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, 243*6ab5eed6SGao Xiang struct page **candidate_bvpage, 244*6ab5eed6SGao Xiang struct page **pagepool) 24506a304cdSGao Xiang { 24605b63d2bSGao Xiang if (iter->cur >= iter->nr) { 24705b63d2bSGao Xiang struct page *nextpage = *candidate_bvpage; 24806a304cdSGao Xiang 24905b63d2bSGao Xiang if (!nextpage) { 250*6ab5eed6SGao Xiang nextpage = erofs_allocpage(pagepool, GFP_NOFS); 25105b63d2bSGao Xiang if (!nextpage) 25205b63d2bSGao Xiang return -ENOMEM; 25305b63d2bSGao Xiang set_page_private(nextpage, Z_EROFS_SHORTLIVED_PAGE); 25405b63d2bSGao Xiang } 25506a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 25605b63d2bSGao Xiang iter->bvset->nextpage = nextpage; 25706a304cdSGao Xiang z_erofs_bvset_flip(iter); 25806a304cdSGao Xiang 25906a304cdSGao Xiang iter->bvset->nextpage = NULL; 26006a304cdSGao Xiang *candidate_bvpage = NULL; 26106a304cdSGao Xiang } 26206a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 26306a304cdSGao Xiang return 0; 26406a304cdSGao Xiang } 26506a304cdSGao Xiang 26606a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 26706a304cdSGao Xiang struct z_erofs_bvec *bvec, 26806a304cdSGao Xiang struct page **old_bvpage) 26906a304cdSGao Xiang { 27006a304cdSGao Xiang if (iter->cur == iter->nr) 27106a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 27206a304cdSGao Xiang else 27306a304cdSGao Xiang *old_bvpage = NULL; 27406a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 27506a304cdSGao Xiang } 27606a304cdSGao Xiang 2779f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 2789f6cc76eSGao Xiang { 2799f6cc76eSGao Xiang int i; 2809f6cc76eSGao Xiang 2819f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 2829f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 2839f6cc76eSGao Xiang continue; 2849f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 2859f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 2869f6cc76eSGao Xiang } 2879f6cc76eSGao Xiang } 2889f6cc76eSGao Xiang 2899f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 2909f6cc76eSGao Xiang { 2919f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 2929f6cc76eSGao Xiang struct z_erofs_pcluster *a; 2939f6cc76eSGao Xiang unsigned int size; 2949f6cc76eSGao Xiang 2959f6cc76eSGao Xiang for (pcs = pcluster_pool; 2969f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 297ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 2989f6cc76eSGao Xiang 2999f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 3009f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 3019f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 3029f6cc76eSGao Xiang if (pcs->slab) 3039f6cc76eSGao Xiang continue; 3049f6cc76eSGao Xiang 3059f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 3069f6cc76eSGao Xiang return -ENOMEM; 3079f6cc76eSGao Xiang } 3089f6cc76eSGao Xiang return 0; 3099f6cc76eSGao Xiang } 3109f6cc76eSGao Xiang 3119f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 3129f6cc76eSGao Xiang { 3139f6cc76eSGao Xiang int i; 3149f6cc76eSGao Xiang 3159f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3169f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3179f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 3189f6cc76eSGao Xiang 3199f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 3209f6cc76eSGao Xiang continue; 3219f6cc76eSGao Xiang 3229f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 3239f6cc76eSGao Xiang if (!pcl) 3249f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 3259f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 3269f6cc76eSGao Xiang return pcl; 3279f6cc76eSGao Xiang } 3289f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 3299f6cc76eSGao Xiang } 3309f6cc76eSGao Xiang 3319f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 3329f6cc76eSGao Xiang { 333cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 3349f6cc76eSGao Xiang int i; 3359f6cc76eSGao Xiang 3369f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3379f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3389f6cc76eSGao Xiang 339cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 3409f6cc76eSGao Xiang continue; 3419f6cc76eSGao Xiang 3429f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 3439f6cc76eSGao Xiang return; 3449f6cc76eSGao Xiang } 3459f6cc76eSGao Xiang DBG_BUGON(1); 3469f6cc76eSGao Xiang } 3479f6cc76eSGao Xiang 34847e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 34947e4937aSGao Xiang 3503fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 3513fffb589SSandeep Dhavale static struct kthread_worker __rcu **z_erofs_pcpu_workers; 3523fffb589SSandeep Dhavale 3533fffb589SSandeep Dhavale static void erofs_destroy_percpu_workers(void) 35447e4937aSGao Xiang { 3553fffb589SSandeep Dhavale struct kthread_worker *worker; 3563fffb589SSandeep Dhavale unsigned int cpu; 3573fffb589SSandeep Dhavale 3583fffb589SSandeep Dhavale for_each_possible_cpu(cpu) { 3593fffb589SSandeep Dhavale worker = rcu_dereference_protected( 3603fffb589SSandeep Dhavale z_erofs_pcpu_workers[cpu], 1); 3613fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 3623fffb589SSandeep Dhavale if (worker) 3633fffb589SSandeep Dhavale kthread_destroy_worker(worker); 3643fffb589SSandeep Dhavale } 3653fffb589SSandeep Dhavale kfree(z_erofs_pcpu_workers); 36647e4937aSGao Xiang } 36747e4937aSGao Xiang 3683fffb589SSandeep Dhavale static struct kthread_worker *erofs_init_percpu_worker(int cpu) 36947e4937aSGao Xiang { 3703fffb589SSandeep Dhavale struct kthread_worker *worker = 3713fffb589SSandeep Dhavale kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 37247e4937aSGao Xiang 3733fffb589SSandeep Dhavale if (IS_ERR(worker)) 3743fffb589SSandeep Dhavale return worker; 3753fffb589SSandeep Dhavale if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 3763fffb589SSandeep Dhavale sched_set_fifo_low(worker->task); 3773fffb589SSandeep Dhavale return worker; 3783fffb589SSandeep Dhavale } 3793fffb589SSandeep Dhavale 3803fffb589SSandeep Dhavale static int erofs_init_percpu_workers(void) 3813fffb589SSandeep Dhavale { 3823fffb589SSandeep Dhavale struct kthread_worker *worker; 3833fffb589SSandeep Dhavale unsigned int cpu; 3843fffb589SSandeep Dhavale 3853fffb589SSandeep Dhavale z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 3863fffb589SSandeep Dhavale sizeof(struct kthread_worker *), GFP_ATOMIC); 3873fffb589SSandeep Dhavale if (!z_erofs_pcpu_workers) 3883fffb589SSandeep Dhavale return -ENOMEM; 3893fffb589SSandeep Dhavale 3903fffb589SSandeep Dhavale for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 3913fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 3923fffb589SSandeep Dhavale if (!IS_ERR(worker)) 3933fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 3943fffb589SSandeep Dhavale } 3953fffb589SSandeep Dhavale return 0; 3963fffb589SSandeep Dhavale } 3973fffb589SSandeep Dhavale #else 3983fffb589SSandeep Dhavale static inline void erofs_destroy_percpu_workers(void) {} 3993fffb589SSandeep Dhavale static inline int erofs_init_percpu_workers(void) { return 0; } 4003fffb589SSandeep Dhavale #endif 4013fffb589SSandeep Dhavale 4023fffb589SSandeep Dhavale #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 4033fffb589SSandeep Dhavale static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 4043fffb589SSandeep Dhavale static enum cpuhp_state erofs_cpuhp_state; 4053fffb589SSandeep Dhavale 4063fffb589SSandeep Dhavale static int erofs_cpu_online(unsigned int cpu) 4073fffb589SSandeep Dhavale { 4083fffb589SSandeep Dhavale struct kthread_worker *worker, *old; 4093fffb589SSandeep Dhavale 4103fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 4113fffb589SSandeep Dhavale if (IS_ERR(worker)) 4123fffb589SSandeep Dhavale return PTR_ERR(worker); 4133fffb589SSandeep Dhavale 4143fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4153fffb589SSandeep Dhavale old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4163fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4173fffb589SSandeep Dhavale if (!old) 4183fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 4193fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4203fffb589SSandeep Dhavale if (old) 4213fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4223fffb589SSandeep Dhavale return 0; 4233fffb589SSandeep Dhavale } 4243fffb589SSandeep Dhavale 4253fffb589SSandeep Dhavale static int erofs_cpu_offline(unsigned int cpu) 4263fffb589SSandeep Dhavale { 4273fffb589SSandeep Dhavale struct kthread_worker *worker; 4283fffb589SSandeep Dhavale 4293fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4303fffb589SSandeep Dhavale worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4313fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4323fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 4333fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4343fffb589SSandeep Dhavale 4353fffb589SSandeep Dhavale synchronize_rcu(); 4363fffb589SSandeep Dhavale if (worker) 4373fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4383fffb589SSandeep Dhavale return 0; 4393fffb589SSandeep Dhavale } 4403fffb589SSandeep Dhavale 4413fffb589SSandeep Dhavale static int erofs_cpu_hotplug_init(void) 4423fffb589SSandeep Dhavale { 4433fffb589SSandeep Dhavale int state; 4443fffb589SSandeep Dhavale 4453fffb589SSandeep Dhavale state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 4463fffb589SSandeep Dhavale "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 4473fffb589SSandeep Dhavale if (state < 0) 4483fffb589SSandeep Dhavale return state; 4493fffb589SSandeep Dhavale 4503fffb589SSandeep Dhavale erofs_cpuhp_state = state; 4513fffb589SSandeep Dhavale return 0; 4523fffb589SSandeep Dhavale } 4533fffb589SSandeep Dhavale 4543fffb589SSandeep Dhavale static void erofs_cpu_hotplug_destroy(void) 4553fffb589SSandeep Dhavale { 4563fffb589SSandeep Dhavale if (erofs_cpuhp_state) 4573fffb589SSandeep Dhavale cpuhp_remove_state_nocalls(erofs_cpuhp_state); 4583fffb589SSandeep Dhavale } 4593fffb589SSandeep Dhavale #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 4603fffb589SSandeep Dhavale static inline int erofs_cpu_hotplug_init(void) { return 0; } 4613fffb589SSandeep Dhavale static inline void erofs_cpu_hotplug_destroy(void) {} 4623fffb589SSandeep Dhavale #endif 4633fffb589SSandeep Dhavale 4643fffb589SSandeep Dhavale void z_erofs_exit_zip_subsystem(void) 4653fffb589SSandeep Dhavale { 4663fffb589SSandeep Dhavale erofs_cpu_hotplug_destroy(); 4673fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4683fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4693fffb589SSandeep Dhavale z_erofs_destroy_pcluster_pool(); 47047e4937aSGao Xiang } 47147e4937aSGao Xiang 47247e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 47347e4937aSGao Xiang { 4749f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 47547e4937aSGao Xiang 4769f6cc76eSGao Xiang if (err) 4773fffb589SSandeep Dhavale goto out_error_pcluster_pool; 4783fffb589SSandeep Dhavale 4793fffb589SSandeep Dhavale z_erofs_workqueue = alloc_workqueue("erofs_worker", 4803fffb589SSandeep Dhavale WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 4818d1b80a7SDan Carpenter if (!z_erofs_workqueue) { 4828d1b80a7SDan Carpenter err = -ENOMEM; 4833fffb589SSandeep Dhavale goto out_error_workqueue_init; 4848d1b80a7SDan Carpenter } 4853fffb589SSandeep Dhavale 4863fffb589SSandeep Dhavale err = erofs_init_percpu_workers(); 4879f6cc76eSGao Xiang if (err) 4883fffb589SSandeep Dhavale goto out_error_pcpu_worker; 4893fffb589SSandeep Dhavale 4903fffb589SSandeep Dhavale err = erofs_cpu_hotplug_init(); 4913fffb589SSandeep Dhavale if (err < 0) 4923fffb589SSandeep Dhavale goto out_error_cpuhp_init; 4933fffb589SSandeep Dhavale return err; 4943fffb589SSandeep Dhavale 4953fffb589SSandeep Dhavale out_error_cpuhp_init: 4963fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4973fffb589SSandeep Dhavale out_error_pcpu_worker: 4983fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4993fffb589SSandeep Dhavale out_error_workqueue_init: 5009f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 5013fffb589SSandeep Dhavale out_error_pcluster_pool: 5029f6cc76eSGao Xiang return err; 50347e4937aSGao Xiang } 50447e4937aSGao Xiang 505db166fc2SGao Xiang enum z_erofs_pclustermode { 506db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 50747e4937aSGao Xiang /* 508db166fc2SGao Xiang * The current pclusters was the tail of an exist chain, in addition 509db166fc2SGao Xiang * that the previous processed chained pclusters are all decided to 51047e4937aSGao Xiang * be hooked up to it. 511db166fc2SGao Xiang * A new chain will be created for the remaining pclusters which are 512db166fc2SGao Xiang * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED, 513db166fc2SGao Xiang * the next pcluster cannot reuse the whole page safely for inplace I/O 514db166fc2SGao Xiang * in the following scenario: 51547e4937aSGao Xiang * ________________________________________________________________ 51647e4937aSGao Xiang * | tail (partial) page | head (partial) page | 517db166fc2SGao Xiang * | (belongs to the next pcl) | (belongs to the current pcl) | 518db166fc2SGao Xiang * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________| 51947e4937aSGao Xiang */ 520db166fc2SGao Xiang Z_EROFS_PCLUSTER_HOOKED, 5210b964600SGao Xiang /* 522db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 5230b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 5240b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 525387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 5260b964600SGao Xiang */ 527db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 52847e4937aSGao Xiang /* 52947e4937aSGao Xiang * The current collection has been linked with the owned chain, and 53047e4937aSGao Xiang * could also be linked with the remaining collections, which means 53147e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 53247e4937aSGao Xiang * the current collection can safely use the whole page (since 53347e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 53447e4937aSGao Xiang * illustrated below: 53547e4937aSGao Xiang * ________________________________________________________________ 53647e4937aSGao Xiang * | tail (partial) page | head (partial) page | 53747e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 538db166fc2SGao Xiang * | PCLUSTER_FOLLOWED or | | 539db166fc2SGao Xiang * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________| 54047e4937aSGao Xiang * 54147e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 54247e4937aSGao Xiang */ 543db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 54447e4937aSGao Xiang }; 54547e4937aSGao Xiang 5465c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 5475c6dcc57SGao Xiang struct inode *const inode; 5485c6dcc57SGao Xiang struct erofs_map_blocks map; 54906a304cdSGao Xiang struct z_erofs_bvec_iter biter; 55047e4937aSGao Xiang 551*6ab5eed6SGao Xiang struct page *pagepool; 55206a304cdSGao Xiang struct page *candidate_bvpage; 55347e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 55447e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 555db166fc2SGao Xiang enum z_erofs_pclustermode mode; 55647e4937aSGao Xiang 55747e4937aSGao Xiang /* used for applying cache strategy on the fly */ 55847e4937aSGao Xiang bool backmost; 55947e4937aSGao Xiang erofs_off_t headoffset; 560ed722fbcSGao Xiang 561ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 562ed722fbcSGao Xiang unsigned int icur; 56347e4937aSGao Xiang }; 56447e4937aSGao Xiang 56547e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 5665c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 567db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 56847e4937aSGao Xiang 5691282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 5701282dea3SGao Xiang { 5711282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 5721282dea3SGao Xiang 5731282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 5741282dea3SGao Xiang return false; 5751282dea3SGao Xiang 5761282dea3SGao Xiang if (fe->backmost) 5771282dea3SGao Xiang return true; 5781282dea3SGao Xiang 5791282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 5801282dea3SGao Xiang fe->map.m_la < fe->headoffset) 5811282dea3SGao Xiang return true; 5821282dea3SGao Xiang 5831282dea3SGao Xiang return false; 5841282dea3SGao Xiang } 5851282dea3SGao Xiang 586*6ab5eed6SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe) 58747e4937aSGao Xiang { 5886f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 5895c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 5901282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 59147e4937aSGao Xiang bool standalone = true; 5926f39d1e1SGao Xiang /* 5936f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 5946f39d1e1SGao Xiang * can be used if low memory otherwise. 5956f39d1e1SGao Xiang */ 5961825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 5971825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 598ed722fbcSGao Xiang unsigned int i; 59947e4937aSGao Xiang 600db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 60147e4937aSGao Xiang return; 60247e4937aSGao Xiang 603ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 60447e4937aSGao Xiang struct page *page; 605b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 6061825c8d7SGao Xiang struct page *newpage = NULL; 60747e4937aSGao Xiang 60847e4937aSGao Xiang /* the compressed page was loaded before */ 609ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 61047e4937aSGao Xiang continue; 61147e4937aSGao Xiang 612ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 61347e4937aSGao Xiang 61447e4937aSGao Xiang if (page) { 615b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 6160b964600SGao Xiang } else { 6170b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 6180b964600SGao Xiang standalone = false; 6191282dea3SGao Xiang if (!shouldalloc) 6201282dea3SGao Xiang continue; 6211282dea3SGao Xiang 6221282dea3SGao Xiang /* 6231282dea3SGao Xiang * try to use cached I/O if page allocation 6241282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 6251282dea3SGao Xiang * to avoid any direct reclaim. 6261282dea3SGao Xiang */ 627*6ab5eed6SGao Xiang newpage = erofs_allocpage(&fe->pagepool, gfp); 6281825c8d7SGao Xiang if (!newpage) 62947e4937aSGao Xiang continue; 6301282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 631b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 63247e4937aSGao Xiang } 63347e4937aSGao Xiang 634b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 63547e4937aSGao Xiang continue; 63647e4937aSGao Xiang 637eaa9172aSGao Xiang if (page) 63847e4937aSGao Xiang put_page(page); 639eaa9172aSGao Xiang else if (newpage) 640*6ab5eed6SGao Xiang erofs_pagepool_add(&fe->pagepool, newpage); 64147e4937aSGao Xiang } 64247e4937aSGao Xiang 6430b964600SGao Xiang /* 6440b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 6450b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 6460b964600SGao Xiang */ 6470b964600SGao Xiang if (standalone) 648db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 64947e4937aSGao Xiang } 65047e4937aSGao Xiang 65147e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 65247e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 65347e4937aSGao Xiang struct erofs_workgroup *grp) 65447e4937aSGao Xiang { 65547e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 65647e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 65747e4937aSGao Xiang int i; 65847e4937aSGao Xiang 659cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 66047e4937aSGao Xiang /* 66147e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 66247e4937aSGao Xiang * therefore no need to worry about available decompression users. 66347e4937aSGao Xiang */ 6649f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 665ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 66647e4937aSGao Xiang 66747e4937aSGao Xiang if (!page) 66847e4937aSGao Xiang continue; 66947e4937aSGao Xiang 67047e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 67147e4937aSGao Xiang if (!trylock_page(page)) 67247e4937aSGao Xiang return -EBUSY; 67347e4937aSGao Xiang 674f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 67547e4937aSGao Xiang continue; 67647e4937aSGao Xiang 67747e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 678ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6796aaa7b06SGao Xiang detach_page_private(page); 68047e4937aSGao Xiang unlock_page(page); 68147e4937aSGao Xiang } 68247e4937aSGao Xiang return 0; 68347e4937aSGao Xiang } 68447e4937aSGao Xiang 685d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 68647e4937aSGao Xiang { 68747e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 688ed722fbcSGao Xiang int ret, i; 68947e4937aSGao Xiang 690ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 691ed722fbcSGao Xiang return 0; 69247e4937aSGao Xiang 693ed722fbcSGao Xiang ret = 0; 694cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 6959f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 696ed722fbcSGao Xiang if (pcl->compressed_bvecs[i].page == page) { 697ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 69847e4937aSGao Xiang ret = 1; 69947e4937aSGao Xiang break; 70047e4937aSGao Xiang } 70147e4937aSGao Xiang } 70247e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 7036aaa7b06SGao Xiang if (ret) 7046aaa7b06SGao Xiang detach_page_private(page); 70547e4937aSGao Xiang return ret; 70647e4937aSGao Xiang } 70747e4937aSGao Xiang 7085c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 709ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 71047e4937aSGao Xiang { 7115c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 71247e4937aSGao Xiang 713ed722fbcSGao Xiang while (fe->icur > 0) { 714ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 715ed722fbcSGao Xiang NULL, bvec->page)) { 716ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 71747e4937aSGao Xiang return true; 718ed722fbcSGao Xiang } 719ed722fbcSGao Xiang } 72047e4937aSGao Xiang return false; 72147e4937aSGao Xiang } 72247e4937aSGao Xiang 72387ca34a7SGao Xiang /* callers must be with pcluster lock held */ 7245c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 7255b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 72647e4937aSGao Xiang { 72747e4937aSGao Xiang int ret; 72847e4937aSGao Xiang 729db166fc2SGao Xiang if (exclusive) { 73006a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 731ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 73247e4937aSGao Xiang return 0; 73306a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 734db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 73506a304cdSGao Xiang !fe->candidate_bvpage) 73606a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 73706a304cdSGao Xiang } 738*6ab5eed6SGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage, 739*6ab5eed6SGao Xiang &fe->pagepool); 74006a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 74106a304cdSGao Xiang return ret; 74247e4937aSGao Xiang } 74347e4937aSGao Xiang 7445c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 74547e4937aSGao Xiang { 7465c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 7475c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 74847e4937aSGao Xiang 749473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 750473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 751473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 75247e4937aSGao Xiang *owned_head = &pcl->next; 753473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 754db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 755473e15b0SGao Xiang return; 756473e15b0SGao Xiang } 757473e15b0SGao Xiang 75847e4937aSGao Xiang /* 759473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 760473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 76147e4937aSGao Xiang */ 762267f2492SGao Xiang if (*owned_head != &pcl->next && pcl != f->tailpcl && 763267f2492SGao Xiang cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 764473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 76547e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 766db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_HOOKED; 7675c6dcc57SGao Xiang f->tailpcl = NULL; 768473e15b0SGao Xiang return; 76947e4937aSGao Xiang } 770473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 771db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 77247e4937aSGao Xiang } 77347e4937aSGao Xiang 77483a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 77547e4937aSGao Xiang { 77683a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 777cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 77847e4937aSGao Xiang struct z_erofs_pcluster *pcl; 77964094a04SGao Xiang struct erofs_workgroup *grp; 78047e4937aSGao Xiang int err; 78147e4937aSGao Xiang 782c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 783c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 7848f899262SGao Xiang DBG_BUGON(1); 7858f899262SGao Xiang return -EFSCORRUPTED; 7868f899262SGao Xiang } 7878f899262SGao Xiang 7889f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 789cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 790cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 7919f6cc76eSGao Xiang if (IS_ERR(pcl)) 7929f6cc76eSGao Xiang return PTR_ERR(pcl); 79347e4937aSGao Xiang 79464094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 7958f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 7962bfab9c0SGao Xiang pcl->length = 0; 7972bfab9c0SGao Xiang pcl->partial = true; 79847e4937aSGao Xiang 79947e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 8005c6dcc57SGao Xiang pcl->next = fe->owned_head; 80187ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 802db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 80347e4937aSGao Xiang 80447e4937aSGao Xiang /* 80547e4937aSGao Xiang * lock all primary followed works before visible to others 80647e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 80747e4937aSGao Xiang */ 80887ca34a7SGao Xiang mutex_init(&pcl->lock); 80987ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 81047e4937aSGao Xiang 811cecf864dSYue Hu if (ztailpacking) { 812cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 8133acea5fcSJingbo Xu pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 814cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 815cecf864dSYue Hu } else { 816cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 817cecf864dSYue Hu 81883a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 81964094a04SGao Xiang if (IS_ERR(grp)) { 82064094a04SGao Xiang err = PTR_ERR(grp); 82164094a04SGao Xiang goto err_out; 82264094a04SGao Xiang } 82364094a04SGao Xiang 82464094a04SGao Xiang if (grp != &pcl->obj) { 8255c6dcc57SGao Xiang fe->pcl = container_of(grp, 826cecf864dSYue Hu struct z_erofs_pcluster, obj); 82764094a04SGao Xiang err = -EEXIST; 82864094a04SGao Xiang goto err_out; 82947e4937aSGao Xiang } 830cecf864dSYue Hu } 83147e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 8325c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 8335c6dcc57SGao Xiang fe->tailpcl = pcl; 8345c6dcc57SGao Xiang fe->owned_head = &pcl->next; 8355c6dcc57SGao Xiang fe->pcl = pcl; 8369e579fc1SGao Xiang return 0; 83764094a04SGao Xiang 83864094a04SGao Xiang err_out: 83987ca34a7SGao Xiang mutex_unlock(&pcl->lock); 8409f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 84164094a04SGao Xiang return err; 84247e4937aSGao Xiang } 84347e4937aSGao Xiang 84483a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 84547e4937aSGao Xiang { 84683a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 8470d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 8489e579fc1SGao Xiang int ret; 84947e4937aSGao Xiang 85087ca34a7SGao Xiang DBG_BUGON(fe->pcl); 85147e4937aSGao Xiang 85287ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 8535c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 8545c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 85547e4937aSGao Xiang 8560d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 8570d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 8580d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 8590d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 86047e4937aSGao Xiang DBG_BUGON(1); 861cecf864dSYue Hu return -EFSCORRUPTED; 862cecf864dSYue Hu } 86347e4937aSGao Xiang 86464094a04SGao Xiang if (grp) { 8655c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 8660d823b42SGao Xiang ret = -EEXIST; 86764094a04SGao Xiang } else { 86883a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 86964094a04SGao Xiang } 87047e4937aSGao Xiang 8710d823b42SGao Xiang if (ret == -EEXIST) { 872267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 873267f2492SGao Xiang /* used to check tail merging loop due to corrupted images */ 874267f2492SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 875267f2492SGao Xiang fe->tailpcl = fe->pcl; 876267f2492SGao Xiang 877267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 8780d823b42SGao Xiang } else if (ret) { 8790d823b42SGao Xiang return ret; 8800d823b42SGao Xiang } 88106a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 882387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 88381382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 884ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 88547e4937aSGao Xiang return 0; 88647e4937aSGao Xiang } 88747e4937aSGao Xiang 88847e4937aSGao Xiang /* 88947e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 89047e4937aSGao Xiang * only after a RCU grace period. 89147e4937aSGao Xiang */ 89247e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 89347e4937aSGao Xiang { 89487ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 89587ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 89647e4937aSGao Xiang } 89747e4937aSGao Xiang 89847e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 89947e4937aSGao Xiang { 90047e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 90147e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 90247e4937aSGao Xiang 90387ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 90447e4937aSGao Xiang } 90547e4937aSGao Xiang 9065c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 90747e4937aSGao Xiang { 90887ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 90947e4937aSGao Xiang 91087ca34a7SGao Xiang if (!pcl) 91147e4937aSGao Xiang return false; 91247e4937aSGao Xiang 91306a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 91487ca34a7SGao Xiang mutex_unlock(&pcl->lock); 91547e4937aSGao Xiang 91605b63d2bSGao Xiang if (fe->candidate_bvpage) 91706a304cdSGao Xiang fe->candidate_bvpage = NULL; 91806a304cdSGao Xiang 91947e4937aSGao Xiang /* 92047e4937aSGao Xiang * if all pending pages are added, don't hold its reference 92147e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 92247e4937aSGao Xiang */ 923db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 92487ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 92547e4937aSGao Xiang 92687ca34a7SGao Xiang fe->pcl = NULL; 92747e4937aSGao Xiang return true; 92847e4937aSGao Xiang } 92947e4937aSGao Xiang 930b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 931b15b2e30SYue Hu struct page *page, unsigned int pageofs, 932b15b2e30SYue Hu unsigned int len) 933b15b2e30SYue Hu { 9343acea5fcSJingbo Xu struct super_block *sb = inode->i_sb; 935b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 936b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 937b15b2e30SYue Hu u8 *src, *dst; 938b15b2e30SYue Hu unsigned int i, cnt; 939b15b2e30SYue Hu 940e5126de1SYue Hu if (!packed_inode) 941e5126de1SYue Hu return -EFSCORRUPTED; 942e5126de1SYue Hu 943eb2c5e41SGao Xiang buf.inode = packed_inode; 944b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 945b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 946b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 9473acea5fcSJingbo Xu sb->s_blocksize - erofs_blkoff(sb, pos)); 948eb2c5e41SGao Xiang src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 949b15b2e30SYue Hu if (IS_ERR(src)) { 950b15b2e30SYue Hu erofs_put_metabuf(&buf); 951b15b2e30SYue Hu return PTR_ERR(src); 952b15b2e30SYue Hu } 953b15b2e30SYue Hu 954b15b2e30SYue Hu dst = kmap_local_page(page); 9553acea5fcSJingbo Xu memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 956b15b2e30SYue Hu kunmap_local(dst); 957b15b2e30SYue Hu pos += cnt; 958b15b2e30SYue Hu } 959b15b2e30SYue Hu erofs_put_metabuf(&buf); 960b15b2e30SYue Hu return 0; 961b15b2e30SYue Hu } 962b15b2e30SYue Hu 96347e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 964*6ab5eed6SGao Xiang struct page *page) 96547e4937aSGao Xiang { 96647e4937aSGao Xiang struct inode *const inode = fe->inode; 96747e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 96847e4937aSGao Xiang const loff_t offset = page_offset(page); 9695b220b20SGao Xiang bool tight = true, exclusive; 9702bfab9c0SGao Xiang unsigned int cur, end, spiltted; 97147e4937aSGao Xiang int err = 0; 97247e4937aSGao Xiang 97347e4937aSGao Xiang /* register locked file pages as online pages in pack */ 97447e4937aSGao Xiang z_erofs_onlinepage_init(page); 97547e4937aSGao Xiang 97647e4937aSGao Xiang spiltted = 0; 97747e4937aSGao Xiang end = PAGE_SIZE; 97847e4937aSGao Xiang repeat: 97947e4937aSGao Xiang cur = end - 1; 98047e4937aSGao Xiang 98139397a46SGao Xiang if (offset + cur < map->m_la || 98239397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 9835c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 98447e4937aSGao Xiang fe->backmost = false; 98547e4937aSGao Xiang map->m_la = offset + cur; 98647e4937aSGao Xiang map->m_llen = 0; 98747e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 9888d8a09b0SGao Xiang if (err) 98967148551SGao Xiang goto out; 99039397a46SGao Xiang } else { 99139397a46SGao Xiang if (fe->pcl) 99239397a46SGao Xiang goto hitted; 99339397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 99439397a46SGao Xiang } 99547e4937aSGao Xiang 996b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 997b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 99847e4937aSGao Xiang goto hitted; 99947e4937aSGao Xiang 100083a386c0SGao Xiang err = z_erofs_collector_begin(fe); 10018d8a09b0SGao Xiang if (err) 100267148551SGao Xiang goto out; 100347e4937aSGao Xiang 10045c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 100509c54379SGao Xiang void *mp; 1006cecf864dSYue Hu 100709c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 10083acea5fcSJingbo Xu erofs_blknr(inode->i_sb, map->m_pa), 10093acea5fcSJingbo Xu EROFS_NO_KMAP); 101009c54379SGao Xiang if (IS_ERR(mp)) { 101109c54379SGao Xiang err = PTR_ERR(mp); 1012cecf864dSYue Hu erofs_err(inode->i_sb, 1013cecf864dSYue Hu "failed to get inline page, err %d", err); 101467148551SGao Xiang goto out; 1015cecf864dSYue Hu } 101609c54379SGao Xiang get_page(fe->map.buf.page); 1017ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 1018ed722fbcSGao Xiang fe->map.buf.page); 1019db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 1020cecf864dSYue Hu } else { 10216f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 1022*6ab5eed6SGao Xiang z_erofs_bind_cache(fe); 1023cecf864dSYue Hu } 102447e4937aSGao Xiang hitted: 1025dc76ea8cSGao Xiang /* 1026dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 1027dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 1028dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 1029387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 1030dc76ea8cSGao Xiang */ 1031db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED && 1032db166fc2SGao Xiang fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1033dc76ea8cSGao Xiang 103447e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 10358d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 103647e4937aSGao Xiang zero_user_segment(page, cur, end); 103747e4937aSGao Xiang goto next_part; 103847e4937aSGao Xiang } 1039b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 1040b15b2e30SYue Hu unsigned int pageofs, skip, len; 1041b15b2e30SYue Hu 1042b15b2e30SYue Hu if (offset > map->m_la) { 1043b15b2e30SYue Hu pageofs = 0; 1044b15b2e30SYue Hu skip = offset - map->m_la; 1045b15b2e30SYue Hu } else { 1046b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 1047b15b2e30SYue Hu skip = 0; 1048b15b2e30SYue Hu } 1049b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 1050b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1051b15b2e30SYue Hu if (err) 1052b15b2e30SYue Hu goto out; 1053b15b2e30SYue Hu ++spiltted; 1054b15b2e30SYue Hu tight = false; 1055b15b2e30SYue Hu goto next_part; 1056b15b2e30SYue Hu } 105747e4937aSGao Xiang 10585b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 105947e4937aSGao Xiang if (cur) 1060db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 106147e4937aSGao Xiang 106206a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 106306a304cdSGao Xiang .page = page, 106406a304cdSGao Xiang .offset = offset - map->m_la, 106506a304cdSGao Xiang .end = end, 10665b220b20SGao Xiang }), exclusive); 106705b63d2bSGao Xiang if (err) 106867148551SGao Xiang goto out; 106947e4937aSGao Xiang 107067148551SGao Xiang z_erofs_onlinepage_split(page); 107147e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 107247e4937aSGao Xiang ++spiltted; 1073267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1074267f2492SGao Xiang fe->pcl->multibases = true; 10752bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 10762bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 10772bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 10782bfab9c0SGao Xiang } 1079e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1080e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1081e7933278SGao Xiang fe->pcl->length == map->m_llen) 1082e7933278SGao Xiang fe->pcl->partial = false; 108347e4937aSGao Xiang next_part: 10842bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 108547e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 10862bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 108747e4937aSGao Xiang 108847e4937aSGao Xiang end = cur; 108947e4937aSGao Xiang if (end > 0) 109047e4937aSGao Xiang goto repeat; 109147e4937aSGao Xiang 109247e4937aSGao Xiang out: 109367148551SGao Xiang if (err) 109467148551SGao Xiang z_erofs_page_mark_eio(page); 109547e4937aSGao Xiang z_erofs_onlinepage_endio(page); 109647e4937aSGao Xiang return err; 109747e4937aSGao Xiang } 109847e4937aSGao Xiang 1099ef4b4b46SYue Hu static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 110040452ffcSHuang Jianan unsigned int readahead_pages) 110140452ffcSHuang Jianan { 1102a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 110340452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 110440452ffcSHuang Jianan !readahead_pages) 110540452ffcSHuang Jianan return true; 110640452ffcSHuang Jianan 110740452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 110840452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 110940452ffcSHuang Jianan return true; 111040452ffcSHuang Jianan 111140452ffcSHuang Jianan return false; 111240452ffcSHuang Jianan } 111340452ffcSHuang Jianan 11146aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 11156aaa7b06SGao Xiang { 11166aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 11176aaa7b06SGao Xiang } 11186aaa7b06SGao Xiang 11194f05687fSGao Xiang struct z_erofs_decompress_backend { 11204f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 11214f05687fSGao Xiang struct super_block *sb; 11224f05687fSGao Xiang struct z_erofs_pcluster *pcl; 11234f05687fSGao Xiang 11244f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 11254f05687fSGao Xiang struct page **decompressed_pages; 11264f05687fSGao Xiang /* pages to keep the compressed data */ 11274f05687fSGao Xiang struct page **compressed_pages; 11284f05687fSGao Xiang 1129267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 11304f05687fSGao Xiang struct page **pagepool; 11312bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 11324f05687fSGao Xiang }; 11334f05687fSGao Xiang 1134267f2492SGao Xiang struct z_erofs_bvec_item { 1135267f2492SGao Xiang struct z_erofs_bvec bvec; 1136267f2492SGao Xiang struct list_head list; 1137267f2492SGao Xiang }; 1138267f2492SGao Xiang 1139267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 11403fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 11413fe96ee0SGao Xiang { 1142267f2492SGao Xiang struct z_erofs_bvec_item *item; 1143267f2492SGao Xiang 1144267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1145267f2492SGao Xiang unsigned int pgnr; 11463fe96ee0SGao Xiang 1147267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 11482bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 114963bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 11503fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1151267f2492SGao Xiang return; 11523fe96ee0SGao Xiang } 115363bbb856SGao Xiang } 11543fe96ee0SGao Xiang 1155267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1156267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1157267f2492SGao Xiang item->bvec = *bvec; 1158267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1159267f2492SGao Xiang } 1160267f2492SGao Xiang 1161267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1162267f2492SGao Xiang int err) 1163267f2492SGao Xiang { 1164267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1165267f2492SGao Xiang struct list_head *p, *n; 1166267f2492SGao Xiang 1167267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1168267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1169267f2492SGao Xiang unsigned int end, cur; 1170267f2492SGao Xiang void *dst, *src; 1171267f2492SGao Xiang 1172267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1173267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1174267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1175267f2492SGao Xiang bvi->bvec.end); 1176267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1177267f2492SGao Xiang while (cur < end) { 1178267f2492SGao Xiang unsigned int pgnr, scur, len; 1179267f2492SGao Xiang 1180267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1181267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1182267f2492SGao Xiang 1183267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1184267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1185267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1186267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1187267f2492SGao Xiang err = -EFSCORRUPTED; 1188267f2492SGao Xiang cur += len; 1189267f2492SGao Xiang continue; 1190267f2492SGao Xiang } 1191267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1192267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1193267f2492SGao Xiang kunmap_local(src); 1194267f2492SGao Xiang cur += len; 1195267f2492SGao Xiang } 1196267f2492SGao Xiang kunmap_local(dst); 1197267f2492SGao Xiang if (err) 1198267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 1199267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 1200267f2492SGao Xiang list_del(p); 1201267f2492SGao Xiang kfree(bvi); 1202267f2492SGao Xiang } 1203267f2492SGao Xiang } 1204267f2492SGao Xiang 1205267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 120642fec235SGao Xiang { 12074f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 120806a304cdSGao Xiang struct z_erofs_bvec_iter biter; 120906a304cdSGao Xiang struct page *old_bvpage; 1210267f2492SGao Xiang int i; 121142fec235SGao Xiang 1212387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 121342fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 121406a304cdSGao Xiang struct z_erofs_bvec bvec; 121542fec235SGao Xiang 121606a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 121742fec235SGao Xiang 121806a304cdSGao Xiang if (old_bvpage) 12194f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 122042fec235SGao Xiang 122106a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1222267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 122342fec235SGao Xiang } 122406a304cdSGao Xiang 122506a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 122606a304cdSGao Xiang if (old_bvpage) 12274f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 122842fec235SGao Xiang } 122942fec235SGao Xiang 12304f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 12314f05687fSGao Xiang bool *overlapped) 123267139e36SGao Xiang { 12334f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 123467139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 123567139e36SGao Xiang int i, err = 0; 123667139e36SGao Xiang 123767139e36SGao Xiang *overlapped = false; 123867139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1239ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1240ed722fbcSGao Xiang struct page *page = bvec->page; 124167139e36SGao Xiang 124267139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 124367139e36SGao Xiang if (!page) { 124467139e36SGao Xiang DBG_BUGON(1); 124567139e36SGao Xiang continue; 124667139e36SGao Xiang } 1247fe3e5914SGao Xiang be->compressed_pages[i] = page; 124867139e36SGao Xiang 124967139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 125067139e36SGao Xiang if (!PageUptodate(page)) 125167139e36SGao Xiang err = -EIO; 125267139e36SGao Xiang continue; 125367139e36SGao Xiang } 125467139e36SGao Xiang 125567139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 125667139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 12574f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 125867139e36SGao Xiang if (!PageUptodate(page)) 125967139e36SGao Xiang err = -EIO; 126067139e36SGao Xiang continue; 126167139e36SGao Xiang } 1262267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 126367139e36SGao Xiang *overlapped = true; 126467139e36SGao Xiang } 126567139e36SGao Xiang } 126667139e36SGao Xiang 1267fe3e5914SGao Xiang if (err) 12684f05687fSGao Xiang return err; 12694f05687fSGao Xiang return 0; 127067139e36SGao Xiang } 127167139e36SGao Xiang 12724f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 12734f05687fSGao Xiang int err) 127447e4937aSGao Xiang { 12754f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 12764f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1277cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1278597e2953SYue Hu const struct z_erofs_decompressor *decompressor = 1279597e2953SYue Hu &erofs_decompressors[pcl->algorithmformat]; 12802bfab9c0SGao Xiang unsigned int i, inputsize; 128167148551SGao Xiang int err2; 12822bfab9c0SGao Xiang struct page *page; 12832bfab9c0SGao Xiang bool overlapped; 128447e4937aSGao Xiang 128587ca34a7SGao Xiang mutex_lock(&pcl->lock); 12862bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 128747e4937aSGao Xiang 1288fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1289fe3e5914SGao Xiang be->decompressed_pages = NULL; 1290fe3e5914SGao Xiang be->compressed_pages = NULL; 1291fe3e5914SGao Xiang be->onstack_used = 0; 12922bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 12934f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 12942bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 12954f05687fSGao Xiang memset(be->decompressed_pages, 0, 12962bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1297fe3e5914SGao Xiang } 1298fe3e5914SGao Xiang 1299fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1300fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1301fe3e5914SGao Xiang 1302fe3e5914SGao Xiang if (!be->decompressed_pages) 13034f05687fSGao Xiang be->decompressed_pages = 1304647dd2c3SGao Xiang kvcalloc(be->nr_pages, sizeof(struct page *), 1305e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1306fe3e5914SGao Xiang if (!be->compressed_pages) 1307fe3e5914SGao Xiang be->compressed_pages = 1308647dd2c3SGao Xiang kvcalloc(pclusterpages, sizeof(struct page *), 1309fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 131047e4937aSGao Xiang 1311267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 13124f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 13134f05687fSGao Xiang if (err2) 13144f05687fSGao Xiang err = err2; 13158d8a09b0SGao Xiang if (err) 131647e4937aSGao Xiang goto out; 131747e4937aSGao Xiang 1318cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1319cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1320cecf864dSYue Hu else 1321cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1322cecf864dSYue Hu 1323597e2953SYue Hu err = decompressor->decompress(&(struct z_erofs_decompress_req) { 13244f05687fSGao Xiang .sb = be->sb, 13254f05687fSGao Xiang .in = be->compressed_pages, 13264f05687fSGao Xiang .out = be->decompressed_pages, 1327cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 132887ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 13299f6cc76eSGao Xiang .inputsize = inputsize, 13302bfab9c0SGao Xiang .outputsize = pcl->length, 133147e4937aSGao Xiang .alg = pcl->algorithmformat, 133247e4937aSGao Xiang .inplace_io = overlapped, 13332bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1334267f2492SGao Xiang .fillgaps = pcl->multibases, 13354f05687fSGao Xiang }, be->pagepool); 133647e4937aSGao Xiang 133747e4937aSGao Xiang out: 1338cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1339cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1340ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1341ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1342cecf864dSYue Hu put_page(page); 1343cecf864dSYue Hu } else { 1344cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1345ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 134647e4937aSGao Xiang 134747e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 134847e4937aSGao Xiang continue; 134947e4937aSGao Xiang 13506aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13514f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1352ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 135347e4937aSGao Xiang } 1354cecf864dSYue Hu } 1355fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1356fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1357647dd2c3SGao Xiang kvfree(be->compressed_pages); 1358267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 135947e4937aSGao Xiang 13602bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 13614f05687fSGao Xiang page = be->decompressed_pages[i]; 136247e4937aSGao Xiang if (!page) 136347e4937aSGao Xiang continue; 136447e4937aSGao Xiang 13656aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 136647e4937aSGao Xiang 13676aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13684f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 136947e4937aSGao Xiang continue; 137067148551SGao Xiang if (err) 137167148551SGao Xiang z_erofs_page_mark_eio(page); 137247e4937aSGao Xiang z_erofs_onlinepage_endio(page); 137347e4937aSGao Xiang } 137447e4937aSGao Xiang 13754f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 1376647dd2c3SGao Xiang kvfree(be->decompressed_pages); 137747e4937aSGao Xiang 13782bfab9c0SGao Xiang pcl->length = 0; 13792bfab9c0SGao Xiang pcl->partial = true; 1380267f2492SGao Xiang pcl->multibases = false; 138106a304cdSGao Xiang pcl->bvset.nextpage = NULL; 138287ca34a7SGao Xiang pcl->vcnt = 0; 138347e4937aSGao Xiang 138487ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 138547e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 138687ca34a7SGao Xiang mutex_unlock(&pcl->lock); 138747e4937aSGao Xiang return err; 138847e4937aSGao Xiang } 138947e4937aSGao Xiang 13900c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1391eaa9172aSGao Xiang struct page **pagepool) 139247e4937aSGao Xiang { 13934f05687fSGao Xiang struct z_erofs_decompress_backend be = { 13944f05687fSGao Xiang .sb = io->sb, 13954f05687fSGao Xiang .pagepool = pagepool, 1396267f2492SGao Xiang .decompressed_secondary_bvecs = 1397267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 13984f05687fSGao Xiang }; 139947e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 140047e4937aSGao Xiang 140147e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 14024f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 140347e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 14044f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */ 140547e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 140647e4937aSGao Xiang 14074f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 14084f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 140947e4937aSGao Xiang 14104f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 14114f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 141247e4937aSGao Xiang } 141347e4937aSGao Xiang } 141447e4937aSGao Xiang 14150c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 141647e4937aSGao Xiang { 1417a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1418a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1419eaa9172aSGao Xiang struct page *pagepool = NULL; 142047e4937aSGao Xiang 1421a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 14220c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 1423eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1424a4b1fab1SGao Xiang kvfree(bgq); 142547e4937aSGao Xiang } 142647e4937aSGao Xiang 14273fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14283fffb589SSandeep Dhavale static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 14293fffb589SSandeep Dhavale { 14303fffb589SSandeep Dhavale z_erofs_decompressqueue_work((struct work_struct *)work); 14313fffb589SSandeep Dhavale } 14323fffb589SSandeep Dhavale #endif 14333fffb589SSandeep Dhavale 14347865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1435cdba5506SGao Xiang int bios) 14367865827cSGao Xiang { 14377865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 14387865827cSGao Xiang 14397865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1440cdba5506SGao Xiang if (io->sync) { 14417865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 144260b30050SHongyu Jin complete(&io->u.done); 14437865827cSGao Xiang return; 14447865827cSGao Xiang } 14457865827cSGao Xiang 14467865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 14477865827cSGao Xiang return; 14483fffb589SSandeep Dhavale /* Use (kthread_)work and sync decompression for atomic contexts only */ 14497865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 14503fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14513fffb589SSandeep Dhavale struct kthread_worker *worker; 14523fffb589SSandeep Dhavale 14533fffb589SSandeep Dhavale rcu_read_lock(); 14543fffb589SSandeep Dhavale worker = rcu_dereference( 14553fffb589SSandeep Dhavale z_erofs_pcpu_workers[raw_smp_processor_id()]); 14563fffb589SSandeep Dhavale if (!worker) { 14573fffb589SSandeep Dhavale INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 14587865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 14593fffb589SSandeep Dhavale } else { 14603fffb589SSandeep Dhavale kthread_queue_work(worker, &io->u.kthread_work); 14613fffb589SSandeep Dhavale } 14623fffb589SSandeep Dhavale rcu_read_unlock(); 14633fffb589SSandeep Dhavale #else 14643fffb589SSandeep Dhavale queue_work(z_erofs_workqueue, &io->u.work); 14653fffb589SSandeep Dhavale #endif 14667865827cSGao Xiang /* enable sync decompression for readahead */ 14677865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 14687865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 14697865827cSGao Xiang return; 14707865827cSGao Xiang } 14717865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 14727865827cSGao Xiang } 14737865827cSGao Xiang 147447e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 147547e4937aSGao Xiang unsigned int nr, 1476eaa9172aSGao Xiang struct page **pagepool, 14779f2731d6SGao Xiang struct address_space *mc) 147847e4937aSGao Xiang { 147947e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 14809f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 148147e4937aSGao Xiang bool tocache = false; 148247e4937aSGao Xiang 148347e4937aSGao Xiang struct address_space *mapping; 148447e4937aSGao Xiang struct page *oldpage, *page; 148547e4937aSGao Xiang int justfound; 148647e4937aSGao Xiang 148747e4937aSGao Xiang repeat: 1488ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 148947e4937aSGao Xiang oldpage = page; 149047e4937aSGao Xiang 149147e4937aSGao Xiang if (!page) 149247e4937aSGao Xiang goto out_allocpage; 149347e4937aSGao Xiang 1494b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1495b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 149647e4937aSGao Xiang 14971825c8d7SGao Xiang /* 14981825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 14991825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 15001825c8d7SGao Xiang */ 15011825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1502ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 15031825c8d7SGao Xiang set_page_private(page, 0); 15041825c8d7SGao Xiang tocache = true; 15051825c8d7SGao Xiang goto out_tocache; 15061825c8d7SGao Xiang } 150747e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 150847e4937aSGao Xiang 150947e4937aSGao Xiang /* 15106aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 151147e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 151247e4937aSGao Xiang */ 151347e4937aSGao Xiang if (mapping && mapping != mc) 151447e4937aSGao Xiang /* ought to be unmanaged pages */ 151547e4937aSGao Xiang goto out; 151647e4937aSGao Xiang 15176aaa7b06SGao Xiang /* directly return for shortlived page as well */ 15186aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 15196aaa7b06SGao Xiang goto out; 15206aaa7b06SGao Xiang 152147e4937aSGao Xiang lock_page(page); 152247e4937aSGao Xiang 152347e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 152447e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 152547e4937aSGao Xiang 152647e4937aSGao Xiang /* the page is still in manage cache */ 152747e4937aSGao Xiang if (page->mapping == mc) { 1528ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 152947e4937aSGao Xiang 153047e4937aSGao Xiang if (!PagePrivate(page)) { 153147e4937aSGao Xiang /* 153247e4937aSGao Xiang * impossible to be !PagePrivate(page) for 153347e4937aSGao Xiang * the current restriction as well if 1534ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 153547e4937aSGao Xiang */ 153647e4937aSGao Xiang DBG_BUGON(!justfound); 153747e4937aSGao Xiang 153847e4937aSGao Xiang justfound = 0; 153947e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 154047e4937aSGao Xiang SetPagePrivate(page); 154147e4937aSGao Xiang } 154247e4937aSGao Xiang 154347e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 154447e4937aSGao Xiang if (PageUptodate(page)) { 154547e4937aSGao Xiang unlock_page(page); 154647e4937aSGao Xiang page = NULL; 154747e4937aSGao Xiang } 154847e4937aSGao Xiang goto out; 154947e4937aSGao Xiang } 155047e4937aSGao Xiang 155147e4937aSGao Xiang /* 155247e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 155347e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 155447e4937aSGao Xiang */ 155547e4937aSGao Xiang DBG_BUGON(page->mapping); 155647e4937aSGao Xiang DBG_BUGON(!justfound); 155747e4937aSGao Xiang 155847e4937aSGao Xiang tocache = true; 155947e4937aSGao Xiang unlock_page(page); 156047e4937aSGao Xiang put_page(page); 156147e4937aSGao Xiang out_allocpage: 15625ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1563ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1564ed722fbcSGao Xiang oldpage, page)) { 1565eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 15665ddcee1fSGao Xiang cond_resched(); 15675ddcee1fSGao Xiang goto repeat; 15685ddcee1fSGao Xiang } 15691825c8d7SGao Xiang out_tocache: 1570bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1571bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1572bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1573bf225074SGao Xiang goto out; 1574a30573b3SGao Xiang } 1575bf225074SGao Xiang attach_page_private(page, pcl); 1576bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1577bf225074SGao Xiang put_page(page); 1578bf225074SGao Xiang 157947e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 158047e4937aSGao Xiang return page; 158147e4937aSGao Xiang } 158247e4937aSGao Xiang 1583cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1584a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 158547e4937aSGao Xiang { 1586a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 158747e4937aSGao Xiang 1588a4b1fab1SGao Xiang if (fg && !*fg) { 1589a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1590a4b1fab1SGao Xiang if (!q) { 1591a4b1fab1SGao Xiang *fg = true; 1592a4b1fab1SGao Xiang goto fg_out; 159347e4937aSGao Xiang } 15943fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 15953fffb589SSandeep Dhavale kthread_init_work(&q->u.kthread_work, 15963fffb589SSandeep Dhavale z_erofs_decompressqueue_kthread_work); 15973fffb589SSandeep Dhavale #else 15980c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 15993fffb589SSandeep Dhavale #endif 1600a4b1fab1SGao Xiang } else { 1601a4b1fab1SGao Xiang fg_out: 1602a4b1fab1SGao Xiang q = fgq; 160360b30050SHongyu Jin init_completion(&fgq->u.done); 1604a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 160567148551SGao Xiang q->eio = false; 1606cdba5506SGao Xiang q->sync = true; 1607a4b1fab1SGao Xiang } 1608a4b1fab1SGao Xiang q->sb = sb; 1609a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1610a4b1fab1SGao Xiang return q; 161147e4937aSGao Xiang } 161247e4937aSGao Xiang 161347e4937aSGao Xiang /* define decompression jobqueue types */ 161447e4937aSGao Xiang enum { 161547e4937aSGao Xiang JQ_BYPASS, 161647e4937aSGao Xiang JQ_SUBMIT, 161747e4937aSGao Xiang NR_JOBQUEUES, 161847e4937aSGao Xiang }; 161947e4937aSGao Xiang 162047e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 162147e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 162247e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 162347e4937aSGao Xiang { 162447e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 162547e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 162647e4937aSGao Xiang 162747e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 162847e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 162947e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 163047e4937aSGao Xiang 163147e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 163247e4937aSGao Xiang 163347e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 163447e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 163547e4937aSGao Xiang 163647e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 163747e4937aSGao Xiang } 163847e4937aSGao Xiang 16397865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 16407865827cSGao Xiang { 1641cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 16427865827cSGao Xiang blk_status_t err = bio->bi_status; 16437865827cSGao Xiang struct bio_vec *bvec; 16447865827cSGao Xiang struct bvec_iter_all iter_all; 16457865827cSGao Xiang 16467865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 16477865827cSGao Xiang struct page *page = bvec->bv_page; 16487865827cSGao Xiang 16497865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 16507865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 16517865827cSGao Xiang 16527865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 16537865827cSGao Xiang if (!err) 16547865827cSGao Xiang SetPageUptodate(page); 16557865827cSGao Xiang unlock_page(page); 16567865827cSGao Xiang } 16577865827cSGao Xiang } 165867148551SGao Xiang if (err) 165967148551SGao Xiang q->eio = true; 1660cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 16617865827cSGao Xiang bio_put(bio); 16627865827cSGao Xiang } 16637865827cSGao Xiang 166483a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1665a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1666ef4b4b46SYue Hu bool *force_fg, bool readahead) 166747e4937aSGao Xiang { 166883a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 166983a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 167047e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1671a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 16725c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1673dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 16743f649ab7SKees Cook pgoff_t last_index; 1675dfeab2e9SGao Xiang struct block_device *last_bdev; 16761e4a2955SGao Xiang unsigned int nr_bios = 0; 16771e4a2955SGao Xiang struct bio *bio = NULL; 167882e60d00SJohannes Weiner unsigned long pflags; 167982e60d00SJohannes Weiner int memstall = 0; 168047e4937aSGao Xiang 1681cdba5506SGao Xiang /* 1682cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1683cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1684cdba5506SGao Xiang */ 1685cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1686cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1687cdba5506SGao Xiang 1688a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1689a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 169047e4937aSGao Xiang 169147e4937aSGao Xiang /* by default, all need io submission */ 169247e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 169347e4937aSGao Xiang 169447e4937aSGao Xiang do { 1695dfeab2e9SGao Xiang struct erofs_map_dev mdev; 169647e4937aSGao Xiang struct z_erofs_pcluster *pcl; 16971e4a2955SGao Xiang pgoff_t cur, end; 16981e4a2955SGao Xiang unsigned int i = 0; 16991e4a2955SGao Xiang bool bypass = true; 170047e4937aSGao Xiang 170147e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 170247e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 170347e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 170447e4937aSGao Xiang 170547e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 170647e4937aSGao Xiang 1707cecf864dSYue Hu /* close the main owned chain at first */ 1708cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1709cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1710cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1711cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1712cecf864dSYue Hu continue; 1713cecf864dSYue Hu } 1714cecf864dSYue Hu 1715dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1716dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 17173acea5fcSJingbo Xu .m_pa = erofs_pos(sb, pcl->obj.index), 1718dfeab2e9SGao Xiang }; 1719dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1720dfeab2e9SGao Xiang 17213acea5fcSJingbo Xu cur = erofs_blknr(sb, mdev.m_pa); 17229f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 172347e4937aSGao Xiang 17241e4a2955SGao Xiang do { 17251e4a2955SGao Xiang struct page *page; 172647e4937aSGao Xiang 1727*6ab5eed6SGao Xiang page = pickup_page_for_submission(pcl, i++, 1728*6ab5eed6SGao Xiang &f->pagepool, mc); 17291e4a2955SGao Xiang if (!page) 17301e4a2955SGao Xiang continue; 173147e4937aSGao Xiang 1732dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1733dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 173447e4937aSGao Xiang submit_bio_retry: 173594e4e153SGao Xiang submit_bio(bio); 173682e60d00SJohannes Weiner if (memstall) { 173782e60d00SJohannes Weiner psi_memstall_leave(&pflags); 173882e60d00SJohannes Weiner memstall = 0; 173982e60d00SJohannes Weiner } 174047e4937aSGao Xiang bio = NULL; 174147e4937aSGao Xiang } 174247e4937aSGao Xiang 174382e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 174499486c51SChristoph Hellwig psi_memstall_enter(&pflags); 174582e60d00SJohannes Weiner memstall = 1; 174682e60d00SJohannes Weiner } 174799486c51SChristoph Hellwig 174847e4937aSGao Xiang if (!bio) { 174907888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 175007888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 17510c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1752dfeab2e9SGao Xiang 1753dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 17541e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 17553acea5fcSJingbo Xu (sb->s_blocksize_bits - 9); 1756cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 1757ef4b4b46SYue Hu if (readahead) 17586ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 175947e4937aSGao Xiang ++nr_bios; 176047e4937aSGao Xiang } 176147e4937aSGao Xiang 17626c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 176347e4937aSGao Xiang goto submit_bio_retry; 176447e4937aSGao Xiang 17651e4a2955SGao Xiang last_index = cur; 17661e4a2955SGao Xiang bypass = false; 17671e4a2955SGao Xiang } while (++cur < end); 176847e4937aSGao Xiang 17691e4a2955SGao Xiang if (!bypass) 177047e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 177147e4937aSGao Xiang else 177247e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 177347e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 177447e4937aSGao Xiang 177599486c51SChristoph Hellwig if (bio) { 177694e4e153SGao Xiang submit_bio(bio); 177782e60d00SJohannes Weiner if (memstall) 177882e60d00SJohannes Weiner psi_memstall_leave(&pflags); 177999486c51SChristoph Hellwig } 178047e4937aSGao Xiang 1781587a67b7SGao Xiang /* 1782587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 17833fffb589SSandeep Dhavale * don't issue decompression but drop it directly instead. 1784587a67b7SGao Xiang */ 1785587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1786587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 17871e4a2955SGao Xiang return; 1788587a67b7SGao Xiang } 1789cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 179047e4937aSGao Xiang } 179147e4937aSGao Xiang 179283a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1793*6ab5eed6SGao Xiang bool force_fg, bool ra) 179447e4937aSGao Xiang { 1795a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 179647e4937aSGao Xiang 17975c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 179847e4937aSGao Xiang return; 1799*6ab5eed6SGao Xiang z_erofs_submit_queue(f, io, &force_fg, ra); 180047e4937aSGao Xiang 18010c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 1802*6ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], &f->pagepool); 180347e4937aSGao Xiang 180447e4937aSGao Xiang if (!force_fg) 180547e4937aSGao Xiang return; 180647e4937aSGao Xiang 180747e4937aSGao Xiang /* wait until all bios are completed */ 180860b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 180947e4937aSGao Xiang 18100c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 1811*6ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], &f->pagepool); 181247e4937aSGao Xiang } 181347e4937aSGao Xiang 181438629291SGao Xiang /* 181538629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 181638629291SGao Xiang * approximate readmore strategies as a start. 181738629291SGao Xiang */ 181838629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 1819*6ab5eed6SGao Xiang struct readahead_control *rac, bool backmost) 182038629291SGao Xiang { 182138629291SGao Xiang struct inode *inode = f->inode; 182238629291SGao Xiang struct erofs_map_blocks *map = &f->map; 1823796e9149SYue Hu erofs_off_t cur, end, headoffset = f->headoffset; 182438629291SGao Xiang int err; 182538629291SGao Xiang 182638629291SGao Xiang if (backmost) { 1827796e9149SYue Hu if (rac) 1828796e9149SYue Hu end = headoffset + readahead_length(rac) - 1; 1829796e9149SYue Hu else 1830796e9149SYue Hu end = headoffset + PAGE_SIZE - 1; 183138629291SGao Xiang map->m_la = end; 1832622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1833622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 183438629291SGao Xiang if (err) 183538629291SGao Xiang return; 183638629291SGao Xiang 1837796e9149SYue Hu /* expand ra for the trailing edge if readahead */ 183838629291SGao Xiang if (rac) { 183938629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1840796e9149SYue Hu readahead_expand(rac, headoffset, cur - headoffset); 184138629291SGao Xiang return; 184238629291SGao Xiang } 184338629291SGao Xiang end = round_up(end, PAGE_SIZE); 184438629291SGao Xiang } else { 184538629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 184638629291SGao Xiang 184738629291SGao Xiang if (!map->m_llen) 184838629291SGao Xiang return; 184938629291SGao Xiang } 185038629291SGao Xiang 185138629291SGao Xiang cur = map->m_la + map->m_llen - 1; 185238629291SGao Xiang while (cur >= end) { 185338629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 185438629291SGao Xiang struct page *page; 185538629291SGao Xiang 185638629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1857aa793b46SGao Xiang if (page) { 185838629291SGao Xiang if (PageUptodate(page)) { 185938629291SGao Xiang unlock_page(page); 1860aa793b46SGao Xiang } else { 1861*6ab5eed6SGao Xiang err = z_erofs_do_read_page(f, page); 186238629291SGao Xiang if (err) 186338629291SGao Xiang erofs_err(inode->i_sb, 186438629291SGao Xiang "readmore error at page %lu @ nid %llu", 186538629291SGao Xiang index, EROFS_I(inode)->nid); 1866aa793b46SGao Xiang } 186738629291SGao Xiang put_page(page); 1868aa793b46SGao Xiang } 1869aa793b46SGao Xiang 187038629291SGao Xiang if (cur < PAGE_SIZE) 187138629291SGao Xiang break; 187238629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 187338629291SGao Xiang } 187438629291SGao Xiang } 187538629291SGao Xiang 1876a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 187747e4937aSGao Xiang { 1878a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 187947e4937aSGao Xiang struct inode *const inode = page->mapping->host; 188040452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 188147e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 188247e4937aSGao Xiang int err; 188347e4937aSGao Xiang 188447e4937aSGao Xiang trace_erofs_readpage(page, false); 188547e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 188647e4937aSGao Xiang 1887*6ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, true); 1888*6ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 1889*6ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, false); 18905c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 189147e4937aSGao Xiang 189247e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 1893*6ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, 0), false); 189447e4937aSGao Xiang 189547e4937aSGao Xiang if (err) 18964f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 189747e4937aSGao Xiang 189809c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1899*6ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 190047e4937aSGao Xiang return err; 190147e4937aSGao Xiang } 190247e4937aSGao Xiang 19030615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 190447e4937aSGao Xiang { 19050615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 190647e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 190747e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1908*6ab5eed6SGao Xiang struct page *head = NULL, *page; 190938629291SGao Xiang unsigned int nr_pages; 191047e4937aSGao Xiang 19110615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 191247e4937aSGao Xiang 1913*6ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, true); 191438629291SGao Xiang nr_pages = readahead_count(rac); 191538629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 191638629291SGao Xiang 19170615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 191847e4937aSGao Xiang set_page_private(page, (unsigned long)head); 191947e4937aSGao Xiang head = page; 192047e4937aSGao Xiang } 192147e4937aSGao Xiang 192247e4937aSGao Xiang while (head) { 192347e4937aSGao Xiang struct page *page = head; 192447e4937aSGao Xiang int err; 192547e4937aSGao Xiang 192647e4937aSGao Xiang /* traversal in reverse order */ 192747e4937aSGao Xiang head = (void *)page_private(page); 192847e4937aSGao Xiang 1929*6ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 1930a5876e24SGao Xiang if (err) 19314f761fa2SGao Xiang erofs_err(inode->i_sb, 19324f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 19334f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 193447e4937aSGao Xiang put_page(page); 193547e4937aSGao Xiang } 1936*6ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, false); 19375c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 193847e4937aSGao Xiang 1939*6ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, nr_pages), true); 194009c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1941*6ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 194247e4937aSGao Xiang } 194347e4937aSGao Xiang 19440c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1945a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 19460615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 194747e4937aSGao Xiang }; 1948