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 96*967c28b2SGao Xiang /* the end of a chain of pclusters */ 97a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE) 98a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_NIL (NULL) 99a9a94d93SGao Xiang 100a9a94d93SGao Xiang struct z_erofs_decompressqueue { 101a9a94d93SGao Xiang struct super_block *sb; 102a9a94d93SGao Xiang atomic_t pending_bios; 103a9a94d93SGao Xiang z_erofs_next_pcluster_t head; 104a9a94d93SGao Xiang 105a9a94d93SGao Xiang union { 106a9a94d93SGao Xiang struct completion done; 107a9a94d93SGao Xiang struct work_struct work; 1083fffb589SSandeep Dhavale struct kthread_work kthread_work; 109a9a94d93SGao Xiang } u; 110a9a94d93SGao Xiang bool eio, sync; 111a9a94d93SGao Xiang }; 112a9a94d93SGao Xiang 113a9a94d93SGao Xiang static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 114a9a94d93SGao Xiang { 115a9a94d93SGao Xiang return !pcl->obj.index; 116a9a94d93SGao Xiang } 117a9a94d93SGao Xiang 118a9a94d93SGao Xiang static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 119a9a94d93SGao Xiang { 120a9a94d93SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) 121a9a94d93SGao Xiang return 1; 122a9a94d93SGao Xiang return pcl->pclusterpages; 123a9a94d93SGao Xiang } 124a9a94d93SGao Xiang 125a9a94d93SGao Xiang /* 126a9a94d93SGao Xiang * bit 30: I/O error occurred on this page 127a9a94d93SGao Xiang * bit 0 - 29: remaining parts to complete this page 128a9a94d93SGao Xiang */ 129a9a94d93SGao Xiang #define Z_EROFS_PAGE_EIO (1 << 30) 130a9a94d93SGao Xiang 131a9a94d93SGao Xiang static inline void z_erofs_onlinepage_init(struct page *page) 132a9a94d93SGao Xiang { 133a9a94d93SGao Xiang union { 134a9a94d93SGao Xiang atomic_t o; 135a9a94d93SGao Xiang unsigned long v; 136a9a94d93SGao Xiang } u = { .o = ATOMIC_INIT(1) }; 137a9a94d93SGao Xiang 138a9a94d93SGao Xiang set_page_private(page, u.v); 139a9a94d93SGao Xiang smp_wmb(); 140a9a94d93SGao Xiang SetPagePrivate(page); 141a9a94d93SGao Xiang } 142a9a94d93SGao Xiang 143a9a94d93SGao Xiang static inline void z_erofs_onlinepage_split(struct page *page) 144a9a94d93SGao Xiang { 145a9a94d93SGao Xiang atomic_inc((atomic_t *)&page->private); 146a9a94d93SGao Xiang } 147a9a94d93SGao Xiang 148a9a94d93SGao Xiang static inline void z_erofs_page_mark_eio(struct page *page) 149a9a94d93SGao Xiang { 150a9a94d93SGao Xiang int orig; 151a9a94d93SGao Xiang 152a9a94d93SGao Xiang do { 153a9a94d93SGao Xiang orig = atomic_read((atomic_t *)&page->private); 154a9a94d93SGao Xiang } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 155a9a94d93SGao Xiang orig | Z_EROFS_PAGE_EIO) != orig); 156a9a94d93SGao Xiang } 157a9a94d93SGao Xiang 158a9a94d93SGao Xiang static inline void z_erofs_onlinepage_endio(struct page *page) 159a9a94d93SGao Xiang { 160a9a94d93SGao Xiang unsigned int v; 161a9a94d93SGao Xiang 162a9a94d93SGao Xiang DBG_BUGON(!PagePrivate(page)); 163a9a94d93SGao Xiang v = atomic_dec_return((atomic_t *)&page->private); 164a9a94d93SGao Xiang if (!(v & ~Z_EROFS_PAGE_EIO)) { 165a9a94d93SGao Xiang set_page_private(page, 0); 166a9a94d93SGao Xiang ClearPagePrivate(page); 167a9a94d93SGao Xiang if (!(v & Z_EROFS_PAGE_EIO)) 168a9a94d93SGao Xiang SetPageUptodate(page); 169a9a94d93SGao Xiang unlock_page(page); 170a9a94d93SGao Xiang } 171a9a94d93SGao Xiang } 172a9a94d93SGao Xiang 173a9a94d93SGao Xiang #define Z_EROFS_ONSTACK_PAGES 32 174a9a94d93SGao Xiang 17547e4937aSGao Xiang /* 1769f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 1779f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 1789f6cc76eSGao Xiang */ 1799f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 1809f6cc76eSGao Xiang struct kmem_cache *slab; 1819f6cc76eSGao Xiang unsigned int maxpages; 1829f6cc76eSGao Xiang char name[48]; 1839f6cc76eSGao Xiang }; 1849f6cc76eSGao Xiang 1859f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 1869f6cc76eSGao Xiang 1879f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 1889f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 1899f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 1909f6cc76eSGao Xiang }; 1919f6cc76eSGao Xiang 19206a304cdSGao Xiang struct z_erofs_bvec_iter { 19306a304cdSGao Xiang struct page *bvpage; 19406a304cdSGao Xiang struct z_erofs_bvset *bvset; 19506a304cdSGao Xiang unsigned int nr, cur; 19606a304cdSGao Xiang }; 19706a304cdSGao Xiang 19806a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 19906a304cdSGao Xiang { 20006a304cdSGao Xiang if (iter->bvpage) 20106a304cdSGao Xiang kunmap_local(iter->bvset); 20206a304cdSGao Xiang return iter->bvpage; 20306a304cdSGao Xiang } 20406a304cdSGao Xiang 20506a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 20606a304cdSGao Xiang { 20706a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 20806a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 20906a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 21006a304cdSGao Xiang struct page *oldpage; 21106a304cdSGao Xiang 21206a304cdSGao Xiang DBG_BUGON(!nextpage); 21306a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 21406a304cdSGao Xiang iter->bvpage = nextpage; 21506a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 21606a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 21706a304cdSGao Xiang iter->cur = 0; 21806a304cdSGao Xiang return oldpage; 21906a304cdSGao Xiang } 22006a304cdSGao Xiang 22106a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 22206a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 22306a304cdSGao Xiang unsigned int bootstrap_nr, 22406a304cdSGao Xiang unsigned int cur) 22506a304cdSGao Xiang { 22606a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 22706a304cdSGao Xiang .nr = bootstrap_nr, 22806a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 22906a304cdSGao Xiang }; 23006a304cdSGao Xiang 23106a304cdSGao Xiang while (cur > iter->nr) { 23206a304cdSGao Xiang cur -= iter->nr; 23306a304cdSGao Xiang z_erofs_bvset_flip(iter); 23406a304cdSGao Xiang } 23506a304cdSGao Xiang iter->cur = cur; 23606a304cdSGao Xiang } 23706a304cdSGao Xiang 23806a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 23906a304cdSGao Xiang struct z_erofs_bvec *bvec, 2406ab5eed6SGao Xiang struct page **candidate_bvpage, 2416ab5eed6SGao Xiang struct page **pagepool) 24206a304cdSGao Xiang { 24305b63d2bSGao Xiang if (iter->cur >= iter->nr) { 24405b63d2bSGao Xiang struct page *nextpage = *candidate_bvpage; 24506a304cdSGao Xiang 24605b63d2bSGao Xiang if (!nextpage) { 2476ab5eed6SGao Xiang nextpage = erofs_allocpage(pagepool, GFP_NOFS); 24805b63d2bSGao Xiang if (!nextpage) 24905b63d2bSGao Xiang return -ENOMEM; 25005b63d2bSGao Xiang set_page_private(nextpage, Z_EROFS_SHORTLIVED_PAGE); 25105b63d2bSGao Xiang } 25206a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 25305b63d2bSGao Xiang iter->bvset->nextpage = nextpage; 25406a304cdSGao Xiang z_erofs_bvset_flip(iter); 25506a304cdSGao Xiang 25606a304cdSGao Xiang iter->bvset->nextpage = NULL; 25706a304cdSGao Xiang *candidate_bvpage = NULL; 25806a304cdSGao Xiang } 25906a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 26006a304cdSGao Xiang return 0; 26106a304cdSGao Xiang } 26206a304cdSGao Xiang 26306a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 26406a304cdSGao Xiang struct z_erofs_bvec *bvec, 26506a304cdSGao Xiang struct page **old_bvpage) 26606a304cdSGao Xiang { 26706a304cdSGao Xiang if (iter->cur == iter->nr) 26806a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 26906a304cdSGao Xiang else 27006a304cdSGao Xiang *old_bvpage = NULL; 27106a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 27206a304cdSGao Xiang } 27306a304cdSGao Xiang 2749f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 2759f6cc76eSGao Xiang { 2769f6cc76eSGao Xiang int i; 2779f6cc76eSGao Xiang 2789f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 2799f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 2809f6cc76eSGao Xiang continue; 2819f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 2829f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 2839f6cc76eSGao Xiang } 2849f6cc76eSGao Xiang } 2859f6cc76eSGao Xiang 2869f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 2879f6cc76eSGao Xiang { 2889f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 2899f6cc76eSGao Xiang struct z_erofs_pcluster *a; 2909f6cc76eSGao Xiang unsigned int size; 2919f6cc76eSGao Xiang 2929f6cc76eSGao Xiang for (pcs = pcluster_pool; 2939f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 294ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 2959f6cc76eSGao Xiang 2969f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 2979f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 2989f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 2999f6cc76eSGao Xiang if (pcs->slab) 3009f6cc76eSGao Xiang continue; 3019f6cc76eSGao Xiang 3029f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 3039f6cc76eSGao Xiang return -ENOMEM; 3049f6cc76eSGao Xiang } 3059f6cc76eSGao Xiang return 0; 3069f6cc76eSGao Xiang } 3079f6cc76eSGao Xiang 3089f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 3099f6cc76eSGao Xiang { 3109f6cc76eSGao Xiang int i; 3119f6cc76eSGao Xiang 3129f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3139f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3149f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 3159f6cc76eSGao Xiang 3169f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 3179f6cc76eSGao Xiang continue; 3189f6cc76eSGao Xiang 3199f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 3209f6cc76eSGao Xiang if (!pcl) 3219f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 3229f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 3239f6cc76eSGao Xiang return pcl; 3249f6cc76eSGao Xiang } 3259f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 3269f6cc76eSGao Xiang } 3279f6cc76eSGao Xiang 3289f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 3299f6cc76eSGao Xiang { 330cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 3319f6cc76eSGao Xiang int i; 3329f6cc76eSGao Xiang 3339f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3349f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3359f6cc76eSGao Xiang 336cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 3379f6cc76eSGao Xiang continue; 3389f6cc76eSGao Xiang 3399f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 3409f6cc76eSGao Xiang return; 3419f6cc76eSGao Xiang } 3429f6cc76eSGao Xiang DBG_BUGON(1); 3439f6cc76eSGao Xiang } 3449f6cc76eSGao Xiang 34547e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 34647e4937aSGao Xiang 3473fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 3483fffb589SSandeep Dhavale static struct kthread_worker __rcu **z_erofs_pcpu_workers; 3493fffb589SSandeep Dhavale 3503fffb589SSandeep Dhavale static void erofs_destroy_percpu_workers(void) 35147e4937aSGao Xiang { 3523fffb589SSandeep Dhavale struct kthread_worker *worker; 3533fffb589SSandeep Dhavale unsigned int cpu; 3543fffb589SSandeep Dhavale 3553fffb589SSandeep Dhavale for_each_possible_cpu(cpu) { 3563fffb589SSandeep Dhavale worker = rcu_dereference_protected( 3573fffb589SSandeep Dhavale z_erofs_pcpu_workers[cpu], 1); 3583fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 3593fffb589SSandeep Dhavale if (worker) 3603fffb589SSandeep Dhavale kthread_destroy_worker(worker); 3613fffb589SSandeep Dhavale } 3623fffb589SSandeep Dhavale kfree(z_erofs_pcpu_workers); 36347e4937aSGao Xiang } 36447e4937aSGao Xiang 3653fffb589SSandeep Dhavale static struct kthread_worker *erofs_init_percpu_worker(int cpu) 36647e4937aSGao Xiang { 3673fffb589SSandeep Dhavale struct kthread_worker *worker = 3683fffb589SSandeep Dhavale kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 36947e4937aSGao Xiang 3703fffb589SSandeep Dhavale if (IS_ERR(worker)) 3713fffb589SSandeep Dhavale return worker; 3723fffb589SSandeep Dhavale if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 3733fffb589SSandeep Dhavale sched_set_fifo_low(worker->task); 3743fffb589SSandeep Dhavale return worker; 3753fffb589SSandeep Dhavale } 3763fffb589SSandeep Dhavale 3773fffb589SSandeep Dhavale static int erofs_init_percpu_workers(void) 3783fffb589SSandeep Dhavale { 3793fffb589SSandeep Dhavale struct kthread_worker *worker; 3803fffb589SSandeep Dhavale unsigned int cpu; 3813fffb589SSandeep Dhavale 3823fffb589SSandeep Dhavale z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 3833fffb589SSandeep Dhavale sizeof(struct kthread_worker *), GFP_ATOMIC); 3843fffb589SSandeep Dhavale if (!z_erofs_pcpu_workers) 3853fffb589SSandeep Dhavale return -ENOMEM; 3863fffb589SSandeep Dhavale 3873fffb589SSandeep Dhavale for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 3883fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 3893fffb589SSandeep Dhavale if (!IS_ERR(worker)) 3903fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 3913fffb589SSandeep Dhavale } 3923fffb589SSandeep Dhavale return 0; 3933fffb589SSandeep Dhavale } 3943fffb589SSandeep Dhavale #else 3953fffb589SSandeep Dhavale static inline void erofs_destroy_percpu_workers(void) {} 3963fffb589SSandeep Dhavale static inline int erofs_init_percpu_workers(void) { return 0; } 3973fffb589SSandeep Dhavale #endif 3983fffb589SSandeep Dhavale 3993fffb589SSandeep Dhavale #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 4003fffb589SSandeep Dhavale static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 4013fffb589SSandeep Dhavale static enum cpuhp_state erofs_cpuhp_state; 4023fffb589SSandeep Dhavale 4033fffb589SSandeep Dhavale static int erofs_cpu_online(unsigned int cpu) 4043fffb589SSandeep Dhavale { 4053fffb589SSandeep Dhavale struct kthread_worker *worker, *old; 4063fffb589SSandeep Dhavale 4073fffb589SSandeep Dhavale worker = erofs_init_percpu_worker(cpu); 4083fffb589SSandeep Dhavale if (IS_ERR(worker)) 4093fffb589SSandeep Dhavale return PTR_ERR(worker); 4103fffb589SSandeep Dhavale 4113fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4123fffb589SSandeep Dhavale old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4133fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4143fffb589SSandeep Dhavale if (!old) 4153fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 4163fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4173fffb589SSandeep Dhavale if (old) 4183fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4193fffb589SSandeep Dhavale return 0; 4203fffb589SSandeep Dhavale } 4213fffb589SSandeep Dhavale 4223fffb589SSandeep Dhavale static int erofs_cpu_offline(unsigned int cpu) 4233fffb589SSandeep Dhavale { 4243fffb589SSandeep Dhavale struct kthread_worker *worker; 4253fffb589SSandeep Dhavale 4263fffb589SSandeep Dhavale spin_lock(&z_erofs_pcpu_worker_lock); 4273fffb589SSandeep Dhavale worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 4283fffb589SSandeep Dhavale lockdep_is_held(&z_erofs_pcpu_worker_lock)); 4293fffb589SSandeep Dhavale rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 4303fffb589SSandeep Dhavale spin_unlock(&z_erofs_pcpu_worker_lock); 4313fffb589SSandeep Dhavale 4323fffb589SSandeep Dhavale synchronize_rcu(); 4333fffb589SSandeep Dhavale if (worker) 4343fffb589SSandeep Dhavale kthread_destroy_worker(worker); 4353fffb589SSandeep Dhavale return 0; 4363fffb589SSandeep Dhavale } 4373fffb589SSandeep Dhavale 4383fffb589SSandeep Dhavale static int erofs_cpu_hotplug_init(void) 4393fffb589SSandeep Dhavale { 4403fffb589SSandeep Dhavale int state; 4413fffb589SSandeep Dhavale 4423fffb589SSandeep Dhavale state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 4433fffb589SSandeep Dhavale "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 4443fffb589SSandeep Dhavale if (state < 0) 4453fffb589SSandeep Dhavale return state; 4463fffb589SSandeep Dhavale 4473fffb589SSandeep Dhavale erofs_cpuhp_state = state; 4483fffb589SSandeep Dhavale return 0; 4493fffb589SSandeep Dhavale } 4503fffb589SSandeep Dhavale 4513fffb589SSandeep Dhavale static void erofs_cpu_hotplug_destroy(void) 4523fffb589SSandeep Dhavale { 4533fffb589SSandeep Dhavale if (erofs_cpuhp_state) 4543fffb589SSandeep Dhavale cpuhp_remove_state_nocalls(erofs_cpuhp_state); 4553fffb589SSandeep Dhavale } 4563fffb589SSandeep Dhavale #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 4573fffb589SSandeep Dhavale static inline int erofs_cpu_hotplug_init(void) { return 0; } 4583fffb589SSandeep Dhavale static inline void erofs_cpu_hotplug_destroy(void) {} 4593fffb589SSandeep Dhavale #endif 4603fffb589SSandeep Dhavale 4613fffb589SSandeep Dhavale void z_erofs_exit_zip_subsystem(void) 4623fffb589SSandeep Dhavale { 4633fffb589SSandeep Dhavale erofs_cpu_hotplug_destroy(); 4643fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4653fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4663fffb589SSandeep Dhavale z_erofs_destroy_pcluster_pool(); 46747e4937aSGao Xiang } 46847e4937aSGao Xiang 46947e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 47047e4937aSGao Xiang { 4719f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 47247e4937aSGao Xiang 4739f6cc76eSGao Xiang if (err) 4743fffb589SSandeep Dhavale goto out_error_pcluster_pool; 4753fffb589SSandeep Dhavale 4763fffb589SSandeep Dhavale z_erofs_workqueue = alloc_workqueue("erofs_worker", 4773fffb589SSandeep Dhavale WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 4788d1b80a7SDan Carpenter if (!z_erofs_workqueue) { 4798d1b80a7SDan Carpenter err = -ENOMEM; 4803fffb589SSandeep Dhavale goto out_error_workqueue_init; 4818d1b80a7SDan Carpenter } 4823fffb589SSandeep Dhavale 4833fffb589SSandeep Dhavale err = erofs_init_percpu_workers(); 4849f6cc76eSGao Xiang if (err) 4853fffb589SSandeep Dhavale goto out_error_pcpu_worker; 4863fffb589SSandeep Dhavale 4873fffb589SSandeep Dhavale err = erofs_cpu_hotplug_init(); 4883fffb589SSandeep Dhavale if (err < 0) 4893fffb589SSandeep Dhavale goto out_error_cpuhp_init; 4903fffb589SSandeep Dhavale return err; 4913fffb589SSandeep Dhavale 4923fffb589SSandeep Dhavale out_error_cpuhp_init: 4933fffb589SSandeep Dhavale erofs_destroy_percpu_workers(); 4943fffb589SSandeep Dhavale out_error_pcpu_worker: 4953fffb589SSandeep Dhavale destroy_workqueue(z_erofs_workqueue); 4963fffb589SSandeep Dhavale out_error_workqueue_init: 4979f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 4983fffb589SSandeep Dhavale out_error_pcluster_pool: 4999f6cc76eSGao Xiang return err; 50047e4937aSGao Xiang } 50147e4937aSGao Xiang 502db166fc2SGao Xiang enum z_erofs_pclustermode { 503db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 50447e4937aSGao Xiang /* 505db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 5060b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 5070b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 508387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 5090b964600SGao Xiang */ 510db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 51147e4937aSGao Xiang /* 51247e4937aSGao Xiang * The current collection has been linked with the owned chain, and 51347e4937aSGao Xiang * could also be linked with the remaining collections, which means 51447e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 51547e4937aSGao Xiang * the current collection can safely use the whole page (since 51647e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 51747e4937aSGao Xiang * illustrated below: 51847e4937aSGao Xiang * ________________________________________________________________ 51947e4937aSGao Xiang * | tail (partial) page | head (partial) page | 52047e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 521*967c28b2SGao Xiang * | | | 522*967c28b2SGao Xiang * |__PCLUSTER_FOLLOWED___|___________PCLUSTER_FOLLOWED____________| 52347e4937aSGao Xiang * 52447e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 52547e4937aSGao Xiang */ 526db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 52747e4937aSGao Xiang }; 52847e4937aSGao Xiang 5295c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 5305c6dcc57SGao Xiang struct inode *const inode; 5315c6dcc57SGao Xiang struct erofs_map_blocks map; 53206a304cdSGao Xiang struct z_erofs_bvec_iter biter; 53347e4937aSGao Xiang 5346ab5eed6SGao Xiang struct page *pagepool; 53506a304cdSGao Xiang struct page *candidate_bvpage; 536*967c28b2SGao Xiang struct z_erofs_pcluster *pcl; 53747e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 538db166fc2SGao Xiang enum z_erofs_pclustermode mode; 53947e4937aSGao Xiang 54047e4937aSGao Xiang /* used for applying cache strategy on the fly */ 54147e4937aSGao Xiang bool backmost; 54247e4937aSGao Xiang erofs_off_t headoffset; 543ed722fbcSGao Xiang 544ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 545ed722fbcSGao Xiang unsigned int icur; 54647e4937aSGao Xiang }; 54747e4937aSGao Xiang 54847e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 5495c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 550db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 55147e4937aSGao Xiang 5521282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 5531282dea3SGao Xiang { 5541282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 5551282dea3SGao Xiang 5561282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 5571282dea3SGao Xiang return false; 5581282dea3SGao Xiang 5591282dea3SGao Xiang if (fe->backmost) 5601282dea3SGao Xiang return true; 5611282dea3SGao Xiang 5621282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 5631282dea3SGao Xiang fe->map.m_la < fe->headoffset) 5641282dea3SGao Xiang return true; 5651282dea3SGao Xiang 5661282dea3SGao Xiang return false; 5671282dea3SGao Xiang } 5681282dea3SGao Xiang 5696ab5eed6SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe) 57047e4937aSGao Xiang { 5716f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 5725c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 5731282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 57447e4937aSGao Xiang bool standalone = true; 5756f39d1e1SGao Xiang /* 5766f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 5776f39d1e1SGao Xiang * can be used if low memory otherwise. 5786f39d1e1SGao Xiang */ 5791825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 5801825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 581ed722fbcSGao Xiang unsigned int i; 58247e4937aSGao Xiang 583db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 58447e4937aSGao Xiang return; 58547e4937aSGao Xiang 586ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 58747e4937aSGao Xiang struct page *page; 588b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 5891825c8d7SGao Xiang struct page *newpage = NULL; 59047e4937aSGao Xiang 59147e4937aSGao Xiang /* the compressed page was loaded before */ 592ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 59347e4937aSGao Xiang continue; 59447e4937aSGao Xiang 595ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 59647e4937aSGao Xiang 59747e4937aSGao Xiang if (page) { 598b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 5990b964600SGao Xiang } else { 6000b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 6010b964600SGao Xiang standalone = false; 6021282dea3SGao Xiang if (!shouldalloc) 6031282dea3SGao Xiang continue; 6041282dea3SGao Xiang 6051282dea3SGao Xiang /* 6061282dea3SGao Xiang * try to use cached I/O if page allocation 6071282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 6081282dea3SGao Xiang * to avoid any direct reclaim. 6091282dea3SGao Xiang */ 6106ab5eed6SGao Xiang newpage = erofs_allocpage(&fe->pagepool, gfp); 6111825c8d7SGao Xiang if (!newpage) 61247e4937aSGao Xiang continue; 6131282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 614b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 61547e4937aSGao Xiang } 61647e4937aSGao Xiang 617b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 61847e4937aSGao Xiang continue; 61947e4937aSGao Xiang 620eaa9172aSGao Xiang if (page) 62147e4937aSGao Xiang put_page(page); 622eaa9172aSGao Xiang else if (newpage) 6236ab5eed6SGao Xiang erofs_pagepool_add(&fe->pagepool, newpage); 62447e4937aSGao Xiang } 62547e4937aSGao Xiang 6260b964600SGao Xiang /* 6270b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 6280b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 6290b964600SGao Xiang */ 6300b964600SGao Xiang if (standalone) 631db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 63247e4937aSGao Xiang } 63347e4937aSGao Xiang 63447e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 63547e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 63647e4937aSGao Xiang struct erofs_workgroup *grp) 63747e4937aSGao Xiang { 63847e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 63947e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 64047e4937aSGao Xiang int i; 64147e4937aSGao Xiang 642cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 64347e4937aSGao Xiang /* 64447e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 64547e4937aSGao Xiang * therefore no need to worry about available decompression users. 64647e4937aSGao Xiang */ 6479f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 648ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 64947e4937aSGao Xiang 65047e4937aSGao Xiang if (!page) 65147e4937aSGao Xiang continue; 65247e4937aSGao Xiang 65347e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 65447e4937aSGao Xiang if (!trylock_page(page)) 65547e4937aSGao Xiang return -EBUSY; 65647e4937aSGao Xiang 657f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 65847e4937aSGao Xiang continue; 65947e4937aSGao Xiang 66047e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 661ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 6626aaa7b06SGao Xiang detach_page_private(page); 66347e4937aSGao Xiang unlock_page(page); 66447e4937aSGao Xiang } 66547e4937aSGao Xiang return 0; 66647e4937aSGao Xiang } 66747e4937aSGao Xiang 668d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 66947e4937aSGao Xiang { 67047e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 671ed722fbcSGao Xiang int ret, i; 67247e4937aSGao Xiang 673ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 674ed722fbcSGao Xiang return 0; 67547e4937aSGao Xiang 676ed722fbcSGao Xiang ret = 0; 677cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 6789f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 679ed722fbcSGao Xiang if (pcl->compressed_bvecs[i].page == page) { 680ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 68147e4937aSGao Xiang ret = 1; 68247e4937aSGao Xiang break; 68347e4937aSGao Xiang } 68447e4937aSGao Xiang } 68547e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 6866aaa7b06SGao Xiang if (ret) 6876aaa7b06SGao Xiang detach_page_private(page); 68847e4937aSGao Xiang return ret; 68947e4937aSGao Xiang } 69047e4937aSGao Xiang 6915c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 692ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 69347e4937aSGao Xiang { 6945c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 69547e4937aSGao Xiang 696ed722fbcSGao Xiang while (fe->icur > 0) { 697ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 698ed722fbcSGao Xiang NULL, bvec->page)) { 699ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 70047e4937aSGao Xiang return true; 701ed722fbcSGao Xiang } 702ed722fbcSGao Xiang } 70347e4937aSGao Xiang return false; 70447e4937aSGao Xiang } 70547e4937aSGao Xiang 70687ca34a7SGao Xiang /* callers must be with pcluster lock held */ 7075c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 7085b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 70947e4937aSGao Xiang { 71047e4937aSGao Xiang int ret; 71147e4937aSGao Xiang 712db166fc2SGao Xiang if (exclusive) { 71306a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 714ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 71547e4937aSGao Xiang return 0; 71606a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 717db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 71806a304cdSGao Xiang !fe->candidate_bvpage) 71906a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 72006a304cdSGao Xiang } 7216ab5eed6SGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage, 7226ab5eed6SGao Xiang &fe->pagepool); 72306a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 72406a304cdSGao Xiang return ret; 72547e4937aSGao Xiang } 72647e4937aSGao Xiang 7275c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 72847e4937aSGao Xiang { 7295c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 7305c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 73147e4937aSGao Xiang 732473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 733473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 734473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 73547e4937aSGao Xiang *owned_head = &pcl->next; 736473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 737db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 738473e15b0SGao Xiang return; 739473e15b0SGao Xiang } 740473e15b0SGao Xiang 741*967c28b2SGao Xiang /* type 2, it belongs to an ongoing chain */ 742db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 74347e4937aSGao Xiang } 74447e4937aSGao Xiang 74583a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 74647e4937aSGao Xiang { 74783a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 748cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 74947e4937aSGao Xiang struct z_erofs_pcluster *pcl; 75064094a04SGao Xiang struct erofs_workgroup *grp; 75147e4937aSGao Xiang int err; 75247e4937aSGao Xiang 753c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 754c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 7558f899262SGao Xiang DBG_BUGON(1); 7568f899262SGao Xiang return -EFSCORRUPTED; 7578f899262SGao Xiang } 7588f899262SGao Xiang 7599f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 760cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 761cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 7629f6cc76eSGao Xiang if (IS_ERR(pcl)) 7639f6cc76eSGao Xiang return PTR_ERR(pcl); 76447e4937aSGao Xiang 76564094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 7668f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 7672bfab9c0SGao Xiang pcl->length = 0; 7682bfab9c0SGao Xiang pcl->partial = true; 76947e4937aSGao Xiang 77047e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 7715c6dcc57SGao Xiang pcl->next = fe->owned_head; 77287ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 773db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 77447e4937aSGao Xiang 77547e4937aSGao Xiang /* 77647e4937aSGao Xiang * lock all primary followed works before visible to others 77747e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 77847e4937aSGao Xiang */ 77987ca34a7SGao Xiang mutex_init(&pcl->lock); 78087ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 78147e4937aSGao Xiang 782cecf864dSYue Hu if (ztailpacking) { 783cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 7843acea5fcSJingbo Xu pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 785cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 786cecf864dSYue Hu } else { 787cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 788cecf864dSYue Hu 78983a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 79064094a04SGao Xiang if (IS_ERR(grp)) { 79164094a04SGao Xiang err = PTR_ERR(grp); 79264094a04SGao Xiang goto err_out; 79364094a04SGao Xiang } 79464094a04SGao Xiang 79564094a04SGao Xiang if (grp != &pcl->obj) { 7965c6dcc57SGao Xiang fe->pcl = container_of(grp, 797cecf864dSYue Hu struct z_erofs_pcluster, obj); 79864094a04SGao Xiang err = -EEXIST; 79964094a04SGao Xiang goto err_out; 80047e4937aSGao Xiang } 801cecf864dSYue Hu } 8025c6dcc57SGao Xiang fe->owned_head = &pcl->next; 8035c6dcc57SGao Xiang fe->pcl = pcl; 8049e579fc1SGao Xiang return 0; 80564094a04SGao Xiang 80664094a04SGao Xiang err_out: 80787ca34a7SGao Xiang mutex_unlock(&pcl->lock); 8089f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 80964094a04SGao Xiang return err; 81047e4937aSGao Xiang } 81147e4937aSGao Xiang 81283a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 81347e4937aSGao Xiang { 81483a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 8150d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 8169e579fc1SGao Xiang int ret; 81747e4937aSGao Xiang 81887ca34a7SGao Xiang DBG_BUGON(fe->pcl); 81947e4937aSGao Xiang 82087ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 8215c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 82247e4937aSGao Xiang 8230d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 8240d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 8250d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 8260d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 82747e4937aSGao Xiang DBG_BUGON(1); 828cecf864dSYue Hu return -EFSCORRUPTED; 829cecf864dSYue Hu } 83047e4937aSGao Xiang 83164094a04SGao Xiang if (grp) { 8325c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 8330d823b42SGao Xiang ret = -EEXIST; 83464094a04SGao Xiang } else { 83583a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 83664094a04SGao Xiang } 83747e4937aSGao Xiang 8380d823b42SGao Xiang if (ret == -EEXIST) { 839267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 840267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 8410d823b42SGao Xiang } else if (ret) { 8420d823b42SGao Xiang return ret; 8430d823b42SGao Xiang } 84406a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 845387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 84681382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 847ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 84847e4937aSGao Xiang return 0; 84947e4937aSGao Xiang } 85047e4937aSGao Xiang 85147e4937aSGao Xiang /* 85247e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 85347e4937aSGao Xiang * only after a RCU grace period. 85447e4937aSGao Xiang */ 85547e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 85647e4937aSGao Xiang { 85787ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 85887ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 85947e4937aSGao Xiang } 86047e4937aSGao Xiang 86147e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 86247e4937aSGao Xiang { 86347e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 86447e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 86547e4937aSGao Xiang 86687ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 86747e4937aSGao Xiang } 86847e4937aSGao Xiang 8695c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 87047e4937aSGao Xiang { 87187ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 87247e4937aSGao Xiang 87387ca34a7SGao Xiang if (!pcl) 87447e4937aSGao Xiang return false; 87547e4937aSGao Xiang 87606a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 87787ca34a7SGao Xiang mutex_unlock(&pcl->lock); 87847e4937aSGao Xiang 87905b63d2bSGao Xiang if (fe->candidate_bvpage) 88006a304cdSGao Xiang fe->candidate_bvpage = NULL; 88106a304cdSGao Xiang 88247e4937aSGao Xiang /* 88347e4937aSGao Xiang * if all pending pages are added, don't hold its reference 88447e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 88547e4937aSGao Xiang */ 886db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 88787ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 88847e4937aSGao Xiang 88987ca34a7SGao Xiang fe->pcl = NULL; 89047e4937aSGao Xiang return true; 89147e4937aSGao Xiang } 89247e4937aSGao Xiang 893b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 894b15b2e30SYue Hu struct page *page, unsigned int pageofs, 895b15b2e30SYue Hu unsigned int len) 896b15b2e30SYue Hu { 8973acea5fcSJingbo Xu struct super_block *sb = inode->i_sb; 898b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 899b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 900b15b2e30SYue Hu u8 *src, *dst; 901b15b2e30SYue Hu unsigned int i, cnt; 902b15b2e30SYue Hu 903e5126de1SYue Hu if (!packed_inode) 904e5126de1SYue Hu return -EFSCORRUPTED; 905e5126de1SYue Hu 906eb2c5e41SGao Xiang buf.inode = packed_inode; 907b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 908b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 909b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 9103acea5fcSJingbo Xu sb->s_blocksize - erofs_blkoff(sb, pos)); 911eb2c5e41SGao Xiang src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 912b15b2e30SYue Hu if (IS_ERR(src)) { 913b15b2e30SYue Hu erofs_put_metabuf(&buf); 914b15b2e30SYue Hu return PTR_ERR(src); 915b15b2e30SYue Hu } 916b15b2e30SYue Hu 917b15b2e30SYue Hu dst = kmap_local_page(page); 9183acea5fcSJingbo Xu memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 919b15b2e30SYue Hu kunmap_local(dst); 920b15b2e30SYue Hu pos += cnt; 921b15b2e30SYue Hu } 922b15b2e30SYue Hu erofs_put_metabuf(&buf); 923b15b2e30SYue Hu return 0; 924b15b2e30SYue Hu } 925b15b2e30SYue Hu 92647e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 9276ab5eed6SGao Xiang struct page *page) 92847e4937aSGao Xiang { 92947e4937aSGao Xiang struct inode *const inode = fe->inode; 93047e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 93147e4937aSGao Xiang const loff_t offset = page_offset(page); 9325b220b20SGao Xiang bool tight = true, exclusive; 9332bfab9c0SGao Xiang unsigned int cur, end, spiltted; 93447e4937aSGao Xiang int err = 0; 93547e4937aSGao Xiang 93647e4937aSGao Xiang /* register locked file pages as online pages in pack */ 93747e4937aSGao Xiang z_erofs_onlinepage_init(page); 93847e4937aSGao Xiang 93947e4937aSGao Xiang spiltted = 0; 94047e4937aSGao Xiang end = PAGE_SIZE; 94147e4937aSGao Xiang repeat: 94247e4937aSGao Xiang cur = end - 1; 94347e4937aSGao Xiang 94439397a46SGao Xiang if (offset + cur < map->m_la || 94539397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 9465c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 94747e4937aSGao Xiang fe->backmost = false; 94847e4937aSGao Xiang map->m_la = offset + cur; 94947e4937aSGao Xiang map->m_llen = 0; 95047e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 9518d8a09b0SGao Xiang if (err) 95267148551SGao Xiang goto out; 95339397a46SGao Xiang } else { 95439397a46SGao Xiang if (fe->pcl) 95539397a46SGao Xiang goto hitted; 95639397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 95739397a46SGao Xiang } 95847e4937aSGao Xiang 959b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 960b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 96147e4937aSGao Xiang goto hitted; 96247e4937aSGao Xiang 96383a386c0SGao Xiang err = z_erofs_collector_begin(fe); 9648d8a09b0SGao Xiang if (err) 96567148551SGao Xiang goto out; 96647e4937aSGao Xiang 9675c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 96809c54379SGao Xiang void *mp; 969cecf864dSYue Hu 97009c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 9713acea5fcSJingbo Xu erofs_blknr(inode->i_sb, map->m_pa), 9723acea5fcSJingbo Xu EROFS_NO_KMAP); 97309c54379SGao Xiang if (IS_ERR(mp)) { 97409c54379SGao Xiang err = PTR_ERR(mp); 975cecf864dSYue Hu erofs_err(inode->i_sb, 976cecf864dSYue Hu "failed to get inline page, err %d", err); 97767148551SGao Xiang goto out; 978cecf864dSYue Hu } 97909c54379SGao Xiang get_page(fe->map.buf.page); 980ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 981ed722fbcSGao Xiang fe->map.buf.page); 982db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 983cecf864dSYue Hu } else { 9846f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 9856ab5eed6SGao Xiang z_erofs_bind_cache(fe); 986cecf864dSYue Hu } 98747e4937aSGao Xiang hitted: 988dc76ea8cSGao Xiang /* 989dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 990dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 991dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 992387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 993dc76ea8cSGao Xiang */ 994*967c28b2SGao Xiang tight &= (fe->mode > Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 995dc76ea8cSGao Xiang 99647e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 9978d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 99847e4937aSGao Xiang zero_user_segment(page, cur, end); 99947e4937aSGao Xiang goto next_part; 100047e4937aSGao Xiang } 1001b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 1002b15b2e30SYue Hu unsigned int pageofs, skip, len; 1003b15b2e30SYue Hu 1004b15b2e30SYue Hu if (offset > map->m_la) { 1005b15b2e30SYue Hu pageofs = 0; 1006b15b2e30SYue Hu skip = offset - map->m_la; 1007b15b2e30SYue Hu } else { 1008b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 1009b15b2e30SYue Hu skip = 0; 1010b15b2e30SYue Hu } 1011b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 1012b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1013b15b2e30SYue Hu if (err) 1014b15b2e30SYue Hu goto out; 1015b15b2e30SYue Hu ++spiltted; 1016b15b2e30SYue Hu tight = false; 1017b15b2e30SYue Hu goto next_part; 1018b15b2e30SYue Hu } 101947e4937aSGao Xiang 10205b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 102147e4937aSGao Xiang if (cur) 1022db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 102347e4937aSGao Xiang 102406a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 102506a304cdSGao Xiang .page = page, 102606a304cdSGao Xiang .offset = offset - map->m_la, 102706a304cdSGao Xiang .end = end, 10285b220b20SGao Xiang }), exclusive); 102905b63d2bSGao Xiang if (err) 103067148551SGao Xiang goto out; 103147e4937aSGao Xiang 103267148551SGao Xiang z_erofs_onlinepage_split(page); 103347e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 103447e4937aSGao Xiang ++spiltted; 1035267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1036267f2492SGao Xiang fe->pcl->multibases = true; 10372bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 10382bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 10392bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 10402bfab9c0SGao Xiang } 1041e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1042e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1043e7933278SGao Xiang fe->pcl->length == map->m_llen) 1044e7933278SGao Xiang fe->pcl->partial = false; 104547e4937aSGao Xiang next_part: 10462bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 104747e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 10482bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 104947e4937aSGao Xiang 105047e4937aSGao Xiang end = cur; 105147e4937aSGao Xiang if (end > 0) 105247e4937aSGao Xiang goto repeat; 105347e4937aSGao Xiang 105447e4937aSGao Xiang out: 105567148551SGao Xiang if (err) 105667148551SGao Xiang z_erofs_page_mark_eio(page); 105747e4937aSGao Xiang z_erofs_onlinepage_endio(page); 105847e4937aSGao Xiang return err; 105947e4937aSGao Xiang } 106047e4937aSGao Xiang 1061ef4b4b46SYue Hu static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 106240452ffcSHuang Jianan unsigned int readahead_pages) 106340452ffcSHuang Jianan { 1064a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 106540452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 106640452ffcSHuang Jianan !readahead_pages) 106740452ffcSHuang Jianan return true; 106840452ffcSHuang Jianan 106940452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 107040452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 107140452ffcSHuang Jianan return true; 107240452ffcSHuang Jianan 107340452ffcSHuang Jianan return false; 107440452ffcSHuang Jianan } 107540452ffcSHuang Jianan 10766aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 10776aaa7b06SGao Xiang { 10786aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 10796aaa7b06SGao Xiang } 10806aaa7b06SGao Xiang 10814f05687fSGao Xiang struct z_erofs_decompress_backend { 10824f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 10834f05687fSGao Xiang struct super_block *sb; 10844f05687fSGao Xiang struct z_erofs_pcluster *pcl; 10854f05687fSGao Xiang 10864f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 10874f05687fSGao Xiang struct page **decompressed_pages; 10884f05687fSGao Xiang /* pages to keep the compressed data */ 10894f05687fSGao Xiang struct page **compressed_pages; 10904f05687fSGao Xiang 1091267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 10924f05687fSGao Xiang struct page **pagepool; 10932bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 10944f05687fSGao Xiang }; 10954f05687fSGao Xiang 1096267f2492SGao Xiang struct z_erofs_bvec_item { 1097267f2492SGao Xiang struct z_erofs_bvec bvec; 1098267f2492SGao Xiang struct list_head list; 1099267f2492SGao Xiang }; 1100267f2492SGao Xiang 1101267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 11023fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 11033fe96ee0SGao Xiang { 1104267f2492SGao Xiang struct z_erofs_bvec_item *item; 1105267f2492SGao Xiang 1106267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1107267f2492SGao Xiang unsigned int pgnr; 11083fe96ee0SGao Xiang 1109267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 11102bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 111163bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 11123fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1113267f2492SGao Xiang return; 11143fe96ee0SGao Xiang } 111563bbb856SGao Xiang } 11163fe96ee0SGao Xiang 1117267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1118267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1119267f2492SGao Xiang item->bvec = *bvec; 1120267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1121267f2492SGao Xiang } 1122267f2492SGao Xiang 1123267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1124267f2492SGao Xiang int err) 1125267f2492SGao Xiang { 1126267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1127267f2492SGao Xiang struct list_head *p, *n; 1128267f2492SGao Xiang 1129267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1130267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1131267f2492SGao Xiang unsigned int end, cur; 1132267f2492SGao Xiang void *dst, *src; 1133267f2492SGao Xiang 1134267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1135267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1136267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1137267f2492SGao Xiang bvi->bvec.end); 1138267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1139267f2492SGao Xiang while (cur < end) { 1140267f2492SGao Xiang unsigned int pgnr, scur, len; 1141267f2492SGao Xiang 1142267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1143267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1144267f2492SGao Xiang 1145267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1146267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1147267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1148267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1149267f2492SGao Xiang err = -EFSCORRUPTED; 1150267f2492SGao Xiang cur += len; 1151267f2492SGao Xiang continue; 1152267f2492SGao Xiang } 1153267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1154267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1155267f2492SGao Xiang kunmap_local(src); 1156267f2492SGao Xiang cur += len; 1157267f2492SGao Xiang } 1158267f2492SGao Xiang kunmap_local(dst); 1159267f2492SGao Xiang if (err) 1160267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 1161267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 1162267f2492SGao Xiang list_del(p); 1163267f2492SGao Xiang kfree(bvi); 1164267f2492SGao Xiang } 1165267f2492SGao Xiang } 1166267f2492SGao Xiang 1167267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 116842fec235SGao Xiang { 11694f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 117006a304cdSGao Xiang struct z_erofs_bvec_iter biter; 117106a304cdSGao Xiang struct page *old_bvpage; 1172267f2492SGao Xiang int i; 117342fec235SGao Xiang 1174387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 117542fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 117606a304cdSGao Xiang struct z_erofs_bvec bvec; 117742fec235SGao Xiang 117806a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 117942fec235SGao Xiang 118006a304cdSGao Xiang if (old_bvpage) 11814f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 118242fec235SGao Xiang 118306a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1184267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 118542fec235SGao Xiang } 118606a304cdSGao Xiang 118706a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 118806a304cdSGao Xiang if (old_bvpage) 11894f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 119042fec235SGao Xiang } 119142fec235SGao Xiang 11924f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 11934f05687fSGao Xiang bool *overlapped) 119467139e36SGao Xiang { 11954f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 119667139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 119767139e36SGao Xiang int i, err = 0; 119867139e36SGao Xiang 119967139e36SGao Xiang *overlapped = false; 120067139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1201ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1202ed722fbcSGao Xiang struct page *page = bvec->page; 120367139e36SGao Xiang 120467139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 120567139e36SGao Xiang if (!page) { 120667139e36SGao Xiang DBG_BUGON(1); 120767139e36SGao Xiang continue; 120867139e36SGao Xiang } 1209fe3e5914SGao Xiang be->compressed_pages[i] = page; 121067139e36SGao Xiang 121167139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 121267139e36SGao Xiang if (!PageUptodate(page)) 121367139e36SGao Xiang err = -EIO; 121467139e36SGao Xiang continue; 121567139e36SGao Xiang } 121667139e36SGao Xiang 121767139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 121867139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 12194f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 122067139e36SGao Xiang if (!PageUptodate(page)) 122167139e36SGao Xiang err = -EIO; 122267139e36SGao Xiang continue; 122367139e36SGao Xiang } 1224267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 122567139e36SGao Xiang *overlapped = true; 122667139e36SGao Xiang } 122767139e36SGao Xiang } 122867139e36SGao Xiang 1229fe3e5914SGao Xiang if (err) 12304f05687fSGao Xiang return err; 12314f05687fSGao Xiang return 0; 123267139e36SGao Xiang } 123367139e36SGao Xiang 12344f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 12354f05687fSGao Xiang int err) 123647e4937aSGao Xiang { 12374f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 12384f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1239cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1240597e2953SYue Hu const struct z_erofs_decompressor *decompressor = 1241597e2953SYue Hu &erofs_decompressors[pcl->algorithmformat]; 12422bfab9c0SGao Xiang unsigned int i, inputsize; 124367148551SGao Xiang int err2; 12442bfab9c0SGao Xiang struct page *page; 12452bfab9c0SGao Xiang bool overlapped; 124647e4937aSGao Xiang 124787ca34a7SGao Xiang mutex_lock(&pcl->lock); 12482bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 124947e4937aSGao Xiang 1250fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1251fe3e5914SGao Xiang be->decompressed_pages = NULL; 1252fe3e5914SGao Xiang be->compressed_pages = NULL; 1253fe3e5914SGao Xiang be->onstack_used = 0; 12542bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 12554f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 12562bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 12574f05687fSGao Xiang memset(be->decompressed_pages, 0, 12582bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1259fe3e5914SGao Xiang } 1260fe3e5914SGao Xiang 1261fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1262fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1263fe3e5914SGao Xiang 1264fe3e5914SGao Xiang if (!be->decompressed_pages) 12654f05687fSGao Xiang be->decompressed_pages = 1266647dd2c3SGao Xiang kvcalloc(be->nr_pages, sizeof(struct page *), 1267e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1268fe3e5914SGao Xiang if (!be->compressed_pages) 1269fe3e5914SGao Xiang be->compressed_pages = 1270647dd2c3SGao Xiang kvcalloc(pclusterpages, sizeof(struct page *), 1271fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 127247e4937aSGao Xiang 1273267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 12744f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 12754f05687fSGao Xiang if (err2) 12764f05687fSGao Xiang err = err2; 12778d8a09b0SGao Xiang if (err) 127847e4937aSGao Xiang goto out; 127947e4937aSGao Xiang 1280cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1281cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1282cecf864dSYue Hu else 1283cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1284cecf864dSYue Hu 1285597e2953SYue Hu err = decompressor->decompress(&(struct z_erofs_decompress_req) { 12864f05687fSGao Xiang .sb = be->sb, 12874f05687fSGao Xiang .in = be->compressed_pages, 12884f05687fSGao Xiang .out = be->decompressed_pages, 1289cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 129087ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 12919f6cc76eSGao Xiang .inputsize = inputsize, 12922bfab9c0SGao Xiang .outputsize = pcl->length, 129347e4937aSGao Xiang .alg = pcl->algorithmformat, 129447e4937aSGao Xiang .inplace_io = overlapped, 12952bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1296267f2492SGao Xiang .fillgaps = pcl->multibases, 12974f05687fSGao Xiang }, be->pagepool); 129847e4937aSGao Xiang 129947e4937aSGao Xiang out: 1300cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1301cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1302ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1303ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1304cecf864dSYue Hu put_page(page); 1305cecf864dSYue Hu } else { 1306cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1307ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 130847e4937aSGao Xiang 130947e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 131047e4937aSGao Xiang continue; 131147e4937aSGao Xiang 13126aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13134f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1314ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 131547e4937aSGao Xiang } 1316cecf864dSYue Hu } 1317fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1318fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1319647dd2c3SGao Xiang kvfree(be->compressed_pages); 1320267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 132147e4937aSGao Xiang 13222bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 13234f05687fSGao Xiang page = be->decompressed_pages[i]; 132447e4937aSGao Xiang if (!page) 132547e4937aSGao Xiang continue; 132647e4937aSGao Xiang 13276aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 132847e4937aSGao Xiang 13296aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 13304f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 133147e4937aSGao Xiang continue; 133267148551SGao Xiang if (err) 133367148551SGao Xiang z_erofs_page_mark_eio(page); 133447e4937aSGao Xiang z_erofs_onlinepage_endio(page); 133547e4937aSGao Xiang } 133647e4937aSGao Xiang 13374f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 1338647dd2c3SGao Xiang kvfree(be->decompressed_pages); 133947e4937aSGao Xiang 13402bfab9c0SGao Xiang pcl->length = 0; 13412bfab9c0SGao Xiang pcl->partial = true; 1342267f2492SGao Xiang pcl->multibases = false; 134306a304cdSGao Xiang pcl->bvset.nextpage = NULL; 134487ca34a7SGao Xiang pcl->vcnt = 0; 134547e4937aSGao Xiang 134687ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 134747e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 134887ca34a7SGao Xiang mutex_unlock(&pcl->lock); 134947e4937aSGao Xiang return err; 135047e4937aSGao Xiang } 135147e4937aSGao Xiang 13520c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1353eaa9172aSGao Xiang struct page **pagepool) 135447e4937aSGao Xiang { 13554f05687fSGao Xiang struct z_erofs_decompress_backend be = { 13564f05687fSGao Xiang .sb = io->sb, 13574f05687fSGao Xiang .pagepool = pagepool, 1358267f2492SGao Xiang .decompressed_secondary_bvecs = 1359267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 13604f05687fSGao Xiang }; 136147e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 136247e4937aSGao Xiang 1363*967c28b2SGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL) { 136447e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 136547e4937aSGao Xiang 13664f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 13674f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 136847e4937aSGao Xiang 13694f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 13704f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 137147e4937aSGao Xiang } 137247e4937aSGao Xiang } 137347e4937aSGao Xiang 13740c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 137547e4937aSGao Xiang { 1376a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1377a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1378eaa9172aSGao Xiang struct page *pagepool = NULL; 137947e4937aSGao Xiang 1380*967c28b2SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL); 13810c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 1382eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1383a4b1fab1SGao Xiang kvfree(bgq); 138447e4937aSGao Xiang } 138547e4937aSGao Xiang 13863fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 13873fffb589SSandeep Dhavale static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 13883fffb589SSandeep Dhavale { 13893fffb589SSandeep Dhavale z_erofs_decompressqueue_work((struct work_struct *)work); 13903fffb589SSandeep Dhavale } 13913fffb589SSandeep Dhavale #endif 13923fffb589SSandeep Dhavale 13937865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1394cdba5506SGao Xiang int bios) 13957865827cSGao Xiang { 13967865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 13977865827cSGao Xiang 13987865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1399cdba5506SGao Xiang if (io->sync) { 14007865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 140160b30050SHongyu Jin complete(&io->u.done); 14027865827cSGao Xiang return; 14037865827cSGao Xiang } 14047865827cSGao Xiang 14057865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 14067865827cSGao Xiang return; 14073fffb589SSandeep Dhavale /* Use (kthread_)work and sync decompression for atomic contexts only */ 14087865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 14093fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 14103fffb589SSandeep Dhavale struct kthread_worker *worker; 14113fffb589SSandeep Dhavale 14123fffb589SSandeep Dhavale rcu_read_lock(); 14133fffb589SSandeep Dhavale worker = rcu_dereference( 14143fffb589SSandeep Dhavale z_erofs_pcpu_workers[raw_smp_processor_id()]); 14153fffb589SSandeep Dhavale if (!worker) { 14163fffb589SSandeep Dhavale INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 14177865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 14183fffb589SSandeep Dhavale } else { 14193fffb589SSandeep Dhavale kthread_queue_work(worker, &io->u.kthread_work); 14203fffb589SSandeep Dhavale } 14213fffb589SSandeep Dhavale rcu_read_unlock(); 14223fffb589SSandeep Dhavale #else 14233fffb589SSandeep Dhavale queue_work(z_erofs_workqueue, &io->u.work); 14243fffb589SSandeep Dhavale #endif 14257865827cSGao Xiang /* enable sync decompression for readahead */ 14267865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 14277865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 14287865827cSGao Xiang return; 14297865827cSGao Xiang } 14307865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 14317865827cSGao Xiang } 14327865827cSGao Xiang 143347e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 143447e4937aSGao Xiang unsigned int nr, 1435eaa9172aSGao Xiang struct page **pagepool, 14369f2731d6SGao Xiang struct address_space *mc) 143747e4937aSGao Xiang { 143847e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 14399f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 144047e4937aSGao Xiang bool tocache = false; 144147e4937aSGao Xiang 144247e4937aSGao Xiang struct address_space *mapping; 144347e4937aSGao Xiang struct page *oldpage, *page; 144447e4937aSGao Xiang int justfound; 144547e4937aSGao Xiang 144647e4937aSGao Xiang repeat: 1447ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 144847e4937aSGao Xiang oldpage = page; 144947e4937aSGao Xiang 145047e4937aSGao Xiang if (!page) 145147e4937aSGao Xiang goto out_allocpage; 145247e4937aSGao Xiang 1453b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1454b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 145547e4937aSGao Xiang 14561825c8d7SGao Xiang /* 14571825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 14581825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 14591825c8d7SGao Xiang */ 14601825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1461ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 14621825c8d7SGao Xiang set_page_private(page, 0); 14631825c8d7SGao Xiang tocache = true; 14641825c8d7SGao Xiang goto out_tocache; 14651825c8d7SGao Xiang } 146647e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 146747e4937aSGao Xiang 146847e4937aSGao Xiang /* 14696aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 147047e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 147147e4937aSGao Xiang */ 147247e4937aSGao Xiang if (mapping && mapping != mc) 147347e4937aSGao Xiang /* ought to be unmanaged pages */ 147447e4937aSGao Xiang goto out; 147547e4937aSGao Xiang 14766aaa7b06SGao Xiang /* directly return for shortlived page as well */ 14776aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 14786aaa7b06SGao Xiang goto out; 14796aaa7b06SGao Xiang 148047e4937aSGao Xiang lock_page(page); 148147e4937aSGao Xiang 148247e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 148347e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 148447e4937aSGao Xiang 148547e4937aSGao Xiang /* the page is still in manage cache */ 148647e4937aSGao Xiang if (page->mapping == mc) { 1487ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 148847e4937aSGao Xiang 148947e4937aSGao Xiang if (!PagePrivate(page)) { 149047e4937aSGao Xiang /* 149147e4937aSGao Xiang * impossible to be !PagePrivate(page) for 149247e4937aSGao Xiang * the current restriction as well if 1493ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 149447e4937aSGao Xiang */ 149547e4937aSGao Xiang DBG_BUGON(!justfound); 149647e4937aSGao Xiang 149747e4937aSGao Xiang justfound = 0; 149847e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 149947e4937aSGao Xiang SetPagePrivate(page); 150047e4937aSGao Xiang } 150147e4937aSGao Xiang 150247e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 150347e4937aSGao Xiang if (PageUptodate(page)) { 150447e4937aSGao Xiang unlock_page(page); 150547e4937aSGao Xiang page = NULL; 150647e4937aSGao Xiang } 150747e4937aSGao Xiang goto out; 150847e4937aSGao Xiang } 150947e4937aSGao Xiang 151047e4937aSGao Xiang /* 151147e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 151247e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 151347e4937aSGao Xiang */ 151447e4937aSGao Xiang DBG_BUGON(page->mapping); 151547e4937aSGao Xiang DBG_BUGON(!justfound); 151647e4937aSGao Xiang 151747e4937aSGao Xiang tocache = true; 151847e4937aSGao Xiang unlock_page(page); 151947e4937aSGao Xiang put_page(page); 152047e4937aSGao Xiang out_allocpage: 15215ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1522ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1523ed722fbcSGao Xiang oldpage, page)) { 1524eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 15255ddcee1fSGao Xiang cond_resched(); 15265ddcee1fSGao Xiang goto repeat; 15275ddcee1fSGao Xiang } 15281825c8d7SGao Xiang out_tocache: 1529bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1530bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1531bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1532bf225074SGao Xiang goto out; 1533a30573b3SGao Xiang } 1534bf225074SGao Xiang attach_page_private(page, pcl); 1535bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1536bf225074SGao Xiang put_page(page); 1537bf225074SGao Xiang 153847e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 153947e4937aSGao Xiang return page; 154047e4937aSGao Xiang } 154147e4937aSGao Xiang 1542cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1543a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 154447e4937aSGao Xiang { 1545a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 154647e4937aSGao Xiang 1547a4b1fab1SGao Xiang if (fg && !*fg) { 1548a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1549a4b1fab1SGao Xiang if (!q) { 1550a4b1fab1SGao Xiang *fg = true; 1551a4b1fab1SGao Xiang goto fg_out; 155247e4937aSGao Xiang } 15533fffb589SSandeep Dhavale #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 15543fffb589SSandeep Dhavale kthread_init_work(&q->u.kthread_work, 15553fffb589SSandeep Dhavale z_erofs_decompressqueue_kthread_work); 15563fffb589SSandeep Dhavale #else 15570c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 15583fffb589SSandeep Dhavale #endif 1559a4b1fab1SGao Xiang } else { 1560a4b1fab1SGao Xiang fg_out: 1561a4b1fab1SGao Xiang q = fgq; 156260b30050SHongyu Jin init_completion(&fgq->u.done); 1563a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 156467148551SGao Xiang q->eio = false; 1565cdba5506SGao Xiang q->sync = true; 1566a4b1fab1SGao Xiang } 1567a4b1fab1SGao Xiang q->sb = sb; 1568*967c28b2SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL; 1569a4b1fab1SGao Xiang return q; 157047e4937aSGao Xiang } 157147e4937aSGao Xiang 157247e4937aSGao Xiang /* define decompression jobqueue types */ 157347e4937aSGao Xiang enum { 157447e4937aSGao Xiang JQ_BYPASS, 157547e4937aSGao Xiang JQ_SUBMIT, 157647e4937aSGao Xiang NR_JOBQUEUES, 157747e4937aSGao Xiang }; 157847e4937aSGao Xiang 157947e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 158047e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 158147e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 158247e4937aSGao Xiang { 158347e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 158447e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 158547e4937aSGao Xiang 1586*967c28b2SGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL); 158747e4937aSGao Xiang 158847e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 158947e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 159047e4937aSGao Xiang 159147e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 159247e4937aSGao Xiang } 159347e4937aSGao Xiang 15947865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 15957865827cSGao Xiang { 1596cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 15977865827cSGao Xiang blk_status_t err = bio->bi_status; 15987865827cSGao Xiang struct bio_vec *bvec; 15997865827cSGao Xiang struct bvec_iter_all iter_all; 16007865827cSGao Xiang 16017865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 16027865827cSGao Xiang struct page *page = bvec->bv_page; 16037865827cSGao Xiang 16047865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 16057865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 16067865827cSGao Xiang 16077865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 16087865827cSGao Xiang if (!err) 16097865827cSGao Xiang SetPageUptodate(page); 16107865827cSGao Xiang unlock_page(page); 16117865827cSGao Xiang } 16127865827cSGao Xiang } 161367148551SGao Xiang if (err) 161467148551SGao Xiang q->eio = true; 1615cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 16167865827cSGao Xiang bio_put(bio); 16177865827cSGao Xiang } 16187865827cSGao Xiang 161983a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1620a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1621ef4b4b46SYue Hu bool *force_fg, bool readahead) 162247e4937aSGao Xiang { 162383a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 162483a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 162547e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1626a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 16275c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1628dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 16293f649ab7SKees Cook pgoff_t last_index; 1630dfeab2e9SGao Xiang struct block_device *last_bdev; 16311e4a2955SGao Xiang unsigned int nr_bios = 0; 16321e4a2955SGao Xiang struct bio *bio = NULL; 163382e60d00SJohannes Weiner unsigned long pflags; 163482e60d00SJohannes Weiner int memstall = 0; 163547e4937aSGao Xiang 1636cdba5506SGao Xiang /* 1637cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1638cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1639cdba5506SGao Xiang */ 1640cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1641cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1642cdba5506SGao Xiang 1643a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1644a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 164547e4937aSGao Xiang 164647e4937aSGao Xiang /* by default, all need io submission */ 164747e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 164847e4937aSGao Xiang 164947e4937aSGao Xiang do { 1650dfeab2e9SGao Xiang struct erofs_map_dev mdev; 165147e4937aSGao Xiang struct z_erofs_pcluster *pcl; 16521e4a2955SGao Xiang pgoff_t cur, end; 16531e4a2955SGao Xiang unsigned int i = 0; 16541e4a2955SGao Xiang bool bypass = true; 165547e4937aSGao Xiang 165647e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 165747e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 1658*967c28b2SGao Xiang owned_head = READ_ONCE(pcl->next); 165947e4937aSGao Xiang 1660cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1661cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1662cecf864dSYue Hu continue; 1663cecf864dSYue Hu } 1664cecf864dSYue Hu 1665dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1666dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 16673acea5fcSJingbo Xu .m_pa = erofs_pos(sb, pcl->obj.index), 1668dfeab2e9SGao Xiang }; 1669dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1670dfeab2e9SGao Xiang 16713acea5fcSJingbo Xu cur = erofs_blknr(sb, mdev.m_pa); 16729f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 167347e4937aSGao Xiang 16741e4a2955SGao Xiang do { 16751e4a2955SGao Xiang struct page *page; 167647e4937aSGao Xiang 16776ab5eed6SGao Xiang page = pickup_page_for_submission(pcl, i++, 16786ab5eed6SGao Xiang &f->pagepool, mc); 16791e4a2955SGao Xiang if (!page) 16801e4a2955SGao Xiang continue; 168147e4937aSGao Xiang 1682dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1683dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 168447e4937aSGao Xiang submit_bio_retry: 168594e4e153SGao Xiang submit_bio(bio); 168682e60d00SJohannes Weiner if (memstall) { 168782e60d00SJohannes Weiner psi_memstall_leave(&pflags); 168882e60d00SJohannes Weiner memstall = 0; 168982e60d00SJohannes Weiner } 169047e4937aSGao Xiang bio = NULL; 169147e4937aSGao Xiang } 169247e4937aSGao Xiang 169382e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 169499486c51SChristoph Hellwig psi_memstall_enter(&pflags); 169582e60d00SJohannes Weiner memstall = 1; 169682e60d00SJohannes Weiner } 169799486c51SChristoph Hellwig 169847e4937aSGao Xiang if (!bio) { 169907888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 170007888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 17010c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1702dfeab2e9SGao Xiang 1703dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 17041e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 17053acea5fcSJingbo Xu (sb->s_blocksize_bits - 9); 1706cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 1707ef4b4b46SYue Hu if (readahead) 17086ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 170947e4937aSGao Xiang ++nr_bios; 171047e4937aSGao Xiang } 171147e4937aSGao Xiang 17126c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 171347e4937aSGao Xiang goto submit_bio_retry; 171447e4937aSGao Xiang 17151e4a2955SGao Xiang last_index = cur; 17161e4a2955SGao Xiang bypass = false; 17171e4a2955SGao Xiang } while (++cur < end); 171847e4937aSGao Xiang 17191e4a2955SGao Xiang if (!bypass) 172047e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 172147e4937aSGao Xiang else 172247e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 172347e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 172447e4937aSGao Xiang 172599486c51SChristoph Hellwig if (bio) { 172694e4e153SGao Xiang submit_bio(bio); 172782e60d00SJohannes Weiner if (memstall) 172882e60d00SJohannes Weiner psi_memstall_leave(&pflags); 172999486c51SChristoph Hellwig } 173047e4937aSGao Xiang 1731587a67b7SGao Xiang /* 1732587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 17333fffb589SSandeep Dhavale * don't issue decompression but drop it directly instead. 1734587a67b7SGao Xiang */ 1735587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1736587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 17371e4a2955SGao Xiang return; 1738587a67b7SGao Xiang } 1739cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 174047e4937aSGao Xiang } 174147e4937aSGao Xiang 174283a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 17436ab5eed6SGao Xiang bool force_fg, bool ra) 174447e4937aSGao Xiang { 1745a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 174647e4937aSGao Xiang 17475c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 174847e4937aSGao Xiang return; 17496ab5eed6SGao Xiang z_erofs_submit_queue(f, io, &force_fg, ra); 175047e4937aSGao Xiang 17510c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 17526ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], &f->pagepool); 175347e4937aSGao Xiang 175447e4937aSGao Xiang if (!force_fg) 175547e4937aSGao Xiang return; 175647e4937aSGao Xiang 175747e4937aSGao Xiang /* wait until all bios are completed */ 175860b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 175947e4937aSGao Xiang 17600c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 17616ab5eed6SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], &f->pagepool); 176247e4937aSGao Xiang } 176347e4937aSGao Xiang 176438629291SGao Xiang /* 176538629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 176638629291SGao Xiang * approximate readmore strategies as a start. 176738629291SGao Xiang */ 176838629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 17696ab5eed6SGao Xiang struct readahead_control *rac, bool backmost) 177038629291SGao Xiang { 177138629291SGao Xiang struct inode *inode = f->inode; 177238629291SGao Xiang struct erofs_map_blocks *map = &f->map; 1773796e9149SYue Hu erofs_off_t cur, end, headoffset = f->headoffset; 177438629291SGao Xiang int err; 177538629291SGao Xiang 177638629291SGao Xiang if (backmost) { 1777796e9149SYue Hu if (rac) 1778796e9149SYue Hu end = headoffset + readahead_length(rac) - 1; 1779796e9149SYue Hu else 1780796e9149SYue Hu end = headoffset + PAGE_SIZE - 1; 178138629291SGao Xiang map->m_la = end; 1782622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1783622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 178438629291SGao Xiang if (err) 178538629291SGao Xiang return; 178638629291SGao Xiang 1787796e9149SYue Hu /* expand ra for the trailing edge if readahead */ 178838629291SGao Xiang if (rac) { 178938629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1790796e9149SYue Hu readahead_expand(rac, headoffset, cur - headoffset); 179138629291SGao Xiang return; 179238629291SGao Xiang } 179338629291SGao Xiang end = round_up(end, PAGE_SIZE); 179438629291SGao Xiang } else { 179538629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 179638629291SGao Xiang 179738629291SGao Xiang if (!map->m_llen) 179838629291SGao Xiang return; 179938629291SGao Xiang } 180038629291SGao Xiang 180138629291SGao Xiang cur = map->m_la + map->m_llen - 1; 180238629291SGao Xiang while (cur >= end) { 180338629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 180438629291SGao Xiang struct page *page; 180538629291SGao Xiang 180638629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1807aa793b46SGao Xiang if (page) { 180838629291SGao Xiang if (PageUptodate(page)) { 180938629291SGao Xiang unlock_page(page); 1810aa793b46SGao Xiang } else { 18116ab5eed6SGao Xiang err = z_erofs_do_read_page(f, page); 181238629291SGao Xiang if (err) 181338629291SGao Xiang erofs_err(inode->i_sb, 181438629291SGao Xiang "readmore error at page %lu @ nid %llu", 181538629291SGao Xiang index, EROFS_I(inode)->nid); 1816aa793b46SGao Xiang } 181738629291SGao Xiang put_page(page); 1818aa793b46SGao Xiang } 1819aa793b46SGao Xiang 182038629291SGao Xiang if (cur < PAGE_SIZE) 182138629291SGao Xiang break; 182238629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 182338629291SGao Xiang } 182438629291SGao Xiang } 182538629291SGao Xiang 1826a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 182747e4937aSGao Xiang { 1828a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 182947e4937aSGao Xiang struct inode *const inode = page->mapping->host; 183040452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 183147e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 183247e4937aSGao Xiang int err; 183347e4937aSGao Xiang 183447e4937aSGao Xiang trace_erofs_readpage(page, false); 183547e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 183647e4937aSGao Xiang 18376ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, true); 18386ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 18396ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, NULL, false); 18405c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 184147e4937aSGao Xiang 184247e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 18436ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, 0), false); 184447e4937aSGao Xiang 184547e4937aSGao Xiang if (err) 18464f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 184747e4937aSGao Xiang 184809c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 18496ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 185047e4937aSGao Xiang return err; 185147e4937aSGao Xiang } 185247e4937aSGao Xiang 18530615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 185447e4937aSGao Xiang { 18550615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 185647e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 185747e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 18586ab5eed6SGao Xiang struct page *head = NULL, *page; 185938629291SGao Xiang unsigned int nr_pages; 186047e4937aSGao Xiang 18610615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 186247e4937aSGao Xiang 18636ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, true); 186438629291SGao Xiang nr_pages = readahead_count(rac); 186538629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 186638629291SGao Xiang 18670615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 186847e4937aSGao Xiang set_page_private(page, (unsigned long)head); 186947e4937aSGao Xiang head = page; 187047e4937aSGao Xiang } 187147e4937aSGao Xiang 187247e4937aSGao Xiang while (head) { 187347e4937aSGao Xiang struct page *page = head; 187447e4937aSGao Xiang int err; 187547e4937aSGao Xiang 187647e4937aSGao Xiang /* traversal in reverse order */ 187747e4937aSGao Xiang head = (void *)page_private(page); 187847e4937aSGao Xiang 18796ab5eed6SGao Xiang err = z_erofs_do_read_page(&f, page); 1880a5876e24SGao Xiang if (err) 18814f761fa2SGao Xiang erofs_err(inode->i_sb, 18824f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 18834f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 188447e4937aSGao Xiang put_page(page); 188547e4937aSGao Xiang } 18866ab5eed6SGao Xiang z_erofs_pcluster_readmore(&f, rac, false); 18875c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 188847e4937aSGao Xiang 18896ab5eed6SGao Xiang z_erofs_runqueue(&f, z_erofs_is_sync_decompress(sbi, nr_pages), true); 189009c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 18916ab5eed6SGao Xiang erofs_release_pages(&f.pagepool); 189247e4937aSGao Xiang } 189347e4937aSGao Xiang 18940c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1895a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 18960615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 189747e4937aSGao Xiang }; 1898