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 94a9a94d93SGao Xiang /* let's avoid the valid 32-bit kernel addresses */ 95a9a94d93SGao Xiang 96967c28b2SGao Xiang /* the end of a chain of pclusters */ 97a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE) 98a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_NIL (NULL) 99a9a94d93SGao Xiang 100a9a94d93SGao Xiang struct z_erofs_decompressqueue { 101a9a94d93SGao Xiang struct super_block *sb; 102a9a94d93SGao Xiang atomic_t pending_bios; 103a9a94d93SGao Xiang z_erofs_next_pcluster_t head; 104a9a94d93SGao Xiang 105a9a94d93SGao Xiang union { 106a9a94d93SGao Xiang struct completion done; 107a9a94d93SGao Xiang struct work_struct work; 1083fffb589SSandeep Dhavale struct kthread_work kthread_work; 109a9a94d93SGao Xiang } u; 110a9a94d93SGao Xiang bool eio, sync; 111a9a94d93SGao Xiang }; 112a9a94d93SGao Xiang 113a9a94d93SGao Xiang static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 114a9a94d93SGao Xiang { 115a9a94d93SGao Xiang return !pcl->obj.index; 116a9a94d93SGao Xiang } 117a9a94d93SGao Xiang 118a9a94d93SGao Xiang static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 119a9a94d93SGao Xiang { 120a9a94d93SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) 121a9a94d93SGao Xiang return 1; 122a9a94d93SGao Xiang return pcl->pclusterpages; 123a9a94d93SGao Xiang } 124a9a94d93SGao Xiang 125a9a94d93SGao Xiang /* 126a9a94d93SGao Xiang * bit 30: I/O error occurred on this page 127a9a94d93SGao Xiang * bit 0 - 29: remaining parts to complete this page 128a9a94d93SGao Xiang */ 129a9a94d93SGao Xiang #define Z_EROFS_PAGE_EIO (1 << 30) 130a9a94d93SGao Xiang 131a9a94d93SGao Xiang static inline void z_erofs_onlinepage_init(struct page *page) 132a9a94d93SGao Xiang { 133a9a94d93SGao Xiang union { 134a9a94d93SGao Xiang atomic_t o; 135a9a94d93SGao Xiang unsigned long v; 136a9a94d93SGao Xiang } u = { .o = ATOMIC_INIT(1) }; 137a9a94d93SGao Xiang 138a9a94d93SGao Xiang set_page_private(page, u.v); 139a9a94d93SGao Xiang smp_wmb(); 140a9a94d93SGao Xiang SetPagePrivate(page); 141a9a94d93SGao Xiang } 142a9a94d93SGao Xiang 143a9a94d93SGao Xiang static inline void z_erofs_onlinepage_split(struct page *page) 144a9a94d93SGao Xiang { 145a9a94d93SGao Xiang atomic_inc((atomic_t *)&page->private); 146a9a94d93SGao Xiang } 147a9a94d93SGao Xiang 148a9a94d93SGao Xiang static inline void z_erofs_page_mark_eio(struct page *page) 149a9a94d93SGao Xiang { 150a9a94d93SGao Xiang int orig; 151a9a94d93SGao Xiang 152a9a94d93SGao Xiang do { 153a9a94d93SGao Xiang orig = atomic_read((atomic_t *)&page->private); 154a9a94d93SGao Xiang } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 155a9a94d93SGao Xiang orig | Z_EROFS_PAGE_EIO) != orig); 156a9a94d93SGao Xiang } 157a9a94d93SGao Xiang 158a9a94d93SGao Xiang static inline void z_erofs_onlinepage_endio(struct page *page) 159a9a94d93SGao Xiang { 160a9a94d93SGao Xiang unsigned int v; 161a9a94d93SGao Xiang 162a9a94d93SGao Xiang DBG_BUGON(!PagePrivate(page)); 163a9a94d93SGao Xiang v = atomic_dec_return((atomic_t *)&page->private); 164a9a94d93SGao Xiang if (!(v & ~Z_EROFS_PAGE_EIO)) { 165a9a94d93SGao Xiang set_page_private(page, 0); 166a9a94d93SGao Xiang ClearPagePrivate(page); 167a9a94d93SGao Xiang if (!(v & Z_EROFS_PAGE_EIO)) 168a9a94d93SGao Xiang SetPageUptodate(page); 169a9a94d93SGao Xiang unlock_page(page); 170a9a94d93SGao Xiang } 171a9a94d93SGao Xiang } 172a9a94d93SGao Xiang 173a9a94d93SGao Xiang #define Z_EROFS_ONSTACK_PAGES 32 174a9a94d93SGao Xiang 17547e4937aSGao Xiang /* 1769f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 1779f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 1789f6cc76eSGao Xiang */ 1799f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 1809f6cc76eSGao Xiang struct kmem_cache *slab; 1819f6cc76eSGao Xiang unsigned int maxpages; 1829f6cc76eSGao Xiang char name[48]; 1839f6cc76eSGao Xiang }; 1849f6cc76eSGao Xiang 1859f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 1869f6cc76eSGao Xiang 1879f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 1889f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 1899f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 1909f6cc76eSGao Xiang }; 1919f6cc76eSGao Xiang 19206a304cdSGao Xiang struct z_erofs_bvec_iter { 19306a304cdSGao Xiang struct page *bvpage; 19406a304cdSGao Xiang struct z_erofs_bvset *bvset; 19506a304cdSGao Xiang unsigned int nr, cur; 19606a304cdSGao Xiang }; 19706a304cdSGao Xiang 19806a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 19906a304cdSGao Xiang { 20006a304cdSGao Xiang if (iter->bvpage) 20106a304cdSGao Xiang kunmap_local(iter->bvset); 20206a304cdSGao Xiang return iter->bvpage; 20306a304cdSGao Xiang } 20406a304cdSGao Xiang 20506a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 20606a304cdSGao Xiang { 20706a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 20806a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 20906a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 21006a304cdSGao Xiang struct page *oldpage; 21106a304cdSGao Xiang 21206a304cdSGao Xiang DBG_BUGON(!nextpage); 21306a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 21406a304cdSGao Xiang iter->bvpage = nextpage; 21506a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 21606a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 21706a304cdSGao Xiang iter->cur = 0; 21806a304cdSGao Xiang return oldpage; 21906a304cdSGao Xiang } 22006a304cdSGao Xiang 22106a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 22206a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 22306a304cdSGao Xiang unsigned int bootstrap_nr, 22406a304cdSGao Xiang unsigned int cur) 22506a304cdSGao Xiang { 22606a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 22706a304cdSGao Xiang .nr = bootstrap_nr, 22806a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 22906a304cdSGao Xiang }; 23006a304cdSGao Xiang 23106a304cdSGao Xiang while (cur > iter->nr) { 23206a304cdSGao Xiang cur -= iter->nr; 23306a304cdSGao Xiang z_erofs_bvset_flip(iter); 23406a304cdSGao Xiang } 23506a304cdSGao Xiang iter->cur = cur; 23606a304cdSGao Xiang } 23706a304cdSGao Xiang 23806a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 23906a304cdSGao Xiang struct z_erofs_bvec *bvec, 2406ab5eed6SGao Xiang struct page **candidate_bvpage, 2416ab5eed6SGao Xiang struct page **pagepool) 24206a304cdSGao Xiang { 24305b63d2bSGao Xiang if (iter->cur >= iter->nr) { 24405b63d2bSGao Xiang struct page *nextpage = *candidate_bvpage; 24506a304cdSGao Xiang 24605b63d2bSGao Xiang if (!nextpage) { 2476ab5eed6SGao Xiang nextpage = erofs_allocpage(pagepool, GFP_NOFS); 24805b63d2bSGao Xiang if (!nextpage) 24905b63d2bSGao Xiang return -ENOMEM; 25005b63d2bSGao Xiang set_page_private(nextpage, Z_EROFS_SHORTLIVED_PAGE); 25105b63d2bSGao Xiang } 25206a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 25305b63d2bSGao Xiang iter->bvset->nextpage = nextpage; 25406a304cdSGao Xiang z_erofs_bvset_flip(iter); 25506a304cdSGao Xiang 25606a304cdSGao Xiang iter->bvset->nextpage = NULL; 25706a304cdSGao Xiang *candidate_bvpage = NULL; 25806a304cdSGao Xiang } 25906a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 26006a304cdSGao Xiang return 0; 26106a304cdSGao Xiang } 26206a304cdSGao Xiang 26306a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 26406a304cdSGao Xiang struct z_erofs_bvec *bvec, 26506a304cdSGao Xiang struct page **old_bvpage) 26606a304cdSGao Xiang { 26706a304cdSGao Xiang if (iter->cur == iter->nr) 26806a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 26906a304cdSGao Xiang else 27006a304cdSGao Xiang *old_bvpage = NULL; 27106a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 27206a304cdSGao Xiang } 27306a304cdSGao Xiang 2749f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 2759f6cc76eSGao Xiang { 2769f6cc76eSGao Xiang int i; 2779f6cc76eSGao Xiang 2789f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 2799f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 2809f6cc76eSGao Xiang continue; 2819f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 2829f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 2839f6cc76eSGao Xiang } 2849f6cc76eSGao Xiang } 2859f6cc76eSGao Xiang 2869f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 2879f6cc76eSGao Xiang { 2889f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 2899f6cc76eSGao Xiang struct z_erofs_pcluster *a; 2909f6cc76eSGao Xiang unsigned int size; 2919f6cc76eSGao Xiang 2929f6cc76eSGao Xiang for (pcs = pcluster_pool; 2939f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 294ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 2959f6cc76eSGao Xiang 2969f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 2979f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 2989f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 2999f6cc76eSGao Xiang if (pcs->slab) 3009f6cc76eSGao Xiang continue; 3019f6cc76eSGao Xiang 3029f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 3039f6cc76eSGao Xiang return -ENOMEM; 3049f6cc76eSGao Xiang } 3059f6cc76eSGao Xiang return 0; 3069f6cc76eSGao Xiang } 3079f6cc76eSGao Xiang 3089f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 3099f6cc76eSGao Xiang { 3109f6cc76eSGao Xiang int i; 3119f6cc76eSGao Xiang 3129f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3139f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3149f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 3159f6cc76eSGao Xiang 3169f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 3179f6cc76eSGao Xiang continue; 3189f6cc76eSGao Xiang 3199f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 3209f6cc76eSGao Xiang if (!pcl) 3219f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 3229f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 3239f6cc76eSGao Xiang return pcl; 3249f6cc76eSGao Xiang } 3259f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 3269f6cc76eSGao Xiang } 3279f6cc76eSGao Xiang 3289f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 3299f6cc76eSGao Xiang { 330cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 3319f6cc76eSGao Xiang int i; 3329f6cc76eSGao Xiang 3339f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3349f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3359f6cc76eSGao Xiang 336cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 3379f6cc76eSGao Xiang continue; 3389f6cc76eSGao Xiang 3399f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 3409f6cc76eSGao Xiang return; 3419f6cc76eSGao Xiang } 3429f6cc76eSGao Xiang DBG_BUGON(1); 3439f6cc76eSGao Xiang } 3449f6cc76eSGao Xiang 34547e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 34647e4937aSGao Xiang 3473fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 3483fffb589SSandeep Dhavale static struct kthread_worker __rcu **z_erofs_pcpu_workers; 3493fffb589SSandeep Dhavale 3503fffb589SSandeep Dhavale static void erofs_destroy_percpu_workers(void) 35147e4937aSGao Xiang { 3523fffb589SSandeep Dhavale struct kthread_worker *worker; 3533fffb589SSandeep Dhavale unsigned int cpu; 3543fffb589SSandeep Dhavale 3553fffb589SSandeep Dhavale for_each_possible_cpu(cpu) { 3563fffb589SSandeep Dhavale worker = rcu_dereference_protected( 3573fffb589SSandeep Dhavale z_erofs_pcpu_workers[cpu], 1); 3583fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 3593fffb589SSandeep Dhavale if (worker) 3603fffb589SSandeep Dhavale kthread_destroy_worker(worker); 3613fffb589SSandeep Dhavale } 3623fffb589SSandeep Dhavale kfree(z_erofs_pcpu_workers); 36347e4937aSGao Xiang } 36447e4937aSGao Xiang 3653fffb589SSandeep Dhavale static struct kthread_worker *erofs_init_percpu_worker(int cpu) 36647e4937aSGao Xiang { 3673fffb589SSandeep Dhavale struct kthread_worker *worker = 3683fffb589SSandeep Dhavale kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 36947e4937aSGao Xiang 3703fffb589SSandeep Dhavale if (IS_ERR(worker)) 3713fffb589SSandeep Dhavale return worker; 3723fffb589SSandeep Dhavale if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 3733fffb589SSandeep Dhavale sched_set_fifo_low(worker->task); 3743fffb589SSandeep Dhavale return worker; 3753fffb589SSandeep Dhavale } 3763fffb589SSandeep Dhavale 3773fffb589SSandeep Dhavale static int erofs_init_percpu_workers(void) 3783fffb589SSandeep Dhavale { 3793fffb589SSandeep Dhavale struct kthread_worker *worker; 3803fffb589SSandeep Dhavale unsigned int cpu; 3813fffb589SSandeep Dhavale 3823fffb589SSandeep Dhavale z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 3833fffb589SSandeep Dhavale sizeof(struct kthread_worker *), GFP_ATOMIC); 3843fffb589SSandeep Dhavale if (!z_erofs_pcpu_workers) 3853fffb589SSandeep Dhavale return -ENOMEM; 3863fffb589SSandeep Dhavale 3873fffb589SSandeep Dhavale for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 3883fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 3893fffb589SSandeep Dhavale if (!IS_ERR(worker)) 3903fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 3913fffb589SSandeep Dhavale } 3923fffb589SSandeep Dhavale return 0; 3933fffb589SSandeep Dhavale } 3943fffb589SSandeep Dhavale #else 3953fffb589SSandeep Dhavale static inline void erofs_destroy_percpu_workers(void) {} 3963fffb589SSandeep Dhavale static inline int erofs_init_percpu_workers(void) { return 0; } 3973fffb589SSandeep Dhavale #endif 3983fffb589SSandeep Dhavale 3993fffb589SSandeep Dhavale #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 4003fffb589SSandeep Dhavale static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 4013fffb589SSandeep Dhavale static enum cpuhp_state erofs_cpuhp_state; 4023fffb589SSandeep Dhavale 4033fffb589SSandeep Dhavale static int erofs_cpu_online(unsigned int cpu) 4043fffb589SSandeep Dhavale { 4053fffb589SSandeep Dhavale struct kthread_worker *worker, *old; 4063fffb589SSandeep Dhavale 4073fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 4083fffb589SSandeep Dhavale if (IS_ERR(worker)) 4093fffb589SSandeep Dhavale return PTR_ERR(worker); 4103fffb589SSandeep Dhavale 4113fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4123fffb589SSandeep Dhavale old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4133fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4143fffb589SSandeep Dhavale if (!old) 4153fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 4163fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4173fffb589SSandeep Dhavale if (old) 4183fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4193fffb589SSandeep Dhavale return 0; 4203fffb589SSandeep Dhavale } 4213fffb589SSandeep Dhavale 4223fffb589SSandeep Dhavale static int erofs_cpu_offline(unsigned int cpu) 4233fffb589SSandeep Dhavale { 4243fffb589SSandeep Dhavale struct kthread_worker *worker; 4253fffb589SSandeep Dhavale 4263fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4273fffb589SSandeep Dhavale worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4283fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4293fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 4303fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4313fffb589SSandeep Dhavale 4323fffb589SSandeep Dhavale synchronize_rcu(); 4333fffb589SSandeep Dhavale if (worker) 4343fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4353fffb589SSandeep Dhavale return 0; 4363fffb589SSandeep Dhavale } 4373fffb589SSandeep Dhavale 4383fffb589SSandeep Dhavale static int erofs_cpu_hotplug_init(void) 4393fffb589SSandeep Dhavale { 4403fffb589SSandeep Dhavale int state; 4413fffb589SSandeep Dhavale 4423fffb589SSandeep Dhavale state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 4433fffb589SSandeep Dhavale "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 4443fffb589SSandeep Dhavale if (state < 0) 4453fffb589SSandeep Dhavale return state; 4463fffb589SSandeep Dhavale 4473fffb589SSandeep Dhavale erofs_cpuhp_state = state; 4483fffb589SSandeep Dhavale return 0; 4493fffb589SSandeep Dhavale } 4503fffb589SSandeep Dhavale 4513fffb589SSandeep Dhavale static void erofs_cpu_hotplug_destroy(void) 4523fffb589SSandeep Dhavale { 4533fffb589SSandeep Dhavale if (erofs_cpuhp_state) 4543fffb589SSandeep Dhavale cpuhp_remove_state_nocalls(erofs_cpuhp_state); 4553fffb589SSandeep Dhavale } 4563fffb589SSandeep Dhavale #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 4573fffb589SSandeep Dhavale static inline int erofs_cpu_hotplug_init(void) { return 0; } 4583fffb589SSandeep Dhavale static inline void erofs_cpu_hotplug_destroy(void) {} 4593fffb589SSandeep Dhavale #endif 4603fffb589SSandeep Dhavale 4613fffb589SSandeep Dhavale void z_erofs_exit_zip_subsystem(void) 4623fffb589SSandeep Dhavale { 4633fffb589SSandeep Dhavale erofs_cpu_hotplug_destroy(); 4643fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4653fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4663fffb589SSandeep Dhavale z_erofs_destroy_pcluster_pool(); 46747e4937aSGao Xiang } 46847e4937aSGao Xiang 46947e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 47047e4937aSGao Xiang { 4719f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 47247e4937aSGao Xiang 4739f6cc76eSGao Xiang if (err) 4743fffb589SSandeep Dhavale goto out_error_pcluster_pool; 4753fffb589SSandeep Dhavale 4763fffb589SSandeep Dhavale z_erofs_workqueue = alloc_workqueue("erofs_worker", 4773fffb589SSandeep Dhavale WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 4788d1b80a7SDan Carpenter if (!z_erofs_workqueue) { 4798d1b80a7SDan Carpenter err = -ENOMEM; 4803fffb589SSandeep Dhavale goto out_error_workqueue_init; 4818d1b80a7SDan Carpenter } 4823fffb589SSandeep Dhavale 4833fffb589SSandeep Dhavale err = erofs_init_percpu_workers(); 4849f6cc76eSGao Xiang if (err) 4853fffb589SSandeep Dhavale goto out_error_pcpu_worker; 4863fffb589SSandeep Dhavale 4873fffb589SSandeep Dhavale err = erofs_cpu_hotplug_init(); 4883fffb589SSandeep Dhavale if (err < 0) 4893fffb589SSandeep Dhavale goto out_error_cpuhp_init; 4903fffb589SSandeep Dhavale return err; 4913fffb589SSandeep Dhavale 4923fffb589SSandeep Dhavale out_error_cpuhp_init: 4933fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4943fffb589SSandeep Dhavale out_error_pcpu_worker: 4953fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4963fffb589SSandeep Dhavale out_error_workqueue_init: 4979f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 4983fffb589SSandeep Dhavale out_error_pcluster_pool: 4999f6cc76eSGao Xiang return err; 50047e4937aSGao Xiang } 50147e4937aSGao Xiang 502db166fc2SGao Xiang enum z_erofs_pclustermode { 503db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 50447e4937aSGao Xiang /* 505db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 5060b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 5070b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 508387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 5090b964600SGao Xiang */ 510db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 51147e4937aSGao Xiang /* 51247e4937aSGao Xiang * The current collection has been linked with the owned chain, and 51347e4937aSGao Xiang * could also be linked with the remaining collections, which means 51447e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 51547e4937aSGao Xiang * the current collection can safely use the whole page (since 51647e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 51747e4937aSGao Xiang * illustrated below: 51847e4937aSGao Xiang * ________________________________________________________________ 51947e4937aSGao Xiang * | tail (partial) page | head (partial) page | 52047e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 521967c28b2SGao Xiang * | | | 522967c28b2SGao Xiang * |__PCLUSTER_FOLLOWED___|___________PCLUSTER_FOLLOWED____________| 52347e4937aSGao Xiang * 52447e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 52547e4937aSGao Xiang */ 526db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 52747e4937aSGao Xiang }; 52847e4937aSGao Xiang 5295c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 5305c6dcc57SGao Xiang struct inode *const inode; 5315c6dcc57SGao Xiang struct erofs_map_blocks map; 53206a304cdSGao Xiang struct z_erofs_bvec_iter biter; 53347e4937aSGao Xiang 5346ab5eed6SGao Xiang struct page *pagepool; 53506a304cdSGao Xiang struct page *candidate_bvpage; 536967c28b2SGao Xiang struct z_erofs_pcluster *pcl; 53747e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 538db166fc2SGao Xiang enum z_erofs_pclustermode mode; 53947e4937aSGao Xiang 54047e4937aSGao Xiang /* used for applying cache strategy on the fly */ 54147e4937aSGao Xiang bool backmost; 54247e4937aSGao Xiang erofs_off_t headoffset; 543ed722fbcSGao Xiang 544ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 545ed722fbcSGao Xiang unsigned int icur; 54647e4937aSGao Xiang }; 54747e4937aSGao Xiang 54847e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 5495c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 550db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 55147e4937aSGao Xiang 5521282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 5531282dea3SGao Xiang { 5541282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 5551282dea3SGao Xiang 5561282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 5571282dea3SGao Xiang return false; 5581282dea3SGao Xiang 5591282dea3SGao Xiang if (fe->backmost) 5601282dea3SGao Xiang return true; 5611282dea3SGao Xiang 5621282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 5631282dea3SGao Xiang fe->map.m_la < fe->headoffset) 5641282dea3SGao Xiang return true; 5651282dea3SGao Xiang 5661282dea3SGao Xiang return false; 5671282dea3SGao Xiang } 5681282dea3SGao Xiang 5696ab5eed6SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe) 57047e4937aSGao Xiang { 5716f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 5725c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 5731282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 57447e4937aSGao Xiang bool standalone = true; 5756f39d1e1SGao Xiang /* 5766f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 5776f39d1e1SGao Xiang * can be used if low memory otherwise. 5786f39d1e1SGao Xiang */ 5791825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 5801825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 581ed722fbcSGao Xiang unsigned int i; 58247e4937aSGao Xiang 583db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 58447e4937aSGao Xiang return; 58547e4937aSGao Xiang 586ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 58747e4937aSGao Xiang struct page *page; 588b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 5891825c8d7SGao Xiang struct page *newpage = NULL; 59047e4937aSGao Xiang 59147e4937aSGao Xiang /* the compressed page was loaded before */ 592ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 59347e4937aSGao Xiang continue; 59447e4937aSGao Xiang 595ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 59647e4937aSGao Xiang 59747e4937aSGao Xiang if (page) { 598b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 5990b964600SGao Xiang } else { 6000b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 6010b964600SGao Xiang standalone = false; 6021282dea3SGao Xiang if (!shouldalloc) 6031282dea3SGao Xiang continue; 6041282dea3SGao Xiang 6051282dea3SGao Xiang /* 6061282dea3SGao Xiang * try to use cached I/O if page allocation 6071282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 6081282dea3SGao Xiang * to avoid any direct reclaim. 6091282dea3SGao Xiang */ 6106ab5eed6SGao Xiang newpage = erofs_allocpage(&fe->pagepool, gfp); 6111825c8d7SGao Xiang if (!newpage) 61247e4937aSGao Xiang continue; 6131282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 614b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 61547e4937aSGao Xiang } 61647e4937aSGao Xiang 617b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 61847e4937aSGao Xiang continue; 61947e4937aSGao Xiang 620eaa9172aSGao Xiang if (page) 62147e4937aSGao Xiang put_page(page); 622eaa9172aSGao Xiang else if (newpage) 6236ab5eed6SGao Xiang erofs_pagepool_add(&fe->pagepool, newpage); 62447e4937aSGao Xiang } 62547e4937aSGao Xiang 6260b964600SGao Xiang /* 6270b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 6280b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 6290b964600SGao Xiang */ 6300b964600SGao Xiang if (standalone) 631db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 63247e4937aSGao Xiang } 63347e4937aSGao Xiang 63447e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 63547e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 63647e4937aSGao Xiang struct erofs_workgroup *grp) 63747e4937aSGao Xiang { 63847e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 63947e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 64047e4937aSGao Xiang int i; 64147e4937aSGao Xiang 642cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 64347e4937aSGao Xiang /* 64447e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 64547e4937aSGao Xiang * therefore no need to worry about available decompression users. 64647e4937aSGao Xiang */ 6479f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 648ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 64947e4937aSGao Xiang 65047e4937aSGao Xiang if (!page) 65147e4937aSGao Xiang continue; 65247e4937aSGao Xiang 65347e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 65447e4937aSGao Xiang if (!trylock_page(page)) 65547e4937aSGao Xiang return -EBUSY; 65647e4937aSGao Xiang 657f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 65847e4937aSGao Xiang continue; 65947e4937aSGao Xiang 66047e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 661ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6626aaa7b06SGao Xiang detach_page_private(page); 66347e4937aSGao Xiang unlock_page(page); 66447e4937aSGao Xiang } 66547e4937aSGao Xiang return 0; 66647e4937aSGao Xiang } 66747e4937aSGao Xiang 668*7b4e372cSGao Xiang static bool z_erofs_cache_release_folio(struct folio *folio, gfp_t gfp) 66947e4937aSGao Xiang { 670*7b4e372cSGao Xiang struct z_erofs_pcluster *pcl = folio_get_private(folio); 671*7b4e372cSGao Xiang bool ret; 672*7b4e372cSGao Xiang int i; 673*7b4e372cSGao Xiang 674*7b4e372cSGao Xiang if (!folio_test_private(folio)) 675*7b4e372cSGao Xiang return true; 67647e4937aSGao Xiang 677ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 678*7b4e372cSGao Xiang return false; 67947e4937aSGao Xiang 680*7b4e372cSGao Xiang ret = false; 681cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 6829f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 683*7b4e372cSGao Xiang if (pcl->compressed_bvecs[i].page == &folio->page) { 684ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 685*7b4e372cSGao Xiang ret = true; 68647e4937aSGao Xiang break; 68747e4937aSGao Xiang } 68847e4937aSGao Xiang } 68947e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 690*7b4e372cSGao Xiang 6916aaa7b06SGao Xiang if (ret) 692*7b4e372cSGao Xiang folio_detach_private(folio); 69347e4937aSGao Xiang return ret; 69447e4937aSGao Xiang } 69547e4937aSGao Xiang 696*7b4e372cSGao Xiang /* 697*7b4e372cSGao Xiang * It will be called only on inode eviction. In case that there are still some 698*7b4e372cSGao Xiang * decompression requests in progress, wait with rescheduling for a bit here. 699*7b4e372cSGao Xiang * An extra lock could be introduced instead but it seems unnecessary. 700*7b4e372cSGao Xiang */ 701*7b4e372cSGao Xiang static void z_erofs_cache_invalidate_folio(struct folio *folio, 702*7b4e372cSGao Xiang size_t offset, size_t length) 703*7b4e372cSGao Xiang { 704*7b4e372cSGao Xiang const size_t stop = length + offset; 705*7b4e372cSGao Xiang 706*7b4e372cSGao Xiang /* Check for potential overflow in debug mode */ 707*7b4e372cSGao Xiang DBG_BUGON(stop > folio_size(folio) || stop < length); 708*7b4e372cSGao Xiang 709*7b4e372cSGao Xiang if (offset == 0 && stop == folio_size(folio)) 710*7b4e372cSGao Xiang while (!z_erofs_cache_release_folio(folio, GFP_NOFS)) 711*7b4e372cSGao Xiang cond_resched(); 712*7b4e372cSGao Xiang } 713*7b4e372cSGao Xiang 714*7b4e372cSGao Xiang static const struct address_space_operations z_erofs_cache_aops = { 715*7b4e372cSGao Xiang .release_folio = z_erofs_cache_release_folio, 716*7b4e372cSGao Xiang .invalidate_folio = z_erofs_cache_invalidate_folio, 717*7b4e372cSGao Xiang }; 718*7b4e372cSGao Xiang 719*7b4e372cSGao Xiang int erofs_init_managed_cache(struct super_block *sb) 720*7b4e372cSGao Xiang { 721*7b4e372cSGao Xiang struct inode *const inode = new_inode(sb); 722*7b4e372cSGao Xiang 723*7b4e372cSGao Xiang if (!inode) 724*7b4e372cSGao Xiang return -ENOMEM; 725*7b4e372cSGao Xiang 726*7b4e372cSGao Xiang set_nlink(inode, 1); 727*7b4e372cSGao Xiang inode->i_size = OFFSET_MAX; 728*7b4e372cSGao Xiang inode->i_mapping->a_ops = &z_erofs_cache_aops; 729*7b4e372cSGao Xiang mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS); 730*7b4e372cSGao Xiang EROFS_SB(sb)->managed_cache = inode; 731*7b4e372cSGao Xiang return 0; 732*7b4e372cSGao Xiang } 733*7b4e372cSGao Xiang 7345c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 735ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 73647e4937aSGao Xiang { 7375c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 73847e4937aSGao Xiang 739ed722fbcSGao Xiang while (fe->icur > 0) { 740ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 741ed722fbcSGao Xiang NULL, bvec->page)) { 742ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 74347e4937aSGao Xiang return true; 744ed722fbcSGao Xiang } 745ed722fbcSGao Xiang } 74647e4937aSGao Xiang return false; 74747e4937aSGao Xiang } 74847e4937aSGao Xiang 74987ca34a7SGao Xiang /* callers must be with pcluster lock held */ 7505c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 7515b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 75247e4937aSGao Xiang { 75347e4937aSGao Xiang int ret; 75447e4937aSGao Xiang 755db166fc2SGao Xiang if (exclusive) { 75606a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 757ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 75847e4937aSGao Xiang return 0; 75906a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 760db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 76106a304cdSGao Xiang !fe->candidate_bvpage) 76206a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 76306a304cdSGao Xiang } 7646ab5eed6SGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage, 7656ab5eed6SGao Xiang &fe->pagepool); 76606a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 76706a304cdSGao Xiang return ret; 76847e4937aSGao Xiang } 76947e4937aSGao Xiang 7705c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 77147e4937aSGao Xiang { 7725c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 7735c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 77447e4937aSGao Xiang 775473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 776473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 777473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 77847e4937aSGao Xiang *owned_head = &pcl->next; 779473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 780db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 781473e15b0SGao Xiang return; 782473e15b0SGao Xiang } 783473e15b0SGao Xiang 784967c28b2SGao Xiang /* type 2, it belongs to an ongoing chain */ 785db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 78647e4937aSGao Xiang } 78747e4937aSGao Xiang 78883a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 78947e4937aSGao Xiang { 79083a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 791cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 79247e4937aSGao Xiang struct z_erofs_pcluster *pcl; 79364094a04SGao Xiang struct erofs_workgroup *grp; 79447e4937aSGao Xiang int err; 79547e4937aSGao Xiang 796c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 797c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 7988f899262SGao Xiang DBG_BUGON(1); 7998f899262SGao Xiang return -EFSCORRUPTED; 8008f899262SGao Xiang } 8018f899262SGao Xiang 8029f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 803cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 804cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 8059f6cc76eSGao Xiang if (IS_ERR(pcl)) 8069f6cc76eSGao Xiang return PTR_ERR(pcl); 80747e4937aSGao Xiang 80864094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 8098f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 8102bfab9c0SGao Xiang pcl->length = 0; 8112bfab9c0SGao Xiang pcl->partial = true; 81247e4937aSGao Xiang 81347e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 8145c6dcc57SGao Xiang pcl->next = fe->owned_head; 81587ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 816db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 81747e4937aSGao Xiang 81847e4937aSGao Xiang /* 81947e4937aSGao Xiang * lock all primary followed works before visible to others 82047e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 82147e4937aSGao Xiang */ 82287ca34a7SGao Xiang mutex_init(&pcl->lock); 82387ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 82447e4937aSGao Xiang 825cecf864dSYue Hu if (ztailpacking) { 826cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 8273acea5fcSJingbo Xu pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 828cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 829cecf864dSYue Hu } else { 830cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 831cecf864dSYue Hu 83283a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 83364094a04SGao Xiang if (IS_ERR(grp)) { 83464094a04SGao Xiang err = PTR_ERR(grp); 83564094a04SGao Xiang goto err_out; 83664094a04SGao Xiang } 83764094a04SGao Xiang 83864094a04SGao Xiang if (grp != &pcl->obj) { 8395c6dcc57SGao Xiang fe->pcl = container_of(grp, 840cecf864dSYue Hu struct z_erofs_pcluster, obj); 84164094a04SGao Xiang err = -EEXIST; 84264094a04SGao Xiang goto err_out; 84347e4937aSGao Xiang } 844cecf864dSYue Hu } 8455c6dcc57SGao Xiang fe->owned_head = &pcl->next; 8465c6dcc57SGao Xiang fe->pcl = pcl; 8479e579fc1SGao Xiang return 0; 84864094a04SGao Xiang 84964094a04SGao Xiang err_out: 85087ca34a7SGao Xiang mutex_unlock(&pcl->lock); 8519f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 85264094a04SGao Xiang return err; 85347e4937aSGao Xiang } 85447e4937aSGao Xiang 85583a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 85647e4937aSGao Xiang { 85783a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 8580d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 8599e579fc1SGao Xiang int ret; 86047e4937aSGao Xiang 86187ca34a7SGao Xiang DBG_BUGON(fe->pcl); 86247e4937aSGao Xiang 86387ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 8645c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 86547e4937aSGao Xiang 8660d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 8670d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 8680d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 8690d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 87047e4937aSGao Xiang DBG_BUGON(1); 871cecf864dSYue Hu return -EFSCORRUPTED; 872cecf864dSYue Hu } 87347e4937aSGao Xiang 87464094a04SGao Xiang if (grp) { 8755c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 8760d823b42SGao Xiang ret = -EEXIST; 87764094a04SGao Xiang } else { 87883a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 87964094a04SGao Xiang } 88047e4937aSGao Xiang 8810d823b42SGao Xiang if (ret == -EEXIST) { 882267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 883267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 8840d823b42SGao Xiang } else if (ret) { 8850d823b42SGao Xiang return ret; 8860d823b42SGao Xiang } 88706a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 888387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 88981382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 890ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 89147e4937aSGao Xiang return 0; 89247e4937aSGao Xiang } 89347e4937aSGao Xiang 89447e4937aSGao Xiang /* 89547e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 89647e4937aSGao Xiang * only after a RCU grace period. 89747e4937aSGao Xiang */ 89847e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 89947e4937aSGao Xiang { 90087ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 90187ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 90247e4937aSGao Xiang } 90347e4937aSGao Xiang 90447e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 90547e4937aSGao Xiang { 90647e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 90747e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 90847e4937aSGao Xiang 90987ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 91047e4937aSGao Xiang } 91147e4937aSGao Xiang 9125c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 91347e4937aSGao Xiang { 91487ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 91547e4937aSGao Xiang 91687ca34a7SGao Xiang if (!pcl) 91747e4937aSGao Xiang return false; 91847e4937aSGao Xiang 91906a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 92087ca34a7SGao Xiang mutex_unlock(&pcl->lock); 92147e4937aSGao Xiang 92205b63d2bSGao Xiang if (fe->candidate_bvpage) 92306a304cdSGao Xiang fe->candidate_bvpage = NULL; 92406a304cdSGao Xiang 92547e4937aSGao Xiang /* 92647e4937aSGao Xiang * if all pending pages are added, don't hold its reference 92747e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 92847e4937aSGao Xiang */ 929db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 93087ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 93147e4937aSGao Xiang 93287ca34a7SGao Xiang fe->pcl = NULL; 93347e4937aSGao Xiang return true; 93447e4937aSGao Xiang } 93547e4937aSGao Xiang 936b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 937b15b2e30SYue Hu struct page *page, unsigned int pageofs, 938b15b2e30SYue Hu unsigned int len) 939b15b2e30SYue Hu { 9403acea5fcSJingbo Xu struct super_block *sb = inode->i_sb; 941b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 942b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 943b15b2e30SYue Hu u8 *src, *dst; 944b15b2e30SYue Hu unsigned int i, cnt; 945b15b2e30SYue Hu 946e5126de1SYue Hu if (!packed_inode) 947e5126de1SYue Hu return -EFSCORRUPTED; 948e5126de1SYue Hu 949eb2c5e41SGao Xiang buf.inode = packed_inode; 950b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 951b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 952b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 9533acea5fcSJingbo Xu sb->s_blocksize - erofs_blkoff(sb, pos)); 954eb2c5e41SGao Xiang src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 955b15b2e30SYue Hu if (IS_ERR(src)) { 956b15b2e30SYue Hu erofs_put_metabuf(&buf); 957b15b2e30SYue Hu return PTR_ERR(src); 958b15b2e30SYue Hu } 959b15b2e30SYue Hu 960b15b2e30SYue Hu dst = kmap_local_page(page); 9613acea5fcSJingbo Xu memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 962b15b2e30SYue Hu kunmap_local(dst); 963b15b2e30SYue Hu pos += cnt; 964b15b2e30SYue Hu } 965b15b2e30SYue Hu erofs_put_metabuf(&buf); 966b15b2e30SYue Hu return 0; 967b15b2e30SYue Hu } 968b15b2e30SYue Hu 96947e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 9706ab5eed6SGao Xiang struct page *page) 97147e4937aSGao Xiang { 97247e4937aSGao Xiang struct inode *const inode = fe->inode; 97347e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 97447e4937aSGao Xiang const loff_t offset = page_offset(page); 9755b220b20SGao Xiang bool tight = true, exclusive; 9762bfab9c0SGao Xiang unsigned int cur, end, spiltted; 97747e4937aSGao Xiang int err = 0; 97847e4937aSGao Xiang 97947e4937aSGao Xiang /* register locked file pages as online pages in pack */ 98047e4937aSGao Xiang z_erofs_onlinepage_init(page); 98147e4937aSGao Xiang 98247e4937aSGao Xiang spiltted = 0; 98347e4937aSGao Xiang end = PAGE_SIZE; 98447e4937aSGao Xiang repeat: 98547e4937aSGao Xiang cur = end - 1; 98647e4937aSGao Xiang 98739397a46SGao Xiang if (offset + cur < map->m_la || 98839397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 9895c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 99047e4937aSGao Xiang fe->backmost = false; 99147e4937aSGao Xiang map->m_la = offset + cur; 99247e4937aSGao Xiang map->m_llen = 0; 99347e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 9948d8a09b0SGao Xiang if (err) 99567148551SGao Xiang goto out; 99639397a46SGao Xiang } else { 99739397a46SGao Xiang if (fe->pcl) 99839397a46SGao Xiang goto hitted; 99939397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 100039397a46SGao Xiang } 100147e4937aSGao Xiang 1002b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 1003b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 100447e4937aSGao Xiang goto hitted; 100547e4937aSGao Xiang 100683a386c0SGao Xiang err = z_erofs_collector_begin(fe); 10078d8a09b0SGao Xiang if (err) 100867148551SGao Xiang goto out; 100947e4937aSGao Xiang 10105c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 101109c54379SGao Xiang void *mp; 1012cecf864dSYue Hu 101309c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 10143acea5fcSJingbo Xu erofs_blknr(inode->i_sb, map->m_pa), 10153acea5fcSJingbo Xu EROFS_NO_KMAP); 101609c54379SGao Xiang if (IS_ERR(mp)) { 101709c54379SGao Xiang err = PTR_ERR(mp); 1018cecf864dSYue Hu erofs_err(inode->i_sb, 1019cecf864dSYue Hu "failed to get inline page, err %d", err); 102067148551SGao Xiang goto out; 1021cecf864dSYue Hu } 102209c54379SGao Xiang get_page(fe->map.buf.page); 1023ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 1024ed722fbcSGao Xiang fe->map.buf.page); 1025db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 1026cecf864dSYue Hu } else { 10276f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 10286ab5eed6SGao Xiang z_erofs_bind_cache(fe); 1029cecf864dSYue Hu } 103047e4937aSGao Xiang hitted: 1031dc76ea8cSGao Xiang /* 1032dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 1033dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 1034dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 1035387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 1036dc76ea8cSGao Xiang */ 1037967c28b2SGao Xiang tight &= (fe->mode > Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1038dc76ea8cSGao Xiang 103947e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 10408d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 104147e4937aSGao Xiang zero_user_segment(page, cur, end); 104247e4937aSGao Xiang goto next_part; 104347e4937aSGao Xiang } 1044b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 1045b15b2e30SYue Hu unsigned int pageofs, skip, len; 1046b15b2e30SYue Hu 1047b15b2e30SYue Hu if (offset > map->m_la) { 1048b15b2e30SYue Hu pageofs = 0; 1049b15b2e30SYue Hu skip = offset - map->m_la; 1050b15b2e30SYue Hu } else { 1051b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 1052b15b2e30SYue Hu skip = 0; 1053b15b2e30SYue Hu } 1054b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 1055b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1056b15b2e30SYue Hu if (err) 1057b15b2e30SYue Hu goto out; 1058b15b2e30SYue Hu ++spiltted; 1059b15b2e30SYue Hu tight = false; 1060b15b2e30SYue Hu goto next_part; 1061b15b2e30SYue Hu } 106247e4937aSGao Xiang 10635b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 106447e4937aSGao Xiang if (cur) 1065db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 106647e4937aSGao Xiang 106706a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 106806a304cdSGao Xiang .page = page, 106906a304cdSGao Xiang .offset = offset - map->m_la, 107006a304cdSGao Xiang .end = end, 10715b220b20SGao Xiang }), exclusive); 107205b63d2bSGao Xiang if (err) 107367148551SGao Xiang goto out; 107447e4937aSGao Xiang 107567148551SGao Xiang z_erofs_onlinepage_split(page); 107647e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 107747e4937aSGao Xiang ++spiltted; 1078267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1079267f2492SGao Xiang fe->pcl->multibases = true; 10802bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 10812bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 10822bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 10832bfab9c0SGao Xiang } 1084e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1085e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1086e7933278SGao Xiang fe->pcl->length == map->m_llen) 1087e7933278SGao Xiang fe->pcl->partial = false; 108847e4937aSGao Xiang next_part: 10892bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 109047e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 10912bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 109247e4937aSGao Xiang 109347e4937aSGao Xiang end = cur; 109447e4937aSGao Xiang if (end > 0) 109547e4937aSGao Xiang goto repeat; 109647e4937aSGao Xiang 109747e4937aSGao Xiang out: 109867148551SGao Xiang if (err) 109967148551SGao Xiang z_erofs_page_mark_eio(page); 110047e4937aSGao Xiang z_erofs_onlinepage_endio(page); 110147e4937aSGao Xiang return err; 110247e4937aSGao Xiang } 110347e4937aSGao Xiang 1104ef4b4b46SYue Hu static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 110540452ffcSHuang Jianan unsigned int readahead_pages) 110640452ffcSHuang Jianan { 1107a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 110840452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 110940452ffcSHuang Jianan !readahead_pages) 111040452ffcSHuang Jianan return true; 111140452ffcSHuang Jianan 111240452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 111340452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 111440452ffcSHuang Jianan return true; 111540452ffcSHuang Jianan 111640452ffcSHuang Jianan return false; 111740452ffcSHuang Jianan } 111840452ffcSHuang Jianan 11196aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 11206aaa7b06SGao Xiang { 11216aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 11226aaa7b06SGao Xiang } 11236aaa7b06SGao Xiang 11244f05687fSGao Xiang struct z_erofs_decompress_backend { 11254f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 11264f05687fSGao Xiang struct super_block *sb; 11274f05687fSGao Xiang struct z_erofs_pcluster *pcl; 11284f05687fSGao Xiang 11294f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 11304f05687fSGao Xiang struct page **decompressed_pages; 11314f05687fSGao Xiang /* pages to keep the compressed data */ 11324f05687fSGao Xiang struct page **compressed_pages; 11334f05687fSGao Xiang 1134267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 11354f05687fSGao Xiang struct page **pagepool; 11362bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 11374f05687fSGao Xiang }; 11384f05687fSGao Xiang 1139267f2492SGao Xiang struct z_erofs_bvec_item { 1140267f2492SGao Xiang struct z_erofs_bvec bvec; 1141267f2492SGao Xiang struct list_head list; 1142267f2492SGao Xiang }; 1143267f2492SGao Xiang 1144267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 11453fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 11463fe96ee0SGao Xiang { 1147267f2492SGao Xiang struct z_erofs_bvec_item *item; 1148267f2492SGao Xiang 1149267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1150267f2492SGao Xiang unsigned int pgnr; 11513fe96ee0SGao Xiang 1152267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 11532bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 115463bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 11553fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1156267f2492SGao Xiang return; 11573fe96ee0SGao Xiang } 115863bbb856SGao Xiang } 11593fe96ee0SGao Xiang 1160267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1161267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1162267f2492SGao Xiang item->bvec = *bvec; 1163267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1164267f2492SGao Xiang } 1165267f2492SGao Xiang 1166267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1167267f2492SGao Xiang int err) 1168267f2492SGao Xiang { 1169267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1170267f2492SGao Xiang struct list_head *p, *n; 1171267f2492SGao Xiang 1172267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1173267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1174267f2492SGao Xiang unsigned int end, cur; 1175267f2492SGao Xiang void *dst, *src; 1176267f2492SGao Xiang 1177267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1178267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1179267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1180267f2492SGao Xiang bvi->bvec.end); 1181267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1182267f2492SGao Xiang while (cur < end) { 1183267f2492SGao Xiang unsigned int pgnr, scur, len; 1184267f2492SGao Xiang 1185267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1186267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1187267f2492SGao Xiang 1188267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1189267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1190267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1191267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1192267f2492SGao Xiang err = -EFSCORRUPTED; 1193267f2492SGao Xiang cur += len; 1194267f2492SGao Xiang continue; 1195267f2492SGao Xiang } 1196267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1197267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1198267f2492SGao Xiang kunmap_local(src); 1199267f2492SGao Xiang cur += len; 1200267f2492SGao Xiang } 1201267f2492SGao Xiang kunmap_local(dst); 1202267f2492SGao Xiang if (err) 1203267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 1204267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 1205267f2492SGao Xiang list_del(p); 1206267f2492SGao Xiang kfree(bvi); 1207267f2492SGao Xiang } 1208267f2492SGao Xiang } 1209267f2492SGao Xiang 1210267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 121142fec235SGao Xiang { 12124f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 121306a304cdSGao Xiang struct z_erofs_bvec_iter biter; 121406a304cdSGao Xiang struct page *old_bvpage; 1215267f2492SGao Xiang int i; 121642fec235SGao Xiang 1217387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 121842fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 121906a304cdSGao Xiang struct z_erofs_bvec bvec; 122042fec235SGao Xiang 122106a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 122242fec235SGao Xiang 122306a304cdSGao Xiang if (old_bvpage) 12244f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 122542fec235SGao Xiang 122606a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1227267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 122842fec235SGao Xiang } 122906a304cdSGao Xiang 123006a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 123106a304cdSGao Xiang if (old_bvpage) 12324f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 123342fec235SGao Xiang } 123442fec235SGao Xiang 12354f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 12364f05687fSGao Xiang bool *overlapped) 123767139e36SGao Xiang { 12384f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 123967139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 124067139e36SGao Xiang int i, err = 0; 124167139e36SGao Xiang 124267139e36SGao Xiang *overlapped = false; 124367139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1244ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1245ed722fbcSGao Xiang struct page *page = bvec->page; 124667139e36SGao Xiang 124767139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 124867139e36SGao Xiang if (!page) { 124967139e36SGao Xiang DBG_BUGON(1); 125067139e36SGao Xiang continue; 125167139e36SGao Xiang } 1252fe3e5914SGao Xiang be->compressed_pages[i] = page; 125367139e36SGao Xiang 125467139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 125567139e36SGao Xiang if (!PageUptodate(page)) 125667139e36SGao Xiang err = -EIO; 125767139e36SGao Xiang continue; 125867139e36SGao Xiang } 125967139e36SGao Xiang 126067139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 126167139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 12624f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 126367139e36SGao Xiang if (!PageUptodate(page)) 126467139e36SGao Xiang err = -EIO; 126567139e36SGao Xiang continue; 126667139e36SGao Xiang } 1267267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 126867139e36SGao Xiang *overlapped = true; 126967139e36SGao Xiang } 127067139e36SGao Xiang } 127167139e36SGao Xiang 1272fe3e5914SGao Xiang if (err) 12734f05687fSGao Xiang return err; 12744f05687fSGao Xiang return 0; 127567139e36SGao Xiang } 127667139e36SGao Xiang 12774f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 12784f05687fSGao Xiang int err) 127947e4937aSGao Xiang { 12804f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 12814f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1282cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1283597e2953SYue Hu const struct z_erofs_decompressor *decompressor = 1284597e2953SYue Hu &erofs_decompressors[pcl->algorithmformat]; 12852bfab9c0SGao Xiang unsigned int i, inputsize; 128667148551SGao Xiang int err2; 12872bfab9c0SGao Xiang struct page *page; 12882bfab9c0SGao Xiang bool overlapped; 128947e4937aSGao Xiang 129087ca34a7SGao Xiang mutex_lock(&pcl->lock); 12912bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 129247e4937aSGao Xiang 1293fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1294fe3e5914SGao Xiang be->decompressed_pages = NULL; 1295fe3e5914SGao Xiang be->compressed_pages = NULL; 1296fe3e5914SGao Xiang be->onstack_used = 0; 12972bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 12984f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 12992bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 13004f05687fSGao Xiang memset(be->decompressed_pages, 0, 13012bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1302fe3e5914SGao Xiang } 1303fe3e5914SGao Xiang 1304fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1305fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1306fe3e5914SGao Xiang 1307fe3e5914SGao Xiang if (!be->decompressed_pages) 13084f05687fSGao Xiang be->decompressed_pages = 1309647dd2c3SGao Xiang kvcalloc(be->nr_pages, sizeof(struct page *), 1310e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1311fe3e5914SGao Xiang if (!be->compressed_pages) 1312fe3e5914SGao Xiang be->compressed_pages = 1313647dd2c3SGao Xiang kvcalloc(pclusterpages, sizeof(struct page *), 1314fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 131547e4937aSGao Xiang 1316267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 13174f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 13184f05687fSGao Xiang if (err2) 13194f05687fSGao Xiang err = err2; 13208d8a09b0SGao Xiang if (err) 132147e4937aSGao Xiang goto out; 132247e4937aSGao Xiang 1323cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1324cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1325cecf864dSYue Hu else 1326cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1327cecf864dSYue Hu 1328597e2953SYue Hu err = decompressor->decompress(&(struct z_erofs_decompress_req) { 13294f05687fSGao Xiang .sb = be->sb, 13304f05687fSGao Xiang .in = be->compressed_pages, 13314f05687fSGao Xiang .out = be->decompressed_pages, 1332cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 133387ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 13349f6cc76eSGao Xiang .inputsize = inputsize, 13352bfab9c0SGao Xiang .outputsize = pcl->length, 133647e4937aSGao Xiang .alg = pcl->algorithmformat, 133747e4937aSGao Xiang .inplace_io = overlapped, 13382bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1339267f2492SGao Xiang .fillgaps = pcl->multibases, 13404f05687fSGao Xiang }, be->pagepool); 134147e4937aSGao Xiang 134247e4937aSGao Xiang out: 1343cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1344cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1345ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1346ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1347cecf864dSYue Hu put_page(page); 1348cecf864dSYue Hu } else { 1349cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1350ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 135147e4937aSGao Xiang 135247e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 135347e4937aSGao Xiang continue; 135447e4937aSGao Xiang 13556aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13564f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1357ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 135847e4937aSGao Xiang } 1359cecf864dSYue Hu } 1360fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1361fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1362647dd2c3SGao Xiang kvfree(be->compressed_pages); 1363267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 136447e4937aSGao Xiang 13652bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 13664f05687fSGao Xiang page = be->decompressed_pages[i]; 136747e4937aSGao Xiang if (!page) 136847e4937aSGao Xiang continue; 136947e4937aSGao Xiang 13706aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 137147e4937aSGao Xiang 13726aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13734f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 137447e4937aSGao Xiang continue; 137567148551SGao Xiang if (err) 137667148551SGao Xiang z_erofs_page_mark_eio(page); 137747e4937aSGao Xiang z_erofs_onlinepage_endio(page); 137847e4937aSGao Xiang } 137947e4937aSGao Xiang 13804f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 1381647dd2c3SGao Xiang kvfree(be->decompressed_pages); 138247e4937aSGao Xiang 13832bfab9c0SGao Xiang pcl->length = 0; 13842bfab9c0SGao Xiang pcl->partial = true; 1385267f2492SGao Xiang pcl->multibases = false; 138606a304cdSGao Xiang pcl->bvset.nextpage = NULL; 138787ca34a7SGao Xiang pcl->vcnt = 0; 138847e4937aSGao Xiang 138987ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 139047e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 139187ca34a7SGao Xiang mutex_unlock(&pcl->lock); 139247e4937aSGao Xiang return err; 139347e4937aSGao Xiang } 139447e4937aSGao Xiang 13950c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1396eaa9172aSGao Xiang struct page **pagepool) 139747e4937aSGao Xiang { 13984f05687fSGao Xiang struct z_erofs_decompress_backend be = { 13994f05687fSGao Xiang .sb = io->sb, 14004f05687fSGao Xiang .pagepool = pagepool, 1401267f2492SGao Xiang .decompressed_secondary_bvecs = 1402267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 14034f05687fSGao Xiang }; 140447e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 140547e4937aSGao Xiang 1406967c28b2SGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL) { 140747e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 140847e4937aSGao Xiang 14094f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 14104f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 141147e4937aSGao Xiang 14124f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 14134f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 141447e4937aSGao Xiang } 141547e4937aSGao Xiang } 141647e4937aSGao Xiang 14170c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 141847e4937aSGao Xiang { 1419a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1420a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1421eaa9172aSGao Xiang struct page *pagepool = NULL; 142247e4937aSGao Xiang 1423967c28b2SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL); 14240c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 1425eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1426a4b1fab1SGao Xiang kvfree(bgq); 142747e4937aSGao Xiang } 142847e4937aSGao Xiang 14293fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14303fffb589SSandeep Dhavale static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 14313fffb589SSandeep Dhavale { 14323fffb589SSandeep Dhavale z_erofs_decompressqueue_work((struct work_struct *)work); 14333fffb589SSandeep Dhavale } 14343fffb589SSandeep Dhavale #endif 14353fffb589SSandeep Dhavale 14367865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1437cdba5506SGao Xiang int bios) 14387865827cSGao Xiang { 14397865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 14407865827cSGao Xiang 14417865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1442cdba5506SGao Xiang if (io->sync) { 14437865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 144460b30050SHongyu Jin complete(&io->u.done); 14457865827cSGao Xiang return; 14467865827cSGao Xiang } 14477865827cSGao Xiang 14487865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 14497865827cSGao Xiang return; 14503fffb589SSandeep Dhavale /* Use (kthread_)work and sync decompression for atomic contexts only */ 14517865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 14523fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14533fffb589SSandeep Dhavale struct kthread_worker *worker; 14543fffb589SSandeep Dhavale 14553fffb589SSandeep Dhavale rcu_read_lock(); 14563fffb589SSandeep Dhavale worker = rcu_dereference( 14573fffb589SSandeep Dhavale z_erofs_pcpu_workers[raw_smp_processor_id()]); 14583fffb589SSandeep Dhavale if (!worker) { 14593fffb589SSandeep Dhavale INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 14607865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 14613fffb589SSandeep Dhavale } else { 14623fffb589SSandeep Dhavale kthread_queue_work(worker, &io->u.kthread_work); 14633fffb589SSandeep Dhavale } 14643fffb589SSandeep Dhavale rcu_read_unlock(); 14653fffb589SSandeep Dhavale #else 14663fffb589SSandeep Dhavale queue_work(z_erofs_workqueue, &io->u.work); 14673fffb589SSandeep Dhavale #endif 14687865827cSGao Xiang /* enable sync decompression for readahead */ 14697865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 14707865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 14717865827cSGao Xiang return; 14727865827cSGao Xiang } 14737865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 14747865827cSGao Xiang } 14757865827cSGao Xiang 147647e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 147747e4937aSGao Xiang unsigned int nr, 1478eaa9172aSGao Xiang struct page **pagepool, 14799f2731d6SGao Xiang struct address_space *mc) 148047e4937aSGao Xiang { 148147e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 14829f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 148347e4937aSGao Xiang bool tocache = false; 148447e4937aSGao Xiang 148547e4937aSGao Xiang struct address_space *mapping; 148647e4937aSGao Xiang struct page *oldpage, *page; 148747e4937aSGao Xiang int justfound; 148847e4937aSGao Xiang 148947e4937aSGao Xiang repeat: 1490ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 149147e4937aSGao Xiang oldpage = page; 149247e4937aSGao Xiang 149347e4937aSGao Xiang if (!page) 149447e4937aSGao Xiang goto out_allocpage; 149547e4937aSGao Xiang 1496b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1497b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 149847e4937aSGao Xiang 14991825c8d7SGao Xiang /* 15001825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 15011825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 15021825c8d7SGao Xiang */ 15031825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1504ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 15051825c8d7SGao Xiang set_page_private(page, 0); 15061825c8d7SGao Xiang tocache = true; 15071825c8d7SGao Xiang goto out_tocache; 15081825c8d7SGao Xiang } 150947e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 151047e4937aSGao Xiang 151147e4937aSGao Xiang /* 15126aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 151347e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 151447e4937aSGao Xiang */ 151547e4937aSGao Xiang if (mapping && mapping != mc) 151647e4937aSGao Xiang /* ought to be unmanaged pages */ 151747e4937aSGao Xiang goto out; 151847e4937aSGao Xiang 15196aaa7b06SGao Xiang /* directly return for shortlived page as well */ 15206aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 15216aaa7b06SGao Xiang goto out; 15226aaa7b06SGao Xiang 152347e4937aSGao Xiang lock_page(page); 152447e4937aSGao Xiang 152547e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 152647e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 152747e4937aSGao Xiang 152847e4937aSGao Xiang /* the page is still in manage cache */ 152947e4937aSGao Xiang if (page->mapping == mc) { 1530ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 153147e4937aSGao Xiang 153247e4937aSGao Xiang if (!PagePrivate(page)) { 153347e4937aSGao Xiang /* 153447e4937aSGao Xiang * impossible to be !PagePrivate(page) for 153547e4937aSGao Xiang * the current restriction as well if 1536ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 153747e4937aSGao Xiang */ 153847e4937aSGao Xiang DBG_BUGON(!justfound); 153947e4937aSGao Xiang 154047e4937aSGao Xiang justfound = 0; 154147e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 154247e4937aSGao Xiang SetPagePrivate(page); 154347e4937aSGao Xiang } 154447e4937aSGao Xiang 154547e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 154647e4937aSGao Xiang if (PageUptodate(page)) { 154747e4937aSGao Xiang unlock_page(page); 154847e4937aSGao Xiang page = NULL; 154947e4937aSGao Xiang } 155047e4937aSGao Xiang goto out; 155147e4937aSGao Xiang } 155247e4937aSGao Xiang 155347e4937aSGao Xiang /* 155447e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 155547e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 155647e4937aSGao Xiang */ 155747e4937aSGao Xiang DBG_BUGON(page->mapping); 155847e4937aSGao Xiang DBG_BUGON(!justfound); 155947e4937aSGao Xiang 156047e4937aSGao Xiang tocache = true; 156147e4937aSGao Xiang unlock_page(page); 156247e4937aSGao Xiang put_page(page); 156347e4937aSGao Xiang out_allocpage: 15645ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1565ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1566ed722fbcSGao Xiang oldpage, page)) { 1567eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 15685ddcee1fSGao Xiang cond_resched(); 15695ddcee1fSGao Xiang goto repeat; 15705ddcee1fSGao Xiang } 15711825c8d7SGao Xiang out_tocache: 1572bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1573bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1574bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1575bf225074SGao Xiang goto out; 1576a30573b3SGao Xiang } 1577bf225074SGao Xiang attach_page_private(page, pcl); 1578bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1579bf225074SGao Xiang put_page(page); 1580bf225074SGao Xiang 158147e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 158247e4937aSGao Xiang return page; 158347e4937aSGao Xiang } 158447e4937aSGao Xiang 1585cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1586a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 158747e4937aSGao Xiang { 1588a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 158947e4937aSGao Xiang 1590a4b1fab1SGao Xiang if (fg && !*fg) { 1591a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1592a4b1fab1SGao Xiang if (!q) { 1593a4b1fab1SGao Xiang *fg = true; 1594a4b1fab1SGao Xiang goto fg_out; 159547e4937aSGao Xiang } 15963fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 15973fffb589SSandeep Dhavale kthread_init_work(&q->u.kthread_work, 15983fffb589SSandeep Dhavale z_erofs_decompressqueue_kthread_work); 15993fffb589SSandeep Dhavale #else 16000c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 16013fffb589SSandeep Dhavale #endif 1602a4b1fab1SGao Xiang } else { 1603a4b1fab1SGao Xiang fg_out: 1604a4b1fab1SGao Xiang q = fgq; 160560b30050SHongyu Jin init_completion(&fgq->u.done); 1606a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 160767148551SGao Xiang q->eio = false; 1608cdba5506SGao Xiang q->sync = true; 1609a4b1fab1SGao Xiang } 1610a4b1fab1SGao Xiang q->sb = sb; 1611967c28b2SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL; 1612a4b1fab1SGao Xiang return q; 161347e4937aSGao Xiang } 161447e4937aSGao Xiang 161547e4937aSGao Xiang /* define decompression jobqueue types */ 161647e4937aSGao Xiang enum { 161747e4937aSGao Xiang JQ_BYPASS, 161847e4937aSGao Xiang JQ_SUBMIT, 161947e4937aSGao Xiang NR_JOBQUEUES, 162047e4937aSGao Xiang }; 162147e4937aSGao Xiang 162247e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 162347e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 162447e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 162547e4937aSGao Xiang { 162647e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 162747e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 162847e4937aSGao Xiang 1629967c28b2SGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL); 163047e4937aSGao Xiang 163147e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 163247e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 163347e4937aSGao Xiang 163447e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 163547e4937aSGao Xiang } 163647e4937aSGao Xiang 16377865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 16387865827cSGao Xiang { 1639cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 16407865827cSGao Xiang blk_status_t err = bio->bi_status; 16417865827cSGao Xiang struct bio_vec *bvec; 16427865827cSGao Xiang struct bvec_iter_all iter_all; 16437865827cSGao Xiang 16447865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 16457865827cSGao Xiang struct page *page = bvec->bv_page; 16467865827cSGao Xiang 16477865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 16487865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 16497865827cSGao Xiang 16507865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 16517865827cSGao Xiang if (!err) 16527865827cSGao Xiang SetPageUptodate(page); 16537865827cSGao Xiang unlock_page(page); 16547865827cSGao Xiang } 16557865827cSGao Xiang } 165667148551SGao Xiang if (err) 165767148551SGao Xiang q->eio = true; 1658cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 16597865827cSGao Xiang bio_put(bio); 16607865827cSGao Xiang } 16617865827cSGao Xiang 166283a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1663a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1664ef4b4b46SYue Hu bool *force_fg, bool readahead) 166547e4937aSGao Xiang { 166683a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 166783a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 166847e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1669a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 16705c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1671dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 16723f649ab7SKees Cook pgoff_t last_index; 1673dfeab2e9SGao Xiang struct block_device *last_bdev; 16741e4a2955SGao Xiang unsigned int nr_bios = 0; 16751e4a2955SGao Xiang struct bio *bio = NULL; 167682e60d00SJohannes Weiner unsigned long pflags; 167782e60d00SJohannes Weiner int memstall = 0; 167847e4937aSGao Xiang 1679cdba5506SGao Xiang /* 1680cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1681cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1682cdba5506SGao Xiang */ 1683cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1684cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1685cdba5506SGao Xiang 1686a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1687a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 168847e4937aSGao Xiang 168947e4937aSGao Xiang /* by default, all need io submission */ 169047e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 169147e4937aSGao Xiang 169247e4937aSGao Xiang do { 1693dfeab2e9SGao Xiang struct erofs_map_dev mdev; 169447e4937aSGao Xiang struct z_erofs_pcluster *pcl; 16951e4a2955SGao Xiang pgoff_t cur, end; 16961e4a2955SGao Xiang unsigned int i = 0; 16971e4a2955SGao Xiang bool bypass = true; 169847e4937aSGao Xiang 169947e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 170047e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 1701967c28b2SGao Xiang owned_head = READ_ONCE(pcl->next); 170247e4937aSGao Xiang 1703cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1704cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1705cecf864dSYue Hu continue; 1706cecf864dSYue Hu } 1707cecf864dSYue Hu 1708dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1709dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 17103acea5fcSJingbo Xu .m_pa = erofs_pos(sb, pcl->obj.index), 1711dfeab2e9SGao Xiang }; 1712dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1713dfeab2e9SGao Xiang 17143acea5fcSJingbo Xu cur = erofs_blknr(sb, mdev.m_pa); 17159f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 171647e4937aSGao Xiang 17171e4a2955SGao Xiang do { 17181e4a2955SGao Xiang struct page *page; 171947e4937aSGao Xiang 17206ab5eed6SGao Xiang page = pickup_page_for_submission(pcl, i++, 17216ab5eed6SGao Xiang &f->pagepool, mc); 17221e4a2955SGao Xiang if (!page) 17231e4a2955SGao Xiang continue; 172447e4937aSGao Xiang 1725dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1726dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 172747e4937aSGao Xiang submit_bio_retry: 172894e4e153SGao Xiang submit_bio(bio); 172982e60d00SJohannes Weiner if (memstall) { 173082e60d00SJohannes Weiner psi_memstall_leave(&pflags); 173182e60d00SJohannes Weiner memstall = 0; 173282e60d00SJohannes Weiner } 173347e4937aSGao Xiang bio = NULL; 173447e4937aSGao Xiang } 173547e4937aSGao Xiang 173682e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 173799486c51SChristoph Hellwig psi_memstall_enter(&pflags); 173882e60d00SJohannes Weiner memstall = 1; 173982e60d00SJohannes Weiner } 174099486c51SChristoph Hellwig 174147e4937aSGao Xiang if (!bio) { 174207888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 174307888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 17440c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1745dfeab2e9SGao Xiang 1746dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 17471e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 17483acea5fcSJingbo Xu (sb->s_blocksize_bits - 9); 1749cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 1750ef4b4b46SYue Hu if (readahead) 17516ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 175247e4937aSGao Xiang ++nr_bios; 175347e4937aSGao Xiang } 175447e4937aSGao Xiang 17556c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 175647e4937aSGao Xiang goto submit_bio_retry; 175747e4937aSGao Xiang 17581e4a2955SGao Xiang last_index = cur; 17591e4a2955SGao Xiang bypass = false; 17601e4a2955SGao Xiang } while (++cur < end); 176147e4937aSGao Xiang 17621e4a2955SGao Xiang if (!bypass) 176347e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 176447e4937aSGao Xiang else 176547e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 176647e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 176747e4937aSGao Xiang 176899486c51SChristoph Hellwig if (bio) { 176994e4e153SGao Xiang submit_bio(bio); 177082e60d00SJohannes Weiner if (memstall) 177182e60d00SJohannes Weiner psi_memstall_leave(&pflags); 177299486c51SChristoph Hellwig } 177347e4937aSGao Xiang 1774587a67b7SGao Xiang /* 1775587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 17763fffb589SSandeep Dhavale * don't issue decompression but drop it directly instead. 1777587a67b7SGao Xiang */ 1778587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1779587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 17801e4a2955SGao Xiang return; 1781587a67b7SGao Xiang } 1782cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 178347e4937aSGao Xiang } 178447e4937aSGao Xiang 178583a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 17866ab5eed6SGao Xiang bool force_fg, bool ra) 178747e4937aSGao Xiang { 1788a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 178947e4937aSGao Xiang 17905c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 179147e4937aSGao Xiang return; 17926ab5eed6SGao Xiang z_erofs_submit_queue(f, io, &force_fg, ra); 179347e4937aSGao Xiang 17940c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 17956ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], &f->pagepool); 179647e4937aSGao Xiang 179747e4937aSGao Xiang if (!force_fg) 179847e4937aSGao Xiang return; 179947e4937aSGao Xiang 180047e4937aSGao Xiang /* wait until all bios are completed */ 180160b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 180247e4937aSGao Xiang 18030c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 18046ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], &f->pagepool); 180547e4937aSGao Xiang } 180647e4937aSGao Xiang 180738629291SGao Xiang /* 180838629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 180938629291SGao Xiang * approximate readmore strategies as a start. 181038629291SGao Xiang */ 181138629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 18126ab5eed6SGao Xiang struct readahead_control *rac, bool backmost) 181338629291SGao Xiang { 181438629291SGao Xiang struct inode *inode = f->inode; 181538629291SGao Xiang struct erofs_map_blocks *map = &f->map; 1816796e9149SYue Hu erofs_off_t cur, end, headoffset = f->headoffset; 181738629291SGao Xiang int err; 181838629291SGao Xiang 181938629291SGao Xiang if (backmost) { 1820796e9149SYue Hu if (rac) 1821796e9149SYue Hu end = headoffset + readahead_length(rac) - 1; 1822796e9149SYue Hu else 1823796e9149SYue Hu end = headoffset + PAGE_SIZE - 1; 182438629291SGao Xiang map->m_la = end; 1825622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1826622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 182738629291SGao Xiang if (err) 182838629291SGao Xiang return; 182938629291SGao Xiang 1830796e9149SYue Hu /* expand ra for the trailing edge if readahead */ 183138629291SGao Xiang if (rac) { 183238629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1833796e9149SYue Hu readahead_expand(rac, headoffset, cur - headoffset); 183438629291SGao Xiang return; 183538629291SGao Xiang } 183638629291SGao Xiang end = round_up(end, PAGE_SIZE); 183738629291SGao Xiang } else { 183838629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 183938629291SGao Xiang 184038629291SGao Xiang if (!map->m_llen) 184138629291SGao Xiang return; 184238629291SGao Xiang } 184338629291SGao Xiang 184438629291SGao Xiang cur = map->m_la + map->m_llen - 1; 184538629291SGao Xiang while (cur >= end) { 184638629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 184738629291SGao Xiang struct page *page; 184838629291SGao Xiang 184938629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1850aa793b46SGao Xiang if (page) { 185138629291SGao Xiang if (PageUptodate(page)) { 185238629291SGao Xiang unlock_page(page); 1853aa793b46SGao Xiang } else { 18546ab5eed6SGao Xiang err = z_erofs_do_read_page(f, page); 185538629291SGao Xiang if (err) 185638629291SGao Xiang erofs_err(inode->i_sb, 185738629291SGao Xiang "readmore error at page %lu @ nid %llu", 185838629291SGao Xiang index, EROFS_I(inode)->nid); 1859aa793b46SGao Xiang } 186038629291SGao Xiang put_page(page); 1861aa793b46SGao Xiang } 1862aa793b46SGao Xiang 186338629291SGao Xiang if (cur < PAGE_SIZE) 186438629291SGao Xiang break; 186538629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 186638629291SGao Xiang } 186738629291SGao Xiang } 186838629291SGao Xiang 1869a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 187047e4937aSGao Xiang { 1871a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 187247e4937aSGao Xiang struct inode *const inode = page->mapping->host; 187340452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 187447e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 187547e4937aSGao Xiang int err; 187647e4937aSGao Xiang 187747e4937aSGao Xiang trace_erofs_readpage(page, false); 187847e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 187947e4937aSGao Xiang 18806ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, true); 18816ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 18826ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, false); 18835c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 188447e4937aSGao Xiang 188547e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 18866ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, 0), false); 188747e4937aSGao Xiang 188847e4937aSGao Xiang if (err) 18894f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 189047e4937aSGao Xiang 189109c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 18926ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 189347e4937aSGao Xiang return err; 189447e4937aSGao Xiang } 189547e4937aSGao Xiang 18960615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 189747e4937aSGao Xiang { 18980615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 189947e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 190047e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 19016ab5eed6SGao Xiang struct page *head = NULL, *page; 190238629291SGao Xiang unsigned int nr_pages; 190347e4937aSGao Xiang 19040615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 190547e4937aSGao Xiang 19066ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, true); 190738629291SGao Xiang nr_pages = readahead_count(rac); 190838629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 190938629291SGao Xiang 19100615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 191147e4937aSGao Xiang set_page_private(page, (unsigned long)head); 191247e4937aSGao Xiang head = page; 191347e4937aSGao Xiang } 191447e4937aSGao Xiang 191547e4937aSGao Xiang while (head) { 191647e4937aSGao Xiang struct page *page = head; 191747e4937aSGao Xiang int err; 191847e4937aSGao Xiang 191947e4937aSGao Xiang /* traversal in reverse order */ 192047e4937aSGao Xiang head = (void *)page_private(page); 192147e4937aSGao Xiang 19226ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 1923a5876e24SGao Xiang if (err) 19244f761fa2SGao Xiang erofs_err(inode->i_sb, 19254f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 19264f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 192747e4937aSGao Xiang put_page(page); 192847e4937aSGao Xiang } 19296ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, false); 19305c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 193147e4937aSGao Xiang 19326ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, nr_pages), true); 193309c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 19346ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 193547e4937aSGao Xiang } 193647e4937aSGao Xiang 19370c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1938a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 19390615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 194047e4937aSGao Xiang }; 1941