147e4937aSGao Xiang // SPDX-License-Identifier: GPL-2.0-only 247e4937aSGao Xiang /* 347e4937aSGao Xiang * Copyright (C) 2018 HUAWEI, Inc. 4592e7cd0SAlexander A. Klimov * https://www.huawei.com/ 506a304cdSGao Xiang * Copyright (C) 2022 Alibaba Cloud 647e4937aSGao Xiang */ 747e4937aSGao Xiang #include "compress.h" 899486c51SChristoph Hellwig #include <linux/psi.h> 93fffb589SSandeep Dhavale #include <linux/cpuhotplug.h> 1047e4937aSGao Xiang #include <trace/events/erofs.h> 1147e4937aSGao Xiang 12a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_MAX_PAGES (Z_EROFS_PCLUSTER_MAX_SIZE / PAGE_SIZE) 13a9a94d93SGao Xiang #define Z_EROFS_INLINE_BVECS 2 14a9a94d93SGao Xiang 15a9a94d93SGao Xiang /* 16a9a94d93SGao Xiang * let's leave a type here in case of introducing 17a9a94d93SGao Xiang * another tagged pointer later. 18a9a94d93SGao Xiang */ 19a9a94d93SGao Xiang typedef void *z_erofs_next_pcluster_t; 20a9a94d93SGao Xiang 21a9a94d93SGao Xiang struct z_erofs_bvec { 22a9a94d93SGao Xiang struct page *page; 23a9a94d93SGao Xiang int offset; 24a9a94d93SGao Xiang unsigned int end; 25a9a94d93SGao Xiang }; 26a9a94d93SGao Xiang 27a9a94d93SGao Xiang #define __Z_EROFS_BVSET(name, total) \ 28a9a94d93SGao Xiang struct name { \ 29a9a94d93SGao Xiang /* point to the next page which contains the following bvecs */ \ 30a9a94d93SGao Xiang struct page *nextpage; \ 31a9a94d93SGao Xiang struct z_erofs_bvec bvec[total]; \ 32a9a94d93SGao Xiang } 33a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset,); 34a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset_inline, Z_EROFS_INLINE_BVECS); 35a9a94d93SGao Xiang 36a9a94d93SGao Xiang /* 37a9a94d93SGao Xiang * Structure fields follow one of the following exclusion rules. 38a9a94d93SGao Xiang * 39a9a94d93SGao Xiang * I: Modifiable by initialization/destruction paths and read-only 40a9a94d93SGao Xiang * for everyone else; 41a9a94d93SGao Xiang * 42a9a94d93SGao Xiang * L: Field should be protected by the pcluster lock; 43a9a94d93SGao Xiang * 44a9a94d93SGao Xiang * A: Field should be accessed / updated in atomic for parallelized code. 45a9a94d93SGao Xiang */ 46a9a94d93SGao Xiang struct z_erofs_pcluster { 47a9a94d93SGao Xiang struct erofs_workgroup obj; 48a9a94d93SGao Xiang struct mutex lock; 49a9a94d93SGao Xiang 50a9a94d93SGao Xiang /* A: point to next chained pcluster or TAILs */ 51a9a94d93SGao Xiang z_erofs_next_pcluster_t next; 52a9a94d93SGao Xiang 53a9a94d93SGao Xiang /* L: the maximum decompression size of this round */ 54a9a94d93SGao Xiang unsigned int length; 55a9a94d93SGao Xiang 56a9a94d93SGao Xiang /* L: total number of bvecs */ 57a9a94d93SGao Xiang unsigned int vcnt; 58a9a94d93SGao Xiang 59a9a94d93SGao Xiang /* I: page offset of start position of decompression */ 60a9a94d93SGao Xiang unsigned short pageofs_out; 61a9a94d93SGao Xiang 62a9a94d93SGao Xiang /* I: page offset of inline compressed data */ 63a9a94d93SGao Xiang unsigned short pageofs_in; 64a9a94d93SGao Xiang 65a9a94d93SGao Xiang union { 66a9a94d93SGao Xiang /* L: inline a certain number of bvec for bootstrap */ 67a9a94d93SGao Xiang struct z_erofs_bvset_inline bvset; 68a9a94d93SGao Xiang 69a9a94d93SGao Xiang /* I: can be used to free the pcluster by RCU. */ 70a9a94d93SGao Xiang struct rcu_head rcu; 71a9a94d93SGao Xiang }; 72a9a94d93SGao Xiang 73a9a94d93SGao Xiang union { 74a9a94d93SGao Xiang /* I: physical cluster size in pages */ 75a9a94d93SGao Xiang unsigned short pclusterpages; 76a9a94d93SGao Xiang 77a9a94d93SGao Xiang /* I: tailpacking inline compressed size */ 78a9a94d93SGao Xiang unsigned short tailpacking_size; 79a9a94d93SGao Xiang }; 80a9a94d93SGao Xiang 81a9a94d93SGao Xiang /* I: compression algorithm format */ 82a9a94d93SGao Xiang unsigned char algorithmformat; 83a9a94d93SGao Xiang 84a9a94d93SGao Xiang /* L: whether partial decompression or not */ 85a9a94d93SGao Xiang bool partial; 86a9a94d93SGao Xiang 87a9a94d93SGao Xiang /* L: indicate several pageofs_outs or not */ 88a9a94d93SGao Xiang bool multibases; 89a9a94d93SGao Xiang 90a9a94d93SGao Xiang /* A: compressed bvecs (can be cached or inplaced pages) */ 91a9a94d93SGao Xiang struct z_erofs_bvec compressed_bvecs[]; 92a9a94d93SGao Xiang }; 93a9a94d93SGao Xiang 94a9a94d93SGao Xiang /* let's avoid the valid 32-bit kernel addresses */ 95a9a94d93SGao Xiang 96a9a94d93SGao Xiang /* the chained workgroup has't submitted io (still open) */ 97a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE) 98a9a94d93SGao Xiang /* the chained workgroup has already submitted io */ 99a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL_CLOSED ((void *)0x5F0EDEAD) 100a9a94d93SGao Xiang 101a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_NIL (NULL) 102a9a94d93SGao Xiang 103a9a94d93SGao Xiang struct z_erofs_decompressqueue { 104a9a94d93SGao Xiang struct super_block *sb; 105a9a94d93SGao Xiang atomic_t pending_bios; 106a9a94d93SGao Xiang z_erofs_next_pcluster_t head; 107a9a94d93SGao Xiang 108a9a94d93SGao Xiang union { 109a9a94d93SGao Xiang struct completion done; 110a9a94d93SGao Xiang struct work_struct work; 1113fffb589SSandeep Dhavale struct kthread_work kthread_work; 112a9a94d93SGao Xiang } u; 113a9a94d93SGao Xiang bool eio, sync; 114a9a94d93SGao Xiang }; 115a9a94d93SGao Xiang 116a9a94d93SGao Xiang static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 117a9a94d93SGao Xiang { 118a9a94d93SGao Xiang return !pcl->obj.index; 119a9a94d93SGao Xiang } 120a9a94d93SGao Xiang 121a9a94d93SGao Xiang static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 122a9a94d93SGao Xiang { 123a9a94d93SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) 124a9a94d93SGao Xiang return 1; 125a9a94d93SGao Xiang return pcl->pclusterpages; 126a9a94d93SGao Xiang } 127a9a94d93SGao Xiang 128a9a94d93SGao Xiang /* 129a9a94d93SGao Xiang * bit 30: I/O error occurred on this page 130a9a94d93SGao Xiang * bit 0 - 29: remaining parts to complete this page 131a9a94d93SGao Xiang */ 132a9a94d93SGao Xiang #define Z_EROFS_PAGE_EIO (1 << 30) 133a9a94d93SGao Xiang 134a9a94d93SGao Xiang static inline void z_erofs_onlinepage_init(struct page *page) 135a9a94d93SGao Xiang { 136a9a94d93SGao Xiang union { 137a9a94d93SGao Xiang atomic_t o; 138a9a94d93SGao Xiang unsigned long v; 139a9a94d93SGao Xiang } u = { .o = ATOMIC_INIT(1) }; 140a9a94d93SGao Xiang 141a9a94d93SGao Xiang set_page_private(page, u.v); 142a9a94d93SGao Xiang smp_wmb(); 143a9a94d93SGao Xiang SetPagePrivate(page); 144a9a94d93SGao Xiang } 145a9a94d93SGao Xiang 146a9a94d93SGao Xiang static inline void z_erofs_onlinepage_split(struct page *page) 147a9a94d93SGao Xiang { 148a9a94d93SGao Xiang atomic_inc((atomic_t *)&page->private); 149a9a94d93SGao Xiang } 150a9a94d93SGao Xiang 151a9a94d93SGao Xiang static inline void z_erofs_page_mark_eio(struct page *page) 152a9a94d93SGao Xiang { 153a9a94d93SGao Xiang int orig; 154a9a94d93SGao Xiang 155a9a94d93SGao Xiang do { 156a9a94d93SGao Xiang orig = atomic_read((atomic_t *)&page->private); 157a9a94d93SGao Xiang } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 158a9a94d93SGao Xiang orig | Z_EROFS_PAGE_EIO) != orig); 159a9a94d93SGao Xiang } 160a9a94d93SGao Xiang 161a9a94d93SGao Xiang static inline void z_erofs_onlinepage_endio(struct page *page) 162a9a94d93SGao Xiang { 163a9a94d93SGao Xiang unsigned int v; 164a9a94d93SGao Xiang 165a9a94d93SGao Xiang DBG_BUGON(!PagePrivate(page)); 166a9a94d93SGao Xiang v = atomic_dec_return((atomic_t *)&page->private); 167a9a94d93SGao Xiang if (!(v & ~Z_EROFS_PAGE_EIO)) { 168a9a94d93SGao Xiang set_page_private(page, 0); 169a9a94d93SGao Xiang ClearPagePrivate(page); 170a9a94d93SGao Xiang if (!(v & Z_EROFS_PAGE_EIO)) 171a9a94d93SGao Xiang SetPageUptodate(page); 172a9a94d93SGao Xiang unlock_page(page); 173a9a94d93SGao Xiang } 174a9a94d93SGao Xiang } 175a9a94d93SGao Xiang 176a9a94d93SGao Xiang #define Z_EROFS_ONSTACK_PAGES 32 177a9a94d93SGao Xiang 17847e4937aSGao Xiang /* 1799f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 1809f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 1819f6cc76eSGao Xiang */ 1829f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 1839f6cc76eSGao Xiang struct kmem_cache *slab; 1849f6cc76eSGao Xiang unsigned int maxpages; 1859f6cc76eSGao Xiang char name[48]; 1869f6cc76eSGao Xiang }; 1879f6cc76eSGao Xiang 1889f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 1899f6cc76eSGao Xiang 1909f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 1919f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 1929f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 1939f6cc76eSGao Xiang }; 1949f6cc76eSGao Xiang 19506a304cdSGao Xiang struct z_erofs_bvec_iter { 19606a304cdSGao Xiang struct page *bvpage; 19706a304cdSGao Xiang struct z_erofs_bvset *bvset; 19806a304cdSGao Xiang unsigned int nr, cur; 19906a304cdSGao Xiang }; 20006a304cdSGao Xiang 20106a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 20206a304cdSGao Xiang { 20306a304cdSGao Xiang if (iter->bvpage) 20406a304cdSGao Xiang kunmap_local(iter->bvset); 20506a304cdSGao Xiang return iter->bvpage; 20606a304cdSGao Xiang } 20706a304cdSGao Xiang 20806a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 20906a304cdSGao Xiang { 21006a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 21106a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 21206a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 21306a304cdSGao Xiang struct page *oldpage; 21406a304cdSGao Xiang 21506a304cdSGao Xiang DBG_BUGON(!nextpage); 21606a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 21706a304cdSGao Xiang iter->bvpage = nextpage; 21806a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 21906a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 22006a304cdSGao Xiang iter->cur = 0; 22106a304cdSGao Xiang return oldpage; 22206a304cdSGao Xiang } 22306a304cdSGao Xiang 22406a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 22506a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 22606a304cdSGao Xiang unsigned int bootstrap_nr, 22706a304cdSGao Xiang unsigned int cur) 22806a304cdSGao Xiang { 22906a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 23006a304cdSGao Xiang .nr = bootstrap_nr, 23106a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 23206a304cdSGao Xiang }; 23306a304cdSGao Xiang 23406a304cdSGao Xiang while (cur > iter->nr) { 23506a304cdSGao Xiang cur -= iter->nr; 23606a304cdSGao Xiang z_erofs_bvset_flip(iter); 23706a304cdSGao Xiang } 23806a304cdSGao Xiang iter->cur = cur; 23906a304cdSGao Xiang } 24006a304cdSGao Xiang 24106a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 24206a304cdSGao Xiang struct z_erofs_bvec *bvec, 24306a304cdSGao Xiang struct page **candidate_bvpage) 24406a304cdSGao Xiang { 24506a304cdSGao Xiang if (iter->cur == iter->nr) { 24606a304cdSGao Xiang if (!*candidate_bvpage) 24706a304cdSGao Xiang return -EAGAIN; 24806a304cdSGao Xiang 24906a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 25006a304cdSGao Xiang iter->bvset->nextpage = *candidate_bvpage; 25106a304cdSGao Xiang z_erofs_bvset_flip(iter); 25206a304cdSGao Xiang 25306a304cdSGao Xiang iter->bvset->nextpage = NULL; 25406a304cdSGao Xiang *candidate_bvpage = NULL; 25506a304cdSGao Xiang } 25606a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 25706a304cdSGao Xiang return 0; 25806a304cdSGao Xiang } 25906a304cdSGao Xiang 26006a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 26106a304cdSGao Xiang struct z_erofs_bvec *bvec, 26206a304cdSGao Xiang struct page **old_bvpage) 26306a304cdSGao Xiang { 26406a304cdSGao Xiang if (iter->cur == iter->nr) 26506a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 26606a304cdSGao Xiang else 26706a304cdSGao Xiang *old_bvpage = NULL; 26806a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 26906a304cdSGao Xiang } 27006a304cdSGao Xiang 2719f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 2729f6cc76eSGao Xiang { 2739f6cc76eSGao Xiang int i; 2749f6cc76eSGao Xiang 2759f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 2769f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 2779f6cc76eSGao Xiang continue; 2789f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 2799f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 2809f6cc76eSGao Xiang } 2819f6cc76eSGao Xiang } 2829f6cc76eSGao Xiang 2839f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 2849f6cc76eSGao Xiang { 2859f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 2869f6cc76eSGao Xiang struct z_erofs_pcluster *a; 2879f6cc76eSGao Xiang unsigned int size; 2889f6cc76eSGao Xiang 2899f6cc76eSGao Xiang for (pcs = pcluster_pool; 2909f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 291ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 2929f6cc76eSGao Xiang 2939f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 2949f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 2959f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 2969f6cc76eSGao Xiang if (pcs->slab) 2979f6cc76eSGao Xiang continue; 2989f6cc76eSGao Xiang 2999f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 3009f6cc76eSGao Xiang return -ENOMEM; 3019f6cc76eSGao Xiang } 3029f6cc76eSGao Xiang return 0; 3039f6cc76eSGao Xiang } 3049f6cc76eSGao Xiang 3059f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 3069f6cc76eSGao Xiang { 3079f6cc76eSGao Xiang int i; 3089f6cc76eSGao Xiang 3099f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3109f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3119f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 3129f6cc76eSGao Xiang 3139f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 3149f6cc76eSGao Xiang continue; 3159f6cc76eSGao Xiang 3169f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 3179f6cc76eSGao Xiang if (!pcl) 3189f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 3199f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 3209f6cc76eSGao Xiang return pcl; 3219f6cc76eSGao Xiang } 3229f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 3239f6cc76eSGao Xiang } 3249f6cc76eSGao Xiang 3259f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 3269f6cc76eSGao Xiang { 327cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 3289f6cc76eSGao Xiang int i; 3299f6cc76eSGao Xiang 3309f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3319f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3329f6cc76eSGao Xiang 333cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 3349f6cc76eSGao Xiang continue; 3359f6cc76eSGao Xiang 3369f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 3379f6cc76eSGao Xiang return; 3389f6cc76eSGao Xiang } 3399f6cc76eSGao Xiang DBG_BUGON(1); 3409f6cc76eSGao Xiang } 3419f6cc76eSGao Xiang 34247e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 34347e4937aSGao Xiang 3443fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 3453fffb589SSandeep Dhavale static struct kthread_worker __rcu **z_erofs_pcpu_workers; 3463fffb589SSandeep Dhavale 3473fffb589SSandeep Dhavale static void erofs_destroy_percpu_workers(void) 34847e4937aSGao Xiang { 3493fffb589SSandeep Dhavale struct kthread_worker *worker; 3503fffb589SSandeep Dhavale unsigned int cpu; 3513fffb589SSandeep Dhavale 3523fffb589SSandeep Dhavale for_each_possible_cpu(cpu) { 3533fffb589SSandeep Dhavale worker = rcu_dereference_protected( 3543fffb589SSandeep Dhavale z_erofs_pcpu_workers[cpu], 1); 3553fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 3563fffb589SSandeep Dhavale if (worker) 3573fffb589SSandeep Dhavale kthread_destroy_worker(worker); 3583fffb589SSandeep Dhavale } 3593fffb589SSandeep Dhavale kfree(z_erofs_pcpu_workers); 36047e4937aSGao Xiang } 36147e4937aSGao Xiang 3623fffb589SSandeep Dhavale static struct kthread_worker *erofs_init_percpu_worker(int cpu) 36347e4937aSGao Xiang { 3643fffb589SSandeep Dhavale struct kthread_worker *worker = 3653fffb589SSandeep Dhavale kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 36647e4937aSGao Xiang 3673fffb589SSandeep Dhavale if (IS_ERR(worker)) 3683fffb589SSandeep Dhavale return worker; 3693fffb589SSandeep Dhavale if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 3703fffb589SSandeep Dhavale sched_set_fifo_low(worker->task); 3713fffb589SSandeep Dhavale return worker; 3723fffb589SSandeep Dhavale } 3733fffb589SSandeep Dhavale 3743fffb589SSandeep Dhavale static int erofs_init_percpu_workers(void) 3753fffb589SSandeep Dhavale { 3763fffb589SSandeep Dhavale struct kthread_worker *worker; 3773fffb589SSandeep Dhavale unsigned int cpu; 3783fffb589SSandeep Dhavale 3793fffb589SSandeep Dhavale z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 3803fffb589SSandeep Dhavale sizeof(struct kthread_worker *), GFP_ATOMIC); 3813fffb589SSandeep Dhavale if (!z_erofs_pcpu_workers) 3823fffb589SSandeep Dhavale return -ENOMEM; 3833fffb589SSandeep Dhavale 3843fffb589SSandeep Dhavale for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 3853fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 3863fffb589SSandeep Dhavale if (!IS_ERR(worker)) 3873fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 3883fffb589SSandeep Dhavale } 3893fffb589SSandeep Dhavale return 0; 3903fffb589SSandeep Dhavale } 3913fffb589SSandeep Dhavale #else 3923fffb589SSandeep Dhavale static inline void erofs_destroy_percpu_workers(void) {} 3933fffb589SSandeep Dhavale static inline int erofs_init_percpu_workers(void) { return 0; } 3943fffb589SSandeep Dhavale #endif 3953fffb589SSandeep Dhavale 3963fffb589SSandeep Dhavale #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 3973fffb589SSandeep Dhavale static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 3983fffb589SSandeep Dhavale static enum cpuhp_state erofs_cpuhp_state; 3993fffb589SSandeep Dhavale 4003fffb589SSandeep Dhavale static int erofs_cpu_online(unsigned int cpu) 4013fffb589SSandeep Dhavale { 4023fffb589SSandeep Dhavale struct kthread_worker *worker, *old; 4033fffb589SSandeep Dhavale 4043fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 4053fffb589SSandeep Dhavale if (IS_ERR(worker)) 4063fffb589SSandeep Dhavale return PTR_ERR(worker); 4073fffb589SSandeep Dhavale 4083fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4093fffb589SSandeep Dhavale old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4103fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4113fffb589SSandeep Dhavale if (!old) 4123fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 4133fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4143fffb589SSandeep Dhavale if (old) 4153fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4163fffb589SSandeep Dhavale return 0; 4173fffb589SSandeep Dhavale } 4183fffb589SSandeep Dhavale 4193fffb589SSandeep Dhavale static int erofs_cpu_offline(unsigned int cpu) 4203fffb589SSandeep Dhavale { 4213fffb589SSandeep Dhavale struct kthread_worker *worker; 4223fffb589SSandeep Dhavale 4233fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4243fffb589SSandeep Dhavale worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4253fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4263fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 4273fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4283fffb589SSandeep Dhavale 4293fffb589SSandeep Dhavale synchronize_rcu(); 4303fffb589SSandeep Dhavale if (worker) 4313fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4323fffb589SSandeep Dhavale return 0; 4333fffb589SSandeep Dhavale } 4343fffb589SSandeep Dhavale 4353fffb589SSandeep Dhavale static int erofs_cpu_hotplug_init(void) 4363fffb589SSandeep Dhavale { 4373fffb589SSandeep Dhavale int state; 4383fffb589SSandeep Dhavale 4393fffb589SSandeep Dhavale state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 4403fffb589SSandeep Dhavale "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 4413fffb589SSandeep Dhavale if (state < 0) 4423fffb589SSandeep Dhavale return state; 4433fffb589SSandeep Dhavale 4443fffb589SSandeep Dhavale erofs_cpuhp_state = state; 4453fffb589SSandeep Dhavale return 0; 4463fffb589SSandeep Dhavale } 4473fffb589SSandeep Dhavale 4483fffb589SSandeep Dhavale static void erofs_cpu_hotplug_destroy(void) 4493fffb589SSandeep Dhavale { 4503fffb589SSandeep Dhavale if (erofs_cpuhp_state) 4513fffb589SSandeep Dhavale cpuhp_remove_state_nocalls(erofs_cpuhp_state); 4523fffb589SSandeep Dhavale } 4533fffb589SSandeep Dhavale #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 4543fffb589SSandeep Dhavale static inline int erofs_cpu_hotplug_init(void) { return 0; } 4553fffb589SSandeep Dhavale static inline void erofs_cpu_hotplug_destroy(void) {} 4563fffb589SSandeep Dhavale #endif 4573fffb589SSandeep Dhavale 4583fffb589SSandeep Dhavale void z_erofs_exit_zip_subsystem(void) 4593fffb589SSandeep Dhavale { 4603fffb589SSandeep Dhavale erofs_cpu_hotplug_destroy(); 4613fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4623fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4633fffb589SSandeep Dhavale z_erofs_destroy_pcluster_pool(); 46447e4937aSGao Xiang } 46547e4937aSGao Xiang 46647e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 46747e4937aSGao Xiang { 4689f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 46947e4937aSGao Xiang 4709f6cc76eSGao Xiang if (err) 4713fffb589SSandeep Dhavale goto out_error_pcluster_pool; 4723fffb589SSandeep Dhavale 4733fffb589SSandeep Dhavale z_erofs_workqueue = alloc_workqueue("erofs_worker", 4743fffb589SSandeep Dhavale WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 4758d1b80a7SDan Carpenter if (!z_erofs_workqueue) { 4768d1b80a7SDan Carpenter err = -ENOMEM; 4773fffb589SSandeep Dhavale goto out_error_workqueue_init; 4788d1b80a7SDan Carpenter } 4793fffb589SSandeep Dhavale 4803fffb589SSandeep Dhavale err = erofs_init_percpu_workers(); 4819f6cc76eSGao Xiang if (err) 4823fffb589SSandeep Dhavale goto out_error_pcpu_worker; 4833fffb589SSandeep Dhavale 4843fffb589SSandeep Dhavale err = erofs_cpu_hotplug_init(); 4853fffb589SSandeep Dhavale if (err < 0) 4863fffb589SSandeep Dhavale goto out_error_cpuhp_init; 4873fffb589SSandeep Dhavale return err; 4883fffb589SSandeep Dhavale 4893fffb589SSandeep Dhavale out_error_cpuhp_init: 4903fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4913fffb589SSandeep Dhavale out_error_pcpu_worker: 4923fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4933fffb589SSandeep Dhavale out_error_workqueue_init: 4949f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 4953fffb589SSandeep Dhavale out_error_pcluster_pool: 4969f6cc76eSGao Xiang return err; 49747e4937aSGao Xiang } 49847e4937aSGao Xiang 499db166fc2SGao Xiang enum z_erofs_pclustermode { 500db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 50147e4937aSGao Xiang /* 502db166fc2SGao Xiang * The current pclusters was the tail of an exist chain, in addition 503db166fc2SGao Xiang * that the previous processed chained pclusters are all decided to 50447e4937aSGao Xiang * be hooked up to it. 505db166fc2SGao Xiang * A new chain will be created for the remaining pclusters which are 506db166fc2SGao Xiang * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED, 507db166fc2SGao Xiang * the next pcluster cannot reuse the whole page safely for inplace I/O 508db166fc2SGao Xiang * in the following scenario: 50947e4937aSGao Xiang * ________________________________________________________________ 51047e4937aSGao Xiang * | tail (partial) page | head (partial) page | 511db166fc2SGao Xiang * | (belongs to the next pcl) | (belongs to the current pcl) | 512db166fc2SGao Xiang * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________| 51347e4937aSGao Xiang */ 514db166fc2SGao Xiang Z_EROFS_PCLUSTER_HOOKED, 5150b964600SGao Xiang /* 516db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 5170b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 5180b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 519387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 5200b964600SGao Xiang */ 521db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 52247e4937aSGao Xiang /* 52347e4937aSGao Xiang * The current collection has been linked with the owned chain, and 52447e4937aSGao Xiang * could also be linked with the remaining collections, which means 52547e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 52647e4937aSGao Xiang * the current collection can safely use the whole page (since 52747e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 52847e4937aSGao Xiang * illustrated below: 52947e4937aSGao Xiang * ________________________________________________________________ 53047e4937aSGao Xiang * | tail (partial) page | head (partial) page | 53147e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 532db166fc2SGao Xiang * | PCLUSTER_FOLLOWED or | | 533db166fc2SGao Xiang * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________| 53447e4937aSGao Xiang * 53547e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 53647e4937aSGao Xiang */ 537db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 53847e4937aSGao Xiang }; 53947e4937aSGao Xiang 5405c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 5415c6dcc57SGao Xiang struct inode *const inode; 5425c6dcc57SGao Xiang struct erofs_map_blocks map; 54306a304cdSGao Xiang struct z_erofs_bvec_iter biter; 54447e4937aSGao Xiang 54506a304cdSGao Xiang struct page *candidate_bvpage; 54647e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 54747e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 548db166fc2SGao Xiang enum z_erofs_pclustermode mode; 54947e4937aSGao Xiang 55047e4937aSGao Xiang /* used for applying cache strategy on the fly */ 55147e4937aSGao Xiang bool backmost; 55247e4937aSGao Xiang erofs_off_t headoffset; 553ed722fbcSGao Xiang 554ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 555ed722fbcSGao Xiang unsigned int icur; 55647e4937aSGao Xiang }; 55747e4937aSGao Xiang 55847e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 5595c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 560db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 56147e4937aSGao Xiang 5621282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 5631282dea3SGao Xiang { 5641282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 5651282dea3SGao Xiang 5661282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 5671282dea3SGao Xiang return false; 5681282dea3SGao Xiang 5691282dea3SGao Xiang if (fe->backmost) 5701282dea3SGao Xiang return true; 5711282dea3SGao Xiang 5721282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 5731282dea3SGao Xiang fe->map.m_la < fe->headoffset) 5741282dea3SGao Xiang return true; 5751282dea3SGao Xiang 5761282dea3SGao Xiang return false; 5771282dea3SGao Xiang } 5781282dea3SGao Xiang 5796f39d1e1SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 580eaa9172aSGao Xiang struct page **pagepool) 58147e4937aSGao Xiang { 5826f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 5835c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 5841282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 58547e4937aSGao Xiang bool standalone = true; 5866f39d1e1SGao Xiang /* 5876f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 5886f39d1e1SGao Xiang * can be used if low memory otherwise. 5896f39d1e1SGao Xiang */ 5901825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 5911825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 592ed722fbcSGao Xiang unsigned int i; 59347e4937aSGao Xiang 594db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 59547e4937aSGao Xiang return; 59647e4937aSGao Xiang 597ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 59847e4937aSGao Xiang struct page *page; 599b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 6001825c8d7SGao Xiang struct page *newpage = NULL; 60147e4937aSGao Xiang 60247e4937aSGao Xiang /* the compressed page was loaded before */ 603ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 60447e4937aSGao Xiang continue; 60547e4937aSGao Xiang 606ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 60747e4937aSGao Xiang 60847e4937aSGao Xiang if (page) { 609b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 6100b964600SGao Xiang } else { 6110b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 6120b964600SGao Xiang standalone = false; 6131282dea3SGao Xiang if (!shouldalloc) 6141282dea3SGao Xiang continue; 6151282dea3SGao Xiang 6161282dea3SGao Xiang /* 6171282dea3SGao Xiang * try to use cached I/O if page allocation 6181282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 6191282dea3SGao Xiang * to avoid any direct reclaim. 6201282dea3SGao Xiang */ 6211825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 6221825c8d7SGao Xiang if (!newpage) 62347e4937aSGao Xiang continue; 6241282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 625b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 62647e4937aSGao Xiang } 62747e4937aSGao Xiang 628b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 62947e4937aSGao Xiang continue; 63047e4937aSGao Xiang 631eaa9172aSGao Xiang if (page) 63247e4937aSGao Xiang put_page(page); 633eaa9172aSGao Xiang else if (newpage) 634eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 63547e4937aSGao Xiang } 63647e4937aSGao Xiang 6370b964600SGao Xiang /* 6380b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 6390b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 6400b964600SGao Xiang */ 6410b964600SGao Xiang if (standalone) 642db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 64347e4937aSGao Xiang } 64447e4937aSGao Xiang 64547e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 64647e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 64747e4937aSGao Xiang struct erofs_workgroup *grp) 64847e4937aSGao Xiang { 64947e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 65047e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 65147e4937aSGao Xiang int i; 65247e4937aSGao Xiang 653cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 65447e4937aSGao Xiang /* 65547e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 65647e4937aSGao Xiang * therefore no need to worry about available decompression users. 65747e4937aSGao Xiang */ 6589f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 659ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 66047e4937aSGao Xiang 66147e4937aSGao Xiang if (!page) 66247e4937aSGao Xiang continue; 66347e4937aSGao Xiang 66447e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 66547e4937aSGao Xiang if (!trylock_page(page)) 66647e4937aSGao Xiang return -EBUSY; 66747e4937aSGao Xiang 668f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 66947e4937aSGao Xiang continue; 67047e4937aSGao Xiang 67147e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 672ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6736aaa7b06SGao Xiang detach_page_private(page); 67447e4937aSGao Xiang unlock_page(page); 67547e4937aSGao Xiang } 67647e4937aSGao Xiang return 0; 67747e4937aSGao Xiang } 67847e4937aSGao Xiang 679d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 68047e4937aSGao Xiang { 68147e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 682ed722fbcSGao Xiang int ret, i; 68347e4937aSGao Xiang 684ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 685ed722fbcSGao Xiang return 0; 68647e4937aSGao Xiang 687ed722fbcSGao Xiang ret = 0; 688cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 6899f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 690ed722fbcSGao Xiang if (pcl->compressed_bvecs[i].page == page) { 691ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 69247e4937aSGao Xiang ret = 1; 69347e4937aSGao Xiang break; 69447e4937aSGao Xiang } 69547e4937aSGao Xiang } 69647e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 6976aaa7b06SGao Xiang if (ret) 6986aaa7b06SGao Xiang detach_page_private(page); 69947e4937aSGao Xiang return ret; 70047e4937aSGao Xiang } 70147e4937aSGao Xiang 7025c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 703ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 70447e4937aSGao Xiang { 7055c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 70647e4937aSGao Xiang 707ed722fbcSGao Xiang while (fe->icur > 0) { 708ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 709ed722fbcSGao Xiang NULL, bvec->page)) { 710ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 71147e4937aSGao Xiang return true; 712ed722fbcSGao Xiang } 713ed722fbcSGao Xiang } 71447e4937aSGao Xiang return false; 71547e4937aSGao Xiang } 71647e4937aSGao Xiang 71787ca34a7SGao Xiang /* callers must be with pcluster lock held */ 7185c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 7195b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 72047e4937aSGao Xiang { 72147e4937aSGao Xiang int ret; 72247e4937aSGao Xiang 723db166fc2SGao Xiang if (exclusive) { 72406a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 725ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 72647e4937aSGao Xiang return 0; 72706a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 728db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 72906a304cdSGao Xiang !fe->candidate_bvpage) 73006a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 73106a304cdSGao Xiang } 73206a304cdSGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 73306a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 73406a304cdSGao Xiang return ret; 73547e4937aSGao Xiang } 73647e4937aSGao Xiang 7375c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 73847e4937aSGao Xiang { 7395c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 7405c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 74147e4937aSGao Xiang 742473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 743473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 744473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 74547e4937aSGao Xiang *owned_head = &pcl->next; 746473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 747db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 748473e15b0SGao Xiang return; 749473e15b0SGao Xiang } 750473e15b0SGao Xiang 75147e4937aSGao Xiang /* 752473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 753473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 75447e4937aSGao Xiang */ 755267f2492SGao Xiang if (*owned_head != &pcl->next && pcl != f->tailpcl && 756267f2492SGao Xiang cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 757473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 75847e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 759db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_HOOKED; 7605c6dcc57SGao Xiang f->tailpcl = NULL; 761473e15b0SGao Xiang return; 76247e4937aSGao Xiang } 763473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 764db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 76547e4937aSGao Xiang } 76647e4937aSGao Xiang 76783a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 76847e4937aSGao Xiang { 76983a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 770cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 77147e4937aSGao Xiang struct z_erofs_pcluster *pcl; 77264094a04SGao Xiang struct erofs_workgroup *grp; 77347e4937aSGao Xiang int err; 77447e4937aSGao Xiang 775c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 776c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 7778f899262SGao Xiang DBG_BUGON(1); 7788f899262SGao Xiang return -EFSCORRUPTED; 7798f899262SGao Xiang } 7808f899262SGao Xiang 7819f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 782cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 783cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 7849f6cc76eSGao Xiang if (IS_ERR(pcl)) 7859f6cc76eSGao Xiang return PTR_ERR(pcl); 78647e4937aSGao Xiang 78764094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 7888f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 7892bfab9c0SGao Xiang pcl->length = 0; 7902bfab9c0SGao Xiang pcl->partial = true; 79147e4937aSGao Xiang 79247e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 7935c6dcc57SGao Xiang pcl->next = fe->owned_head; 79487ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 795db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 79647e4937aSGao Xiang 79747e4937aSGao Xiang /* 79847e4937aSGao Xiang * lock all primary followed works before visible to others 79947e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 80047e4937aSGao Xiang */ 80187ca34a7SGao Xiang mutex_init(&pcl->lock); 80287ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 80347e4937aSGao Xiang 804cecf864dSYue Hu if (ztailpacking) { 805cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 8063acea5fcSJingbo Xu pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 807cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 808cecf864dSYue Hu } else { 809cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 810cecf864dSYue Hu 81183a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 81264094a04SGao Xiang if (IS_ERR(grp)) { 81364094a04SGao Xiang err = PTR_ERR(grp); 81464094a04SGao Xiang goto err_out; 81564094a04SGao Xiang } 81664094a04SGao Xiang 81764094a04SGao Xiang if (grp != &pcl->obj) { 8185c6dcc57SGao Xiang fe->pcl = container_of(grp, 819cecf864dSYue Hu struct z_erofs_pcluster, obj); 82064094a04SGao Xiang err = -EEXIST; 82164094a04SGao Xiang goto err_out; 82247e4937aSGao Xiang } 823cecf864dSYue Hu } 82447e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 8255c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 8265c6dcc57SGao Xiang fe->tailpcl = pcl; 8275c6dcc57SGao Xiang fe->owned_head = &pcl->next; 8285c6dcc57SGao Xiang fe->pcl = pcl; 8299e579fc1SGao Xiang return 0; 83064094a04SGao Xiang 83164094a04SGao Xiang err_out: 83287ca34a7SGao Xiang mutex_unlock(&pcl->lock); 8339f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 83464094a04SGao Xiang return err; 83547e4937aSGao Xiang } 83647e4937aSGao Xiang 83783a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 83847e4937aSGao Xiang { 83983a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 8400d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 8419e579fc1SGao Xiang int ret; 84247e4937aSGao Xiang 84387ca34a7SGao Xiang DBG_BUGON(fe->pcl); 84447e4937aSGao Xiang 84587ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 8465c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 8475c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 84847e4937aSGao Xiang 8490d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 8500d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 8510d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 8520d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 85347e4937aSGao Xiang DBG_BUGON(1); 854cecf864dSYue Hu return -EFSCORRUPTED; 855cecf864dSYue Hu } 85647e4937aSGao Xiang 85764094a04SGao Xiang if (grp) { 8585c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 8590d823b42SGao Xiang ret = -EEXIST; 86064094a04SGao Xiang } else { 86183a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 86264094a04SGao Xiang } 86347e4937aSGao Xiang 8640d823b42SGao Xiang if (ret == -EEXIST) { 865267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 866267f2492SGao Xiang /* used to check tail merging loop due to corrupted images */ 867267f2492SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 868267f2492SGao Xiang fe->tailpcl = fe->pcl; 869267f2492SGao Xiang 870267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 8710d823b42SGao Xiang } else if (ret) { 8720d823b42SGao Xiang return ret; 8730d823b42SGao Xiang } 87406a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 875387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 87681382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 877ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 87847e4937aSGao Xiang return 0; 87947e4937aSGao Xiang } 88047e4937aSGao Xiang 88147e4937aSGao Xiang /* 88247e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 88347e4937aSGao Xiang * only after a RCU grace period. 88447e4937aSGao Xiang */ 88547e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 88647e4937aSGao Xiang { 88787ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 88887ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 88947e4937aSGao Xiang } 89047e4937aSGao Xiang 89147e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 89247e4937aSGao Xiang { 89347e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 89447e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 89547e4937aSGao Xiang 89687ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 89747e4937aSGao Xiang } 89847e4937aSGao Xiang 8995c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 90047e4937aSGao Xiang { 90187ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 90247e4937aSGao Xiang 90387ca34a7SGao Xiang if (!pcl) 90447e4937aSGao Xiang return false; 90547e4937aSGao Xiang 90606a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 90787ca34a7SGao Xiang mutex_unlock(&pcl->lock); 90847e4937aSGao Xiang 90906a304cdSGao Xiang if (fe->candidate_bvpage) { 91006a304cdSGao Xiang DBG_BUGON(z_erofs_is_shortlived_page(fe->candidate_bvpage)); 91106a304cdSGao Xiang fe->candidate_bvpage = NULL; 91206a304cdSGao Xiang } 91306a304cdSGao Xiang 91447e4937aSGao Xiang /* 91547e4937aSGao Xiang * if all pending pages are added, don't hold its reference 91647e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 91747e4937aSGao Xiang */ 918db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 91987ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 92047e4937aSGao Xiang 92187ca34a7SGao Xiang fe->pcl = NULL; 92247e4937aSGao Xiang return true; 92347e4937aSGao Xiang } 92447e4937aSGao Xiang 925b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 926b15b2e30SYue Hu struct page *page, unsigned int pageofs, 927b15b2e30SYue Hu unsigned int len) 928b15b2e30SYue Hu { 9293acea5fcSJingbo Xu struct super_block *sb = inode->i_sb; 930b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 931b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 932b15b2e30SYue Hu u8 *src, *dst; 933b15b2e30SYue Hu unsigned int i, cnt; 934b15b2e30SYue Hu 935e5126de1SYue Hu if (!packed_inode) 936e5126de1SYue Hu return -EFSCORRUPTED; 937e5126de1SYue Hu 938eb2c5e41SGao Xiang buf.inode = packed_inode; 939b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 940b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 941b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 9423acea5fcSJingbo Xu sb->s_blocksize - erofs_blkoff(sb, pos)); 943eb2c5e41SGao Xiang src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 944b15b2e30SYue Hu if (IS_ERR(src)) { 945b15b2e30SYue Hu erofs_put_metabuf(&buf); 946b15b2e30SYue Hu return PTR_ERR(src); 947b15b2e30SYue Hu } 948b15b2e30SYue Hu 949b15b2e30SYue Hu dst = kmap_local_page(page); 9503acea5fcSJingbo Xu memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 951b15b2e30SYue Hu kunmap_local(dst); 952b15b2e30SYue Hu pos += cnt; 953b15b2e30SYue Hu } 954b15b2e30SYue Hu erofs_put_metabuf(&buf); 955b15b2e30SYue Hu return 0; 956b15b2e30SYue Hu } 957b15b2e30SYue Hu 95847e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 959eaa9172aSGao Xiang struct page *page, struct page **pagepool) 96047e4937aSGao Xiang { 96147e4937aSGao Xiang struct inode *const inode = fe->inode; 96247e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 96347e4937aSGao Xiang const loff_t offset = page_offset(page); 9645b220b20SGao Xiang bool tight = true, exclusive; 9652bfab9c0SGao Xiang unsigned int cur, end, spiltted; 96647e4937aSGao Xiang int err = 0; 96747e4937aSGao Xiang 96847e4937aSGao Xiang /* register locked file pages as online pages in pack */ 96947e4937aSGao Xiang z_erofs_onlinepage_init(page); 97047e4937aSGao Xiang 97147e4937aSGao Xiang spiltted = 0; 97247e4937aSGao Xiang end = PAGE_SIZE; 97347e4937aSGao Xiang repeat: 97447e4937aSGao Xiang cur = end - 1; 97547e4937aSGao Xiang 97639397a46SGao Xiang if (offset + cur < map->m_la || 97739397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 9785c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 97947e4937aSGao Xiang fe->backmost = false; 98047e4937aSGao Xiang map->m_la = offset + cur; 98147e4937aSGao Xiang map->m_llen = 0; 98247e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 9838d8a09b0SGao Xiang if (err) 98467148551SGao Xiang goto out; 98539397a46SGao Xiang } else { 98639397a46SGao Xiang if (fe->pcl) 98739397a46SGao Xiang goto hitted; 98839397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 98939397a46SGao Xiang } 99047e4937aSGao Xiang 991b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 992b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 99347e4937aSGao Xiang goto hitted; 99447e4937aSGao Xiang 99583a386c0SGao Xiang err = z_erofs_collector_begin(fe); 9968d8a09b0SGao Xiang if (err) 99767148551SGao Xiang goto out; 99847e4937aSGao Xiang 9995c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 100009c54379SGao Xiang void *mp; 1001cecf864dSYue Hu 100209c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 10033acea5fcSJingbo Xu erofs_blknr(inode->i_sb, map->m_pa), 10043acea5fcSJingbo Xu EROFS_NO_KMAP); 100509c54379SGao Xiang if (IS_ERR(mp)) { 100609c54379SGao Xiang err = PTR_ERR(mp); 1007cecf864dSYue Hu erofs_err(inode->i_sb, 1008cecf864dSYue Hu "failed to get inline page, err %d", err); 100967148551SGao Xiang goto out; 1010cecf864dSYue Hu } 101109c54379SGao Xiang get_page(fe->map.buf.page); 1012ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 1013ed722fbcSGao Xiang fe->map.buf.page); 1014db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 1015cecf864dSYue Hu } else { 10166f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 10171282dea3SGao Xiang z_erofs_bind_cache(fe, pagepool); 1018cecf864dSYue Hu } 101947e4937aSGao Xiang hitted: 1020dc76ea8cSGao Xiang /* 1021dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 1022dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 1023dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 1024387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 1025dc76ea8cSGao Xiang */ 1026db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED && 1027db166fc2SGao Xiang fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1028dc76ea8cSGao Xiang 102947e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 10308d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 103147e4937aSGao Xiang zero_user_segment(page, cur, end); 103247e4937aSGao Xiang goto next_part; 103347e4937aSGao Xiang } 1034b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 1035b15b2e30SYue Hu unsigned int pageofs, skip, len; 1036b15b2e30SYue Hu 1037b15b2e30SYue Hu if (offset > map->m_la) { 1038b15b2e30SYue Hu pageofs = 0; 1039b15b2e30SYue Hu skip = offset - map->m_la; 1040b15b2e30SYue Hu } else { 1041b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 1042b15b2e30SYue Hu skip = 0; 1043b15b2e30SYue Hu } 1044b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 1045b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1046b15b2e30SYue Hu if (err) 1047b15b2e30SYue Hu goto out; 1048b15b2e30SYue Hu ++spiltted; 1049b15b2e30SYue Hu tight = false; 1050b15b2e30SYue Hu goto next_part; 1051b15b2e30SYue Hu } 105247e4937aSGao Xiang 10535b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 105447e4937aSGao Xiang if (cur) 1055db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 105647e4937aSGao Xiang 105747e4937aSGao Xiang retry: 105806a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 105906a304cdSGao Xiang .page = page, 106006a304cdSGao Xiang .offset = offset - map->m_la, 106106a304cdSGao Xiang .end = end, 10625b220b20SGao Xiang }), exclusive); 106306a304cdSGao Xiang /* should allocate an additional short-lived page for bvset */ 106406a304cdSGao Xiang if (err == -EAGAIN && !fe->candidate_bvpage) { 106506a304cdSGao Xiang fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL); 106606a304cdSGao Xiang set_page_private(fe->candidate_bvpage, 106706a304cdSGao Xiang Z_EROFS_SHORTLIVED_PAGE); 106847e4937aSGao Xiang goto retry; 106947e4937aSGao Xiang } 107047e4937aSGao Xiang 107106a304cdSGao Xiang if (err) { 107206a304cdSGao Xiang DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage); 107367148551SGao Xiang goto out; 107406a304cdSGao Xiang } 107547e4937aSGao Xiang 107667148551SGao Xiang z_erofs_onlinepage_split(page); 107747e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 107847e4937aSGao Xiang ++spiltted; 1079267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1080267f2492SGao Xiang fe->pcl->multibases = true; 10812bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 10822bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 10832bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 10842bfab9c0SGao Xiang } 1085e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1086e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1087e7933278SGao Xiang fe->pcl->length == map->m_llen) 1088e7933278SGao Xiang fe->pcl->partial = false; 108947e4937aSGao Xiang next_part: 10902bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 109147e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 10922bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 109347e4937aSGao Xiang 109447e4937aSGao Xiang end = cur; 109547e4937aSGao Xiang if (end > 0) 109647e4937aSGao Xiang goto repeat; 109747e4937aSGao Xiang 109847e4937aSGao Xiang out: 109967148551SGao Xiang if (err) 110067148551SGao Xiang z_erofs_page_mark_eio(page); 110147e4937aSGao Xiang z_erofs_onlinepage_endio(page); 110247e4937aSGao Xiang return err; 110347e4937aSGao Xiang } 110447e4937aSGao Xiang 1105ef4b4b46SYue Hu static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 110640452ffcSHuang Jianan unsigned int readahead_pages) 110740452ffcSHuang Jianan { 1108a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 110940452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 111040452ffcSHuang Jianan !readahead_pages) 111140452ffcSHuang Jianan return true; 111240452ffcSHuang Jianan 111340452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 111440452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 111540452ffcSHuang Jianan return true; 111640452ffcSHuang Jianan 111740452ffcSHuang Jianan return false; 111840452ffcSHuang Jianan } 111940452ffcSHuang Jianan 11206aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 11216aaa7b06SGao Xiang { 11226aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 11236aaa7b06SGao Xiang } 11246aaa7b06SGao Xiang 11254f05687fSGao Xiang struct z_erofs_decompress_backend { 11264f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 11274f05687fSGao Xiang struct super_block *sb; 11284f05687fSGao Xiang struct z_erofs_pcluster *pcl; 11294f05687fSGao Xiang 11304f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 11314f05687fSGao Xiang struct page **decompressed_pages; 11324f05687fSGao Xiang /* pages to keep the compressed data */ 11334f05687fSGao Xiang struct page **compressed_pages; 11344f05687fSGao Xiang 1135267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 11364f05687fSGao Xiang struct page **pagepool; 11372bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 11384f05687fSGao Xiang }; 11394f05687fSGao Xiang 1140267f2492SGao Xiang struct z_erofs_bvec_item { 1141267f2492SGao Xiang struct z_erofs_bvec bvec; 1142267f2492SGao Xiang struct list_head list; 1143267f2492SGao Xiang }; 1144267f2492SGao Xiang 1145267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 11463fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 11473fe96ee0SGao Xiang { 1148267f2492SGao Xiang struct z_erofs_bvec_item *item; 1149267f2492SGao Xiang 1150267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1151267f2492SGao Xiang unsigned int pgnr; 11523fe96ee0SGao Xiang 1153267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 11542bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 115563bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 11563fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1157267f2492SGao Xiang return; 11583fe96ee0SGao Xiang } 115963bbb856SGao Xiang } 11603fe96ee0SGao Xiang 1161267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1162267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1163267f2492SGao Xiang item->bvec = *bvec; 1164267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1165267f2492SGao Xiang } 1166267f2492SGao Xiang 1167267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1168267f2492SGao Xiang int err) 1169267f2492SGao Xiang { 1170267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1171267f2492SGao Xiang struct list_head *p, *n; 1172267f2492SGao Xiang 1173267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1174267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1175267f2492SGao Xiang unsigned int end, cur; 1176267f2492SGao Xiang void *dst, *src; 1177267f2492SGao Xiang 1178267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1179267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1180267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1181267f2492SGao Xiang bvi->bvec.end); 1182267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1183267f2492SGao Xiang while (cur < end) { 1184267f2492SGao Xiang unsigned int pgnr, scur, len; 1185267f2492SGao Xiang 1186267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1187267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1188267f2492SGao Xiang 1189267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1190267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1191267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1192267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1193267f2492SGao Xiang err = -EFSCORRUPTED; 1194267f2492SGao Xiang cur += len; 1195267f2492SGao Xiang continue; 1196267f2492SGao Xiang } 1197267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1198267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1199267f2492SGao Xiang kunmap_local(src); 1200267f2492SGao Xiang cur += len; 1201267f2492SGao Xiang } 1202267f2492SGao Xiang kunmap_local(dst); 1203267f2492SGao Xiang if (err) 1204267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 1205267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 1206267f2492SGao Xiang list_del(p); 1207267f2492SGao Xiang kfree(bvi); 1208267f2492SGao Xiang } 1209267f2492SGao Xiang } 1210267f2492SGao Xiang 1211267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 121242fec235SGao Xiang { 12134f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 121406a304cdSGao Xiang struct z_erofs_bvec_iter biter; 121506a304cdSGao Xiang struct page *old_bvpage; 1216267f2492SGao Xiang int i; 121742fec235SGao Xiang 1218387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 121942fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 122006a304cdSGao Xiang struct z_erofs_bvec bvec; 122142fec235SGao Xiang 122206a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 122342fec235SGao Xiang 122406a304cdSGao Xiang if (old_bvpage) 12254f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 122642fec235SGao Xiang 122706a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1228267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 122942fec235SGao Xiang } 123006a304cdSGao Xiang 123106a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 123206a304cdSGao Xiang if (old_bvpage) 12334f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 123442fec235SGao Xiang } 123542fec235SGao Xiang 12364f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 12374f05687fSGao Xiang bool *overlapped) 123867139e36SGao Xiang { 12394f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 124067139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 124167139e36SGao Xiang int i, err = 0; 124267139e36SGao Xiang 124367139e36SGao Xiang *overlapped = false; 124467139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1245ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1246ed722fbcSGao Xiang struct page *page = bvec->page; 124767139e36SGao Xiang 124867139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 124967139e36SGao Xiang if (!page) { 125067139e36SGao Xiang DBG_BUGON(1); 125167139e36SGao Xiang continue; 125267139e36SGao Xiang } 1253fe3e5914SGao Xiang be->compressed_pages[i] = page; 125467139e36SGao Xiang 125567139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 125667139e36SGao Xiang if (!PageUptodate(page)) 125767139e36SGao Xiang err = -EIO; 125867139e36SGao Xiang continue; 125967139e36SGao Xiang } 126067139e36SGao Xiang 126167139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 126267139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 12634f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 126467139e36SGao Xiang if (!PageUptodate(page)) 126567139e36SGao Xiang err = -EIO; 126667139e36SGao Xiang continue; 126767139e36SGao Xiang } 1268267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 126967139e36SGao Xiang *overlapped = true; 127067139e36SGao Xiang } 127167139e36SGao Xiang } 127267139e36SGao Xiang 1273fe3e5914SGao Xiang if (err) 12744f05687fSGao Xiang return err; 12754f05687fSGao Xiang return 0; 127667139e36SGao Xiang } 127767139e36SGao Xiang 12784f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 12794f05687fSGao Xiang int err) 128047e4937aSGao Xiang { 12814f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 12824f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1283cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1284597e2953SYue Hu const struct z_erofs_decompressor *decompressor = 1285597e2953SYue Hu &erofs_decompressors[pcl->algorithmformat]; 12862bfab9c0SGao Xiang unsigned int i, inputsize; 128767148551SGao Xiang int err2; 12882bfab9c0SGao Xiang struct page *page; 12892bfab9c0SGao Xiang bool overlapped; 129047e4937aSGao Xiang 129187ca34a7SGao Xiang mutex_lock(&pcl->lock); 12922bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 129347e4937aSGao Xiang 1294fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1295fe3e5914SGao Xiang be->decompressed_pages = NULL; 1296fe3e5914SGao Xiang be->compressed_pages = NULL; 1297fe3e5914SGao Xiang be->onstack_used = 0; 12982bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 12994f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 13002bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 13014f05687fSGao Xiang memset(be->decompressed_pages, 0, 13022bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1303fe3e5914SGao Xiang } 1304fe3e5914SGao Xiang 1305fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1306fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1307fe3e5914SGao Xiang 1308fe3e5914SGao Xiang if (!be->decompressed_pages) 13094f05687fSGao Xiang be->decompressed_pages = 1310647dd2c3SGao Xiang kvcalloc(be->nr_pages, sizeof(struct page *), 1311e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1312fe3e5914SGao Xiang if (!be->compressed_pages) 1313fe3e5914SGao Xiang be->compressed_pages = 1314647dd2c3SGao Xiang kvcalloc(pclusterpages, sizeof(struct page *), 1315fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 131647e4937aSGao Xiang 1317267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 13184f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 13194f05687fSGao Xiang if (err2) 13204f05687fSGao Xiang err = err2; 13218d8a09b0SGao Xiang if (err) 132247e4937aSGao Xiang goto out; 132347e4937aSGao Xiang 1324cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1325cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1326cecf864dSYue Hu else 1327cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1328cecf864dSYue Hu 1329597e2953SYue Hu err = decompressor->decompress(&(struct z_erofs_decompress_req) { 13304f05687fSGao Xiang .sb = be->sb, 13314f05687fSGao Xiang .in = be->compressed_pages, 13324f05687fSGao Xiang .out = be->decompressed_pages, 1333cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 133487ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 13359f6cc76eSGao Xiang .inputsize = inputsize, 13362bfab9c0SGao Xiang .outputsize = pcl->length, 133747e4937aSGao Xiang .alg = pcl->algorithmformat, 133847e4937aSGao Xiang .inplace_io = overlapped, 13392bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1340267f2492SGao Xiang .fillgaps = pcl->multibases, 13414f05687fSGao Xiang }, be->pagepool); 134247e4937aSGao Xiang 134347e4937aSGao Xiang out: 1344cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1345cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1346ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1347ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1348cecf864dSYue Hu put_page(page); 1349cecf864dSYue Hu } else { 1350cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1351ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 135247e4937aSGao Xiang 135347e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 135447e4937aSGao Xiang continue; 135547e4937aSGao Xiang 13566aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13574f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1358ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 135947e4937aSGao Xiang } 1360cecf864dSYue Hu } 1361fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1362fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1363647dd2c3SGao Xiang kvfree(be->compressed_pages); 1364267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 136547e4937aSGao Xiang 13662bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 13674f05687fSGao Xiang page = be->decompressed_pages[i]; 136847e4937aSGao Xiang if (!page) 136947e4937aSGao Xiang continue; 137047e4937aSGao Xiang 13716aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 137247e4937aSGao Xiang 13736aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13744f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 137547e4937aSGao Xiang continue; 137667148551SGao Xiang if (err) 137767148551SGao Xiang z_erofs_page_mark_eio(page); 137847e4937aSGao Xiang z_erofs_onlinepage_endio(page); 137947e4937aSGao Xiang } 138047e4937aSGao Xiang 13814f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 1382647dd2c3SGao Xiang kvfree(be->decompressed_pages); 138347e4937aSGao Xiang 13842bfab9c0SGao Xiang pcl->length = 0; 13852bfab9c0SGao Xiang pcl->partial = true; 1386267f2492SGao Xiang pcl->multibases = false; 138706a304cdSGao Xiang pcl->bvset.nextpage = NULL; 138887ca34a7SGao Xiang pcl->vcnt = 0; 138947e4937aSGao Xiang 139087ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 139147e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 139287ca34a7SGao Xiang mutex_unlock(&pcl->lock); 139347e4937aSGao Xiang return err; 139447e4937aSGao Xiang } 139547e4937aSGao Xiang 13960c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1397eaa9172aSGao Xiang struct page **pagepool) 139847e4937aSGao Xiang { 13994f05687fSGao Xiang struct z_erofs_decompress_backend be = { 14004f05687fSGao Xiang .sb = io->sb, 14014f05687fSGao Xiang .pagepool = pagepool, 1402267f2492SGao Xiang .decompressed_secondary_bvecs = 1403267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 14044f05687fSGao Xiang }; 140547e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 140647e4937aSGao Xiang 140747e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 14084f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 140947e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 14104f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */ 141147e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 141247e4937aSGao Xiang 14134f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 14144f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 141547e4937aSGao Xiang 14164f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 14174f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 141847e4937aSGao Xiang } 141947e4937aSGao Xiang } 142047e4937aSGao Xiang 14210c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 142247e4937aSGao Xiang { 1423a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1424a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1425eaa9172aSGao Xiang struct page *pagepool = NULL; 142647e4937aSGao Xiang 1427a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 14280c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 1429eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1430a4b1fab1SGao Xiang kvfree(bgq); 143147e4937aSGao Xiang } 143247e4937aSGao Xiang 14333fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14343fffb589SSandeep Dhavale static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 14353fffb589SSandeep Dhavale { 14363fffb589SSandeep Dhavale z_erofs_decompressqueue_work((struct work_struct *)work); 14373fffb589SSandeep Dhavale } 14383fffb589SSandeep Dhavale #endif 14393fffb589SSandeep Dhavale 14407865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1441cdba5506SGao Xiang int bios) 14427865827cSGao Xiang { 14437865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 14447865827cSGao Xiang 14457865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1446cdba5506SGao Xiang if (io->sync) { 14477865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 144860b30050SHongyu Jin complete(&io->u.done); 14497865827cSGao Xiang return; 14507865827cSGao Xiang } 14517865827cSGao Xiang 14527865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 14537865827cSGao Xiang return; 14543fffb589SSandeep Dhavale /* Use (kthread_)work and sync decompression for atomic contexts only */ 14557865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 14563fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14573fffb589SSandeep Dhavale struct kthread_worker *worker; 14583fffb589SSandeep Dhavale 14593fffb589SSandeep Dhavale rcu_read_lock(); 14603fffb589SSandeep Dhavale worker = rcu_dereference( 14613fffb589SSandeep Dhavale z_erofs_pcpu_workers[raw_smp_processor_id()]); 14623fffb589SSandeep Dhavale if (!worker) { 14633fffb589SSandeep Dhavale INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 14647865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 14653fffb589SSandeep Dhavale } else { 14663fffb589SSandeep Dhavale kthread_queue_work(worker, &io->u.kthread_work); 14673fffb589SSandeep Dhavale } 14683fffb589SSandeep Dhavale rcu_read_unlock(); 14693fffb589SSandeep Dhavale #else 14703fffb589SSandeep Dhavale queue_work(z_erofs_workqueue, &io->u.work); 14713fffb589SSandeep Dhavale #endif 14727865827cSGao Xiang /* enable sync decompression for readahead */ 14737865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 14747865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 14757865827cSGao Xiang return; 14767865827cSGao Xiang } 14777865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 14787865827cSGao Xiang } 14797865827cSGao Xiang 148047e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 148147e4937aSGao Xiang unsigned int nr, 1482eaa9172aSGao Xiang struct page **pagepool, 14839f2731d6SGao Xiang struct address_space *mc) 148447e4937aSGao Xiang { 148547e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 14869f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 148747e4937aSGao Xiang bool tocache = false; 148847e4937aSGao Xiang 148947e4937aSGao Xiang struct address_space *mapping; 149047e4937aSGao Xiang struct page *oldpage, *page; 149147e4937aSGao Xiang int justfound; 149247e4937aSGao Xiang 149347e4937aSGao Xiang repeat: 1494ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 149547e4937aSGao Xiang oldpage = page; 149647e4937aSGao Xiang 149747e4937aSGao Xiang if (!page) 149847e4937aSGao Xiang goto out_allocpage; 149947e4937aSGao Xiang 1500b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1501b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 150247e4937aSGao Xiang 15031825c8d7SGao Xiang /* 15041825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 15051825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 15061825c8d7SGao Xiang */ 15071825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1508ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 15091825c8d7SGao Xiang set_page_private(page, 0); 15101825c8d7SGao Xiang tocache = true; 15111825c8d7SGao Xiang goto out_tocache; 15121825c8d7SGao Xiang } 151347e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 151447e4937aSGao Xiang 151547e4937aSGao Xiang /* 15166aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 151747e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 151847e4937aSGao Xiang */ 151947e4937aSGao Xiang if (mapping && mapping != mc) 152047e4937aSGao Xiang /* ought to be unmanaged pages */ 152147e4937aSGao Xiang goto out; 152247e4937aSGao Xiang 15236aaa7b06SGao Xiang /* directly return for shortlived page as well */ 15246aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 15256aaa7b06SGao Xiang goto out; 15266aaa7b06SGao Xiang 152747e4937aSGao Xiang lock_page(page); 152847e4937aSGao Xiang 152947e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 153047e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 153147e4937aSGao Xiang 153247e4937aSGao Xiang /* the page is still in manage cache */ 153347e4937aSGao Xiang if (page->mapping == mc) { 1534ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 153547e4937aSGao Xiang 153647e4937aSGao Xiang if (!PagePrivate(page)) { 153747e4937aSGao Xiang /* 153847e4937aSGao Xiang * impossible to be !PagePrivate(page) for 153947e4937aSGao Xiang * the current restriction as well if 1540ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 154147e4937aSGao Xiang */ 154247e4937aSGao Xiang DBG_BUGON(!justfound); 154347e4937aSGao Xiang 154447e4937aSGao Xiang justfound = 0; 154547e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 154647e4937aSGao Xiang SetPagePrivate(page); 154747e4937aSGao Xiang } 154847e4937aSGao Xiang 154947e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 155047e4937aSGao Xiang if (PageUptodate(page)) { 155147e4937aSGao Xiang unlock_page(page); 155247e4937aSGao Xiang page = NULL; 155347e4937aSGao Xiang } 155447e4937aSGao Xiang goto out; 155547e4937aSGao Xiang } 155647e4937aSGao Xiang 155747e4937aSGao Xiang /* 155847e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 155947e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 156047e4937aSGao Xiang */ 156147e4937aSGao Xiang DBG_BUGON(page->mapping); 156247e4937aSGao Xiang DBG_BUGON(!justfound); 156347e4937aSGao Xiang 156447e4937aSGao Xiang tocache = true; 156547e4937aSGao Xiang unlock_page(page); 156647e4937aSGao Xiang put_page(page); 156747e4937aSGao Xiang out_allocpage: 15685ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1569ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1570ed722fbcSGao Xiang oldpage, page)) { 1571eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 15725ddcee1fSGao Xiang cond_resched(); 15735ddcee1fSGao Xiang goto repeat; 15745ddcee1fSGao Xiang } 15751825c8d7SGao Xiang out_tocache: 1576bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1577bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1578bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1579bf225074SGao Xiang goto out; 1580a30573b3SGao Xiang } 1581bf225074SGao Xiang attach_page_private(page, pcl); 1582bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1583bf225074SGao Xiang put_page(page); 1584bf225074SGao Xiang 158547e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 158647e4937aSGao Xiang return page; 158747e4937aSGao Xiang } 158847e4937aSGao Xiang 1589cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1590a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 159147e4937aSGao Xiang { 1592a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 159347e4937aSGao Xiang 1594a4b1fab1SGao Xiang if (fg && !*fg) { 1595a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1596a4b1fab1SGao Xiang if (!q) { 1597a4b1fab1SGao Xiang *fg = true; 1598a4b1fab1SGao Xiang goto fg_out; 159947e4937aSGao Xiang } 16003fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 16013fffb589SSandeep Dhavale kthread_init_work(&q->u.kthread_work, 16023fffb589SSandeep Dhavale z_erofs_decompressqueue_kthread_work); 16033fffb589SSandeep Dhavale #else 16040c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 16053fffb589SSandeep Dhavale #endif 1606a4b1fab1SGao Xiang } else { 1607a4b1fab1SGao Xiang fg_out: 1608a4b1fab1SGao Xiang q = fgq; 160960b30050SHongyu Jin init_completion(&fgq->u.done); 1610a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 161167148551SGao Xiang q->eio = false; 1612cdba5506SGao Xiang q->sync = true; 1613a4b1fab1SGao Xiang } 1614a4b1fab1SGao Xiang q->sb = sb; 1615a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1616a4b1fab1SGao Xiang return q; 161747e4937aSGao Xiang } 161847e4937aSGao Xiang 161947e4937aSGao Xiang /* define decompression jobqueue types */ 162047e4937aSGao Xiang enum { 162147e4937aSGao Xiang JQ_BYPASS, 162247e4937aSGao Xiang JQ_SUBMIT, 162347e4937aSGao Xiang NR_JOBQUEUES, 162447e4937aSGao Xiang }; 162547e4937aSGao Xiang 162647e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 162747e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 162847e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 162947e4937aSGao Xiang { 163047e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 163147e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 163247e4937aSGao Xiang 163347e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 163447e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 163547e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 163647e4937aSGao Xiang 163747e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 163847e4937aSGao Xiang 163947e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 164047e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 164147e4937aSGao Xiang 164247e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 164347e4937aSGao Xiang } 164447e4937aSGao Xiang 16457865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 16467865827cSGao Xiang { 1647cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 16487865827cSGao Xiang blk_status_t err = bio->bi_status; 16497865827cSGao Xiang struct bio_vec *bvec; 16507865827cSGao Xiang struct bvec_iter_all iter_all; 16517865827cSGao Xiang 16527865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 16537865827cSGao Xiang struct page *page = bvec->bv_page; 16547865827cSGao Xiang 16557865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 16567865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 16577865827cSGao Xiang 16587865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 16597865827cSGao Xiang if (!err) 16607865827cSGao Xiang SetPageUptodate(page); 16617865827cSGao Xiang unlock_page(page); 16627865827cSGao Xiang } 16637865827cSGao Xiang } 166467148551SGao Xiang if (err) 166567148551SGao Xiang q->eio = true; 1666cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 16677865827cSGao Xiang bio_put(bio); 16687865827cSGao Xiang } 16697865827cSGao Xiang 167083a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1671eaa9172aSGao Xiang struct page **pagepool, 1672a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1673ef4b4b46SYue Hu bool *force_fg, bool readahead) 167447e4937aSGao Xiang { 167583a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 167683a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 167747e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1678a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 16795c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1680dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 16813f649ab7SKees Cook pgoff_t last_index; 1682dfeab2e9SGao Xiang struct block_device *last_bdev; 16831e4a2955SGao Xiang unsigned int nr_bios = 0; 16841e4a2955SGao Xiang struct bio *bio = NULL; 168582e60d00SJohannes Weiner unsigned long pflags; 168682e60d00SJohannes Weiner int memstall = 0; 168747e4937aSGao Xiang 1688cdba5506SGao Xiang /* 1689cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1690cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1691cdba5506SGao Xiang */ 1692cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1693cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1694cdba5506SGao Xiang 1695a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1696a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 169747e4937aSGao Xiang 169847e4937aSGao Xiang /* by default, all need io submission */ 169947e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 170047e4937aSGao Xiang 170147e4937aSGao Xiang do { 1702dfeab2e9SGao Xiang struct erofs_map_dev mdev; 170347e4937aSGao Xiang struct z_erofs_pcluster *pcl; 17041e4a2955SGao Xiang pgoff_t cur, end; 17051e4a2955SGao Xiang unsigned int i = 0; 17061e4a2955SGao Xiang bool bypass = true; 170747e4937aSGao Xiang 170847e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 170947e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 171047e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 171147e4937aSGao Xiang 171247e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 171347e4937aSGao Xiang 1714cecf864dSYue Hu /* close the main owned chain at first */ 1715cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1716cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1717cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1718cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1719cecf864dSYue Hu continue; 1720cecf864dSYue Hu } 1721cecf864dSYue Hu 1722dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1723dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 17243acea5fcSJingbo Xu .m_pa = erofs_pos(sb, pcl->obj.index), 1725dfeab2e9SGao Xiang }; 1726dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1727dfeab2e9SGao Xiang 17283acea5fcSJingbo Xu cur = erofs_blknr(sb, mdev.m_pa); 17299f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 173047e4937aSGao Xiang 17311e4a2955SGao Xiang do { 17321e4a2955SGao Xiang struct page *page; 173347e4937aSGao Xiang 17341e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 173583a386c0SGao Xiang mc); 17361e4a2955SGao Xiang if (!page) 17371e4a2955SGao Xiang continue; 173847e4937aSGao Xiang 1739dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1740dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 174147e4937aSGao Xiang submit_bio_retry: 174294e4e153SGao Xiang submit_bio(bio); 174382e60d00SJohannes Weiner if (memstall) { 174482e60d00SJohannes Weiner psi_memstall_leave(&pflags); 174582e60d00SJohannes Weiner memstall = 0; 174682e60d00SJohannes Weiner } 174747e4937aSGao Xiang bio = NULL; 174847e4937aSGao Xiang } 174947e4937aSGao Xiang 175082e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 175199486c51SChristoph Hellwig psi_memstall_enter(&pflags); 175282e60d00SJohannes Weiner memstall = 1; 175382e60d00SJohannes Weiner } 175499486c51SChristoph Hellwig 175547e4937aSGao Xiang if (!bio) { 175607888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 175707888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 17580c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1759dfeab2e9SGao Xiang 1760dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 17611e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 17623acea5fcSJingbo Xu (sb->s_blocksize_bits - 9); 1763cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 1764ef4b4b46SYue Hu if (readahead) 17656ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 176647e4937aSGao Xiang ++nr_bios; 176747e4937aSGao Xiang } 176847e4937aSGao Xiang 17696c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 177047e4937aSGao Xiang goto submit_bio_retry; 177147e4937aSGao Xiang 17721e4a2955SGao Xiang last_index = cur; 17731e4a2955SGao Xiang bypass = false; 17741e4a2955SGao Xiang } while (++cur < end); 177547e4937aSGao Xiang 17761e4a2955SGao Xiang if (!bypass) 177747e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 177847e4937aSGao Xiang else 177947e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 178047e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 178147e4937aSGao Xiang 178299486c51SChristoph Hellwig if (bio) { 178394e4e153SGao Xiang submit_bio(bio); 178482e60d00SJohannes Weiner if (memstall) 178582e60d00SJohannes Weiner psi_memstall_leave(&pflags); 178699486c51SChristoph Hellwig } 178747e4937aSGao Xiang 1788587a67b7SGao Xiang /* 1789587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 17903fffb589SSandeep Dhavale * don't issue decompression but drop it directly instead. 1791587a67b7SGao Xiang */ 1792587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1793587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 17941e4a2955SGao Xiang return; 1795587a67b7SGao Xiang } 1796cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 179747e4937aSGao Xiang } 179847e4937aSGao Xiang 179983a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1800ef4b4b46SYue Hu struct page **pagepool, bool force_fg, bool ra) 180147e4937aSGao Xiang { 1802a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 180347e4937aSGao Xiang 18045c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 180547e4937aSGao Xiang return; 1806ef4b4b46SYue Hu z_erofs_submit_queue(f, pagepool, io, &force_fg, ra); 180747e4937aSGao Xiang 18080c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 18090c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 181047e4937aSGao Xiang 181147e4937aSGao Xiang if (!force_fg) 181247e4937aSGao Xiang return; 181347e4937aSGao Xiang 181447e4937aSGao Xiang /* wait until all bios are completed */ 181560b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 181647e4937aSGao Xiang 18170c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 18180c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 181947e4937aSGao Xiang } 182047e4937aSGao Xiang 182138629291SGao Xiang /* 182238629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 182338629291SGao Xiang * approximate readmore strategies as a start. 182438629291SGao Xiang */ 182538629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 182638629291SGao Xiang struct readahead_control *rac, 1827*796e9149SYue Hu struct page **pagepool, bool backmost) 182838629291SGao Xiang { 182938629291SGao Xiang struct inode *inode = f->inode; 183038629291SGao Xiang struct erofs_map_blocks *map = &f->map; 1831*796e9149SYue Hu erofs_off_t cur, end, headoffset = f->headoffset; 183238629291SGao Xiang int err; 183338629291SGao Xiang 183438629291SGao Xiang if (backmost) { 1835*796e9149SYue Hu if (rac) 1836*796e9149SYue Hu end = headoffset + readahead_length(rac) - 1; 1837*796e9149SYue Hu else 1838*796e9149SYue Hu end = headoffset + PAGE_SIZE - 1; 183938629291SGao Xiang map->m_la = end; 1840622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1841622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 184238629291SGao Xiang if (err) 184338629291SGao Xiang return; 184438629291SGao Xiang 1845*796e9149SYue Hu /* expand ra for the trailing edge if readahead */ 184638629291SGao Xiang if (rac) { 184738629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1848*796e9149SYue Hu readahead_expand(rac, headoffset, cur - headoffset); 184938629291SGao Xiang return; 185038629291SGao Xiang } 185138629291SGao Xiang end = round_up(end, PAGE_SIZE); 185238629291SGao Xiang } else { 185338629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 185438629291SGao Xiang 185538629291SGao Xiang if (!map->m_llen) 185638629291SGao Xiang return; 185738629291SGao Xiang } 185838629291SGao Xiang 185938629291SGao Xiang cur = map->m_la + map->m_llen - 1; 186038629291SGao Xiang while (cur >= end) { 186138629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 186238629291SGao Xiang struct page *page; 186338629291SGao Xiang 186438629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1865aa793b46SGao Xiang if (page) { 186638629291SGao Xiang if (PageUptodate(page)) { 186738629291SGao Xiang unlock_page(page); 1868aa793b46SGao Xiang } else { 186938629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 187038629291SGao Xiang if (err) 187138629291SGao Xiang erofs_err(inode->i_sb, 187238629291SGao Xiang "readmore error at page %lu @ nid %llu", 187338629291SGao Xiang index, EROFS_I(inode)->nid); 1874aa793b46SGao Xiang } 187538629291SGao Xiang put_page(page); 1876aa793b46SGao Xiang } 1877aa793b46SGao Xiang 187838629291SGao Xiang if (cur < PAGE_SIZE) 187938629291SGao Xiang break; 188038629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 188138629291SGao Xiang } 188238629291SGao Xiang } 188338629291SGao Xiang 1884a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 188547e4937aSGao Xiang { 1886a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 188747e4937aSGao Xiang struct inode *const inode = page->mapping->host; 188840452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 188947e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1890eaa9172aSGao Xiang struct page *pagepool = NULL; 189147e4937aSGao Xiang int err; 189247e4937aSGao Xiang 189347e4937aSGao Xiang trace_erofs_readpage(page, false); 189447e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 189547e4937aSGao Xiang 1896*796e9149SYue Hu z_erofs_pcluster_readmore(&f, NULL, &pagepool, true); 18971825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1898*796e9149SYue Hu z_erofs_pcluster_readmore(&f, NULL, &pagepool, false); 189938629291SGao Xiang 19005c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 190147e4937aSGao Xiang 190247e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 1903ef4b4b46SYue Hu z_erofs_runqueue(&f, &pagepool, z_erofs_is_sync_decompress(sbi, 0), 1904ef4b4b46SYue Hu false); 190547e4937aSGao Xiang 190647e4937aSGao Xiang if (err) 19074f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 190847e4937aSGao Xiang 190909c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1910eaa9172aSGao Xiang erofs_release_pages(&pagepool); 191147e4937aSGao Xiang return err; 191247e4937aSGao Xiang } 191347e4937aSGao Xiang 19140615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 191547e4937aSGao Xiang { 19160615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 191747e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 191847e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1919eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 192038629291SGao Xiang unsigned int nr_pages; 192147e4937aSGao Xiang 19220615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 192347e4937aSGao Xiang 1924*796e9149SYue Hu z_erofs_pcluster_readmore(&f, rac, &pagepool, true); 192538629291SGao Xiang nr_pages = readahead_count(rac); 192638629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 192738629291SGao Xiang 19280615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 192947e4937aSGao Xiang set_page_private(page, (unsigned long)head); 193047e4937aSGao Xiang head = page; 193147e4937aSGao Xiang } 193247e4937aSGao Xiang 193347e4937aSGao Xiang while (head) { 193447e4937aSGao Xiang struct page *page = head; 193547e4937aSGao Xiang int err; 193647e4937aSGao Xiang 193747e4937aSGao Xiang /* traversal in reverse order */ 193847e4937aSGao Xiang head = (void *)page_private(page); 193947e4937aSGao Xiang 19401825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1941a5876e24SGao Xiang if (err) 19424f761fa2SGao Xiang erofs_err(inode->i_sb, 19434f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 19444f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 194547e4937aSGao Xiang put_page(page); 194647e4937aSGao Xiang } 1947*796e9149SYue Hu z_erofs_pcluster_readmore(&f, rac, &pagepool, false); 19485c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 194947e4937aSGao Xiang 195083a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 1951ef4b4b46SYue Hu z_erofs_is_sync_decompress(sbi, nr_pages), true); 195209c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1953eaa9172aSGao Xiang erofs_release_pages(&pagepool); 195447e4937aSGao Xiang } 195547e4937aSGao Xiang 19560c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1957a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 19580615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 195947e4937aSGao Xiang }; 1960