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> 1047e4937aSGao Xiang 1147e4937aSGao Xiang #include <trace/events/erofs.h> 1247e4937aSGao Xiang 1347e4937aSGao Xiang /* 149f6cc76eSGao Xiang * since pclustersize is variable for big pcluster feature, introduce slab 159f6cc76eSGao Xiang * pools implementation for different pcluster sizes. 169f6cc76eSGao Xiang */ 179f6cc76eSGao Xiang struct z_erofs_pcluster_slab { 189f6cc76eSGao Xiang struct kmem_cache *slab; 199f6cc76eSGao Xiang unsigned int maxpages; 209f6cc76eSGao Xiang char name[48]; 219f6cc76eSGao Xiang }; 229f6cc76eSGao Xiang 239f6cc76eSGao Xiang #define _PCLP(n) { .maxpages = n } 249f6cc76eSGao Xiang 259f6cc76eSGao Xiang static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 269f6cc76eSGao Xiang _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 279f6cc76eSGao Xiang _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 289f6cc76eSGao Xiang }; 299f6cc76eSGao Xiang 30*387bab87SGao Xiang /* (obsoleted) page type for online pages */ 31*387bab87SGao Xiang enum z_erofs_page_type { 32*387bab87SGao Xiang /* including Z_EROFS_VLE_PAGE_TAIL_EXCLUSIVE */ 33*387bab87SGao Xiang Z_EROFS_PAGE_TYPE_EXCLUSIVE, 34*387bab87SGao Xiang 35*387bab87SGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED, 36*387bab87SGao Xiang 37*387bab87SGao Xiang Z_EROFS_VLE_PAGE_TYPE_HEAD, 38*387bab87SGao Xiang Z_EROFS_VLE_PAGE_TYPE_MAX 39*387bab87SGao Xiang }; 40*387bab87SGao Xiang 4106a304cdSGao Xiang struct z_erofs_bvec_iter { 4206a304cdSGao Xiang struct page *bvpage; 4306a304cdSGao Xiang struct z_erofs_bvset *bvset; 4406a304cdSGao Xiang unsigned int nr, cur; 4506a304cdSGao Xiang }; 4606a304cdSGao Xiang 4706a304cdSGao Xiang static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 4806a304cdSGao Xiang { 4906a304cdSGao Xiang if (iter->bvpage) 5006a304cdSGao Xiang kunmap_local(iter->bvset); 5106a304cdSGao Xiang return iter->bvpage; 5206a304cdSGao Xiang } 5306a304cdSGao Xiang 5406a304cdSGao Xiang static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 5506a304cdSGao Xiang { 5606a304cdSGao Xiang unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 5706a304cdSGao Xiang /* have to access nextpage in advance, otherwise it will be unmapped */ 5806a304cdSGao Xiang struct page *nextpage = iter->bvset->nextpage; 5906a304cdSGao Xiang struct page *oldpage; 6006a304cdSGao Xiang 6106a304cdSGao Xiang DBG_BUGON(!nextpage); 6206a304cdSGao Xiang oldpage = z_erofs_bvec_iter_end(iter); 6306a304cdSGao Xiang iter->bvpage = nextpage; 6406a304cdSGao Xiang iter->bvset = kmap_local_page(nextpage); 6506a304cdSGao Xiang iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 6606a304cdSGao Xiang iter->cur = 0; 6706a304cdSGao Xiang return oldpage; 6806a304cdSGao Xiang } 6906a304cdSGao Xiang 7006a304cdSGao Xiang static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 7106a304cdSGao Xiang struct z_erofs_bvset_inline *bvset, 7206a304cdSGao Xiang unsigned int bootstrap_nr, 7306a304cdSGao Xiang unsigned int cur) 7406a304cdSGao Xiang { 7506a304cdSGao Xiang *iter = (struct z_erofs_bvec_iter) { 7606a304cdSGao Xiang .nr = bootstrap_nr, 7706a304cdSGao Xiang .bvset = (struct z_erofs_bvset *)bvset, 7806a304cdSGao Xiang }; 7906a304cdSGao Xiang 8006a304cdSGao Xiang while (cur > iter->nr) { 8106a304cdSGao Xiang cur -= iter->nr; 8206a304cdSGao Xiang z_erofs_bvset_flip(iter); 8306a304cdSGao Xiang } 8406a304cdSGao Xiang iter->cur = cur; 8506a304cdSGao Xiang } 8606a304cdSGao Xiang 8706a304cdSGao Xiang static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 8806a304cdSGao Xiang struct z_erofs_bvec *bvec, 8906a304cdSGao Xiang struct page **candidate_bvpage) 9006a304cdSGao Xiang { 9106a304cdSGao Xiang if (iter->cur == iter->nr) { 9206a304cdSGao Xiang if (!*candidate_bvpage) 9306a304cdSGao Xiang return -EAGAIN; 9406a304cdSGao Xiang 9506a304cdSGao Xiang DBG_BUGON(iter->bvset->nextpage); 9606a304cdSGao Xiang iter->bvset->nextpage = *candidate_bvpage; 9706a304cdSGao Xiang z_erofs_bvset_flip(iter); 9806a304cdSGao Xiang 9906a304cdSGao Xiang iter->bvset->nextpage = NULL; 10006a304cdSGao Xiang *candidate_bvpage = NULL; 10106a304cdSGao Xiang } 10206a304cdSGao Xiang iter->bvset->bvec[iter->cur++] = *bvec; 10306a304cdSGao Xiang return 0; 10406a304cdSGao Xiang } 10506a304cdSGao Xiang 10606a304cdSGao Xiang static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 10706a304cdSGao Xiang struct z_erofs_bvec *bvec, 10806a304cdSGao Xiang struct page **old_bvpage) 10906a304cdSGao Xiang { 11006a304cdSGao Xiang if (iter->cur == iter->nr) 11106a304cdSGao Xiang *old_bvpage = z_erofs_bvset_flip(iter); 11206a304cdSGao Xiang else 11306a304cdSGao Xiang *old_bvpage = NULL; 11406a304cdSGao Xiang *bvec = iter->bvset->bvec[iter->cur++]; 11506a304cdSGao Xiang } 11606a304cdSGao Xiang 1179f6cc76eSGao Xiang static void z_erofs_destroy_pcluster_pool(void) 1189f6cc76eSGao Xiang { 1199f6cc76eSGao Xiang int i; 1209f6cc76eSGao Xiang 1219f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 1229f6cc76eSGao Xiang if (!pcluster_pool[i].slab) 1239f6cc76eSGao Xiang continue; 1249f6cc76eSGao Xiang kmem_cache_destroy(pcluster_pool[i].slab); 1259f6cc76eSGao Xiang pcluster_pool[i].slab = NULL; 1269f6cc76eSGao Xiang } 1279f6cc76eSGao Xiang } 1289f6cc76eSGao Xiang 1299f6cc76eSGao Xiang static int z_erofs_create_pcluster_pool(void) 1309f6cc76eSGao Xiang { 1319f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs; 1329f6cc76eSGao Xiang struct z_erofs_pcluster *a; 1339f6cc76eSGao Xiang unsigned int size; 1349f6cc76eSGao Xiang 1359f6cc76eSGao Xiang for (pcs = pcluster_pool; 1369f6cc76eSGao Xiang pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 1379f6cc76eSGao Xiang size = struct_size(a, compressed_pages, pcs->maxpages); 1389f6cc76eSGao Xiang 1399f6cc76eSGao Xiang sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 1409f6cc76eSGao Xiang pcs->slab = kmem_cache_create(pcs->name, size, 0, 1419f6cc76eSGao Xiang SLAB_RECLAIM_ACCOUNT, NULL); 1429f6cc76eSGao Xiang if (pcs->slab) 1439f6cc76eSGao Xiang continue; 1449f6cc76eSGao Xiang 1459f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 1469f6cc76eSGao Xiang return -ENOMEM; 1479f6cc76eSGao Xiang } 1489f6cc76eSGao Xiang return 0; 1499f6cc76eSGao Xiang } 1509f6cc76eSGao Xiang 1519f6cc76eSGao Xiang static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 1529f6cc76eSGao Xiang { 1539f6cc76eSGao Xiang int i; 1549f6cc76eSGao Xiang 1559f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 1569f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 1579f6cc76eSGao Xiang struct z_erofs_pcluster *pcl; 1589f6cc76eSGao Xiang 1599f6cc76eSGao Xiang if (nrpages > pcs->maxpages) 1609f6cc76eSGao Xiang continue; 1619f6cc76eSGao Xiang 1629f6cc76eSGao Xiang pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 1639f6cc76eSGao Xiang if (!pcl) 1649f6cc76eSGao Xiang return ERR_PTR(-ENOMEM); 1659f6cc76eSGao Xiang pcl->pclusterpages = nrpages; 1669f6cc76eSGao Xiang return pcl; 1679f6cc76eSGao Xiang } 1689f6cc76eSGao Xiang return ERR_PTR(-EINVAL); 1699f6cc76eSGao Xiang } 1709f6cc76eSGao Xiang 1719f6cc76eSGao Xiang static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 1729f6cc76eSGao Xiang { 173cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1749f6cc76eSGao Xiang int i; 1759f6cc76eSGao Xiang 1769f6cc76eSGao Xiang for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 1779f6cc76eSGao Xiang struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 1789f6cc76eSGao Xiang 179cecf864dSYue Hu if (pclusterpages > pcs->maxpages) 1809f6cc76eSGao Xiang continue; 1819f6cc76eSGao Xiang 1829f6cc76eSGao Xiang kmem_cache_free(pcs->slab, pcl); 1839f6cc76eSGao Xiang return; 1849f6cc76eSGao Xiang } 1859f6cc76eSGao Xiang DBG_BUGON(1); 1869f6cc76eSGao Xiang } 1879f6cc76eSGao Xiang 18847e4937aSGao Xiang /* how to allocate cached pages for a pcluster */ 18947e4937aSGao Xiang enum z_erofs_cache_alloctype { 19047e4937aSGao Xiang DONTALLOC, /* don't allocate any cached pages */ 1911825c8d7SGao Xiang /* 1921825c8d7SGao Xiang * try to use cached I/O if page allocation succeeds or fallback 1931825c8d7SGao Xiang * to in-place I/O instead to avoid any direct reclaim. 1941825c8d7SGao Xiang */ 1951825c8d7SGao Xiang TRYALLOC, 19647e4937aSGao Xiang }; 19747e4937aSGao Xiang 19847e4937aSGao Xiang /* 19947e4937aSGao Xiang * tagged pointer with 1-bit tag for all compressed pages 20047e4937aSGao Xiang * tag 0 - the page is just found with an extra page reference 20147e4937aSGao Xiang */ 20247e4937aSGao Xiang typedef tagptr1_t compressed_page_t; 20347e4937aSGao Xiang 20447e4937aSGao Xiang #define tag_compressed_page_justfound(page) \ 20547e4937aSGao Xiang tagptr_fold(compressed_page_t, page, 1) 20647e4937aSGao Xiang 20747e4937aSGao Xiang static struct workqueue_struct *z_erofs_workqueue __read_mostly; 20847e4937aSGao Xiang 20947e4937aSGao Xiang void z_erofs_exit_zip_subsystem(void) 21047e4937aSGao Xiang { 21147e4937aSGao Xiang destroy_workqueue(z_erofs_workqueue); 2129f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 21347e4937aSGao Xiang } 21447e4937aSGao Xiang 21599634bf3SGao Xiang static inline int z_erofs_init_workqueue(void) 21647e4937aSGao Xiang { 21747e4937aSGao Xiang const unsigned int onlinecpus = num_possible_cpus(); 21847e4937aSGao Xiang 21947e4937aSGao Xiang /* 22047e4937aSGao Xiang * no need to spawn too many threads, limiting threads could minimum 22147e4937aSGao Xiang * scheduling overhead, perhaps per-CPU threads should be better? 22247e4937aSGao Xiang */ 2230e62ea33SGao Xiang z_erofs_workqueue = alloc_workqueue("erofs_unzipd", 2240e62ea33SGao Xiang WQ_UNBOUND | WQ_HIGHPRI, 22547e4937aSGao Xiang onlinecpus + onlinecpus / 4); 22647e4937aSGao Xiang return z_erofs_workqueue ? 0 : -ENOMEM; 22747e4937aSGao Xiang } 22847e4937aSGao Xiang 22947e4937aSGao Xiang int __init z_erofs_init_zip_subsystem(void) 23047e4937aSGao Xiang { 2319f6cc76eSGao Xiang int err = z_erofs_create_pcluster_pool(); 23247e4937aSGao Xiang 2339f6cc76eSGao Xiang if (err) 2349f6cc76eSGao Xiang return err; 2359f6cc76eSGao Xiang err = z_erofs_init_workqueue(); 2369f6cc76eSGao Xiang if (err) 2379f6cc76eSGao Xiang z_erofs_destroy_pcluster_pool(); 2389f6cc76eSGao Xiang return err; 23947e4937aSGao Xiang } 24047e4937aSGao Xiang 24147e4937aSGao Xiang enum z_erofs_collectmode { 24247e4937aSGao Xiang COLLECT_SECONDARY, 24347e4937aSGao Xiang COLLECT_PRIMARY, 24447e4937aSGao Xiang /* 24547e4937aSGao Xiang * The current collection was the tail of an exist chain, in addition 24647e4937aSGao Xiang * that the previous processed chained collections are all decided to 24747e4937aSGao Xiang * be hooked up to it. 24847e4937aSGao Xiang * A new chain will be created for the remaining collections which are 24947e4937aSGao Xiang * not processed yet, therefore different from COLLECT_PRIMARY_FOLLOWED, 25047e4937aSGao Xiang * the next collection cannot reuse the whole page safely in 25147e4937aSGao Xiang * the following scenario: 25247e4937aSGao Xiang * ________________________________________________________________ 25347e4937aSGao Xiang * | tail (partial) page | head (partial) page | 25447e4937aSGao Xiang * | (belongs to the next cl) | (belongs to the current cl) | 25547e4937aSGao Xiang * |_______PRIMARY_FOLLOWED_______|________PRIMARY_HOOKED___________| 25647e4937aSGao Xiang */ 25747e4937aSGao Xiang COLLECT_PRIMARY_HOOKED, 2580b964600SGao Xiang /* 2590b964600SGao Xiang * a weak form of COLLECT_PRIMARY_FOLLOWED, the difference is that it 2600b964600SGao Xiang * could be dispatched into bypass queue later due to uptodated managed 2610b964600SGao Xiang * pages. All related online pages cannot be reused for inplace I/O (or 262*387bab87SGao Xiang * bvpage) since it can be directly decoded without I/O submission. 2630b964600SGao Xiang */ 26447e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED_NOINPLACE, 26547e4937aSGao Xiang /* 26647e4937aSGao Xiang * The current collection has been linked with the owned chain, and 26747e4937aSGao Xiang * could also be linked with the remaining collections, which means 26847e4937aSGao Xiang * if the processing page is the tail page of the collection, thus 26947e4937aSGao Xiang * the current collection can safely use the whole page (since 27047e4937aSGao Xiang * the previous collection is under control) for in-place I/O, as 27147e4937aSGao Xiang * illustrated below: 27247e4937aSGao Xiang * ________________________________________________________________ 27347e4937aSGao Xiang * | tail (partial) page | head (partial) page | 27447e4937aSGao Xiang * | (of the current cl) | (of the previous collection) | 27547e4937aSGao Xiang * | PRIMARY_FOLLOWED or | | 27647e4937aSGao Xiang * |_____PRIMARY_HOOKED___|____________PRIMARY_FOLLOWED____________| 27747e4937aSGao Xiang * 27847e4937aSGao Xiang * [ (*) the above page can be used as inplace I/O. ] 27947e4937aSGao Xiang */ 28047e4937aSGao Xiang COLLECT_PRIMARY_FOLLOWED, 28147e4937aSGao Xiang }; 28247e4937aSGao Xiang 2835c6dcc57SGao Xiang struct z_erofs_decompress_frontend { 2845c6dcc57SGao Xiang struct inode *const inode; 2855c6dcc57SGao Xiang struct erofs_map_blocks map; 28606a304cdSGao Xiang struct z_erofs_bvec_iter biter; 28747e4937aSGao Xiang 28806a304cdSGao Xiang struct page *candidate_bvpage; 28947e4937aSGao Xiang struct z_erofs_pcluster *pcl, *tailpcl; 29081382f5fSGao Xiang /* a pointer used to pick up inplace I/O pages */ 29181382f5fSGao Xiang struct page **icpage_ptr; 29247e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 29347e4937aSGao Xiang 29447e4937aSGao Xiang enum z_erofs_collectmode mode; 29547e4937aSGao Xiang 2966ea5aad3SGao Xiang bool readahead; 29747e4937aSGao Xiang /* used for applying cache strategy on the fly */ 29847e4937aSGao Xiang bool backmost; 29947e4937aSGao Xiang erofs_off_t headoffset; 30047e4937aSGao Xiang }; 30147e4937aSGao Xiang 30247e4937aSGao Xiang #define DECOMPRESS_FRONTEND_INIT(__i) { \ 3035c6dcc57SGao Xiang .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 3044398d3c3SWeizhao Ouyang .mode = COLLECT_PRIMARY_FOLLOWED, .backmost = true } 30547e4937aSGao Xiang 30647e4937aSGao Xiang static struct page *z_pagemap_global[Z_EROFS_VMAP_GLOBAL_PAGES]; 30747e4937aSGao Xiang static DEFINE_MUTEX(z_pagemap_global_lock); 30847e4937aSGao Xiang 3096f39d1e1SGao Xiang static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 3101825c8d7SGao Xiang enum z_erofs_cache_alloctype type, 311eaa9172aSGao Xiang struct page **pagepool) 31247e4937aSGao Xiang { 3136f39d1e1SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 3145c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 31547e4937aSGao Xiang bool standalone = true; 3166f39d1e1SGao Xiang /* 3176f39d1e1SGao Xiang * optimistic allocation without direct reclaim since inplace I/O 3186f39d1e1SGao Xiang * can be used if low memory otherwise. 3196f39d1e1SGao Xiang */ 3201825c8d7SGao Xiang gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 3211825c8d7SGao Xiang __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 32281382f5fSGao Xiang struct page **pages; 32381382f5fSGao Xiang pgoff_t index; 32447e4937aSGao Xiang 3255c6dcc57SGao Xiang if (fe->mode < COLLECT_PRIMARY_FOLLOWED) 32647e4937aSGao Xiang return; 32747e4937aSGao Xiang 32881382f5fSGao Xiang pages = pcl->compressed_pages; 32981382f5fSGao Xiang index = pcl->obj.index; 33081382f5fSGao Xiang for (; index < pcl->obj.index + pcl->pclusterpages; ++index, ++pages) { 33147e4937aSGao Xiang struct page *page; 33247e4937aSGao Xiang compressed_page_t t; 3331825c8d7SGao Xiang struct page *newpage = NULL; 33447e4937aSGao Xiang 33547e4937aSGao Xiang /* the compressed page was loaded before */ 33647e4937aSGao Xiang if (READ_ONCE(*pages)) 33747e4937aSGao Xiang continue; 33847e4937aSGao Xiang 33947e4937aSGao Xiang page = find_get_page(mc, index); 34047e4937aSGao Xiang 34147e4937aSGao Xiang if (page) { 34247e4937aSGao Xiang t = tag_compressed_page_justfound(page); 3430b964600SGao Xiang } else { 3440b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 3450b964600SGao Xiang standalone = false; 3460b964600SGao Xiang switch (type) { 3470b964600SGao Xiang case TRYALLOC: 3481825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 3491825c8d7SGao Xiang if (!newpage) 35047e4937aSGao Xiang continue; 3510b964600SGao Xiang set_page_private(newpage, 3520b964600SGao Xiang Z_EROFS_PREALLOCATED_PAGE); 3530b964600SGao Xiang t = tag_compressed_page_justfound(newpage); 3540b964600SGao Xiang break; 3550b964600SGao Xiang default: /* DONTALLOC */ 3560b964600SGao Xiang continue; 3570b964600SGao Xiang } 35847e4937aSGao Xiang } 35947e4937aSGao Xiang 36047e4937aSGao Xiang if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t))) 36147e4937aSGao Xiang continue; 36247e4937aSGao Xiang 363eaa9172aSGao Xiang if (page) 36447e4937aSGao Xiang put_page(page); 365eaa9172aSGao Xiang else if (newpage) 366eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 36747e4937aSGao Xiang } 36847e4937aSGao Xiang 3690b964600SGao Xiang /* 3700b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 3710b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 3720b964600SGao Xiang */ 3730b964600SGao Xiang if (standalone) 3745c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 37547e4937aSGao Xiang } 37647e4937aSGao Xiang 37747e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 37847e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 37947e4937aSGao Xiang struct erofs_workgroup *grp) 38047e4937aSGao Xiang { 38147e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 38247e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 38347e4937aSGao Xiang int i; 38447e4937aSGao Xiang 385cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 38647e4937aSGao Xiang /* 38747e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 38847e4937aSGao Xiang * therefore no need to worry about available decompression users. 38947e4937aSGao Xiang */ 3909f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 39147e4937aSGao Xiang struct page *page = pcl->compressed_pages[i]; 39247e4937aSGao Xiang 39347e4937aSGao Xiang if (!page) 39447e4937aSGao Xiang continue; 39547e4937aSGao Xiang 39647e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 39747e4937aSGao Xiang if (!trylock_page(page)) 39847e4937aSGao Xiang return -EBUSY; 39947e4937aSGao Xiang 400f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 40147e4937aSGao Xiang continue; 40247e4937aSGao Xiang 40347e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 40447e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 4056aaa7b06SGao Xiang detach_page_private(page); 40647e4937aSGao Xiang unlock_page(page); 40747e4937aSGao Xiang } 40847e4937aSGao Xiang return 0; 40947e4937aSGao Xiang } 41047e4937aSGao Xiang 411d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 41247e4937aSGao Xiang { 41347e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 41447e4937aSGao Xiang int ret = 0; /* 0 - busy */ 41547e4937aSGao Xiang 41647e4937aSGao Xiang if (erofs_workgroup_try_to_freeze(&pcl->obj, 1)) { 41747e4937aSGao Xiang unsigned int i; 41847e4937aSGao Xiang 419cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 4209f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 42147e4937aSGao Xiang if (pcl->compressed_pages[i] == page) { 42247e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[i], NULL); 42347e4937aSGao Xiang ret = 1; 42447e4937aSGao Xiang break; 42547e4937aSGao Xiang } 42647e4937aSGao Xiang } 42747e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 42847e4937aSGao Xiang 4296aaa7b06SGao Xiang if (ret) 4306aaa7b06SGao Xiang detach_page_private(page); 43147e4937aSGao Xiang } 43247e4937aSGao Xiang return ret; 43347e4937aSGao Xiang } 43447e4937aSGao Xiang 43547e4937aSGao Xiang /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 4365c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 43747e4937aSGao Xiang struct page *page) 43847e4937aSGao Xiang { 4395c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 44047e4937aSGao Xiang 4415c6dcc57SGao Xiang while (fe->icpage_ptr > pcl->compressed_pages) 4425c6dcc57SGao Xiang if (!cmpxchg(--fe->icpage_ptr, NULL, page)) 44347e4937aSGao Xiang return true; 44447e4937aSGao Xiang return false; 44547e4937aSGao Xiang } 44647e4937aSGao Xiang 44787ca34a7SGao Xiang /* callers must be with pcluster lock held */ 4485c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 44906a304cdSGao Xiang struct z_erofs_bvec *bvec, 45006a304cdSGao Xiang enum z_erofs_page_type type) 45147e4937aSGao Xiang { 45247e4937aSGao Xiang int ret; 45347e4937aSGao Xiang 4545c6dcc57SGao Xiang if (fe->mode >= COLLECT_PRIMARY && 45506a304cdSGao Xiang type == Z_EROFS_PAGE_TYPE_EXCLUSIVE) { 45606a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 45706a304cdSGao Xiang if (z_erofs_try_inplace_io(fe, bvec->page)) 45847e4937aSGao Xiang return 0; 45906a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 46006a304cdSGao Xiang if (fe->mode >= COLLECT_PRIMARY_FOLLOWED && 46106a304cdSGao Xiang !fe->candidate_bvpage) 46206a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 46306a304cdSGao Xiang } 46406a304cdSGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 46506a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 46606a304cdSGao Xiang return ret; 46747e4937aSGao Xiang } 46847e4937aSGao Xiang 4695c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 47047e4937aSGao Xiang { 4715c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 4725c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 47347e4937aSGao Xiang 474473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 475473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 476473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 47747e4937aSGao Xiang *owned_head = &pcl->next; 478473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 4795c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_FOLLOWED; 480473e15b0SGao Xiang return; 481473e15b0SGao Xiang } 482473e15b0SGao Xiang 48347e4937aSGao Xiang /* 484473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 485473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 48647e4937aSGao Xiang */ 48747e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 488473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 48947e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 4905c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_HOOKED; 4915c6dcc57SGao Xiang f->tailpcl = NULL; 492473e15b0SGao Xiang return; 49347e4937aSGao Xiang } 494473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 4955c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY; 49647e4937aSGao Xiang } 49747e4937aSGao Xiang 49883a386c0SGao Xiang static int z_erofs_lookup_pcluster(struct z_erofs_decompress_frontend *fe) 49947e4937aSGao Xiang { 50083a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 5015c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 50247e4937aSGao Xiang unsigned int length; 50347e4937aSGao Xiang 50464094a04SGao Xiang /* to avoid unexpected loop formed by corrupted images */ 5055c6dcc57SGao Xiang if (fe->owned_head == &pcl->next || pcl == fe->tailpcl) { 50647e4937aSGao Xiang DBG_BUGON(1); 5079e579fc1SGao Xiang return -EFSCORRUPTED; 50847e4937aSGao Xiang } 50947e4937aSGao Xiang 51087ca34a7SGao Xiang if (pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) { 51147e4937aSGao Xiang DBG_BUGON(1); 5129e579fc1SGao Xiang return -EFSCORRUPTED; 51347e4937aSGao Xiang } 51447e4937aSGao Xiang 51547e4937aSGao Xiang length = READ_ONCE(pcl->length); 51647e4937aSGao Xiang if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 51747e4937aSGao Xiang if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 51847e4937aSGao Xiang DBG_BUGON(1); 5199e579fc1SGao Xiang return -EFSCORRUPTED; 52047e4937aSGao Xiang } 52147e4937aSGao Xiang } else { 52247e4937aSGao Xiang unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 52347e4937aSGao Xiang 52447e4937aSGao Xiang if (map->m_flags & EROFS_MAP_FULL_MAPPED) 52547e4937aSGao Xiang llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 52647e4937aSGao Xiang 52747e4937aSGao Xiang while (llen > length && 52847e4937aSGao Xiang length != cmpxchg_relaxed(&pcl->length, length, llen)) { 52947e4937aSGao Xiang cpu_relax(); 53047e4937aSGao Xiang length = READ_ONCE(pcl->length); 53147e4937aSGao Xiang } 53247e4937aSGao Xiang } 53387ca34a7SGao Xiang mutex_lock(&pcl->lock); 53447e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 5355c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 5365c6dcc57SGao Xiang fe->tailpcl = pcl; 537473e15b0SGao Xiang 5385c6dcc57SGao Xiang z_erofs_try_to_claim_pcluster(fe); 5399e579fc1SGao Xiang return 0; 54047e4937aSGao Xiang } 54147e4937aSGao Xiang 54283a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 54347e4937aSGao Xiang { 54483a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 545cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 54647e4937aSGao Xiang struct z_erofs_pcluster *pcl; 54764094a04SGao Xiang struct erofs_workgroup *grp; 54847e4937aSGao Xiang int err; 54947e4937aSGao Xiang 5508f899262SGao Xiang if (!(map->m_flags & EROFS_MAP_ENCODED)) { 5518f899262SGao Xiang DBG_BUGON(1); 5528f899262SGao Xiang return -EFSCORRUPTED; 5538f899262SGao Xiang } 5548f899262SGao Xiang 5559f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 556cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 557cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 5589f6cc76eSGao Xiang if (IS_ERR(pcl)) 5599f6cc76eSGao Xiang return PTR_ERR(pcl); 56047e4937aSGao Xiang 56164094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 5628f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 56347e4937aSGao Xiang pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 56447e4937aSGao Xiang (map->m_flags & EROFS_MAP_FULL_MAPPED ? 56547e4937aSGao Xiang Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 56647e4937aSGao Xiang 56747e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 5685c6dcc57SGao Xiang pcl->next = fe->owned_head; 56987ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 5705c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED; 57147e4937aSGao Xiang 57247e4937aSGao Xiang /* 57347e4937aSGao Xiang * lock all primary followed works before visible to others 57447e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 57547e4937aSGao Xiang */ 57687ca34a7SGao Xiang mutex_init(&pcl->lock); 57787ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 57847e4937aSGao Xiang 579cecf864dSYue Hu if (ztailpacking) { 580cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 581cecf864dSYue Hu pcl->pageofs_in = erofs_blkoff(map->m_pa); 582cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 583cecf864dSYue Hu } else { 584cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 585cecf864dSYue Hu 58683a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 58764094a04SGao Xiang if (IS_ERR(grp)) { 58864094a04SGao Xiang err = PTR_ERR(grp); 58964094a04SGao Xiang goto err_out; 59064094a04SGao Xiang } 59164094a04SGao Xiang 59264094a04SGao Xiang if (grp != &pcl->obj) { 5935c6dcc57SGao Xiang fe->pcl = container_of(grp, 594cecf864dSYue Hu struct z_erofs_pcluster, obj); 59564094a04SGao Xiang err = -EEXIST; 59664094a04SGao Xiang goto err_out; 59747e4937aSGao Xiang } 598cecf864dSYue Hu } 59947e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 6005c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 6015c6dcc57SGao Xiang fe->tailpcl = pcl; 6025c6dcc57SGao Xiang fe->owned_head = &pcl->next; 6035c6dcc57SGao Xiang fe->pcl = pcl; 6049e579fc1SGao Xiang return 0; 60564094a04SGao Xiang 60664094a04SGao Xiang err_out: 60787ca34a7SGao Xiang mutex_unlock(&pcl->lock); 6089f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 60964094a04SGao Xiang return err; 61047e4937aSGao Xiang } 61147e4937aSGao Xiang 61283a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 61347e4937aSGao Xiang { 61483a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 6150d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 6169e579fc1SGao Xiang int ret; 61747e4937aSGao Xiang 61887ca34a7SGao Xiang DBG_BUGON(fe->pcl); 61947e4937aSGao Xiang 62087ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 6215c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 6225c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 62347e4937aSGao Xiang 6240d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 6250d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 6260d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 6270d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 62847e4937aSGao Xiang DBG_BUGON(1); 629cecf864dSYue Hu return -EFSCORRUPTED; 630cecf864dSYue Hu } 63147e4937aSGao Xiang 63264094a04SGao Xiang if (grp) { 6335c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 6340d823b42SGao Xiang ret = -EEXIST; 63564094a04SGao Xiang } else { 63683a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 63764094a04SGao Xiang } 63847e4937aSGao Xiang 6390d823b42SGao Xiang if (ret == -EEXIST) { 64083a386c0SGao Xiang ret = z_erofs_lookup_pcluster(fe); 64164094a04SGao Xiang if (ret) { 6425c6dcc57SGao Xiang erofs_workgroup_put(&fe->pcl->obj); 64364094a04SGao Xiang return ret; 64464094a04SGao Xiang } 6450d823b42SGao Xiang } else if (ret) { 6460d823b42SGao Xiang return ret; 6470d823b42SGao Xiang } 64806a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 649*387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 65081382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 6515c6dcc57SGao Xiang fe->icpage_ptr = fe->pcl->compressed_pages + 6525c6dcc57SGao Xiang z_erofs_pclusterpages(fe->pcl); 65347e4937aSGao Xiang return 0; 65447e4937aSGao Xiang } 65547e4937aSGao Xiang 65647e4937aSGao Xiang /* 65747e4937aSGao Xiang * keep in mind that no referenced pclusters will be freed 65847e4937aSGao Xiang * only after a RCU grace period. 65947e4937aSGao Xiang */ 66047e4937aSGao Xiang static void z_erofs_rcu_callback(struct rcu_head *head) 66147e4937aSGao Xiang { 66287ca34a7SGao Xiang z_erofs_free_pcluster(container_of(head, 66387ca34a7SGao Xiang struct z_erofs_pcluster, rcu)); 66447e4937aSGao Xiang } 66547e4937aSGao Xiang 66647e4937aSGao Xiang void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 66747e4937aSGao Xiang { 66847e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 66947e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 67047e4937aSGao Xiang 67187ca34a7SGao Xiang call_rcu(&pcl->rcu, z_erofs_rcu_callback); 67247e4937aSGao Xiang } 67347e4937aSGao Xiang 6745c6dcc57SGao Xiang static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 67547e4937aSGao Xiang { 67687ca34a7SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 67747e4937aSGao Xiang 67887ca34a7SGao Xiang if (!pcl) 67947e4937aSGao Xiang return false; 68047e4937aSGao Xiang 68106a304cdSGao Xiang z_erofs_bvec_iter_end(&fe->biter); 68287ca34a7SGao Xiang mutex_unlock(&pcl->lock); 68347e4937aSGao Xiang 68406a304cdSGao Xiang if (fe->candidate_bvpage) { 68506a304cdSGao Xiang DBG_BUGON(z_erofs_is_shortlived_page(fe->candidate_bvpage)); 68606a304cdSGao Xiang fe->candidate_bvpage = NULL; 68706a304cdSGao Xiang } 68806a304cdSGao Xiang 68947e4937aSGao Xiang /* 69047e4937aSGao Xiang * if all pending pages are added, don't hold its reference 69147e4937aSGao Xiang * any longer if the pcluster isn't hosted by ourselves. 69247e4937aSGao Xiang */ 6935c6dcc57SGao Xiang if (fe->mode < COLLECT_PRIMARY_FOLLOWED_NOINPLACE) 69487ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 69547e4937aSGao Xiang 69687ca34a7SGao Xiang fe->pcl = NULL; 69747e4937aSGao Xiang return true; 69847e4937aSGao Xiang } 69947e4937aSGao Xiang 70047e4937aSGao Xiang static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe, 70147e4937aSGao Xiang unsigned int cachestrategy, 70247e4937aSGao Xiang erofs_off_t la) 70347e4937aSGao Xiang { 70447e4937aSGao Xiang if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 70547e4937aSGao Xiang return false; 70647e4937aSGao Xiang 70747e4937aSGao Xiang if (fe->backmost) 70847e4937aSGao Xiang return true; 70947e4937aSGao Xiang 71047e4937aSGao Xiang return cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 71147e4937aSGao Xiang la < fe->headoffset; 71247e4937aSGao Xiang } 71347e4937aSGao Xiang 71447e4937aSGao Xiang static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 715eaa9172aSGao Xiang struct page *page, struct page **pagepool) 71647e4937aSGao Xiang { 71747e4937aSGao Xiang struct inode *const inode = fe->inode; 718bda17a45SGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 71947e4937aSGao Xiang struct erofs_map_blocks *const map = &fe->map; 72047e4937aSGao Xiang const loff_t offset = page_offset(page); 721dc76ea8cSGao Xiang bool tight = true; 72247e4937aSGao Xiang 72347e4937aSGao Xiang enum z_erofs_cache_alloctype cache_strategy; 72447e4937aSGao Xiang enum z_erofs_page_type page_type; 72547e4937aSGao Xiang unsigned int cur, end, spiltted, index; 72647e4937aSGao Xiang int err = 0; 72747e4937aSGao Xiang 72847e4937aSGao Xiang /* register locked file pages as online pages in pack */ 72947e4937aSGao Xiang z_erofs_onlinepage_init(page); 73047e4937aSGao Xiang 73147e4937aSGao Xiang spiltted = 0; 73247e4937aSGao Xiang end = PAGE_SIZE; 73347e4937aSGao Xiang repeat: 73447e4937aSGao Xiang cur = end - 1; 73547e4937aSGao Xiang 73639397a46SGao Xiang if (offset + cur < map->m_la || 73739397a46SGao Xiang offset + cur >= map->m_la + map->m_llen) { 73839397a46SGao Xiang erofs_dbg("out-of-range map @ pos %llu", offset + cur); 73947e4937aSGao Xiang 7405c6dcc57SGao Xiang if (z_erofs_collector_end(fe)) 74147e4937aSGao Xiang fe->backmost = false; 74247e4937aSGao Xiang map->m_la = offset + cur; 74347e4937aSGao Xiang map->m_llen = 0; 74447e4937aSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 0); 7458d8a09b0SGao Xiang if (err) 74647e4937aSGao Xiang goto err_out; 74739397a46SGao Xiang } else { 74839397a46SGao Xiang if (fe->pcl) 74939397a46SGao Xiang goto hitted; 75039397a46SGao Xiang /* didn't get a valid pcluster previously (very rare) */ 75139397a46SGao Xiang } 75247e4937aSGao Xiang 7538d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) 75447e4937aSGao Xiang goto hitted; 75547e4937aSGao Xiang 75683a386c0SGao Xiang err = z_erofs_collector_begin(fe); 7578d8a09b0SGao Xiang if (err) 75847e4937aSGao Xiang goto err_out; 75947e4937aSGao Xiang 7605c6dcc57SGao Xiang if (z_erofs_is_inline_pcluster(fe->pcl)) { 76109c54379SGao Xiang void *mp; 762cecf864dSYue Hu 76309c54379SGao Xiang mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 76409c54379SGao Xiang erofs_blknr(map->m_pa), EROFS_NO_KMAP); 76509c54379SGao Xiang if (IS_ERR(mp)) { 76609c54379SGao Xiang err = PTR_ERR(mp); 767cecf864dSYue Hu erofs_err(inode->i_sb, 768cecf864dSYue Hu "failed to get inline page, err %d", err); 769cecf864dSYue Hu goto err_out; 770cecf864dSYue Hu } 77109c54379SGao Xiang get_page(fe->map.buf.page); 7725c6dcc57SGao Xiang WRITE_ONCE(fe->pcl->compressed_pages[0], fe->map.buf.page); 7735c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 774cecf864dSYue Hu } else { 7756f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 776cecf864dSYue Hu if (should_alloc_managed_pages(fe, sbi->opt.cache_strategy, 777cecf864dSYue Hu map->m_la)) 7781825c8d7SGao Xiang cache_strategy = TRYALLOC; 77947e4937aSGao Xiang else 78047e4937aSGao Xiang cache_strategy = DONTALLOC; 78147e4937aSGao Xiang 7826f39d1e1SGao Xiang z_erofs_bind_cache(fe, cache_strategy, pagepool); 783cecf864dSYue Hu } 78447e4937aSGao Xiang hitted: 785dc76ea8cSGao Xiang /* 786dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 787dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 788dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 789*387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 790dc76ea8cSGao Xiang */ 7915c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_HOOKED && 7925c6dcc57SGao Xiang fe->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 793dc76ea8cSGao Xiang 79447e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 7958d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 79647e4937aSGao Xiang zero_user_segment(page, cur, end); 79747e4937aSGao Xiang goto next_part; 79847e4937aSGao Xiang } 79947e4937aSGao Xiang 80047e4937aSGao Xiang /* let's derive page type */ 80147e4937aSGao Xiang page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 80247e4937aSGao Xiang (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 80347e4937aSGao Xiang (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 80447e4937aSGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 80547e4937aSGao Xiang 80647e4937aSGao Xiang if (cur) 8075c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_FOLLOWED); 80847e4937aSGao Xiang 80947e4937aSGao Xiang retry: 81006a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 81106a304cdSGao Xiang .page = page, 81206a304cdSGao Xiang .offset = offset - map->m_la, 81306a304cdSGao Xiang .end = end, 81406a304cdSGao Xiang }), page_type); 81506a304cdSGao Xiang /* should allocate an additional short-lived page for bvset */ 81606a304cdSGao Xiang if (err == -EAGAIN && !fe->candidate_bvpage) { 81706a304cdSGao Xiang fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL); 81806a304cdSGao Xiang set_page_private(fe->candidate_bvpage, 81906a304cdSGao Xiang Z_EROFS_SHORTLIVED_PAGE); 82047e4937aSGao Xiang goto retry; 82147e4937aSGao Xiang } 82247e4937aSGao Xiang 82306a304cdSGao Xiang if (err) { 82406a304cdSGao Xiang DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage); 82547e4937aSGao Xiang goto err_out; 82606a304cdSGao Xiang } 82747e4937aSGao Xiang 82847e4937aSGao Xiang index = page->index - (map->m_la >> PAGE_SHIFT); 82947e4937aSGao Xiang 83047e4937aSGao Xiang z_erofs_onlinepage_fixup(page, index, true); 83147e4937aSGao Xiang 83247e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 83347e4937aSGao Xiang ++spiltted; 83447e4937aSGao Xiang /* also update nr_pages */ 83587ca34a7SGao Xiang fe->pcl->nr_pages = max_t(pgoff_t, fe->pcl->nr_pages, index + 1); 83647e4937aSGao Xiang next_part: 83747e4937aSGao Xiang /* can be used for verification */ 83847e4937aSGao Xiang map->m_llen = offset + cur - map->m_la; 83947e4937aSGao Xiang 84047e4937aSGao Xiang end = cur; 84147e4937aSGao Xiang if (end > 0) 84247e4937aSGao Xiang goto repeat; 84347e4937aSGao Xiang 84447e4937aSGao Xiang out: 84547e4937aSGao Xiang z_erofs_onlinepage_endio(page); 84647e4937aSGao Xiang 8474f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 84847e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 84947e4937aSGao Xiang return err; 85047e4937aSGao Xiang 85147e4937aSGao Xiang /* if some error occurred while processing this page */ 85247e4937aSGao Xiang err_out: 85347e4937aSGao Xiang SetPageError(page); 85447e4937aSGao Xiang goto out; 85547e4937aSGao Xiang } 85647e4937aSGao Xiang 85740452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 85840452ffcSHuang Jianan unsigned int readahead_pages) 85940452ffcSHuang Jianan { 860a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 86140452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 86240452ffcSHuang Jianan !readahead_pages) 86340452ffcSHuang Jianan return true; 86440452ffcSHuang Jianan 86540452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 86640452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 86740452ffcSHuang Jianan return true; 86840452ffcSHuang Jianan 86940452ffcSHuang Jianan return false; 87040452ffcSHuang Jianan } 87140452ffcSHuang Jianan 8726aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 8736aaa7b06SGao Xiang { 8746aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 8756aaa7b06SGao Xiang } 8766aaa7b06SGao Xiang 87742fec235SGao Xiang static int z_erofs_parse_out_bvecs(struct z_erofs_pcluster *pcl, 87842fec235SGao Xiang struct page **pages, struct page **pagepool) 87942fec235SGao Xiang { 88006a304cdSGao Xiang struct z_erofs_bvec_iter biter; 88106a304cdSGao Xiang struct page *old_bvpage; 88242fec235SGao Xiang int i, err = 0; 88342fec235SGao Xiang 884*387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 88542fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 88606a304cdSGao Xiang struct z_erofs_bvec bvec; 88742fec235SGao Xiang unsigned int pagenr; 88842fec235SGao Xiang 88906a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 89042fec235SGao Xiang 89106a304cdSGao Xiang if (old_bvpage) 89206a304cdSGao Xiang z_erofs_put_shortlivedpage(pagepool, old_bvpage); 89342fec235SGao Xiang 89406a304cdSGao Xiang pagenr = (bvec.offset + pcl->pageofs_out) >> PAGE_SHIFT; 89542fec235SGao Xiang DBG_BUGON(pagenr >= pcl->nr_pages); 89606a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 89742fec235SGao Xiang /* 89842fec235SGao Xiang * currently EROFS doesn't support multiref(dedup), 89942fec235SGao Xiang * so here erroring out one multiref page. 90042fec235SGao Xiang */ 90142fec235SGao Xiang if (pages[pagenr]) { 90242fec235SGao Xiang DBG_BUGON(1); 90342fec235SGao Xiang SetPageError(pages[pagenr]); 90442fec235SGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 90542fec235SGao Xiang err = -EFSCORRUPTED; 90642fec235SGao Xiang } 90706a304cdSGao Xiang pages[pagenr] = bvec.page; 90842fec235SGao Xiang } 90906a304cdSGao Xiang 91006a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 91106a304cdSGao Xiang if (old_bvpage) 91206a304cdSGao Xiang z_erofs_put_shortlivedpage(pagepool, old_bvpage); 91342fec235SGao Xiang return err; 91442fec235SGao Xiang } 91542fec235SGao Xiang 91647e4937aSGao Xiang static int z_erofs_decompress_pcluster(struct super_block *sb, 91747e4937aSGao Xiang struct z_erofs_pcluster *pcl, 918eaa9172aSGao Xiang struct page **pagepool) 91947e4937aSGao Xiang { 92047e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 921cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 9229f6cc76eSGao Xiang unsigned int i, inputsize, outputsize, llen, nr_pages; 92347e4937aSGao Xiang struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 92447e4937aSGao Xiang struct page **pages, **compressed_pages, *page; 92547e4937aSGao Xiang 92647e4937aSGao Xiang bool overlapped, partial; 92747e4937aSGao Xiang int err; 92847e4937aSGao Xiang 92947e4937aSGao Xiang might_sleep(); 93087ca34a7SGao Xiang DBG_BUGON(!READ_ONCE(pcl->nr_pages)); 93147e4937aSGao Xiang 93287ca34a7SGao Xiang mutex_lock(&pcl->lock); 93387ca34a7SGao Xiang nr_pages = pcl->nr_pages; 93447e4937aSGao Xiang 9358d8a09b0SGao Xiang if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 93647e4937aSGao Xiang pages = pages_onstack; 93747e4937aSGao Xiang } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 93847e4937aSGao Xiang mutex_trylock(&z_pagemap_global_lock)) { 93947e4937aSGao Xiang pages = z_pagemap_global; 94047e4937aSGao Xiang } else { 94147e4937aSGao Xiang gfp_t gfp_flags = GFP_KERNEL; 94247e4937aSGao Xiang 94347e4937aSGao Xiang if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 94447e4937aSGao Xiang gfp_flags |= __GFP_NOFAIL; 94547e4937aSGao Xiang 94647e4937aSGao Xiang pages = kvmalloc_array(nr_pages, sizeof(struct page *), 94747e4937aSGao Xiang gfp_flags); 94847e4937aSGao Xiang 94947e4937aSGao Xiang /* fallback to global pagemap for the lowmem scenario */ 9508d8a09b0SGao Xiang if (!pages) { 95147e4937aSGao Xiang mutex_lock(&z_pagemap_global_lock); 95247e4937aSGao Xiang pages = z_pagemap_global; 95347e4937aSGao Xiang } 95447e4937aSGao Xiang } 95547e4937aSGao Xiang 95647e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) 95747e4937aSGao Xiang pages[i] = NULL; 95847e4937aSGao Xiang 95942fec235SGao Xiang err = z_erofs_parse_out_bvecs(pcl, pages, pagepool); 96047e4937aSGao Xiang 96147e4937aSGao Xiang overlapped = false; 96247e4937aSGao Xiang compressed_pages = pcl->compressed_pages; 96347e4937aSGao Xiang 964cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 96547e4937aSGao Xiang unsigned int pagenr; 96647e4937aSGao Xiang 96747e4937aSGao Xiang page = compressed_pages[i]; 96847e4937aSGao Xiang /* all compressed pages ought to be valid */ 96947e4937aSGao Xiang DBG_BUGON(!page); 97047e4937aSGao Xiang 971cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 972cecf864dSYue Hu if (!PageUptodate(page)) 973cecf864dSYue Hu err = -EIO; 974cecf864dSYue Hu continue; 975cecf864dSYue Hu } 976cecf864dSYue Hu 977cecf864dSYue Hu DBG_BUGON(z_erofs_page_is_invalidated(page)); 9786aaa7b06SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 97947e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) { 9808d8a09b0SGao Xiang if (!PageUptodate(page)) 98147e4937aSGao Xiang err = -EIO; 98247e4937aSGao Xiang continue; 98347e4937aSGao Xiang } 98447e4937aSGao Xiang 98547e4937aSGao Xiang /* 98647e4937aSGao Xiang * only if non-head page can be selected 98747e4937aSGao Xiang * for inplace decompression 98847e4937aSGao Xiang */ 98947e4937aSGao Xiang pagenr = z_erofs_onlinepage_index(page); 99047e4937aSGao Xiang 99147e4937aSGao Xiang DBG_BUGON(pagenr >= nr_pages); 9928d8a09b0SGao Xiang if (pages[pagenr]) { 99347e4937aSGao Xiang DBG_BUGON(1); 99447e4937aSGao Xiang SetPageError(pages[pagenr]); 99547e4937aSGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 99647e4937aSGao Xiang err = -EFSCORRUPTED; 99747e4937aSGao Xiang } 99847e4937aSGao Xiang pages[pagenr] = page; 99947e4937aSGao Xiang 100047e4937aSGao Xiang overlapped = true; 100147e4937aSGao Xiang } 100247e4937aSGao Xiang 10036aaa7b06SGao Xiang /* PG_error needs checking for all non-managed pages */ 10048d8a09b0SGao Xiang if (PageError(page)) { 100547e4937aSGao Xiang DBG_BUGON(PageUptodate(page)); 100647e4937aSGao Xiang err = -EIO; 100747e4937aSGao Xiang } 100847e4937aSGao Xiang } 100947e4937aSGao Xiang 10108d8a09b0SGao Xiang if (err) 101147e4937aSGao Xiang goto out; 101247e4937aSGao Xiang 101347e4937aSGao Xiang llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 101487ca34a7SGao Xiang if (nr_pages << PAGE_SHIFT >= pcl->pageofs_out + llen) { 101547e4937aSGao Xiang outputsize = llen; 101647e4937aSGao Xiang partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 101747e4937aSGao Xiang } else { 101887ca34a7SGao Xiang outputsize = (nr_pages << PAGE_SHIFT) - pcl->pageofs_out; 101947e4937aSGao Xiang partial = true; 102047e4937aSGao Xiang } 102147e4937aSGao Xiang 1022cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1023cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1024cecf864dSYue Hu else 1025cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1026cecf864dSYue Hu 102747e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 102847e4937aSGao Xiang .sb = sb, 102947e4937aSGao Xiang .in = compressed_pages, 103047e4937aSGao Xiang .out = pages, 1031cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 103287ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 10339f6cc76eSGao Xiang .inputsize = inputsize, 103447e4937aSGao Xiang .outputsize = outputsize, 103547e4937aSGao Xiang .alg = pcl->algorithmformat, 103647e4937aSGao Xiang .inplace_io = overlapped, 103747e4937aSGao Xiang .partial_decoding = partial 103847e4937aSGao Xiang }, pagepool); 103947e4937aSGao Xiang 104047e4937aSGao Xiang out: 1041cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1042cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1043cecf864dSYue Hu page = compressed_pages[0]; 1044cecf864dSYue Hu WRITE_ONCE(compressed_pages[0], NULL); 1045cecf864dSYue Hu put_page(page); 1046cecf864dSYue Hu } else { 1047cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 104847e4937aSGao Xiang page = compressed_pages[i]; 104947e4937aSGao Xiang 105047e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 105147e4937aSGao Xiang continue; 105247e4937aSGao Xiang 10536aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10546aaa7b06SGao Xiang (void)z_erofs_put_shortlivedpage(pagepool, page); 105547e4937aSGao Xiang WRITE_ONCE(compressed_pages[i], NULL); 105647e4937aSGao Xiang } 1057cecf864dSYue Hu } 105847e4937aSGao Xiang 105947e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) { 106047e4937aSGao Xiang page = pages[i]; 106147e4937aSGao Xiang if (!page) 106247e4937aSGao Xiang continue; 106347e4937aSGao Xiang 10646aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 106547e4937aSGao Xiang 10666aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10676aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 106847e4937aSGao Xiang continue; 106947e4937aSGao Xiang 10708d8a09b0SGao Xiang if (err < 0) 107147e4937aSGao Xiang SetPageError(page); 107247e4937aSGao Xiang 107347e4937aSGao Xiang z_erofs_onlinepage_endio(page); 107447e4937aSGao Xiang } 107547e4937aSGao Xiang 107647e4937aSGao Xiang if (pages == z_pagemap_global) 107747e4937aSGao Xiang mutex_unlock(&z_pagemap_global_lock); 10788d8a09b0SGao Xiang else if (pages != pages_onstack) 107947e4937aSGao Xiang kvfree(pages); 108047e4937aSGao Xiang 108187ca34a7SGao Xiang pcl->nr_pages = 0; 108206a304cdSGao Xiang pcl->bvset.nextpage = NULL; 108387ca34a7SGao Xiang pcl->vcnt = 0; 108447e4937aSGao Xiang 108587ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 108647e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 108787ca34a7SGao Xiang mutex_unlock(&pcl->lock); 108847e4937aSGao Xiang return err; 108947e4937aSGao Xiang } 109047e4937aSGao Xiang 10910c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1092eaa9172aSGao Xiang struct page **pagepool) 109347e4937aSGao Xiang { 109447e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 109547e4937aSGao Xiang 109647e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 109747e4937aSGao Xiang struct z_erofs_pcluster *pcl; 109847e4937aSGao Xiang 109947e4937aSGao Xiang /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 110047e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 110147e4937aSGao Xiang 110247e4937aSGao Xiang /* no possible that 'owned' equals NULL */ 110347e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 110447e4937aSGao Xiang 110547e4937aSGao Xiang pcl = container_of(owned, struct z_erofs_pcluster, next); 110647e4937aSGao Xiang owned = READ_ONCE(pcl->next); 110747e4937aSGao Xiang 1108a4b1fab1SGao Xiang z_erofs_decompress_pcluster(io->sb, pcl, pagepool); 110987ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 111047e4937aSGao Xiang } 111147e4937aSGao Xiang } 111247e4937aSGao Xiang 11130c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 111447e4937aSGao Xiang { 1115a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1116a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1117eaa9172aSGao Xiang struct page *pagepool = NULL; 111847e4937aSGao Xiang 1119a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 11200c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 112147e4937aSGao Xiang 1122eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1123a4b1fab1SGao Xiang kvfree(bgq); 112447e4937aSGao Xiang } 112547e4937aSGao Xiang 11267865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 11277865827cSGao Xiang bool sync, int bios) 11287865827cSGao Xiang { 11297865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 11307865827cSGao Xiang 11317865827cSGao Xiang /* wake up the caller thread for sync decompression */ 11327865827cSGao Xiang if (sync) { 11337865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 113460b30050SHongyu Jin complete(&io->u.done); 113560b30050SHongyu Jin 11367865827cSGao Xiang return; 11377865827cSGao Xiang } 11387865827cSGao Xiang 11397865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 11407865827cSGao Xiang return; 11417865827cSGao Xiang /* Use workqueue and sync decompression for atomic contexts only */ 11427865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 11437865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 11447865827cSGao Xiang /* enable sync decompression for readahead */ 11457865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 11467865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 11477865827cSGao Xiang return; 11487865827cSGao Xiang } 11497865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 11507865827cSGao Xiang } 11517865827cSGao Xiang 115247e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 115347e4937aSGao Xiang unsigned int nr, 1154eaa9172aSGao Xiang struct page **pagepool, 11559f2731d6SGao Xiang struct address_space *mc) 115647e4937aSGao Xiang { 115747e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 11589f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 115947e4937aSGao Xiang bool tocache = false; 116047e4937aSGao Xiang 116147e4937aSGao Xiang struct address_space *mapping; 116247e4937aSGao Xiang struct page *oldpage, *page; 116347e4937aSGao Xiang 116447e4937aSGao Xiang compressed_page_t t; 116547e4937aSGao Xiang int justfound; 116647e4937aSGao Xiang 116747e4937aSGao Xiang repeat: 116847e4937aSGao Xiang page = READ_ONCE(pcl->compressed_pages[nr]); 116947e4937aSGao Xiang oldpage = page; 117047e4937aSGao Xiang 117147e4937aSGao Xiang if (!page) 117247e4937aSGao Xiang goto out_allocpage; 117347e4937aSGao Xiang 117447e4937aSGao Xiang /* process the target tagged pointer */ 117547e4937aSGao Xiang t = tagptr_init(compressed_page_t, page); 117647e4937aSGao Xiang justfound = tagptr_unfold_tags(t); 117747e4937aSGao Xiang page = tagptr_unfold_ptr(t); 117847e4937aSGao Xiang 11791825c8d7SGao Xiang /* 11801825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 11811825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 11821825c8d7SGao Xiang */ 11831825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 11841825c8d7SGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 11851825c8d7SGao Xiang set_page_private(page, 0); 11861825c8d7SGao Xiang tocache = true; 11871825c8d7SGao Xiang goto out_tocache; 11881825c8d7SGao Xiang } 118947e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 119047e4937aSGao Xiang 119147e4937aSGao Xiang /* 11926aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 119347e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 119447e4937aSGao Xiang */ 119547e4937aSGao Xiang if (mapping && mapping != mc) 119647e4937aSGao Xiang /* ought to be unmanaged pages */ 119747e4937aSGao Xiang goto out; 119847e4937aSGao Xiang 11996aaa7b06SGao Xiang /* directly return for shortlived page as well */ 12006aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 12016aaa7b06SGao Xiang goto out; 12026aaa7b06SGao Xiang 120347e4937aSGao Xiang lock_page(page); 120447e4937aSGao Xiang 120547e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 120647e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 120747e4937aSGao Xiang 120847e4937aSGao Xiang /* the page is still in manage cache */ 120947e4937aSGao Xiang if (page->mapping == mc) { 121047e4937aSGao Xiang WRITE_ONCE(pcl->compressed_pages[nr], page); 121147e4937aSGao Xiang 121247e4937aSGao Xiang ClearPageError(page); 121347e4937aSGao Xiang if (!PagePrivate(page)) { 121447e4937aSGao Xiang /* 121547e4937aSGao Xiang * impossible to be !PagePrivate(page) for 121647e4937aSGao Xiang * the current restriction as well if 121747e4937aSGao Xiang * the page is already in compressed_pages[]. 121847e4937aSGao Xiang */ 121947e4937aSGao Xiang DBG_BUGON(!justfound); 122047e4937aSGao Xiang 122147e4937aSGao Xiang justfound = 0; 122247e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 122347e4937aSGao Xiang SetPagePrivate(page); 122447e4937aSGao Xiang } 122547e4937aSGao Xiang 122647e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 122747e4937aSGao Xiang if (PageUptodate(page)) { 122847e4937aSGao Xiang unlock_page(page); 122947e4937aSGao Xiang page = NULL; 123047e4937aSGao Xiang } 123147e4937aSGao Xiang goto out; 123247e4937aSGao Xiang } 123347e4937aSGao Xiang 123447e4937aSGao Xiang /* 123547e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 123647e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 123747e4937aSGao Xiang */ 123847e4937aSGao Xiang DBG_BUGON(page->mapping); 123947e4937aSGao Xiang DBG_BUGON(!justfound); 124047e4937aSGao Xiang 124147e4937aSGao Xiang tocache = true; 124247e4937aSGao Xiang unlock_page(page); 124347e4937aSGao Xiang put_page(page); 124447e4937aSGao Xiang out_allocpage: 12455ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 12465ddcee1fSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_pages[nr], oldpage, page)) { 1247eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 12485ddcee1fSGao Xiang cond_resched(); 12495ddcee1fSGao Xiang goto repeat; 12505ddcee1fSGao Xiang } 12511825c8d7SGao Xiang out_tocache: 1252bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1253bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1254bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1255bf225074SGao Xiang goto out; 1256a30573b3SGao Xiang } 1257bf225074SGao Xiang attach_page_private(page, pcl); 1258bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1259bf225074SGao Xiang put_page(page); 1260bf225074SGao Xiang 126147e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 126247e4937aSGao Xiang return page; 126347e4937aSGao Xiang } 126447e4937aSGao Xiang 1265a4b1fab1SGao Xiang static struct z_erofs_decompressqueue * 1266a4b1fab1SGao Xiang jobqueue_init(struct super_block *sb, 1267a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 126847e4937aSGao Xiang { 1269a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 127047e4937aSGao Xiang 1271a4b1fab1SGao Xiang if (fg && !*fg) { 1272a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1273a4b1fab1SGao Xiang if (!q) { 1274a4b1fab1SGao Xiang *fg = true; 1275a4b1fab1SGao Xiang goto fg_out; 127647e4937aSGao Xiang } 12770c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1278a4b1fab1SGao Xiang } else { 1279a4b1fab1SGao Xiang fg_out: 1280a4b1fab1SGao Xiang q = fgq; 128160b30050SHongyu Jin init_completion(&fgq->u.done); 1282a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 1283a4b1fab1SGao Xiang } 1284a4b1fab1SGao Xiang q->sb = sb; 1285a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1286a4b1fab1SGao Xiang return q; 128747e4937aSGao Xiang } 128847e4937aSGao Xiang 128947e4937aSGao Xiang /* define decompression jobqueue types */ 129047e4937aSGao Xiang enum { 129147e4937aSGao Xiang JQ_BYPASS, 129247e4937aSGao Xiang JQ_SUBMIT, 129347e4937aSGao Xiang NR_JOBQUEUES, 129447e4937aSGao Xiang }; 129547e4937aSGao Xiang 129647e4937aSGao Xiang static void *jobqueueset_init(struct super_block *sb, 1297a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[], 1298a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 129947e4937aSGao Xiang { 130047e4937aSGao Xiang /* 130147e4937aSGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 130247e4937aSGao Xiang * no need to read from device for all pclusters in this queue. 130347e4937aSGao Xiang */ 1304a4b1fab1SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1305a4b1fab1SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 130647e4937aSGao Xiang 1307a4b1fab1SGao Xiang return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 130847e4937aSGao Xiang } 130947e4937aSGao Xiang 131047e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 131147e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 131247e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 131347e4937aSGao Xiang { 131447e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 131547e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 131647e4937aSGao Xiang 131747e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 131847e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 131947e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 132047e4937aSGao Xiang 132147e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 132247e4937aSGao Xiang 132347e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 132447e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 132547e4937aSGao Xiang 132647e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 132747e4937aSGao Xiang } 132847e4937aSGao Xiang 13297865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 13307865827cSGao Xiang { 13317865827cSGao Xiang tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 13327865827cSGao Xiang struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 13337865827cSGao Xiang blk_status_t err = bio->bi_status; 13347865827cSGao Xiang struct bio_vec *bvec; 13357865827cSGao Xiang struct bvec_iter_all iter_all; 13367865827cSGao Xiang 13377865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 13387865827cSGao Xiang struct page *page = bvec->bv_page; 13397865827cSGao Xiang 13407865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 13417865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 13427865827cSGao Xiang 13437865827cSGao Xiang if (err) 13447865827cSGao Xiang SetPageError(page); 13457865827cSGao Xiang 13467865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 13477865827cSGao Xiang if (!err) 13487865827cSGao Xiang SetPageUptodate(page); 13497865827cSGao Xiang unlock_page(page); 13507865827cSGao Xiang } 13517865827cSGao Xiang } 13527865827cSGao Xiang z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 13537865827cSGao Xiang bio_put(bio); 13547865827cSGao Xiang } 13557865827cSGao Xiang 135683a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1357eaa9172aSGao Xiang struct page **pagepool, 1358a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1359a4b1fab1SGao Xiang bool *force_fg) 136047e4937aSGao Xiang { 136183a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 136283a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 136347e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1364a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 136547e4937aSGao Xiang void *bi_private; 13665c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1367dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 13683f649ab7SKees Cook pgoff_t last_index; 1369dfeab2e9SGao Xiang struct block_device *last_bdev; 13701e4a2955SGao Xiang unsigned int nr_bios = 0; 13711e4a2955SGao Xiang struct bio *bio = NULL; 137247e4937aSGao Xiang 1373a4b1fab1SGao Xiang bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1374a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1375a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 137647e4937aSGao Xiang 137747e4937aSGao Xiang /* by default, all need io submission */ 137847e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 137947e4937aSGao Xiang 138047e4937aSGao Xiang do { 1381dfeab2e9SGao Xiang struct erofs_map_dev mdev; 138247e4937aSGao Xiang struct z_erofs_pcluster *pcl; 13831e4a2955SGao Xiang pgoff_t cur, end; 13841e4a2955SGao Xiang unsigned int i = 0; 13851e4a2955SGao Xiang bool bypass = true; 138647e4937aSGao Xiang 138747e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 138847e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 138947e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 139047e4937aSGao Xiang 139147e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 139247e4937aSGao Xiang 1393cecf864dSYue Hu /* close the main owned chain at first */ 1394cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1395cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1396cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1397cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1398cecf864dSYue Hu continue; 1399cecf864dSYue Hu } 1400cecf864dSYue Hu 1401dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1402dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1403dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1404dfeab2e9SGao Xiang }; 1405dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1406dfeab2e9SGao Xiang 1407dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 14089f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 140947e4937aSGao Xiang 14101e4a2955SGao Xiang do { 14111e4a2955SGao Xiang struct page *page; 141247e4937aSGao Xiang 14131e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 141483a386c0SGao Xiang mc); 14151e4a2955SGao Xiang if (!page) 14161e4a2955SGao Xiang continue; 141747e4937aSGao Xiang 1418dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1419dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 142047e4937aSGao Xiang submit_bio_retry: 142194e4e153SGao Xiang submit_bio(bio); 142247e4937aSGao Xiang bio = NULL; 142347e4937aSGao Xiang } 142447e4937aSGao Xiang 142547e4937aSGao Xiang if (!bio) { 142607888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 142707888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 14280c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1429dfeab2e9SGao Xiang 1430dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 14311e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1432a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1433a5c0b780SGao Xiang bio->bi_private = bi_private; 14346ea5aad3SGao Xiang if (f->readahead) 14356ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 143647e4937aSGao Xiang ++nr_bios; 143747e4937aSGao Xiang } 143847e4937aSGao Xiang 14396c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 144047e4937aSGao Xiang goto submit_bio_retry; 144147e4937aSGao Xiang 14421e4a2955SGao Xiang last_index = cur; 14431e4a2955SGao Xiang bypass = false; 14441e4a2955SGao Xiang } while (++cur < end); 144547e4937aSGao Xiang 14461e4a2955SGao Xiang if (!bypass) 144747e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 144847e4937aSGao Xiang else 144947e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 145047e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 145147e4937aSGao Xiang 145247e4937aSGao Xiang if (bio) 145394e4e153SGao Xiang submit_bio(bio); 145447e4937aSGao Xiang 1455587a67b7SGao Xiang /* 1456587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1457587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1458587a67b7SGao Xiang */ 1459587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1460587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 14611e4a2955SGao Xiang return; 1462587a67b7SGao Xiang } 1463a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 146447e4937aSGao Xiang } 146547e4937aSGao Xiang 146683a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1467eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 146847e4937aSGao Xiang { 1469a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 147047e4937aSGao Xiang 14715c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 147247e4937aSGao Xiang return; 147383a386c0SGao Xiang z_erofs_submit_queue(f, pagepool, io, &force_fg); 147447e4937aSGao Xiang 14750c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 14760c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 147747e4937aSGao Xiang 147847e4937aSGao Xiang if (!force_fg) 147947e4937aSGao Xiang return; 148047e4937aSGao Xiang 148147e4937aSGao Xiang /* wait until all bios are completed */ 148260b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 148347e4937aSGao Xiang 14840c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 14850c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 148647e4937aSGao Xiang } 148747e4937aSGao Xiang 148838629291SGao Xiang /* 148938629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 149038629291SGao Xiang * approximate readmore strategies as a start. 149138629291SGao Xiang */ 149238629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 149338629291SGao Xiang struct readahead_control *rac, 149438629291SGao Xiang erofs_off_t end, 1495eaa9172aSGao Xiang struct page **pagepool, 149638629291SGao Xiang bool backmost) 149738629291SGao Xiang { 149838629291SGao Xiang struct inode *inode = f->inode; 149938629291SGao Xiang struct erofs_map_blocks *map = &f->map; 150038629291SGao Xiang erofs_off_t cur; 150138629291SGao Xiang int err; 150238629291SGao Xiang 150338629291SGao Xiang if (backmost) { 150438629291SGao Xiang map->m_la = end; 1505622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1506622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 150738629291SGao Xiang if (err) 150838629291SGao Xiang return; 150938629291SGao Xiang 151038629291SGao Xiang /* expend ra for the trailing edge if readahead */ 151138629291SGao Xiang if (rac) { 151238629291SGao Xiang loff_t newstart = readahead_pos(rac); 151338629291SGao Xiang 151438629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 151538629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 151638629291SGao Xiang return; 151738629291SGao Xiang } 151838629291SGao Xiang end = round_up(end, PAGE_SIZE); 151938629291SGao Xiang } else { 152038629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 152138629291SGao Xiang 152238629291SGao Xiang if (!map->m_llen) 152338629291SGao Xiang return; 152438629291SGao Xiang } 152538629291SGao Xiang 152638629291SGao Xiang cur = map->m_la + map->m_llen - 1; 152738629291SGao Xiang while (cur >= end) { 152838629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 152938629291SGao Xiang struct page *page; 153038629291SGao Xiang 153138629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1532aa793b46SGao Xiang if (page) { 153338629291SGao Xiang if (PageUptodate(page)) { 153438629291SGao Xiang unlock_page(page); 1535aa793b46SGao Xiang } else { 153638629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 153738629291SGao Xiang if (err) 153838629291SGao Xiang erofs_err(inode->i_sb, 153938629291SGao Xiang "readmore error at page %lu @ nid %llu", 154038629291SGao Xiang index, EROFS_I(inode)->nid); 1541aa793b46SGao Xiang } 154238629291SGao Xiang put_page(page); 1543aa793b46SGao Xiang } 1544aa793b46SGao Xiang 154538629291SGao Xiang if (cur < PAGE_SIZE) 154638629291SGao Xiang break; 154738629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 154838629291SGao Xiang } 154938629291SGao Xiang } 155038629291SGao Xiang 1551a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 155247e4937aSGao Xiang { 1553a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 155447e4937aSGao Xiang struct inode *const inode = page->mapping->host; 155540452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 155647e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1557eaa9172aSGao Xiang struct page *pagepool = NULL; 155847e4937aSGao Xiang int err; 155947e4937aSGao Xiang 156047e4937aSGao Xiang trace_erofs_readpage(page, false); 156147e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 156247e4937aSGao Xiang 156338629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 156438629291SGao Xiang &pagepool, true); 15651825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 156638629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 156738629291SGao Xiang 15685c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 156947e4937aSGao Xiang 157047e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 157183a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 157240452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 157347e4937aSGao Xiang 157447e4937aSGao Xiang if (err) 15754f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 157647e4937aSGao Xiang 157709c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1578eaa9172aSGao Xiang erofs_release_pages(&pagepool); 157947e4937aSGao Xiang return err; 158047e4937aSGao Xiang } 158147e4937aSGao Xiang 15820615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 158347e4937aSGao Xiang { 15840615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 158547e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 158647e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1587eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 158838629291SGao Xiang unsigned int nr_pages; 158947e4937aSGao Xiang 15906ea5aad3SGao Xiang f.readahead = true; 15910615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 159247e4937aSGao Xiang 159338629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 159438629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 159538629291SGao Xiang nr_pages = readahead_count(rac); 159638629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 159738629291SGao Xiang 15980615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 159947e4937aSGao Xiang set_page_private(page, (unsigned long)head); 160047e4937aSGao Xiang head = page; 160147e4937aSGao Xiang } 160247e4937aSGao Xiang 160347e4937aSGao Xiang while (head) { 160447e4937aSGao Xiang struct page *page = head; 160547e4937aSGao Xiang int err; 160647e4937aSGao Xiang 160747e4937aSGao Xiang /* traversal in reverse order */ 160847e4937aSGao Xiang head = (void *)page_private(page); 160947e4937aSGao Xiang 16101825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1611a5876e24SGao Xiang if (err) 16124f761fa2SGao Xiang erofs_err(inode->i_sb, 16134f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 16144f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 161547e4937aSGao Xiang put_page(page); 161647e4937aSGao Xiang } 161738629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 16185c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 161947e4937aSGao Xiang 162083a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 162140452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 162209c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1623eaa9172aSGao Xiang erofs_release_pages(&pagepool); 162447e4937aSGao Xiang } 162547e4937aSGao Xiang 16260c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1627a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 16280615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 162947e4937aSGao Xiang }; 1630