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 */ 95*43d86ec9SGao 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 146a9a94d93SGao Xiang static inline void z_erofs_page_mark_eio(struct page *page) 147a9a94d93SGao Xiang { 148a9a94d93SGao Xiang int orig; 149a9a94d93SGao Xiang 150a9a94d93SGao Xiang do { 151a9a94d93SGao Xiang orig = atomic_read((atomic_t *)&page->private); 152a9a94d93SGao Xiang } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 153a9a94d93SGao Xiang orig | Z_EROFS_PAGE_EIO) != orig); 154a9a94d93SGao Xiang } 155a9a94d93SGao Xiang 156a9a94d93SGao Xiang static inline void z_erofs_onlinepage_endio(struct page *page) 157a9a94d93SGao Xiang { 158a9a94d93SGao Xiang unsigned int v; 159a9a94d93SGao Xiang 160a9a94d93SGao Xiang DBG_BUGON(!PagePrivate(page)); 161a9a94d93SGao Xiang v = atomic_dec_return((atomic_t *)&page->private); 162a9a94d93SGao Xiang if (!(v & ~Z_EROFS_PAGE_EIO)) { 163a9a94d93SGao Xiang set_page_private(page, 0); 164a9a94d93SGao Xiang ClearPagePrivate(page); 165a9a94d93SGao Xiang if (!(v & Z_EROFS_PAGE_EIO)) 166a9a94d93SGao Xiang SetPageUptodate(page); 167a9a94d93SGao Xiang unlock_page(page); 168a9a94d93SGao Xiang } 169a9a94d93SGao Xiang } 170a9a94d93SGao Xiang 171a9a94d93SGao Xiang #define Z_EROFS_ONSTACK_PAGES 32 172a9a94d93SGao Xiang 17347e4937aSGao Xiang /* 1749f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 1759f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 1769f6cc76eSGao Xiang */ 1779f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 1789f6cc76eSGao Xiang struct kmem_cache *slab; 1799f6cc76eSGao Xiang unsigned int maxpages; 1809f6cc76eSGao Xiang char name[48]; 1819f6cc76eSGao Xiang }; 1829f6cc76eSGao Xiang 1839f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 1849f6cc76eSGao Xiang 1859f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 1869f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 1879f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 1889f6cc76eSGao Xiang }; 1899f6cc76eSGao Xiang 19006a304cdSGao Xiang struct z_erofs_bvec_iter { 19106a304cdSGao Xiang struct page *bvpage; 19206a304cdSGao Xiang struct z_erofs_bvset *bvset; 19306a304cdSGao Xiang unsigned int nr, cur; 19406a304cdSGao Xiang }; 19506a304cdSGao Xiang 19606a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 19706a304cdSGao Xiang { 19806a304cdSGao Xiang if (iter->bvpage) 19906a304cdSGao Xiang kunmap_local(iter->bvset); 20006a304cdSGao Xiang return iter->bvpage; 20106a304cdSGao Xiang } 20206a304cdSGao Xiang 20306a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 20406a304cdSGao Xiang { 20506a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 20606a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 20706a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 20806a304cdSGao Xiang struct page *oldpage; 20906a304cdSGao Xiang 21006a304cdSGao Xiang DBG_BUGON(!nextpage); 21106a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 21206a304cdSGao Xiang iter->bvpage = nextpage; 21306a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 21406a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 21506a304cdSGao Xiang iter->cur = 0; 21606a304cdSGao Xiang return oldpage; 21706a304cdSGao Xiang } 21806a304cdSGao Xiang 21906a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 22006a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 22106a304cdSGao Xiang unsigned int bootstrap_nr, 22206a304cdSGao Xiang unsigned int cur) 22306a304cdSGao Xiang { 22406a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 22506a304cdSGao Xiang .nr = bootstrap_nr, 22606a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 22706a304cdSGao Xiang }; 22806a304cdSGao Xiang 22906a304cdSGao Xiang while (cur > iter->nr) { 23006a304cdSGao Xiang cur -= iter->nr; 23106a304cdSGao Xiang z_erofs_bvset_flip(iter); 23206a304cdSGao Xiang } 23306a304cdSGao Xiang iter->cur = cur; 23406a304cdSGao Xiang } 23506a304cdSGao Xiang 23606a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 23706a304cdSGao Xiang struct z_erofs_bvec *bvec, 2386ab5eed6SGao Xiang struct page **candidate_bvpage, 2396ab5eed6SGao Xiang struct page **pagepool) 24006a304cdSGao Xiang { 24105b63d2bSGao Xiang if (iter->cur >= iter->nr) { 24205b63d2bSGao Xiang struct page *nextpage = *candidate_bvpage; 24306a304cdSGao Xiang 24405b63d2bSGao Xiang if (!nextpage) { 2456ab5eed6SGao Xiang nextpage = erofs_allocpage(pagepool, GFP_NOFS); 24605b63d2bSGao Xiang if (!nextpage) 24705b63d2bSGao Xiang return -ENOMEM; 24805b63d2bSGao Xiang set_page_private(nextpage, Z_EROFS_SHORTLIVED_PAGE); 24905b63d2bSGao Xiang } 25006a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 25105b63d2bSGao Xiang iter->bvset->nextpage = nextpage; 25206a304cdSGao Xiang z_erofs_bvset_flip(iter); 25306a304cdSGao Xiang 25406a304cdSGao Xiang iter->bvset->nextpage = NULL; 25506a304cdSGao Xiang *candidate_bvpage = NULL; 25606a304cdSGao Xiang } 25706a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 25806a304cdSGao Xiang return 0; 25906a304cdSGao Xiang } 26006a304cdSGao Xiang 26106a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 26206a304cdSGao Xiang struct z_erofs_bvec *bvec, 26306a304cdSGao Xiang struct page **old_bvpage) 26406a304cdSGao Xiang { 26506a304cdSGao Xiang if (iter->cur == iter->nr) 26606a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 26706a304cdSGao Xiang else 26806a304cdSGao Xiang *old_bvpage = NULL; 26906a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 27006a304cdSGao Xiang } 27106a304cdSGao Xiang 2729f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 2739f6cc76eSGao Xiang { 2749f6cc76eSGao Xiang int i; 2759f6cc76eSGao Xiang 2769f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 2779f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 2789f6cc76eSGao Xiang continue; 2799f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 2809f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 2819f6cc76eSGao Xiang } 2829f6cc76eSGao Xiang } 2839f6cc76eSGao Xiang 2849f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 2859f6cc76eSGao Xiang { 2869f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 2879f6cc76eSGao Xiang struct z_erofs_pcluster *a; 2889f6cc76eSGao Xiang unsigned int size; 2899f6cc76eSGao Xiang 2909f6cc76eSGao Xiang for (pcs = pcluster_pool; 2919f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 292ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 2939f6cc76eSGao Xiang 2949f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 2959f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 2969f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 2979f6cc76eSGao Xiang if (pcs->slab) 2989f6cc76eSGao Xiang continue; 2999f6cc76eSGao Xiang 3009f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 3019f6cc76eSGao Xiang return -ENOMEM; 3029f6cc76eSGao Xiang } 3039f6cc76eSGao Xiang return 0; 3049f6cc76eSGao Xiang } 3059f6cc76eSGao Xiang 3069f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 3079f6cc76eSGao Xiang { 3089f6cc76eSGao Xiang int i; 3099f6cc76eSGao Xiang 3109f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3119f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3129f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 3139f6cc76eSGao Xiang 3149f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 3159f6cc76eSGao Xiang continue; 3169f6cc76eSGao Xiang 3179f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 3189f6cc76eSGao Xiang if (!pcl) 3199f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 3209f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 3219f6cc76eSGao Xiang return pcl; 3229f6cc76eSGao Xiang } 3239f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 3249f6cc76eSGao Xiang } 3259f6cc76eSGao Xiang 3269f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 3279f6cc76eSGao Xiang { 328cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 3299f6cc76eSGao Xiang int i; 3309f6cc76eSGao Xiang 3319f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3329f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3339f6cc76eSGao Xiang 334cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 3359f6cc76eSGao Xiang continue; 3369f6cc76eSGao Xiang 3379f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 3389f6cc76eSGao Xiang return; 3399f6cc76eSGao Xiang } 3409f6cc76eSGao Xiang DBG_BUGON(1); 3419f6cc76eSGao Xiang } 3429f6cc76eSGao Xiang 34347e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 34447e4937aSGao Xiang 3453fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 3463fffb589SSandeep Dhavale static struct kthread_worker __rcu **z_erofs_pcpu_workers; 3473fffb589SSandeep Dhavale 3483fffb589SSandeep Dhavale static void erofs_destroy_percpu_workers(void) 34947e4937aSGao Xiang { 3503fffb589SSandeep Dhavale struct kthread_worker *worker; 3513fffb589SSandeep Dhavale unsigned int cpu; 3523fffb589SSandeep Dhavale 3533fffb589SSandeep Dhavale for_each_possible_cpu(cpu) { 3543fffb589SSandeep Dhavale worker = rcu_dereference_protected( 3553fffb589SSandeep Dhavale z_erofs_pcpu_workers[cpu], 1); 3563fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 3573fffb589SSandeep Dhavale if (worker) 3583fffb589SSandeep Dhavale kthread_destroy_worker(worker); 3593fffb589SSandeep Dhavale } 3603fffb589SSandeep Dhavale kfree(z_erofs_pcpu_workers); 36147e4937aSGao Xiang } 36247e4937aSGao Xiang 3633fffb589SSandeep Dhavale static struct kthread_worker *erofs_init_percpu_worker(int cpu) 36447e4937aSGao Xiang { 3653fffb589SSandeep Dhavale struct kthread_worker *worker = 3663fffb589SSandeep Dhavale kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 36747e4937aSGao Xiang 3683fffb589SSandeep Dhavale if (IS_ERR(worker)) 3693fffb589SSandeep Dhavale return worker; 3703fffb589SSandeep Dhavale if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 3713fffb589SSandeep Dhavale sched_set_fifo_low(worker->task); 3723fffb589SSandeep Dhavale return worker; 3733fffb589SSandeep Dhavale } 3743fffb589SSandeep Dhavale 3753fffb589SSandeep Dhavale static int erofs_init_percpu_workers(void) 3763fffb589SSandeep Dhavale { 3773fffb589SSandeep Dhavale struct kthread_worker *worker; 3783fffb589SSandeep Dhavale unsigned int cpu; 3793fffb589SSandeep Dhavale 3803fffb589SSandeep Dhavale z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 3813fffb589SSandeep Dhavale sizeof(struct kthread_worker *), GFP_ATOMIC); 3823fffb589SSandeep Dhavale if (!z_erofs_pcpu_workers) 3833fffb589SSandeep Dhavale return -ENOMEM; 3843fffb589SSandeep Dhavale 3853fffb589SSandeep Dhavale for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 3863fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 3873fffb589SSandeep Dhavale if (!IS_ERR(worker)) 3883fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 3893fffb589SSandeep Dhavale } 3903fffb589SSandeep Dhavale return 0; 3913fffb589SSandeep Dhavale } 3923fffb589SSandeep Dhavale #else 3933fffb589SSandeep Dhavale static inline void erofs_destroy_percpu_workers(void) {} 3943fffb589SSandeep Dhavale static inline int erofs_init_percpu_workers(void) { return 0; } 3953fffb589SSandeep Dhavale #endif 3963fffb589SSandeep Dhavale 3973fffb589SSandeep Dhavale #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 3983fffb589SSandeep Dhavale static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 3993fffb589SSandeep Dhavale static enum cpuhp_state erofs_cpuhp_state; 4003fffb589SSandeep Dhavale 4013fffb589SSandeep Dhavale static int erofs_cpu_online(unsigned int cpu) 4023fffb589SSandeep Dhavale { 4033fffb589SSandeep Dhavale struct kthread_worker *worker, *old; 4043fffb589SSandeep Dhavale 4053fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 4063fffb589SSandeep Dhavale if (IS_ERR(worker)) 4073fffb589SSandeep Dhavale return PTR_ERR(worker); 4083fffb589SSandeep Dhavale 4093fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4103fffb589SSandeep Dhavale old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4113fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4123fffb589SSandeep Dhavale if (!old) 4133fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 4143fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4153fffb589SSandeep Dhavale if (old) 4163fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4173fffb589SSandeep Dhavale return 0; 4183fffb589SSandeep Dhavale } 4193fffb589SSandeep Dhavale 4203fffb589SSandeep Dhavale static int erofs_cpu_offline(unsigned int cpu) 4213fffb589SSandeep Dhavale { 4223fffb589SSandeep Dhavale struct kthread_worker *worker; 4233fffb589SSandeep Dhavale 4243fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4253fffb589SSandeep Dhavale worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4263fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4273fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 4283fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4293fffb589SSandeep Dhavale 4303fffb589SSandeep Dhavale synchronize_rcu(); 4313fffb589SSandeep Dhavale if (worker) 4323fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4333fffb589SSandeep Dhavale return 0; 4343fffb589SSandeep Dhavale } 4353fffb589SSandeep Dhavale 4363fffb589SSandeep Dhavale static int erofs_cpu_hotplug_init(void) 4373fffb589SSandeep Dhavale { 4383fffb589SSandeep Dhavale int state; 4393fffb589SSandeep Dhavale 4403fffb589SSandeep Dhavale state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 4413fffb589SSandeep Dhavale "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 4423fffb589SSandeep Dhavale if (state < 0) 4433fffb589SSandeep Dhavale return state; 4443fffb589SSandeep Dhavale 4453fffb589SSandeep Dhavale erofs_cpuhp_state = state; 4463fffb589SSandeep Dhavale return 0; 4473fffb589SSandeep Dhavale } 4483fffb589SSandeep Dhavale 4493fffb589SSandeep Dhavale static void erofs_cpu_hotplug_destroy(void) 4503fffb589SSandeep Dhavale { 4513fffb589SSandeep Dhavale if (erofs_cpuhp_state) 4523fffb589SSandeep Dhavale cpuhp_remove_state_nocalls(erofs_cpuhp_state); 4533fffb589SSandeep Dhavale } 4543fffb589SSandeep Dhavale #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 4553fffb589SSandeep Dhavale static inline int erofs_cpu_hotplug_init(void) { return 0; } 4563fffb589SSandeep Dhavale static inline void erofs_cpu_hotplug_destroy(void) {} 4573fffb589SSandeep Dhavale #endif 4583fffb589SSandeep Dhavale 4593fffb589SSandeep Dhavale void z_erofs_exit_zip_subsystem(void) 4603fffb589SSandeep Dhavale { 4613fffb589SSandeep Dhavale erofs_cpu_hotplug_destroy(); 4623fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4633fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4643fffb589SSandeep Dhavale z_erofs_destroy_pcluster_pool(); 46547e4937aSGao Xiang } 46647e4937aSGao Xiang 46747e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 46847e4937aSGao Xiang { 4699f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 47047e4937aSGao Xiang 4719f6cc76eSGao Xiang if (err) 4723fffb589SSandeep Dhavale goto out_error_pcluster_pool; 4733fffb589SSandeep Dhavale 4743fffb589SSandeep Dhavale z_erofs_workqueue = alloc_workqueue("erofs_worker", 4753fffb589SSandeep Dhavale WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 4768d1b80a7SDan Carpenter if (!z_erofs_workqueue) { 4778d1b80a7SDan Carpenter err = -ENOMEM; 4783fffb589SSandeep Dhavale goto out_error_workqueue_init; 4798d1b80a7SDan Carpenter } 4803fffb589SSandeep Dhavale 4813fffb589SSandeep Dhavale err = erofs_init_percpu_workers(); 4829f6cc76eSGao Xiang if (err) 4833fffb589SSandeep Dhavale goto out_error_pcpu_worker; 4843fffb589SSandeep Dhavale 4853fffb589SSandeep Dhavale err = erofs_cpu_hotplug_init(); 4863fffb589SSandeep Dhavale if (err < 0) 4873fffb589SSandeep Dhavale goto out_error_cpuhp_init; 4883fffb589SSandeep Dhavale return err; 4893fffb589SSandeep Dhavale 4903fffb589SSandeep Dhavale out_error_cpuhp_init: 4913fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4923fffb589SSandeep Dhavale out_error_pcpu_worker: 4933fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4943fffb589SSandeep Dhavale out_error_workqueue_init: 4959f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 4963fffb589SSandeep Dhavale out_error_pcluster_pool: 4979f6cc76eSGao Xiang return err; 49847e4937aSGao Xiang } 49947e4937aSGao Xiang 500db166fc2SGao Xiang enum z_erofs_pclustermode { 501db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 50247e4937aSGao Xiang /* 503db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 5040b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 5050b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 506387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 5070b964600SGao Xiang */ 508db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 50947e4937aSGao Xiang /* 51047e4937aSGao Xiang * The current collection has been linked with the owned chain, and 51147e4937aSGao Xiang * could also be linked with the remaining collections, which means 51247e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 51347e4937aSGao Xiang * the current collection can safely use the whole page (since 51447e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 51547e4937aSGao Xiang * illustrated below: 51647e4937aSGao Xiang * ________________________________________________________________ 51747e4937aSGao Xiang * | tail (partial) page | head (partial) page | 51847e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 519967c28b2SGao Xiang * | | | 520967c28b2SGao Xiang * |__PCLUSTER_FOLLOWED___|___________PCLUSTER_FOLLOWED____________| 52147e4937aSGao Xiang * 52247e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 52347e4937aSGao Xiang */ 524db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 52547e4937aSGao Xiang }; 52647e4937aSGao Xiang 5275c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 5285c6dcc57SGao Xiang struct inode *const inode; 5295c6dcc57SGao Xiang struct erofs_map_blocks map; 53006a304cdSGao Xiang struct z_erofs_bvec_iter biter; 53147e4937aSGao Xiang 5326ab5eed6SGao Xiang struct page *pagepool; 53306a304cdSGao Xiang struct page *candidate_bvpage; 534967c28b2SGao Xiang struct z_erofs_pcluster *pcl; 53547e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 536db166fc2SGao Xiang enum z_erofs_pclustermode mode; 53747e4937aSGao Xiang 53847e4937aSGao Xiang /* used for applying cache strategy on the fly */ 53947e4937aSGao Xiang bool backmost; 54047e4937aSGao Xiang erofs_off_t headoffset; 541ed722fbcSGao Xiang 542ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 543ed722fbcSGao Xiang unsigned int icur; 54447e4937aSGao Xiang }; 54547e4937aSGao Xiang 54647e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 5475c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 548db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 54947e4937aSGao Xiang 5501282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 5511282dea3SGao Xiang { 5521282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 5531282dea3SGao Xiang 5541282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 5551282dea3SGao Xiang return false; 5561282dea3SGao Xiang 5571282dea3SGao Xiang if (fe->backmost) 5581282dea3SGao Xiang return true; 5591282dea3SGao Xiang 5601282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 5611282dea3SGao Xiang fe->map.m_la < fe->headoffset) 5621282dea3SGao Xiang return true; 5631282dea3SGao Xiang 5641282dea3SGao Xiang return false; 5651282dea3SGao Xiang } 5661282dea3SGao Xiang 5676ab5eed6SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe) 56847e4937aSGao Xiang { 5696f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 5705c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 5711282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 57247e4937aSGao Xiang bool standalone = true; 5736f39d1e1SGao Xiang /* 5746f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 5756f39d1e1SGao Xiang * can be used if low memory otherwise. 5766f39d1e1SGao Xiang */ 5771825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 5781825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 579ed722fbcSGao Xiang unsigned int i; 58047e4937aSGao Xiang 581db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 58247e4937aSGao Xiang return; 58347e4937aSGao Xiang 584ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 58547e4937aSGao Xiang struct page *page; 586b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 5871825c8d7SGao Xiang struct page *newpage = NULL; 58847e4937aSGao Xiang 58947e4937aSGao Xiang /* the compressed page was loaded before */ 590ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 59147e4937aSGao Xiang continue; 59247e4937aSGao Xiang 593ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 59447e4937aSGao Xiang 59547e4937aSGao Xiang if (page) { 596b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 5970b964600SGao Xiang } else { 5980b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 5990b964600SGao Xiang standalone = false; 6001282dea3SGao Xiang if (!shouldalloc) 6011282dea3SGao Xiang continue; 6021282dea3SGao Xiang 6031282dea3SGao Xiang /* 6041282dea3SGao Xiang * try to use cached I/O if page allocation 6051282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 6061282dea3SGao Xiang * to avoid any direct reclaim. 6071282dea3SGao Xiang */ 6086ab5eed6SGao Xiang newpage = erofs_allocpage(&fe->pagepool, gfp); 6091825c8d7SGao Xiang if (!newpage) 61047e4937aSGao Xiang continue; 6111282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 612b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 61347e4937aSGao Xiang } 61447e4937aSGao Xiang 615b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 61647e4937aSGao Xiang continue; 61747e4937aSGao Xiang 618eaa9172aSGao Xiang if (page) 61947e4937aSGao Xiang put_page(page); 620eaa9172aSGao Xiang else if (newpage) 6216ab5eed6SGao Xiang erofs_pagepool_add(&fe->pagepool, newpage); 62247e4937aSGao Xiang } 62347e4937aSGao Xiang 6240b964600SGao Xiang /* 6250b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 6260b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 6270b964600SGao Xiang */ 6280b964600SGao Xiang if (standalone) 629db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 63047e4937aSGao Xiang } 63147e4937aSGao Xiang 63247e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 63347e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 63447e4937aSGao Xiang struct erofs_workgroup *grp) 63547e4937aSGao Xiang { 63647e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 63747e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 63847e4937aSGao Xiang int i; 63947e4937aSGao Xiang 640cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 64147e4937aSGao Xiang /* 6427674a42fSGao Xiang * refcount of workgroup is now freezed as 0, 64347e4937aSGao Xiang * therefore no need to worry about available decompression users. 64447e4937aSGao Xiang */ 6459f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 646ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 64747e4937aSGao Xiang 64847e4937aSGao Xiang if (!page) 64947e4937aSGao Xiang continue; 65047e4937aSGao Xiang 65147e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 65247e4937aSGao Xiang if (!trylock_page(page)) 65347e4937aSGao Xiang return -EBUSY; 65447e4937aSGao Xiang 655f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 65647e4937aSGao Xiang continue; 65747e4937aSGao Xiang 65847e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 659ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6606aaa7b06SGao Xiang detach_page_private(page); 66147e4937aSGao Xiang unlock_page(page); 66247e4937aSGao Xiang } 66347e4937aSGao Xiang return 0; 66447e4937aSGao Xiang } 66547e4937aSGao Xiang 6667b4e372cSGao Xiang static bool z_erofs_cache_release_folio(struct folio *folio, gfp_t gfp) 66747e4937aSGao Xiang { 6687b4e372cSGao Xiang struct z_erofs_pcluster *pcl = folio_get_private(folio); 6697b4e372cSGao Xiang bool ret; 6707b4e372cSGao Xiang int i; 6717b4e372cSGao Xiang 6727b4e372cSGao Xiang if (!folio_test_private(folio)) 6737b4e372cSGao Xiang return true; 67447e4937aSGao Xiang 6757b4e372cSGao Xiang ret = false; 6767674a42fSGao Xiang spin_lock(&pcl->obj.lockref.lock); 6777674a42fSGao Xiang if (pcl->obj.lockref.count > 0) 6787674a42fSGao Xiang goto out; 6797674a42fSGao Xiang 680cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 6819f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 6827b4e372cSGao Xiang if (pcl->compressed_bvecs[i].page == &folio->page) { 683ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6847b4e372cSGao Xiang ret = true; 68547e4937aSGao Xiang break; 68647e4937aSGao Xiang } 68747e4937aSGao Xiang } 6886aaa7b06SGao Xiang if (ret) 6897b4e372cSGao Xiang folio_detach_private(folio); 6907674a42fSGao Xiang out: 6917674a42fSGao Xiang spin_unlock(&pcl->obj.lockref.lock); 69247e4937aSGao Xiang return ret; 69347e4937aSGao Xiang } 69447e4937aSGao Xiang 6957b4e372cSGao Xiang /* 6967b4e372cSGao Xiang * It will be called only on inode eviction. In case that there are still some 6977b4e372cSGao Xiang * decompression requests in progress, wait with rescheduling for a bit here. 6987b4e372cSGao Xiang * An extra lock could be introduced instead but it seems unnecessary. 6997b4e372cSGao Xiang */ 7007b4e372cSGao Xiang static void z_erofs_cache_invalidate_folio(struct folio *folio, 7017b4e372cSGao Xiang size_t offset, size_t length) 7027b4e372cSGao Xiang { 7037b4e372cSGao Xiang const size_t stop = length + offset; 7047b4e372cSGao Xiang 7057b4e372cSGao Xiang /* Check for potential overflow in debug mode */ 7067b4e372cSGao Xiang DBG_BUGON(stop > folio_size(folio) || stop < length); 7077b4e372cSGao Xiang 7087b4e372cSGao Xiang if (offset == 0 && stop == folio_size(folio)) 7097b4e372cSGao Xiang while (!z_erofs_cache_release_folio(folio, GFP_NOFS)) 7107b4e372cSGao Xiang cond_resched(); 7117b4e372cSGao Xiang } 7127b4e372cSGao Xiang 7137b4e372cSGao Xiang static const struct address_space_operations z_erofs_cache_aops = { 7147b4e372cSGao Xiang .release_folio = z_erofs_cache_release_folio, 7157b4e372cSGao Xiang .invalidate_folio = z_erofs_cache_invalidate_folio, 7167b4e372cSGao Xiang }; 7177b4e372cSGao Xiang 7187b4e372cSGao Xiang int erofs_init_managed_cache(struct super_block *sb) 7197b4e372cSGao Xiang { 7207b4e372cSGao Xiang struct inode *const inode = new_inode(sb); 7217b4e372cSGao Xiang 7227b4e372cSGao Xiang if (!inode) 7237b4e372cSGao Xiang return -ENOMEM; 7247b4e372cSGao Xiang 7257b4e372cSGao Xiang set_nlink(inode, 1); 7267b4e372cSGao Xiang inode->i_size = OFFSET_MAX; 7277b4e372cSGao Xiang inode->i_mapping->a_ops = &z_erofs_cache_aops; 7287b4e372cSGao Xiang mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS); 7297b4e372cSGao Xiang EROFS_SB(sb)->managed_cache = inode; 7307b4e372cSGao Xiang return 0; 7317b4e372cSGao Xiang } 7327b4e372cSGao Xiang 7335c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 734ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 73547e4937aSGao Xiang { 7365c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 73747e4937aSGao Xiang 738ed722fbcSGao Xiang while (fe->icur > 0) { 739ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 740ed722fbcSGao Xiang NULL, bvec->page)) { 741ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 74247e4937aSGao Xiang return true; 743ed722fbcSGao Xiang } 744ed722fbcSGao Xiang } 74547e4937aSGao Xiang return false; 74647e4937aSGao Xiang } 74747e4937aSGao Xiang 74887ca34a7SGao Xiang /* callers must be with pcluster lock held */ 7495c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 7505b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 75147e4937aSGao Xiang { 75247e4937aSGao Xiang int ret; 75347e4937aSGao Xiang 754db166fc2SGao Xiang if (exclusive) { 75506a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 756ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 75747e4937aSGao Xiang return 0; 75806a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 759db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 76006a304cdSGao Xiang !fe->candidate_bvpage) 76106a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 76206a304cdSGao Xiang } 7636ab5eed6SGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage, 7646ab5eed6SGao Xiang &fe->pagepool); 76506a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 76606a304cdSGao Xiang return ret; 76747e4937aSGao Xiang } 76847e4937aSGao Xiang 7695c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 77047e4937aSGao Xiang { 7715c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 7725c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 77347e4937aSGao Xiang 774473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 775473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 776473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 77747e4937aSGao Xiang *owned_head = &pcl->next; 778473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 779db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 780473e15b0SGao Xiang return; 781473e15b0SGao Xiang } 782473e15b0SGao Xiang 783967c28b2SGao Xiang /* type 2, it belongs to an ongoing chain */ 784db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 78547e4937aSGao Xiang } 78647e4937aSGao Xiang 78783a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 78847e4937aSGao Xiang { 78983a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 790cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 79147e4937aSGao Xiang struct z_erofs_pcluster *pcl; 79264094a04SGao Xiang struct erofs_workgroup *grp; 79347e4937aSGao Xiang int err; 79447e4937aSGao Xiang 795c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 796c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 7978f899262SGao Xiang DBG_BUGON(1); 7988f899262SGao Xiang return -EFSCORRUPTED; 7998f899262SGao Xiang } 8008f899262SGao Xiang 8019f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 802cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 803cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 8049f6cc76eSGao Xiang if (IS_ERR(pcl)) 8059f6cc76eSGao Xiang return PTR_ERR(pcl); 80647e4937aSGao Xiang 8077674a42fSGao Xiang spin_lock_init(&pcl->obj.lockref.lock); 8088f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 8092bfab9c0SGao Xiang pcl->length = 0; 8102bfab9c0SGao Xiang pcl->partial = true; 81147e4937aSGao Xiang 81247e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 8135c6dcc57SGao Xiang pcl->next = fe->owned_head; 81487ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 815db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 81647e4937aSGao Xiang 81747e4937aSGao Xiang /* 81847e4937aSGao Xiang * lock all primary followed works before visible to others 81947e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 82047e4937aSGao Xiang */ 82187ca34a7SGao Xiang mutex_init(&pcl->lock); 82287ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 82347e4937aSGao Xiang 824cecf864dSYue Hu if (ztailpacking) { 825cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 8263acea5fcSJingbo Xu pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 827cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 828cecf864dSYue Hu } else { 829cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 830cecf864dSYue Hu 83183a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 83264094a04SGao Xiang if (IS_ERR(grp)) { 83364094a04SGao Xiang err = PTR_ERR(grp); 83464094a04SGao Xiang goto err_out; 83564094a04SGao Xiang } 83664094a04SGao Xiang 83764094a04SGao Xiang if (grp != &pcl->obj) { 8385c6dcc57SGao Xiang fe->pcl = container_of(grp, 839cecf864dSYue Hu struct z_erofs_pcluster, obj); 84064094a04SGao Xiang err = -EEXIST; 84164094a04SGao Xiang goto err_out; 84247e4937aSGao Xiang } 843cecf864dSYue Hu } 8445c6dcc57SGao Xiang fe->owned_head = &pcl->next; 8455c6dcc57SGao Xiang fe->pcl = pcl; 8469e579fc1SGao Xiang return 0; 84764094a04SGao Xiang 84864094a04SGao Xiang err_out: 84987ca34a7SGao Xiang mutex_unlock(&pcl->lock); 8509f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 85164094a04SGao Xiang return err; 85247e4937aSGao Xiang } 85347e4937aSGao Xiang 85483a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 85547e4937aSGao Xiang { 85683a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 8570d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 8589e579fc1SGao Xiang int ret; 85947e4937aSGao Xiang 86087ca34a7SGao Xiang DBG_BUGON(fe->pcl); 86147e4937aSGao Xiang 86287ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 8635c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 86447e4937aSGao Xiang 8650d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 8660d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 8670d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 8680d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 86947e4937aSGao Xiang DBG_BUGON(1); 870cecf864dSYue Hu return -EFSCORRUPTED; 871cecf864dSYue Hu } 87247e4937aSGao Xiang 87364094a04SGao Xiang if (grp) { 8745c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 8750d823b42SGao Xiang ret = -EEXIST; 87664094a04SGao Xiang } else { 87783a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 87864094a04SGao Xiang } 87947e4937aSGao Xiang 8800d823b42SGao Xiang if (ret == -EEXIST) { 881267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 882267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 8830d823b42SGao Xiang } else if (ret) { 8840d823b42SGao Xiang return ret; 8850d823b42SGao Xiang } 88606a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 887387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 88881382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 889ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 89047e4937aSGao Xiang return 0; 89147e4937aSGao Xiang } 89247e4937aSGao Xiang 89347e4937aSGao Xiang /* 89447e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 89547e4937aSGao Xiang * only after a RCU grace period. 89647e4937aSGao Xiang */ 89747e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 89847e4937aSGao Xiang { 89987ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 90087ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 90147e4937aSGao Xiang } 90247e4937aSGao Xiang 90347e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 90447e4937aSGao Xiang { 90547e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 90647e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 90747e4937aSGao Xiang 90887ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 90947e4937aSGao Xiang } 91047e4937aSGao Xiang 9115c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 91247e4937aSGao Xiang { 91387ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 91447e4937aSGao Xiang 91587ca34a7SGao Xiang if (!pcl) 91647e4937aSGao Xiang return false; 91747e4937aSGao Xiang 91806a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 91987ca34a7SGao Xiang mutex_unlock(&pcl->lock); 92047e4937aSGao Xiang 92105b63d2bSGao Xiang if (fe->candidate_bvpage) 92206a304cdSGao Xiang fe->candidate_bvpage = NULL; 92306a304cdSGao Xiang 92447e4937aSGao Xiang /* 92547e4937aSGao Xiang * if all pending pages are added, don't hold its reference 92647e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 92747e4937aSGao Xiang */ 928db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 92987ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 93047e4937aSGao Xiang 93187ca34a7SGao Xiang fe->pcl = NULL; 93247e4937aSGao Xiang return true; 93347e4937aSGao Xiang } 93447e4937aSGao Xiang 935b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 936b15b2e30SYue Hu struct page *page, unsigned int pageofs, 937b15b2e30SYue Hu unsigned int len) 938b15b2e30SYue Hu { 9393acea5fcSJingbo Xu struct super_block *sb = inode->i_sb; 940b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 941b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 942b15b2e30SYue Hu u8 *src, *dst; 943b15b2e30SYue Hu unsigned int i, cnt; 944b15b2e30SYue Hu 945e5126de1SYue Hu if (!packed_inode) 946e5126de1SYue Hu return -EFSCORRUPTED; 947e5126de1SYue Hu 948eb2c5e41SGao Xiang buf.inode = packed_inode; 949b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 950b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 951b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 9523acea5fcSJingbo Xu sb->s_blocksize - erofs_blkoff(sb, pos)); 953eb2c5e41SGao Xiang src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 954b15b2e30SYue Hu if (IS_ERR(src)) { 955b15b2e30SYue Hu erofs_put_metabuf(&buf); 956b15b2e30SYue Hu return PTR_ERR(src); 957b15b2e30SYue Hu } 958b15b2e30SYue Hu 959b15b2e30SYue Hu dst = kmap_local_page(page); 9603acea5fcSJingbo Xu memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 961b15b2e30SYue Hu kunmap_local(dst); 962b15b2e30SYue Hu pos += cnt; 963b15b2e30SYue Hu } 964b15b2e30SYue Hu erofs_put_metabuf(&buf); 965b15b2e30SYue Hu return 0; 966b15b2e30SYue Hu } 967b15b2e30SYue Hu 96847e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 9696ab5eed6SGao Xiang struct page *page) 97047e4937aSGao Xiang { 97147e4937aSGao Xiang struct inode *const inode = fe->inode; 97247e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 97347e4937aSGao Xiang const loff_t offset = page_offset(page); 9745b220b20SGao Xiang bool tight = true, exclusive; 9752bfab9c0SGao Xiang unsigned int cur, end, spiltted; 97647e4937aSGao Xiang int err = 0; 97747e4937aSGao Xiang 97847e4937aSGao Xiang /* register locked file pages as online pages in pack */ 97947e4937aSGao Xiang z_erofs_onlinepage_init(page); 98047e4937aSGao Xiang 98147e4937aSGao Xiang spiltted = 0; 98247e4937aSGao Xiang end = PAGE_SIZE; 98347e4937aSGao Xiang repeat: 98447e4937aSGao Xiang cur = end - 1; 98547e4937aSGao Xiang 98639397a46SGao Xiang if (offset + cur < map->m_la || 98739397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 9885c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 98947e4937aSGao Xiang fe->backmost = false; 99047e4937aSGao Xiang map->m_la = offset + cur; 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 } else { 99639397a46SGao Xiang if (fe->pcl) 99739397a46SGao Xiang goto hitted; 99839397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 99939397a46SGao Xiang } 100047e4937aSGao Xiang 1001b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 1002b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 100347e4937aSGao Xiang goto hitted; 100447e4937aSGao Xiang 100583a386c0SGao Xiang err = z_erofs_collector_begin(fe); 10068d8a09b0SGao Xiang if (err) 100767148551SGao Xiang goto out; 100847e4937aSGao Xiang 10095c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 101009c54379SGao Xiang void *mp; 1011cecf864dSYue Hu 101209c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 10133acea5fcSJingbo Xu erofs_blknr(inode->i_sb, map->m_pa), 10143acea5fcSJingbo Xu EROFS_NO_KMAP); 101509c54379SGao Xiang if (IS_ERR(mp)) { 101609c54379SGao Xiang err = PTR_ERR(mp); 1017cecf864dSYue Hu erofs_err(inode->i_sb, 1018cecf864dSYue Hu "failed to get inline page, err %d", err); 101967148551SGao Xiang goto out; 1020cecf864dSYue Hu } 102109c54379SGao Xiang get_page(fe->map.buf.page); 1022ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 1023ed722fbcSGao Xiang fe->map.buf.page); 1024db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 1025cecf864dSYue Hu } else { 10266f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 10276ab5eed6SGao Xiang z_erofs_bind_cache(fe); 1028cecf864dSYue Hu } 102947e4937aSGao Xiang hitted: 1030dc76ea8cSGao Xiang /* 1031dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 1032dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 1033dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 1034387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 1035dc76ea8cSGao Xiang */ 1036967c28b2SGao Xiang tight &= (fe->mode > Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1037dc76ea8cSGao Xiang 103847e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 10398d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 104047e4937aSGao Xiang zero_user_segment(page, cur, end); 104147e4937aSGao Xiang goto next_part; 104247e4937aSGao Xiang } 1043b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 1044b15b2e30SYue Hu unsigned int pageofs, skip, len; 1045b15b2e30SYue Hu 1046b15b2e30SYue Hu if (offset > map->m_la) { 1047b15b2e30SYue Hu pageofs = 0; 1048b15b2e30SYue Hu skip = offset - map->m_la; 1049b15b2e30SYue Hu } else { 1050b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 1051b15b2e30SYue Hu skip = 0; 1052b15b2e30SYue Hu } 1053b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 1054b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1055b15b2e30SYue Hu if (err) 1056b15b2e30SYue Hu goto out; 1057b15b2e30SYue Hu ++spiltted; 1058b15b2e30SYue Hu tight = false; 1059b15b2e30SYue Hu goto next_part; 1060b15b2e30SYue Hu } 106147e4937aSGao Xiang 10625b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 106347e4937aSGao Xiang if (cur) 1064db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 106547e4937aSGao Xiang 106606a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 106706a304cdSGao Xiang .page = page, 106806a304cdSGao Xiang .offset = offset - map->m_la, 106906a304cdSGao Xiang .end = end, 10705b220b20SGao Xiang }), exclusive); 107105b63d2bSGao Xiang if (err) 107267148551SGao Xiang goto out; 107347e4937aSGao Xiang 107467148551SGao Xiang z_erofs_onlinepage_split(page); 107547e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 107647e4937aSGao Xiang ++spiltted; 1077267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1078267f2492SGao Xiang fe->pcl->multibases = true; 10792bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 10802bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 10812bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 10822bfab9c0SGao Xiang } 1083e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1084e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1085e7933278SGao Xiang fe->pcl->length == map->m_llen) 1086e7933278SGao Xiang fe->pcl->partial = false; 108747e4937aSGao Xiang next_part: 10882bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 108947e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 10902bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 109147e4937aSGao Xiang 109247e4937aSGao Xiang end = cur; 109347e4937aSGao Xiang if (end > 0) 109447e4937aSGao Xiang goto repeat; 109547e4937aSGao Xiang 109647e4937aSGao Xiang out: 109767148551SGao Xiang if (err) 109867148551SGao Xiang z_erofs_page_mark_eio(page); 109947e4937aSGao Xiang z_erofs_onlinepage_endio(page); 110047e4937aSGao Xiang return err; 110147e4937aSGao Xiang } 110247e4937aSGao Xiang 1103ef4b4b46SYue Hu static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 110440452ffcSHuang Jianan unsigned int readahead_pages) 110540452ffcSHuang Jianan { 1106a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 110740452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 110840452ffcSHuang Jianan !readahead_pages) 110940452ffcSHuang Jianan return true; 111040452ffcSHuang Jianan 111140452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 111240452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 111340452ffcSHuang Jianan return true; 111440452ffcSHuang Jianan 111540452ffcSHuang Jianan return false; 111640452ffcSHuang Jianan } 111740452ffcSHuang Jianan 11186aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 11196aaa7b06SGao Xiang { 11206aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 11216aaa7b06SGao Xiang } 11226aaa7b06SGao Xiang 11234f05687fSGao Xiang struct z_erofs_decompress_backend { 11244f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 11254f05687fSGao Xiang struct super_block *sb; 11264f05687fSGao Xiang struct z_erofs_pcluster *pcl; 11274f05687fSGao Xiang 11284f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 11294f05687fSGao Xiang struct page **decompressed_pages; 11304f05687fSGao Xiang /* pages to keep the compressed data */ 11314f05687fSGao Xiang struct page **compressed_pages; 11324f05687fSGao Xiang 1133267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 11344f05687fSGao Xiang struct page **pagepool; 11352bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 11364f05687fSGao Xiang }; 11374f05687fSGao Xiang 1138267f2492SGao Xiang struct z_erofs_bvec_item { 1139267f2492SGao Xiang struct z_erofs_bvec bvec; 1140267f2492SGao Xiang struct list_head list; 1141267f2492SGao Xiang }; 1142267f2492SGao Xiang 1143267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 11443fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 11453fe96ee0SGao Xiang { 1146267f2492SGao Xiang struct z_erofs_bvec_item *item; 1147267f2492SGao Xiang 1148267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1149267f2492SGao Xiang unsigned int pgnr; 11503fe96ee0SGao Xiang 1151267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 11522bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 115363bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 11543fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1155267f2492SGao Xiang return; 11563fe96ee0SGao Xiang } 115763bbb856SGao Xiang } 11583fe96ee0SGao Xiang 1159267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1160267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1161267f2492SGao Xiang item->bvec = *bvec; 1162267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1163267f2492SGao Xiang } 1164267f2492SGao Xiang 1165267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1166267f2492SGao Xiang int err) 1167267f2492SGao Xiang { 1168267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1169267f2492SGao Xiang struct list_head *p, *n; 1170267f2492SGao Xiang 1171267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1172267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1173267f2492SGao Xiang unsigned int end, cur; 1174267f2492SGao Xiang void *dst, *src; 1175267f2492SGao Xiang 1176267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1177267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1178267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1179267f2492SGao Xiang bvi->bvec.end); 1180267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1181267f2492SGao Xiang while (cur < end) { 1182267f2492SGao Xiang unsigned int pgnr, scur, len; 1183267f2492SGao Xiang 1184267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1185267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1186267f2492SGao Xiang 1187267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1188267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1189267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1190267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1191267f2492SGao Xiang err = -EFSCORRUPTED; 1192267f2492SGao Xiang cur += len; 1193267f2492SGao Xiang continue; 1194267f2492SGao Xiang } 1195267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1196267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1197267f2492SGao Xiang kunmap_local(src); 1198267f2492SGao Xiang cur += len; 1199267f2492SGao Xiang } 1200267f2492SGao Xiang kunmap_local(dst); 1201267f2492SGao Xiang if (err) 1202267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 1203267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 1204267f2492SGao Xiang list_del(p); 1205267f2492SGao Xiang kfree(bvi); 1206267f2492SGao Xiang } 1207267f2492SGao Xiang } 1208267f2492SGao Xiang 1209267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 121042fec235SGao Xiang { 12114f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 121206a304cdSGao Xiang struct z_erofs_bvec_iter biter; 121306a304cdSGao Xiang struct page *old_bvpage; 1214267f2492SGao Xiang int i; 121542fec235SGao Xiang 1216387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 121742fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 121806a304cdSGao Xiang struct z_erofs_bvec bvec; 121942fec235SGao Xiang 122006a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 122142fec235SGao Xiang 122206a304cdSGao Xiang if (old_bvpage) 12234f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 122442fec235SGao Xiang 122506a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1226267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 122742fec235SGao Xiang } 122806a304cdSGao Xiang 122906a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 123006a304cdSGao Xiang if (old_bvpage) 12314f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 123242fec235SGao Xiang } 123342fec235SGao Xiang 12344f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 12354f05687fSGao Xiang bool *overlapped) 123667139e36SGao Xiang { 12374f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 123867139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 123967139e36SGao Xiang int i, err = 0; 124067139e36SGao Xiang 124167139e36SGao Xiang *overlapped = false; 124267139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1243ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1244ed722fbcSGao Xiang struct page *page = bvec->page; 124567139e36SGao Xiang 124667139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 124767139e36SGao Xiang if (!page) { 124867139e36SGao Xiang DBG_BUGON(1); 124967139e36SGao Xiang continue; 125067139e36SGao Xiang } 1251fe3e5914SGao Xiang be->compressed_pages[i] = page; 125267139e36SGao Xiang 125367139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 125467139e36SGao Xiang if (!PageUptodate(page)) 125567139e36SGao Xiang err = -EIO; 125667139e36SGao Xiang continue; 125767139e36SGao Xiang } 125867139e36SGao Xiang 125967139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 126067139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 12614f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 126267139e36SGao Xiang if (!PageUptodate(page)) 126367139e36SGao Xiang err = -EIO; 126467139e36SGao Xiang continue; 126567139e36SGao Xiang } 1266267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 126767139e36SGao Xiang *overlapped = true; 126867139e36SGao Xiang } 126967139e36SGao Xiang } 127067139e36SGao Xiang 1271fe3e5914SGao Xiang if (err) 12724f05687fSGao Xiang return err; 12734f05687fSGao Xiang return 0; 127467139e36SGao Xiang } 127567139e36SGao Xiang 12764f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 12774f05687fSGao Xiang int err) 127847e4937aSGao Xiang { 12794f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 12804f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1281cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1282597e2953SYue Hu const struct z_erofs_decompressor *decompressor = 1283597e2953SYue Hu &erofs_decompressors[pcl->algorithmformat]; 12842bfab9c0SGao Xiang unsigned int i, inputsize; 128567148551SGao Xiang int err2; 12862bfab9c0SGao Xiang struct page *page; 12872bfab9c0SGao Xiang bool overlapped; 128847e4937aSGao Xiang 128987ca34a7SGao Xiang mutex_lock(&pcl->lock); 12902bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 129147e4937aSGao Xiang 1292fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1293fe3e5914SGao Xiang be->decompressed_pages = NULL; 1294fe3e5914SGao Xiang be->compressed_pages = NULL; 1295fe3e5914SGao Xiang be->onstack_used = 0; 12962bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 12974f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 12982bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 12994f05687fSGao Xiang memset(be->decompressed_pages, 0, 13002bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1301fe3e5914SGao Xiang } 1302fe3e5914SGao Xiang 1303fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1304fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1305fe3e5914SGao Xiang 1306fe3e5914SGao Xiang if (!be->decompressed_pages) 13074f05687fSGao Xiang be->decompressed_pages = 1308647dd2c3SGao Xiang kvcalloc(be->nr_pages, sizeof(struct page *), 1309e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1310fe3e5914SGao Xiang if (!be->compressed_pages) 1311fe3e5914SGao Xiang be->compressed_pages = 1312647dd2c3SGao Xiang kvcalloc(pclusterpages, sizeof(struct page *), 1313fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 131447e4937aSGao Xiang 1315267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 13164f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 13174f05687fSGao Xiang if (err2) 13184f05687fSGao Xiang err = err2; 13198d8a09b0SGao Xiang if (err) 132047e4937aSGao Xiang goto out; 132147e4937aSGao Xiang 1322cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1323cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1324cecf864dSYue Hu else 1325cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1326cecf864dSYue Hu 1327597e2953SYue Hu err = decompressor->decompress(&(struct z_erofs_decompress_req) { 13284f05687fSGao Xiang .sb = be->sb, 13294f05687fSGao Xiang .in = be->compressed_pages, 13304f05687fSGao Xiang .out = be->decompressed_pages, 1331cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 133287ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 13339f6cc76eSGao Xiang .inputsize = inputsize, 13342bfab9c0SGao Xiang .outputsize = pcl->length, 133547e4937aSGao Xiang .alg = pcl->algorithmformat, 133647e4937aSGao Xiang .inplace_io = overlapped, 13372bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1338267f2492SGao Xiang .fillgaps = pcl->multibases, 13394f05687fSGao Xiang }, be->pagepool); 134047e4937aSGao Xiang 134147e4937aSGao Xiang out: 1342cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1343cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1344ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1345ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1346cecf864dSYue Hu put_page(page); 1347cecf864dSYue Hu } else { 1348cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1349ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 135047e4937aSGao Xiang 135147e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 135247e4937aSGao Xiang continue; 135347e4937aSGao Xiang 13546aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13554f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1356ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 135747e4937aSGao Xiang } 1358cecf864dSYue Hu } 1359fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1360fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1361647dd2c3SGao Xiang kvfree(be->compressed_pages); 1362267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 136347e4937aSGao Xiang 13642bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 13654f05687fSGao Xiang page = be->decompressed_pages[i]; 136647e4937aSGao Xiang if (!page) 136747e4937aSGao Xiang continue; 136847e4937aSGao Xiang 13696aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 137047e4937aSGao Xiang 13716aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13724f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 137347e4937aSGao Xiang continue; 137467148551SGao Xiang if (err) 137567148551SGao Xiang z_erofs_page_mark_eio(page); 137647e4937aSGao Xiang z_erofs_onlinepage_endio(page); 137747e4937aSGao Xiang } 137847e4937aSGao Xiang 13794f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 1380647dd2c3SGao Xiang kvfree(be->decompressed_pages); 138147e4937aSGao Xiang 13822bfab9c0SGao Xiang pcl->length = 0; 13832bfab9c0SGao Xiang pcl->partial = true; 1384267f2492SGao Xiang pcl->multibases = false; 138506a304cdSGao Xiang pcl->bvset.nextpage = NULL; 138687ca34a7SGao Xiang pcl->vcnt = 0; 138747e4937aSGao Xiang 138887ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 138947e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 139087ca34a7SGao Xiang mutex_unlock(&pcl->lock); 139147e4937aSGao Xiang return err; 139247e4937aSGao Xiang } 139347e4937aSGao Xiang 13940c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1395eaa9172aSGao Xiang struct page **pagepool) 139647e4937aSGao Xiang { 13974f05687fSGao Xiang struct z_erofs_decompress_backend be = { 13984f05687fSGao Xiang .sb = io->sb, 13994f05687fSGao Xiang .pagepool = pagepool, 1400267f2492SGao Xiang .decompressed_secondary_bvecs = 1401267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 14024f05687fSGao Xiang }; 140347e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 140447e4937aSGao Xiang 1405967c28b2SGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL) { 140647e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 140747e4937aSGao Xiang 14084f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 14094f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 141047e4937aSGao Xiang 14114f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 14124f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 141347e4937aSGao Xiang } 141447e4937aSGao Xiang } 141547e4937aSGao Xiang 14160c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 141747e4937aSGao Xiang { 1418a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1419a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1420eaa9172aSGao Xiang struct page *pagepool = NULL; 142147e4937aSGao Xiang 1422967c28b2SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL); 14230c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 1424eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1425a4b1fab1SGao Xiang kvfree(bgq); 142647e4937aSGao Xiang } 142747e4937aSGao Xiang 14283fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14293fffb589SSandeep Dhavale static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 14303fffb589SSandeep Dhavale { 14313fffb589SSandeep Dhavale z_erofs_decompressqueue_work((struct work_struct *)work); 14323fffb589SSandeep Dhavale } 14333fffb589SSandeep Dhavale #endif 14343fffb589SSandeep Dhavale 14357865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1436cdba5506SGao Xiang int bios) 14377865827cSGao Xiang { 14387865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 14397865827cSGao Xiang 14407865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1441cdba5506SGao Xiang if (io->sync) { 14427865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 144360b30050SHongyu Jin complete(&io->u.done); 14447865827cSGao Xiang return; 14457865827cSGao Xiang } 14467865827cSGao Xiang 14477865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 14487865827cSGao Xiang return; 14493fffb589SSandeep Dhavale /* Use (kthread_)work and sync decompression for atomic contexts only */ 14507865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 14513fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14523fffb589SSandeep Dhavale struct kthread_worker *worker; 14533fffb589SSandeep Dhavale 14543fffb589SSandeep Dhavale rcu_read_lock(); 14553fffb589SSandeep Dhavale worker = rcu_dereference( 14563fffb589SSandeep Dhavale z_erofs_pcpu_workers[raw_smp_processor_id()]); 14573fffb589SSandeep Dhavale if (!worker) { 14583fffb589SSandeep Dhavale INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 14597865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 14603fffb589SSandeep Dhavale } else { 14613fffb589SSandeep Dhavale kthread_queue_work(worker, &io->u.kthread_work); 14623fffb589SSandeep Dhavale } 14633fffb589SSandeep Dhavale rcu_read_unlock(); 14643fffb589SSandeep Dhavale #else 14653fffb589SSandeep Dhavale queue_work(z_erofs_workqueue, &io->u.work); 14663fffb589SSandeep Dhavale #endif 14677865827cSGao Xiang /* enable sync decompression for readahead */ 14687865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 14697865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 14707865827cSGao Xiang return; 14717865827cSGao Xiang } 14727865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 14737865827cSGao Xiang } 14747865827cSGao Xiang 147547e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 147647e4937aSGao Xiang unsigned int nr, 1477eaa9172aSGao Xiang struct page **pagepool, 14789f2731d6SGao Xiang struct address_space *mc) 147947e4937aSGao Xiang { 148047e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 14819f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 148247e4937aSGao Xiang bool tocache = false; 148347e4937aSGao Xiang 148447e4937aSGao Xiang struct address_space *mapping; 148547e4937aSGao Xiang struct page *oldpage, *page; 148647e4937aSGao Xiang int justfound; 148747e4937aSGao Xiang 148847e4937aSGao Xiang repeat: 1489ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 149047e4937aSGao Xiang oldpage = page; 149147e4937aSGao Xiang 149247e4937aSGao Xiang if (!page) 149347e4937aSGao Xiang goto out_allocpage; 149447e4937aSGao Xiang 1495b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1496b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 149747e4937aSGao Xiang 14981825c8d7SGao Xiang /* 14991825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 15001825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 15011825c8d7SGao Xiang */ 15021825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1503ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 15041825c8d7SGao Xiang set_page_private(page, 0); 15051825c8d7SGao Xiang tocache = true; 15061825c8d7SGao Xiang goto out_tocache; 15071825c8d7SGao Xiang } 150847e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 150947e4937aSGao Xiang 151047e4937aSGao Xiang /* 15116aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 151247e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 151347e4937aSGao Xiang */ 151447e4937aSGao Xiang if (mapping && mapping != mc) 151547e4937aSGao Xiang /* ought to be unmanaged pages */ 151647e4937aSGao Xiang goto out; 151747e4937aSGao Xiang 15186aaa7b06SGao Xiang /* directly return for shortlived page as well */ 15196aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 15206aaa7b06SGao Xiang goto out; 15216aaa7b06SGao Xiang 152247e4937aSGao Xiang lock_page(page); 152347e4937aSGao Xiang 152447e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 152547e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 152647e4937aSGao Xiang 152747e4937aSGao Xiang /* the page is still in manage cache */ 152847e4937aSGao Xiang if (page->mapping == mc) { 1529ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 153047e4937aSGao Xiang 153147e4937aSGao Xiang if (!PagePrivate(page)) { 153247e4937aSGao Xiang /* 153347e4937aSGao Xiang * impossible to be !PagePrivate(page) for 153447e4937aSGao Xiang * the current restriction as well if 1535ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 153647e4937aSGao Xiang */ 153747e4937aSGao Xiang DBG_BUGON(!justfound); 153847e4937aSGao Xiang 153947e4937aSGao Xiang justfound = 0; 154047e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 154147e4937aSGao Xiang SetPagePrivate(page); 154247e4937aSGao Xiang } 154347e4937aSGao Xiang 154447e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 154547e4937aSGao Xiang if (PageUptodate(page)) { 154647e4937aSGao Xiang unlock_page(page); 154747e4937aSGao Xiang page = NULL; 154847e4937aSGao Xiang } 154947e4937aSGao Xiang goto out; 155047e4937aSGao Xiang } 155147e4937aSGao Xiang 155247e4937aSGao Xiang /* 155347e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 155447e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 155547e4937aSGao Xiang */ 155647e4937aSGao Xiang DBG_BUGON(page->mapping); 155747e4937aSGao Xiang DBG_BUGON(!justfound); 155847e4937aSGao Xiang 155947e4937aSGao Xiang tocache = true; 156047e4937aSGao Xiang unlock_page(page); 156147e4937aSGao Xiang put_page(page); 156247e4937aSGao Xiang out_allocpage: 15635ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1564ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1565ed722fbcSGao Xiang oldpage, page)) { 1566eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 15675ddcee1fSGao Xiang cond_resched(); 15685ddcee1fSGao Xiang goto repeat; 15695ddcee1fSGao Xiang } 15701825c8d7SGao Xiang out_tocache: 1571bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1572bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1573bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1574bf225074SGao Xiang goto out; 1575a30573b3SGao Xiang } 1576bf225074SGao Xiang attach_page_private(page, pcl); 1577bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1578bf225074SGao Xiang put_page(page); 1579bf225074SGao Xiang 158047e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 158147e4937aSGao Xiang return page; 158247e4937aSGao Xiang } 158347e4937aSGao Xiang 1584cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1585a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 158647e4937aSGao Xiang { 1587a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 158847e4937aSGao Xiang 1589a4b1fab1SGao Xiang if (fg && !*fg) { 1590a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1591a4b1fab1SGao Xiang if (!q) { 1592a4b1fab1SGao Xiang *fg = true; 1593a4b1fab1SGao Xiang goto fg_out; 159447e4937aSGao Xiang } 15953fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 15963fffb589SSandeep Dhavale kthread_init_work(&q->u.kthread_work, 15973fffb589SSandeep Dhavale z_erofs_decompressqueue_kthread_work); 15983fffb589SSandeep Dhavale #else 15990c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 16003fffb589SSandeep Dhavale #endif 1601a4b1fab1SGao Xiang } else { 1602a4b1fab1SGao Xiang fg_out: 1603a4b1fab1SGao Xiang q = fgq; 160460b30050SHongyu Jin init_completion(&fgq->u.done); 1605a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 160667148551SGao Xiang q->eio = false; 1607cdba5506SGao Xiang q->sync = true; 1608a4b1fab1SGao Xiang } 1609a4b1fab1SGao Xiang q->sb = sb; 1610967c28b2SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL; 1611a4b1fab1SGao Xiang return q; 161247e4937aSGao Xiang } 161347e4937aSGao Xiang 161447e4937aSGao Xiang /* define decompression jobqueue types */ 161547e4937aSGao Xiang enum { 161647e4937aSGao Xiang JQ_BYPASS, 161747e4937aSGao Xiang JQ_SUBMIT, 161847e4937aSGao Xiang NR_JOBQUEUES, 161947e4937aSGao Xiang }; 162047e4937aSGao Xiang 162147e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 162247e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 162347e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 162447e4937aSGao Xiang { 162547e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 162647e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 162747e4937aSGao Xiang 1628967c28b2SGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL); 162947e4937aSGao Xiang 163047e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 163147e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 163247e4937aSGao Xiang 163347e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 163447e4937aSGao Xiang } 163547e4937aSGao Xiang 16367865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 16377865827cSGao Xiang { 1638cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 16397865827cSGao Xiang blk_status_t err = bio->bi_status; 16407865827cSGao Xiang struct bio_vec *bvec; 16417865827cSGao Xiang struct bvec_iter_all iter_all; 16427865827cSGao Xiang 16437865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 16447865827cSGao Xiang struct page *page = bvec->bv_page; 16457865827cSGao Xiang 16467865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 16477865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 16487865827cSGao Xiang 16497865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 16507865827cSGao Xiang if (!err) 16517865827cSGao Xiang SetPageUptodate(page); 16527865827cSGao Xiang unlock_page(page); 16537865827cSGao Xiang } 16547865827cSGao Xiang } 165567148551SGao Xiang if (err) 165667148551SGao Xiang q->eio = true; 1657cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 16587865827cSGao Xiang bio_put(bio); 16597865827cSGao Xiang } 16607865827cSGao Xiang 166183a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1662a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1663ef4b4b46SYue Hu bool *force_fg, bool readahead) 166447e4937aSGao Xiang { 166583a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 166683a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 166747e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1668a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 16695c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1670dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 16713f649ab7SKees Cook pgoff_t last_index; 1672dfeab2e9SGao Xiang struct block_device *last_bdev; 16731e4a2955SGao Xiang unsigned int nr_bios = 0; 16741e4a2955SGao Xiang struct bio *bio = NULL; 167582e60d00SJohannes Weiner unsigned long pflags; 167682e60d00SJohannes Weiner int memstall = 0; 167747e4937aSGao Xiang 1678cdba5506SGao Xiang /* 1679cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1680cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1681cdba5506SGao Xiang */ 1682cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1683cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1684cdba5506SGao Xiang 1685a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1686a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 168747e4937aSGao Xiang 168847e4937aSGao Xiang /* by default, all need io submission */ 168947e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 169047e4937aSGao Xiang 169147e4937aSGao Xiang do { 1692dfeab2e9SGao Xiang struct erofs_map_dev mdev; 169347e4937aSGao Xiang struct z_erofs_pcluster *pcl; 16941e4a2955SGao Xiang pgoff_t cur, end; 16951e4a2955SGao Xiang unsigned int i = 0; 16961e4a2955SGao Xiang bool bypass = true; 169747e4937aSGao Xiang 169847e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 169947e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 1700967c28b2SGao Xiang owned_head = READ_ONCE(pcl->next); 170147e4937aSGao Xiang 1702cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1703cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1704cecf864dSYue Hu continue; 1705cecf864dSYue Hu } 1706cecf864dSYue Hu 1707dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1708dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 17093acea5fcSJingbo Xu .m_pa = erofs_pos(sb, pcl->obj.index), 1710dfeab2e9SGao Xiang }; 1711dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1712dfeab2e9SGao Xiang 17133acea5fcSJingbo Xu cur = erofs_blknr(sb, mdev.m_pa); 17149f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 171547e4937aSGao Xiang 17161e4a2955SGao Xiang do { 17171e4a2955SGao Xiang struct page *page; 171847e4937aSGao Xiang 17196ab5eed6SGao Xiang page = pickup_page_for_submission(pcl, i++, 17206ab5eed6SGao Xiang &f->pagepool, mc); 17211e4a2955SGao Xiang if (!page) 17221e4a2955SGao Xiang continue; 172347e4937aSGao Xiang 1724dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1725dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 172647e4937aSGao Xiang submit_bio_retry: 172794e4e153SGao Xiang submit_bio(bio); 172882e60d00SJohannes Weiner if (memstall) { 172982e60d00SJohannes Weiner psi_memstall_leave(&pflags); 173082e60d00SJohannes Weiner memstall = 0; 173182e60d00SJohannes Weiner } 173247e4937aSGao Xiang bio = NULL; 173347e4937aSGao Xiang } 173447e4937aSGao Xiang 173582e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 173699486c51SChristoph Hellwig psi_memstall_enter(&pflags); 173782e60d00SJohannes Weiner memstall = 1; 173882e60d00SJohannes Weiner } 173999486c51SChristoph Hellwig 174047e4937aSGao Xiang if (!bio) { 174107888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 174207888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 17430c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1744dfeab2e9SGao Xiang 1745dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 17461e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 17473acea5fcSJingbo Xu (sb->s_blocksize_bits - 9); 1748cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 1749ef4b4b46SYue Hu if (readahead) 17506ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 175147e4937aSGao Xiang ++nr_bios; 175247e4937aSGao Xiang } 175347e4937aSGao Xiang 17546c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 175547e4937aSGao Xiang goto submit_bio_retry; 175647e4937aSGao Xiang 17571e4a2955SGao Xiang last_index = cur; 17581e4a2955SGao Xiang bypass = false; 17591e4a2955SGao Xiang } while (++cur < end); 176047e4937aSGao Xiang 17611e4a2955SGao Xiang if (!bypass) 176247e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 176347e4937aSGao Xiang else 176447e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 176547e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 176647e4937aSGao Xiang 176799486c51SChristoph Hellwig if (bio) { 176894e4e153SGao Xiang submit_bio(bio); 176982e60d00SJohannes Weiner if (memstall) 177082e60d00SJohannes Weiner psi_memstall_leave(&pflags); 177199486c51SChristoph Hellwig } 177247e4937aSGao Xiang 1773587a67b7SGao Xiang /* 1774587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 17753fffb589SSandeep Dhavale * don't issue decompression but drop it directly instead. 1776587a67b7SGao Xiang */ 1777587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1778587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 17791e4a2955SGao Xiang return; 1780587a67b7SGao Xiang } 1781cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 178247e4937aSGao Xiang } 178347e4937aSGao Xiang 178483a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 17856ab5eed6SGao Xiang bool force_fg, bool ra) 178647e4937aSGao Xiang { 1787a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 178847e4937aSGao Xiang 17895c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 179047e4937aSGao Xiang return; 17916ab5eed6SGao Xiang z_erofs_submit_queue(f, io, &force_fg, ra); 179247e4937aSGao Xiang 17930c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 17946ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], &f->pagepool); 179547e4937aSGao Xiang 179647e4937aSGao Xiang if (!force_fg) 179747e4937aSGao Xiang return; 179847e4937aSGao Xiang 179947e4937aSGao Xiang /* wait until all bios are completed */ 180060b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 180147e4937aSGao Xiang 18020c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 18036ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], &f->pagepool); 180447e4937aSGao Xiang } 180547e4937aSGao Xiang 180638629291SGao Xiang /* 180738629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 180838629291SGao Xiang * approximate readmore strategies as a start. 180938629291SGao Xiang */ 181038629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 18116ab5eed6SGao Xiang struct readahead_control *rac, bool backmost) 181238629291SGao Xiang { 181338629291SGao Xiang struct inode *inode = f->inode; 181438629291SGao Xiang struct erofs_map_blocks *map = &f->map; 1815796e9149SYue Hu erofs_off_t cur, end, headoffset = f->headoffset; 181638629291SGao Xiang int err; 181738629291SGao Xiang 181838629291SGao Xiang if (backmost) { 1819796e9149SYue Hu if (rac) 1820796e9149SYue Hu end = headoffset + readahead_length(rac) - 1; 1821796e9149SYue Hu else 1822796e9149SYue Hu end = headoffset + PAGE_SIZE - 1; 182338629291SGao Xiang map->m_la = end; 1824622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1825622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 182638629291SGao Xiang if (err) 182738629291SGao Xiang return; 182838629291SGao Xiang 1829796e9149SYue Hu /* expand ra for the trailing edge if readahead */ 183038629291SGao Xiang if (rac) { 183138629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1832796e9149SYue Hu readahead_expand(rac, headoffset, cur - headoffset); 183338629291SGao Xiang return; 183438629291SGao Xiang } 183538629291SGao Xiang end = round_up(end, PAGE_SIZE); 183638629291SGao Xiang } else { 183738629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 183838629291SGao Xiang 183938629291SGao Xiang if (!map->m_llen) 184038629291SGao Xiang return; 184138629291SGao Xiang } 184238629291SGao Xiang 184338629291SGao Xiang cur = map->m_la + map->m_llen - 1; 184438629291SGao Xiang while (cur >= end) { 184538629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 184638629291SGao Xiang struct page *page; 184738629291SGao Xiang 184838629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1849aa793b46SGao Xiang if (page) { 185038629291SGao Xiang if (PageUptodate(page)) { 185138629291SGao Xiang unlock_page(page); 1852aa793b46SGao Xiang } else { 18536ab5eed6SGao Xiang err = z_erofs_do_read_page(f, page); 185438629291SGao Xiang if (err) 185538629291SGao Xiang erofs_err(inode->i_sb, 185638629291SGao Xiang "readmore error at page %lu @ nid %llu", 185738629291SGao Xiang index, EROFS_I(inode)->nid); 1858aa793b46SGao Xiang } 185938629291SGao Xiang put_page(page); 1860aa793b46SGao Xiang } 1861aa793b46SGao Xiang 186238629291SGao Xiang if (cur < PAGE_SIZE) 186338629291SGao Xiang break; 186438629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 186538629291SGao Xiang } 186638629291SGao Xiang } 186738629291SGao Xiang 1868a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 186947e4937aSGao Xiang { 1870a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 187147e4937aSGao Xiang struct inode *const inode = page->mapping->host; 187240452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 187347e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 187447e4937aSGao Xiang int err; 187547e4937aSGao Xiang 187647e4937aSGao Xiang trace_erofs_readpage(page, false); 187747e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 187847e4937aSGao Xiang 18796ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, true); 18806ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 18816ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, false); 18825c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 188347e4937aSGao Xiang 188447e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 18856ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, 0), false); 188647e4937aSGao Xiang 188747e4937aSGao Xiang if (err) 18884f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 188947e4937aSGao Xiang 189009c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 18916ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 189247e4937aSGao Xiang return err; 189347e4937aSGao Xiang } 189447e4937aSGao Xiang 18950615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 189647e4937aSGao Xiang { 18970615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 189847e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 189947e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 19006ab5eed6SGao Xiang struct page *head = NULL, *page; 190138629291SGao Xiang unsigned int nr_pages; 190247e4937aSGao Xiang 19030615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 190447e4937aSGao Xiang 19056ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, true); 190638629291SGao Xiang nr_pages = readahead_count(rac); 190738629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 190838629291SGao Xiang 19090615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 191047e4937aSGao Xiang set_page_private(page, (unsigned long)head); 191147e4937aSGao Xiang head = page; 191247e4937aSGao Xiang } 191347e4937aSGao Xiang 191447e4937aSGao Xiang while (head) { 191547e4937aSGao Xiang struct page *page = head; 191647e4937aSGao Xiang int err; 191747e4937aSGao Xiang 191847e4937aSGao Xiang /* traversal in reverse order */ 191947e4937aSGao Xiang head = (void *)page_private(page); 192047e4937aSGao Xiang 19216ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 1922a5876e24SGao Xiang if (err) 19234f761fa2SGao Xiang erofs_err(inode->i_sb, 19244f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 19254f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 192647e4937aSGao Xiang put_page(page); 192747e4937aSGao Xiang } 19286ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, false); 19295c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 193047e4937aSGao Xiang 19316ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, nr_pages), true); 193209c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 19336ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 193447e4937aSGao Xiang } 193547e4937aSGao Xiang 19360c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1937a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 19380615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 193947e4937aSGao Xiang }; 1940