747e4937aSGao Xiang 67847e4937aSGao Xiang end = cur; 67947e4937aSGao Xiang if (end > 0) 68047e4937aSGao Xiang goto repeat; 68147e4937aSGao Xiang 68247e4937aSGao Xiang out: 68347e4937aSGao Xiang z_erofs_onlinepage_endio(page); 68447e4937aSGao Xiang 6854f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 68647e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 68747e4937aSGao Xiang return err; 68847e4937aSGao Xiang 68947e4937aSGao Xiang /* if some error occurred while processing this page */ 69047e4937aSGao Xiang err_out: 69147e4937aSGao Xiang SetPageError(page); 69247e4937aSGao Xiang goto out; 69347e4937aSGao Xiang } 69447e4937aSGao Xiang 695a4b1fab1SGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 696a4b1fab1SGao Xiang bool sync, int bios) 69747e4937aSGao Xiang { 698a4b1fab1SGao Xiang /* wake up the caller thread for sync decompression */ 699a4b1fab1SGao Xiang if (sync) { 70047e4937aSGao Xiang unsigned long flags; 70147e4937aSGao Xiang 70247e4937aSGao Xiang spin_lock_irqsave(&io->u.wait.lock, flags); 70347e4937aSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 70447e4937aSGao Xiang wake_up_locked(&io->u.wait); 70547e4937aSGao Xiang spin_unlock_irqrestore(&io->u.wait.lock, flags); 70647e4937aSGao Xiang return; 70747e4937aSGao Xiang } 70847e4937aSGao Xiang 70947e4937aSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 71047e4937aSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 71147e4937aSGao Xiang } 71247e4937aSGao Xiang 7130c638f70SGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 71447e4937aSGao Xiang { 715a4b1fab1SGao Xiang tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 716a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 71747e4937aSGao Xiang blk_status_t err = bio->bi_status; 71847e4937aSGao Xiang struct bio_vec *bvec; 71947e4937aSGao Xiang struct bvec_iter_all iter_all; 72047e4937aSGao Xiang 72147e4937aSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 72247e4937aSGao Xiang struct page *page = bvec->bv_page; 72347e4937aSGao Xiang 72447e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 72547e4937aSGao Xiang DBG_BUGON(!page->mapping); 72647e4937aSGao Xiang 7278d8a09b0SGao Xiang if (err) 72847e4937aSGao Xiang SetPageError(page); 72947e4937aSGao Xiang 730a4b1fab1SGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 731a4b1fab1SGao Xiang if (!err) 732a4b1fab1SGao Xiang SetPageUptodate(page); 73347e4937aSGao Xiang unlock_page(page); 73447e4937aSGao Xiang } 735a4b1fab1SGao Xiang } 736a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 73747e4937aSGao Xiang bio_put(bio); 73847e4937aSGao Xiang } 73947e4937aSGao Xiang 74047e4937aSGao Xiang static int z_erofs_decompress_pcluster(struct super_block *sb, 74147e4937aSGao Xiang struct z_erofs_pcluster *pcl, 74247e4937aSGao Xiang struct list_head *pagepool) 74347e4937aSGao Xiang { 74447e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 74547e4937aSGao Xiang const unsigned int clusterpages = BIT(pcl->clusterbits); 74647e4937aSGao Xiang struct z_erofs_pagevec_ctor ctor; 74747e4937aSGao Xiang unsigned int i, outputsize, llen, nr_pages; 74847e4937aSGao Xiang struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 74947e4937aSGao Xiang struct page **pages, **compressed_pages, *page; 75047e4937aSGao Xiang 75147e4937aSGao Xiang enum z_erofs_page_type page_type; 75247e4937aSGao Xiang bool overlapped, partial; 75347e4937aSGao Xiang struct z_erofs_collection *cl; 75447e4937aSGao Xiang int err; 75547e4937aSGao Xiang 75647e4937aSGao Xiang might_sleep(); 75747e4937aSGao Xiang cl = z_erofs_primarycollection(pcl); 75847e4937aSGao Xiang DBG_BUGON(!READ_ONCE(cl->nr_pages)); 75947e4937aSGao Xiang 76047e4937aSGao Xiang mutex_lock(&cl->lock); 76147e4937aSGao Xiang nr_pages = cl->nr_pages; 76247e4937aSGao Xiang 7638d8a09b0SGao Xiang if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 76447e4937aSGao Xiang pages = pages_onstack; 76547e4937aSGao Xiang } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 76647e4937aSGao Xiang mutex_trylock(&z_pagemap_global_lock)) { 76747e4937aSGao Xiang pages = z_pagemap_global; 76847e4937aSGao Xiang } else { 76947e4937aSGao Xiang gfp_t gfp_flags = GFP_KERNEL; 77047e4937aSGao Xiang 77147e4937aSGao Xiang if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 77247e4937aSGao Xiang gfp_flags |= __GFP_NOFAIL; 77347e4937aSGao Xiang 77447e4937aSGao Xiang pages = kvmalloc_array(nr_pages, sizeof(struct page *), 77547e4937aSGao Xiang gfp_flags); 77647e4937aSGao Xiang 77747e4937aSGao Xiang /* fallback to global pagemap for the lowmem scenario */ 7788d8a09b0SGao Xiang if (!pages) { 77947e4937aSGao Xiang mutex_lock(&z_pagemap_global_lock); 78047e4937aSGao Xiang pages = z_pagemap_global; 78147e4937aSGao Xiang } 78247e4937aSGao Xiang } 78347e4937aSGao Xiang 78447e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) 78547e4937aSGao Xiang pages[i] = NULL; 78647e4937aSGao Xiang 78747e4937aSGao Xiang err = 0; 78847e4937aSGao Xiang z_erofs_pagevec_ctor_init(&ctor, Z_EROFS_NR_INLINE_PAGEVECS, 78947e4937aSGao Xiang cl->pagevec, 0); 79047e4937aSGao Xiang 79147e4937aSGao Xiang for (i = 0; i < cl->vcnt; ++i) { 79247e4937aSGao Xiang unsigned int pagenr; 79347e4937aSGao Xiang 79447e4937aSGao Xiang page = z_erofs_pagevec_dequeue(&ctor, &page_type); 79547e4937aSGao Xiang 79647e4937aSGao Xiang /* all pages in pagevec ought to be valid */ 79747e4937aSGao Xiang DBG_BUGON(!page); 79847e4937aSGao Xiang DBG_BUGON(!page->mapping); 79947e4937aSGao Xiang 80047e4937aSGao Xiang if (z_erofs_put_stagingpage(pagepool, page)) 80147e4937aSGao Xiang continue; 80247e4937aSGao Xiang 80347e4937aSGao Xiang if (page_type == Z_EROFS_VLE_PAGE_TYPE_HEAD) 80447e4937aSGao Xiang pagenr = 0; 80547e4937aSGao Xiang else 80647e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 80747e4937aSGao Xiang 80847e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 80947e4937aSGao Xiang 81047e4937aSGao Xiang /* 81147e4937aSGao Xiang * currently EROFS doesn't support multiref(dedup), 81247e4937aSGao Xiang * so here erroring out one multiref page. 81347e4937aSGao Xiang */ 8148d8a09b0SGao Xiang if (pages[pagenr]) { 81547e4937aSGao Xiang DBG_BUGON(1); 81647e4937aSGao Xiang SetPageError(pages[pagenr]); 81747e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 81847e4937aSGao Xiang err = -EFSCORRUPTED; 81947e4937aSGao Xiang } 82047e4937aSGao Xiang pages[pagenr] = page; 82147e4937aSGao Xiang } 82247e4937aSGao Xiang z_erofs_pagevec_ctor_exit(&ctor, true); 82347e4937aSGao Xiang 82447e4937aSGao Xiang overlapped = false; 82547e4937aSGao Xiang compressed_pages = pcl->compressed_pages; 82647e4937aSGao Xiang 82747e4937aSGao Xiang for (i = 0; i < clusterpages; ++i) { 82847e4937aSGao Xiang unsigned int pagenr; 82947e4937aSGao Xiang 83047e4937aSGao Xiang page = compressed_pages[i]; 83147e4937aSGao Xiang 83247e4937aSGao Xiang /* all compressed pages ought to be valid */ 83347e4937aSGao Xiang DBG_BUGON(!page); 83447e4937aSGao Xiang DBG_BUGON(!page->mapping); 83547e4937aSGao Xiang 83647e4937aSGao Xiang if (!z_erofs_page_is_staging(page)) { 83747e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) { 8388d8a09b0SGao Xiang if (!PageUptodate(page)) 83947e4937aSGao Xiang err = -EIO; 84047e4937aSGao Xiang continue; 84147e4937aSGao Xiang } 84247e4937aSGao Xiang 84347e4937aSGao Xiang /* 84447e4937aSGao Xiang * only if non-head page can be selected 84547e4937aSGao Xiang * for inplace decompression 84647e4937aSGao Xiang */ 84747e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 84847e4937aSGao Xiang 84947e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 8508d8a09b0SGao Xiang if (pages[pagenr]) { 85147e4937aSGao Xiang DBG_BUGON(1); 85247e4937aSGao Xiang SetPageError(pages[pagenr]); 85347e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 85447e4937aSGao Xiang err = -EFSCORRUPTED; 85547e4937aSGao Xiang } 85647e4937aSGao Xiang pages[pagenr] = page; 85747e4937aSGao Xiang 85847e4937aSGao Xiang overlapped = true; 85947e4937aSGao Xiang } 86047e4937aSGao Xiang 86147e4937aSGao Xiang /* PG_error needs checking for inplaced and staging pages */ 8628d8a09b0SGao Xiang if (PageError(page)) { 86347e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 86447e4937aSGao Xiang err = -EIO; 86547e4937aSGao Xiang } 86647e4937aSGao Xiang } 86747e4937aSGao Xiang 8688d8a09b0SGao Xiang if (err) 86947e4937aSGao Xiang goto out; 87047e4937aSGao Xiang 87147e4937aSGao Xiang llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 87247e4937aSGao Xiang if (nr_pages << PAGE_SHIFT >= cl->pageofs + llen) { 87347e4937aSGao Xiang outputsize = llen; 87447e4937aSGao Xiang partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 87547e4937aSGao Xiang } else { 87647e4937aSGao Xiang outputsize = (nr_pages << PAGE_SHIFT) - cl->pageofs; 87747e4937aSGao Xiang partial = true; 87847e4937aSGao Xiang } 87947e4937aSGao Xiang 88047e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 88147e4937aSGao Xiang .sb = sb, 88247e4937aSGao Xiang .in = compressed_pages, 88347e4937aSGao Xiang .out = pages, 88447e4937aSGao Xiang .pageofs_out = cl->pageofs, 88547e4937aSGao Xiang .inputsize = PAGE_SIZE, 88647e4937aSGao Xiang .outputsize = outputsize, 88747e4937aSGao Xiang .alg = pcl->algorithmformat, 88847e4937aSGao Xiang .inplace_io = overlapped, 88947e4937aSGao Xiang .partial_decoding = partial 89047e4937aSGao Xiang }, pagepool); 89147e4937aSGao Xiang 89247e4937aSGao Xiang out: 89347e4937aSGao Xiang /* must handle all compressed pages before endding pages */ 89447e4937aSGao Xiang for (i = 0; i < clusterpages; ++i) { 89547e4937aSGao Xiang page = compressed_pages[i]; 89647e4937aSGao Xiang 89747e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 89847e4937aSGao Xiang continue; 89947e4937aSGao Xiang 90047e4937aSGao Xiang /* recycle all individual staging pages */ 90147e4937aSGao Xiang (void)z_erofs_put_stagingpage(pagepool, page); 90247e4937aSGao Xiang 90347e4937aSGao Xiang WRITE_ONCE(compressed_pages[i], NULL); 90447e4937aSGao Xiang } 90547e4937aSGao Xiang 90647e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) { 90747e4937aSGao Xiang page = pages[i]; 90847e4937aSGao Xiang if (!page) 90947e4937aSGao Xiang continue; 91047e4937aSGao Xiang 91147e4937aSGao Xiang DBG_BUGON(!page->mapping); 91247e4937aSGao Xiang 91347e4937aSGao Xiang /* recycle all individual staging pages */ 91447e4937aSGao Xiang if (z_erofs_put_stagingpage(pagepool, page)) 91547e4937aSGao Xiang continue; 91647e4937aSGao Xiang 9178d8a09b0SGao Xiang if (err < 0) 91847e4937aSGao Xiang SetPageError(page); 91947e4937aSGao Xiang 92047e4937aSGao Xiang z_erofs_onlinepage_endio(page); 92147e4937aSGao Xiang } 92247e4937aSGao Xiang 92347e4937aSGao Xiang if (pages == z_pagemap_global) 92447e4937aSGao Xiang mutex_unlock(&z_pagemap_global_lock); 9258d8a09b0SGao Xiang else if (pages != pages_onstack) 92647e4937aSGao Xiang kvfree(pages); 92747e4937aSGao Xiang 92847e4937aSGao Xiang cl->nr_pages = 0; 92947e4937aSGao Xiang cl->vcnt = 0; 93047e4937aSGao Xiang 93147e4937aSGao Xiang /* all cl locks MUST be taken before the following line */ 93247e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 93347e4937aSGao Xiang 93447e4937aSGao Xiang /* all cl locks SHOULD be released right now */ 93547e4937aSGao Xiang mutex_unlock(&cl->lock); 93647e4937aSGao Xiang 93747e4937aSGao Xiang z_erofs_collection_put(cl); 93847e4937aSGao Xiang return err; 93947e4937aSGao Xiang } 94047e4937aSGao Xiang 9410c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 94247e4937aSGao Xiang struct list_head *pagepool) 94347e4937aSGao Xiang { 94447e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 94547e4937aSGao Xiang 94647e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 94747e4937aSGao Xiang struct z_erofs_pcluster *pcl; 94847e4937aSGao Xiang 94947e4937aSGao Xiang /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 95047e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 95147e4937aSGao Xiang 95247e4937aSGao Xiang /* no possible that 'owned' equals NULL */ 95347e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 95447e4937aSGao Xiang 95547e4937aSGao Xiang pcl = container_of(owned, struct z_erofs_pcluster, next); 95647e4937aSGao Xiang owned = READ_ONCE(pcl->next); 95747e4937aSGao Xiang 958a4b1fab1SGao Xiang z_erofs_decompress_pcluster(io->sb, pcl, pagepool); 95947e4937aSGao Xiang } 96047e4937aSGao Xiang } 96147e4937aSGao Xiang 9620c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 96347e4937aSGao Xiang { 964a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 965a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 96647e4937aSGao Xiang LIST_HEAD(pagepool); 96747e4937aSGao Xiang 968a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 9690c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 97047e4937aSGao Xiang 97147e4937aSGao Xiang put_pages_list(&pagepool); 972a4b1fab1SGao Xiang kvfree(bgq); 97347e4937aSGao Xiang } 97447e4937aSGao Xiang 97547e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 97647e4937aSGao Xiang unsigned int nr, 97747e4937aSGao Xiang struct list_head *pagepool, 97847e4937aSGao Xiang struct address_space *mc, 97947e4937aSGao Xiang gfp_t gfp) 98047e4937aSGao Xiang { 98147e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 98247e4937aSGao Xiang bool tocache = false; 98347e4937aSGao Xiang 98447e4937aSGao Xiang struct address_space *mapping; 98547e4937aSGao Xiang struct page *oldpage, *page; 98647e4937aSGao Xiang 98747e4937aSGao Xiang compressed_page_t t; 98847e4937aSGao Xiang int justfound; 98947e4937aSGao Xiang 99047e4937aSGao Xiang repeat: 99147e4937aSGao Xiang page = READ_ONCE(pcl->compressed_pages[nr]); 99247e4937aSGao Xiang oldpage = page; 99347e4937aSGao Xiang 99447e4937aSGao Xiang if (!page) 99547e4937aSGao Xiang goto out_allocpage; 99647e4937aSGao Xiang 99747e4937aSGao Xiang /* 99847e4937aSGao Xiang * the cached page has not been allocated and 99947e4937aSGao Xiang * an placeholder is out there, prepare it now. 100047e4937aSGao Xiang */ 1001bda17a45SGao Xiang if (page == PAGE_UNALLOCATED) { 100247e4937aSGao Xiang tocache = true; 100347e4937aSGao Xiang goto out_allocpage; 100447e4937aSGao Xiang } 100547e4937aSGao Xiang 100647e4937aSGao Xiang /* process the target tagged pointer */ 100747e4937aSGao Xiang t = tagptr_init(compressed_page_t, page); 100847e4937aSGao Xiang justfound = tagptr_unfold_tags(t); 100947e4937aSGao Xiang page = tagptr_unfold_ptr(t); 101047e4937aSGao Xiang 101147e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 101247e4937aSGao Xiang 101347e4937aSGao Xiang /* 101447e4937aSGao Xiang * unmanaged (file) pages are all locked solidly, 101547e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 101647e4937aSGao Xiang */ 101747e4937aSGao Xiang if (mapping && mapping != mc) 101847e4937aSGao Xiang /* ought to be unmanaged pages */ 101947e4937aSGao Xiang goto out; 102047e4937aSGao Xiang 102147e4937aSGao Xiang lock_page(page); 102247e4937aSGao Xiang 102347e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 102447e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 102547e4937aSGao Xiang 102647e4937aSGao Xiang /* the page is still in manage cache */ 102747e4937aSGao Xiang if (page->mapping == mc) { 102847e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 102947e4937aSGao Xiang 103047e4937aSGao Xiang ClearPageError(page); 103147e4937aSGao Xiang if (!PagePrivate(page)) { 103247e4937aSGao Xiang /* 103347e4937aSGao Xiang * impossible to be !PagePrivate(page) for 103447e4937aSGao Xiang * the current restriction as well if 103547e4937aSGao Xiang * the page is already in compressed_pages[]. 103647e4937aSGao Xiang */ 103747e4937aSGao Xiang DBG_BUGON(!justfound); 103847e4937aSGao Xiang 103947e4937aSGao Xiang justfound = 0; 104047e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 104147e4937aSGao Xiang SetPagePrivate(page); 104247e4937aSGao Xiang } 104347e4937aSGao Xiang 104447e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 104547e4937aSGao Xiang if (PageUptodate(page)) { 104647e4937aSGao Xiang unlock_page(page); 104747e4937aSGao Xiang page = NULL; 104847e4937aSGao Xiang } 104947e4937aSGao Xiang goto out; 105047e4937aSGao Xiang } 105147e4937aSGao Xiang 105247e4937aSGao Xiang /* 105347e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 105447e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 105547e4937aSGao Xiang */ 105647e4937aSGao Xiang DBG_BUGON(page->mapping); 105747e4937aSGao Xiang DBG_BUGON(!justfound); 105847e4937aSGao Xiang 105947e4937aSGao Xiang tocache = true; 106047e4937aSGao Xiang unlock_page(page); 106147e4937aSGao Xiang put_page(page); 106247e4937aSGao Xiang out_allocpage: 10635ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 10645ddcee1fSGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 10655ddcee1fSGao Xiang /* non-LRU / non-movable temporary page is needed */ 106647e4937aSGao Xiang page->mapping = Z_EROFS_MAPPING_STAGING; 10675ddcee1fSGao Xiang tocache = false; 106847e4937aSGao Xiang } 106947e4937aSGao Xiang 10705ddcee1fSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_pages[nr], oldpage, page)) { 10715ddcee1fSGao Xiang if (tocache) { 10725ddcee1fSGao Xiang /* since it added to managed cache successfully */ 10735ddcee1fSGao Xiang unlock_page(page); 10745ddcee1fSGao Xiang put_page(page); 10755ddcee1fSGao Xiang } else { 10765ddcee1fSGao Xiang list_add(&page->lru, pagepool); 10775ddcee1fSGao Xiang } 10785ddcee1fSGao Xiang cond_resched(); 10795ddcee1fSGao Xiang goto repeat; 10805ddcee1fSGao Xiang } 108147e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 108247e4937aSGao Xiang SetPagePrivate(page); 108347e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 108447e4937aSGao Xiang return page; 108547e4937aSGao Xiang } 108647e4937aSGao Xiang 1087a4b1fab1SGao Xiang static struct z_erofs_decompressqueue * 1088a4b1fab1SGao Xiang jobqueue_init(struct super_block *sb, 1089a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 109047e4937aSGao Xiang { 1091a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 109247e4937aSGao Xiang 1093a4b1fab1SGao Xiang if (fg && !*fg) { 1094a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1095a4b1fab1SGao Xiang if (!q) { 1096a4b1fab1SGao Xiang *fg = true; 1097a4b1fab1SGao Xiang goto fg_out; 109847e4937aSGao Xiang } 10990c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1100a4b1fab1SGao Xiang } else { 1101a4b1fab1SGao Xiang fg_out: 1102a4b1fab1SGao Xiang q = fgq; 1103a4b1fab1SGao Xiang init_waitqueue_head(&fgq->u.wait); 1104a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 1105a4b1fab1SGao Xiang } 1106a4b1fab1SGao Xiang q->sb = sb; 1107a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1108a4b1fab1SGao Xiang return q; 110947e4937aSGao Xiang } 111047e4937aSGao Xiang 111147e4937aSGao Xiang /* define decompression jobqueue types */ 111247e4937aSGao Xiang enum { 111347e4937aSGao Xiang JQ_BYPASS, 111447e4937aSGao Xiang JQ_SUBMIT, 111547e4937aSGao Xiang NR_JOBQUEUES, 111647e4937aSGao Xiang }; 111747e4937aSGao Xiang 111847e4937aSGao Xiang static void *jobqueueset_init(struct super_block *sb, 1119a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[], 1120a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 112147e4937aSGao Xiang { 112247e4937aSGao Xiang /* 112347e4937aSGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 112447e4937aSGao Xiang * no need to read from device for all pclusters in this queue. 112547e4937aSGao Xiang */ 1126a4b1fab1SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1127a4b1fab1SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 112847e4937aSGao Xiang 1129a4b1fab1SGao Xiang return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 113047e4937aSGao Xiang } 113147e4937aSGao Xiang 113247e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 113347e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 113447e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 113547e4937aSGao Xiang { 113647e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 113747e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 113847e4937aSGao Xiang 113947e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 114047e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 114147e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 114247e4937aSGao Xiang 114347e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 114447e4937aSGao Xiang 114547e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 114647e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 114747e4937aSGao Xiang 114847e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 114947e4937aSGao Xiang } 115047e4937aSGao Xiang 1151a4b1fab1SGao Xiang static bool postsubmit_is_all_bypassed(struct z_erofs_decompressqueue *q[], 1152a4b1fab1SGao Xiang unsigned int nr_bios, bool force_fg) 115347e4937aSGao Xiang { 115447e4937aSGao Xiang /* 115547e4937aSGao Xiang * although background is preferred, no one is pending for submission. 115647e4937aSGao Xiang * don't issue workqueue for decompression but drop it directly instead. 115747e4937aSGao Xiang */ 115847e4937aSGao Xiang if (force_fg || nr_bios) 115947e4937aSGao Xiang return false; 116047e4937aSGao Xiang 1161a4b1fab1SGao Xiang kvfree(q[JQ_SUBMIT]); 116247e4937aSGao Xiang return true; 116347e4937aSGao Xiang } 116447e4937aSGao Xiang 11650c638f70SGao Xiang static bool z_erofs_submit_queue(struct super_block *sb, 116647e4937aSGao Xiang z_erofs_next_pcluster_t owned_head, 116747e4937aSGao Xiang struct list_head *pagepool, 1168a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1169a4b1fab1SGao Xiang bool *force_fg) 117047e4937aSGao Xiang { 1171bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 117247e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1173a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 117447e4937aSGao Xiang struct bio *bio; 117547e4937aSGao Xiang void *bi_private; 117647e4937aSGao Xiang /* since bio will be NULL, no need to initialize last_index */ 117747e4937aSGao Xiang pgoff_t uninitialized_var(last_index); 117847e4937aSGao Xiang bool force_submit = false; 117947e4937aSGao Xiang unsigned int nr_bios; 118047e4937aSGao Xiang 11818d8a09b0SGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 118247e4937aSGao Xiang return false; 118347e4937aSGao Xiang 118447e4937aSGao Xiang force_submit = false; 118547e4937aSGao Xiang bio = NULL; 118647e4937aSGao Xiang nr_bios = 0; 1187a4b1fab1SGao Xiang bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1188a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1189a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 119047e4937aSGao Xiang 119147e4937aSGao Xiang /* by default, all need io submission */ 119247e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 119347e4937aSGao Xiang 119447e4937aSGao Xiang do { 119547e4937aSGao Xiang struct z_erofs_pcluster *pcl; 119647e4937aSGao Xiang unsigned int clusterpages; 119747e4937aSGao Xiang pgoff_t first_index; 119847e4937aSGao Xiang struct page *page; 119947e4937aSGao Xiang unsigned int i = 0, bypass = 0; 120047e4937aSGao Xiang int err; 120147e4937aSGao Xiang 120247e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 120347e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 120447e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 120547e4937aSGao Xiang 120647e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 120747e4937aSGao Xiang 120847e4937aSGao Xiang clusterpages = BIT(pcl->clusterbits); 120947e4937aSGao Xiang 121047e4937aSGao Xiang /* close the main owned chain at first */ 121147e4937aSGao Xiang owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 121247e4937aSGao Xiang Z_EROFS_PCLUSTER_TAIL_CLOSED); 121347e4937aSGao Xiang 121447e4937aSGao Xiang first_index = pcl->obj.index; 121547e4937aSGao Xiang force_submit |= (first_index != last_index + 1); 121647e4937aSGao Xiang 121747e4937aSGao Xiang repeat: 121847e4937aSGao Xiang page = pickup_page_for_submission(pcl, i, pagepool, 121947e4937aSGao Xiang MNGD_MAPPING(sbi), 122047e4937aSGao Xiang GFP_NOFS); 122147e4937aSGao Xiang if (!page) { 122247e4937aSGao Xiang force_submit = true; 122347e4937aSGao Xiang ++bypass; 122447e4937aSGao Xiang goto skippage; 122547e4937aSGao Xiang } 122647e4937aSGao Xiang 122747e4937aSGao Xiang if (bio && force_submit) { 122847e4937aSGao Xiang submit_bio_retry: 122994e4e153SGao Xiang submit_bio(bio); 123047e4937aSGao Xiang bio = NULL; 123147e4937aSGao Xiang } 123247e4937aSGao Xiang 123347e4937aSGao Xiang if (!bio) { 1234a5c0b780SGao Xiang bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); 1235a5c0b780SGao Xiang 12360c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1237a5c0b780SGao Xiang bio_set_dev(bio, sb->s_bdev); 1238a5c0b780SGao Xiang bio->bi_iter.bi_sector = (sector_t)(first_index + i) << 1239a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1240a5c0b780SGao Xiang bio->bi_private = bi_private; 124194e4e153SGao Xiang bio->bi_opf = REQ_OP_READ; 124294e4e153SGao Xiang 124347e4937aSGao Xiang ++nr_bios; 124447e4937aSGao Xiang } 124547e4937aSGao Xiang 124647e4937aSGao Xiang err = bio_add_page(bio, page, PAGE_SIZE, 0); 124747e4937aSGao Xiang if (err < PAGE_SIZE) 124847e4937aSGao Xiang goto submit_bio_retry; 124947e4937aSGao Xiang 125047e4937aSGao Xiang force_submit = false; 125147e4937aSGao Xiang last_index = first_index + i; 125247e4937aSGao Xiang skippage: 125347e4937aSGao Xiang if (++i < clusterpages) 125447e4937aSGao Xiang goto repeat; 125547e4937aSGao Xiang 125647e4937aSGao Xiang if (bypass < clusterpages) 125747e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 125847e4937aSGao Xiang else 125947e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 126047e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 126147e4937aSGao Xiang 126247e4937aSGao Xiang if (bio) 126394e4e153SGao Xiang submit_bio(bio); 126447e4937aSGao Xiang 1265a4b1fab1SGao Xiang if (postsubmit_is_all_bypassed(q, nr_bios, *force_fg)) 126647e4937aSGao Xiang return true; 126747e4937aSGao Xiang 1268a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 126947e4937aSGao Xiang return true; 127047e4937aSGao Xiang } 127147e4937aSGao Xiang 12720c638f70SGao Xiang static void z_erofs_runqueue(struct super_block *sb, 127347e4937aSGao Xiang struct z_erofs_collector *clt, 12740c638f70SGao Xiang struct list_head *pagepool, bool force_fg) 127547e4937aSGao Xiang { 1276a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 127747e4937aSGao Xiang 12780c638f70SGao Xiang if (!z_erofs_submit_queue(sb, clt->owned_head, 1279a4b1fab1SGao Xiang pagepool, io, &force_fg)) 128047e4937aSGao Xiang return; 128147e4937aSGao Xiang 12820c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 12830c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 128447e4937aSGao Xiang 128547e4937aSGao Xiang if (!force_fg) 128647e4937aSGao Xiang return; 128747e4937aSGao Xiang 128847e4937aSGao Xiang /* wait until all bios are completed */ 1289a93f8c36SGao Xiang io_wait_event(io[JQ_SUBMIT].u.wait, 129047e4937aSGao Xiang !atomic_read(&io[JQ_SUBMIT].pending_bios)); 129147e4937aSGao Xiang 12920c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 12930c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 129447e4937aSGao Xiang } 129547e4937aSGao Xiang 12960c638f70SGao Xiang static int z_erofs_readpage(struct file *file, struct page *page) 129747e4937aSGao Xiang { 129847e4937aSGao Xiang struct inode *const inode = page->mapping->host; 129947e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 130047e4937aSGao Xiang int err; 130147e4937aSGao Xiang LIST_HEAD(pagepool); 130247e4937aSGao Xiang 130347e4937aSGao Xiang trace_erofs_readpage(page, false); 130447e4937aSGao Xiang 130547e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 130647e4937aSGao Xiang 130747e4937aSGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 130847e4937aSGao Xiang (void)z_erofs_collector_end(&f.clt); 130947e4937aSGao Xiang 131047e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 13110c638f70SGao Xiang z_erofs_runqueue(inode->i_sb, &f.clt, &pagepool, true); 131247e4937aSGao Xiang 131347e4937aSGao Xiang if (err) 13144f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 131547e4937aSGao Xiang 131647e4937aSGao Xiang if (f.map.mpage) 131747e4937aSGao Xiang put_page(f.map.mpage); 131847e4937aSGao Xiang 131947e4937aSGao Xiang /* clean up the remaining free pages */ 132047e4937aSGao Xiang put_pages_list(&pagepool); 132147e4937aSGao Xiang return err; 132247e4937aSGao Xiang } 132347e4937aSGao Xiang 132447e4937aSGao Xiang static bool should_decompress_synchronously(struct erofs_sb_info *sbi, 132547e4937aSGao Xiang unsigned int nr) 132647e4937aSGao Xiang { 132747e4937aSGao Xiang return nr <= sbi->max_sync_decompress_pages; 132847e4937aSGao Xiang } 132947e4937aSGao Xiang 13300c638f70SGao Xiang static int z_erofs_readpages(struct file *filp, struct address_space *mapping, 13310c638f70SGao Xiang struct list_head *pages, unsigned int nr_pages) 133247e4937aSGao Xiang { 133347e4937aSGao Xiang struct inode *const inode = mapping->host; 133447e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 133547e4937aSGao Xiang 133647e4937aSGao Xiang bool sync = should_decompress_synchronously(sbi, nr_pages); 133747e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 133847e4937aSGao Xiang gfp_t gfp = mapping_gfp_constraint(mapping, GFP_KERNEL); 133947e4937aSGao Xiang struct page *head = NULL; 134047e4937aSGao Xiang LIST_HEAD(pagepool); 134147e4937aSGao Xiang 134247e4937aSGao Xiang trace_erofs_readpages(mapping->host, lru_to_page(pages), 134347e4937aSGao Xiang nr_pages, false); 134447e4937aSGao Xiang 134547e4937aSGao Xiang f.headoffset = (erofs_off_t)lru_to_page(pages)->index << PAGE_SHIFT; 134647e4937aSGao Xiang 134747e4937aSGao Xiang for (; nr_pages; --nr_pages) { 134847e4937aSGao Xiang struct page *page = lru_to_page(pages); 134947e4937aSGao Xiang 135047e4937aSGao Xiang prefetchw(&page->flags); 135147e4937aSGao Xiang list_del(&page->lru); 135247e4937aSGao Xiang 135347e4937aSGao Xiang /* 135447e4937aSGao Xiang * A pure asynchronous readahead is indicated if 135547e4937aSGao Xiang * a PG_readahead marked page is hitted at first. 135647e4937aSGao Xiang * Let's also do asynchronous decompression for this case. 135747e4937aSGao Xiang */ 135847e4937aSGao Xiang sync &= !(PageReadahead(page) && !head); 135947e4937aSGao Xiang 136047e4937aSGao Xiang if (add_to_page_cache_lru(page, mapping, page->index, gfp)) { 136147e4937aSGao Xiang list_add(&page->lru, &pagepool); 136247e4937aSGao Xiang continue; 136347e4937aSGao Xiang } 136447e4937aSGao Xiang 136547e4937aSGao Xiang set_page_private(page, (unsigned long)head); 136647e4937aSGao Xiang head = page; 136747e4937aSGao Xiang } 136847e4937aSGao Xiang 136947e4937aSGao Xiang while (head) { 137047e4937aSGao Xiang struct page *page = head; 137147e4937aSGao Xiang int err; 137247e4937aSGao Xiang 137347e4937aSGao Xiang /* traversal in reverse order */ 137447e4937aSGao Xiang head = (void *)page_private(page); 137547e4937aSGao Xiang 137647e4937aSGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1377a5876e24SGao Xiang if (err) 13784f761fa2SGao Xiang erofs_err(inode->i_sb, 13794f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 13804f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 138147e4937aSGao Xiang put_page(page); 138247e4937aSGao Xiang } 138347e4937aSGao Xiang 138447e4937aSGao Xiang (void)z_erofs_collector_end(&f.clt); 138547e4937aSGao Xiang 13860c638f70SGao Xiang z_erofs_runqueue(inode->i_sb, &f.clt, &pagepool, sync); 138747e4937aSGao Xiang 138847e4937aSGao Xiang if (f.map.mpage) 138947e4937aSGao Xiang put_page(f.map.mpage); 139047e4937aSGao Xiang 139147e4937aSGao Xiang /* clean up the remaining free pages */ 139247e4937aSGao Xiang put_pages_list(&pagepool); 139347e4937aSGao Xiang return 0; 139447e4937aSGao Xiang } 139547e4937aSGao Xiang 13960c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 13970c638f70SGao Xiang .readpage = z_erofs_readpage, 13980c638f70SGao Xiang .readpages = z_erofs_readpages, 139947e4937aSGao Xiang }; 140047e4937aSGao Xiang 1401