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 "zdata.h" 847e4937aSGao Xiang #include "compress.h" 947e4937aSGao Xiang #include <linux/prefetch.h> 1099486c51SChristoph Hellwig #include <linux/psi.h> 1147e4937aSGao Xiang 1247e4937aSGao Xiang #include <trace/events/erofs.h> 1347e4937aSGao Xiang 1447e4937aSGao Xiang /* 159f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 169f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 179f6cc76eSGao Xiang */ 189f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 199f6cc76eSGao Xiang struct kmem_cache *slab; 209f6cc76eSGao Xiang unsigned int maxpages; 219f6cc76eSGao Xiang char name[48]; 229f6cc76eSGao Xiang }; 239f6cc76eSGao Xiang 249f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 259f6cc76eSGao Xiang 269f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 279f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 289f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 299f6cc76eSGao Xiang }; 309f6cc76eSGao Xiang 3106a304cdSGao Xiang struct z_erofs_bvec_iter { 3206a304cdSGao Xiang struct page *bvpage; 3306a304cdSGao Xiang struct z_erofs_bvset *bvset; 3406a304cdSGao Xiang unsigned int nr, cur; 3506a304cdSGao Xiang }; 3606a304cdSGao Xiang 3706a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 3806a304cdSGao Xiang { 3906a304cdSGao Xiang if (iter->bvpage) 4006a304cdSGao Xiang kunmap_local(iter->bvset); 4106a304cdSGao Xiang return iter->bvpage; 4206a304cdSGao Xiang } 4306a304cdSGao Xiang 4406a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 4506a304cdSGao Xiang { 4606a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 4706a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 4806a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 4906a304cdSGao Xiang struct page *oldpage; 5006a304cdSGao Xiang 5106a304cdSGao Xiang DBG_BUGON(!nextpage); 5206a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 5306a304cdSGao Xiang iter->bvpage = nextpage; 5406a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 5506a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 5606a304cdSGao Xiang iter->cur = 0; 5706a304cdSGao Xiang return oldpage; 5806a304cdSGao Xiang } 5906a304cdSGao Xiang 6006a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 6106a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 6206a304cdSGao Xiang unsigned int bootstrap_nr, 6306a304cdSGao Xiang unsigned int cur) 6406a304cdSGao Xiang { 6506a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 6606a304cdSGao Xiang .nr = bootstrap_nr, 6706a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 6806a304cdSGao Xiang }; 6906a304cdSGao Xiang 7006a304cdSGao Xiang while (cur > iter->nr) { 7106a304cdSGao Xiang cur -= iter->nr; 7206a304cdSGao Xiang z_erofs_bvset_flip(iter); 7306a304cdSGao Xiang } 7406a304cdSGao Xiang iter->cur = cur; 7506a304cdSGao Xiang } 7606a304cdSGao Xiang 7706a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 7806a304cdSGao Xiang struct z_erofs_bvec *bvec, 7906a304cdSGao Xiang struct page **candidate_bvpage) 8006a304cdSGao Xiang { 8106a304cdSGao Xiang if (iter->cur == iter->nr) { 8206a304cdSGao Xiang if (!*candidate_bvpage) 8306a304cdSGao Xiang return -EAGAIN; 8406a304cdSGao Xiang 8506a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 8606a304cdSGao Xiang iter->bvset->nextpage = *candidate_bvpage; 8706a304cdSGao Xiang z_erofs_bvset_flip(iter); 8806a304cdSGao Xiang 8906a304cdSGao Xiang iter->bvset->nextpage = NULL; 9006a304cdSGao Xiang *candidate_bvpage = NULL; 9106a304cdSGao Xiang } 9206a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 9306a304cdSGao Xiang return 0; 9406a304cdSGao Xiang } 9506a304cdSGao Xiang 9606a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 9706a304cdSGao Xiang struct z_erofs_bvec *bvec, 9806a304cdSGao Xiang struct page **old_bvpage) 9906a304cdSGao Xiang { 10006a304cdSGao Xiang if (iter->cur == iter->nr) 10106a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 10206a304cdSGao Xiang else 10306a304cdSGao Xiang *old_bvpage = NULL; 10406a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 10506a304cdSGao Xiang } 10606a304cdSGao Xiang 1079f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 1089f6cc76eSGao Xiang { 1099f6cc76eSGao Xiang int i; 1109f6cc76eSGao Xiang 1119f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 1129f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 1139f6cc76eSGao Xiang continue; 1149f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 1159f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 1169f6cc76eSGao Xiang } 1179f6cc76eSGao Xiang } 1189f6cc76eSGao Xiang 1199f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 1209f6cc76eSGao Xiang { 1219f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 1229f6cc76eSGao Xiang struct z_erofs_pcluster *a; 1239f6cc76eSGao Xiang unsigned int size; 1249f6cc76eSGao Xiang 1259f6cc76eSGao Xiang for (pcs = pcluster_pool; 1269f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 127ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, pcs->maxpages); 1289f6cc76eSGao Xiang 1299f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 1309f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 1319f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 1329f6cc76eSGao Xiang if (pcs->slab) 1339f6cc76eSGao Xiang continue; 1349f6cc76eSGao Xiang 1359f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 1369f6cc76eSGao Xiang return -ENOMEM; 1379f6cc76eSGao Xiang } 1389f6cc76eSGao Xiang return 0; 1399f6cc76eSGao Xiang } 1409f6cc76eSGao Xiang 1419f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 1429f6cc76eSGao Xiang { 1439f6cc76eSGao Xiang int i; 1449f6cc76eSGao Xiang 1459f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 1469f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 1479f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 1489f6cc76eSGao Xiang 1499f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 1509f6cc76eSGao Xiang continue; 1519f6cc76eSGao Xiang 1529f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 1539f6cc76eSGao Xiang if (!pcl) 1549f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 1559f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 1569f6cc76eSGao Xiang return pcl; 1579f6cc76eSGao Xiang } 1589f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 1599f6cc76eSGao Xiang } 1609f6cc76eSGao Xiang 1619f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 1629f6cc76eSGao Xiang { 163cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1649f6cc76eSGao Xiang int i; 1659f6cc76eSGao Xiang 1669f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 1679f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 1689f6cc76eSGao Xiang 169cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 1709f6cc76eSGao Xiang continue; 1719f6cc76eSGao Xiang 1729f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 1739f6cc76eSGao Xiang return; 1749f6cc76eSGao Xiang } 1759f6cc76eSGao Xiang DBG_BUGON(1); 1769f6cc76eSGao Xiang } 1779f6cc76eSGao Xiang 17847e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 17947e4937aSGao Xiang 18047e4937aSGao Xiang void z_erofs_exit_zip_subsystem(void) 18147e4937aSGao Xiang { 18247e4937aSGao Xiang destroy_workqueue(z_erofs_workqueue); 1839f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 18447e4937aSGao Xiang } 18547e4937aSGao Xiang 18699634bf3SGao Xiang static inline int z_erofs_init_workqueue(void) 18747e4937aSGao Xiang { 18847e4937aSGao Xiang const unsigned int onlinecpus = num_possible_cpus(); 18947e4937aSGao Xiang 19047e4937aSGao Xiang /* 19147e4937aSGao Xiang * no need to spawn too many threads, limiting threads could minimum 19247e4937aSGao Xiang * scheduling overhead, perhaps per-CPU threads should be better? 19347e4937aSGao Xiang */ 1940e62ea33SGao Xiang z_erofs_workqueue = alloc_workqueue("erofs_unzipd", 1950e62ea33SGao Xiang WQ_UNBOUND | WQ_HIGHPRI, 19647e4937aSGao Xiang onlinecpus + onlinecpus / 4); 19747e4937aSGao Xiang return z_erofs_workqueue ? 0 : -ENOMEM; 19847e4937aSGao Xiang } 19947e4937aSGao Xiang 20047e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 20147e4937aSGao Xiang { 2029f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 20347e4937aSGao Xiang 2049f6cc76eSGao Xiang if (err) 2059f6cc76eSGao Xiang return err; 2069f6cc76eSGao Xiang err = z_erofs_init_workqueue(); 2079f6cc76eSGao Xiang if (err) 2089f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 2099f6cc76eSGao Xiang return err; 21047e4937aSGao Xiang } 21147e4937aSGao Xiang 212db166fc2SGao Xiang enum z_erofs_pclustermode { 213db166fc2SGao Xiang Z_EROFS_PCLUSTER_INFLIGHT, 21447e4937aSGao Xiang /* 215db166fc2SGao Xiang * The current pclusters was the tail of an exist chain, in addition 216db166fc2SGao Xiang * that the previous processed chained pclusters are all decided to 21747e4937aSGao Xiang * be hooked up to it. 218db166fc2SGao Xiang * A new chain will be created for the remaining pclusters which are 219db166fc2SGao Xiang * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED, 220db166fc2SGao Xiang * the next pcluster cannot reuse the whole page safely for inplace I/O 221db166fc2SGao Xiang * in the following scenario: 22247e4937aSGao Xiang * ________________________________________________________________ 22347e4937aSGao Xiang * | tail (partial) page | head (partial) page | 224db166fc2SGao Xiang * | (belongs to the next pcl) | (belongs to the current pcl) | 225db166fc2SGao Xiang * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________| 22647e4937aSGao Xiang */ 227db166fc2SGao Xiang Z_EROFS_PCLUSTER_HOOKED, 2280b964600SGao Xiang /* 229db166fc2SGao Xiang * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 2300b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 2310b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 232387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 2330b964600SGao Xiang */ 234db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 23547e4937aSGao Xiang /* 23647e4937aSGao Xiang * The current collection has been linked with the owned chain, and 23747e4937aSGao Xiang * could also be linked with the remaining collections, which means 23847e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 23947e4937aSGao Xiang * the current collection can safely use the whole page (since 24047e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 24147e4937aSGao Xiang * illustrated below: 24247e4937aSGao Xiang * ________________________________________________________________ 24347e4937aSGao Xiang * | tail (partial) page | head (partial) page | 24447e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 245db166fc2SGao Xiang * | PCLUSTER_FOLLOWED or | | 246db166fc2SGao Xiang * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________| 24747e4937aSGao Xiang * 24847e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 24947e4937aSGao Xiang */ 250db166fc2SGao Xiang Z_EROFS_PCLUSTER_FOLLOWED, 25147e4937aSGao Xiang }; 25247e4937aSGao Xiang 2535c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 2545c6dcc57SGao Xiang struct inode *const inode; 2555c6dcc57SGao Xiang struct erofs_map_blocks map; 25606a304cdSGao Xiang struct z_erofs_bvec_iter biter; 25747e4937aSGao Xiang 25806a304cdSGao Xiang struct page *candidate_bvpage; 25947e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 26047e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 261db166fc2SGao Xiang enum z_erofs_pclustermode mode; 26247e4937aSGao Xiang 2636ea5aad3SGao Xiang bool readahead; 26447e4937aSGao Xiang /* used for applying cache strategy on the fly */ 26547e4937aSGao Xiang bool backmost; 26647e4937aSGao Xiang erofs_off_t headoffset; 267ed722fbcSGao Xiang 268ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 269ed722fbcSGao Xiang unsigned int icur; 27047e4937aSGao Xiang }; 27147e4937aSGao Xiang 27247e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 2735c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 274db166fc2SGao Xiang .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 27547e4937aSGao Xiang 2761282dea3SGao Xiang static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 2771282dea3SGao Xiang { 2781282dea3SGao Xiang unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 2791282dea3SGao Xiang 2801282dea3SGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 2811282dea3SGao Xiang return false; 2821282dea3SGao Xiang 2831282dea3SGao Xiang if (fe->backmost) 2841282dea3SGao Xiang return true; 2851282dea3SGao Xiang 2861282dea3SGao Xiang if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 2871282dea3SGao Xiang fe->map.m_la < fe->headoffset) 2881282dea3SGao Xiang return true; 2891282dea3SGao Xiang 2901282dea3SGao Xiang return false; 2911282dea3SGao Xiang } 2921282dea3SGao Xiang 2936f39d1e1SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 294eaa9172aSGao Xiang struct page **pagepool) 29547e4937aSGao Xiang { 2966f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 2975c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 2981282dea3SGao Xiang bool shouldalloc = z_erofs_should_alloc_cache(fe); 29947e4937aSGao Xiang bool standalone = true; 3006f39d1e1SGao Xiang /* 3016f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 3026f39d1e1SGao Xiang * can be used if low memory otherwise. 3036f39d1e1SGao Xiang */ 3041825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 3051825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 306ed722fbcSGao Xiang unsigned int i; 30747e4937aSGao Xiang 308db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 30947e4937aSGao Xiang return; 31047e4937aSGao Xiang 311ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 31247e4937aSGao Xiang struct page *page; 313*b1ed220cSGao Xiang void *t; /* mark pages just found for debugging */ 3141825c8d7SGao Xiang struct page *newpage = NULL; 31547e4937aSGao Xiang 31647e4937aSGao Xiang /* the compressed page was loaded before */ 317ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 31847e4937aSGao Xiang continue; 31947e4937aSGao Xiang 320ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 32147e4937aSGao Xiang 32247e4937aSGao Xiang if (page) { 323*b1ed220cSGao Xiang t = (void *)((unsigned long)page | 1); 3240b964600SGao Xiang } else { 3250b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 3260b964600SGao Xiang standalone = false; 3271282dea3SGao Xiang if (!shouldalloc) 3281282dea3SGao Xiang continue; 3291282dea3SGao Xiang 3301282dea3SGao Xiang /* 3311282dea3SGao Xiang * try to use cached I/O if page allocation 3321282dea3SGao Xiang * succeeds or fallback to in-place I/O instead 3331282dea3SGao Xiang * to avoid any direct reclaim. 3341282dea3SGao Xiang */ 3351825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 3361825c8d7SGao Xiang if (!newpage) 33747e4937aSGao Xiang continue; 3381282dea3SGao Xiang set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 339*b1ed220cSGao Xiang t = (void *)((unsigned long)newpage | 1); 34047e4937aSGao Xiang } 34147e4937aSGao Xiang 342*b1ed220cSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 34347e4937aSGao Xiang continue; 34447e4937aSGao Xiang 345eaa9172aSGao Xiang if (page) 34647e4937aSGao Xiang put_page(page); 347eaa9172aSGao Xiang else if (newpage) 348eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 34947e4937aSGao Xiang } 35047e4937aSGao Xiang 3510b964600SGao Xiang /* 3520b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 3530b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 3540b964600SGao Xiang */ 3550b964600SGao Xiang if (standalone) 356db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 35747e4937aSGao Xiang } 35847e4937aSGao Xiang 35947e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 36047e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 36147e4937aSGao Xiang struct erofs_workgroup *grp) 36247e4937aSGao Xiang { 36347e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 36447e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 36547e4937aSGao Xiang int i; 36647e4937aSGao Xiang 367cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 36847e4937aSGao Xiang /* 36947e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 37047e4937aSGao Xiang * therefore no need to worry about available decompression users. 37147e4937aSGao Xiang */ 3729f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 373ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 37447e4937aSGao Xiang 37547e4937aSGao Xiang if (!page) 37647e4937aSGao Xiang continue; 37747e4937aSGao Xiang 37847e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 37947e4937aSGao Xiang if (!trylock_page(page)) 38047e4937aSGao Xiang return -EBUSY; 38147e4937aSGao Xiang 382f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 38347e4937aSGao Xiang continue; 38447e4937aSGao Xiang 38547e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 386ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 3876aaa7b06SGao Xiang detach_page_private(page); 38847e4937aSGao Xiang unlock_page(page); 38947e4937aSGao Xiang } 39047e4937aSGao Xiang return 0; 39147e4937aSGao Xiang } 39247e4937aSGao Xiang 393d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 39447e4937aSGao Xiang { 39547e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 396ed722fbcSGao Xiang int ret, i; 39747e4937aSGao Xiang 398ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 399ed722fbcSGao Xiang return 0; 40047e4937aSGao Xiang 401ed722fbcSGao Xiang ret = 0; 402cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 4039f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 404ed722fbcSGao Xiang if (pcl->compressed_bvecs[i].page == page) { 405ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 40647e4937aSGao Xiang ret = 1; 40747e4937aSGao Xiang break; 40847e4937aSGao Xiang } 40947e4937aSGao Xiang } 41047e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 4116aaa7b06SGao Xiang if (ret) 4126aaa7b06SGao Xiang detach_page_private(page); 41347e4937aSGao Xiang return ret; 41447e4937aSGao Xiang } 41547e4937aSGao Xiang 4165c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 417ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 41847e4937aSGao Xiang { 4195c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 42047e4937aSGao Xiang 421ed722fbcSGao Xiang while (fe->icur > 0) { 422ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 423ed722fbcSGao Xiang NULL, bvec->page)) { 424ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 42547e4937aSGao Xiang return true; 426ed722fbcSGao Xiang } 427ed722fbcSGao Xiang } 42847e4937aSGao Xiang return false; 42947e4937aSGao Xiang } 43047e4937aSGao Xiang 43187ca34a7SGao Xiang /* callers must be with pcluster lock held */ 4325c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 4335b220b20SGao Xiang struct z_erofs_bvec *bvec, bool exclusive) 43447e4937aSGao Xiang { 43547e4937aSGao Xiang int ret; 43647e4937aSGao Xiang 437db166fc2SGao Xiang if (exclusive) { 43806a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 439ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 44047e4937aSGao Xiang return 0; 44106a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 442db166fc2SGao Xiang if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 44306a304cdSGao Xiang !fe->candidate_bvpage) 44406a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 44506a304cdSGao Xiang } 44606a304cdSGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 44706a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 44806a304cdSGao Xiang return ret; 44947e4937aSGao Xiang } 45047e4937aSGao Xiang 4515c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 45247e4937aSGao Xiang { 4535c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 4545c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 45547e4937aSGao Xiang 456473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 457473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 458473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 45947e4937aSGao Xiang *owned_head = &pcl->next; 460473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 461db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 462473e15b0SGao Xiang return; 463473e15b0SGao Xiang } 464473e15b0SGao Xiang 46547e4937aSGao Xiang /* 466473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 467473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 46847e4937aSGao Xiang */ 469267f2492SGao Xiang if (*owned_head != &pcl->next && pcl != f->tailpcl && 470267f2492SGao Xiang cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 471473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 47247e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 473db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_HOOKED; 4745c6dcc57SGao Xiang f->tailpcl = NULL; 475473e15b0SGao Xiang return; 47647e4937aSGao Xiang } 477473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 478db166fc2SGao Xiang f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 47947e4937aSGao Xiang } 48047e4937aSGao Xiang 48183a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 48247e4937aSGao Xiang { 48383a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 484cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 48547e4937aSGao Xiang struct z_erofs_pcluster *pcl; 48664094a04SGao Xiang struct erofs_workgroup *grp; 48747e4937aSGao Xiang int err; 48847e4937aSGao Xiang 489c42c0ffeSChen Zhongjin if (!(map->m_flags & EROFS_MAP_ENCODED) || 490c42c0ffeSChen Zhongjin (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 4918f899262SGao Xiang DBG_BUGON(1); 4928f899262SGao Xiang return -EFSCORRUPTED; 4938f899262SGao Xiang } 4948f899262SGao Xiang 4959f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 496cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 497cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 4989f6cc76eSGao Xiang if (IS_ERR(pcl)) 4999f6cc76eSGao Xiang return PTR_ERR(pcl); 50047e4937aSGao Xiang 50164094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 5028f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 5032bfab9c0SGao Xiang pcl->length = 0; 5042bfab9c0SGao Xiang pcl->partial = true; 50547e4937aSGao Xiang 50647e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 5075c6dcc57SGao Xiang pcl->next = fe->owned_head; 50887ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 509db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 51047e4937aSGao Xiang 51147e4937aSGao Xiang /* 51247e4937aSGao Xiang * lock all primary followed works before visible to others 51347e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 51447e4937aSGao Xiang */ 51587ca34a7SGao Xiang mutex_init(&pcl->lock); 51687ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 51747e4937aSGao Xiang 518cecf864dSYue Hu if (ztailpacking) { 519cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 520cecf864dSYue Hu pcl->pageofs_in = erofs_blkoff(map->m_pa); 521cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 522cecf864dSYue Hu } else { 523cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 524cecf864dSYue Hu 52583a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 52664094a04SGao Xiang if (IS_ERR(grp)) { 52764094a04SGao Xiang err = PTR_ERR(grp); 52864094a04SGao Xiang goto err_out; 52964094a04SGao Xiang } 53064094a04SGao Xiang 53164094a04SGao Xiang if (grp != &pcl->obj) { 5325c6dcc57SGao Xiang fe->pcl = container_of(grp, 533cecf864dSYue Hu struct z_erofs_pcluster, obj); 53464094a04SGao Xiang err = -EEXIST; 53564094a04SGao Xiang goto err_out; 53647e4937aSGao Xiang } 537cecf864dSYue Hu } 53847e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 5395c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 5405c6dcc57SGao Xiang fe->tailpcl = pcl; 5415c6dcc57SGao Xiang fe->owned_head = &pcl->next; 5425c6dcc57SGao Xiang fe->pcl = pcl; 5439e579fc1SGao Xiang return 0; 54464094a04SGao Xiang 54564094a04SGao Xiang err_out: 54687ca34a7SGao Xiang mutex_unlock(&pcl->lock); 5479f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 54864094a04SGao Xiang return err; 54947e4937aSGao Xiang } 55047e4937aSGao Xiang 55183a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 55247e4937aSGao Xiang { 55383a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 5540d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 5559e579fc1SGao Xiang int ret; 55647e4937aSGao Xiang 55787ca34a7SGao Xiang DBG_BUGON(fe->pcl); 55847e4937aSGao Xiang 55987ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 5605c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 5615c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 56247e4937aSGao Xiang 5630d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 5640d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 5650d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 5660d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 56747e4937aSGao Xiang DBG_BUGON(1); 568cecf864dSYue Hu return -EFSCORRUPTED; 569cecf864dSYue Hu } 57047e4937aSGao Xiang 57164094a04SGao Xiang if (grp) { 5725c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 5730d823b42SGao Xiang ret = -EEXIST; 57464094a04SGao Xiang } else { 57583a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 57664094a04SGao Xiang } 57747e4937aSGao Xiang 5780d823b42SGao Xiang if (ret == -EEXIST) { 579267f2492SGao Xiang mutex_lock(&fe->pcl->lock); 580267f2492SGao Xiang /* used to check tail merging loop due to corrupted images */ 581267f2492SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 582267f2492SGao Xiang fe->tailpcl = fe->pcl; 583267f2492SGao Xiang 584267f2492SGao Xiang z_erofs_try_to_claim_pcluster(fe); 5850d823b42SGao Xiang } else if (ret) { 5860d823b42SGao Xiang return ret; 5870d823b42SGao Xiang } 58806a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 589387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 59081382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 591ed722fbcSGao Xiang fe->icur = z_erofs_pclusterpages(fe->pcl); 59247e4937aSGao Xiang return 0; 59347e4937aSGao Xiang } 59447e4937aSGao Xiang 59547e4937aSGao Xiang /* 59647e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 59747e4937aSGao Xiang * only after a RCU grace period. 59847e4937aSGao Xiang */ 59947e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 60047e4937aSGao Xiang { 60187ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 60287ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 60347e4937aSGao Xiang } 60447e4937aSGao Xiang 60547e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 60647e4937aSGao Xiang { 60747e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 60847e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 60947e4937aSGao Xiang 61087ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 61147e4937aSGao Xiang } 61247e4937aSGao Xiang 6135c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 61447e4937aSGao Xiang { 61587ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 61647e4937aSGao Xiang 61787ca34a7SGao Xiang if (!pcl) 61847e4937aSGao Xiang return false; 61947e4937aSGao Xiang 62006a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 62187ca34a7SGao Xiang mutex_unlock(&pcl->lock); 62247e4937aSGao Xiang 62306a304cdSGao Xiang if (fe->candidate_bvpage) { 62406a304cdSGao Xiang DBG_BUGON(z_erofs_is_shortlived_page(fe->candidate_bvpage)); 62506a304cdSGao Xiang fe->candidate_bvpage = NULL; 62606a304cdSGao Xiang } 62706a304cdSGao Xiang 62847e4937aSGao Xiang /* 62947e4937aSGao Xiang * if all pending pages are added, don't hold its reference 63047e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 63147e4937aSGao Xiang */ 632db166fc2SGao Xiang if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 63387ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 63447e4937aSGao Xiang 63587ca34a7SGao Xiang fe->pcl = NULL; 63647e4937aSGao Xiang return true; 63747e4937aSGao Xiang } 63847e4937aSGao Xiang 639b15b2e30SYue Hu static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 640b15b2e30SYue Hu struct page *page, unsigned int pageofs, 641b15b2e30SYue Hu unsigned int len) 642b15b2e30SYue Hu { 643b15b2e30SYue Hu struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 644b15b2e30SYue Hu struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 645b15b2e30SYue Hu u8 *src, *dst; 646b15b2e30SYue Hu unsigned int i, cnt; 647b15b2e30SYue Hu 648e5126de1SYue Hu if (!packed_inode) 649e5126de1SYue Hu return -EFSCORRUPTED; 650e5126de1SYue Hu 651b15b2e30SYue Hu pos += EROFS_I(inode)->z_fragmentoff; 652b15b2e30SYue Hu for (i = 0; i < len; i += cnt) { 653b15b2e30SYue Hu cnt = min_t(unsigned int, len - i, 654b15b2e30SYue Hu EROFS_BLKSIZ - erofs_blkoff(pos)); 655b15b2e30SYue Hu src = erofs_bread(&buf, packed_inode, 656b15b2e30SYue Hu erofs_blknr(pos), EROFS_KMAP); 657b15b2e30SYue Hu if (IS_ERR(src)) { 658b15b2e30SYue Hu erofs_put_metabuf(&buf); 659b15b2e30SYue Hu return PTR_ERR(src); 660b15b2e30SYue Hu } 661b15b2e30SYue Hu 662b15b2e30SYue Hu dst = kmap_local_page(page); 663b15b2e30SYue Hu memcpy(dst + pageofs + i, src + erofs_blkoff(pos), cnt); 664b15b2e30SYue Hu kunmap_local(dst); 665b15b2e30SYue Hu pos += cnt; 666b15b2e30SYue Hu } 667b15b2e30SYue Hu erofs_put_metabuf(&buf); 668b15b2e30SYue Hu return 0; 669b15b2e30SYue Hu } 670b15b2e30SYue Hu 67147e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 672eaa9172aSGao Xiang struct page *page, struct page **pagepool) 67347e4937aSGao Xiang { 67447e4937aSGao Xiang struct inode *const inode = fe->inode; 67547e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 67647e4937aSGao Xiang const loff_t offset = page_offset(page); 6775b220b20SGao Xiang bool tight = true, exclusive; 6782bfab9c0SGao Xiang unsigned int cur, end, spiltted; 67947e4937aSGao Xiang int err = 0; 68047e4937aSGao Xiang 68147e4937aSGao Xiang /* register locked file pages as online pages in pack */ 68247e4937aSGao Xiang z_erofs_onlinepage_init(page); 68347e4937aSGao Xiang 68447e4937aSGao Xiang spiltted = 0; 68547e4937aSGao Xiang end = PAGE_SIZE; 68647e4937aSGao Xiang repeat: 68747e4937aSGao Xiang cur = end - 1; 68847e4937aSGao Xiang 68939397a46SGao Xiang if (offset + cur < map->m_la || 69039397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 69139397a46SGao Xiang erofs_dbg("out-of-range map @ pos %llu", offset + cur); 69247e4937aSGao Xiang 6935c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 69447e4937aSGao Xiang fe->backmost = false; 69547e4937aSGao Xiang map->m_la = offset + cur; 69647e4937aSGao Xiang map->m_llen = 0; 69747e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 6988d8a09b0SGao Xiang if (err) 69967148551SGao Xiang goto out; 70039397a46SGao Xiang } else { 70139397a46SGao Xiang if (fe->pcl) 70239397a46SGao Xiang goto hitted; 70339397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 70439397a46SGao Xiang } 70547e4937aSGao Xiang 706b15b2e30SYue Hu if (!(map->m_flags & EROFS_MAP_MAPPED) || 707b15b2e30SYue Hu map->m_flags & EROFS_MAP_FRAGMENT) 70847e4937aSGao Xiang goto hitted; 70947e4937aSGao Xiang 71083a386c0SGao Xiang err = z_erofs_collector_begin(fe); 7118d8a09b0SGao Xiang if (err) 71267148551SGao Xiang goto out; 71347e4937aSGao Xiang 7145c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 71509c54379SGao Xiang void *mp; 716cecf864dSYue Hu 71709c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 71809c54379SGao Xiang erofs_blknr(map->m_pa), EROFS_NO_KMAP); 71909c54379SGao Xiang if (IS_ERR(mp)) { 72009c54379SGao Xiang err = PTR_ERR(mp); 721cecf864dSYue Hu erofs_err(inode->i_sb, 722cecf864dSYue Hu "failed to get inline page, err %d", err); 72367148551SGao Xiang goto out; 724cecf864dSYue Hu } 72509c54379SGao Xiang get_page(fe->map.buf.page); 726ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 727ed722fbcSGao Xiang fe->map.buf.page); 728db166fc2SGao Xiang fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 729cecf864dSYue Hu } else { 7306f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 7311282dea3SGao Xiang z_erofs_bind_cache(fe, pagepool); 732cecf864dSYue Hu } 73347e4937aSGao Xiang hitted: 734dc76ea8cSGao Xiang /* 735dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 736dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 737dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 738387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 739dc76ea8cSGao Xiang */ 740db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED && 741db166fc2SGao Xiang fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 742dc76ea8cSGao Xiang 74347e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 7448d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 74547e4937aSGao Xiang zero_user_segment(page, cur, end); 74647e4937aSGao Xiang goto next_part; 74747e4937aSGao Xiang } 748b15b2e30SYue Hu if (map->m_flags & EROFS_MAP_FRAGMENT) { 749b15b2e30SYue Hu unsigned int pageofs, skip, len; 750b15b2e30SYue Hu 751b15b2e30SYue Hu if (offset > map->m_la) { 752b15b2e30SYue Hu pageofs = 0; 753b15b2e30SYue Hu skip = offset - map->m_la; 754b15b2e30SYue Hu } else { 755b15b2e30SYue Hu pageofs = map->m_la & ~PAGE_MASK; 756b15b2e30SYue Hu skip = 0; 757b15b2e30SYue Hu } 758b15b2e30SYue Hu len = min_t(unsigned int, map->m_llen - skip, end - cur); 759b15b2e30SYue Hu err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 760b15b2e30SYue Hu if (err) 761b15b2e30SYue Hu goto out; 762b15b2e30SYue Hu ++spiltted; 763b15b2e30SYue Hu tight = false; 764b15b2e30SYue Hu goto next_part; 765b15b2e30SYue Hu } 76647e4937aSGao Xiang 7675b220b20SGao Xiang exclusive = (!cur && (!spiltted || tight)); 76847e4937aSGao Xiang if (cur) 769db166fc2SGao Xiang tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 77047e4937aSGao Xiang 77147e4937aSGao Xiang retry: 77206a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 77306a304cdSGao Xiang .page = page, 77406a304cdSGao Xiang .offset = offset - map->m_la, 77506a304cdSGao Xiang .end = end, 7765b220b20SGao Xiang }), exclusive); 77706a304cdSGao Xiang /* should allocate an additional short-lived page for bvset */ 77806a304cdSGao Xiang if (err == -EAGAIN && !fe->candidate_bvpage) { 77906a304cdSGao Xiang fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL); 78006a304cdSGao Xiang set_page_private(fe->candidate_bvpage, 78106a304cdSGao Xiang Z_EROFS_SHORTLIVED_PAGE); 78247e4937aSGao Xiang goto retry; 78347e4937aSGao Xiang } 78447e4937aSGao Xiang 78506a304cdSGao Xiang if (err) { 78606a304cdSGao Xiang DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage); 78767148551SGao Xiang goto out; 78806a304cdSGao Xiang } 78947e4937aSGao Xiang 79067148551SGao Xiang z_erofs_onlinepage_split(page); 79147e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 79247e4937aSGao Xiang ++spiltted; 793267f2492SGao Xiang if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 794267f2492SGao Xiang fe->pcl->multibases = true; 7952bfab9c0SGao Xiang if (fe->pcl->length < offset + end - map->m_la) { 7962bfab9c0SGao Xiang fe->pcl->length = offset + end - map->m_la; 7972bfab9c0SGao Xiang fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 7982bfab9c0SGao Xiang } 799e7933278SGao Xiang if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 800e7933278SGao Xiang !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 801e7933278SGao Xiang fe->pcl->length == map->m_llen) 802e7933278SGao Xiang fe->pcl->partial = false; 80347e4937aSGao Xiang next_part: 8042bfab9c0SGao Xiang /* shorten the remaining extent to update progress */ 80547e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 8062bfab9c0SGao Xiang map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 80747e4937aSGao Xiang 80847e4937aSGao Xiang end = cur; 80947e4937aSGao Xiang if (end > 0) 81047e4937aSGao Xiang goto repeat; 81147e4937aSGao Xiang 81247e4937aSGao Xiang out: 81367148551SGao Xiang if (err) 81467148551SGao Xiang z_erofs_page_mark_eio(page); 81547e4937aSGao Xiang z_erofs_onlinepage_endio(page); 81647e4937aSGao Xiang 8174f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 81847e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 81947e4937aSGao Xiang return err; 82047e4937aSGao Xiang } 82147e4937aSGao Xiang 82240452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 82340452ffcSHuang Jianan unsigned int readahead_pages) 82440452ffcSHuang Jianan { 825a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 82640452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 82740452ffcSHuang Jianan !readahead_pages) 82840452ffcSHuang Jianan return true; 82940452ffcSHuang Jianan 83040452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 83140452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 83240452ffcSHuang Jianan return true; 83340452ffcSHuang Jianan 83440452ffcSHuang Jianan return false; 83540452ffcSHuang Jianan } 83640452ffcSHuang Jianan 8376aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 8386aaa7b06SGao Xiang { 8396aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 8406aaa7b06SGao Xiang } 8416aaa7b06SGao Xiang 8424f05687fSGao Xiang struct z_erofs_decompress_backend { 8434f05687fSGao Xiang struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 8444f05687fSGao Xiang struct super_block *sb; 8454f05687fSGao Xiang struct z_erofs_pcluster *pcl; 8464f05687fSGao Xiang 8474f05687fSGao Xiang /* pages with the longest decompressed length for deduplication */ 8484f05687fSGao Xiang struct page **decompressed_pages; 8494f05687fSGao Xiang /* pages to keep the compressed data */ 8504f05687fSGao Xiang struct page **compressed_pages; 8514f05687fSGao Xiang 852267f2492SGao Xiang struct list_head decompressed_secondary_bvecs; 8534f05687fSGao Xiang struct page **pagepool; 8542bfab9c0SGao Xiang unsigned int onstack_used, nr_pages; 8554f05687fSGao Xiang }; 8564f05687fSGao Xiang 857267f2492SGao Xiang struct z_erofs_bvec_item { 858267f2492SGao Xiang struct z_erofs_bvec bvec; 859267f2492SGao Xiang struct list_head list; 860267f2492SGao Xiang }; 861267f2492SGao Xiang 862267f2492SGao Xiang static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 8633fe96ee0SGao Xiang struct z_erofs_bvec *bvec) 8643fe96ee0SGao Xiang { 865267f2492SGao Xiang struct z_erofs_bvec_item *item; 866267f2492SGao Xiang 867267f2492SGao Xiang if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 868267f2492SGao Xiang unsigned int pgnr; 8693fe96ee0SGao Xiang 870267f2492SGao Xiang pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 8712bfab9c0SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 87263bbb856SGao Xiang if (!be->decompressed_pages[pgnr]) { 8733fe96ee0SGao Xiang be->decompressed_pages[pgnr] = bvec->page; 874267f2492SGao Xiang return; 8753fe96ee0SGao Xiang } 87663bbb856SGao Xiang } 8773fe96ee0SGao Xiang 878267f2492SGao Xiang /* (cold path) one pcluster is requested multiple times */ 879267f2492SGao Xiang item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 880267f2492SGao Xiang item->bvec = *bvec; 881267f2492SGao Xiang list_add(&item->list, &be->decompressed_secondary_bvecs); 882267f2492SGao Xiang } 883267f2492SGao Xiang 884267f2492SGao Xiang static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 885267f2492SGao Xiang int err) 886267f2492SGao Xiang { 887267f2492SGao Xiang unsigned int off0 = be->pcl->pageofs_out; 888267f2492SGao Xiang struct list_head *p, *n; 889267f2492SGao Xiang 890267f2492SGao Xiang list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 891267f2492SGao Xiang struct z_erofs_bvec_item *bvi; 892267f2492SGao Xiang unsigned int end, cur; 893267f2492SGao Xiang void *dst, *src; 894267f2492SGao Xiang 895267f2492SGao Xiang bvi = container_of(p, struct z_erofs_bvec_item, list); 896267f2492SGao Xiang cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 897267f2492SGao Xiang end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 898267f2492SGao Xiang bvi->bvec.end); 899267f2492SGao Xiang dst = kmap_local_page(bvi->bvec.page); 900267f2492SGao Xiang while (cur < end) { 901267f2492SGao Xiang unsigned int pgnr, scur, len; 902267f2492SGao Xiang 903267f2492SGao Xiang pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 904267f2492SGao Xiang DBG_BUGON(pgnr >= be->nr_pages); 905267f2492SGao Xiang 906267f2492SGao Xiang scur = bvi->bvec.offset + cur - 907267f2492SGao Xiang ((pgnr << PAGE_SHIFT) - off0); 908267f2492SGao Xiang len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 909267f2492SGao Xiang if (!be->decompressed_pages[pgnr]) { 910267f2492SGao Xiang err = -EFSCORRUPTED; 911267f2492SGao Xiang cur += len; 912267f2492SGao Xiang continue; 913267f2492SGao Xiang } 914267f2492SGao Xiang src = kmap_local_page(be->decompressed_pages[pgnr]); 915267f2492SGao Xiang memcpy(dst + cur, src + scur, len); 916267f2492SGao Xiang kunmap_local(src); 917267f2492SGao Xiang cur += len; 918267f2492SGao Xiang } 919267f2492SGao Xiang kunmap_local(dst); 920267f2492SGao Xiang if (err) 921267f2492SGao Xiang z_erofs_page_mark_eio(bvi->bvec.page); 922267f2492SGao Xiang z_erofs_onlinepage_endio(bvi->bvec.page); 923267f2492SGao Xiang list_del(p); 924267f2492SGao Xiang kfree(bvi); 925267f2492SGao Xiang } 926267f2492SGao Xiang } 927267f2492SGao Xiang 928267f2492SGao Xiang static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 92942fec235SGao Xiang { 9304f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 93106a304cdSGao Xiang struct z_erofs_bvec_iter biter; 93206a304cdSGao Xiang struct page *old_bvpage; 933267f2492SGao Xiang int i; 93442fec235SGao Xiang 935387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 93642fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 93706a304cdSGao Xiang struct z_erofs_bvec bvec; 93842fec235SGao Xiang 93906a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 94042fec235SGao Xiang 94106a304cdSGao Xiang if (old_bvpage) 9424f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 94342fec235SGao Xiang 94406a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 945267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, &bvec); 94642fec235SGao Xiang } 94706a304cdSGao Xiang 94806a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 94906a304cdSGao Xiang if (old_bvpage) 9504f05687fSGao Xiang z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 95142fec235SGao Xiang } 95242fec235SGao Xiang 9534f05687fSGao Xiang static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 9544f05687fSGao Xiang bool *overlapped) 95567139e36SGao Xiang { 9564f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 95767139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 95867139e36SGao Xiang int i, err = 0; 95967139e36SGao Xiang 96067139e36SGao Xiang *overlapped = false; 96167139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 962ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 963ed722fbcSGao Xiang struct page *page = bvec->page; 96467139e36SGao Xiang 96567139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 96667139e36SGao Xiang if (!page) { 96767139e36SGao Xiang DBG_BUGON(1); 96867139e36SGao Xiang continue; 96967139e36SGao Xiang } 970fe3e5914SGao Xiang be->compressed_pages[i] = page; 97167139e36SGao Xiang 97267139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 97367139e36SGao Xiang if (!PageUptodate(page)) 97467139e36SGao Xiang err = -EIO; 97567139e36SGao Xiang continue; 97667139e36SGao Xiang } 97767139e36SGao Xiang 97867139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 97967139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 9804f05687fSGao Xiang if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 98167139e36SGao Xiang if (!PageUptodate(page)) 98267139e36SGao Xiang err = -EIO; 98367139e36SGao Xiang continue; 98467139e36SGao Xiang } 985267f2492SGao Xiang z_erofs_do_decompressed_bvec(be, bvec); 98667139e36SGao Xiang *overlapped = true; 98767139e36SGao Xiang } 98867139e36SGao Xiang } 98967139e36SGao Xiang 990fe3e5914SGao Xiang if (err) 9914f05687fSGao Xiang return err; 9924f05687fSGao Xiang return 0; 99367139e36SGao Xiang } 99467139e36SGao Xiang 9954f05687fSGao Xiang static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 9964f05687fSGao Xiang int err) 99747e4937aSGao Xiang { 9984f05687fSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 9994f05687fSGao Xiang struct z_erofs_pcluster *pcl = be->pcl; 1000cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 10012bfab9c0SGao Xiang unsigned int i, inputsize; 100267148551SGao Xiang int err2; 10032bfab9c0SGao Xiang struct page *page; 10042bfab9c0SGao Xiang bool overlapped; 100547e4937aSGao Xiang 100687ca34a7SGao Xiang mutex_lock(&pcl->lock); 10072bfab9c0SGao Xiang be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 100847e4937aSGao Xiang 1009fe3e5914SGao Xiang /* allocate (de)compressed page arrays if cannot be kept on stack */ 1010fe3e5914SGao Xiang be->decompressed_pages = NULL; 1011fe3e5914SGao Xiang be->compressed_pages = NULL; 1012fe3e5914SGao Xiang be->onstack_used = 0; 10132bfab9c0SGao Xiang if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 10144f05687fSGao Xiang be->decompressed_pages = be->onstack_pages; 10152bfab9c0SGao Xiang be->onstack_used = be->nr_pages; 10164f05687fSGao Xiang memset(be->decompressed_pages, 0, 10172bfab9c0SGao Xiang sizeof(struct page *) * be->nr_pages); 1018fe3e5914SGao Xiang } 1019fe3e5914SGao Xiang 1020fe3e5914SGao Xiang if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1021fe3e5914SGao Xiang be->compressed_pages = be->onstack_pages + be->onstack_used; 1022fe3e5914SGao Xiang 1023fe3e5914SGao Xiang if (!be->decompressed_pages) 10244f05687fSGao Xiang be->decompressed_pages = 102512724ba3SGao Xiang kcalloc(be->nr_pages, sizeof(struct page *), 1026e7368187SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 1027fe3e5914SGao Xiang if (!be->compressed_pages) 1028fe3e5914SGao Xiang be->compressed_pages = 102912724ba3SGao Xiang kcalloc(pclusterpages, sizeof(struct page *), 1030fe3e5914SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 103147e4937aSGao Xiang 1032267f2492SGao Xiang z_erofs_parse_out_bvecs(be); 10334f05687fSGao Xiang err2 = z_erofs_parse_in_bvecs(be, &overlapped); 10344f05687fSGao Xiang if (err2) 10354f05687fSGao Xiang err = err2; 10368d8a09b0SGao Xiang if (err) 103747e4937aSGao Xiang goto out; 103847e4937aSGao Xiang 1039cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1040cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1041cecf864dSYue Hu else 1042cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1043cecf864dSYue Hu 104447e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 10454f05687fSGao Xiang .sb = be->sb, 10464f05687fSGao Xiang .in = be->compressed_pages, 10474f05687fSGao Xiang .out = be->decompressed_pages, 1048cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 104987ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 10509f6cc76eSGao Xiang .inputsize = inputsize, 10512bfab9c0SGao Xiang .outputsize = pcl->length, 105247e4937aSGao Xiang .alg = pcl->algorithmformat, 105347e4937aSGao Xiang .inplace_io = overlapped, 10542bfab9c0SGao Xiang .partial_decoding = pcl->partial, 1055267f2492SGao Xiang .fillgaps = pcl->multibases, 10564f05687fSGao Xiang }, be->pagepool); 105747e4937aSGao Xiang 105847e4937aSGao Xiang out: 1059cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1060cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1061ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1062ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1063cecf864dSYue Hu put_page(page); 1064cecf864dSYue Hu } else { 1065cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1066ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 106747e4937aSGao Xiang 106847e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 106947e4937aSGao Xiang continue; 107047e4937aSGao Xiang 10716aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10724f05687fSGao Xiang (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1073ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 107447e4937aSGao Xiang } 1075cecf864dSYue Hu } 1076fe3e5914SGao Xiang if (be->compressed_pages < be->onstack_pages || 1077fe3e5914SGao Xiang be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 107812724ba3SGao Xiang kfree(be->compressed_pages); 1079267f2492SGao Xiang z_erofs_fill_other_copies(be, err); 108047e4937aSGao Xiang 10812bfab9c0SGao Xiang for (i = 0; i < be->nr_pages; ++i) { 10824f05687fSGao Xiang page = be->decompressed_pages[i]; 108347e4937aSGao Xiang if (!page) 108447e4937aSGao Xiang continue; 108547e4937aSGao Xiang 10866aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 108747e4937aSGao Xiang 10886aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10894f05687fSGao Xiang if (z_erofs_put_shortlivedpage(be->pagepool, page)) 109047e4937aSGao Xiang continue; 109167148551SGao Xiang if (err) 109267148551SGao Xiang z_erofs_page_mark_eio(page); 109347e4937aSGao Xiang z_erofs_onlinepage_endio(page); 109447e4937aSGao Xiang } 109547e4937aSGao Xiang 10964f05687fSGao Xiang if (be->decompressed_pages != be->onstack_pages) 109712724ba3SGao Xiang kfree(be->decompressed_pages); 109847e4937aSGao Xiang 10992bfab9c0SGao Xiang pcl->length = 0; 11002bfab9c0SGao Xiang pcl->partial = true; 1101267f2492SGao Xiang pcl->multibases = false; 110206a304cdSGao Xiang pcl->bvset.nextpage = NULL; 110387ca34a7SGao Xiang pcl->vcnt = 0; 110447e4937aSGao Xiang 110587ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 110647e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 110787ca34a7SGao Xiang mutex_unlock(&pcl->lock); 110847e4937aSGao Xiang return err; 110947e4937aSGao Xiang } 111047e4937aSGao Xiang 11110c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1112eaa9172aSGao Xiang struct page **pagepool) 111347e4937aSGao Xiang { 11144f05687fSGao Xiang struct z_erofs_decompress_backend be = { 11154f05687fSGao Xiang .sb = io->sb, 11164f05687fSGao Xiang .pagepool = pagepool, 1117267f2492SGao Xiang .decompressed_secondary_bvecs = 1118267f2492SGao Xiang LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 11194f05687fSGao Xiang }; 112047e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 112147e4937aSGao Xiang 112247e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 11234f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 112447e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 11254f05687fSGao Xiang /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */ 112647e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 112747e4937aSGao Xiang 11284f05687fSGao Xiang be.pcl = container_of(owned, struct z_erofs_pcluster, next); 11294f05687fSGao Xiang owned = READ_ONCE(be.pcl->next); 113047e4937aSGao Xiang 11314f05687fSGao Xiang z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 11324f05687fSGao Xiang erofs_workgroup_put(&be.pcl->obj); 113347e4937aSGao Xiang } 113447e4937aSGao Xiang } 113547e4937aSGao Xiang 11360c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 113747e4937aSGao Xiang { 1138a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1139a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1140eaa9172aSGao Xiang struct page *pagepool = NULL; 114147e4937aSGao Xiang 1142a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 11430c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 114447e4937aSGao Xiang 1145eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1146a4b1fab1SGao Xiang kvfree(bgq); 114747e4937aSGao Xiang } 114847e4937aSGao Xiang 11497865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1150cdba5506SGao Xiang int bios) 11517865827cSGao Xiang { 11527865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 11537865827cSGao Xiang 11547865827cSGao Xiang /* wake up the caller thread for sync decompression */ 1155cdba5506SGao Xiang if (io->sync) { 11567865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 115760b30050SHongyu Jin complete(&io->u.done); 11587865827cSGao Xiang return; 11597865827cSGao Xiang } 11607865827cSGao Xiang 11617865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 11627865827cSGao Xiang return; 11637865827cSGao Xiang /* Use workqueue and sync decompression for atomic contexts only */ 11647865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 11657865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 11667865827cSGao Xiang /* enable sync decompression for readahead */ 11677865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 11687865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 11697865827cSGao Xiang return; 11707865827cSGao Xiang } 11717865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 11727865827cSGao Xiang } 11737865827cSGao Xiang 117447e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 117547e4937aSGao Xiang unsigned int nr, 1176eaa9172aSGao Xiang struct page **pagepool, 11779f2731d6SGao Xiang struct address_space *mc) 117847e4937aSGao Xiang { 117947e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 11809f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 118147e4937aSGao Xiang bool tocache = false; 118247e4937aSGao Xiang 118347e4937aSGao Xiang struct address_space *mapping; 118447e4937aSGao Xiang struct page *oldpage, *page; 118547e4937aSGao Xiang int justfound; 118647e4937aSGao Xiang 118747e4937aSGao Xiang repeat: 1188ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 118947e4937aSGao Xiang oldpage = page; 119047e4937aSGao Xiang 119147e4937aSGao Xiang if (!page) 119247e4937aSGao Xiang goto out_allocpage; 119347e4937aSGao Xiang 1194*b1ed220cSGao Xiang justfound = (unsigned long)page & 1UL; 1195*b1ed220cSGao Xiang page = (struct page *)((unsigned long)page & ~1UL); 119647e4937aSGao Xiang 11971825c8d7SGao Xiang /* 11981825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 11991825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 12001825c8d7SGao Xiang */ 12011825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1202ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 12031825c8d7SGao Xiang set_page_private(page, 0); 12041825c8d7SGao Xiang tocache = true; 12051825c8d7SGao Xiang goto out_tocache; 12061825c8d7SGao Xiang } 120747e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 120847e4937aSGao Xiang 120947e4937aSGao Xiang /* 12106aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 121147e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 121247e4937aSGao Xiang */ 121347e4937aSGao Xiang if (mapping && mapping != mc) 121447e4937aSGao Xiang /* ought to be unmanaged pages */ 121547e4937aSGao Xiang goto out; 121647e4937aSGao Xiang 12176aaa7b06SGao Xiang /* directly return for shortlived page as well */ 12186aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 12196aaa7b06SGao Xiang goto out; 12206aaa7b06SGao Xiang 122147e4937aSGao Xiang lock_page(page); 122247e4937aSGao Xiang 122347e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 122447e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 122547e4937aSGao Xiang 122647e4937aSGao Xiang /* the page is still in manage cache */ 122747e4937aSGao Xiang if (page->mapping == mc) { 1228ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 122947e4937aSGao Xiang 123047e4937aSGao Xiang if (!PagePrivate(page)) { 123147e4937aSGao Xiang /* 123247e4937aSGao Xiang * impossible to be !PagePrivate(page) for 123347e4937aSGao Xiang * the current restriction as well if 1234ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 123547e4937aSGao Xiang */ 123647e4937aSGao Xiang DBG_BUGON(!justfound); 123747e4937aSGao Xiang 123847e4937aSGao Xiang justfound = 0; 123947e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 124047e4937aSGao Xiang SetPagePrivate(page); 124147e4937aSGao Xiang } 124247e4937aSGao Xiang 124347e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 124447e4937aSGao Xiang if (PageUptodate(page)) { 124547e4937aSGao Xiang unlock_page(page); 124647e4937aSGao Xiang page = NULL; 124747e4937aSGao Xiang } 124847e4937aSGao Xiang goto out; 124947e4937aSGao Xiang } 125047e4937aSGao Xiang 125147e4937aSGao Xiang /* 125247e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 125347e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 125447e4937aSGao Xiang */ 125547e4937aSGao Xiang DBG_BUGON(page->mapping); 125647e4937aSGao Xiang DBG_BUGON(!justfound); 125747e4937aSGao Xiang 125847e4937aSGao Xiang tocache = true; 125947e4937aSGao Xiang unlock_page(page); 126047e4937aSGao Xiang put_page(page); 126147e4937aSGao Xiang out_allocpage: 12625ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1263ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1264ed722fbcSGao Xiang oldpage, page)) { 1265eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 12665ddcee1fSGao Xiang cond_resched(); 12675ddcee1fSGao Xiang goto repeat; 12685ddcee1fSGao Xiang } 12691825c8d7SGao Xiang out_tocache: 1270bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1271bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1272bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1273bf225074SGao Xiang goto out; 1274a30573b3SGao Xiang } 1275bf225074SGao Xiang attach_page_private(page, pcl); 1276bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1277bf225074SGao Xiang put_page(page); 1278bf225074SGao Xiang 127947e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 128047e4937aSGao Xiang return page; 128147e4937aSGao Xiang } 128247e4937aSGao Xiang 1283cdba5506SGao Xiang static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1284a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 128547e4937aSGao Xiang { 1286a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 128747e4937aSGao Xiang 1288a4b1fab1SGao Xiang if (fg && !*fg) { 1289a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1290a4b1fab1SGao Xiang if (!q) { 1291a4b1fab1SGao Xiang *fg = true; 1292a4b1fab1SGao Xiang goto fg_out; 129347e4937aSGao Xiang } 12940c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1295a4b1fab1SGao Xiang } else { 1296a4b1fab1SGao Xiang fg_out: 1297a4b1fab1SGao Xiang q = fgq; 129860b30050SHongyu Jin init_completion(&fgq->u.done); 1299a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 130067148551SGao Xiang q->eio = false; 1301cdba5506SGao Xiang q->sync = true; 1302a4b1fab1SGao Xiang } 1303a4b1fab1SGao Xiang q->sb = sb; 1304a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1305a4b1fab1SGao Xiang return q; 130647e4937aSGao Xiang } 130747e4937aSGao Xiang 130847e4937aSGao Xiang /* define decompression jobqueue types */ 130947e4937aSGao Xiang enum { 131047e4937aSGao Xiang JQ_BYPASS, 131147e4937aSGao Xiang JQ_SUBMIT, 131247e4937aSGao Xiang NR_JOBQUEUES, 131347e4937aSGao Xiang }; 131447e4937aSGao Xiang 131547e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 131647e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 131747e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 131847e4937aSGao Xiang { 131947e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 132047e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 132147e4937aSGao Xiang 132247e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 132347e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 132447e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 132547e4937aSGao Xiang 132647e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 132747e4937aSGao Xiang 132847e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 132947e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 133047e4937aSGao Xiang 133147e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 133247e4937aSGao Xiang } 133347e4937aSGao Xiang 13347865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 13357865827cSGao Xiang { 1336cdba5506SGao Xiang struct z_erofs_decompressqueue *q = bio->bi_private; 13377865827cSGao Xiang blk_status_t err = bio->bi_status; 13387865827cSGao Xiang struct bio_vec *bvec; 13397865827cSGao Xiang struct bvec_iter_all iter_all; 13407865827cSGao Xiang 13417865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 13427865827cSGao Xiang struct page *page = bvec->bv_page; 13437865827cSGao Xiang 13447865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 13457865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 13467865827cSGao Xiang 13477865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 13487865827cSGao Xiang if (!err) 13497865827cSGao Xiang SetPageUptodate(page); 13507865827cSGao Xiang unlock_page(page); 13517865827cSGao Xiang } 13527865827cSGao Xiang } 135367148551SGao Xiang if (err) 135467148551SGao Xiang q->eio = true; 1355cdba5506SGao Xiang z_erofs_decompress_kickoff(q, -1); 13567865827cSGao Xiang bio_put(bio); 13577865827cSGao Xiang } 13587865827cSGao Xiang 135983a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1360eaa9172aSGao Xiang struct page **pagepool, 1361a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1362a4b1fab1SGao Xiang bool *force_fg) 136347e4937aSGao Xiang { 136483a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 136583a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 136647e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1367a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 13685c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1369dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 13703f649ab7SKees Cook pgoff_t last_index; 1371dfeab2e9SGao Xiang struct block_device *last_bdev; 13721e4a2955SGao Xiang unsigned int nr_bios = 0; 13731e4a2955SGao Xiang struct bio *bio = NULL; 137482e60d00SJohannes Weiner unsigned long pflags; 137582e60d00SJohannes Weiner int memstall = 0; 137647e4937aSGao Xiang 1377cdba5506SGao Xiang /* 1378cdba5506SGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 1379cdba5506SGao Xiang * no need to read from device for all pclusters in this queue. 1380cdba5506SGao Xiang */ 1381cdba5506SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1382cdba5506SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1383cdba5506SGao Xiang 1384a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1385a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 138647e4937aSGao Xiang 138747e4937aSGao Xiang /* by default, all need io submission */ 138847e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 138947e4937aSGao Xiang 139047e4937aSGao Xiang do { 1391dfeab2e9SGao Xiang struct erofs_map_dev mdev; 139247e4937aSGao Xiang struct z_erofs_pcluster *pcl; 13931e4a2955SGao Xiang pgoff_t cur, end; 13941e4a2955SGao Xiang unsigned int i = 0; 13951e4a2955SGao Xiang bool bypass = true; 139647e4937aSGao Xiang 139747e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 139847e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 139947e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 140047e4937aSGao Xiang 140147e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 140247e4937aSGao Xiang 1403cecf864dSYue Hu /* close the main owned chain at first */ 1404cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1405cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1406cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1407cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1408cecf864dSYue Hu continue; 1409cecf864dSYue Hu } 1410cecf864dSYue Hu 1411dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1412dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1413dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1414dfeab2e9SGao Xiang }; 1415dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1416dfeab2e9SGao Xiang 1417dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 14189f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 141947e4937aSGao Xiang 14201e4a2955SGao Xiang do { 14211e4a2955SGao Xiang struct page *page; 142247e4937aSGao Xiang 14231e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 142483a386c0SGao Xiang mc); 14251e4a2955SGao Xiang if (!page) 14261e4a2955SGao Xiang continue; 142747e4937aSGao Xiang 1428dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1429dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 143047e4937aSGao Xiang submit_bio_retry: 143194e4e153SGao Xiang submit_bio(bio); 143282e60d00SJohannes Weiner if (memstall) { 143382e60d00SJohannes Weiner psi_memstall_leave(&pflags); 143482e60d00SJohannes Weiner memstall = 0; 143582e60d00SJohannes Weiner } 143647e4937aSGao Xiang bio = NULL; 143747e4937aSGao Xiang } 143847e4937aSGao Xiang 143982e60d00SJohannes Weiner if (unlikely(PageWorkingset(page)) && !memstall) { 144099486c51SChristoph Hellwig psi_memstall_enter(&pflags); 144182e60d00SJohannes Weiner memstall = 1; 144282e60d00SJohannes Weiner } 144399486c51SChristoph Hellwig 144447e4937aSGao Xiang if (!bio) { 144507888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 144607888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 14470c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1448dfeab2e9SGao Xiang 1449dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 14501e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1451a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1452cdba5506SGao Xiang bio->bi_private = q[JQ_SUBMIT]; 14536ea5aad3SGao Xiang if (f->readahead) 14546ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 145547e4937aSGao Xiang ++nr_bios; 145647e4937aSGao Xiang } 145747e4937aSGao Xiang 14586c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 145947e4937aSGao Xiang goto submit_bio_retry; 146047e4937aSGao Xiang 14611e4a2955SGao Xiang last_index = cur; 14621e4a2955SGao Xiang bypass = false; 14631e4a2955SGao Xiang } while (++cur < end); 146447e4937aSGao Xiang 14651e4a2955SGao Xiang if (!bypass) 146647e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 146747e4937aSGao Xiang else 146847e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 146947e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 147047e4937aSGao Xiang 147199486c51SChristoph Hellwig if (bio) { 147294e4e153SGao Xiang submit_bio(bio); 147382e60d00SJohannes Weiner if (memstall) 147482e60d00SJohannes Weiner psi_memstall_leave(&pflags); 147599486c51SChristoph Hellwig } 147647e4937aSGao Xiang 1477587a67b7SGao Xiang /* 1478587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1479587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1480587a67b7SGao Xiang */ 1481587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1482587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 14831e4a2955SGao Xiang return; 1484587a67b7SGao Xiang } 1485cdba5506SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 148647e4937aSGao Xiang } 148747e4937aSGao Xiang 148883a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1489eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 149047e4937aSGao Xiang { 1491a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 149247e4937aSGao Xiang 14935c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 149447e4937aSGao Xiang return; 149583a386c0SGao Xiang z_erofs_submit_queue(f, pagepool, io, &force_fg); 149647e4937aSGao Xiang 14970c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 14980c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 149947e4937aSGao Xiang 150047e4937aSGao Xiang if (!force_fg) 150147e4937aSGao Xiang return; 150247e4937aSGao Xiang 150347e4937aSGao Xiang /* wait until all bios are completed */ 150460b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 150547e4937aSGao Xiang 15060c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 15070c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 150847e4937aSGao Xiang } 150947e4937aSGao Xiang 151038629291SGao Xiang /* 151138629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 151238629291SGao Xiang * approximate readmore strategies as a start. 151338629291SGao Xiang */ 151438629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 151538629291SGao Xiang struct readahead_control *rac, 151638629291SGao Xiang erofs_off_t end, 1517eaa9172aSGao Xiang struct page **pagepool, 151838629291SGao Xiang bool backmost) 151938629291SGao Xiang { 152038629291SGao Xiang struct inode *inode = f->inode; 152138629291SGao Xiang struct erofs_map_blocks *map = &f->map; 152238629291SGao Xiang erofs_off_t cur; 152338629291SGao Xiang int err; 152438629291SGao Xiang 152538629291SGao Xiang if (backmost) { 152638629291SGao Xiang map->m_la = end; 1527622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1528622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 152938629291SGao Xiang if (err) 153038629291SGao Xiang return; 153138629291SGao Xiang 153238629291SGao Xiang /* expend ra for the trailing edge if readahead */ 153338629291SGao Xiang if (rac) { 153438629291SGao Xiang loff_t newstart = readahead_pos(rac); 153538629291SGao Xiang 153638629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 153738629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 153838629291SGao Xiang return; 153938629291SGao Xiang } 154038629291SGao Xiang end = round_up(end, PAGE_SIZE); 154138629291SGao Xiang } else { 154238629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 154338629291SGao Xiang 154438629291SGao Xiang if (!map->m_llen) 154538629291SGao Xiang return; 154638629291SGao Xiang } 154738629291SGao Xiang 154838629291SGao Xiang cur = map->m_la + map->m_llen - 1; 154938629291SGao Xiang while (cur >= end) { 155038629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 155138629291SGao Xiang struct page *page; 155238629291SGao Xiang 155338629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1554aa793b46SGao Xiang if (page) { 155538629291SGao Xiang if (PageUptodate(page)) { 155638629291SGao Xiang unlock_page(page); 1557aa793b46SGao Xiang } else { 155838629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 155938629291SGao Xiang if (err) 156038629291SGao Xiang erofs_err(inode->i_sb, 156138629291SGao Xiang "readmore error at page %lu @ nid %llu", 156238629291SGao Xiang index, EROFS_I(inode)->nid); 1563aa793b46SGao Xiang } 156438629291SGao Xiang put_page(page); 1565aa793b46SGao Xiang } 1566aa793b46SGao Xiang 156738629291SGao Xiang if (cur < PAGE_SIZE) 156838629291SGao Xiang break; 156938629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 157038629291SGao Xiang } 157138629291SGao Xiang } 157238629291SGao Xiang 1573a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 157447e4937aSGao Xiang { 1575a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 157647e4937aSGao Xiang struct inode *const inode = page->mapping->host; 157740452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 157847e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1579eaa9172aSGao Xiang struct page *pagepool = NULL; 158047e4937aSGao Xiang int err; 158147e4937aSGao Xiang 158247e4937aSGao Xiang trace_erofs_readpage(page, false); 158347e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 158447e4937aSGao Xiang 158538629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 158638629291SGao Xiang &pagepool, true); 15871825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 158838629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 158938629291SGao Xiang 15905c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 159147e4937aSGao Xiang 159247e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 159383a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 159440452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 159547e4937aSGao Xiang 159647e4937aSGao Xiang if (err) 15974f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 159847e4937aSGao Xiang 159909c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1600eaa9172aSGao Xiang erofs_release_pages(&pagepool); 160147e4937aSGao Xiang return err; 160247e4937aSGao Xiang } 160347e4937aSGao Xiang 16040615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 160547e4937aSGao Xiang { 16060615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 160747e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 160847e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1609eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 161038629291SGao Xiang unsigned int nr_pages; 161147e4937aSGao Xiang 16126ea5aad3SGao Xiang f.readahead = true; 16130615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 161447e4937aSGao Xiang 161538629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 161638629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 161738629291SGao Xiang nr_pages = readahead_count(rac); 161838629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 161938629291SGao Xiang 16200615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 162147e4937aSGao Xiang set_page_private(page, (unsigned long)head); 162247e4937aSGao Xiang head = page; 162347e4937aSGao Xiang } 162447e4937aSGao Xiang 162547e4937aSGao Xiang while (head) { 162647e4937aSGao Xiang struct page *page = head; 162747e4937aSGao Xiang int err; 162847e4937aSGao Xiang 162947e4937aSGao Xiang /* traversal in reverse order */ 163047e4937aSGao Xiang head = (void *)page_private(page); 163147e4937aSGao Xiang 16321825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1633a5876e24SGao Xiang if (err) 16344f761fa2SGao Xiang erofs_err(inode->i_sb, 16354f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 16364f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 163747e4937aSGao Xiang put_page(page); 163847e4937aSGao Xiang } 163938629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 16405c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 164147e4937aSGao Xiang 164283a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 164340452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 164409c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1645eaa9172aSGao Xiang erofs_release_pages(&pagepool); 164647e4937aSGao Xiang } 164747e4937aSGao Xiang 16480c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1649a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 16500615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 165147e4937aSGao Xiang }; 1652