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" 847e4937aSGao Xiang #include <linux/prefetch.h> 999486c51SChristoph Hellwig #include <linux/psi.h> 103fffb589SSandeep Dhavale #include <linux/cpuhotplug.h> 1147e4937aSGao Xiang #include <trace/events/erofs.h> 1247e4937aSGao Xiang 13a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_MAX_PAGES (Z_EROFS_PCLUSTER_MAX_SIZE / PAGE_SIZE) 14a9a94d93SGao Xiang #define Z_EROFS_INLINE_BVECS 2 15a9a94d93SGao Xiang 16a9a94d93SGao Xiang /* 17a9a94d93SGao Xiang * let's leave a type here in case of introducing 18a9a94d93SGao Xiang * another tagged pointer later. 19a9a94d93SGao Xiang */ 20a9a94d93SGao Xiang typedef void *z_erofs_next_pcluster_t; 21a9a94d93SGao Xiang 22a9a94d93SGao Xiang struct z_erofs_bvec { 23a9a94d93SGao Xiang struct page *page; 24a9a94d93SGao Xiang int offset; 25a9a94d93SGao Xiang unsigned int end; 26a9a94d93SGao Xiang }; 27a9a94d93SGao Xiang 28a9a94d93SGao Xiang #define __Z_EROFS_BVSET(name, total) \ 29a9a94d93SGao Xiang struct name { \ 30a9a94d93SGao Xiang /* point to the next page which contains the following bvecs */ \ 31a9a94d93SGao Xiang struct page *nextpage; \ 32a9a94d93SGao Xiang struct z_erofs_bvec bvec[total]; \ 33a9a94d93SGao Xiang } 34a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset,); 35a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset_inline, Z_EROFS_INLINE_BVECS); 36a9a94d93SGao Xiang 37a9a94d93SGao Xiang /* 38a9a94d93SGao Xiang * Structure fields follow one of the following exclusion rules. 39a9a94d93SGao Xiang * 40a9a94d93SGao Xiang * I: Modifiable by initialization/destruction paths and read-only 41a9a94d93SGao Xiang * for everyone else; 42a9a94d93SGao Xiang * 43a9a94d93SGao Xiang * L: Field should be protected by the pcluster lock; 44a9a94d93SGao Xiang * 45a9a94d93SGao Xiang * A: Field should be accessed / updated in atomic for parallelized code. 46a9a94d93SGao Xiang */ 47a9a94d93SGao Xiang struct z_erofs_pcluster { 48a9a94d93SGao Xiang struct erofs_workgroup obj; 49a9a94d93SGao Xiang struct mutex lock; 50a9a94d93SGao Xiang 51a9a94d93SGao Xiang /* A: point to next chained pcluster or TAILs */ 52a9a94d93SGao Xiang z_erofs_next_pcluster_t next; 53a9a94d93SGao Xiang 54a9a94d93SGao Xiang /* L: the maximum decompression size of this round */ 55a9a94d93SGao Xiang unsigned int length; 56a9a94d93SGao Xiang 57a9a94d93SGao Xiang /* L: total number of bvecs */ 58a9a94d93SGao Xiang unsigned int vcnt; 59a9a94d93SGao Xiang 60a9a94d93SGao Xiang /* I: page offset of start position of decompression */ 61a9a94d93SGao Xiang unsigned short pageofs_out; 62a9a94d93SGao Xiang 63a9a94d93SGao Xiang /* I: page offset of inline compressed data */ 64a9a94d93SGao Xiang unsigned short pageofs_in; 65a9a94d93SGao Xiang 66a9a94d93SGao Xiang union { 67a9a94d93SGao Xiang /* L: inline a certain number of bvec for bootstrap */ 68a9a94d93SGao Xiang struct z_erofs_bvset_inline bvset; 69a9a94d93SGao Xiang 70a9a94d93SGao Xiang /* I: can be used to free the pcluster by RCU. */ 71a9a94d93SGao Xiang struct rcu_head rcu; 72a9a94d93SGao Xiang }; 73a9a94d93SGao Xiang 74a9a94d93SGao Xiang union { 75a9a94d93SGao Xiang /* I: physical cluster size in pages */ 76a9a94d93SGao Xiang unsigned short pclusterpages; 77a9a94d93SGao Xiang 78a9a94d93SGao Xiang /* I: tailpacking inline compressed size */ 79a9a94d93SGao Xiang unsigned short tailpacking_size; 80a9a94d93SGao Xiang }; 81a9a94d93SGao Xiang 82a9a94d93SGao Xiang /* I: compression algorithm format */ 83a9a94d93SGao Xiang unsigned char algorithmformat; 84a9a94d93SGao Xiang 85a9a94d93SGao Xiang /* L: whether partial decompression or not */ 86a9a94d93SGao Xiang bool partial; 87a9a94d93SGao Xiang 88a9a94d93SGao Xiang /* L: indicate several pageofs_outs or not */ 89a9a94d93SGao Xiang bool multibases; 90a9a94d93SGao Xiang 91a9a94d93SGao Xiang /* A: compressed bvecs (can be cached or inplaced pages) */ 92a9a94d93SGao Xiang struct z_erofs_bvec compressed_bvecs[]; 93a9a94d93SGao Xiang }; 94a9a94d93SGao Xiang 95a9a94d93SGao Xiang /* let's avoid the valid 32-bit kernel addresses */ 96a9a94d93SGao Xiang 97a9a94d93SGao Xiang /* the chained workgroup has't submitted io (still open) */ 98a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE) 99a9a94d93SGao Xiang /* the chained workgroup has already submitted io */ 100a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL_CLOSED ((void *)0x5F0EDEAD) 101a9a94d93SGao Xiang 102a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_NIL (NULL) 103a9a94d93SGao Xiang 104a9a94d93SGao Xiang struct z_erofs_decompressqueue { 105a9a94d93SGao Xiang struct super_block *sb; 106a9a94d93SGao Xiang atomic_t pending_bios; 107a9a94d93SGao Xiang z_erofs_next_pcluster_t head; 108a9a94d93SGao Xiang 109a9a94d93SGao Xiang union { 110a9a94d93SGao Xiang struct completion done; 111a9a94d93SGao Xiang struct work_struct work; 1123fffb589SSandeep Dhavale struct kthread_work kthread_work; 113a9a94d93SGao Xiang } u; 114a9a94d93SGao Xiang bool eio, sync; 115a9a94d93SGao Xiang }; 116a9a94d93SGao Xiang 117a9a94d93SGao Xiang static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 118a9a94d93SGao Xiang { 119a9a94d93SGao Xiang return !pcl->obj.index; 120a9a94d93SGao Xiang } 121a9a94d93SGao Xiang 122a9a94d93SGao Xiang static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 123a9a94d93SGao Xiang { 124a9a94d93SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) 125a9a94d93SGao Xiang return 1; 126a9a94d93SGao Xiang return pcl->pclusterpages; 127a9a94d93SGao Xiang } 128a9a94d93SGao Xiang 129a9a94d93SGao Xiang /* 130a9a94d93SGao Xiang * bit 30: I/O error occurred on this page 131a9a94d93SGao Xiang * bit 0 - 29: remaining parts to complete this page 132a9a94d93SGao Xiang */ 133a9a94d93SGao Xiang #define Z_EROFS_PAGE_EIO (1 << 30) 134a9a94d93SGao Xiang 135a9a94d93SGao Xiang static inline void z_erofs_onlinepage_init(struct page *page) 136a9a94d93SGao Xiang { 137a9a94d93SGao Xiang union { 138a9a94d93SGao Xiang atomic_t o; 139a9a94d93SGao Xiang unsigned long v; 140a9a94d93SGao Xiang } u = { .o = ATOMIC_INIT(1) }; 141a9a94d93SGao Xiang 142a9a94d93SGao Xiang set_page_private(page, u.v); 143a9a94d93SGao Xiang smp_wmb(); 144a9a94d93SGao Xiang SetPagePrivate(page); 145a9a94d93SGao Xiang } 146a9a94d93SGao Xiang 147a9a94d93SGao Xiang static inline void z_erofs_onlinepage_split(struct page *page) 148a9a94d93SGao Xiang { 149a9a94d93SGao Xiang atomic_inc((atomic_t *)&page->private); 150a9a94d93SGao Xiang } 151a9a94d93SGao Xiang 152a9a94d93SGao Xiang static inline void z_erofs_page_mark_eio(struct page *page) 153a9a94d93SGao Xiang { 154a9a94d93SGao Xiang int orig; 155a9a94d93SGao Xiang 156a9a94d93SGao Xiang do { 157a9a94d93SGao Xiang orig = atomic_read((atomic_t *)&page->private); 158a9a94d93SGao Xiang } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 159a9a94d93SGao Xiang orig | Z_EROFS_PAGE_EIO) != orig); 160a9a94d93SGao Xiang } 161a9a94d93SGao Xiang 162a9a94d93SGao Xiang static inline void z_erofs_onlinepage_endio(struct page *page) 163a9a94d93SGao Xiang { 164a9a94d93SGao Xiang unsigned int v; 165a9a94d93SGao Xiang 166a9a94d93SGao Xiang DBG_BUGON(!PagePrivate(page)); 167a9a94d93SGao Xiang v = atomic_dec_return((atomic_t *)&page->private); 168a9a94d93SGao Xiang if (!(v & ~Z_EROFS_PAGE_EIO)) { 169a9a94d93SGao Xiang set_page_private(page, 0); 170a9a94d93SGao Xiang ClearPagePrivate(page); 171a9a94d93SGao Xiang if (!(v & Z_EROFS_PAGE_EIO)) 172a9a94d93SGao Xiang SetPageUptodate(page); 173a9a94d93SGao Xiang unlock_page(page); 174a9a94d93SGao Xiang } 175a9a94d93SGao Xiang } 176a9a94d93SGao Xiang 177a9a94d93SGao Xiang #define Z_EROFS_ONSTACK_PAGES 32 178a9a94d93SGao Xiang 17947e4937aSGao Xiang /* 1809f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 1819f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 1829f6cc76eSGao Xiang */ 1839f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 1849f6cc76eSGao Xiang struct kmem_cache *slab; 1859f6cc76eSGao Xiang unsigned int maxpages; 1869f6cc76eSGao Xiang char name[48]; 1879f6cc76eSGao Xiang }; 1889f6cc76eSGao Xiang 1899f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 1909f6cc76eSGao Xiang 1919f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 1929f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 1939f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 1949f6cc76eSGao Xiang }; 1959f6cc76eSGao Xiang 19606a304cdSGao Xiang struct z_erofs_bvec_iter { 19706a304cdSGao Xiang struct page *bvpage; 19806a304cdSGao Xiang struct z_erofs_bvset *bvset; 19906a304cdSGao Xiang unsigned int nr, cur; 20006a304cdSGao Xiang }; 20106a304cdSGao Xiang 20206a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 20306a304cdSGao Xiang { 20406a304cdSGao Xiang if (iter->bvpage) 20506a304cdSGao Xiang kunmap_local(iter->bvset); 20606a304cdSGao Xiang return iter->bvpage; 20706a304cdSGao Xiang } 20806a304cdSGao Xiang 20906a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 21006a304cdSGao Xiang { 21106a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 21206a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 21306a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 21406a304cdSGao Xiang struct page *oldpage; 21506a304cdSGao Xiang 21606a304cdSGao Xiang DBG_BUGON(!nextpage); 21706a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 21806a304cdSGao Xiang iter->bvpage = nextpage; 21906a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 22006a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 22106a304cdSGao Xiang iter->cur = 0; 22206a304cdSGao Xiang return oldpage; 22306a304cdSGao Xiang } 22406a304cdSGao Xiang 22506a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 22606a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 22706a304cdSGao Xiang unsigned int bootstrap_nr, 22806a304cdSGao Xiang unsigned int cur) 22906a304cdSGao Xiang { 23006a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 23106a304cdSGao Xiang .nr = bootstrap_nr, 23206a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 23306a304cdSGao Xiang }; 23406a304cdSGao Xiang 23506a304cdSGao Xiang while (cur > iter->nr) { 23606a304cdSGao Xiang cur -= iter->nr; 23706a304cdSGao Xiang z_erofs_bvset_flip(iter); 23806a304cdSGao Xiang } 23906a304cdSGao Xiang iter->cur = cur; 24006a304cdSGao Xiang } 24106a304cdSGao Xiang 24206a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 24306a304cdSGao Xiang struct z_erofs_bvec *bvec, 24406a304cdSGao Xiang struct page **candidate_bvpage) 24506a304cdSGao Xiang { 24606a304cdSGao Xiang if (iter->cur == iter->nr) { 24706a304cdSGao Xiang if (!*candidate_bvpage) 24806a304cdSGao Xiang return -EAGAIN; 24906a304cdSGao Xiang 25006a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 25106a304cdSGao Xiang iter->bvset->nextpage = *candidate_bvpage; 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 else 3733fffb589SSandeep Dhavale sched_set_normal(worker->task, 0); 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 * The current pclusters was the tail of an exist chain, in addition 506db166fc2SGao Xiang * that the previous processed chained pclusters are all decided to 50747e4937aSGao Xiang * be hooked up to it. 508db166fc2SGao Xiang * A new chain will be created for the remaining pclusters which are 509db166fc2SGao Xiang * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED, 510db166fc2SGao Xiang * the next pcluster cannot reuse the whole page safely for inplace I/O 511db166fc2SGao Xiang * in the following scenario: 51247e4937aSGao Xiang * ________________________________________________________________ 51347e4937aSGao Xiang * | tail (partial) page | head (partial) page | 514db166fc2SGao Xiang * | (belongs to the next pcl) | (belongs to the current pcl) | 515db166fc2SGao Xiang * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________| 51647e4937aSGao Xiang */ 517db166fc2SGao Xiang Z_EROFS_PCLUSTER_HOOKED, 5180b964600SGao Xiang /* 519db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 5200b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 5210b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 522387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 5230b964600SGao Xiang */ 524db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 52547e4937aSGao Xiang /* 52647e4937aSGao Xiang * The current collection has been linked with the owned chain, and 52747e4937aSGao Xiang * could also be linked with the remaining collections, which means 52847e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 52947e4937aSGao Xiang * the current collection can safely use the whole page (since 53047e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 53147e4937aSGao Xiang * illustrated below: 53247e4937aSGao Xiang * ________________________________________________________________ 53347e4937aSGao Xiang * | tail (partial) page | head (partial) page | 53447e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 535db166fc2SGao Xiang * | PCLUSTER_FOLLOWED or | | 536db166fc2SGao Xiang * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________| 53747e4937aSGao Xiang * 53847e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 53947e4937aSGao Xiang */ 540db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 54147e4937aSGao Xiang }; 54247e4937aSGao Xiang 5435c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 5445c6dcc57SGao Xiang struct inode *const inode; 5455c6dcc57SGao Xiang struct erofs_map_blocks map; 54606a304cdSGao Xiang struct z_erofs_bvec_iter biter; 54747e4937aSGao Xiang 54806a304cdSGao Xiang struct page *candidate_bvpage; 54947e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 55047e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 551db166fc2SGao Xiang enum z_erofs_pclustermode mode; 55247e4937aSGao Xiang 5536ea5aad3SGao Xiang bool readahead; 55447e4937aSGao Xiang /* used for applying cache strategy on the fly */ 55547e4937aSGao Xiang bool backmost; 55647e4937aSGao Xiang erofs_off_t headoffset; 557ed722fbcSGao Xiang 558ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 559ed722fbcSGao Xiang unsigned int icur; 56047e4937aSGao Xiang }; 56147e4937aSGao Xiang 56247e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 5635c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 564db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 56547e4937aSGao Xiang 5661282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 5671282dea3SGao Xiang { 5681282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 5691282dea3SGao Xiang 5701282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 5711282dea3SGao Xiang return false; 5721282dea3SGao Xiang 5731282dea3SGao Xiang if (fe->backmost) 5741282dea3SGao Xiang return true; 5751282dea3SGao Xiang 5761282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 5771282dea3SGao Xiang fe->map.m_la < fe->headoffset) 5781282dea3SGao Xiang return true; 5791282dea3SGao Xiang 5801282dea3SGao Xiang return false; 5811282dea3SGao Xiang } 5821282dea3SGao Xiang 5836f39d1e1SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 584eaa9172aSGao Xiang struct page **pagepool) 58547e4937aSGao Xiang { 5866f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 5875c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 5881282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 58947e4937aSGao Xiang bool standalone = true; 5906f39d1e1SGao Xiang /* 5916f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 5926f39d1e1SGao Xiang * can be used if low memory otherwise. 5936f39d1e1SGao Xiang */ 5941825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 5951825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 596ed722fbcSGao Xiang unsigned int i; 59747e4937aSGao Xiang 598db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 59947e4937aSGao Xiang return; 60047e4937aSGao Xiang 601ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 60247e4937aSGao Xiang struct page *page; 603b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 6041825c8d7SGao Xiang struct page *newpage = NULL; 60547e4937aSGao Xiang 60647e4937aSGao Xiang /* the compressed page was loaded before */ 607ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 60847e4937aSGao Xiang continue; 60947e4937aSGao Xiang 610ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 61147e4937aSGao Xiang 61247e4937aSGao Xiang if (page) { 613b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 6140b964600SGao Xiang } else { 6150b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 6160b964600SGao Xiang standalone = false; 6171282dea3SGao Xiang if (!shouldalloc) 6181282dea3SGao Xiang continue; 6191282dea3SGao Xiang 6201282dea3SGao Xiang /* 6211282dea3SGao Xiang * try to use cached I/O if page allocation 6221282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 6231282dea3SGao Xiang * to avoid any direct reclaim. 6241282dea3SGao Xiang */ 6251825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 6261825c8d7SGao Xiang if (!newpage) 62747e4937aSGao Xiang continue; 6281282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 629b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 63047e4937aSGao Xiang } 63147e4937aSGao Xiang 632b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 63347e4937aSGao Xiang continue; 63447e4937aSGao Xiang 635eaa9172aSGao Xiang if (page) 63647e4937aSGao Xiang put_page(page); 637eaa9172aSGao Xiang else if (newpage) 638eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 63947e4937aSGao Xiang } 64047e4937aSGao Xiang 6410b964600SGao Xiang /* 6420b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 6430b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 6440b964600SGao Xiang */ 6450b964600SGao Xiang if (standalone) 646db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 64747e4937aSGao Xiang } 64847e4937aSGao Xiang 64947e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 65047e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 65147e4937aSGao Xiang struct erofs_workgroup *grp) 65247e4937aSGao Xiang { 65347e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 65447e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 65547e4937aSGao Xiang int i; 65647e4937aSGao Xiang 657cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 65847e4937aSGao Xiang /* 65947e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 66047e4937aSGao Xiang * therefore no need to worry about available decompression users. 66147e4937aSGao Xiang */ 6629f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 663ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 66447e4937aSGao Xiang 66547e4937aSGao Xiang if (!page) 66647e4937aSGao Xiang continue; 66747e4937aSGao Xiang 66847e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 66947e4937aSGao Xiang if (!trylock_page(page)) 67047e4937aSGao Xiang return -EBUSY; 67147e4937aSGao Xiang 672f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 67347e4937aSGao Xiang continue; 67447e4937aSGao Xiang 67547e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 676ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6776aaa7b06SGao Xiang detach_page_private(page); 67847e4937aSGao Xiang unlock_page(page); 67947e4937aSGao Xiang } 68047e4937aSGao Xiang return 0; 68147e4937aSGao Xiang } 68247e4937aSGao Xiang 683d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 68447e4937aSGao Xiang { 68547e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 686ed722fbcSGao Xiang int ret, i; 68747e4937aSGao Xiang 688ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 689ed722fbcSGao Xiang return 0; 69047e4937aSGao Xiang 691ed722fbcSGao Xiang ret = 0; 692cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 6939f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 694ed722fbcSGao Xiang if (pcl->compressed_bvecs[i].page == page) { 695ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 69647e4937aSGao Xiang ret = 1; 69747e4937aSGao Xiang break; 69847e4937aSGao Xiang } 69947e4937aSGao Xiang } 70047e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 7016aaa7b06SGao Xiang if (ret) 7026aaa7b06SGao Xiang detach_page_private(page); 70347e4937aSGao Xiang return ret; 70447e4937aSGao Xiang } 70547e4937aSGao Xiang 7065c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 707ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 70847e4937aSGao Xiang { 7095c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 71047e4937aSGao Xiang 711ed722fbcSGao Xiang while (fe->icur > 0) { 712ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 713ed722fbcSGao Xiang NULL, bvec->page)) { 714ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 71547e4937aSGao Xiang return true; 716ed722fbcSGao Xiang } 717ed722fbcSGao Xiang } 71847e4937aSGao Xiang return false; 71947e4937aSGao Xiang } 72047e4937aSGao Xiang 72187ca34a7SGao Xiang /* callers must be with pcluster lock held */ 7225c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 7235b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 72447e4937aSGao Xiang { 72547e4937aSGao Xiang int ret; 72647e4937aSGao Xiang 727db166fc2SGao Xiang if (exclusive) { 72806a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 729ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 73047e4937aSGao Xiang return 0; 73106a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 732db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 73306a304cdSGao Xiang !fe->candidate_bvpage) 73406a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 73506a304cdSGao Xiang } 73606a304cdSGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 73706a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 73806a304cdSGao Xiang return ret; 73947e4937aSGao Xiang } 74047e4937aSGao Xiang 7415c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 74247e4937aSGao Xiang { 7435c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 7445c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 74547e4937aSGao Xiang 746473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 747473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 748473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 74947e4937aSGao Xiang *owned_head = &pcl->next; 750473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 751db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 752473e15b0SGao Xiang return; 753473e15b0SGao Xiang } 754473e15b0SGao Xiang 75547e4937aSGao Xiang /* 756473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 757473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 75847e4937aSGao Xiang */ 759267f2492SGao Xiang if (*owned_head != &pcl->next && pcl != f->tailpcl && 760267f2492SGao Xiang cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 761473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 76247e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 763db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_HOOKED; 7645c6dcc57SGao Xiang f->tailpcl = NULL; 765473e15b0SGao Xiang return; 76647e4937aSGao Xiang } 767473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 768db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 76947e4937aSGao Xiang } 77047e4937aSGao Xiang 77183a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 77247e4937aSGao Xiang { 77383a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 774cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 77547e4937aSGao Xiang struct z_erofs_pcluster *pcl; 77664094a04SGao Xiang struct erofs_workgroup *grp; 77747e4937aSGao Xiang int err; 77847e4937aSGao Xiang 779c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 780c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 7818f899262SGao Xiang DBG_BUGON(1); 7828f899262SGao Xiang return -EFSCORRUPTED; 7838f899262SGao Xiang } 7848f899262SGao Xiang 7859f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 786cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 787cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 7889f6cc76eSGao Xiang if (IS_ERR(pcl)) 7899f6cc76eSGao Xiang return PTR_ERR(pcl); 79047e4937aSGao Xiang 79164094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 7928f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 7932bfab9c0SGao Xiang pcl->length = 0; 7942bfab9c0SGao Xiang pcl->partial = true; 79547e4937aSGao Xiang 79647e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 7975c6dcc57SGao Xiang pcl->next = fe->owned_head; 79887ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 799db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 80047e4937aSGao Xiang 80147e4937aSGao Xiang /* 80247e4937aSGao Xiang * lock all primary followed works before visible to others 80347e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 80447e4937aSGao Xiang */ 80587ca34a7SGao Xiang mutex_init(&pcl->lock); 80687ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 80747e4937aSGao Xiang 808cecf864dSYue Hu if (ztailpacking) { 809cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 810*3acea5fcSJingbo Xu pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 811cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 812cecf864dSYue Hu } else { 813cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 814cecf864dSYue Hu 81583a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 81664094a04SGao Xiang if (IS_ERR(grp)) { 81764094a04SGao Xiang err = PTR_ERR(grp); 81864094a04SGao Xiang goto err_out; 81964094a04SGao Xiang } 82064094a04SGao Xiang 82164094a04SGao Xiang if (grp != &pcl->obj) { 8225c6dcc57SGao Xiang fe->pcl = container_of(grp, 823cecf864dSYue Hu struct z_erofs_pcluster, obj); 82464094a04SGao Xiang err = -EEXIST; 82564094a04SGao Xiang goto err_out; 82647e4937aSGao Xiang } 827cecf864dSYue Hu } 82847e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 8295c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 8305c6dcc57SGao Xiang fe->tailpcl = pcl; 8315c6dcc57SGao Xiang fe->owned_head = &pcl->next; 8325c6dcc57SGao Xiang fe->pcl = pcl; 8339e579fc1SGao Xiang return 0; 83464094a04SGao Xiang 83564094a04SGao Xiang err_out: 83687ca34a7SGao Xiang mutex_unlock(&pcl->lock); 8379f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 83864094a04SGao Xiang return err; 83947e4937aSGao Xiang } 84047e4937aSGao Xiang 84183a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 84247e4937aSGao Xiang { 84383a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 8440d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 8459e579fc1SGao Xiang int ret; 84647e4937aSGao Xiang 84787ca34a7SGao Xiang DBG_BUGON(fe->pcl); 84847e4937aSGao Xiang 84987ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 8505c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 8515c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 85247e4937aSGao Xiang 8530d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 8540d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 8550d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 8560d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 85747e4937aSGao Xiang DBG_BUGON(1); 858cecf864dSYue Hu return -EFSCORRUPTED; 859cecf864dSYue Hu } 86047e4937aSGao Xiang 86164094a04SGao Xiang if (grp) { 8625c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 8630d823b42SGao Xiang ret = -EEXIST; 86464094a04SGao Xiang } else { 86583a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 86664094a04SGao Xiang } 86747e4937aSGao Xiang 8680d823b42SGao Xiang if (ret == -EEXIST) { 869267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 870267f2492SGao Xiang /* used to check tail merging loop due to corrupted images */ 871267f2492SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 872267f2492SGao Xiang fe->tailpcl = fe->pcl; 873267f2492SGao Xiang 874267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 8750d823b42SGao Xiang } else if (ret) { 8760d823b42SGao Xiang return ret; 8770d823b42SGao Xiang } 87806a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 879387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 88081382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 881ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 88247e4937aSGao Xiang return 0; 88347e4937aSGao Xiang } 88447e4937aSGao Xiang 88547e4937aSGao Xiang /* 88647e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 88747e4937aSGao Xiang * only after a RCU grace period. 88847e4937aSGao Xiang */ 88947e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 89047e4937aSGao Xiang { 89187ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 89287ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 89347e4937aSGao Xiang } 89447e4937aSGao Xiang 89547e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 89647e4937aSGao Xiang { 89747e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 89847e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 89947e4937aSGao Xiang 90087ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 90147e4937aSGao Xiang } 90247e4937aSGao Xiang 9035c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 90447e4937aSGao Xiang { 90587ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 90647e4937aSGao Xiang 90787ca34a7SGao Xiang if (!pcl) 90847e4937aSGao Xiang return false; 90947e4937aSGao Xiang 91006a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 91187ca34a7SGao Xiang mutex_unlock(&pcl->lock); 91247e4937aSGao Xiang 91306a304cdSGao Xiang if (fe->candidate_bvpage) { 91406a304cdSGao Xiang DBG_BUGON(z_erofs_is_shortlived_page(fe->candidate_bvpage)); 91506a304cdSGao Xiang fe->candidate_bvpage = NULL; 91606a304cdSGao Xiang } 91706a304cdSGao Xiang 91847e4937aSGao Xiang /* 91947e4937aSGao Xiang * if all pending pages are added, don't hold its reference 92047e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 92147e4937aSGao Xiang */ 922db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 92387ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 92447e4937aSGao Xiang 92587ca34a7SGao Xiang fe->pcl = NULL; 92647e4937aSGao Xiang return true; 92747e4937aSGao Xiang } 92847e4937aSGao Xiang 929b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 930b15b2e30SYue Hu struct page *page, unsigned int pageofs, 931b15b2e30SYue Hu unsigned int len) 932b15b2e30SYue Hu { 933*3acea5fcSJingbo Xu struct super_block *sb = inode->i_sb; 934b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 935b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 936b15b2e30SYue Hu u8 *src, *dst; 937b15b2e30SYue Hu unsigned int i, cnt; 938b15b2e30SYue Hu 939e5126de1SYue Hu if (!packed_inode) 940e5126de1SYue Hu return -EFSCORRUPTED; 941e5126de1SYue Hu 942b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 943b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 944b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 945*3acea5fcSJingbo Xu sb->s_blocksize - erofs_blkoff(sb, pos)); 946b15b2e30SYue Hu src = erofs_bread(&buf, packed_inode, 947*3acea5fcSJingbo Xu erofs_blknr(sb, pos), EROFS_KMAP); 948b15b2e30SYue Hu if (IS_ERR(src)) { 949b15b2e30SYue Hu erofs_put_metabuf(&buf); 950b15b2e30SYue Hu return PTR_ERR(src); 951b15b2e30SYue Hu } 952b15b2e30SYue Hu 953b15b2e30SYue Hu dst = kmap_local_page(page); 954*3acea5fcSJingbo Xu memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 955b15b2e30SYue Hu kunmap_local(dst); 956b15b2e30SYue Hu pos += cnt; 957b15b2e30SYue Hu } 958b15b2e30SYue Hu erofs_put_metabuf(&buf); 959b15b2e30SYue Hu return 0; 960b15b2e30SYue Hu } 961b15b2e30SYue Hu 96247e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 963eaa9172aSGao Xiang struct page *page, struct page **pagepool) 96447e4937aSGao Xiang { 96547e4937aSGao Xiang struct inode *const inode = fe->inode; 96647e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 96747e4937aSGao Xiang const loff_t offset = page_offset(page); 9685b220b20SGao Xiang bool tight = true, exclusive; 9692bfab9c0SGao Xiang unsigned int cur, end, spiltted; 97047e4937aSGao Xiang int err = 0; 97147e4937aSGao Xiang 97247e4937aSGao Xiang /* register locked file pages as online pages in pack */ 97347e4937aSGao Xiang z_erofs_onlinepage_init(page); 97447e4937aSGao Xiang 97547e4937aSGao Xiang spiltted = 0; 97647e4937aSGao Xiang end = PAGE_SIZE; 97747e4937aSGao Xiang repeat: 97847e4937aSGao Xiang cur = end - 1; 97947e4937aSGao Xiang 98039397a46SGao Xiang if (offset + cur < map->m_la || 98139397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 98239397a46SGao Xiang erofs_dbg("out-of-range map @ pos %llu", offset + cur); 98347e4937aSGao Xiang 9845c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 98547e4937aSGao Xiang fe->backmost = false; 98647e4937aSGao Xiang map->m_la = offset + cur; 98747e4937aSGao Xiang map->m_llen = 0; 98847e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 9898d8a09b0SGao Xiang if (err) 99067148551SGao Xiang goto out; 99139397a46SGao Xiang } else { 99239397a46SGao Xiang if (fe->pcl) 99339397a46SGao Xiang goto hitted; 99439397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 99539397a46SGao Xiang } 99647e4937aSGao Xiang 997b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 998b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 99947e4937aSGao Xiang goto hitted; 100047e4937aSGao Xiang 100183a386c0SGao Xiang err = z_erofs_collector_begin(fe); 10028d8a09b0SGao Xiang if (err) 100367148551SGao Xiang goto out; 100447e4937aSGao Xiang 10055c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 100609c54379SGao Xiang void *mp; 1007cecf864dSYue Hu 100809c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 1009*3acea5fcSJingbo Xu erofs_blknr(inode->i_sb, map->m_pa), 1010*3acea5fcSJingbo Xu EROFS_NO_KMAP); 101109c54379SGao Xiang if (IS_ERR(mp)) { 101209c54379SGao Xiang err = PTR_ERR(mp); 1013cecf864dSYue Hu erofs_err(inode->i_sb, 1014cecf864dSYue Hu "failed to get inline page, err %d", err); 101567148551SGao Xiang goto out; 1016cecf864dSYue Hu } 101709c54379SGao Xiang get_page(fe->map.buf.page); 1018ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 1019ed722fbcSGao Xiang fe->map.buf.page); 1020db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 1021cecf864dSYue Hu } else { 10226f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 10231282dea3SGao Xiang z_erofs_bind_cache(fe, pagepool); 1024cecf864dSYue Hu } 102547e4937aSGao Xiang hitted: 1026dc76ea8cSGao Xiang /* 1027dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 1028dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 1029dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 1030387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 1031dc76ea8cSGao Xiang */ 1032db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED && 1033db166fc2SGao Xiang fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1034dc76ea8cSGao Xiang 103547e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 10368d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 103747e4937aSGao Xiang zero_user_segment(page, cur, end); 103847e4937aSGao Xiang goto next_part; 103947e4937aSGao Xiang } 1040b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 1041b15b2e30SYue Hu unsigned int pageofs, skip, len; 1042b15b2e30SYue Hu 1043b15b2e30SYue Hu if (offset > map->m_la) { 1044b15b2e30SYue Hu pageofs = 0; 1045b15b2e30SYue Hu skip = offset - map->m_la; 1046b15b2e30SYue Hu } else { 1047b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 1048b15b2e30SYue Hu skip = 0; 1049b15b2e30SYue Hu } 1050b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 1051b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1052b15b2e30SYue Hu if (err) 1053b15b2e30SYue Hu goto out; 1054b15b2e30SYue Hu ++spiltted; 1055b15b2e30SYue Hu tight = false; 1056b15b2e30SYue Hu goto next_part; 1057b15b2e30SYue Hu } 105847e4937aSGao Xiang 10595b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 106047e4937aSGao Xiang if (cur) 1061db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 106247e4937aSGao Xiang 106347e4937aSGao Xiang retry: 106406a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 106506a304cdSGao Xiang .page = page, 106606a304cdSGao Xiang .offset = offset - map->m_la, 106706a304cdSGao Xiang .end = end, 10685b220b20SGao Xiang }), exclusive); 106906a304cdSGao Xiang /* should allocate an additional short-lived page for bvset */ 107006a304cdSGao Xiang if (err == -EAGAIN && !fe->candidate_bvpage) { 107106a304cdSGao Xiang fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL); 107206a304cdSGao Xiang set_page_private(fe->candidate_bvpage, 107306a304cdSGao Xiang Z_EROFS_SHORTLIVED_PAGE); 107447e4937aSGao Xiang goto retry; 107547e4937aSGao Xiang } 107647e4937aSGao Xiang 107706a304cdSGao Xiang if (err) { 107806a304cdSGao Xiang DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage); 107967148551SGao Xiang goto out; 108006a304cdSGao Xiang } 108147e4937aSGao Xiang 108267148551SGao Xiang z_erofs_onlinepage_split(page); 108347e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 108447e4937aSGao Xiang ++spiltted; 1085267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1086267f2492SGao Xiang fe->pcl->multibases = true; 10872bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 10882bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 10892bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 10902bfab9c0SGao Xiang } 1091e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1092e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1093e7933278SGao Xiang fe->pcl->length == map->m_llen) 1094e7933278SGao Xiang fe->pcl->partial = false; 109547e4937aSGao Xiang next_part: 10962bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 109747e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 10982bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 109947e4937aSGao Xiang 110047e4937aSGao Xiang end = cur; 110147e4937aSGao Xiang if (end > 0) 110247e4937aSGao Xiang goto repeat; 110347e4937aSGao Xiang 110447e4937aSGao Xiang out: 110567148551SGao Xiang if (err) 110667148551SGao Xiang z_erofs_page_mark_eio(page); 110747e4937aSGao Xiang z_erofs_onlinepage_endio(page); 110847e4937aSGao Xiang 11094f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 111047e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 111147e4937aSGao Xiang return err; 111247e4937aSGao Xiang } 111347e4937aSGao Xiang 111440452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 111540452ffcSHuang Jianan unsigned int readahead_pages) 111640452ffcSHuang Jianan { 1117a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 111840452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 111940452ffcSHuang Jianan !readahead_pages) 112040452ffcSHuang Jianan return true; 112140452ffcSHuang Jianan 112240452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 112340452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 112440452ffcSHuang Jianan return true; 112540452ffcSHuang Jianan 112640452ffcSHuang Jianan return false; 112740452ffcSHuang Jianan } 112840452ffcSHuang Jianan 11296aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 11306aaa7b06SGao Xiang { 11316aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 11326aaa7b06SGao Xiang } 11336aaa7b06SGao Xiang 11344f05687fSGao Xiang struct z_erofs_decompress_backend { 11354f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 11364f05687fSGao Xiang struct super_block *sb; 11374f05687fSGao Xiang struct z_erofs_pcluster *pcl; 11384f05687fSGao Xiang 11394f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 11404f05687fSGao Xiang struct page **decompressed_pages; 11414f05687fSGao Xiang /* pages to keep the compressed data */ 11424f05687fSGao Xiang struct page **compressed_pages; 11434f05687fSGao Xiang 1144267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 11454f05687fSGao Xiang struct page **pagepool; 11462bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 11474f05687fSGao Xiang }; 11484f05687fSGao Xiang 1149267f2492SGao Xiang struct z_erofs_bvec_item { 1150267f2492SGao Xiang struct z_erofs_bvec bvec; 1151267f2492SGao Xiang struct list_head list; 1152267f2492SGao Xiang }; 1153267f2492SGao Xiang 1154267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 11553fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 11563fe96ee0SGao Xiang { 1157267f2492SGao Xiang struct z_erofs_bvec_item *item; 1158267f2492SGao Xiang 1159267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1160267f2492SGao Xiang unsigned int pgnr; 11613fe96ee0SGao Xiang 1162267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 11632bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 116463bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 11653fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1166267f2492SGao Xiang return; 11673fe96ee0SGao Xiang } 116863bbb856SGao Xiang } 11693fe96ee0SGao Xiang 1170267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1171267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1172267f2492SGao Xiang item->bvec = *bvec; 1173267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1174267f2492SGao Xiang } 1175267f2492SGao Xiang 1176267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1177267f2492SGao Xiang int err) 1178267f2492SGao Xiang { 1179267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1180267f2492SGao Xiang struct list_head *p, *n; 1181267f2492SGao Xiang 1182267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1183267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1184267f2492SGao Xiang unsigned int end, cur; 1185267f2492SGao Xiang void *dst, *src; 1186267f2492SGao Xiang 1187267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1188267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1189267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1190267f2492SGao Xiang bvi->bvec.end); 1191267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1192267f2492SGao Xiang while (cur < end) { 1193267f2492SGao Xiang unsigned int pgnr, scur, len; 1194267f2492SGao Xiang 1195267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1196267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1197267f2492SGao Xiang 1198267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1199267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1200267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1201267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1202267f2492SGao Xiang err = -EFSCORRUPTED; 1203267f2492SGao Xiang cur += len; 1204267f2492SGao Xiang continue; 1205267f2492SGao Xiang } 1206267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1207267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1208267f2492SGao Xiang kunmap_local(src); 1209267f2492SGao Xiang cur += len; 1210267f2492SGao Xiang } 1211267f2492SGao Xiang kunmap_local(dst); 1212267f2492SGao Xiang if (err) 1213267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 1214267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 1215267f2492SGao Xiang list_del(p); 1216267f2492SGao Xiang kfree(bvi); 1217267f2492SGao Xiang } 1218267f2492SGao Xiang } 1219267f2492SGao Xiang 1220267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 122142fec235SGao Xiang { 12224f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 122306a304cdSGao Xiang struct z_erofs_bvec_iter biter; 122406a304cdSGao Xiang struct page *old_bvpage; 1225267f2492SGao Xiang int i; 122642fec235SGao Xiang 1227387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 122842fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 122906a304cdSGao Xiang struct z_erofs_bvec bvec; 123042fec235SGao Xiang 123106a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 123242fec235SGao Xiang 123306a304cdSGao Xiang if (old_bvpage) 12344f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 123542fec235SGao Xiang 123606a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1237267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 123842fec235SGao Xiang } 123906a304cdSGao Xiang 124006a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 124106a304cdSGao Xiang if (old_bvpage) 12424f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 124342fec235SGao Xiang } 124442fec235SGao Xiang 12454f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 12464f05687fSGao Xiang bool *overlapped) 124767139e36SGao Xiang { 12484f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 124967139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 125067139e36SGao Xiang int i, err = 0; 125167139e36SGao Xiang 125267139e36SGao Xiang *overlapped = false; 125367139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1254ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1255ed722fbcSGao Xiang struct page *page = bvec->page; 125667139e36SGao Xiang 125767139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 125867139e36SGao Xiang if (!page) { 125967139e36SGao Xiang DBG_BUGON(1); 126067139e36SGao Xiang continue; 126167139e36SGao Xiang } 1262fe3e5914SGao Xiang be->compressed_pages[i] = page; 126367139e36SGao Xiang 126467139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 126567139e36SGao Xiang if (!PageUptodate(page)) 126667139e36SGao Xiang err = -EIO; 126767139e36SGao Xiang continue; 126867139e36SGao Xiang } 126967139e36SGao Xiang 127067139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 127167139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 12724f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 127367139e36SGao Xiang if (!PageUptodate(page)) 127467139e36SGao Xiang err = -EIO; 127567139e36SGao Xiang continue; 127667139e36SGao Xiang } 1277267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 127867139e36SGao Xiang *overlapped = true; 127967139e36SGao Xiang } 128067139e36SGao Xiang } 128167139e36SGao Xiang 1282fe3e5914SGao Xiang if (err) 12834f05687fSGao Xiang return err; 12844f05687fSGao Xiang return 0; 128567139e36SGao Xiang } 128667139e36SGao Xiang 12874f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 12884f05687fSGao Xiang int err) 128947e4937aSGao Xiang { 12904f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 12914f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1292cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 12932bfab9c0SGao Xiang unsigned int i, inputsize; 129467148551SGao Xiang int err2; 12952bfab9c0SGao Xiang struct page *page; 12962bfab9c0SGao Xiang bool overlapped; 129747e4937aSGao Xiang 129887ca34a7SGao Xiang mutex_lock(&pcl->lock); 12992bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 130047e4937aSGao Xiang 1301fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1302fe3e5914SGao Xiang be->decompressed_pages = NULL; 1303fe3e5914SGao Xiang be->compressed_pages = NULL; 1304fe3e5914SGao Xiang be->onstack_used = 0; 13052bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 13064f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 13072bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 13084f05687fSGao Xiang memset(be->decompressed_pages, 0, 13092bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1310fe3e5914SGao Xiang } 1311fe3e5914SGao Xiang 1312fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1313fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1314fe3e5914SGao Xiang 1315fe3e5914SGao Xiang if (!be->decompressed_pages) 13164f05687fSGao Xiang be->decompressed_pages = 1317647dd2c3SGao Xiang kvcalloc(be->nr_pages, sizeof(struct page *), 1318e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1319fe3e5914SGao Xiang if (!be->compressed_pages) 1320fe3e5914SGao Xiang be->compressed_pages = 1321647dd2c3SGao Xiang kvcalloc(pclusterpages, sizeof(struct page *), 1322fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 132347e4937aSGao Xiang 1324267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 13254f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 13264f05687fSGao Xiang if (err2) 13274f05687fSGao Xiang err = err2; 13288d8a09b0SGao Xiang if (err) 132947e4937aSGao Xiang goto out; 133047e4937aSGao Xiang 1331cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1332cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1333cecf864dSYue Hu else 1334cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1335cecf864dSYue Hu 133647e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 13374f05687fSGao Xiang .sb = be->sb, 13384f05687fSGao Xiang .in = be->compressed_pages, 13394f05687fSGao Xiang .out = be->decompressed_pages, 1340cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 134187ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 13429f6cc76eSGao Xiang .inputsize = inputsize, 13432bfab9c0SGao Xiang .outputsize = pcl->length, 134447e4937aSGao Xiang .alg = pcl->algorithmformat, 134547e4937aSGao Xiang .inplace_io = overlapped, 13462bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1347267f2492SGao Xiang .fillgaps = pcl->multibases, 13484f05687fSGao Xiang }, be->pagepool); 134947e4937aSGao Xiang 135047e4937aSGao Xiang out: 1351cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1352cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1353ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1354ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1355cecf864dSYue Hu put_page(page); 1356cecf864dSYue Hu } else { 1357cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1358ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 135947e4937aSGao Xiang 136047e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 136147e4937aSGao Xiang continue; 136247e4937aSGao Xiang 13636aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13644f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1365ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 136647e4937aSGao Xiang } 1367cecf864dSYue Hu } 1368fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1369fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1370647dd2c3SGao Xiang kvfree(be->compressed_pages); 1371267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 137247e4937aSGao Xiang 13732bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 13744f05687fSGao Xiang page = be->decompressed_pages[i]; 137547e4937aSGao Xiang if (!page) 137647e4937aSGao Xiang continue; 137747e4937aSGao Xiang 13786aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 137947e4937aSGao Xiang 13806aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13814f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 138247e4937aSGao Xiang continue; 138367148551SGao Xiang if (err) 138467148551SGao Xiang z_erofs_page_mark_eio(page); 138547e4937aSGao Xiang z_erofs_onlinepage_endio(page); 138647e4937aSGao Xiang } 138747e4937aSGao Xiang 13884f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 1389647dd2c3SGao Xiang kvfree(be->decompressed_pages); 139047e4937aSGao Xiang 13912bfab9c0SGao Xiang pcl->length = 0; 13922bfab9c0SGao Xiang pcl->partial = true; 1393267f2492SGao Xiang pcl->multibases = false; 139406a304cdSGao Xiang pcl->bvset.nextpage = NULL; 139587ca34a7SGao Xiang pcl->vcnt = 0; 139647e4937aSGao Xiang 139787ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 139847e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 139987ca34a7SGao Xiang mutex_unlock(&pcl->lock); 140047e4937aSGao Xiang return err; 140147e4937aSGao Xiang } 140247e4937aSGao Xiang 14030c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1404eaa9172aSGao Xiang struct page **pagepool) 140547e4937aSGao Xiang { 14064f05687fSGao Xiang struct z_erofs_decompress_backend be = { 14074f05687fSGao Xiang .sb = io->sb, 14084f05687fSGao Xiang .pagepool = pagepool, 1409267f2492SGao Xiang .decompressed_secondary_bvecs = 1410267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 14114f05687fSGao Xiang }; 141247e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 141347e4937aSGao Xiang 141447e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 14154f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 141647e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 14174f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */ 141847e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 141947e4937aSGao Xiang 14204f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 14214f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 142247e4937aSGao Xiang 14234f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 14244f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 142547e4937aSGao Xiang } 142647e4937aSGao Xiang } 142747e4937aSGao Xiang 14280c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 142947e4937aSGao Xiang { 1430a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1431a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1432eaa9172aSGao Xiang struct page *pagepool = NULL; 143347e4937aSGao Xiang 1434a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 14350c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 1436eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1437a4b1fab1SGao Xiang kvfree(bgq); 143847e4937aSGao Xiang } 143947e4937aSGao Xiang 14403fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14413fffb589SSandeep Dhavale static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 14423fffb589SSandeep Dhavale { 14433fffb589SSandeep Dhavale z_erofs_decompressqueue_work((struct work_struct *)work); 14443fffb589SSandeep Dhavale } 14453fffb589SSandeep Dhavale #endif 14463fffb589SSandeep Dhavale 14477865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1448cdba5506SGao Xiang int bios) 14497865827cSGao Xiang { 14507865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 14517865827cSGao Xiang 14527865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1453cdba5506SGao Xiang if (io->sync) { 14547865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 145560b30050SHongyu Jin complete(&io->u.done); 14567865827cSGao Xiang return; 14577865827cSGao Xiang } 14587865827cSGao Xiang 14597865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 14607865827cSGao Xiang return; 14613fffb589SSandeep Dhavale /* Use (kthread_)work and sync decompression for atomic contexts only */ 14627865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 14633fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14643fffb589SSandeep Dhavale struct kthread_worker *worker; 14653fffb589SSandeep Dhavale 14663fffb589SSandeep Dhavale rcu_read_lock(); 14673fffb589SSandeep Dhavale worker = rcu_dereference( 14683fffb589SSandeep Dhavale z_erofs_pcpu_workers[raw_smp_processor_id()]); 14693fffb589SSandeep Dhavale if (!worker) { 14703fffb589SSandeep Dhavale INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 14717865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 14723fffb589SSandeep Dhavale } else { 14733fffb589SSandeep Dhavale kthread_queue_work(worker, &io->u.kthread_work); 14743fffb589SSandeep Dhavale } 14753fffb589SSandeep Dhavale rcu_read_unlock(); 14763fffb589SSandeep Dhavale #else 14773fffb589SSandeep Dhavale queue_work(z_erofs_workqueue, &io->u.work); 14783fffb589SSandeep Dhavale #endif 14797865827cSGao Xiang /* enable sync decompression for readahead */ 14807865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 14817865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 14827865827cSGao Xiang return; 14837865827cSGao Xiang } 14847865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 14857865827cSGao Xiang } 14867865827cSGao Xiang 148747e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 148847e4937aSGao Xiang unsigned int nr, 1489eaa9172aSGao Xiang struct page **pagepool, 14909f2731d6SGao Xiang struct address_space *mc) 149147e4937aSGao Xiang { 149247e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 14939f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 149447e4937aSGao Xiang bool tocache = false; 149547e4937aSGao Xiang 149647e4937aSGao Xiang struct address_space *mapping; 149747e4937aSGao Xiang struct page *oldpage, *page; 149847e4937aSGao Xiang int justfound; 149947e4937aSGao Xiang 150047e4937aSGao Xiang repeat: 1501ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 150247e4937aSGao Xiang oldpage = page; 150347e4937aSGao Xiang 150447e4937aSGao Xiang if (!page) 150547e4937aSGao Xiang goto out_allocpage; 150647e4937aSGao Xiang 1507b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1508b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 150947e4937aSGao Xiang 15101825c8d7SGao Xiang /* 15111825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 15121825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 15131825c8d7SGao Xiang */ 15141825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1515ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 15161825c8d7SGao Xiang set_page_private(page, 0); 15171825c8d7SGao Xiang tocache = true; 15181825c8d7SGao Xiang goto out_tocache; 15191825c8d7SGao Xiang } 152047e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 152147e4937aSGao Xiang 152247e4937aSGao Xiang /* 15236aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 152447e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 152547e4937aSGao Xiang */ 152647e4937aSGao Xiang if (mapping && mapping != mc) 152747e4937aSGao Xiang /* ought to be unmanaged pages */ 152847e4937aSGao Xiang goto out; 152947e4937aSGao Xiang 15306aaa7b06SGao Xiang /* directly return for shortlived page as well */ 15316aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 15326aaa7b06SGao Xiang goto out; 15336aaa7b06SGao Xiang 153447e4937aSGao Xiang lock_page(page); 153547e4937aSGao Xiang 153647e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 153747e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 153847e4937aSGao Xiang 153947e4937aSGao Xiang /* the page is still in manage cache */ 154047e4937aSGao Xiang if (page->mapping == mc) { 1541ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 154247e4937aSGao Xiang 154347e4937aSGao Xiang if (!PagePrivate(page)) { 154447e4937aSGao Xiang /* 154547e4937aSGao Xiang * impossible to be !PagePrivate(page) for 154647e4937aSGao Xiang * the current restriction as well if 1547ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 154847e4937aSGao Xiang */ 154947e4937aSGao Xiang DBG_BUGON(!justfound); 155047e4937aSGao Xiang 155147e4937aSGao Xiang justfound = 0; 155247e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 155347e4937aSGao Xiang SetPagePrivate(page); 155447e4937aSGao Xiang } 155547e4937aSGao Xiang 155647e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 155747e4937aSGao Xiang if (PageUptodate(page)) { 155847e4937aSGao Xiang unlock_page(page); 155947e4937aSGao Xiang page = NULL; 156047e4937aSGao Xiang } 156147e4937aSGao Xiang goto out; 156247e4937aSGao Xiang } 156347e4937aSGao Xiang 156447e4937aSGao Xiang /* 156547e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 156647e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 156747e4937aSGao Xiang */ 156847e4937aSGao Xiang DBG_BUGON(page->mapping); 156947e4937aSGao Xiang DBG_BUGON(!justfound); 157047e4937aSGao Xiang 157147e4937aSGao Xiang tocache = true; 157247e4937aSGao Xiang unlock_page(page); 157347e4937aSGao Xiang put_page(page); 157447e4937aSGao Xiang out_allocpage: 15755ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1576ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1577ed722fbcSGao Xiang oldpage, page)) { 1578eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 15795ddcee1fSGao Xiang cond_resched(); 15805ddcee1fSGao Xiang goto repeat; 15815ddcee1fSGao Xiang } 15821825c8d7SGao Xiang out_tocache: 1583bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1584bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1585bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1586bf225074SGao Xiang goto out; 1587a30573b3SGao Xiang } 1588bf225074SGao Xiang attach_page_private(page, pcl); 1589bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1590bf225074SGao Xiang put_page(page); 1591bf225074SGao Xiang 159247e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 159347e4937aSGao Xiang return page; 159447e4937aSGao Xiang } 159547e4937aSGao Xiang 1596cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1597a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 159847e4937aSGao Xiang { 1599a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 160047e4937aSGao Xiang 1601a4b1fab1SGao Xiang if (fg && !*fg) { 1602a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1603a4b1fab1SGao Xiang if (!q) { 1604a4b1fab1SGao Xiang *fg = true; 1605a4b1fab1SGao Xiang goto fg_out; 160647e4937aSGao Xiang } 16073fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 16083fffb589SSandeep Dhavale kthread_init_work(&q->u.kthread_work, 16093fffb589SSandeep Dhavale z_erofs_decompressqueue_kthread_work); 16103fffb589SSandeep Dhavale #else 16110c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 16123fffb589SSandeep Dhavale #endif 1613a4b1fab1SGao Xiang } else { 1614a4b1fab1SGao Xiang fg_out: 1615a4b1fab1SGao Xiang q = fgq; 161660b30050SHongyu Jin init_completion(&fgq->u.done); 1617a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 161867148551SGao Xiang q->eio = false; 1619cdba5506SGao Xiang q->sync = true; 1620a4b1fab1SGao Xiang } 1621a4b1fab1SGao Xiang q->sb = sb; 1622a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1623a4b1fab1SGao Xiang return q; 162447e4937aSGao Xiang } 162547e4937aSGao Xiang 162647e4937aSGao Xiang /* define decompression jobqueue types */ 162747e4937aSGao Xiang enum { 162847e4937aSGao Xiang JQ_BYPASS, 162947e4937aSGao Xiang JQ_SUBMIT, 163047e4937aSGao Xiang NR_JOBQUEUES, 163147e4937aSGao Xiang }; 163247e4937aSGao Xiang 163347e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 163447e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 163547e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 163647e4937aSGao Xiang { 163747e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 163847e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 163947e4937aSGao Xiang 164047e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 164147e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 164247e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 164347e4937aSGao Xiang 164447e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 164547e4937aSGao Xiang 164647e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 164747e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 164847e4937aSGao Xiang 164947e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 165047e4937aSGao Xiang } 165147e4937aSGao Xiang 16527865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 16537865827cSGao Xiang { 1654cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 16557865827cSGao Xiang blk_status_t err = bio->bi_status; 16567865827cSGao Xiang struct bio_vec *bvec; 16577865827cSGao Xiang struct bvec_iter_all iter_all; 16587865827cSGao Xiang 16597865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 16607865827cSGao Xiang struct page *page = bvec->bv_page; 16617865827cSGao Xiang 16627865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 16637865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 16647865827cSGao Xiang 16657865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 16667865827cSGao Xiang if (!err) 16677865827cSGao Xiang SetPageUptodate(page); 16687865827cSGao Xiang unlock_page(page); 16697865827cSGao Xiang } 16707865827cSGao Xiang } 167167148551SGao Xiang if (err) 167267148551SGao Xiang q->eio = true; 1673cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 16747865827cSGao Xiang bio_put(bio); 16757865827cSGao Xiang } 16767865827cSGao Xiang 167783a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1678eaa9172aSGao Xiang struct page **pagepool, 1679a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1680a4b1fab1SGao Xiang bool *force_fg) 168147e4937aSGao Xiang { 168283a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 168383a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 168447e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1685a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 16865c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1687dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 16883f649ab7SKees Cook pgoff_t last_index; 1689dfeab2e9SGao Xiang struct block_device *last_bdev; 16901e4a2955SGao Xiang unsigned int nr_bios = 0; 16911e4a2955SGao Xiang struct bio *bio = NULL; 169282e60d00SJohannes Weiner unsigned long pflags; 169382e60d00SJohannes Weiner int memstall = 0; 169447e4937aSGao Xiang 1695cdba5506SGao Xiang /* 1696cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1697cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1698cdba5506SGao Xiang */ 1699cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1700cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1701cdba5506SGao Xiang 1702a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1703a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 170447e4937aSGao Xiang 170547e4937aSGao Xiang /* by default, all need io submission */ 170647e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 170747e4937aSGao Xiang 170847e4937aSGao Xiang do { 1709dfeab2e9SGao Xiang struct erofs_map_dev mdev; 171047e4937aSGao Xiang struct z_erofs_pcluster *pcl; 17111e4a2955SGao Xiang pgoff_t cur, end; 17121e4a2955SGao Xiang unsigned int i = 0; 17131e4a2955SGao Xiang bool bypass = true; 171447e4937aSGao Xiang 171547e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 171647e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 171747e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 171847e4937aSGao Xiang 171947e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 172047e4937aSGao Xiang 1721cecf864dSYue Hu /* close the main owned chain at first */ 1722cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1723cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1724cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1725cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1726cecf864dSYue Hu continue; 1727cecf864dSYue Hu } 1728cecf864dSYue Hu 1729dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1730dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1731*3acea5fcSJingbo Xu .m_pa = erofs_pos(sb, pcl->obj.index), 1732dfeab2e9SGao Xiang }; 1733dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1734dfeab2e9SGao Xiang 1735*3acea5fcSJingbo Xu cur = erofs_blknr(sb, mdev.m_pa); 17369f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 173747e4937aSGao Xiang 17381e4a2955SGao Xiang do { 17391e4a2955SGao Xiang struct page *page; 174047e4937aSGao Xiang 17411e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 174283a386c0SGao Xiang mc); 17431e4a2955SGao Xiang if (!page) 17441e4a2955SGao Xiang continue; 174547e4937aSGao Xiang 1746dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1747dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 174847e4937aSGao Xiang submit_bio_retry: 174994e4e153SGao Xiang submit_bio(bio); 175082e60d00SJohannes Weiner if (memstall) { 175182e60d00SJohannes Weiner psi_memstall_leave(&pflags); 175282e60d00SJohannes Weiner memstall = 0; 175382e60d00SJohannes Weiner } 175447e4937aSGao Xiang bio = NULL; 175547e4937aSGao Xiang } 175647e4937aSGao Xiang 175782e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 175899486c51SChristoph Hellwig psi_memstall_enter(&pflags); 175982e60d00SJohannes Weiner memstall = 1; 176082e60d00SJohannes Weiner } 176199486c51SChristoph Hellwig 176247e4937aSGao Xiang if (!bio) { 176307888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 176407888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 17650c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1766dfeab2e9SGao Xiang 1767dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 17681e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1769*3acea5fcSJingbo Xu (sb->s_blocksize_bits - 9); 1770cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 17716ea5aad3SGao Xiang if (f->readahead) 17726ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 177347e4937aSGao Xiang ++nr_bios; 177447e4937aSGao Xiang } 177547e4937aSGao Xiang 17766c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 177747e4937aSGao Xiang goto submit_bio_retry; 177847e4937aSGao Xiang 17791e4a2955SGao Xiang last_index = cur; 17801e4a2955SGao Xiang bypass = false; 17811e4a2955SGao Xiang } while (++cur < end); 178247e4937aSGao Xiang 17831e4a2955SGao Xiang if (!bypass) 178447e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 178547e4937aSGao Xiang else 178647e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 178747e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 178847e4937aSGao Xiang 178999486c51SChristoph Hellwig if (bio) { 179094e4e153SGao Xiang submit_bio(bio); 179182e60d00SJohannes Weiner if (memstall) 179282e60d00SJohannes Weiner psi_memstall_leave(&pflags); 179399486c51SChristoph Hellwig } 179447e4937aSGao Xiang 1795587a67b7SGao Xiang /* 1796587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 17973fffb589SSandeep Dhavale * don't issue decompression but drop it directly instead. 1798587a67b7SGao Xiang */ 1799587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1800587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 18011e4a2955SGao Xiang return; 1802587a67b7SGao Xiang } 1803cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 180447e4937aSGao Xiang } 180547e4937aSGao Xiang 180683a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1807eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 180847e4937aSGao Xiang { 1809a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 181047e4937aSGao Xiang 18115c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 181247e4937aSGao Xiang return; 181383a386c0SGao Xiang z_erofs_submit_queue(f, pagepool, io, &force_fg); 181447e4937aSGao Xiang 18150c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 18160c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 181747e4937aSGao Xiang 181847e4937aSGao Xiang if (!force_fg) 181947e4937aSGao Xiang return; 182047e4937aSGao Xiang 182147e4937aSGao Xiang /* wait until all bios are completed */ 182260b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 182347e4937aSGao Xiang 18240c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 18250c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 182647e4937aSGao Xiang } 182747e4937aSGao Xiang 182838629291SGao Xiang /* 182938629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 183038629291SGao Xiang * approximate readmore strategies as a start. 183138629291SGao Xiang */ 183238629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 183338629291SGao Xiang struct readahead_control *rac, 183438629291SGao Xiang erofs_off_t end, 1835eaa9172aSGao Xiang struct page **pagepool, 183638629291SGao Xiang bool backmost) 183738629291SGao Xiang { 183838629291SGao Xiang struct inode *inode = f->inode; 183938629291SGao Xiang struct erofs_map_blocks *map = &f->map; 184038629291SGao Xiang erofs_off_t cur; 184138629291SGao Xiang int err; 184238629291SGao Xiang 184338629291SGao Xiang if (backmost) { 184438629291SGao Xiang map->m_la = end; 1845622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1846622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 184738629291SGao Xiang if (err) 184838629291SGao Xiang return; 184938629291SGao Xiang 185038629291SGao Xiang /* expend ra for the trailing edge if readahead */ 185138629291SGao Xiang if (rac) { 185238629291SGao Xiang loff_t newstart = readahead_pos(rac); 185338629291SGao Xiang 185438629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 185538629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 185638629291SGao Xiang return; 185738629291SGao Xiang } 185838629291SGao Xiang end = round_up(end, PAGE_SIZE); 185938629291SGao Xiang } else { 186038629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 186138629291SGao Xiang 186238629291SGao Xiang if (!map->m_llen) 186338629291SGao Xiang return; 186438629291SGao Xiang } 186538629291SGao Xiang 186638629291SGao Xiang cur = map->m_la + map->m_llen - 1; 186738629291SGao Xiang while (cur >= end) { 186838629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 186938629291SGao Xiang struct page *page; 187038629291SGao Xiang 187138629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1872aa793b46SGao Xiang if (page) { 187338629291SGao Xiang if (PageUptodate(page)) { 187438629291SGao Xiang unlock_page(page); 1875aa793b46SGao Xiang } else { 187638629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 187738629291SGao Xiang if (err) 187838629291SGao Xiang erofs_err(inode->i_sb, 187938629291SGao Xiang "readmore error at page %lu @ nid %llu", 188038629291SGao Xiang index, EROFS_I(inode)->nid); 1881aa793b46SGao Xiang } 188238629291SGao Xiang put_page(page); 1883aa793b46SGao Xiang } 1884aa793b46SGao Xiang 188538629291SGao Xiang if (cur < PAGE_SIZE) 188638629291SGao Xiang break; 188738629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 188838629291SGao Xiang } 188938629291SGao Xiang } 189038629291SGao Xiang 1891a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 189247e4937aSGao Xiang { 1893a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 189447e4937aSGao Xiang struct inode *const inode = page->mapping->host; 189540452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 189647e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1897eaa9172aSGao Xiang struct page *pagepool = NULL; 189847e4937aSGao Xiang int err; 189947e4937aSGao Xiang 190047e4937aSGao Xiang trace_erofs_readpage(page, false); 190147e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 190247e4937aSGao Xiang 190338629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 190438629291SGao Xiang &pagepool, true); 19051825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 190638629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 190738629291SGao Xiang 19085c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 190947e4937aSGao Xiang 191047e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 191183a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 191240452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 191347e4937aSGao Xiang 191447e4937aSGao Xiang if (err) 19154f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 191647e4937aSGao Xiang 191709c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1918eaa9172aSGao Xiang erofs_release_pages(&pagepool); 191947e4937aSGao Xiang return err; 192047e4937aSGao Xiang } 192147e4937aSGao Xiang 19220615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 192347e4937aSGao Xiang { 19240615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 192547e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 192647e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1927eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 192838629291SGao Xiang unsigned int nr_pages; 192947e4937aSGao Xiang 19306ea5aad3SGao Xiang f.readahead = true; 19310615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 193247e4937aSGao Xiang 193338629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 193438629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 193538629291SGao Xiang nr_pages = readahead_count(rac); 193638629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 193738629291SGao Xiang 19380615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 193947e4937aSGao Xiang set_page_private(page, (unsigned long)head); 194047e4937aSGao Xiang head = page; 194147e4937aSGao Xiang } 194247e4937aSGao Xiang 194347e4937aSGao Xiang while (head) { 194447e4937aSGao Xiang struct page *page = head; 194547e4937aSGao Xiang int err; 194647e4937aSGao Xiang 194747e4937aSGao Xiang /* traversal in reverse order */ 194847e4937aSGao Xiang head = (void *)page_private(page); 194947e4937aSGao Xiang 19501825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1951a5876e24SGao Xiang if (err) 19524f761fa2SGao Xiang erofs_err(inode->i_sb, 19534f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 19544f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 195547e4937aSGao Xiang put_page(page); 195647e4937aSGao Xiang } 195738629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 19585c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 195947e4937aSGao Xiang 196083a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 196140452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 196209c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1963eaa9172aSGao Xiang erofs_release_pages(&pagepool); 196447e4937aSGao Xiang } 196547e4937aSGao Xiang 19660c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1967a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 19680615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 196947e4937aSGao Xiang }; 1970