147e4937aSGao Xiang // SPDX-License-Identifier: GPL-2.0-only 247e4937aSGao Xiang /* 347e4937aSGao Xiang * Copyright (C) 2018 HUAWEI, Inc. 4592e7cd0SAlexander A. Klimov * https://www.huawei.com/ 506a304cdSGao Xiang * Copyright (C) 2022 Alibaba Cloud 647e4937aSGao Xiang */ 747e4937aSGao Xiang #include "compress.h" 847e4937aSGao Xiang #include <linux/prefetch.h> 999486c51SChristoph Hellwig #include <linux/psi.h> 1047e4937aSGao Xiang 1147e4937aSGao Xiang #include <trace/events/erofs.h> 1247e4937aSGao Xiang 13*a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_MAX_PAGES (Z_EROFS_PCLUSTER_MAX_SIZE / PAGE_SIZE) 14*a9a94d93SGao Xiang #define Z_EROFS_INLINE_BVECS 2 15*a9a94d93SGao Xiang 16*a9a94d93SGao Xiang /* 17*a9a94d93SGao Xiang * let's leave a type here in case of introducing 18*a9a94d93SGao Xiang * another tagged pointer later. 19*a9a94d93SGao Xiang */ 20*a9a94d93SGao Xiang typedef void *z_erofs_next_pcluster_t; 21*a9a94d93SGao Xiang 22*a9a94d93SGao Xiang struct z_erofs_bvec { 23*a9a94d93SGao Xiang struct page *page; 24*a9a94d93SGao Xiang int offset; 25*a9a94d93SGao Xiang unsigned int end; 26*a9a94d93SGao Xiang }; 27*a9a94d93SGao Xiang 28*a9a94d93SGao Xiang #define __Z_EROFS_BVSET(name, total) \ 29*a9a94d93SGao Xiang struct name { \ 30*a9a94d93SGao Xiang /* point to the next page which contains the following bvecs */ \ 31*a9a94d93SGao Xiang struct page *nextpage; \ 32*a9a94d93SGao Xiang struct z_erofs_bvec bvec[total]; \ 33*a9a94d93SGao Xiang } 34*a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset,); 35*a9a94d93SGao Xiang __Z_EROFS_BVSET(z_erofs_bvset_inline, Z_EROFS_INLINE_BVECS); 36*a9a94d93SGao Xiang 37*a9a94d93SGao Xiang /* 38*a9a94d93SGao Xiang * Structure fields follow one of the following exclusion rules. 39*a9a94d93SGao Xiang * 40*a9a94d93SGao Xiang * I: Modifiable by initialization/destruction paths and read-only 41*a9a94d93SGao Xiang * for everyone else; 42*a9a94d93SGao Xiang * 43*a9a94d93SGao Xiang * L: Field should be protected by the pcluster lock; 44*a9a94d93SGao Xiang * 45*a9a94d93SGao Xiang * A: Field should be accessed / updated in atomic for parallelized code. 46*a9a94d93SGao Xiang */ 47*a9a94d93SGao Xiang struct z_erofs_pcluster { 48*a9a94d93SGao Xiang struct erofs_workgroup obj; 49*a9a94d93SGao Xiang struct mutex lock; 50*a9a94d93SGao Xiang 51*a9a94d93SGao Xiang /* A: point to next chained pcluster or TAILs */ 52*a9a94d93SGao Xiang z_erofs_next_pcluster_t next; 53*a9a94d93SGao Xiang 54*a9a94d93SGao Xiang /* L: the maximum decompression size of this round */ 55*a9a94d93SGao Xiang unsigned int length; 56*a9a94d93SGao Xiang 57*a9a94d93SGao Xiang /* L: total number of bvecs */ 58*a9a94d93SGao Xiang unsigned int vcnt; 59*a9a94d93SGao Xiang 60*a9a94d93SGao Xiang /* I: page offset of start position of decompression */ 61*a9a94d93SGao Xiang unsigned short pageofs_out; 62*a9a94d93SGao Xiang 63*a9a94d93SGao Xiang /* I: page offset of inline compressed data */ 64*a9a94d93SGao Xiang unsigned short pageofs_in; 65*a9a94d93SGao Xiang 66*a9a94d93SGao Xiang union { 67*a9a94d93SGao Xiang /* L: inline a certain number of bvec for bootstrap */ 68*a9a94d93SGao Xiang struct z_erofs_bvset_inline bvset; 69*a9a94d93SGao Xiang 70*a9a94d93SGao Xiang /* I: can be used to free the pcluster by RCU. */ 71*a9a94d93SGao Xiang struct rcu_head rcu; 72*a9a94d93SGao Xiang }; 73*a9a94d93SGao Xiang 74*a9a94d93SGao Xiang union { 75*a9a94d93SGao Xiang /* I: physical cluster size in pages */ 76*a9a94d93SGao Xiang unsigned short pclusterpages; 77*a9a94d93SGao Xiang 78*a9a94d93SGao Xiang /* I: tailpacking inline compressed size */ 79*a9a94d93SGao Xiang unsigned short tailpacking_size; 80*a9a94d93SGao Xiang }; 81*a9a94d93SGao Xiang 82*a9a94d93SGao Xiang /* I: compression algorithm format */ 83*a9a94d93SGao Xiang unsigned char algorithmformat; 84*a9a94d93SGao Xiang 85*a9a94d93SGao Xiang /* L: whether partial decompression or not */ 86*a9a94d93SGao Xiang bool partial; 87*a9a94d93SGao Xiang 88*a9a94d93SGao Xiang /* L: indicate several pageofs_outs or not */ 89*a9a94d93SGao Xiang bool multibases; 90*a9a94d93SGao Xiang 91*a9a94d93SGao Xiang /* A: compressed bvecs (can be cached or inplaced pages) */ 92*a9a94d93SGao Xiang struct z_erofs_bvec compressed_bvecs[]; 93*a9a94d93SGao Xiang }; 94*a9a94d93SGao Xiang 95*a9a94d93SGao Xiang /* let's avoid the valid 32-bit kernel addresses */ 96*a9a94d93SGao Xiang 97*a9a94d93SGao Xiang /* the chained workgroup has't submitted io (still open) */ 98*a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE) 99*a9a94d93SGao Xiang /* the chained workgroup has already submitted io */ 100*a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_TAIL_CLOSED ((void *)0x5F0EDEAD) 101*a9a94d93SGao Xiang 102*a9a94d93SGao Xiang #define Z_EROFS_PCLUSTER_NIL (NULL) 103*a9a94d93SGao Xiang 104*a9a94d93SGao Xiang struct z_erofs_decompressqueue { 105*a9a94d93SGao Xiang struct super_block *sb; 106*a9a94d93SGao Xiang atomic_t pending_bios; 107*a9a94d93SGao Xiang z_erofs_next_pcluster_t head; 108*a9a94d93SGao Xiang 109*a9a94d93SGao Xiang union { 110*a9a94d93SGao Xiang struct completion done; 111*a9a94d93SGao Xiang struct work_struct work; 112*a9a94d93SGao Xiang } u; 113*a9a94d93SGao Xiang bool eio, sync; 114*a9a94d93SGao Xiang }; 115*a9a94d93SGao Xiang 116*a9a94d93SGao Xiang static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 117*a9a94d93SGao Xiang { 118*a9a94d93SGao Xiang return !pcl->obj.index; 119*a9a94d93SGao Xiang } 120*a9a94d93SGao Xiang 121*a9a94d93SGao Xiang static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 122*a9a94d93SGao Xiang { 123*a9a94d93SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) 124*a9a94d93SGao Xiang return 1; 125*a9a94d93SGao Xiang return pcl->pclusterpages; 126*a9a94d93SGao Xiang } 127*a9a94d93SGao Xiang 128*a9a94d93SGao Xiang /* 129*a9a94d93SGao Xiang * bit 30: I/O error occurred on this page 130*a9a94d93SGao Xiang * bit 0 - 29: remaining parts to complete this page 131*a9a94d93SGao Xiang */ 132*a9a94d93SGao Xiang #define Z_EROFS_PAGE_EIO (1 << 30) 133*a9a94d93SGao Xiang 134*a9a94d93SGao Xiang static inline void z_erofs_onlinepage_init(struct page *page) 135*a9a94d93SGao Xiang { 136*a9a94d93SGao Xiang union { 137*a9a94d93SGao Xiang atomic_t o; 138*a9a94d93SGao Xiang unsigned long v; 139*a9a94d93SGao Xiang } u = { .o = ATOMIC_INIT(1) }; 140*a9a94d93SGao Xiang 141*a9a94d93SGao Xiang set_page_private(page, u.v); 142*a9a94d93SGao Xiang smp_wmb(); 143*a9a94d93SGao Xiang SetPagePrivate(page); 144*a9a94d93SGao Xiang } 145*a9a94d93SGao Xiang 146*a9a94d93SGao Xiang static inline void z_erofs_onlinepage_split(struct page *page) 147*a9a94d93SGao Xiang { 148*a9a94d93SGao Xiang atomic_inc((atomic_t *)&page->private); 149*a9a94d93SGao Xiang } 150*a9a94d93SGao Xiang 151*a9a94d93SGao Xiang static inline void z_erofs_page_mark_eio(struct page *page) 152*a9a94d93SGao Xiang { 153*a9a94d93SGao Xiang int orig; 154*a9a94d93SGao Xiang 155*a9a94d93SGao Xiang do { 156*a9a94d93SGao Xiang orig = atomic_read((atomic_t *)&page->private); 157*a9a94d93SGao Xiang } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 158*a9a94d93SGao Xiang orig | Z_EROFS_PAGE_EIO) != orig); 159*a9a94d93SGao Xiang } 160*a9a94d93SGao Xiang 161*a9a94d93SGao Xiang static inline void z_erofs_onlinepage_endio(struct page *page) 162*a9a94d93SGao Xiang { 163*a9a94d93SGao Xiang unsigned int v; 164*a9a94d93SGao Xiang 165*a9a94d93SGao Xiang DBG_BUGON(!PagePrivate(page)); 166*a9a94d93SGao Xiang v = atomic_dec_return((atomic_t *)&page->private); 167*a9a94d93SGao Xiang if (!(v & ~Z_EROFS_PAGE_EIO)) { 168*a9a94d93SGao Xiang set_page_private(page, 0); 169*a9a94d93SGao Xiang ClearPagePrivate(page); 170*a9a94d93SGao Xiang if (!(v & Z_EROFS_PAGE_EIO)) 171*a9a94d93SGao Xiang SetPageUptodate(page); 172*a9a94d93SGao Xiang unlock_page(page); 173*a9a94d93SGao Xiang } 174*a9a94d93SGao Xiang } 175*a9a94d93SGao Xiang 176*a9a94d93SGao Xiang #define Z_EROFS_ONSTACK_PAGES 32 177*a9a94d93SGao Xiang 17847e4937aSGao Xiang /* 1799f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 1809f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 1819f6cc76eSGao Xiang */ 1829f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 1839f6cc76eSGao Xiang struct kmem_cache *slab; 1849f6cc76eSGao Xiang unsigned int maxpages; 1859f6cc76eSGao Xiang char name[48]; 1869f6cc76eSGao Xiang }; 1879f6cc76eSGao Xiang 1889f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 1899f6cc76eSGao Xiang 1909f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 1919f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 1929f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 1939f6cc76eSGao Xiang }; 1949f6cc76eSGao Xiang 19506a304cdSGao Xiang struct z_erofs_bvec_iter { 19606a304cdSGao Xiang struct page *bvpage; 19706a304cdSGao Xiang struct z_erofs_bvset *bvset; 19806a304cdSGao Xiang unsigned int nr, cur; 19906a304cdSGao Xiang }; 20006a304cdSGao Xiang 20106a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 20206a304cdSGao Xiang { 20306a304cdSGao Xiang if (iter->bvpage) 20406a304cdSGao Xiang kunmap_local(iter->bvset); 20506a304cdSGao Xiang return iter->bvpage; 20606a304cdSGao Xiang } 20706a304cdSGao Xiang 20806a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 20906a304cdSGao Xiang { 21006a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 21106a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 21206a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 21306a304cdSGao Xiang struct page *oldpage; 21406a304cdSGao Xiang 21506a304cdSGao Xiang DBG_BUGON(!nextpage); 21606a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 21706a304cdSGao Xiang iter->bvpage = nextpage; 21806a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 21906a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 22006a304cdSGao Xiang iter->cur = 0; 22106a304cdSGao Xiang return oldpage; 22206a304cdSGao Xiang } 22306a304cdSGao Xiang 22406a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 22506a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 22606a304cdSGao Xiang unsigned int bootstrap_nr, 22706a304cdSGao Xiang unsigned int cur) 22806a304cdSGao Xiang { 22906a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 23006a304cdSGao Xiang .nr = bootstrap_nr, 23106a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 23206a304cdSGao Xiang }; 23306a304cdSGao Xiang 23406a304cdSGao Xiang while (cur > iter->nr) { 23506a304cdSGao Xiang cur -= iter->nr; 23606a304cdSGao Xiang z_erofs_bvset_flip(iter); 23706a304cdSGao Xiang } 23806a304cdSGao Xiang iter->cur = cur; 23906a304cdSGao Xiang } 24006a304cdSGao Xiang 24106a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 24206a304cdSGao Xiang struct z_erofs_bvec *bvec, 24306a304cdSGao Xiang struct page **candidate_bvpage) 24406a304cdSGao Xiang { 24506a304cdSGao Xiang if (iter->cur == iter->nr) { 24606a304cdSGao Xiang if (!*candidate_bvpage) 24706a304cdSGao Xiang return -EAGAIN; 24806a304cdSGao Xiang 24906a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 25006a304cdSGao Xiang iter->bvset->nextpage = *candidate_bvpage; 25106a304cdSGao Xiang z_erofs_bvset_flip(iter); 25206a304cdSGao Xiang 25306a304cdSGao Xiang iter->bvset->nextpage = NULL; 25406a304cdSGao Xiang *candidate_bvpage = NULL; 25506a304cdSGao Xiang } 25606a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 25706a304cdSGao Xiang return 0; 25806a304cdSGao Xiang } 25906a304cdSGao Xiang 26006a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 26106a304cdSGao Xiang struct z_erofs_bvec *bvec, 26206a304cdSGao Xiang struct page **old_bvpage) 26306a304cdSGao Xiang { 26406a304cdSGao Xiang if (iter->cur == iter->nr) 26506a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 26606a304cdSGao Xiang else 26706a304cdSGao Xiang *old_bvpage = NULL; 26806a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 26906a304cdSGao Xiang } 27006a304cdSGao Xiang 2719f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 2729f6cc76eSGao Xiang { 2739f6cc76eSGao Xiang int i; 2749f6cc76eSGao Xiang 2759f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 2769f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 2779f6cc76eSGao Xiang continue; 2789f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 2799f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 2809f6cc76eSGao Xiang } 2819f6cc76eSGao Xiang } 2829f6cc76eSGao Xiang 2839f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 2849f6cc76eSGao Xiang { 2859f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 2869f6cc76eSGao Xiang struct z_erofs_pcluster *a; 2879f6cc76eSGao Xiang unsigned int size; 2889f6cc76eSGao Xiang 2899f6cc76eSGao Xiang for (pcs = pcluster_pool; 2909f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 291ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 2929f6cc76eSGao Xiang 2939f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 2949f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 2959f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 2969f6cc76eSGao Xiang if (pcs->slab) 2979f6cc76eSGao Xiang continue; 2989f6cc76eSGao Xiang 2999f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 3009f6cc76eSGao Xiang return -ENOMEM; 3019f6cc76eSGao Xiang } 3029f6cc76eSGao Xiang return 0; 3039f6cc76eSGao Xiang } 3049f6cc76eSGao Xiang 3059f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 3069f6cc76eSGao Xiang { 3079f6cc76eSGao Xiang int i; 3089f6cc76eSGao Xiang 3099f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3109f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3119f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 3129f6cc76eSGao Xiang 3139f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 3149f6cc76eSGao Xiang continue; 3159f6cc76eSGao Xiang 3169f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 3179f6cc76eSGao Xiang if (!pcl) 3189f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 3199f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 3209f6cc76eSGao Xiang return pcl; 3219f6cc76eSGao Xiang } 3229f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 3239f6cc76eSGao Xiang } 3249f6cc76eSGao Xiang 3259f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 3269f6cc76eSGao Xiang { 327cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 3289f6cc76eSGao Xiang int i; 3299f6cc76eSGao Xiang 3309f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 3319f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 3329f6cc76eSGao Xiang 333cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 3349f6cc76eSGao Xiang continue; 3359f6cc76eSGao Xiang 3369f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 3379f6cc76eSGao Xiang return; 3389f6cc76eSGao Xiang } 3399f6cc76eSGao Xiang DBG_BUGON(1); 3409f6cc76eSGao Xiang } 3419f6cc76eSGao Xiang 34247e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 34347e4937aSGao Xiang 34447e4937aSGao Xiang void z_erofs_exit_zip_subsystem(void) 34547e4937aSGao Xiang { 34647e4937aSGao Xiang destroy_workqueue(z_erofs_workqueue); 3479f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 34847e4937aSGao Xiang } 34947e4937aSGao Xiang 35099634bf3SGao Xiang static inline int z_erofs_init_workqueue(void) 35147e4937aSGao Xiang { 35247e4937aSGao Xiang const unsigned int onlinecpus = num_possible_cpus(); 35347e4937aSGao Xiang 35447e4937aSGao Xiang /* 35547e4937aSGao Xiang * no need to spawn too many threads, limiting threads could minimum 35647e4937aSGao Xiang * scheduling overhead, perhaps per-CPU threads should be better? 35747e4937aSGao Xiang */ 3580e62ea33SGao Xiang z_erofs_workqueue = alloc_workqueue("erofs_unzipd", 3590e62ea33SGao Xiang WQ_UNBOUND | WQ_HIGHPRI, 36047e4937aSGao Xiang onlinecpus + onlinecpus / 4); 36147e4937aSGao Xiang return z_erofs_workqueue ? 0 : -ENOMEM; 36247e4937aSGao Xiang } 36347e4937aSGao Xiang 36447e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 36547e4937aSGao Xiang { 3669f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 36747e4937aSGao Xiang 3689f6cc76eSGao Xiang if (err) 3699f6cc76eSGao Xiang return err; 3709f6cc76eSGao Xiang err = z_erofs_init_workqueue(); 3719f6cc76eSGao Xiang if (err) 3729f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 3739f6cc76eSGao Xiang return err; 37447e4937aSGao Xiang } 37547e4937aSGao Xiang 376db166fc2SGao Xiang enum z_erofs_pclustermode { 377db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 37847e4937aSGao Xiang /* 379db166fc2SGao Xiang * The current pclusters was the tail of an exist chain, in addition 380db166fc2SGao Xiang * that the previous processed chained pclusters are all decided to 38147e4937aSGao Xiang * be hooked up to it. 382db166fc2SGao Xiang * A new chain will be created for the remaining pclusters which are 383db166fc2SGao Xiang * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED, 384db166fc2SGao Xiang * the next pcluster cannot reuse the whole page safely for inplace I/O 385db166fc2SGao Xiang * in the following scenario: 38647e4937aSGao Xiang * ________________________________________________________________ 38747e4937aSGao Xiang * | tail (partial) page | head (partial) page | 388db166fc2SGao Xiang * | (belongs to the next pcl) | (belongs to the current pcl) | 389db166fc2SGao Xiang * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________| 39047e4937aSGao Xiang */ 391db166fc2SGao Xiang Z_EROFS_PCLUSTER_HOOKED, 3920b964600SGao Xiang /* 393db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 3940b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 3950b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 396387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 3970b964600SGao Xiang */ 398db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 39947e4937aSGao Xiang /* 40047e4937aSGao Xiang * The current collection has been linked with the owned chain, and 40147e4937aSGao Xiang * could also be linked with the remaining collections, which means 40247e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 40347e4937aSGao Xiang * the current collection can safely use the whole page (since 40447e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 40547e4937aSGao Xiang * illustrated below: 40647e4937aSGao Xiang * ________________________________________________________________ 40747e4937aSGao Xiang * | tail (partial) page | head (partial) page | 40847e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 409db166fc2SGao Xiang * | PCLUSTER_FOLLOWED or | | 410db166fc2SGao Xiang * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________| 41147e4937aSGao Xiang * 41247e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 41347e4937aSGao Xiang */ 414db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 41547e4937aSGao Xiang }; 41647e4937aSGao Xiang 4175c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 4185c6dcc57SGao Xiang struct inode *const inode; 4195c6dcc57SGao Xiang struct erofs_map_blocks map; 42006a304cdSGao Xiang struct z_erofs_bvec_iter biter; 42147e4937aSGao Xiang 42206a304cdSGao Xiang struct page *candidate_bvpage; 42347e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 42447e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 425db166fc2SGao Xiang enum z_erofs_pclustermode mode; 42647e4937aSGao Xiang 4276ea5aad3SGao Xiang bool readahead; 42847e4937aSGao Xiang /* used for applying cache strategy on the fly */ 42947e4937aSGao Xiang bool backmost; 43047e4937aSGao Xiang erofs_off_t headoffset; 431ed722fbcSGao Xiang 432ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 433ed722fbcSGao Xiang unsigned int icur; 43447e4937aSGao Xiang }; 43547e4937aSGao Xiang 43647e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 4375c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 438db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 43947e4937aSGao Xiang 4401282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 4411282dea3SGao Xiang { 4421282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 4431282dea3SGao Xiang 4441282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 4451282dea3SGao Xiang return false; 4461282dea3SGao Xiang 4471282dea3SGao Xiang if (fe->backmost) 4481282dea3SGao Xiang return true; 4491282dea3SGao Xiang 4501282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 4511282dea3SGao Xiang fe->map.m_la < fe->headoffset) 4521282dea3SGao Xiang return true; 4531282dea3SGao Xiang 4541282dea3SGao Xiang return false; 4551282dea3SGao Xiang } 4561282dea3SGao Xiang 4576f39d1e1SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 458eaa9172aSGao Xiang struct page **pagepool) 45947e4937aSGao Xiang { 4606f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 4615c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 4621282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 46347e4937aSGao Xiang bool standalone = true; 4646f39d1e1SGao Xiang /* 4656f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 4666f39d1e1SGao Xiang * can be used if low memory otherwise. 4676f39d1e1SGao Xiang */ 4681825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 4691825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 470ed722fbcSGao Xiang unsigned int i; 47147e4937aSGao Xiang 472db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 47347e4937aSGao Xiang return; 47447e4937aSGao Xiang 475ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 47647e4937aSGao Xiang struct page *page; 477b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 4781825c8d7SGao Xiang struct page *newpage = NULL; 47947e4937aSGao Xiang 48047e4937aSGao Xiang /* the compressed page was loaded before */ 481ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 48247e4937aSGao Xiang continue; 48347e4937aSGao Xiang 484ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 48547e4937aSGao Xiang 48647e4937aSGao Xiang if (page) { 487b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 4880b964600SGao Xiang } else { 4890b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 4900b964600SGao Xiang standalone = false; 4911282dea3SGao Xiang if (!shouldalloc) 4921282dea3SGao Xiang continue; 4931282dea3SGao Xiang 4941282dea3SGao Xiang /* 4951282dea3SGao Xiang * try to use cached I/O if page allocation 4961282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 4971282dea3SGao Xiang * to avoid any direct reclaim. 4981282dea3SGao Xiang */ 4991825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 5001825c8d7SGao Xiang if (!newpage) 50147e4937aSGao Xiang continue; 5021282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 503b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 50447e4937aSGao Xiang } 50547e4937aSGao Xiang 506b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 50747e4937aSGao Xiang continue; 50847e4937aSGao Xiang 509eaa9172aSGao Xiang if (page) 51047e4937aSGao Xiang put_page(page); 511eaa9172aSGao Xiang else if (newpage) 512eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 51347e4937aSGao Xiang } 51447e4937aSGao Xiang 5150b964600SGao Xiang /* 5160b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 5170b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 5180b964600SGao Xiang */ 5190b964600SGao Xiang if (standalone) 520db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 52147e4937aSGao Xiang } 52247e4937aSGao Xiang 52347e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 52447e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 52547e4937aSGao Xiang struct erofs_workgroup *grp) 52647e4937aSGao Xiang { 52747e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 52847e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 52947e4937aSGao Xiang int i; 53047e4937aSGao Xiang 531cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 53247e4937aSGao Xiang /* 53347e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 53447e4937aSGao Xiang * therefore no need to worry about available decompression users. 53547e4937aSGao Xiang */ 5369f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 537ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 53847e4937aSGao Xiang 53947e4937aSGao Xiang if (!page) 54047e4937aSGao Xiang continue; 54147e4937aSGao Xiang 54247e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 54347e4937aSGao Xiang if (!trylock_page(page)) 54447e4937aSGao Xiang return -EBUSY; 54547e4937aSGao Xiang 546f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 54747e4937aSGao Xiang continue; 54847e4937aSGao Xiang 54947e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 550ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 5516aaa7b06SGao Xiang detach_page_private(page); 55247e4937aSGao Xiang unlock_page(page); 55347e4937aSGao Xiang } 55447e4937aSGao Xiang return 0; 55547e4937aSGao Xiang } 55647e4937aSGao Xiang 557d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 55847e4937aSGao Xiang { 55947e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 560ed722fbcSGao Xiang int ret, i; 56147e4937aSGao Xiang 562ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 563ed722fbcSGao Xiang return 0; 56447e4937aSGao Xiang 565ed722fbcSGao Xiang ret = 0; 566cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 5679f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 568ed722fbcSGao Xiang if (pcl->compressed_bvecs[i].page == page) { 569ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 57047e4937aSGao Xiang ret = 1; 57147e4937aSGao Xiang break; 57247e4937aSGao Xiang } 57347e4937aSGao Xiang } 57447e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 5756aaa7b06SGao Xiang if (ret) 5766aaa7b06SGao Xiang detach_page_private(page); 57747e4937aSGao Xiang return ret; 57847e4937aSGao Xiang } 57947e4937aSGao Xiang 5805c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 581ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 58247e4937aSGao Xiang { 5835c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 58447e4937aSGao Xiang 585ed722fbcSGao Xiang while (fe->icur > 0) { 586ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 587ed722fbcSGao Xiang NULL, bvec->page)) { 588ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 58947e4937aSGao Xiang return true; 590ed722fbcSGao Xiang } 591ed722fbcSGao Xiang } 59247e4937aSGao Xiang return false; 59347e4937aSGao Xiang } 59447e4937aSGao Xiang 59587ca34a7SGao Xiang /* callers must be with pcluster lock held */ 5965c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 5975b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 59847e4937aSGao Xiang { 59947e4937aSGao Xiang int ret; 60047e4937aSGao Xiang 601db166fc2SGao Xiang if (exclusive) { 60206a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 603ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 60447e4937aSGao Xiang return 0; 60506a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 606db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 60706a304cdSGao Xiang !fe->candidate_bvpage) 60806a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 60906a304cdSGao Xiang } 61006a304cdSGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 61106a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 61206a304cdSGao Xiang return ret; 61347e4937aSGao Xiang } 61447e4937aSGao Xiang 6155c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 61647e4937aSGao Xiang { 6175c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 6185c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 61947e4937aSGao Xiang 620473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 621473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 622473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 62347e4937aSGao Xiang *owned_head = &pcl->next; 624473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 625db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 626473e15b0SGao Xiang return; 627473e15b0SGao Xiang } 628473e15b0SGao Xiang 62947e4937aSGao Xiang /* 630473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 631473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 63247e4937aSGao Xiang */ 633267f2492SGao Xiang if (*owned_head != &pcl->next && pcl != f->tailpcl && 634267f2492SGao Xiang cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 635473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 63647e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 637db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_HOOKED; 6385c6dcc57SGao Xiang f->tailpcl = NULL; 639473e15b0SGao Xiang return; 64047e4937aSGao Xiang } 641473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 642db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 64347e4937aSGao Xiang } 64447e4937aSGao Xiang 64583a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 64647e4937aSGao Xiang { 64783a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 648cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 64947e4937aSGao Xiang struct z_erofs_pcluster *pcl; 65064094a04SGao Xiang struct erofs_workgroup *grp; 65147e4937aSGao Xiang int err; 65247e4937aSGao Xiang 653c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 654c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 6558f899262SGao Xiang DBG_BUGON(1); 6568f899262SGao Xiang return -EFSCORRUPTED; 6578f899262SGao Xiang } 6588f899262SGao Xiang 6599f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 660cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 661cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 6629f6cc76eSGao Xiang if (IS_ERR(pcl)) 6639f6cc76eSGao Xiang return PTR_ERR(pcl); 66447e4937aSGao Xiang 66564094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 6668f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 6672bfab9c0SGao Xiang pcl->length = 0; 6682bfab9c0SGao Xiang pcl->partial = true; 66947e4937aSGao Xiang 67047e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 6715c6dcc57SGao Xiang pcl->next = fe->owned_head; 67287ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 673db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 67447e4937aSGao Xiang 67547e4937aSGao Xiang /* 67647e4937aSGao Xiang * lock all primary followed works before visible to others 67747e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 67847e4937aSGao Xiang */ 67987ca34a7SGao Xiang mutex_init(&pcl->lock); 68087ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 68147e4937aSGao Xiang 682cecf864dSYue Hu if (ztailpacking) { 683cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 684cecf864dSYue Hu pcl->pageofs_in = erofs_blkoff(map->m_pa); 685cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 686cecf864dSYue Hu } else { 687cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 688cecf864dSYue Hu 68983a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 69064094a04SGao Xiang if (IS_ERR(grp)) { 69164094a04SGao Xiang err = PTR_ERR(grp); 69264094a04SGao Xiang goto err_out; 69364094a04SGao Xiang } 69464094a04SGao Xiang 69564094a04SGao Xiang if (grp != &pcl->obj) { 6965c6dcc57SGao Xiang fe->pcl = container_of(grp, 697cecf864dSYue Hu struct z_erofs_pcluster, obj); 69864094a04SGao Xiang err = -EEXIST; 69964094a04SGao Xiang goto err_out; 70047e4937aSGao Xiang } 701cecf864dSYue Hu } 70247e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 7035c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 7045c6dcc57SGao Xiang fe->tailpcl = pcl; 7055c6dcc57SGao Xiang fe->owned_head = &pcl->next; 7065c6dcc57SGao Xiang fe->pcl = pcl; 7079e579fc1SGao Xiang return 0; 70864094a04SGao Xiang 70964094a04SGao Xiang err_out: 71087ca34a7SGao Xiang mutex_unlock(&pcl->lock); 7119f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 71264094a04SGao Xiang return err; 71347e4937aSGao Xiang } 71447e4937aSGao Xiang 71583a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 71647e4937aSGao Xiang { 71783a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 7180d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 7199e579fc1SGao Xiang int ret; 72047e4937aSGao Xiang 72187ca34a7SGao Xiang DBG_BUGON(fe->pcl); 72247e4937aSGao Xiang 72387ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 7245c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 7255c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 72647e4937aSGao Xiang 7270d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 7280d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 7290d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 7300d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 73147e4937aSGao Xiang DBG_BUGON(1); 732cecf864dSYue Hu return -EFSCORRUPTED; 733cecf864dSYue Hu } 73447e4937aSGao Xiang 73564094a04SGao Xiang if (grp) { 7365c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 7370d823b42SGao Xiang ret = -EEXIST; 73864094a04SGao Xiang } else { 73983a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 74064094a04SGao Xiang } 74147e4937aSGao Xiang 7420d823b42SGao Xiang if (ret == -EEXIST) { 743267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 744267f2492SGao Xiang /* used to check tail merging loop due to corrupted images */ 745267f2492SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 746267f2492SGao Xiang fe->tailpcl = fe->pcl; 747267f2492SGao Xiang 748267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 7490d823b42SGao Xiang } else if (ret) { 7500d823b42SGao Xiang return ret; 7510d823b42SGao Xiang } 75206a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 753387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 75481382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 755ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 75647e4937aSGao Xiang return 0; 75747e4937aSGao Xiang } 75847e4937aSGao Xiang 75947e4937aSGao Xiang /* 76047e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 76147e4937aSGao Xiang * only after a RCU grace period. 76247e4937aSGao Xiang */ 76347e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 76447e4937aSGao Xiang { 76587ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 76687ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 76747e4937aSGao Xiang } 76847e4937aSGao Xiang 76947e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 77047e4937aSGao Xiang { 77147e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 77247e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 77347e4937aSGao Xiang 77487ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 77547e4937aSGao Xiang } 77647e4937aSGao Xiang 7775c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 77847e4937aSGao Xiang { 77987ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 78047e4937aSGao Xiang 78187ca34a7SGao Xiang if (!pcl) 78247e4937aSGao Xiang return false; 78347e4937aSGao Xiang 78406a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 78587ca34a7SGao Xiang mutex_unlock(&pcl->lock); 78647e4937aSGao Xiang 78706a304cdSGao Xiang if (fe->candidate_bvpage) { 78806a304cdSGao Xiang DBG_BUGON(z_erofs_is_shortlived_page(fe->candidate_bvpage)); 78906a304cdSGao Xiang fe->candidate_bvpage = NULL; 79006a304cdSGao Xiang } 79106a304cdSGao Xiang 79247e4937aSGao Xiang /* 79347e4937aSGao Xiang * if all pending pages are added, don't hold its reference 79447e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 79547e4937aSGao Xiang */ 796db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 79787ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 79847e4937aSGao Xiang 79987ca34a7SGao Xiang fe->pcl = NULL; 80047e4937aSGao Xiang return true; 80147e4937aSGao Xiang } 80247e4937aSGao Xiang 803b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 804b15b2e30SYue Hu struct page *page, unsigned int pageofs, 805b15b2e30SYue Hu unsigned int len) 806b15b2e30SYue Hu { 807b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 808b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 809b15b2e30SYue Hu u8 *src, *dst; 810b15b2e30SYue Hu unsigned int i, cnt; 811b15b2e30SYue Hu 812e5126de1SYue Hu if (!packed_inode) 813e5126de1SYue Hu return -EFSCORRUPTED; 814e5126de1SYue Hu 815b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 816b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 817b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 818b15b2e30SYue Hu EROFS_BLKSIZ - erofs_blkoff(pos)); 819b15b2e30SYue Hu src = erofs_bread(&buf, packed_inode, 820b15b2e30SYue Hu erofs_blknr(pos), EROFS_KMAP); 821b15b2e30SYue Hu if (IS_ERR(src)) { 822b15b2e30SYue Hu erofs_put_metabuf(&buf); 823b15b2e30SYue Hu return PTR_ERR(src); 824b15b2e30SYue Hu } 825b15b2e30SYue Hu 826b15b2e30SYue Hu dst = kmap_local_page(page); 827b15b2e30SYue Hu memcpy(dst + pageofs + i, src + erofs_blkoff(pos), cnt); 828b15b2e30SYue Hu kunmap_local(dst); 829b15b2e30SYue Hu pos += cnt; 830b15b2e30SYue Hu } 831b15b2e30SYue Hu erofs_put_metabuf(&buf); 832b15b2e30SYue Hu return 0; 833b15b2e30SYue Hu } 834b15b2e30SYue Hu 83547e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 836eaa9172aSGao Xiang struct page *page, struct page **pagepool) 83747e4937aSGao Xiang { 83847e4937aSGao Xiang struct inode *const inode = fe->inode; 83947e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 84047e4937aSGao Xiang const loff_t offset = page_offset(page); 8415b220b20SGao Xiang bool tight = true, exclusive; 8422bfab9c0SGao Xiang unsigned int cur, end, spiltted; 84347e4937aSGao Xiang int err = 0; 84447e4937aSGao Xiang 84547e4937aSGao Xiang /* register locked file pages as online pages in pack */ 84647e4937aSGao Xiang z_erofs_onlinepage_init(page); 84747e4937aSGao Xiang 84847e4937aSGao Xiang spiltted = 0; 84947e4937aSGao Xiang end = PAGE_SIZE; 85047e4937aSGao Xiang repeat: 85147e4937aSGao Xiang cur = end - 1; 85247e4937aSGao Xiang 85339397a46SGao Xiang if (offset + cur < map->m_la || 85439397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 85539397a46SGao Xiang erofs_dbg("out-of-range map @ pos %llu", offset + cur); 85647e4937aSGao Xiang 8575c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 85847e4937aSGao Xiang fe->backmost = false; 85947e4937aSGao Xiang map->m_la = offset + cur; 86047e4937aSGao Xiang map->m_llen = 0; 86147e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 8628d8a09b0SGao Xiang if (err) 86367148551SGao Xiang goto out; 86439397a46SGao Xiang } else { 86539397a46SGao Xiang if (fe->pcl) 86639397a46SGao Xiang goto hitted; 86739397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 86839397a46SGao Xiang } 86947e4937aSGao Xiang 870b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 871b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 87247e4937aSGao Xiang goto hitted; 87347e4937aSGao Xiang 87483a386c0SGao Xiang err = z_erofs_collector_begin(fe); 8758d8a09b0SGao Xiang if (err) 87667148551SGao Xiang goto out; 87747e4937aSGao Xiang 8785c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 87909c54379SGao Xiang void *mp; 880cecf864dSYue Hu 88109c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 88209c54379SGao Xiang erofs_blknr(map->m_pa), EROFS_NO_KMAP); 88309c54379SGao Xiang if (IS_ERR(mp)) { 88409c54379SGao Xiang err = PTR_ERR(mp); 885cecf864dSYue Hu erofs_err(inode->i_sb, 886cecf864dSYue Hu "failed to get inline page, err %d", err); 88767148551SGao Xiang goto out; 888cecf864dSYue Hu } 88909c54379SGao Xiang get_page(fe->map.buf.page); 890ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 891ed722fbcSGao Xiang fe->map.buf.page); 892db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 893cecf864dSYue Hu } else { 8946f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 8951282dea3SGao Xiang z_erofs_bind_cache(fe, pagepool); 896cecf864dSYue Hu } 89747e4937aSGao Xiang hitted: 898dc76ea8cSGao Xiang /* 899dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 900dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 901dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 902387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 903dc76ea8cSGao Xiang */ 904db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED && 905db166fc2SGao Xiang fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 906dc76ea8cSGao Xiang 90747e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 9088d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 90947e4937aSGao Xiang zero_user_segment(page, cur, end); 91047e4937aSGao Xiang goto next_part; 91147e4937aSGao Xiang } 912b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 913b15b2e30SYue Hu unsigned int pageofs, skip, len; 914b15b2e30SYue Hu 915b15b2e30SYue Hu if (offset > map->m_la) { 916b15b2e30SYue Hu pageofs = 0; 917b15b2e30SYue Hu skip = offset - map->m_la; 918b15b2e30SYue Hu } else { 919b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 920b15b2e30SYue Hu skip = 0; 921b15b2e30SYue Hu } 922b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 923b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 924b15b2e30SYue Hu if (err) 925b15b2e30SYue Hu goto out; 926b15b2e30SYue Hu ++spiltted; 927b15b2e30SYue Hu tight = false; 928b15b2e30SYue Hu goto next_part; 929b15b2e30SYue Hu } 93047e4937aSGao Xiang 9315b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 93247e4937aSGao Xiang if (cur) 933db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 93447e4937aSGao Xiang 93547e4937aSGao Xiang retry: 93606a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 93706a304cdSGao Xiang .page = page, 93806a304cdSGao Xiang .offset = offset - map->m_la, 93906a304cdSGao Xiang .end = end, 9405b220b20SGao Xiang }), exclusive); 94106a304cdSGao Xiang /* should allocate an additional short-lived page for bvset */ 94206a304cdSGao Xiang if (err == -EAGAIN && !fe->candidate_bvpage) { 94306a304cdSGao Xiang fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL); 94406a304cdSGao Xiang set_page_private(fe->candidate_bvpage, 94506a304cdSGao Xiang Z_EROFS_SHORTLIVED_PAGE); 94647e4937aSGao Xiang goto retry; 94747e4937aSGao Xiang } 94847e4937aSGao Xiang 94906a304cdSGao Xiang if (err) { 95006a304cdSGao Xiang DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage); 95167148551SGao Xiang goto out; 95206a304cdSGao Xiang } 95347e4937aSGao Xiang 95467148551SGao Xiang z_erofs_onlinepage_split(page); 95547e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 95647e4937aSGao Xiang ++spiltted; 957267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 958267f2492SGao Xiang fe->pcl->multibases = true; 9592bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 9602bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 9612bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 9622bfab9c0SGao Xiang } 963e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 964e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 965e7933278SGao Xiang fe->pcl->length == map->m_llen) 966e7933278SGao Xiang fe->pcl->partial = false; 96747e4937aSGao Xiang next_part: 9682bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 96947e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 9702bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 97147e4937aSGao Xiang 97247e4937aSGao Xiang end = cur; 97347e4937aSGao Xiang if (end > 0) 97447e4937aSGao Xiang goto repeat; 97547e4937aSGao Xiang 97647e4937aSGao Xiang out: 97767148551SGao Xiang if (err) 97867148551SGao Xiang z_erofs_page_mark_eio(page); 97947e4937aSGao Xiang z_erofs_onlinepage_endio(page); 98047e4937aSGao Xiang 9814f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 98247e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 98347e4937aSGao Xiang return err; 98447e4937aSGao Xiang } 98547e4937aSGao Xiang 98640452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 98740452ffcSHuang Jianan unsigned int readahead_pages) 98840452ffcSHuang Jianan { 989a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 99040452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 99140452ffcSHuang Jianan !readahead_pages) 99240452ffcSHuang Jianan return true; 99340452ffcSHuang Jianan 99440452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 99540452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 99640452ffcSHuang Jianan return true; 99740452ffcSHuang Jianan 99840452ffcSHuang Jianan return false; 99940452ffcSHuang Jianan } 100040452ffcSHuang Jianan 10016aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 10026aaa7b06SGao Xiang { 10036aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 10046aaa7b06SGao Xiang } 10056aaa7b06SGao Xiang 10064f05687fSGao Xiang struct z_erofs_decompress_backend { 10074f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 10084f05687fSGao Xiang struct super_block *sb; 10094f05687fSGao Xiang struct z_erofs_pcluster *pcl; 10104f05687fSGao Xiang 10114f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 10124f05687fSGao Xiang struct page **decompressed_pages; 10134f05687fSGao Xiang /* pages to keep the compressed data */ 10144f05687fSGao Xiang struct page **compressed_pages; 10154f05687fSGao Xiang 1016267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 10174f05687fSGao Xiang struct page **pagepool; 10182bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 10194f05687fSGao Xiang }; 10204f05687fSGao Xiang 1021267f2492SGao Xiang struct z_erofs_bvec_item { 1022267f2492SGao Xiang struct z_erofs_bvec bvec; 1023267f2492SGao Xiang struct list_head list; 1024267f2492SGao Xiang }; 1025267f2492SGao Xiang 1026267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 10273fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 10283fe96ee0SGao Xiang { 1029267f2492SGao Xiang struct z_erofs_bvec_item *item; 1030267f2492SGao Xiang 1031267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1032267f2492SGao Xiang unsigned int pgnr; 10333fe96ee0SGao Xiang 1034267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 10352bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 103663bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 10373fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 1038267f2492SGao Xiang return; 10393fe96ee0SGao Xiang } 104063bbb856SGao Xiang } 10413fe96ee0SGao Xiang 1042267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 1043267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1044267f2492SGao Xiang item->bvec = *bvec; 1045267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 1046267f2492SGao Xiang } 1047267f2492SGao Xiang 1048267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1049267f2492SGao Xiang int err) 1050267f2492SGao Xiang { 1051267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 1052267f2492SGao Xiang struct list_head *p, *n; 1053267f2492SGao Xiang 1054267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1055267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 1056267f2492SGao Xiang unsigned int end, cur; 1057267f2492SGao Xiang void *dst, *src; 1058267f2492SGao Xiang 1059267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 1060267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1061267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1062267f2492SGao Xiang bvi->bvec.end); 1063267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 1064267f2492SGao Xiang while (cur < end) { 1065267f2492SGao Xiang unsigned int pgnr, scur, len; 1066267f2492SGao Xiang 1067267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1068267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 1069267f2492SGao Xiang 1070267f2492SGao Xiang scur = bvi->bvec.offset + cur - 1071267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 1072267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1073267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 1074267f2492SGao Xiang err = -EFSCORRUPTED; 1075267f2492SGao Xiang cur += len; 1076267f2492SGao Xiang continue; 1077267f2492SGao Xiang } 1078267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 1079267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 1080267f2492SGao Xiang kunmap_local(src); 1081267f2492SGao Xiang cur += len; 1082267f2492SGao Xiang } 1083267f2492SGao Xiang kunmap_local(dst); 1084267f2492SGao Xiang if (err) 1085267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 1086267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 1087267f2492SGao Xiang list_del(p); 1088267f2492SGao Xiang kfree(bvi); 1089267f2492SGao Xiang } 1090267f2492SGao Xiang } 1091267f2492SGao Xiang 1092267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 109342fec235SGao Xiang { 10944f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 109506a304cdSGao Xiang struct z_erofs_bvec_iter biter; 109606a304cdSGao Xiang struct page *old_bvpage; 1097267f2492SGao Xiang int i; 109842fec235SGao Xiang 1099387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 110042fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 110106a304cdSGao Xiang struct z_erofs_bvec bvec; 110242fec235SGao Xiang 110306a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 110442fec235SGao Xiang 110506a304cdSGao Xiang if (old_bvpage) 11064f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 110742fec235SGao Xiang 110806a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1109267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 111042fec235SGao Xiang } 111106a304cdSGao Xiang 111206a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 111306a304cdSGao Xiang if (old_bvpage) 11144f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 111542fec235SGao Xiang } 111642fec235SGao Xiang 11174f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 11184f05687fSGao Xiang bool *overlapped) 111967139e36SGao Xiang { 11204f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 112167139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 112267139e36SGao Xiang int i, err = 0; 112367139e36SGao Xiang 112467139e36SGao Xiang *overlapped = false; 112567139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 1126ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1127ed722fbcSGao Xiang struct page *page = bvec->page; 112867139e36SGao Xiang 112967139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 113067139e36SGao Xiang if (!page) { 113167139e36SGao Xiang DBG_BUGON(1); 113267139e36SGao Xiang continue; 113367139e36SGao Xiang } 1134fe3e5914SGao Xiang be->compressed_pages[i] = page; 113567139e36SGao Xiang 113667139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 113767139e36SGao Xiang if (!PageUptodate(page)) 113867139e36SGao Xiang err = -EIO; 113967139e36SGao Xiang continue; 114067139e36SGao Xiang } 114167139e36SGao Xiang 114267139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 114367139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 11444f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 114567139e36SGao Xiang if (!PageUptodate(page)) 114667139e36SGao Xiang err = -EIO; 114767139e36SGao Xiang continue; 114867139e36SGao Xiang } 1149267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 115067139e36SGao Xiang *overlapped = true; 115167139e36SGao Xiang } 115267139e36SGao Xiang } 115367139e36SGao Xiang 1154fe3e5914SGao Xiang if (err) 11554f05687fSGao Xiang return err; 11564f05687fSGao Xiang return 0; 115767139e36SGao Xiang } 115867139e36SGao Xiang 11594f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 11604f05687fSGao Xiang int err) 116147e4937aSGao Xiang { 11624f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 11634f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1164cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 11652bfab9c0SGao Xiang unsigned int i, inputsize; 116667148551SGao Xiang int err2; 11672bfab9c0SGao Xiang struct page *page; 11682bfab9c0SGao Xiang bool overlapped; 116947e4937aSGao Xiang 117087ca34a7SGao Xiang mutex_lock(&pcl->lock); 11712bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 117247e4937aSGao Xiang 1173fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1174fe3e5914SGao Xiang be->decompressed_pages = NULL; 1175fe3e5914SGao Xiang be->compressed_pages = NULL; 1176fe3e5914SGao Xiang be->onstack_used = 0; 11772bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 11784f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 11792bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 11804f05687fSGao Xiang memset(be->decompressed_pages, 0, 11812bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1182fe3e5914SGao Xiang } 1183fe3e5914SGao Xiang 1184fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1185fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1186fe3e5914SGao Xiang 1187fe3e5914SGao Xiang if (!be->decompressed_pages) 11884f05687fSGao Xiang be->decompressed_pages = 118912724ba3SGao Xiang kcalloc(be->nr_pages, sizeof(struct page *), 1190e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1191fe3e5914SGao Xiang if (!be->compressed_pages) 1192fe3e5914SGao Xiang be->compressed_pages = 119312724ba3SGao Xiang kcalloc(pclusterpages, sizeof(struct page *), 1194fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 119547e4937aSGao Xiang 1196267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 11974f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 11984f05687fSGao Xiang if (err2) 11994f05687fSGao Xiang err = err2; 12008d8a09b0SGao Xiang if (err) 120147e4937aSGao Xiang goto out; 120247e4937aSGao Xiang 1203cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1204cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1205cecf864dSYue Hu else 1206cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1207cecf864dSYue Hu 120847e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 12094f05687fSGao Xiang .sb = be->sb, 12104f05687fSGao Xiang .in = be->compressed_pages, 12114f05687fSGao Xiang .out = be->decompressed_pages, 1212cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 121387ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 12149f6cc76eSGao Xiang .inputsize = inputsize, 12152bfab9c0SGao Xiang .outputsize = pcl->length, 121647e4937aSGao Xiang .alg = pcl->algorithmformat, 121747e4937aSGao Xiang .inplace_io = overlapped, 12182bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1219267f2492SGao Xiang .fillgaps = pcl->multibases, 12204f05687fSGao Xiang }, be->pagepool); 122147e4937aSGao Xiang 122247e4937aSGao Xiang out: 1223cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1224cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1225ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1226ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1227cecf864dSYue Hu put_page(page); 1228cecf864dSYue Hu } else { 1229cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1230ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 123147e4937aSGao Xiang 123247e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 123347e4937aSGao Xiang continue; 123447e4937aSGao Xiang 12356aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 12364f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1237ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 123847e4937aSGao Xiang } 1239cecf864dSYue Hu } 1240fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1241fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 124212724ba3SGao Xiang kfree(be->compressed_pages); 1243267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 124447e4937aSGao Xiang 12452bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 12464f05687fSGao Xiang page = be->decompressed_pages[i]; 124747e4937aSGao Xiang if (!page) 124847e4937aSGao Xiang continue; 124947e4937aSGao Xiang 12506aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 125147e4937aSGao Xiang 12526aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 12534f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 125447e4937aSGao Xiang continue; 125567148551SGao Xiang if (err) 125667148551SGao Xiang z_erofs_page_mark_eio(page); 125747e4937aSGao Xiang z_erofs_onlinepage_endio(page); 125847e4937aSGao Xiang } 125947e4937aSGao Xiang 12604f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 126112724ba3SGao Xiang kfree(be->decompressed_pages); 126247e4937aSGao Xiang 12632bfab9c0SGao Xiang pcl->length = 0; 12642bfab9c0SGao Xiang pcl->partial = true; 1265267f2492SGao Xiang pcl->multibases = false; 126606a304cdSGao Xiang pcl->bvset.nextpage = NULL; 126787ca34a7SGao Xiang pcl->vcnt = 0; 126847e4937aSGao Xiang 126987ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 127047e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 127187ca34a7SGao Xiang mutex_unlock(&pcl->lock); 127247e4937aSGao Xiang return err; 127347e4937aSGao Xiang } 127447e4937aSGao Xiang 12750c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1276eaa9172aSGao Xiang struct page **pagepool) 127747e4937aSGao Xiang { 12784f05687fSGao Xiang struct z_erofs_decompress_backend be = { 12794f05687fSGao Xiang .sb = io->sb, 12804f05687fSGao Xiang .pagepool = pagepool, 1281267f2492SGao Xiang .decompressed_secondary_bvecs = 1282267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 12834f05687fSGao Xiang }; 128447e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 128547e4937aSGao Xiang 128647e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 12874f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 128847e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 12894f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */ 129047e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 129147e4937aSGao Xiang 12924f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 12934f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 129447e4937aSGao Xiang 12954f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 12964f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 129747e4937aSGao Xiang } 129847e4937aSGao Xiang } 129947e4937aSGao Xiang 13000c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 130147e4937aSGao Xiang { 1302a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1303a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1304eaa9172aSGao Xiang struct page *pagepool = NULL; 130547e4937aSGao Xiang 1306a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 13070c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 130847e4937aSGao Xiang 1309eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1310a4b1fab1SGao Xiang kvfree(bgq); 131147e4937aSGao Xiang } 131247e4937aSGao Xiang 13137865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1314cdba5506SGao Xiang int bios) 13157865827cSGao Xiang { 13167865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 13177865827cSGao Xiang 13187865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1319cdba5506SGao Xiang if (io->sync) { 13207865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 132160b30050SHongyu Jin complete(&io->u.done); 13227865827cSGao Xiang return; 13237865827cSGao Xiang } 13247865827cSGao Xiang 13257865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 13267865827cSGao Xiang return; 13277865827cSGao Xiang /* Use workqueue and sync decompression for atomic contexts only */ 13287865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 13297865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 13307865827cSGao Xiang /* enable sync decompression for readahead */ 13317865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 13327865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 13337865827cSGao Xiang return; 13347865827cSGao Xiang } 13357865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 13367865827cSGao Xiang } 13377865827cSGao Xiang 133847e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 133947e4937aSGao Xiang unsigned int nr, 1340eaa9172aSGao Xiang struct page **pagepool, 13419f2731d6SGao Xiang struct address_space *mc) 134247e4937aSGao Xiang { 134347e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 13449f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 134547e4937aSGao Xiang bool tocache = false; 134647e4937aSGao Xiang 134747e4937aSGao Xiang struct address_space *mapping; 134847e4937aSGao Xiang struct page *oldpage, *page; 134947e4937aSGao Xiang int justfound; 135047e4937aSGao Xiang 135147e4937aSGao Xiang repeat: 1352ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 135347e4937aSGao Xiang oldpage = page; 135447e4937aSGao Xiang 135547e4937aSGao Xiang if (!page) 135647e4937aSGao Xiang goto out_allocpage; 135747e4937aSGao Xiang 1358b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1359b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 136047e4937aSGao Xiang 13611825c8d7SGao Xiang /* 13621825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 13631825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 13641825c8d7SGao Xiang */ 13651825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1366ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 13671825c8d7SGao Xiang set_page_private(page, 0); 13681825c8d7SGao Xiang tocache = true; 13691825c8d7SGao Xiang goto out_tocache; 13701825c8d7SGao Xiang } 137147e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 137247e4937aSGao Xiang 137347e4937aSGao Xiang /* 13746aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 137547e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 137647e4937aSGao Xiang */ 137747e4937aSGao Xiang if (mapping && mapping != mc) 137847e4937aSGao Xiang /* ought to be unmanaged pages */ 137947e4937aSGao Xiang goto out; 138047e4937aSGao Xiang 13816aaa7b06SGao Xiang /* directly return for shortlived page as well */ 13826aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 13836aaa7b06SGao Xiang goto out; 13846aaa7b06SGao Xiang 138547e4937aSGao Xiang lock_page(page); 138647e4937aSGao Xiang 138747e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 138847e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 138947e4937aSGao Xiang 139047e4937aSGao Xiang /* the page is still in manage cache */ 139147e4937aSGao Xiang if (page->mapping == mc) { 1392ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 139347e4937aSGao Xiang 139447e4937aSGao Xiang if (!PagePrivate(page)) { 139547e4937aSGao Xiang /* 139647e4937aSGao Xiang * impossible to be !PagePrivate(page) for 139747e4937aSGao Xiang * the current restriction as well if 1398ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 139947e4937aSGao Xiang */ 140047e4937aSGao Xiang DBG_BUGON(!justfound); 140147e4937aSGao Xiang 140247e4937aSGao Xiang justfound = 0; 140347e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 140447e4937aSGao Xiang SetPagePrivate(page); 140547e4937aSGao Xiang } 140647e4937aSGao Xiang 140747e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 140847e4937aSGao Xiang if (PageUptodate(page)) { 140947e4937aSGao Xiang unlock_page(page); 141047e4937aSGao Xiang page = NULL; 141147e4937aSGao Xiang } 141247e4937aSGao Xiang goto out; 141347e4937aSGao Xiang } 141447e4937aSGao Xiang 141547e4937aSGao Xiang /* 141647e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 141747e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 141847e4937aSGao Xiang */ 141947e4937aSGao Xiang DBG_BUGON(page->mapping); 142047e4937aSGao Xiang DBG_BUGON(!justfound); 142147e4937aSGao Xiang 142247e4937aSGao Xiang tocache = true; 142347e4937aSGao Xiang unlock_page(page); 142447e4937aSGao Xiang put_page(page); 142547e4937aSGao Xiang out_allocpage: 14265ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1427ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1428ed722fbcSGao Xiang oldpage, page)) { 1429eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 14305ddcee1fSGao Xiang cond_resched(); 14315ddcee1fSGao Xiang goto repeat; 14325ddcee1fSGao Xiang } 14331825c8d7SGao Xiang out_tocache: 1434bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1435bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1436bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1437bf225074SGao Xiang goto out; 1438a30573b3SGao Xiang } 1439bf225074SGao Xiang attach_page_private(page, pcl); 1440bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1441bf225074SGao Xiang put_page(page); 1442bf225074SGao Xiang 144347e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 144447e4937aSGao Xiang return page; 144547e4937aSGao Xiang } 144647e4937aSGao Xiang 1447cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1448a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 144947e4937aSGao Xiang { 1450a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 145147e4937aSGao Xiang 1452a4b1fab1SGao Xiang if (fg && !*fg) { 1453a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1454a4b1fab1SGao Xiang if (!q) { 1455a4b1fab1SGao Xiang *fg = true; 1456a4b1fab1SGao Xiang goto fg_out; 145747e4937aSGao Xiang } 14580c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1459a4b1fab1SGao Xiang } else { 1460a4b1fab1SGao Xiang fg_out: 1461a4b1fab1SGao Xiang q = fgq; 146260b30050SHongyu Jin init_completion(&fgq->u.done); 1463a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 146467148551SGao Xiang q->eio = false; 1465cdba5506SGao Xiang q->sync = true; 1466a4b1fab1SGao Xiang } 1467a4b1fab1SGao Xiang q->sb = sb; 1468a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1469a4b1fab1SGao Xiang return q; 147047e4937aSGao Xiang } 147147e4937aSGao Xiang 147247e4937aSGao Xiang /* define decompression jobqueue types */ 147347e4937aSGao Xiang enum { 147447e4937aSGao Xiang JQ_BYPASS, 147547e4937aSGao Xiang JQ_SUBMIT, 147647e4937aSGao Xiang NR_JOBQUEUES, 147747e4937aSGao Xiang }; 147847e4937aSGao Xiang 147947e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 148047e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 148147e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 148247e4937aSGao Xiang { 148347e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 148447e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 148547e4937aSGao Xiang 148647e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 148747e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 148847e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 148947e4937aSGao Xiang 149047e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 149147e4937aSGao Xiang 149247e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 149347e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 149447e4937aSGao Xiang 149547e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 149647e4937aSGao Xiang } 149747e4937aSGao Xiang 14987865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 14997865827cSGao Xiang { 1500cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 15017865827cSGao Xiang blk_status_t err = bio->bi_status; 15027865827cSGao Xiang struct bio_vec *bvec; 15037865827cSGao Xiang struct bvec_iter_all iter_all; 15047865827cSGao Xiang 15057865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 15067865827cSGao Xiang struct page *page = bvec->bv_page; 15077865827cSGao Xiang 15087865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 15097865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 15107865827cSGao Xiang 15117865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 15127865827cSGao Xiang if (!err) 15137865827cSGao Xiang SetPageUptodate(page); 15147865827cSGao Xiang unlock_page(page); 15157865827cSGao Xiang } 15167865827cSGao Xiang } 151767148551SGao Xiang if (err) 151867148551SGao Xiang q->eio = true; 1519cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 15207865827cSGao Xiang bio_put(bio); 15217865827cSGao Xiang } 15227865827cSGao Xiang 152383a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1524eaa9172aSGao Xiang struct page **pagepool, 1525a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1526a4b1fab1SGao Xiang bool *force_fg) 152747e4937aSGao Xiang { 152883a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 152983a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 153047e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1531a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 15325c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1533dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 15343f649ab7SKees Cook pgoff_t last_index; 1535dfeab2e9SGao Xiang struct block_device *last_bdev; 15361e4a2955SGao Xiang unsigned int nr_bios = 0; 15371e4a2955SGao Xiang struct bio *bio = NULL; 153882e60d00SJohannes Weiner unsigned long pflags; 153982e60d00SJohannes Weiner int memstall = 0; 154047e4937aSGao Xiang 1541cdba5506SGao Xiang /* 1542cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1543cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1544cdba5506SGao Xiang */ 1545cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1546cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1547cdba5506SGao Xiang 1548a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1549a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 155047e4937aSGao Xiang 155147e4937aSGao Xiang /* by default, all need io submission */ 155247e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 155347e4937aSGao Xiang 155447e4937aSGao Xiang do { 1555dfeab2e9SGao Xiang struct erofs_map_dev mdev; 155647e4937aSGao Xiang struct z_erofs_pcluster *pcl; 15571e4a2955SGao Xiang pgoff_t cur, end; 15581e4a2955SGao Xiang unsigned int i = 0; 15591e4a2955SGao Xiang bool bypass = true; 156047e4937aSGao Xiang 156147e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 156247e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 156347e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 156447e4937aSGao Xiang 156547e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 156647e4937aSGao Xiang 1567cecf864dSYue Hu /* close the main owned chain at first */ 1568cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1569cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1570cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1571cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1572cecf864dSYue Hu continue; 1573cecf864dSYue Hu } 1574cecf864dSYue Hu 1575dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1576dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1577dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1578dfeab2e9SGao Xiang }; 1579dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1580dfeab2e9SGao Xiang 1581dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 15829f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 158347e4937aSGao Xiang 15841e4a2955SGao Xiang do { 15851e4a2955SGao Xiang struct page *page; 158647e4937aSGao Xiang 15871e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 158883a386c0SGao Xiang mc); 15891e4a2955SGao Xiang if (!page) 15901e4a2955SGao Xiang continue; 159147e4937aSGao Xiang 1592dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1593dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 159447e4937aSGao Xiang submit_bio_retry: 159594e4e153SGao Xiang submit_bio(bio); 159682e60d00SJohannes Weiner if (memstall) { 159782e60d00SJohannes Weiner psi_memstall_leave(&pflags); 159882e60d00SJohannes Weiner memstall = 0; 159982e60d00SJohannes Weiner } 160047e4937aSGao Xiang bio = NULL; 160147e4937aSGao Xiang } 160247e4937aSGao Xiang 160382e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 160499486c51SChristoph Hellwig psi_memstall_enter(&pflags); 160582e60d00SJohannes Weiner memstall = 1; 160682e60d00SJohannes Weiner } 160799486c51SChristoph Hellwig 160847e4937aSGao Xiang if (!bio) { 160907888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 161007888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 16110c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1612dfeab2e9SGao Xiang 1613dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 16141e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1615a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1616cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 16176ea5aad3SGao Xiang if (f->readahead) 16186ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 161947e4937aSGao Xiang ++nr_bios; 162047e4937aSGao Xiang } 162147e4937aSGao Xiang 16226c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 162347e4937aSGao Xiang goto submit_bio_retry; 162447e4937aSGao Xiang 16251e4a2955SGao Xiang last_index = cur; 16261e4a2955SGao Xiang bypass = false; 16271e4a2955SGao Xiang } while (++cur < end); 162847e4937aSGao Xiang 16291e4a2955SGao Xiang if (!bypass) 163047e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 163147e4937aSGao Xiang else 163247e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 163347e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 163447e4937aSGao Xiang 163599486c51SChristoph Hellwig if (bio) { 163694e4e153SGao Xiang submit_bio(bio); 163782e60d00SJohannes Weiner if (memstall) 163882e60d00SJohannes Weiner psi_memstall_leave(&pflags); 163999486c51SChristoph Hellwig } 164047e4937aSGao Xiang 1641587a67b7SGao Xiang /* 1642587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1643587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1644587a67b7SGao Xiang */ 1645587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1646587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 16471e4a2955SGao Xiang return; 1648587a67b7SGao Xiang } 1649cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 165047e4937aSGao Xiang } 165147e4937aSGao Xiang 165283a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1653eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 165447e4937aSGao Xiang { 1655a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 165647e4937aSGao Xiang 16575c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 165847e4937aSGao Xiang return; 165983a386c0SGao Xiang z_erofs_submit_queue(f, pagepool, io, &force_fg); 166047e4937aSGao Xiang 16610c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 16620c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 166347e4937aSGao Xiang 166447e4937aSGao Xiang if (!force_fg) 166547e4937aSGao Xiang return; 166647e4937aSGao Xiang 166747e4937aSGao Xiang /* wait until all bios are completed */ 166860b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 166947e4937aSGao Xiang 16700c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 16710c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 167247e4937aSGao Xiang } 167347e4937aSGao Xiang 167438629291SGao Xiang /* 167538629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 167638629291SGao Xiang * approximate readmore strategies as a start. 167738629291SGao Xiang */ 167838629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 167938629291SGao Xiang struct readahead_control *rac, 168038629291SGao Xiang erofs_off_t end, 1681eaa9172aSGao Xiang struct page **pagepool, 168238629291SGao Xiang bool backmost) 168338629291SGao Xiang { 168438629291SGao Xiang struct inode *inode = f->inode; 168538629291SGao Xiang struct erofs_map_blocks *map = &f->map; 168638629291SGao Xiang erofs_off_t cur; 168738629291SGao Xiang int err; 168838629291SGao Xiang 168938629291SGao Xiang if (backmost) { 169038629291SGao Xiang map->m_la = end; 1691622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1692622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 169338629291SGao Xiang if (err) 169438629291SGao Xiang return; 169538629291SGao Xiang 169638629291SGao Xiang /* expend ra for the trailing edge if readahead */ 169738629291SGao Xiang if (rac) { 169838629291SGao Xiang loff_t newstart = readahead_pos(rac); 169938629291SGao Xiang 170038629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 170138629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 170238629291SGao Xiang return; 170338629291SGao Xiang } 170438629291SGao Xiang end = round_up(end, PAGE_SIZE); 170538629291SGao Xiang } else { 170638629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 170738629291SGao Xiang 170838629291SGao Xiang if (!map->m_llen) 170938629291SGao Xiang return; 171038629291SGao Xiang } 171138629291SGao Xiang 171238629291SGao Xiang cur = map->m_la + map->m_llen - 1; 171338629291SGao Xiang while (cur >= end) { 171438629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 171538629291SGao Xiang struct page *page; 171638629291SGao Xiang 171738629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1718aa793b46SGao Xiang if (page) { 171938629291SGao Xiang if (PageUptodate(page)) { 172038629291SGao Xiang unlock_page(page); 1721aa793b46SGao Xiang } else { 172238629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 172338629291SGao Xiang if (err) 172438629291SGao Xiang erofs_err(inode->i_sb, 172538629291SGao Xiang "readmore error at page %lu @ nid %llu", 172638629291SGao Xiang index, EROFS_I(inode)->nid); 1727aa793b46SGao Xiang } 172838629291SGao Xiang put_page(page); 1729aa793b46SGao Xiang } 1730aa793b46SGao Xiang 173138629291SGao Xiang if (cur < PAGE_SIZE) 173238629291SGao Xiang break; 173338629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 173438629291SGao Xiang } 173538629291SGao Xiang } 173638629291SGao Xiang 1737a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 173847e4937aSGao Xiang { 1739a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 174047e4937aSGao Xiang struct inode *const inode = page->mapping->host; 174140452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 174247e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1743eaa9172aSGao Xiang struct page *pagepool = NULL; 174447e4937aSGao Xiang int err; 174547e4937aSGao Xiang 174647e4937aSGao Xiang trace_erofs_readpage(page, false); 174747e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 174847e4937aSGao Xiang 174938629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 175038629291SGao Xiang &pagepool, true); 17511825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 175238629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 175338629291SGao Xiang 17545c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 175547e4937aSGao Xiang 175647e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 175783a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 175840452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 175947e4937aSGao Xiang 176047e4937aSGao Xiang if (err) 17614f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 176247e4937aSGao Xiang 176309c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1764eaa9172aSGao Xiang erofs_release_pages(&pagepool); 176547e4937aSGao Xiang return err; 176647e4937aSGao Xiang } 176747e4937aSGao Xiang 17680615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 176947e4937aSGao Xiang { 17700615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 177147e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 177247e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1773eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 177438629291SGao Xiang unsigned int nr_pages; 177547e4937aSGao Xiang 17766ea5aad3SGao Xiang f.readahead = true; 17770615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 177847e4937aSGao Xiang 177938629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 178038629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 178138629291SGao Xiang nr_pages = readahead_count(rac); 178238629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 178338629291SGao Xiang 17840615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 178547e4937aSGao Xiang set_page_private(page, (unsigned long)head); 178647e4937aSGao Xiang head = page; 178747e4937aSGao Xiang } 178847e4937aSGao Xiang 178947e4937aSGao Xiang while (head) { 179047e4937aSGao Xiang struct page *page = head; 179147e4937aSGao Xiang int err; 179247e4937aSGao Xiang 179347e4937aSGao Xiang /* traversal in reverse order */ 179447e4937aSGao Xiang head = (void *)page_private(page); 179547e4937aSGao Xiang 17961825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1797a5876e24SGao Xiang if (err) 17984f761fa2SGao Xiang erofs_err(inode->i_sb, 17994f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 18004f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 180147e4937aSGao Xiang put_page(page); 180247e4937aSGao Xiang } 180338629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 18045c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 180547e4937aSGao Xiang 180683a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 180740452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 180809c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1809eaa9172aSGao Xiang erofs_release_pages(&pagepool); 181047e4937aSGao Xiang } 181147e4937aSGao Xiang 18120c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1813a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 18140615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 181547e4937aSGao Xiang }; 1816