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/ 506a304cdSGao Xiang * Copyright (C) 2022 Alibaba Cloud 647e4937aSGao Xiang */ 747e4937aSGao Xiang #include "compress.h" 899486c51SChristoph Hellwig #include <linux/psi.h> 93fffb589SSandeep Dhavale #include <linux/cpuhotplug.h> 1047e4937aSGao Xiang #include <trace/events/erofs.h> 1147e4937aSGao Xiang 12a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_MAX_PAGES (Z_EROFS_PCLUSTER_MAX_SIZE / PAGE_SIZE) 13a9a94d93SGao Xiang #define Z_EROFS_INLINE_BVECS 2 14a9a94d93SGao Xiang 15a9a94d93SGao Xiang /* 16a9a94d93SGao Xiang * let's leave a type here in case of introducing 17a9a94d93SGao Xiang * another tagged pointer later. 18a9a94d93SGao Xiang */ 19a9a94d93SGao Xiang typedef void *z_erofs_next_pcluster_t; 20a9a94d93SGao Xiang 21a9a94d93SGao Xiang struct z_erofs_bvec { 22a9a94d93SGao Xiang struct page *page; 23a9a94d93SGao Xiang int offset; 24a9a94d93SGao Xiang unsigned int end; 25a9a94d93SGao Xiang }; 26a9a94d93SGao Xiang 27a9a94d93SGao Xiang #define __Z_EROFS_BVSET(name, total) \ 28a9a94d93SGao Xiang struct name { \ 29a9a94d93SGao Xiang /* point to the next page which contains the following bvecs */ \ 30a9a94d93SGao Xiang struct page *nextpage; \ 31a9a94d93SGao Xiang struct z_erofs_bvec bvec[total]; \ 32a9a94d93SGao Xiang } 33a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset,); 34a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset_inline, Z_EROFS_INLINE_BVECS); 35a9a94d93SGao Xiang 36a9a94d93SGao Xiang /* 37a9a94d93SGao Xiang * Structure fields follow one of the following exclusion rules. 38a9a94d93SGao Xiang * 39a9a94d93SGao Xiang * I: Modifiable by initialization/destruction paths and read-only 40a9a94d93SGao Xiang * for everyone else; 41a9a94d93SGao Xiang * 42a9a94d93SGao Xiang * L: Field should be protected by the pcluster lock; 43a9a94d93SGao Xiang * 44a9a94d93SGao Xiang * A: Field should be accessed / updated in atomic for parallelized code. 45a9a94d93SGao Xiang */ 46a9a94d93SGao Xiang struct z_erofs_pcluster { 47a9a94d93SGao Xiang struct erofs_workgroup obj; 48a9a94d93SGao Xiang struct mutex lock; 49a9a94d93SGao Xiang 50a9a94d93SGao Xiang /* A: point to next chained pcluster or TAILs */ 51a9a94d93SGao Xiang z_erofs_next_pcluster_t next; 52a9a94d93SGao Xiang 53a9a94d93SGao Xiang /* L: the maximum decompression size of this round */ 54a9a94d93SGao Xiang unsigned int length; 55a9a94d93SGao Xiang 56a9a94d93SGao Xiang /* L: total number of bvecs */ 57a9a94d93SGao Xiang unsigned int vcnt; 58a9a94d93SGao Xiang 59a9a94d93SGao Xiang /* I: page offset of start position of decompression */ 60a9a94d93SGao Xiang unsigned short pageofs_out; 61a9a94d93SGao Xiang 62a9a94d93SGao Xiang /* I: page offset of inline compressed data */ 63a9a94d93SGao Xiang unsigned short pageofs_in; 64a9a94d93SGao Xiang 65a9a94d93SGao Xiang union { 66a9a94d93SGao Xiang /* L: inline a certain number of bvec for bootstrap */ 67a9a94d93SGao Xiang struct z_erofs_bvset_inline bvset; 68a9a94d93SGao Xiang 69a9a94d93SGao Xiang /* I: can be used to free the pcluster by RCU. */ 70a9a94d93SGao Xiang struct rcu_head rcu; 71a9a94d93SGao Xiang }; 72a9a94d93SGao Xiang 73a9a94d93SGao Xiang union { 74a9a94d93SGao Xiang /* I: physical cluster size in pages */ 75a9a94d93SGao Xiang unsigned short pclusterpages; 76a9a94d93SGao Xiang 77a9a94d93SGao Xiang /* I: tailpacking inline compressed size */ 78a9a94d93SGao Xiang unsigned short tailpacking_size; 79a9a94d93SGao Xiang }; 80a9a94d93SGao Xiang 81a9a94d93SGao Xiang /* I: compression algorithm format */ 82a9a94d93SGao Xiang unsigned char algorithmformat; 83a9a94d93SGao Xiang 84a9a94d93SGao Xiang /* L: whether partial decompression or not */ 85a9a94d93SGao Xiang bool partial; 86a9a94d93SGao Xiang 87a9a94d93SGao Xiang /* L: indicate several pageofs_outs or not */ 88a9a94d93SGao Xiang bool multibases; 89a9a94d93SGao Xiang 90a9a94d93SGao Xiang /* A: compressed bvecs (can be cached or inplaced pages) */ 91a9a94d93SGao Xiang struct z_erofs_bvec compressed_bvecs[]; 92a9a94d93SGao Xiang }; 93a9a94d93SGao Xiang 94967c28b2SGao Xiang /* the end of a chain of pclusters */ 9543d86ec9SGao Xiang #define Z_EROFS_PCLUSTER_TAIL ((void *) 0x700 + POISON_POINTER_DELTA) 96a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_NIL (NULL) 97a9a94d93SGao Xiang 98a9a94d93SGao Xiang struct z_erofs_decompressqueue { 99a9a94d93SGao Xiang struct super_block *sb; 100a9a94d93SGao Xiang atomic_t pending_bios; 101a9a94d93SGao Xiang z_erofs_next_pcluster_t head; 102a9a94d93SGao Xiang 103a9a94d93SGao Xiang union { 104a9a94d93SGao Xiang struct completion done; 105a9a94d93SGao Xiang struct work_struct work; 1063fffb589SSandeep Dhavale struct kthread_work kthread_work; 107a9a94d93SGao Xiang } u; 108a9a94d93SGao Xiang bool eio, sync; 109a9a94d93SGao Xiang }; 110a9a94d93SGao Xiang 111a9a94d93SGao Xiang static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 112a9a94d93SGao Xiang { 113a9a94d93SGao Xiang return !pcl->obj.index; 114a9a94d93SGao Xiang } 115a9a94d93SGao Xiang 116a9a94d93SGao Xiang static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 117a9a94d93SGao Xiang { 118a9a94d93SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) 119a9a94d93SGao Xiang return 1; 120a9a94d93SGao Xiang return pcl->pclusterpages; 121a9a94d93SGao Xiang } 122a9a94d93SGao Xiang 123a9a94d93SGao Xiang /* 124a9a94d93SGao Xiang * bit 30: I/O error occurred on this page 125a9a94d93SGao Xiang * bit 0 - 29: remaining parts to complete this page 126a9a94d93SGao Xiang */ 127a9a94d93SGao Xiang #define Z_EROFS_PAGE_EIO (1 << 30) 128a9a94d93SGao Xiang 129a9a94d93SGao Xiang static inline void z_erofs_onlinepage_init(struct page *page) 130a9a94d93SGao Xiang { 131a9a94d93SGao Xiang union { 132a9a94d93SGao Xiang atomic_t o; 133a9a94d93SGao Xiang unsigned long v; 134a9a94d93SGao Xiang } u = { .o = ATOMIC_INIT(1) }; 135a9a94d93SGao Xiang 136a9a94d93SGao Xiang set_page_private(page, u.v); 137a9a94d93SGao Xiang smp_wmb(); 138a9a94d93SGao Xiang SetPagePrivate(page); 139a9a94d93SGao Xiang } 140a9a94d93SGao Xiang 141a9a94d93SGao Xiang static inline void z_erofs_onlinepage_split(struct page *page) 142a9a94d93SGao Xiang { 143a9a94d93SGao Xiang atomic_inc((atomic_t *)&page->private); 144a9a94d93SGao Xiang } 145a9a94d93SGao Xiang 146*9a05c6a8SGao Xiang static void z_erofs_onlinepage_endio(struct page *page, int err) 147a9a94d93SGao Xiang { 148*9a05c6a8SGao Xiang int orig, v; 149*9a05c6a8SGao Xiang 150*9a05c6a8SGao Xiang DBG_BUGON(!PagePrivate(page)); 151a9a94d93SGao Xiang 152a9a94d93SGao Xiang do { 153a9a94d93SGao Xiang orig = atomic_read((atomic_t *)&page->private); 154*9a05c6a8SGao Xiang v = (orig - 1) | (err ? Z_EROFS_PAGE_EIO : 0); 155*9a05c6a8SGao Xiang } while (atomic_cmpxchg((atomic_t *)&page->private, orig, v) != orig); 156a9a94d93SGao Xiang 157a9a94d93SGao Xiang if (!(v & ~Z_EROFS_PAGE_EIO)) { 158a9a94d93SGao Xiang set_page_private(page, 0); 159a9a94d93SGao Xiang ClearPagePrivate(page); 160a9a94d93SGao Xiang if (!(v & Z_EROFS_PAGE_EIO)) 161a9a94d93SGao Xiang SetPageUptodate(page); 162a9a94d93SGao Xiang unlock_page(page); 163a9a94d93SGao Xiang } 164a9a94d93SGao Xiang } 165a9a94d93SGao Xiang 166a9a94d93SGao Xiang #define Z_EROFS_ONSTACK_PAGES 32 167a9a94d93SGao Xiang 16847e4937aSGao Xiang /* 1699f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 1709f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 1719f6cc76eSGao Xiang */ 1729f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 1739f6cc76eSGao Xiang struct kmem_cache *slab; 1749f6cc76eSGao Xiang unsigned int maxpages; 1759f6cc76eSGao Xiang char name[48]; 1769f6cc76eSGao Xiang }; 1779f6cc76eSGao Xiang 1789f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 1799f6cc76eSGao Xiang 1809f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 1819f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 1829f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 1839f6cc76eSGao Xiang }; 1849f6cc76eSGao Xiang 18506a304cdSGao Xiang struct z_erofs_bvec_iter { 18606a304cdSGao Xiang struct page *bvpage; 18706a304cdSGao Xiang struct z_erofs_bvset *bvset; 18806a304cdSGao Xiang unsigned int nr, cur; 18906a304cdSGao Xiang }; 19006a304cdSGao Xiang 19106a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 19206a304cdSGao Xiang { 19306a304cdSGao Xiang if (iter->bvpage) 19406a304cdSGao Xiang kunmap_local(iter->bvset); 19506a304cdSGao Xiang return iter->bvpage; 19606a304cdSGao Xiang } 19706a304cdSGao Xiang 19806a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 19906a304cdSGao Xiang { 20006a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 20106a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 20206a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 20306a304cdSGao Xiang struct page *oldpage; 20406a304cdSGao Xiang 20506a304cdSGao Xiang DBG_BUGON(!nextpage); 20606a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 20706a304cdSGao Xiang iter->bvpage = nextpage; 20806a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 20906a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 21006a304cdSGao Xiang iter->cur = 0; 21106a304cdSGao Xiang return oldpage; 21206a304cdSGao Xiang } 21306a304cdSGao Xiang 21406a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 21506a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 21606a304cdSGao Xiang unsigned int bootstrap_nr, 21706a304cdSGao Xiang unsigned int cur) 21806a304cdSGao Xiang { 21906a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 22006a304cdSGao Xiang .nr = bootstrap_nr, 22106a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 22206a304cdSGao Xiang }; 22306a304cdSGao Xiang 22406a304cdSGao Xiang while (cur > iter->nr) { 22506a304cdSGao Xiang cur -= iter->nr; 22606a304cdSGao Xiang z_erofs_bvset_flip(iter); 22706a304cdSGao Xiang } 22806a304cdSGao Xiang iter->cur = cur; 22906a304cdSGao Xiang } 23006a304cdSGao Xiang 23106a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 23206a304cdSGao Xiang struct z_erofs_bvec *bvec, 2336ab5eed6SGao Xiang struct page **candidate_bvpage, 2346ab5eed6SGao Xiang struct page **pagepool) 23506a304cdSGao Xiang { 23605b63d2bSGao Xiang if (iter->cur >= iter->nr) { 23705b63d2bSGao Xiang struct page *nextpage = *candidate_bvpage; 23806a304cdSGao Xiang 23905b63d2bSGao Xiang if (!nextpage) { 2406ab5eed6SGao Xiang nextpage = erofs_allocpage(pagepool, GFP_NOFS); 24105b63d2bSGao Xiang if (!nextpage) 24205b63d2bSGao Xiang return -ENOMEM; 24305b63d2bSGao Xiang set_page_private(nextpage, Z_EROFS_SHORTLIVED_PAGE); 24405b63d2bSGao Xiang } 24506a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 24605b63d2bSGao Xiang iter->bvset->nextpage = nextpage; 24706a304cdSGao Xiang z_erofs_bvset_flip(iter); 24806a304cdSGao Xiang 24906a304cdSGao Xiang iter->bvset->nextpage = NULL; 25006a304cdSGao Xiang *candidate_bvpage = NULL; 25106a304cdSGao Xiang } 25206a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 25306a304cdSGao Xiang return 0; 25406a304cdSGao Xiang } 25506a304cdSGao Xiang 25606a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 25706a304cdSGao Xiang struct z_erofs_bvec *bvec, 25806a304cdSGao Xiang struct page **old_bvpage) 25906a304cdSGao Xiang { 26006a304cdSGao Xiang if (iter->cur == iter->nr) 26106a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 26206a304cdSGao Xiang else 26306a304cdSGao Xiang *old_bvpage = NULL; 26406a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 26506a304cdSGao Xiang } 26606a304cdSGao Xiang 2679f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 2689f6cc76eSGao Xiang { 2699f6cc76eSGao Xiang int i; 2709f6cc76eSGao Xiang 2719f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 2729f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 2739f6cc76eSGao Xiang continue; 2749f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 2759f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 2769f6cc76eSGao Xiang } 2779f6cc76eSGao Xiang } 2789f6cc76eSGao Xiang 2799f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 2809f6cc76eSGao Xiang { 2819f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 2829f6cc76eSGao Xiang struct z_erofs_pcluster *a; 2839f6cc76eSGao Xiang unsigned int size; 2849f6cc76eSGao Xiang 2859f6cc76eSGao Xiang for (pcs = pcluster_pool; 2869f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 287ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 2889f6cc76eSGao Xiang 2899f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 2909f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 2919f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 2929f6cc76eSGao Xiang if (pcs->slab) 2939f6cc76eSGao Xiang continue; 2949f6cc76eSGao Xiang 2959f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 2969f6cc76eSGao Xiang return -ENOMEM; 2979f6cc76eSGao Xiang } 2989f6cc76eSGao Xiang return 0; 2999f6cc76eSGao Xiang } 3009f6cc76eSGao Xiang 3019f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 3029f6cc76eSGao Xiang { 3039f6cc76eSGao Xiang int i; 3049f6cc76eSGao Xiang 3059f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3069f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3079f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 3089f6cc76eSGao Xiang 3099f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 3109f6cc76eSGao Xiang continue; 3119f6cc76eSGao Xiang 3129f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 3139f6cc76eSGao Xiang if (!pcl) 3149f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 3159f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 3169f6cc76eSGao Xiang return pcl; 3179f6cc76eSGao Xiang } 3189f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 3199f6cc76eSGao Xiang } 3209f6cc76eSGao Xiang 3219f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 3229f6cc76eSGao Xiang { 323cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 3249f6cc76eSGao Xiang int i; 3259f6cc76eSGao Xiang 3269f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3279f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3289f6cc76eSGao Xiang 329cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 3309f6cc76eSGao Xiang continue; 3319f6cc76eSGao Xiang 3329f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 3339f6cc76eSGao Xiang return; 3349f6cc76eSGao Xiang } 3359f6cc76eSGao Xiang DBG_BUGON(1); 3369f6cc76eSGao Xiang } 3379f6cc76eSGao Xiang 33847e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 33947e4937aSGao Xiang 3403fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 3413fffb589SSandeep Dhavale static struct kthread_worker __rcu **z_erofs_pcpu_workers; 3423fffb589SSandeep Dhavale 3433fffb589SSandeep Dhavale static void erofs_destroy_percpu_workers(void) 34447e4937aSGao Xiang { 3453fffb589SSandeep Dhavale struct kthread_worker *worker; 3463fffb589SSandeep Dhavale unsigned int cpu; 3473fffb589SSandeep Dhavale 3483fffb589SSandeep Dhavale for_each_possible_cpu(cpu) { 3493fffb589SSandeep Dhavale worker = rcu_dereference_protected( 3503fffb589SSandeep Dhavale z_erofs_pcpu_workers[cpu], 1); 3513fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 3523fffb589SSandeep Dhavale if (worker) 3533fffb589SSandeep Dhavale kthread_destroy_worker(worker); 3543fffb589SSandeep Dhavale } 3553fffb589SSandeep Dhavale kfree(z_erofs_pcpu_workers); 35647e4937aSGao Xiang } 35747e4937aSGao Xiang 3583fffb589SSandeep Dhavale static struct kthread_worker *erofs_init_percpu_worker(int cpu) 35947e4937aSGao Xiang { 3603fffb589SSandeep Dhavale struct kthread_worker *worker = 3613fffb589SSandeep Dhavale kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 36247e4937aSGao Xiang 3633fffb589SSandeep Dhavale if (IS_ERR(worker)) 3643fffb589SSandeep Dhavale return worker; 3653fffb589SSandeep Dhavale if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 3663fffb589SSandeep Dhavale sched_set_fifo_low(worker->task); 3673fffb589SSandeep Dhavale return worker; 3683fffb589SSandeep Dhavale } 3693fffb589SSandeep Dhavale 3703fffb589SSandeep Dhavale static int erofs_init_percpu_workers(void) 3713fffb589SSandeep Dhavale { 3723fffb589SSandeep Dhavale struct kthread_worker *worker; 3733fffb589SSandeep Dhavale unsigned int cpu; 3743fffb589SSandeep Dhavale 3753fffb589SSandeep Dhavale z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 3763fffb589SSandeep Dhavale sizeof(struct kthread_worker *), GFP_ATOMIC); 3773fffb589SSandeep Dhavale if (!z_erofs_pcpu_workers) 3783fffb589SSandeep Dhavale return -ENOMEM; 3793fffb589SSandeep Dhavale 3803fffb589SSandeep Dhavale for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 3813fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 3823fffb589SSandeep Dhavale if (!IS_ERR(worker)) 3833fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 3843fffb589SSandeep Dhavale } 3853fffb589SSandeep Dhavale return 0; 3863fffb589SSandeep Dhavale } 3873fffb589SSandeep Dhavale #else 3883fffb589SSandeep Dhavale static inline void erofs_destroy_percpu_workers(void) {} 3893fffb589SSandeep Dhavale static inline int erofs_init_percpu_workers(void) { return 0; } 3903fffb589SSandeep Dhavale #endif 3913fffb589SSandeep Dhavale 3923fffb589SSandeep Dhavale #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 3933fffb589SSandeep Dhavale static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 3943fffb589SSandeep Dhavale static enum cpuhp_state erofs_cpuhp_state; 3953fffb589SSandeep Dhavale 3963fffb589SSandeep Dhavale static int erofs_cpu_online(unsigned int cpu) 3973fffb589SSandeep Dhavale { 3983fffb589SSandeep Dhavale struct kthread_worker *worker, *old; 3993fffb589SSandeep Dhavale 4003fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 4013fffb589SSandeep Dhavale if (IS_ERR(worker)) 4023fffb589SSandeep Dhavale return PTR_ERR(worker); 4033fffb589SSandeep Dhavale 4043fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4053fffb589SSandeep Dhavale old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4063fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4073fffb589SSandeep Dhavale if (!old) 4083fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 4093fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4103fffb589SSandeep Dhavale if (old) 4113fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4123fffb589SSandeep Dhavale return 0; 4133fffb589SSandeep Dhavale } 4143fffb589SSandeep Dhavale 4153fffb589SSandeep Dhavale static int erofs_cpu_offline(unsigned int cpu) 4163fffb589SSandeep Dhavale { 4173fffb589SSandeep Dhavale struct kthread_worker *worker; 4183fffb589SSandeep Dhavale 4193fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4203fffb589SSandeep Dhavale worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4213fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4223fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 4233fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4243fffb589SSandeep Dhavale 4253fffb589SSandeep Dhavale synchronize_rcu(); 4263fffb589SSandeep Dhavale if (worker) 4273fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4283fffb589SSandeep Dhavale return 0; 4293fffb589SSandeep Dhavale } 4303fffb589SSandeep Dhavale 4313fffb589SSandeep Dhavale static int erofs_cpu_hotplug_init(void) 4323fffb589SSandeep Dhavale { 4333fffb589SSandeep Dhavale int state; 4343fffb589SSandeep Dhavale 4353fffb589SSandeep Dhavale state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 4363fffb589SSandeep Dhavale "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 4373fffb589SSandeep Dhavale if (state < 0) 4383fffb589SSandeep Dhavale return state; 4393fffb589SSandeep Dhavale 4403fffb589SSandeep Dhavale erofs_cpuhp_state = state; 4413fffb589SSandeep Dhavale return 0; 4423fffb589SSandeep Dhavale } 4433fffb589SSandeep Dhavale 4443fffb589SSandeep Dhavale static void erofs_cpu_hotplug_destroy(void) 4453fffb589SSandeep Dhavale { 4463fffb589SSandeep Dhavale if (erofs_cpuhp_state) 4473fffb589SSandeep Dhavale cpuhp_remove_state_nocalls(erofs_cpuhp_state); 4483fffb589SSandeep Dhavale } 4493fffb589SSandeep Dhavale #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 4503fffb589SSandeep Dhavale static inline int erofs_cpu_hotplug_init(void) { return 0; } 4513fffb589SSandeep Dhavale static inline void erofs_cpu_hotplug_destroy(void) {} 4523fffb589SSandeep Dhavale #endif 4533fffb589SSandeep Dhavale 4543fffb589SSandeep Dhavale void z_erofs_exit_zip_subsystem(void) 4553fffb589SSandeep Dhavale { 4563fffb589SSandeep Dhavale erofs_cpu_hotplug_destroy(); 4573fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4583fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4593fffb589SSandeep Dhavale z_erofs_destroy_pcluster_pool(); 46047e4937aSGao Xiang } 46147e4937aSGao Xiang 46247e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 46347e4937aSGao Xiang { 4649f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 46547e4937aSGao Xiang 4669f6cc76eSGao Xiang if (err) 4673fffb589SSandeep Dhavale goto out_error_pcluster_pool; 4683fffb589SSandeep Dhavale 4693fffb589SSandeep Dhavale z_erofs_workqueue = alloc_workqueue("erofs_worker", 4703fffb589SSandeep Dhavale WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 4718d1b80a7SDan Carpenter if (!z_erofs_workqueue) { 4728d1b80a7SDan Carpenter err = -ENOMEM; 4733fffb589SSandeep Dhavale goto out_error_workqueue_init; 4748d1b80a7SDan Carpenter } 4753fffb589SSandeep Dhavale 4763fffb589SSandeep Dhavale err = erofs_init_percpu_workers(); 4779f6cc76eSGao Xiang if (err) 4783fffb589SSandeep Dhavale goto out_error_pcpu_worker; 4793fffb589SSandeep Dhavale 4803fffb589SSandeep Dhavale err = erofs_cpu_hotplug_init(); 4813fffb589SSandeep Dhavale if (err < 0) 4823fffb589SSandeep Dhavale goto out_error_cpuhp_init; 4833fffb589SSandeep Dhavale return err; 4843fffb589SSandeep Dhavale 4853fffb589SSandeep Dhavale out_error_cpuhp_init: 4863fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4873fffb589SSandeep Dhavale out_error_pcpu_worker: 4883fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4893fffb589SSandeep Dhavale out_error_workqueue_init: 4909f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 4913fffb589SSandeep Dhavale out_error_pcluster_pool: 4929f6cc76eSGao Xiang return err; 49347e4937aSGao Xiang } 49447e4937aSGao Xiang 495db166fc2SGao Xiang enum z_erofs_pclustermode { 496db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 49747e4937aSGao Xiang /* 498db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 4990b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 5000b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 501387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 5020b964600SGao Xiang */ 503db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 50447e4937aSGao Xiang /* 505dcba1b23SGao Xiang * The pcluster was just linked to a decompression chain by us. It can 506dcba1b23SGao Xiang * also be linked with the remaining pclusters, which means if the 507dcba1b23SGao Xiang * processing page is the tail page of a pcluster, this pcluster can 508dcba1b23SGao Xiang * safely use the whole page (since the previous pcluster is within the 509dcba1b23SGao Xiang * same chain) for in-place I/O, as illustrated below: 510dcba1b23SGao Xiang * ___________________________________________________ 51147e4937aSGao Xiang * | tail (partial) page | head (partial) page | 512dcba1b23SGao Xiang * | (of the current pcl) | (of the previous pcl) | 513dcba1b23SGao Xiang * |___PCLUSTER_FOLLOWED___|_____PCLUSTER_FOLLOWED_____| 51447e4937aSGao Xiang * 515dcba1b23SGao Xiang * [ (*) the page above can be used as inplace I/O. ] 51647e4937aSGao Xiang */ 517db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 51847e4937aSGao Xiang }; 51947e4937aSGao Xiang 5205c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 5215c6dcc57SGao Xiang struct inode *const inode; 5225c6dcc57SGao Xiang struct erofs_map_blocks map; 52306a304cdSGao Xiang struct z_erofs_bvec_iter biter; 52447e4937aSGao Xiang 5256ab5eed6SGao Xiang struct page *pagepool; 52606a304cdSGao Xiang struct page *candidate_bvpage; 527967c28b2SGao Xiang struct z_erofs_pcluster *pcl; 52847e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 529db166fc2SGao Xiang enum z_erofs_pclustermode mode; 53047e4937aSGao Xiang 53147e4937aSGao Xiang /* used for applying cache strategy on the fly */ 53247e4937aSGao Xiang bool backmost; 53347e4937aSGao Xiang erofs_off_t headoffset; 534ed722fbcSGao Xiang 535ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 536ed722fbcSGao Xiang unsigned int icur; 53747e4937aSGao Xiang }; 53847e4937aSGao Xiang 53947e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 5405c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 541db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 54247e4937aSGao Xiang 5431282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 5441282dea3SGao Xiang { 5451282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 5461282dea3SGao Xiang 5471282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 5481282dea3SGao Xiang return false; 5491282dea3SGao Xiang 5501282dea3SGao Xiang if (fe->backmost) 5511282dea3SGao Xiang return true; 5521282dea3SGao Xiang 5531282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 5541282dea3SGao Xiang fe->map.m_la < fe->headoffset) 5551282dea3SGao Xiang return true; 5561282dea3SGao Xiang 5571282dea3SGao Xiang return false; 5581282dea3SGao Xiang } 5591282dea3SGao Xiang 5606ab5eed6SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe) 56147e4937aSGao Xiang { 5626f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 5635c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 5641282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 56547e4937aSGao Xiang bool standalone = true; 5666f39d1e1SGao Xiang /* 5676f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 5686f39d1e1SGao Xiang * can be used if low memory otherwise. 5696f39d1e1SGao Xiang */ 5701825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 5711825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 572ed722fbcSGao Xiang unsigned int i; 57347e4937aSGao Xiang 574db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 57547e4937aSGao Xiang return; 57647e4937aSGao Xiang 577ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 57847e4937aSGao Xiang struct page *page; 579b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 5801825c8d7SGao Xiang struct page *newpage = NULL; 58147e4937aSGao Xiang 58247e4937aSGao Xiang /* the compressed page was loaded before */ 583ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 58447e4937aSGao Xiang continue; 58547e4937aSGao Xiang 586ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 58747e4937aSGao Xiang 58847e4937aSGao Xiang if (page) { 589b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 5900b964600SGao Xiang } else { 5910b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 5920b964600SGao Xiang standalone = false; 5931282dea3SGao Xiang if (!shouldalloc) 5941282dea3SGao Xiang continue; 5951282dea3SGao Xiang 5961282dea3SGao Xiang /* 5971282dea3SGao Xiang * try to use cached I/O if page allocation 5981282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 5991282dea3SGao Xiang * to avoid any direct reclaim. 6001282dea3SGao Xiang */ 6016ab5eed6SGao Xiang newpage = erofs_allocpage(&fe->pagepool, gfp); 6021825c8d7SGao Xiang if (!newpage) 60347e4937aSGao Xiang continue; 6041282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 605b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 60647e4937aSGao Xiang } 60747e4937aSGao Xiang 608b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 60947e4937aSGao Xiang continue; 61047e4937aSGao Xiang 611eaa9172aSGao Xiang if (page) 61247e4937aSGao Xiang put_page(page); 613eaa9172aSGao Xiang else if (newpage) 6146ab5eed6SGao Xiang erofs_pagepool_add(&fe->pagepool, newpage); 61547e4937aSGao Xiang } 61647e4937aSGao Xiang 6170b964600SGao Xiang /* 6180b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 6190b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 6200b964600SGao Xiang */ 6210b964600SGao Xiang if (standalone) 622db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 62347e4937aSGao Xiang } 62447e4937aSGao Xiang 62547e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 62647e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 62747e4937aSGao Xiang struct erofs_workgroup *grp) 62847e4937aSGao Xiang { 62947e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 63047e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 63147e4937aSGao Xiang int i; 63247e4937aSGao Xiang 633cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 63447e4937aSGao Xiang /* 6357674a42fSGao Xiang * refcount of workgroup is now freezed as 0, 63647e4937aSGao Xiang * therefore no need to worry about available decompression users. 63747e4937aSGao Xiang */ 6389f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 639ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 64047e4937aSGao Xiang 64147e4937aSGao Xiang if (!page) 64247e4937aSGao Xiang continue; 64347e4937aSGao Xiang 64447e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 64547e4937aSGao Xiang if (!trylock_page(page)) 64647e4937aSGao Xiang return -EBUSY; 64747e4937aSGao Xiang 648f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 64947e4937aSGao Xiang continue; 65047e4937aSGao Xiang 65147e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 652ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6536aaa7b06SGao Xiang detach_page_private(page); 65447e4937aSGao Xiang unlock_page(page); 65547e4937aSGao Xiang } 65647e4937aSGao Xiang return 0; 65747e4937aSGao Xiang } 65847e4937aSGao Xiang 6597b4e372cSGao Xiang static bool z_erofs_cache_release_folio(struct folio *folio, gfp_t gfp) 66047e4937aSGao Xiang { 6617b4e372cSGao Xiang struct z_erofs_pcluster *pcl = folio_get_private(folio); 6627b4e372cSGao Xiang bool ret; 6637b4e372cSGao Xiang int i; 6647b4e372cSGao Xiang 6657b4e372cSGao Xiang if (!folio_test_private(folio)) 6667b4e372cSGao Xiang return true; 66747e4937aSGao Xiang 6687b4e372cSGao Xiang ret = false; 6697674a42fSGao Xiang spin_lock(&pcl->obj.lockref.lock); 6707674a42fSGao Xiang if (pcl->obj.lockref.count > 0) 6717674a42fSGao Xiang goto out; 6727674a42fSGao Xiang 673cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 6749f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 6757b4e372cSGao Xiang if (pcl->compressed_bvecs[i].page == &folio->page) { 676ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6777b4e372cSGao Xiang ret = true; 67847e4937aSGao Xiang break; 67947e4937aSGao Xiang } 68047e4937aSGao Xiang } 6816aaa7b06SGao Xiang if (ret) 6827b4e372cSGao Xiang folio_detach_private(folio); 6837674a42fSGao Xiang out: 6847674a42fSGao Xiang spin_unlock(&pcl->obj.lockref.lock); 68547e4937aSGao Xiang return ret; 68647e4937aSGao Xiang } 68747e4937aSGao Xiang 6887b4e372cSGao Xiang /* 6897b4e372cSGao Xiang * It will be called only on inode eviction. In case that there are still some 6907b4e372cSGao Xiang * decompression requests in progress, wait with rescheduling for a bit here. 6917b4e372cSGao Xiang * An extra lock could be introduced instead but it seems unnecessary. 6927b4e372cSGao Xiang */ 6937b4e372cSGao Xiang static void z_erofs_cache_invalidate_folio(struct folio *folio, 6947b4e372cSGao Xiang size_t offset, size_t length) 6957b4e372cSGao Xiang { 6967b4e372cSGao Xiang const size_t stop = length + offset; 6977b4e372cSGao Xiang 6987b4e372cSGao Xiang /* Check for potential overflow in debug mode */ 6997b4e372cSGao Xiang DBG_BUGON(stop > folio_size(folio) || stop < length); 7007b4e372cSGao Xiang 7017b4e372cSGao Xiang if (offset == 0 && stop == folio_size(folio)) 7027b4e372cSGao Xiang while (!z_erofs_cache_release_folio(folio, GFP_NOFS)) 7037b4e372cSGao Xiang cond_resched(); 7047b4e372cSGao Xiang } 7057b4e372cSGao Xiang 7067b4e372cSGao Xiang static const struct address_space_operations z_erofs_cache_aops = { 7077b4e372cSGao Xiang .release_folio = z_erofs_cache_release_folio, 7087b4e372cSGao Xiang .invalidate_folio = z_erofs_cache_invalidate_folio, 7097b4e372cSGao Xiang }; 7107b4e372cSGao Xiang 7117b4e372cSGao Xiang int erofs_init_managed_cache(struct super_block *sb) 7127b4e372cSGao Xiang { 7137b4e372cSGao Xiang struct inode *const inode = new_inode(sb); 7147b4e372cSGao Xiang 7157b4e372cSGao Xiang if (!inode) 7167b4e372cSGao Xiang return -ENOMEM; 7177b4e372cSGao Xiang 7187b4e372cSGao Xiang set_nlink(inode, 1); 7197b4e372cSGao Xiang inode->i_size = OFFSET_MAX; 7207b4e372cSGao Xiang inode->i_mapping->a_ops = &z_erofs_cache_aops; 7217b4e372cSGao Xiang mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS); 7227b4e372cSGao Xiang EROFS_SB(sb)->managed_cache = inode; 7237b4e372cSGao Xiang return 0; 7247b4e372cSGao Xiang } 7257b4e372cSGao Xiang 7265c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 727ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 72847e4937aSGao Xiang { 7295c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 73047e4937aSGao Xiang 731ed722fbcSGao Xiang while (fe->icur > 0) { 732ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 733ed722fbcSGao Xiang NULL, bvec->page)) { 734ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 73547e4937aSGao Xiang return true; 736ed722fbcSGao Xiang } 737ed722fbcSGao Xiang } 73847e4937aSGao Xiang return false; 73947e4937aSGao Xiang } 74047e4937aSGao Xiang 74187ca34a7SGao Xiang /* callers must be with pcluster lock held */ 7425c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 7435b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 74447e4937aSGao Xiang { 74547e4937aSGao Xiang int ret; 74647e4937aSGao Xiang 747db166fc2SGao Xiang if (exclusive) { 74806a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 749ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 75047e4937aSGao Xiang return 0; 75106a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 752db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 75306a304cdSGao Xiang !fe->candidate_bvpage) 75406a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 75506a304cdSGao Xiang } 7566ab5eed6SGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage, 7576ab5eed6SGao Xiang &fe->pagepool); 75806a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 75906a304cdSGao Xiang return ret; 76047e4937aSGao Xiang } 76147e4937aSGao Xiang 7625c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 76347e4937aSGao Xiang { 7645c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 7655c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 76647e4937aSGao Xiang 767473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 768473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 769473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 77047e4937aSGao Xiang *owned_head = &pcl->next; 771473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 772db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 773473e15b0SGao Xiang return; 774473e15b0SGao Xiang } 775473e15b0SGao Xiang 776967c28b2SGao Xiang /* type 2, it belongs to an ongoing chain */ 777db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 77847e4937aSGao Xiang } 77947e4937aSGao Xiang 78083a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 78147e4937aSGao Xiang { 78283a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 783cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 78447e4937aSGao Xiang struct z_erofs_pcluster *pcl; 78564094a04SGao Xiang struct erofs_workgroup *grp; 78647e4937aSGao Xiang int err; 78747e4937aSGao Xiang 788c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 789c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 7908f899262SGao Xiang DBG_BUGON(1); 7918f899262SGao Xiang return -EFSCORRUPTED; 7928f899262SGao Xiang } 7938f899262SGao Xiang 7949f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 795cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 796cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 7979f6cc76eSGao Xiang if (IS_ERR(pcl)) 7989f6cc76eSGao Xiang return PTR_ERR(pcl); 79947e4937aSGao Xiang 8007674a42fSGao Xiang spin_lock_init(&pcl->obj.lockref.lock); 8018f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 8022bfab9c0SGao Xiang pcl->length = 0; 8032bfab9c0SGao Xiang pcl->partial = true; 80447e4937aSGao Xiang 80547e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 8065c6dcc57SGao Xiang pcl->next = fe->owned_head; 80787ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 808db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 80947e4937aSGao Xiang 81047e4937aSGao Xiang /* 81147e4937aSGao Xiang * lock all primary followed works before visible to others 81247e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 81347e4937aSGao Xiang */ 81487ca34a7SGao Xiang mutex_init(&pcl->lock); 81587ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 81647e4937aSGao Xiang 817cecf864dSYue Hu if (ztailpacking) { 818cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 8193acea5fcSJingbo Xu pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 820cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 821cecf864dSYue Hu } else { 822cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 823cecf864dSYue Hu 82483a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 82564094a04SGao Xiang if (IS_ERR(grp)) { 82664094a04SGao Xiang err = PTR_ERR(grp); 82764094a04SGao Xiang goto err_out; 82864094a04SGao Xiang } 82964094a04SGao Xiang 83064094a04SGao Xiang if (grp != &pcl->obj) { 8315c6dcc57SGao Xiang fe->pcl = container_of(grp, 832cecf864dSYue Hu struct z_erofs_pcluster, obj); 83364094a04SGao Xiang err = -EEXIST; 83464094a04SGao Xiang goto err_out; 83547e4937aSGao Xiang } 836cecf864dSYue Hu } 8375c6dcc57SGao Xiang fe->owned_head = &pcl->next; 8385c6dcc57SGao Xiang fe->pcl = pcl; 8399e579fc1SGao Xiang return 0; 84064094a04SGao Xiang 84164094a04SGao Xiang err_out: 84287ca34a7SGao Xiang mutex_unlock(&pcl->lock); 8439f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 84464094a04SGao Xiang return err; 84547e4937aSGao Xiang } 84647e4937aSGao Xiang 847dcba1b23SGao Xiang static int z_erofs_pcluster_begin(struct z_erofs_decompress_frontend *fe) 84847e4937aSGao Xiang { 84983a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 850aeebae9dSGao Xiang struct super_block *sb = fe->inode->i_sb; 851aeebae9dSGao Xiang erofs_blk_t blknr = erofs_blknr(sb, map->m_pa); 8520d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 8539e579fc1SGao Xiang int ret; 85447e4937aSGao Xiang 85587ca34a7SGao Xiang DBG_BUGON(fe->pcl); 85647e4937aSGao Xiang 85787ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 8585c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 85947e4937aSGao Xiang 8600d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 861aeebae9dSGao Xiang grp = erofs_find_workgroup(sb, blknr); 8620d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 86347e4937aSGao Xiang DBG_BUGON(1); 864cecf864dSYue Hu return -EFSCORRUPTED; 865cecf864dSYue Hu } 86647e4937aSGao Xiang 86764094a04SGao Xiang if (grp) { 8685c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 8690d823b42SGao Xiang ret = -EEXIST; 87064094a04SGao Xiang } else { 87183a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 87264094a04SGao Xiang } 87347e4937aSGao Xiang 8740d823b42SGao Xiang if (ret == -EEXIST) { 875267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 876267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 8770d823b42SGao Xiang } else if (ret) { 8780d823b42SGao Xiang return ret; 8790d823b42SGao Xiang } 880aeebae9dSGao Xiang 88106a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 882387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 883aeebae9dSGao Xiang if (!z_erofs_is_inline_pcluster(fe->pcl)) { 884aeebae9dSGao Xiang /* bind cache first when cached decompression is preferred */ 885aeebae9dSGao Xiang z_erofs_bind_cache(fe); 886aeebae9dSGao Xiang } else { 887aeebae9dSGao Xiang void *mptr; 888aeebae9dSGao Xiang 889aeebae9dSGao Xiang mptr = erofs_read_metabuf(&map->buf, sb, blknr, EROFS_NO_KMAP); 890aeebae9dSGao Xiang if (IS_ERR(mptr)) { 891aeebae9dSGao Xiang ret = PTR_ERR(mptr); 892aeebae9dSGao Xiang erofs_err(sb, "failed to get inline data %d", ret); 893aeebae9dSGao Xiang return ret; 894aeebae9dSGao Xiang } 895aeebae9dSGao Xiang get_page(map->buf.page); 896aeebae9dSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, map->buf.page); 897aeebae9dSGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 898aeebae9dSGao Xiang } 899aeebae9dSGao Xiang /* file-backed inplace I/O pages are traversed in reverse order */ 900ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 90147e4937aSGao Xiang return 0; 90247e4937aSGao Xiang } 90347e4937aSGao Xiang 90447e4937aSGao Xiang /* 90547e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 90647e4937aSGao Xiang * only after a RCU grace period. 90747e4937aSGao Xiang */ 90847e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 90947e4937aSGao Xiang { 91087ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 91187ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 91247e4937aSGao Xiang } 91347e4937aSGao Xiang 91447e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 91547e4937aSGao Xiang { 91647e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 91747e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 91847e4937aSGao Xiang 91987ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 92047e4937aSGao Xiang } 92147e4937aSGao Xiang 922dcba1b23SGao Xiang static void z_erofs_pcluster_end(struct z_erofs_decompress_frontend *fe) 92347e4937aSGao Xiang { 92487ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 92547e4937aSGao Xiang 92687ca34a7SGao Xiang if (!pcl) 927dcba1b23SGao Xiang return; 92847e4937aSGao Xiang 92906a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 93087ca34a7SGao Xiang mutex_unlock(&pcl->lock); 93147e4937aSGao Xiang 93205b63d2bSGao Xiang if (fe->candidate_bvpage) 93306a304cdSGao Xiang fe->candidate_bvpage = NULL; 93406a304cdSGao Xiang 93547e4937aSGao Xiang /* 93647e4937aSGao Xiang * if all pending pages are added, don't hold its reference 93747e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 93847e4937aSGao Xiang */ 939db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 94087ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 94147e4937aSGao Xiang 94287ca34a7SGao Xiang fe->pcl = NULL; 943dcba1b23SGao Xiang fe->backmost = false; 94447e4937aSGao Xiang } 94547e4937aSGao Xiang 9468b00be16SGao Xiang static int z_erofs_read_fragment(struct super_block *sb, struct page *page, 9478b00be16SGao Xiang unsigned int cur, unsigned int end, erofs_off_t pos) 948b15b2e30SYue Hu { 9498b00be16SGao Xiang struct inode *packed_inode = EROFS_SB(sb)->packed_inode; 950b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 9518b00be16SGao Xiang unsigned int cnt; 9528b00be16SGao Xiang u8 *src; 953b15b2e30SYue Hu 954e5126de1SYue Hu if (!packed_inode) 955e5126de1SYue Hu return -EFSCORRUPTED; 956e5126de1SYue Hu 957eb2c5e41SGao Xiang buf.inode = packed_inode; 9588b00be16SGao Xiang for (; cur < end; cur += cnt, pos += cnt) { 9598b00be16SGao Xiang cnt = min_t(unsigned int, end - cur, 9603acea5fcSJingbo Xu sb->s_blocksize - erofs_blkoff(sb, pos)); 961eb2c5e41SGao Xiang src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 962b15b2e30SYue Hu if (IS_ERR(src)) { 963b15b2e30SYue Hu erofs_put_metabuf(&buf); 964b15b2e30SYue Hu return PTR_ERR(src); 965b15b2e30SYue Hu } 9668b00be16SGao Xiang memcpy_to_page(page, cur, src + erofs_blkoff(sb, pos), cnt); 967b15b2e30SYue Hu } 968b15b2e30SYue Hu erofs_put_metabuf(&buf); 969b15b2e30SYue Hu return 0; 970b15b2e30SYue Hu } 971b15b2e30SYue Hu 97247e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 9736ab5eed6SGao Xiang struct page *page) 97447e4937aSGao Xiang { 97547e4937aSGao Xiang struct inode *const inode = fe->inode; 97647e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 97747e4937aSGao Xiang const loff_t offset = page_offset(page); 9785b220b20SGao Xiang bool tight = true, exclusive; 979e4c1cf52SGao Xiang unsigned int cur, end, len, split; 98047e4937aSGao Xiang int err = 0; 98147e4937aSGao Xiang 98247e4937aSGao Xiang z_erofs_onlinepage_init(page); 98347e4937aSGao Xiang 984e4c1cf52SGao Xiang split = 0; 98547e4937aSGao Xiang end = PAGE_SIZE; 98647e4937aSGao Xiang repeat: 987e4c1cf52SGao Xiang if (offset + end - 1 < map->m_la || 988e4c1cf52SGao Xiang offset + end - 1 >= map->m_la + map->m_llen) { 989dcba1b23SGao Xiang z_erofs_pcluster_end(fe); 990e4c1cf52SGao Xiang map->m_la = offset + end - 1; 99147e4937aSGao Xiang map->m_llen = 0; 99247e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 9938d8a09b0SGao Xiang if (err) 99467148551SGao Xiang goto out; 99539397a46SGao Xiang } 99647e4937aSGao Xiang 997e4c1cf52SGao Xiang cur = offset > map->m_la ? 0 : map->m_la - offset; 998e4c1cf52SGao Xiang /* bump split parts first to avoid several separate cases */ 999e4c1cf52SGao Xiang ++split; 1000dc76ea8cSGao Xiang 10018d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 100247e4937aSGao Xiang zero_user_segment(page, cur, end); 1003e4c1cf52SGao Xiang tight = false; 100447e4937aSGao Xiang goto next_part; 100547e4937aSGao Xiang } 1006e4c1cf52SGao Xiang 1007b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 10088b00be16SGao Xiang erofs_off_t fpos = offset + cur - map->m_la; 1009b15b2e30SYue Hu 10108b00be16SGao Xiang len = min_t(unsigned int, map->m_llen - fpos, end - cur); 10118b00be16SGao Xiang err = z_erofs_read_fragment(inode->i_sb, page, cur, cur + len, 10128b00be16SGao Xiang EROFS_I(inode)->z_fragmentoff + fpos); 1013b15b2e30SYue Hu if (err) 1014b15b2e30SYue Hu goto out; 1015b15b2e30SYue Hu tight = false; 1016b15b2e30SYue Hu goto next_part; 1017b15b2e30SYue Hu } 101847e4937aSGao Xiang 1019e4c1cf52SGao Xiang if (!fe->pcl) { 1020e4c1cf52SGao Xiang err = z_erofs_pcluster_begin(fe); 1021e4c1cf52SGao Xiang if (err) 1022e4c1cf52SGao Xiang goto out; 1023e4c1cf52SGao Xiang } 1024e4c1cf52SGao Xiang 1025e4c1cf52SGao Xiang /* 1026e4c1cf52SGao Xiang * Ensure the current partial page belongs to this submit chain rather 1027e4c1cf52SGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 1028e4c1cf52SGao Xiang * those chains are handled asynchronously thus the page cannot be used 1029e4c1cf52SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 1030e4c1cf52SGao Xiang */ 1031e4c1cf52SGao Xiang tight &= (fe->mode > Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1032e4c1cf52SGao Xiang exclusive = (!cur && ((split <= 1) || tight)); 103347e4937aSGao Xiang if (cur) 1034db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 103547e4937aSGao Xiang 103606a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 103706a304cdSGao Xiang .page = page, 103806a304cdSGao Xiang .offset = offset - map->m_la, 103906a304cdSGao Xiang .end = end, 10405b220b20SGao Xiang }), exclusive); 104105b63d2bSGao Xiang if (err) 104267148551SGao Xiang goto out; 104347e4937aSGao Xiang 104467148551SGao Xiang z_erofs_onlinepage_split(page); 1045267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1046267f2492SGao Xiang fe->pcl->multibases = true; 10472bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 10482bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 10492bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 10502bfab9c0SGao Xiang } 1051e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1052e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1053e7933278SGao Xiang fe->pcl->length == map->m_llen) 1054e7933278SGao Xiang fe->pcl->partial = false; 105547e4937aSGao Xiang next_part: 10562bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 105747e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 10582bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 105947e4937aSGao Xiang 106047e4937aSGao Xiang end = cur; 106147e4937aSGao Xiang if (end > 0) 106247e4937aSGao Xiang goto repeat; 106347e4937aSGao Xiang 106447e4937aSGao Xiang out: 1065*9a05c6a8SGao Xiang z_erofs_onlinepage_endio(page, err); 106647e4937aSGao Xiang return err; 106747e4937aSGao Xiang } 106847e4937aSGao Xiang 1069ef4b4b46SYue Hu static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 107040452ffcSHuang Jianan unsigned int readahead_pages) 107140452ffcSHuang Jianan { 1072a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 107340452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 107440452ffcSHuang Jianan !readahead_pages) 107540452ffcSHuang Jianan return true; 107640452ffcSHuang Jianan 107740452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 107840452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 107940452ffcSHuang Jianan return true; 108040452ffcSHuang Jianan 108140452ffcSHuang Jianan return false; 108240452ffcSHuang Jianan } 108340452ffcSHuang Jianan 10846aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 10856aaa7b06SGao Xiang { 10866aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 10876aaa7b06SGao Xiang } 10886aaa7b06SGao Xiang 10894f05687fSGao Xiang struct z_erofs_decompress_backend { 10904f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 10914f05687fSGao Xiang struct super_block *sb; 10924f05687fSGao Xiang struct z_erofs_pcluster *pcl; 10934f05687fSGao Xiang 10944f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 10954f05687fSGao Xiang struct page **decompressed_pages; 10964f05687fSGao Xiang /* pages to keep the compressed data */ 10974f05687fSGao Xiang struct page **compressed_pages; 10984f05687fSGao Xiang 1099267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 11004f05687fSGao Xiang struct page **pagepool; 11012bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 11024f05687fSGao Xiang }; 11034f05687fSGao Xiang 1104267f2492SGao Xiang struct z_erofs_bvec_item { 1105267f2492SGao Xiang struct z_erofs_bvec bvec; 1106267f2492SGao Xiang struct list_head list; 1107267f2492SGao Xiang }; 1108267f2492SGao Xiang 1109267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 11103fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 11113fe96ee0SGao Xiang { 1112267f2492SGao Xiang struct z_erofs_bvec_item *item; 1113267f2492SGao Xiang unsigned int pgnr; 11143fe96ee0SGao Xiang 111594c43de7SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK) && 111694c43de7SGao Xiang (bvec->end == PAGE_SIZE || 111794c43de7SGao Xiang bvec->offset + bvec->end == be->pcl->length)) { 1118267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 11192bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 112063bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 11213fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1122267f2492SGao Xiang return; 11233fe96ee0SGao Xiang } 112463bbb856SGao Xiang } 11253fe96ee0SGao Xiang 1126267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1127267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1128267f2492SGao Xiang item->bvec = *bvec; 1129267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1130267f2492SGao Xiang } 1131267f2492SGao Xiang 1132267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1133267f2492SGao Xiang int err) 1134267f2492SGao Xiang { 1135267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1136267f2492SGao Xiang struct list_head *p, *n; 1137267f2492SGao Xiang 1138267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1139267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1140267f2492SGao Xiang unsigned int end, cur; 1141267f2492SGao Xiang void *dst, *src; 1142267f2492SGao Xiang 1143267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1144267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1145267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1146267f2492SGao Xiang bvi->bvec.end); 1147267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1148267f2492SGao Xiang while (cur < end) { 1149267f2492SGao Xiang unsigned int pgnr, scur, len; 1150267f2492SGao Xiang 1151267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1152267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1153267f2492SGao Xiang 1154267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1155267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1156267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1157267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1158267f2492SGao Xiang err = -EFSCORRUPTED; 1159267f2492SGao Xiang cur += len; 1160267f2492SGao Xiang continue; 1161267f2492SGao Xiang } 1162267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1163267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1164267f2492SGao Xiang kunmap_local(src); 1165267f2492SGao Xiang cur += len; 1166267f2492SGao Xiang } 1167267f2492SGao Xiang kunmap_local(dst); 1168*9a05c6a8SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page, err); 1169267f2492SGao Xiang list_del(p); 1170267f2492SGao Xiang kfree(bvi); 1171267f2492SGao Xiang } 1172267f2492SGao Xiang } 1173267f2492SGao Xiang 1174267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 117542fec235SGao Xiang { 11764f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 117706a304cdSGao Xiang struct z_erofs_bvec_iter biter; 117806a304cdSGao Xiang struct page *old_bvpage; 1179267f2492SGao Xiang int i; 118042fec235SGao Xiang 1181387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 118242fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 118306a304cdSGao Xiang struct z_erofs_bvec bvec; 118442fec235SGao Xiang 118506a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 118642fec235SGao Xiang 118706a304cdSGao Xiang if (old_bvpage) 11884f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 118942fec235SGao Xiang 119006a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1191267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 119242fec235SGao Xiang } 119306a304cdSGao Xiang 119406a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 119506a304cdSGao Xiang if (old_bvpage) 11964f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 119742fec235SGao Xiang } 119842fec235SGao Xiang 11994f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 12004f05687fSGao Xiang bool *overlapped) 120167139e36SGao Xiang { 12024f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 120367139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 120467139e36SGao Xiang int i, err = 0; 120567139e36SGao Xiang 120667139e36SGao Xiang *overlapped = false; 120767139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1208ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1209ed722fbcSGao Xiang struct page *page = bvec->page; 121067139e36SGao Xiang 121167139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 121267139e36SGao Xiang if (!page) { 121367139e36SGao Xiang DBG_BUGON(1); 121467139e36SGao Xiang continue; 121567139e36SGao Xiang } 1216fe3e5914SGao Xiang be->compressed_pages[i] = page; 121767139e36SGao Xiang 121867139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 121967139e36SGao Xiang if (!PageUptodate(page)) 122067139e36SGao Xiang err = -EIO; 122167139e36SGao Xiang continue; 122267139e36SGao Xiang } 122367139e36SGao Xiang 122467139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 122567139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 12264f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 122767139e36SGao Xiang if (!PageUptodate(page)) 122867139e36SGao Xiang err = -EIO; 122967139e36SGao Xiang continue; 123067139e36SGao Xiang } 1231267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 123267139e36SGao Xiang *overlapped = true; 123367139e36SGao Xiang } 123467139e36SGao Xiang } 123567139e36SGao Xiang 1236fe3e5914SGao Xiang if (err) 12374f05687fSGao Xiang return err; 12384f05687fSGao Xiang return 0; 123967139e36SGao Xiang } 124067139e36SGao Xiang 12414f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 12424f05687fSGao Xiang int err) 124347e4937aSGao Xiang { 12444f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 12454f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1246cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1247597e2953SYue Hu const struct z_erofs_decompressor *decompressor = 1248597e2953SYue Hu &erofs_decompressors[pcl->algorithmformat]; 12492bfab9c0SGao Xiang unsigned int i, inputsize; 125067148551SGao Xiang int err2; 12512bfab9c0SGao Xiang struct page *page; 12522bfab9c0SGao Xiang bool overlapped; 125347e4937aSGao Xiang 125487ca34a7SGao Xiang mutex_lock(&pcl->lock); 12552bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 125647e4937aSGao Xiang 1257fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1258fe3e5914SGao Xiang be->decompressed_pages = NULL; 1259fe3e5914SGao Xiang be->compressed_pages = NULL; 1260fe3e5914SGao Xiang be->onstack_used = 0; 12612bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 12624f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 12632bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 12644f05687fSGao Xiang memset(be->decompressed_pages, 0, 12652bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1266fe3e5914SGao Xiang } 1267fe3e5914SGao Xiang 1268fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1269fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1270fe3e5914SGao Xiang 1271fe3e5914SGao Xiang if (!be->decompressed_pages) 12724f05687fSGao Xiang be->decompressed_pages = 1273647dd2c3SGao Xiang kvcalloc(be->nr_pages, sizeof(struct page *), 1274e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1275fe3e5914SGao Xiang if (!be->compressed_pages) 1276fe3e5914SGao Xiang be->compressed_pages = 1277647dd2c3SGao Xiang kvcalloc(pclusterpages, sizeof(struct page *), 1278fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 127947e4937aSGao Xiang 1280267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 12814f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 12824f05687fSGao Xiang if (err2) 12834f05687fSGao Xiang err = err2; 12848d8a09b0SGao Xiang if (err) 128547e4937aSGao Xiang goto out; 128647e4937aSGao Xiang 1287cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1288cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1289cecf864dSYue Hu else 1290cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1291cecf864dSYue Hu 1292597e2953SYue Hu err = decompressor->decompress(&(struct z_erofs_decompress_req) { 12934f05687fSGao Xiang .sb = be->sb, 12944f05687fSGao Xiang .in = be->compressed_pages, 12954f05687fSGao Xiang .out = be->decompressed_pages, 1296cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 129787ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 12989f6cc76eSGao Xiang .inputsize = inputsize, 12992bfab9c0SGao Xiang .outputsize = pcl->length, 130047e4937aSGao Xiang .alg = pcl->algorithmformat, 130147e4937aSGao Xiang .inplace_io = overlapped, 13022bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1303267f2492SGao Xiang .fillgaps = pcl->multibases, 13044f05687fSGao Xiang }, be->pagepool); 130547e4937aSGao Xiang 130647e4937aSGao Xiang out: 1307cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1308cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1309ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1310ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1311cecf864dSYue Hu put_page(page); 1312cecf864dSYue Hu } else { 1313cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1314ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 131547e4937aSGao Xiang 131647e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 131747e4937aSGao Xiang continue; 131847e4937aSGao Xiang 13196aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13204f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1321ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 132247e4937aSGao Xiang } 1323cecf864dSYue Hu } 1324fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1325fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1326647dd2c3SGao Xiang kvfree(be->compressed_pages); 1327267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 132847e4937aSGao Xiang 13292bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 13304f05687fSGao Xiang page = be->decompressed_pages[i]; 133147e4937aSGao Xiang if (!page) 133247e4937aSGao Xiang continue; 133347e4937aSGao Xiang 13346aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 133547e4937aSGao Xiang 13366aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13374f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 133847e4937aSGao Xiang continue; 1339*9a05c6a8SGao Xiang z_erofs_onlinepage_endio(page, err); 134047e4937aSGao Xiang } 134147e4937aSGao Xiang 13424f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 1343647dd2c3SGao Xiang kvfree(be->decompressed_pages); 134447e4937aSGao Xiang 13452bfab9c0SGao Xiang pcl->length = 0; 13462bfab9c0SGao Xiang pcl->partial = true; 1347267f2492SGao Xiang pcl->multibases = false; 134806a304cdSGao Xiang pcl->bvset.nextpage = NULL; 134987ca34a7SGao Xiang pcl->vcnt = 0; 135047e4937aSGao Xiang 135187ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 135247e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 135387ca34a7SGao Xiang mutex_unlock(&pcl->lock); 135447e4937aSGao Xiang return err; 135547e4937aSGao Xiang } 135647e4937aSGao Xiang 13570c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1358eaa9172aSGao Xiang struct page **pagepool) 135947e4937aSGao Xiang { 13604f05687fSGao Xiang struct z_erofs_decompress_backend be = { 13614f05687fSGao Xiang .sb = io->sb, 13624f05687fSGao Xiang .pagepool = pagepool, 1363267f2492SGao Xiang .decompressed_secondary_bvecs = 1364267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 13654f05687fSGao Xiang }; 136647e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 136747e4937aSGao Xiang 1368967c28b2SGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL) { 136947e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 137047e4937aSGao Xiang 13714f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 13724f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 137347e4937aSGao Xiang 13744f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 13754f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 137647e4937aSGao Xiang } 137747e4937aSGao Xiang } 137847e4937aSGao Xiang 13790c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 138047e4937aSGao Xiang { 1381a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1382a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1383eaa9172aSGao Xiang struct page *pagepool = NULL; 138447e4937aSGao Xiang 1385967c28b2SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL); 13860c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 1387eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1388a4b1fab1SGao Xiang kvfree(bgq); 138947e4937aSGao Xiang } 139047e4937aSGao Xiang 13913fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 13923fffb589SSandeep Dhavale static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 13933fffb589SSandeep Dhavale { 13943fffb589SSandeep Dhavale z_erofs_decompressqueue_work((struct work_struct *)work); 13953fffb589SSandeep Dhavale } 13963fffb589SSandeep Dhavale #endif 13973fffb589SSandeep Dhavale 13987865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1399cdba5506SGao Xiang int bios) 14007865827cSGao Xiang { 14017865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 14027865827cSGao Xiang 14037865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1404cdba5506SGao Xiang if (io->sync) { 14057865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 140660b30050SHongyu Jin complete(&io->u.done); 14077865827cSGao Xiang return; 14087865827cSGao Xiang } 14097865827cSGao Xiang 14107865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 14117865827cSGao Xiang return; 14123fffb589SSandeep Dhavale /* Use (kthread_)work and sync decompression for atomic contexts only */ 141312d0a24aSSandeep Dhavale if (!in_task() || irqs_disabled() || rcu_read_lock_any_held()) { 14143fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14153fffb589SSandeep Dhavale struct kthread_worker *worker; 14163fffb589SSandeep Dhavale 14173fffb589SSandeep Dhavale rcu_read_lock(); 14183fffb589SSandeep Dhavale worker = rcu_dereference( 14193fffb589SSandeep Dhavale z_erofs_pcpu_workers[raw_smp_processor_id()]); 14203fffb589SSandeep Dhavale if (!worker) { 14213fffb589SSandeep Dhavale INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 14227865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 14233fffb589SSandeep Dhavale } else { 14243fffb589SSandeep Dhavale kthread_queue_work(worker, &io->u.kthread_work); 14253fffb589SSandeep Dhavale } 14263fffb589SSandeep Dhavale rcu_read_unlock(); 14273fffb589SSandeep Dhavale #else 14283fffb589SSandeep Dhavale queue_work(z_erofs_workqueue, &io->u.work); 14293fffb589SSandeep Dhavale #endif 14307865827cSGao Xiang /* enable sync decompression for readahead */ 14317865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 14327865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 14337865827cSGao Xiang return; 14347865827cSGao Xiang } 14357865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 14367865827cSGao Xiang } 14377865827cSGao Xiang 143847e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 143947e4937aSGao Xiang unsigned int nr, 1440eaa9172aSGao Xiang struct page **pagepool, 14419f2731d6SGao Xiang struct address_space *mc) 144247e4937aSGao Xiang { 144347e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 14449f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 144547e4937aSGao Xiang bool tocache = false; 144647e4937aSGao Xiang 144747e4937aSGao Xiang struct address_space *mapping; 144847e4937aSGao Xiang struct page *oldpage, *page; 144947e4937aSGao Xiang int justfound; 145047e4937aSGao Xiang 145147e4937aSGao Xiang repeat: 1452ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 145347e4937aSGao Xiang oldpage = page; 145447e4937aSGao Xiang 145547e4937aSGao Xiang if (!page) 145647e4937aSGao Xiang goto out_allocpage; 145747e4937aSGao Xiang 1458b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1459b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 146047e4937aSGao Xiang 14611825c8d7SGao Xiang /* 14621825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 14631825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 14641825c8d7SGao Xiang */ 14651825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1466ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 14671825c8d7SGao Xiang set_page_private(page, 0); 14681825c8d7SGao Xiang tocache = true; 14691825c8d7SGao Xiang goto out_tocache; 14701825c8d7SGao Xiang } 147147e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 147247e4937aSGao Xiang 147347e4937aSGao Xiang /* 14746aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 147547e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 147647e4937aSGao Xiang */ 147747e4937aSGao Xiang if (mapping && mapping != mc) 147847e4937aSGao Xiang /* ought to be unmanaged pages */ 147947e4937aSGao Xiang goto out; 148047e4937aSGao Xiang 14816aaa7b06SGao Xiang /* directly return for shortlived page as well */ 14826aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 14836aaa7b06SGao Xiang goto out; 14846aaa7b06SGao Xiang 148547e4937aSGao Xiang lock_page(page); 148647e4937aSGao Xiang 148747e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 148847e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 148947e4937aSGao Xiang 149047e4937aSGao Xiang /* the page is still in manage cache */ 149147e4937aSGao Xiang if (page->mapping == mc) { 1492ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 149347e4937aSGao Xiang 149447e4937aSGao Xiang if (!PagePrivate(page)) { 149547e4937aSGao Xiang /* 149647e4937aSGao Xiang * impossible to be !PagePrivate(page) for 149747e4937aSGao Xiang * the current restriction as well if 1498ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 149947e4937aSGao Xiang */ 150047e4937aSGao Xiang DBG_BUGON(!justfound); 150147e4937aSGao Xiang 150247e4937aSGao Xiang justfound = 0; 150347e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 150447e4937aSGao Xiang SetPagePrivate(page); 150547e4937aSGao Xiang } 150647e4937aSGao Xiang 150747e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 150847e4937aSGao Xiang if (PageUptodate(page)) { 150947e4937aSGao Xiang unlock_page(page); 151047e4937aSGao Xiang page = NULL; 151147e4937aSGao Xiang } 151247e4937aSGao Xiang goto out; 151347e4937aSGao Xiang } 151447e4937aSGao Xiang 151547e4937aSGao Xiang /* 151647e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 151747e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 151847e4937aSGao Xiang */ 151947e4937aSGao Xiang DBG_BUGON(page->mapping); 152047e4937aSGao Xiang DBG_BUGON(!justfound); 152147e4937aSGao Xiang 152247e4937aSGao Xiang tocache = true; 152347e4937aSGao Xiang unlock_page(page); 152447e4937aSGao Xiang put_page(page); 152547e4937aSGao Xiang out_allocpage: 15265ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1527ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1528ed722fbcSGao Xiang oldpage, page)) { 1529eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 15305ddcee1fSGao Xiang cond_resched(); 15315ddcee1fSGao Xiang goto repeat; 15325ddcee1fSGao Xiang } 15331825c8d7SGao Xiang out_tocache: 1534bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1535bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1536bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1537bf225074SGao Xiang goto out; 1538a30573b3SGao Xiang } 1539bf225074SGao Xiang attach_page_private(page, pcl); 1540bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1541bf225074SGao Xiang put_page(page); 1542bf225074SGao Xiang 154347e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 154447e4937aSGao Xiang return page; 154547e4937aSGao Xiang } 154647e4937aSGao Xiang 1547cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1548a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 154947e4937aSGao Xiang { 1550a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 155147e4937aSGao Xiang 1552a4b1fab1SGao Xiang if (fg && !*fg) { 1553a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1554a4b1fab1SGao Xiang if (!q) { 1555a4b1fab1SGao Xiang *fg = true; 1556a4b1fab1SGao Xiang goto fg_out; 155747e4937aSGao Xiang } 15583fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 15593fffb589SSandeep Dhavale kthread_init_work(&q->u.kthread_work, 15603fffb589SSandeep Dhavale z_erofs_decompressqueue_kthread_work); 15613fffb589SSandeep Dhavale #else 15620c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 15633fffb589SSandeep Dhavale #endif 1564a4b1fab1SGao Xiang } else { 1565a4b1fab1SGao Xiang fg_out: 1566a4b1fab1SGao Xiang q = fgq; 156760b30050SHongyu Jin init_completion(&fgq->u.done); 1568a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 156967148551SGao Xiang q->eio = false; 1570cdba5506SGao Xiang q->sync = true; 1571a4b1fab1SGao Xiang } 1572a4b1fab1SGao Xiang q->sb = sb; 1573967c28b2SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL; 1574a4b1fab1SGao Xiang return q; 157547e4937aSGao Xiang } 157647e4937aSGao Xiang 157747e4937aSGao Xiang /* define decompression jobqueue types */ 157847e4937aSGao Xiang enum { 157947e4937aSGao Xiang JQ_BYPASS, 158047e4937aSGao Xiang JQ_SUBMIT, 158147e4937aSGao Xiang NR_JOBQUEUES, 158247e4937aSGao Xiang }; 158347e4937aSGao Xiang 158447e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 158547e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 158647e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 158747e4937aSGao Xiang { 158847e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 158947e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 159047e4937aSGao Xiang 1591967c28b2SGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL); 159247e4937aSGao Xiang 159347e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 159447e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 159547e4937aSGao Xiang 159647e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 159747e4937aSGao Xiang } 159847e4937aSGao Xiang 15997865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 16007865827cSGao Xiang { 1601cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 16027865827cSGao Xiang blk_status_t err = bio->bi_status; 16037865827cSGao Xiang struct bio_vec *bvec; 16047865827cSGao Xiang struct bvec_iter_all iter_all; 16057865827cSGao Xiang 16067865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 16077865827cSGao Xiang struct page *page = bvec->bv_page; 16087865827cSGao Xiang 16097865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 16107865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 16117865827cSGao Xiang 16127865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 16137865827cSGao Xiang if (!err) 16147865827cSGao Xiang SetPageUptodate(page); 16157865827cSGao Xiang unlock_page(page); 16167865827cSGao Xiang } 16177865827cSGao Xiang } 161867148551SGao Xiang if (err) 161967148551SGao Xiang q->eio = true; 1620cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 16217865827cSGao Xiang bio_put(bio); 16227865827cSGao Xiang } 16237865827cSGao Xiang 162483a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1625a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1626ef4b4b46SYue Hu bool *force_fg, bool readahead) 162747e4937aSGao Xiang { 162883a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 162983a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 163047e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1631a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 16325c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1633dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 16343f649ab7SKees Cook pgoff_t last_index; 1635dfeab2e9SGao Xiang struct block_device *last_bdev; 16361e4a2955SGao Xiang unsigned int nr_bios = 0; 16371e4a2955SGao Xiang struct bio *bio = NULL; 163882e60d00SJohannes Weiner unsigned long pflags; 163982e60d00SJohannes Weiner int memstall = 0; 164047e4937aSGao Xiang 1641cdba5506SGao Xiang /* 1642cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1643cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1644cdba5506SGao Xiang */ 1645cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1646cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1647cdba5506SGao Xiang 1648a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1649a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 165047e4937aSGao Xiang 165147e4937aSGao Xiang /* by default, all need io submission */ 165247e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 165347e4937aSGao Xiang 165447e4937aSGao Xiang do { 1655dfeab2e9SGao Xiang struct erofs_map_dev mdev; 165647e4937aSGao Xiang struct z_erofs_pcluster *pcl; 16571e4a2955SGao Xiang pgoff_t cur, end; 16581e4a2955SGao Xiang unsigned int i = 0; 16591e4a2955SGao Xiang bool bypass = true; 166047e4937aSGao Xiang 166147e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 166247e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 1663967c28b2SGao Xiang owned_head = READ_ONCE(pcl->next); 166447e4937aSGao Xiang 1665cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1666cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1667cecf864dSYue Hu continue; 1668cecf864dSYue Hu } 1669cecf864dSYue Hu 1670dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1671dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 16723acea5fcSJingbo Xu .m_pa = erofs_pos(sb, pcl->obj.index), 1673dfeab2e9SGao Xiang }; 1674dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1675dfeab2e9SGao Xiang 16763acea5fcSJingbo Xu cur = erofs_blknr(sb, mdev.m_pa); 16779f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 167847e4937aSGao Xiang 16791e4a2955SGao Xiang do { 16801e4a2955SGao Xiang struct page *page; 168147e4937aSGao Xiang 16826ab5eed6SGao Xiang page = pickup_page_for_submission(pcl, i++, 16836ab5eed6SGao Xiang &f->pagepool, mc); 16841e4a2955SGao Xiang if (!page) 16851e4a2955SGao Xiang continue; 168647e4937aSGao Xiang 1687dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1688dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 168947e4937aSGao Xiang submit_bio_retry: 169094e4e153SGao Xiang submit_bio(bio); 169182e60d00SJohannes Weiner if (memstall) { 169282e60d00SJohannes Weiner psi_memstall_leave(&pflags); 169382e60d00SJohannes Weiner memstall = 0; 169482e60d00SJohannes Weiner } 169547e4937aSGao Xiang bio = NULL; 169647e4937aSGao Xiang } 169747e4937aSGao Xiang 169882e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 169999486c51SChristoph Hellwig psi_memstall_enter(&pflags); 170082e60d00SJohannes Weiner memstall = 1; 170182e60d00SJohannes Weiner } 170299486c51SChristoph Hellwig 170347e4937aSGao Xiang if (!bio) { 170407888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 170507888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 17060c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1707dfeab2e9SGao Xiang 1708dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 17091e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 17103acea5fcSJingbo Xu (sb->s_blocksize_bits - 9); 1711cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 1712ef4b4b46SYue Hu if (readahead) 17136ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 171447e4937aSGao Xiang ++nr_bios; 171547e4937aSGao Xiang } 171647e4937aSGao Xiang 17176c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 171847e4937aSGao Xiang goto submit_bio_retry; 171947e4937aSGao Xiang 17201e4a2955SGao Xiang last_index = cur; 17211e4a2955SGao Xiang bypass = false; 17221e4a2955SGao Xiang } while (++cur < end); 172347e4937aSGao Xiang 17241e4a2955SGao Xiang if (!bypass) 172547e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 172647e4937aSGao Xiang else 172747e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 172847e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 172947e4937aSGao Xiang 173099486c51SChristoph Hellwig if (bio) { 173194e4e153SGao Xiang submit_bio(bio); 173282e60d00SJohannes Weiner if (memstall) 173382e60d00SJohannes Weiner psi_memstall_leave(&pflags); 173499486c51SChristoph Hellwig } 173547e4937aSGao Xiang 1736587a67b7SGao Xiang /* 1737587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 17383fffb589SSandeep Dhavale * don't issue decompression but drop it directly instead. 1739587a67b7SGao Xiang */ 1740587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1741587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 17421e4a2955SGao Xiang return; 1743587a67b7SGao Xiang } 1744cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 174547e4937aSGao Xiang } 174647e4937aSGao Xiang 174783a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 17486ab5eed6SGao Xiang bool force_fg, bool ra) 174947e4937aSGao Xiang { 1750a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 175147e4937aSGao Xiang 17525c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 175347e4937aSGao Xiang return; 17546ab5eed6SGao Xiang z_erofs_submit_queue(f, io, &force_fg, ra); 175547e4937aSGao Xiang 17560c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 17576ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], &f->pagepool); 175847e4937aSGao Xiang 175947e4937aSGao Xiang if (!force_fg) 176047e4937aSGao Xiang return; 176147e4937aSGao Xiang 176247e4937aSGao Xiang /* wait until all bios are completed */ 176360b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 176447e4937aSGao Xiang 17650c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 17666ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], &f->pagepool); 176747e4937aSGao Xiang } 176847e4937aSGao Xiang 176938629291SGao Xiang /* 177038629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 177138629291SGao Xiang * approximate readmore strategies as a start. 177238629291SGao Xiang */ 177338629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 17746ab5eed6SGao Xiang struct readahead_control *rac, bool backmost) 177538629291SGao Xiang { 177638629291SGao Xiang struct inode *inode = f->inode; 177738629291SGao Xiang struct erofs_map_blocks *map = &f->map; 1778796e9149SYue Hu erofs_off_t cur, end, headoffset = f->headoffset; 177938629291SGao Xiang int err; 178038629291SGao Xiang 178138629291SGao Xiang if (backmost) { 1782796e9149SYue Hu if (rac) 1783796e9149SYue Hu end = headoffset + readahead_length(rac) - 1; 1784796e9149SYue Hu else 1785796e9149SYue Hu end = headoffset + PAGE_SIZE - 1; 178638629291SGao Xiang map->m_la = end; 1787622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1788622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 178938629291SGao Xiang if (err) 179038629291SGao Xiang return; 179138629291SGao Xiang 1792796e9149SYue Hu /* expand ra for the trailing edge if readahead */ 179338629291SGao Xiang if (rac) { 179438629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1795796e9149SYue Hu readahead_expand(rac, headoffset, cur - headoffset); 179638629291SGao Xiang return; 179738629291SGao Xiang } 179838629291SGao Xiang end = round_up(end, PAGE_SIZE); 179938629291SGao Xiang } else { 180038629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 180138629291SGao Xiang 180238629291SGao Xiang if (!map->m_llen) 180338629291SGao Xiang return; 180438629291SGao Xiang } 180538629291SGao Xiang 180638629291SGao Xiang cur = map->m_la + map->m_llen - 1; 1807936aa701SChunhai Guo while ((cur >= end) && (cur < i_size_read(inode))) { 180838629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 180938629291SGao Xiang struct page *page; 181038629291SGao Xiang 181138629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1812aa793b46SGao Xiang if (page) { 1813e3157bb5SFerry Meng if (PageUptodate(page)) 181438629291SGao Xiang unlock_page(page); 1815e3157bb5SFerry Meng else 1816e3157bb5SFerry Meng (void)z_erofs_do_read_page(f, page); 181738629291SGao Xiang put_page(page); 1818aa793b46SGao Xiang } 1819aa793b46SGao Xiang 182038629291SGao Xiang if (cur < PAGE_SIZE) 182138629291SGao Xiang break; 182238629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 182338629291SGao Xiang } 182438629291SGao Xiang } 182538629291SGao Xiang 1826a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 182747e4937aSGao Xiang { 1828a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 182947e4937aSGao Xiang struct inode *const inode = page->mapping->host; 183040452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 183147e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 183247e4937aSGao Xiang int err; 183347e4937aSGao Xiang 183447e4937aSGao Xiang trace_erofs_readpage(page, false); 183547e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 183647e4937aSGao Xiang 18376ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, true); 18386ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 18396ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, false); 1840dcba1b23SGao Xiang z_erofs_pcluster_end(&f); 184147e4937aSGao Xiang 184247e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 18436ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, 0), false); 184447e4937aSGao Xiang 1845e3157bb5SFerry Meng if (err && err != -EINTR) 1846e3157bb5SFerry Meng erofs_err(inode->i_sb, "read error %d @ %lu of nid %llu", 1847e3157bb5SFerry Meng err, folio->index, EROFS_I(inode)->nid); 184847e4937aSGao Xiang 184909c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 18506ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 185147e4937aSGao Xiang return err; 185247e4937aSGao Xiang } 185347e4937aSGao Xiang 18540615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 185547e4937aSGao Xiang { 18560615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 185747e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 185847e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 18596ab5eed6SGao Xiang struct page *head = NULL, *page; 186038629291SGao Xiang unsigned int nr_pages; 186147e4937aSGao Xiang 18620615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 186347e4937aSGao Xiang 18646ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, true); 186538629291SGao Xiang nr_pages = readahead_count(rac); 186638629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 186738629291SGao Xiang 18680615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 186947e4937aSGao Xiang set_page_private(page, (unsigned long)head); 187047e4937aSGao Xiang head = page; 187147e4937aSGao Xiang } 187247e4937aSGao Xiang 187347e4937aSGao Xiang while (head) { 187447e4937aSGao Xiang struct page *page = head; 187547e4937aSGao Xiang int err; 187647e4937aSGao Xiang 187747e4937aSGao Xiang /* traversal in reverse order */ 187847e4937aSGao Xiang head = (void *)page_private(page); 187947e4937aSGao Xiang 18806ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 1881e3157bb5SFerry Meng if (err && err != -EINTR) 1882e3157bb5SFerry Meng erofs_err(inode->i_sb, "readahead error %d @ %lu of nid %llu", 1883e3157bb5SFerry Meng err, page->index, EROFS_I(inode)->nid); 188447e4937aSGao Xiang put_page(page); 188547e4937aSGao Xiang } 18866ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, false); 1887dcba1b23SGao Xiang z_erofs_pcluster_end(&f); 188847e4937aSGao Xiang 18896ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, nr_pages), true); 189009c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 18916ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 189247e4937aSGao Xiang } 189347e4937aSGao Xiang 18940c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1895a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 18960615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 189747e4937aSGao Xiang }; 1898