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 30387bab87SGao Xiang /* (obsoleted) page type for online pages */ 31387bab87SGao Xiang enum z_erofs_page_type { 32387bab87SGao Xiang /* including Z_EROFS_VLE_PAGE_TAIL_EXCLUSIVE */ 33387bab87SGao Xiang Z_EROFS_PAGE_TYPE_EXCLUSIVE, 34387bab87SGao Xiang 35387bab87SGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED, 36387bab87SGao Xiang 37387bab87SGao Xiang Z_EROFS_VLE_PAGE_TYPE_HEAD, 38387bab87SGao Xiang Z_EROFS_VLE_PAGE_TYPE_MAX 39387bab87SGao Xiang }; 40387bab87SGao 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) { 137ed722fbcSGao Xiang size = struct_size(a, compressed_bvecs, 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 262387bab87SGao 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; 29047e4937aSGao Xiang z_erofs_next_pcluster_t owned_head; 29147e4937aSGao Xiang enum z_erofs_collectmode mode; 29247e4937aSGao Xiang 2936ea5aad3SGao Xiang bool readahead; 29447e4937aSGao Xiang /* used for applying cache strategy on the fly */ 29547e4937aSGao Xiang bool backmost; 29647e4937aSGao Xiang erofs_off_t headoffset; 297ed722fbcSGao Xiang 298ed722fbcSGao Xiang /* a pointer used to pick up inplace I/O pages */ 299ed722fbcSGao Xiang unsigned int icur; 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; 322ed722fbcSGao Xiang unsigned int i; 32347e4937aSGao Xiang 3245c6dcc57SGao Xiang if (fe->mode < COLLECT_PRIMARY_FOLLOWED) 32547e4937aSGao Xiang return; 32647e4937aSGao Xiang 327ed722fbcSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 32847e4937aSGao Xiang struct page *page; 32947e4937aSGao Xiang compressed_page_t t; 3301825c8d7SGao Xiang struct page *newpage = NULL; 33147e4937aSGao Xiang 33247e4937aSGao Xiang /* the compressed page was loaded before */ 333ed722fbcSGao Xiang if (READ_ONCE(pcl->compressed_bvecs[i].page)) 33447e4937aSGao Xiang continue; 33547e4937aSGao Xiang 336ed722fbcSGao Xiang page = find_get_page(mc, pcl->obj.index + i); 33747e4937aSGao Xiang 33847e4937aSGao Xiang if (page) { 33947e4937aSGao Xiang t = tag_compressed_page_justfound(page); 3400b964600SGao Xiang } else { 3410b964600SGao Xiang /* I/O is needed, no possible to decompress directly */ 3420b964600SGao Xiang standalone = false; 3430b964600SGao Xiang switch (type) { 3440b964600SGao Xiang case TRYALLOC: 3451825c8d7SGao Xiang newpage = erofs_allocpage(pagepool, gfp); 3461825c8d7SGao Xiang if (!newpage) 34747e4937aSGao Xiang continue; 3480b964600SGao Xiang set_page_private(newpage, 3490b964600SGao Xiang Z_EROFS_PREALLOCATED_PAGE); 3500b964600SGao Xiang t = tag_compressed_page_justfound(newpage); 3510b964600SGao Xiang break; 3520b964600SGao Xiang default: /* DONTALLOC */ 3530b964600SGao Xiang continue; 3540b964600SGao Xiang } 35547e4937aSGao Xiang } 35647e4937aSGao Xiang 357ed722fbcSGao Xiang if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, 358ed722fbcSGao Xiang tagptr_cast_ptr(t))) 35947e4937aSGao Xiang continue; 36047e4937aSGao Xiang 361eaa9172aSGao Xiang if (page) 36247e4937aSGao Xiang put_page(page); 363eaa9172aSGao Xiang else if (newpage) 364eaa9172aSGao Xiang erofs_pagepool_add(pagepool, newpage); 36547e4937aSGao Xiang } 36647e4937aSGao Xiang 3670b964600SGao Xiang /* 3680b964600SGao Xiang * don't do inplace I/O if all compressed pages are available in 3690b964600SGao Xiang * managed cache since it can be moved to the bypass queue instead. 3700b964600SGao Xiang */ 3710b964600SGao Xiang if (standalone) 3725c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 37347e4937aSGao Xiang } 37447e4937aSGao Xiang 37547e4937aSGao Xiang /* called by erofs_shrinker to get rid of all compressed_pages */ 37647e4937aSGao Xiang int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 37747e4937aSGao Xiang struct erofs_workgroup *grp) 37847e4937aSGao Xiang { 37947e4937aSGao Xiang struct z_erofs_pcluster *const pcl = 38047e4937aSGao Xiang container_of(grp, struct z_erofs_pcluster, obj); 38147e4937aSGao Xiang int i; 38247e4937aSGao Xiang 383cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 38447e4937aSGao Xiang /* 38547e4937aSGao Xiang * refcount of workgroup is now freezed as 1, 38647e4937aSGao Xiang * therefore no need to worry about available decompression users. 38747e4937aSGao Xiang */ 3889f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 389ed722fbcSGao Xiang struct page *page = pcl->compressed_bvecs[i].page; 39047e4937aSGao Xiang 39147e4937aSGao Xiang if (!page) 39247e4937aSGao Xiang continue; 39347e4937aSGao Xiang 39447e4937aSGao Xiang /* block other users from reclaiming or migrating the page */ 39547e4937aSGao Xiang if (!trylock_page(page)) 39647e4937aSGao Xiang return -EBUSY; 39747e4937aSGao Xiang 398f4d4e5fcSYue Hu if (!erofs_page_is_managed(sbi, page)) 39947e4937aSGao Xiang continue; 40047e4937aSGao Xiang 40147e4937aSGao Xiang /* barrier is implied in the following 'unlock_page' */ 402ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 4036aaa7b06SGao Xiang detach_page_private(page); 40447e4937aSGao Xiang unlock_page(page); 40547e4937aSGao Xiang } 40647e4937aSGao Xiang return 0; 40747e4937aSGao Xiang } 40847e4937aSGao Xiang 409d252ff3dSYue Hu int erofs_try_to_free_cached_page(struct page *page) 41047e4937aSGao Xiang { 41147e4937aSGao Xiang struct z_erofs_pcluster *const pcl = (void *)page_private(page); 412ed722fbcSGao Xiang int ret, i; 41347e4937aSGao Xiang 414ed722fbcSGao Xiang if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 415ed722fbcSGao Xiang return 0; 41647e4937aSGao Xiang 417ed722fbcSGao Xiang ret = 0; 418cecf864dSYue Hu DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 4199f6cc76eSGao Xiang for (i = 0; i < pcl->pclusterpages; ++i) { 420ed722fbcSGao Xiang if (pcl->compressed_bvecs[i].page == page) { 421ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 42247e4937aSGao Xiang ret = 1; 42347e4937aSGao Xiang break; 42447e4937aSGao Xiang } 42547e4937aSGao Xiang } 42647e4937aSGao Xiang erofs_workgroup_unfreeze(&pcl->obj, 1); 4276aaa7b06SGao Xiang if (ret) 4286aaa7b06SGao Xiang detach_page_private(page); 42947e4937aSGao Xiang return ret; 43047e4937aSGao Xiang } 43147e4937aSGao Xiang 43247e4937aSGao Xiang /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 4335c6dcc57SGao Xiang static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 434ed722fbcSGao Xiang struct z_erofs_bvec *bvec) 43547e4937aSGao Xiang { 4365c6dcc57SGao Xiang struct z_erofs_pcluster *const pcl = fe->pcl; 43747e4937aSGao Xiang 438ed722fbcSGao Xiang while (fe->icur > 0) { 439ed722fbcSGao Xiang if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 440ed722fbcSGao Xiang NULL, bvec->page)) { 441ed722fbcSGao Xiang pcl->compressed_bvecs[fe->icur] = *bvec; 44247e4937aSGao Xiang return true; 443ed722fbcSGao Xiang } 444ed722fbcSGao Xiang } 44547e4937aSGao Xiang return false; 44647e4937aSGao Xiang } 44747e4937aSGao Xiang 44887ca34a7SGao Xiang /* callers must be with pcluster lock held */ 4495c6dcc57SGao Xiang static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 45006a304cdSGao Xiang struct z_erofs_bvec *bvec, 45106a304cdSGao Xiang enum z_erofs_page_type type) 45247e4937aSGao Xiang { 45347e4937aSGao Xiang int ret; 45447e4937aSGao Xiang 4555c6dcc57SGao Xiang if (fe->mode >= COLLECT_PRIMARY && 45606a304cdSGao Xiang type == Z_EROFS_PAGE_TYPE_EXCLUSIVE) { 45706a304cdSGao Xiang /* give priority for inplaceio to use file pages first */ 458ed722fbcSGao Xiang if (z_erofs_try_inplace_io(fe, bvec)) 45947e4937aSGao Xiang return 0; 46006a304cdSGao Xiang /* otherwise, check if it can be used as a bvpage */ 46106a304cdSGao Xiang if (fe->mode >= COLLECT_PRIMARY_FOLLOWED && 46206a304cdSGao Xiang !fe->candidate_bvpage) 46306a304cdSGao Xiang fe->candidate_bvpage = bvec->page; 46406a304cdSGao Xiang } 46506a304cdSGao Xiang ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 46606a304cdSGao Xiang fe->pcl->vcnt += (ret >= 0); 46706a304cdSGao Xiang return ret; 46847e4937aSGao Xiang } 46947e4937aSGao Xiang 4705c6dcc57SGao Xiang static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 47147e4937aSGao Xiang { 4725c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = f->pcl; 4735c6dcc57SGao Xiang z_erofs_next_pcluster_t *owned_head = &f->owned_head; 47447e4937aSGao Xiang 475473e15b0SGao Xiang /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 476473e15b0SGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 477473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_NIL) { 47847e4937aSGao Xiang *owned_head = &pcl->next; 479473e15b0SGao Xiang /* so we can attach this pcluster to our submission chain. */ 4805c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_FOLLOWED; 481473e15b0SGao Xiang return; 482473e15b0SGao Xiang } 483473e15b0SGao Xiang 48447e4937aSGao Xiang /* 485473e15b0SGao Xiang * type 2, link to the end of an existing open chain, be careful 486473e15b0SGao Xiang * that its submission is controlled by the original attached chain. 48747e4937aSGao Xiang */ 48847e4937aSGao Xiang if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 489473e15b0SGao Xiang *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 49047e4937aSGao Xiang *owned_head = Z_EROFS_PCLUSTER_TAIL; 4915c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY_HOOKED; 4925c6dcc57SGao Xiang f->tailpcl = NULL; 493473e15b0SGao Xiang return; 49447e4937aSGao Xiang } 495473e15b0SGao Xiang /* type 3, it belongs to a chain, but it isn't the end of the chain */ 4965c6dcc57SGao Xiang f->mode = COLLECT_PRIMARY; 49747e4937aSGao Xiang } 49847e4937aSGao Xiang 49983a386c0SGao Xiang static int z_erofs_lookup_pcluster(struct z_erofs_decompress_frontend *fe) 50047e4937aSGao Xiang { 50183a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 5025c6dcc57SGao Xiang struct z_erofs_pcluster *pcl = fe->pcl; 50347e4937aSGao Xiang unsigned int length; 50447e4937aSGao Xiang 50564094a04SGao Xiang /* to avoid unexpected loop formed by corrupted images */ 5065c6dcc57SGao Xiang if (fe->owned_head == &pcl->next || pcl == fe->tailpcl) { 50747e4937aSGao Xiang DBG_BUGON(1); 5089e579fc1SGao Xiang return -EFSCORRUPTED; 50947e4937aSGao Xiang } 51047e4937aSGao Xiang 51187ca34a7SGao Xiang if (pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) { 51247e4937aSGao Xiang DBG_BUGON(1); 5139e579fc1SGao Xiang return -EFSCORRUPTED; 51447e4937aSGao Xiang } 51547e4937aSGao Xiang 51647e4937aSGao Xiang length = READ_ONCE(pcl->length); 51747e4937aSGao Xiang if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 51847e4937aSGao Xiang if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 51947e4937aSGao Xiang DBG_BUGON(1); 5209e579fc1SGao Xiang return -EFSCORRUPTED; 52147e4937aSGao Xiang } 52247e4937aSGao Xiang } else { 52347e4937aSGao Xiang unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 52447e4937aSGao Xiang 52547e4937aSGao Xiang if (map->m_flags & EROFS_MAP_FULL_MAPPED) 52647e4937aSGao Xiang llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 52747e4937aSGao Xiang 52847e4937aSGao Xiang while (llen > length && 52947e4937aSGao Xiang length != cmpxchg_relaxed(&pcl->length, length, llen)) { 53047e4937aSGao Xiang cpu_relax(); 53147e4937aSGao Xiang length = READ_ONCE(pcl->length); 53247e4937aSGao Xiang } 53347e4937aSGao Xiang } 53487ca34a7SGao Xiang mutex_lock(&pcl->lock); 53547e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 5365c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 5375c6dcc57SGao Xiang fe->tailpcl = pcl; 538473e15b0SGao Xiang 5395c6dcc57SGao Xiang z_erofs_try_to_claim_pcluster(fe); 5409e579fc1SGao Xiang return 0; 54147e4937aSGao Xiang } 54247e4937aSGao Xiang 54383a386c0SGao Xiang static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 54447e4937aSGao Xiang { 54583a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 546cecf864dSYue Hu bool ztailpacking = map->m_flags & EROFS_MAP_META; 54747e4937aSGao Xiang struct z_erofs_pcluster *pcl; 54864094a04SGao Xiang struct erofs_workgroup *grp; 54947e4937aSGao Xiang int err; 55047e4937aSGao Xiang 5518f899262SGao Xiang if (!(map->m_flags & EROFS_MAP_ENCODED)) { 5528f899262SGao Xiang DBG_BUGON(1); 5538f899262SGao Xiang return -EFSCORRUPTED; 5548f899262SGao Xiang } 5558f899262SGao Xiang 5569f6cc76eSGao Xiang /* no available pcluster, let's allocate one */ 557cecf864dSYue Hu pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 558cecf864dSYue Hu map->m_plen >> PAGE_SHIFT); 5599f6cc76eSGao Xiang if (IS_ERR(pcl)) 5609f6cc76eSGao Xiang return PTR_ERR(pcl); 56147e4937aSGao Xiang 56264094a04SGao Xiang atomic_set(&pcl->obj.refcount, 1); 5638f899262SGao Xiang pcl->algorithmformat = map->m_algorithmformat; 56447e4937aSGao Xiang pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 56547e4937aSGao Xiang (map->m_flags & EROFS_MAP_FULL_MAPPED ? 56647e4937aSGao Xiang Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 56747e4937aSGao Xiang 56847e4937aSGao Xiang /* new pclusters should be claimed as type 1, primary and followed */ 5695c6dcc57SGao Xiang pcl->next = fe->owned_head; 57087ca34a7SGao Xiang pcl->pageofs_out = map->m_la & ~PAGE_MASK; 5715c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED; 57247e4937aSGao Xiang 57347e4937aSGao Xiang /* 57447e4937aSGao Xiang * lock all primary followed works before visible to others 57547e4937aSGao Xiang * and mutex_trylock *never* fails for a new pcluster. 57647e4937aSGao Xiang */ 57787ca34a7SGao Xiang mutex_init(&pcl->lock); 57887ca34a7SGao Xiang DBG_BUGON(!mutex_trylock(&pcl->lock)); 57947e4937aSGao Xiang 580cecf864dSYue Hu if (ztailpacking) { 581cecf864dSYue Hu pcl->obj.index = 0; /* which indicates ztailpacking */ 582cecf864dSYue Hu pcl->pageofs_in = erofs_blkoff(map->m_pa); 583cecf864dSYue Hu pcl->tailpacking_size = map->m_plen; 584cecf864dSYue Hu } else { 585cecf864dSYue Hu pcl->obj.index = map->m_pa >> PAGE_SHIFT; 586cecf864dSYue Hu 58783a386c0SGao Xiang grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 58864094a04SGao Xiang if (IS_ERR(grp)) { 58964094a04SGao Xiang err = PTR_ERR(grp); 59064094a04SGao Xiang goto err_out; 59164094a04SGao Xiang } 59264094a04SGao Xiang 59364094a04SGao Xiang if (grp != &pcl->obj) { 5945c6dcc57SGao Xiang fe->pcl = container_of(grp, 595cecf864dSYue Hu struct z_erofs_pcluster, obj); 59664094a04SGao Xiang err = -EEXIST; 59764094a04SGao Xiang goto err_out; 59847e4937aSGao Xiang } 599cecf864dSYue Hu } 60047e4937aSGao Xiang /* used to check tail merging loop due to corrupted images */ 6015c6dcc57SGao Xiang if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 6025c6dcc57SGao Xiang fe->tailpcl = pcl; 6035c6dcc57SGao Xiang fe->owned_head = &pcl->next; 6045c6dcc57SGao Xiang fe->pcl = pcl; 6059e579fc1SGao Xiang return 0; 60664094a04SGao Xiang 60764094a04SGao Xiang err_out: 60887ca34a7SGao Xiang mutex_unlock(&pcl->lock); 6099f6cc76eSGao Xiang z_erofs_free_pcluster(pcl); 61064094a04SGao Xiang return err; 61147e4937aSGao Xiang } 61247e4937aSGao Xiang 61383a386c0SGao Xiang static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 61447e4937aSGao Xiang { 61583a386c0SGao Xiang struct erofs_map_blocks *map = &fe->map; 6160d823b42SGao Xiang struct erofs_workgroup *grp = NULL; 6179e579fc1SGao Xiang int ret; 61847e4937aSGao Xiang 61987ca34a7SGao Xiang DBG_BUGON(fe->pcl); 62047e4937aSGao Xiang 62187ca34a7SGao Xiang /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 6225c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 6235c6dcc57SGao Xiang DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 62447e4937aSGao Xiang 6250d823b42SGao Xiang if (!(map->m_flags & EROFS_MAP_META)) { 6260d823b42SGao Xiang grp = erofs_find_workgroup(fe->inode->i_sb, 6270d823b42SGao Xiang map->m_pa >> PAGE_SHIFT); 6280d823b42SGao Xiang } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 62947e4937aSGao Xiang DBG_BUGON(1); 630cecf864dSYue Hu return -EFSCORRUPTED; 631cecf864dSYue Hu } 63247e4937aSGao Xiang 63364094a04SGao Xiang if (grp) { 6345c6dcc57SGao Xiang fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 6350d823b42SGao Xiang ret = -EEXIST; 63664094a04SGao Xiang } else { 63783a386c0SGao Xiang ret = z_erofs_register_pcluster(fe); 63864094a04SGao Xiang } 63947e4937aSGao Xiang 6400d823b42SGao Xiang if (ret == -EEXIST) { 64183a386c0SGao Xiang ret = z_erofs_lookup_pcluster(fe); 64264094a04SGao Xiang if (ret) { 6435c6dcc57SGao Xiang erofs_workgroup_put(&fe->pcl->obj); 64464094a04SGao Xiang return ret; 64564094a04SGao Xiang } 6460d823b42SGao Xiang } else if (ret) { 6470d823b42SGao Xiang return ret; 6480d823b42SGao Xiang } 64906a304cdSGao Xiang z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 650387bab87SGao Xiang Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 65181382f5fSGao Xiang /* since file-backed online pages are traversed in reverse order */ 652ed722fbcSGao Xiang fe->icur = 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) 746*67148551SGao Xiang goto 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) 758*67148551SGao Xiang goto 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); 769*67148551SGao Xiang goto out; 770cecf864dSYue Hu } 77109c54379SGao Xiang get_page(fe->map.buf.page); 772ed722fbcSGao Xiang WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 773ed722fbcSGao Xiang fe->map.buf.page); 7745c6dcc57SGao Xiang fe->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 775cecf864dSYue Hu } else { 7766f39d1e1SGao Xiang /* bind cache first when cached decompression is preferred */ 777cecf864dSYue Hu if (should_alloc_managed_pages(fe, sbi->opt.cache_strategy, 778cecf864dSYue Hu map->m_la)) 7791825c8d7SGao Xiang cache_strategy = TRYALLOC; 78047e4937aSGao Xiang else 78147e4937aSGao Xiang cache_strategy = DONTALLOC; 78247e4937aSGao Xiang 7836f39d1e1SGao Xiang z_erofs_bind_cache(fe, cache_strategy, pagepool); 784cecf864dSYue Hu } 78547e4937aSGao Xiang hitted: 786dc76ea8cSGao Xiang /* 787dc76ea8cSGao Xiang * Ensure the current partial page belongs to this submit chain rather 788dc76ea8cSGao Xiang * than other concurrent submit chains or the noio(bypass) chain since 789dc76ea8cSGao Xiang * those chains are handled asynchronously thus the page cannot be used 790387bab87SGao Xiang * for inplace I/O or bvpage (should be processed in a strict order.) 791dc76ea8cSGao Xiang */ 7925c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_HOOKED && 7935c6dcc57SGao Xiang fe->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 794dc76ea8cSGao Xiang 79547e4937aSGao Xiang cur = end - min_t(unsigned int, offset + end - map->m_la, end); 7968d8a09b0SGao Xiang if (!(map->m_flags & EROFS_MAP_MAPPED)) { 79747e4937aSGao Xiang zero_user_segment(page, cur, end); 79847e4937aSGao Xiang goto next_part; 79947e4937aSGao Xiang } 80047e4937aSGao Xiang 80147e4937aSGao Xiang /* let's derive page type */ 80247e4937aSGao Xiang page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 80347e4937aSGao Xiang (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 80447e4937aSGao Xiang (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 80547e4937aSGao Xiang Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 80647e4937aSGao Xiang 80747e4937aSGao Xiang if (cur) 8085c6dcc57SGao Xiang tight &= (fe->mode >= COLLECT_PRIMARY_FOLLOWED); 80947e4937aSGao Xiang 81047e4937aSGao Xiang retry: 81106a304cdSGao Xiang err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 81206a304cdSGao Xiang .page = page, 81306a304cdSGao Xiang .offset = offset - map->m_la, 81406a304cdSGao Xiang .end = end, 81506a304cdSGao Xiang }), page_type); 81606a304cdSGao Xiang /* should allocate an additional short-lived page for bvset */ 81706a304cdSGao Xiang if (err == -EAGAIN && !fe->candidate_bvpage) { 81806a304cdSGao Xiang fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL); 81906a304cdSGao Xiang set_page_private(fe->candidate_bvpage, 82006a304cdSGao Xiang Z_EROFS_SHORTLIVED_PAGE); 82147e4937aSGao Xiang goto retry; 82247e4937aSGao Xiang } 82347e4937aSGao Xiang 82406a304cdSGao Xiang if (err) { 82506a304cdSGao Xiang DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage); 826*67148551SGao Xiang goto out; 82706a304cdSGao Xiang } 82847e4937aSGao Xiang 829*67148551SGao Xiang z_erofs_onlinepage_split(page); 83047e4937aSGao Xiang /* bump up the number of spiltted parts of a page */ 83147e4937aSGao Xiang ++spiltted; 832*67148551SGao Xiang 83347e4937aSGao Xiang /* also update nr_pages */ 834*67148551SGao Xiang index = page->index - (map->m_la >> PAGE_SHIFT); 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: 845*67148551SGao Xiang if (err) 846*67148551SGao Xiang z_erofs_page_mark_eio(page); 84747e4937aSGao Xiang z_erofs_onlinepage_endio(page); 84847e4937aSGao Xiang 8494f761fa2SGao Xiang erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 85047e4937aSGao Xiang __func__, page, spiltted, map->m_llen); 85147e4937aSGao Xiang return err; 85247e4937aSGao Xiang } 85347e4937aSGao Xiang 85440452ffcSHuang Jianan static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi, 85540452ffcSHuang Jianan unsigned int readahead_pages) 85640452ffcSHuang Jianan { 857a2e20a25SMatthew Wilcox (Oracle) /* auto: enable for read_folio, disable for readahead */ 85840452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 85940452ffcSHuang Jianan !readahead_pages) 86040452ffcSHuang Jianan return true; 86140452ffcSHuang Jianan 86240452ffcSHuang Jianan if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 86340452ffcSHuang Jianan (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 86440452ffcSHuang Jianan return true; 86540452ffcSHuang Jianan 86640452ffcSHuang Jianan return false; 86740452ffcSHuang Jianan } 86840452ffcSHuang Jianan 8696aaa7b06SGao Xiang static bool z_erofs_page_is_invalidated(struct page *page) 8706aaa7b06SGao Xiang { 8716aaa7b06SGao Xiang return !page->mapping && !z_erofs_is_shortlived_page(page); 8726aaa7b06SGao Xiang } 8736aaa7b06SGao Xiang 87442fec235SGao Xiang static int z_erofs_parse_out_bvecs(struct z_erofs_pcluster *pcl, 87542fec235SGao Xiang struct page **pages, struct page **pagepool) 87642fec235SGao Xiang { 87706a304cdSGao Xiang struct z_erofs_bvec_iter biter; 87806a304cdSGao Xiang struct page *old_bvpage; 87942fec235SGao Xiang int i, err = 0; 88042fec235SGao Xiang 881387bab87SGao Xiang z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 88242fec235SGao Xiang for (i = 0; i < pcl->vcnt; ++i) { 88306a304cdSGao Xiang struct z_erofs_bvec bvec; 88442fec235SGao Xiang unsigned int pagenr; 88542fec235SGao Xiang 88606a304cdSGao Xiang z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 88742fec235SGao Xiang 88806a304cdSGao Xiang if (old_bvpage) 88906a304cdSGao Xiang z_erofs_put_shortlivedpage(pagepool, old_bvpage); 89042fec235SGao Xiang 89106a304cdSGao Xiang pagenr = (bvec.offset + pcl->pageofs_out) >> PAGE_SHIFT; 89242fec235SGao Xiang DBG_BUGON(pagenr >= pcl->nr_pages); 89306a304cdSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 89442fec235SGao Xiang /* 89542fec235SGao Xiang * currently EROFS doesn't support multiref(dedup), 89642fec235SGao Xiang * so here erroring out one multiref page. 89742fec235SGao Xiang */ 89842fec235SGao Xiang if (pages[pagenr]) { 89942fec235SGao Xiang DBG_BUGON(1); 900*67148551SGao Xiang z_erofs_page_mark_eio(pages[pagenr]); 90142fec235SGao Xiang z_erofs_onlinepage_endio(pages[pagenr]); 90242fec235SGao Xiang err = -EFSCORRUPTED; 90342fec235SGao Xiang } 90406a304cdSGao Xiang pages[pagenr] = bvec.page; 90542fec235SGao Xiang } 90606a304cdSGao Xiang 90706a304cdSGao Xiang old_bvpage = z_erofs_bvec_iter_end(&biter); 90806a304cdSGao Xiang if (old_bvpage) 90906a304cdSGao Xiang z_erofs_put_shortlivedpage(pagepool, old_bvpage); 91042fec235SGao Xiang return err; 91142fec235SGao Xiang } 91242fec235SGao Xiang 91367139e36SGao Xiang static struct page **z_erofs_parse_in_bvecs(struct erofs_sb_info *sbi, 91467139e36SGao Xiang struct z_erofs_pcluster *pcl, struct page **pages, 91567139e36SGao Xiang struct page **pagepool, bool *overlapped) 91667139e36SGao Xiang { 91767139e36SGao Xiang unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 91867139e36SGao Xiang struct page **compressed_pages; 91967139e36SGao Xiang int i, err = 0; 92067139e36SGao Xiang 92167139e36SGao Xiang /* XXX: will have a better approach in the following commits */ 92267139e36SGao Xiang compressed_pages = kmalloc_array(pclusterpages, sizeof(struct page *), 92367139e36SGao Xiang GFP_KERNEL | __GFP_NOFAIL); 92467139e36SGao Xiang *overlapped = false; 92567139e36SGao Xiang 92667139e36SGao Xiang for (i = 0; i < pclusterpages; ++i) { 927ed722fbcSGao Xiang struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 928ed722fbcSGao Xiang struct page *page = bvec->page; 929ed722fbcSGao Xiang unsigned int pgnr; 93067139e36SGao Xiang 93167139e36SGao Xiang /* compressed pages ought to be present before decompressing */ 93267139e36SGao Xiang if (!page) { 93367139e36SGao Xiang DBG_BUGON(1); 93467139e36SGao Xiang continue; 93567139e36SGao Xiang } 93667139e36SGao Xiang compressed_pages[i] = page; 93767139e36SGao Xiang 93867139e36SGao Xiang if (z_erofs_is_inline_pcluster(pcl)) { 93967139e36SGao Xiang if (!PageUptodate(page)) 94067139e36SGao Xiang err = -EIO; 94167139e36SGao Xiang continue; 94267139e36SGao Xiang } 94367139e36SGao Xiang 94467139e36SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 94567139e36SGao Xiang if (!z_erofs_is_shortlived_page(page)) { 94667139e36SGao Xiang if (erofs_page_is_managed(sbi, page)) { 94767139e36SGao Xiang if (!PageUptodate(page)) 94867139e36SGao Xiang err = -EIO; 94967139e36SGao Xiang continue; 95067139e36SGao Xiang } 95167139e36SGao Xiang 952ed722fbcSGao Xiang pgnr = (bvec->offset + pcl->pageofs_out) >> PAGE_SHIFT; 953ed722fbcSGao Xiang DBG_BUGON(pgnr >= pcl->nr_pages); 954ed722fbcSGao Xiang if (pages[pgnr]) { 95567139e36SGao Xiang DBG_BUGON(1); 956*67148551SGao Xiang z_erofs_page_mark_eio(pages[pgnr]); 957ed722fbcSGao Xiang z_erofs_onlinepage_endio(pages[pgnr]); 95867139e36SGao Xiang err = -EFSCORRUPTED; 95967139e36SGao Xiang } 960ed722fbcSGao Xiang pages[pgnr] = page; 96167139e36SGao Xiang *overlapped = true; 96267139e36SGao Xiang } 96367139e36SGao Xiang } 96467139e36SGao Xiang 96567139e36SGao Xiang if (err) { 96667139e36SGao Xiang kfree(compressed_pages); 96767139e36SGao Xiang return ERR_PTR(err); 96867139e36SGao Xiang } 96967139e36SGao Xiang return compressed_pages; 97067139e36SGao Xiang } 97167139e36SGao Xiang 97247e4937aSGao Xiang static int z_erofs_decompress_pcluster(struct super_block *sb, 97347e4937aSGao Xiang struct z_erofs_pcluster *pcl, 974*67148551SGao Xiang struct page **pagepool, int err) 97547e4937aSGao Xiang { 97647e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(sb); 977cecf864dSYue Hu unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 9789f6cc76eSGao Xiang unsigned int i, inputsize, outputsize, llen, nr_pages; 97947e4937aSGao Xiang struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 98047e4937aSGao Xiang struct page **pages, **compressed_pages, *page; 981*67148551SGao Xiang int err2; 98247e4937aSGao Xiang bool overlapped, partial; 98347e4937aSGao Xiang 98447e4937aSGao Xiang might_sleep(); 98587ca34a7SGao Xiang DBG_BUGON(!READ_ONCE(pcl->nr_pages)); 98647e4937aSGao Xiang 98787ca34a7SGao Xiang mutex_lock(&pcl->lock); 98887ca34a7SGao Xiang nr_pages = pcl->nr_pages; 98947e4937aSGao Xiang 9908d8a09b0SGao Xiang if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 99147e4937aSGao Xiang pages = pages_onstack; 99247e4937aSGao Xiang } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 99347e4937aSGao Xiang mutex_trylock(&z_pagemap_global_lock)) { 99447e4937aSGao Xiang pages = z_pagemap_global; 99547e4937aSGao Xiang } else { 99647e4937aSGao Xiang gfp_t gfp_flags = GFP_KERNEL; 99747e4937aSGao Xiang 99847e4937aSGao Xiang if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 99947e4937aSGao Xiang gfp_flags |= __GFP_NOFAIL; 100047e4937aSGao Xiang 100147e4937aSGao Xiang pages = kvmalloc_array(nr_pages, sizeof(struct page *), 100247e4937aSGao Xiang gfp_flags); 100347e4937aSGao Xiang 100447e4937aSGao Xiang /* fallback to global pagemap for the lowmem scenario */ 10058d8a09b0SGao Xiang if (!pages) { 100647e4937aSGao Xiang mutex_lock(&z_pagemap_global_lock); 100747e4937aSGao Xiang pages = z_pagemap_global; 100847e4937aSGao Xiang } 100947e4937aSGao Xiang } 101047e4937aSGao Xiang 101147e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) 101247e4937aSGao Xiang pages[i] = NULL; 101347e4937aSGao Xiang 1014*67148551SGao Xiang err2 = z_erofs_parse_out_bvecs(pcl, pages, pagepool); 1015*67148551SGao Xiang if (err2) 1016*67148551SGao Xiang err = err2; 101767139e36SGao Xiang compressed_pages = z_erofs_parse_in_bvecs(sbi, pcl, pages, 101867139e36SGao Xiang pagepool, &overlapped); 101967139e36SGao Xiang if (IS_ERR(compressed_pages)) { 102067139e36SGao Xiang err = PTR_ERR(compressed_pages); 102167139e36SGao Xiang compressed_pages = NULL; 102247e4937aSGao Xiang } 102347e4937aSGao Xiang 10248d8a09b0SGao Xiang if (err) 102547e4937aSGao Xiang goto out; 102647e4937aSGao Xiang 102747e4937aSGao Xiang llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 102887ca34a7SGao Xiang if (nr_pages << PAGE_SHIFT >= pcl->pageofs_out + llen) { 102947e4937aSGao Xiang outputsize = llen; 103047e4937aSGao Xiang partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 103147e4937aSGao Xiang } else { 103287ca34a7SGao Xiang outputsize = (nr_pages << PAGE_SHIFT) - pcl->pageofs_out; 103347e4937aSGao Xiang partial = true; 103447e4937aSGao Xiang } 103547e4937aSGao Xiang 1036cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) 1037cecf864dSYue Hu inputsize = pcl->tailpacking_size; 1038cecf864dSYue Hu else 1039cecf864dSYue Hu inputsize = pclusterpages * PAGE_SIZE; 1040cecf864dSYue Hu 104147e4937aSGao Xiang err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 104247e4937aSGao Xiang .sb = sb, 104347e4937aSGao Xiang .in = compressed_pages, 104447e4937aSGao Xiang .out = pages, 1045cecf864dSYue Hu .pageofs_in = pcl->pageofs_in, 104687ca34a7SGao Xiang .pageofs_out = pcl->pageofs_out, 10479f6cc76eSGao Xiang .inputsize = inputsize, 104847e4937aSGao Xiang .outputsize = outputsize, 104947e4937aSGao Xiang .alg = pcl->algorithmformat, 105047e4937aSGao Xiang .inplace_io = overlapped, 105147e4937aSGao Xiang .partial_decoding = partial 105247e4937aSGao Xiang }, pagepool); 105347e4937aSGao Xiang 105447e4937aSGao Xiang out: 1055cecf864dSYue Hu /* must handle all compressed pages before actual file pages */ 1056cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1057ed722fbcSGao Xiang page = pcl->compressed_bvecs[0].page; 1058ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1059cecf864dSYue Hu put_page(page); 1060cecf864dSYue Hu } else { 1061cecf864dSYue Hu for (i = 0; i < pclusterpages; ++i) { 1062ed722fbcSGao Xiang page = pcl->compressed_bvecs[i].page; 106347e4937aSGao Xiang 106447e4937aSGao Xiang if (erofs_page_is_managed(sbi, page)) 106547e4937aSGao Xiang continue; 106647e4937aSGao Xiang 10676aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10686aaa7b06SGao Xiang (void)z_erofs_put_shortlivedpage(pagepool, page); 1069ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 107047e4937aSGao Xiang } 1071cecf864dSYue Hu } 107267139e36SGao Xiang kfree(compressed_pages); 107347e4937aSGao Xiang 107447e4937aSGao Xiang for (i = 0; i < nr_pages; ++i) { 107547e4937aSGao Xiang page = pages[i]; 107647e4937aSGao Xiang if (!page) 107747e4937aSGao Xiang continue; 107847e4937aSGao Xiang 10796aaa7b06SGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 108047e4937aSGao Xiang 10816aaa7b06SGao Xiang /* recycle all individual short-lived pages */ 10826aaa7b06SGao Xiang if (z_erofs_put_shortlivedpage(pagepool, page)) 108347e4937aSGao Xiang continue; 1084*67148551SGao Xiang if (err) 1085*67148551SGao Xiang z_erofs_page_mark_eio(page); 108647e4937aSGao Xiang z_erofs_onlinepage_endio(page); 108747e4937aSGao Xiang } 108847e4937aSGao Xiang 108947e4937aSGao Xiang if (pages == z_pagemap_global) 109047e4937aSGao Xiang mutex_unlock(&z_pagemap_global_lock); 10918d8a09b0SGao Xiang else if (pages != pages_onstack) 109247e4937aSGao Xiang kvfree(pages); 109347e4937aSGao Xiang 109487ca34a7SGao Xiang pcl->nr_pages = 0; 109506a304cdSGao Xiang pcl->bvset.nextpage = NULL; 109687ca34a7SGao Xiang pcl->vcnt = 0; 109747e4937aSGao Xiang 109887ca34a7SGao Xiang /* pcluster lock MUST be taken before the following line */ 109947e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 110087ca34a7SGao Xiang mutex_unlock(&pcl->lock); 110147e4937aSGao Xiang return err; 110247e4937aSGao Xiang } 110347e4937aSGao Xiang 11040c638f70SGao Xiang static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1105eaa9172aSGao Xiang struct page **pagepool) 110647e4937aSGao Xiang { 110747e4937aSGao Xiang z_erofs_next_pcluster_t owned = io->head; 110847e4937aSGao Xiang 110947e4937aSGao Xiang while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 111047e4937aSGao Xiang struct z_erofs_pcluster *pcl; 111147e4937aSGao Xiang 111247e4937aSGao Xiang /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 111347e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 111447e4937aSGao Xiang 111547e4937aSGao Xiang /* no possible that 'owned' equals NULL */ 111647e4937aSGao Xiang DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 111747e4937aSGao Xiang 111847e4937aSGao Xiang pcl = container_of(owned, struct z_erofs_pcluster, next); 111947e4937aSGao Xiang owned = READ_ONCE(pcl->next); 112047e4937aSGao Xiang 1121*67148551SGao Xiang z_erofs_decompress_pcluster(io->sb, pcl, pagepool, 1122*67148551SGao Xiang io->eio ? -EIO : 0); 112387ca34a7SGao Xiang erofs_workgroup_put(&pcl->obj); 112447e4937aSGao Xiang } 112547e4937aSGao Xiang } 112647e4937aSGao Xiang 11270c638f70SGao Xiang static void z_erofs_decompressqueue_work(struct work_struct *work) 112847e4937aSGao Xiang { 1129a4b1fab1SGao Xiang struct z_erofs_decompressqueue *bgq = 1130a4b1fab1SGao Xiang container_of(work, struct z_erofs_decompressqueue, u.work); 1131eaa9172aSGao Xiang struct page *pagepool = NULL; 113247e4937aSGao Xiang 1133a4b1fab1SGao Xiang DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 11340c638f70SGao Xiang z_erofs_decompress_queue(bgq, &pagepool); 113547e4937aSGao Xiang 1136eaa9172aSGao Xiang erofs_release_pages(&pagepool); 1137a4b1fab1SGao Xiang kvfree(bgq); 113847e4937aSGao Xiang } 113947e4937aSGao Xiang 11407865827cSGao Xiang static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 11417865827cSGao Xiang bool sync, int bios) 11427865827cSGao Xiang { 11437865827cSGao Xiang struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 11447865827cSGao Xiang 11457865827cSGao Xiang /* wake up the caller thread for sync decompression */ 11467865827cSGao Xiang if (sync) { 11477865827cSGao Xiang if (!atomic_add_return(bios, &io->pending_bios)) 114860b30050SHongyu Jin complete(&io->u.done); 114960b30050SHongyu Jin 11507865827cSGao Xiang return; 11517865827cSGao Xiang } 11527865827cSGao Xiang 11537865827cSGao Xiang if (atomic_add_return(bios, &io->pending_bios)) 11547865827cSGao Xiang return; 11557865827cSGao Xiang /* Use workqueue and sync decompression for atomic contexts only */ 11567865827cSGao Xiang if (in_atomic() || irqs_disabled()) { 11577865827cSGao Xiang queue_work(z_erofs_workqueue, &io->u.work); 11587865827cSGao Xiang /* enable sync decompression for readahead */ 11597865827cSGao Xiang if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 11607865827cSGao Xiang sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 11617865827cSGao Xiang return; 11627865827cSGao Xiang } 11637865827cSGao Xiang z_erofs_decompressqueue_work(&io->u.work); 11647865827cSGao Xiang } 11657865827cSGao Xiang 116647e4937aSGao Xiang static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 116747e4937aSGao Xiang unsigned int nr, 1168eaa9172aSGao Xiang struct page **pagepool, 11699f2731d6SGao Xiang struct address_space *mc) 117047e4937aSGao Xiang { 117147e4937aSGao Xiang const pgoff_t index = pcl->obj.index; 11729f2731d6SGao Xiang gfp_t gfp = mapping_gfp_mask(mc); 117347e4937aSGao Xiang bool tocache = false; 117447e4937aSGao Xiang 117547e4937aSGao Xiang struct address_space *mapping; 117647e4937aSGao Xiang struct page *oldpage, *page; 117747e4937aSGao Xiang 117847e4937aSGao Xiang compressed_page_t t; 117947e4937aSGao Xiang int justfound; 118047e4937aSGao Xiang 118147e4937aSGao Xiang repeat: 1182ed722fbcSGao Xiang page = READ_ONCE(pcl->compressed_bvecs[nr].page); 118347e4937aSGao Xiang oldpage = page; 118447e4937aSGao Xiang 118547e4937aSGao Xiang if (!page) 118647e4937aSGao Xiang goto out_allocpage; 118747e4937aSGao Xiang 118847e4937aSGao Xiang /* process the target tagged pointer */ 118947e4937aSGao Xiang t = tagptr_init(compressed_page_t, page); 119047e4937aSGao Xiang justfound = tagptr_unfold_tags(t); 119147e4937aSGao Xiang page = tagptr_unfold_ptr(t); 119247e4937aSGao Xiang 11931825c8d7SGao Xiang /* 11941825c8d7SGao Xiang * preallocated cached pages, which is used to avoid direct reclaim 11951825c8d7SGao Xiang * otherwise, it will go inplace I/O path instead. 11961825c8d7SGao Xiang */ 11971825c8d7SGao Xiang if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1198ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 11991825c8d7SGao Xiang set_page_private(page, 0); 12001825c8d7SGao Xiang tocache = true; 12011825c8d7SGao Xiang goto out_tocache; 12021825c8d7SGao Xiang } 120347e4937aSGao Xiang mapping = READ_ONCE(page->mapping); 120447e4937aSGao Xiang 120547e4937aSGao Xiang /* 12066aaa7b06SGao Xiang * file-backed online pages in plcuster are all locked steady, 120747e4937aSGao Xiang * therefore it is impossible for `mapping' to be NULL. 120847e4937aSGao Xiang */ 120947e4937aSGao Xiang if (mapping && mapping != mc) 121047e4937aSGao Xiang /* ought to be unmanaged pages */ 121147e4937aSGao Xiang goto out; 121247e4937aSGao Xiang 12136aaa7b06SGao Xiang /* directly return for shortlived page as well */ 12146aaa7b06SGao Xiang if (z_erofs_is_shortlived_page(page)) 12156aaa7b06SGao Xiang goto out; 12166aaa7b06SGao Xiang 121747e4937aSGao Xiang lock_page(page); 121847e4937aSGao Xiang 121947e4937aSGao Xiang /* only true if page reclaim goes wrong, should never happen */ 122047e4937aSGao Xiang DBG_BUGON(justfound && PagePrivate(page)); 122147e4937aSGao Xiang 122247e4937aSGao Xiang /* the page is still in manage cache */ 122347e4937aSGao Xiang if (page->mapping == mc) { 1224ed722fbcSGao Xiang WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 122547e4937aSGao Xiang 122647e4937aSGao Xiang if (!PagePrivate(page)) { 122747e4937aSGao Xiang /* 122847e4937aSGao Xiang * impossible to be !PagePrivate(page) for 122947e4937aSGao Xiang * the current restriction as well if 1230ed722fbcSGao Xiang * the page is already in compressed_bvecs[]. 123147e4937aSGao Xiang */ 123247e4937aSGao Xiang DBG_BUGON(!justfound); 123347e4937aSGao Xiang 123447e4937aSGao Xiang justfound = 0; 123547e4937aSGao Xiang set_page_private(page, (unsigned long)pcl); 123647e4937aSGao Xiang SetPagePrivate(page); 123747e4937aSGao Xiang } 123847e4937aSGao Xiang 123947e4937aSGao Xiang /* no need to submit io if it is already up-to-date */ 124047e4937aSGao Xiang if (PageUptodate(page)) { 124147e4937aSGao Xiang unlock_page(page); 124247e4937aSGao Xiang page = NULL; 124347e4937aSGao Xiang } 124447e4937aSGao Xiang goto out; 124547e4937aSGao Xiang } 124647e4937aSGao Xiang 124747e4937aSGao Xiang /* 124847e4937aSGao Xiang * the managed page has been truncated, it's unsafe to 124947e4937aSGao Xiang * reuse this one, let's allocate a new cache-managed page. 125047e4937aSGao Xiang */ 125147e4937aSGao Xiang DBG_BUGON(page->mapping); 125247e4937aSGao Xiang DBG_BUGON(!justfound); 125347e4937aSGao Xiang 125447e4937aSGao Xiang tocache = true; 125547e4937aSGao Xiang unlock_page(page); 125647e4937aSGao Xiang put_page(page); 125747e4937aSGao Xiang out_allocpage: 12585ddcee1fSGao Xiang page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1259ed722fbcSGao Xiang if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1260ed722fbcSGao Xiang oldpage, page)) { 1261eaa9172aSGao Xiang erofs_pagepool_add(pagepool, page); 12625ddcee1fSGao Xiang cond_resched(); 12635ddcee1fSGao Xiang goto repeat; 12645ddcee1fSGao Xiang } 12651825c8d7SGao Xiang out_tocache: 1266bf225074SGao Xiang if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1267bf225074SGao Xiang /* turn into temporary page if fails (1 ref) */ 1268bf225074SGao Xiang set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1269bf225074SGao Xiang goto out; 1270a30573b3SGao Xiang } 1271bf225074SGao Xiang attach_page_private(page, pcl); 1272bf225074SGao Xiang /* drop a refcount added by allocpage (then we have 2 refs here) */ 1273bf225074SGao Xiang put_page(page); 1274bf225074SGao Xiang 127547e4937aSGao Xiang out: /* the only exit (for tracing and debugging) */ 127647e4937aSGao Xiang return page; 127747e4937aSGao Xiang } 127847e4937aSGao Xiang 1279a4b1fab1SGao Xiang static struct z_erofs_decompressqueue * 1280a4b1fab1SGao Xiang jobqueue_init(struct super_block *sb, 1281a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 128247e4937aSGao Xiang { 1283a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q; 128447e4937aSGao Xiang 1285a4b1fab1SGao Xiang if (fg && !*fg) { 1286a4b1fab1SGao Xiang q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1287a4b1fab1SGao Xiang if (!q) { 1288a4b1fab1SGao Xiang *fg = true; 1289a4b1fab1SGao Xiang goto fg_out; 129047e4937aSGao Xiang } 12910c638f70SGao Xiang INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1292a4b1fab1SGao Xiang } else { 1293a4b1fab1SGao Xiang fg_out: 1294a4b1fab1SGao Xiang q = fgq; 129560b30050SHongyu Jin init_completion(&fgq->u.done); 1296a4b1fab1SGao Xiang atomic_set(&fgq->pending_bios, 0); 1297*67148551SGao Xiang q->eio = false; 1298a4b1fab1SGao Xiang } 1299a4b1fab1SGao Xiang q->sb = sb; 1300a4b1fab1SGao Xiang q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1301a4b1fab1SGao Xiang return q; 130247e4937aSGao Xiang } 130347e4937aSGao Xiang 130447e4937aSGao Xiang /* define decompression jobqueue types */ 130547e4937aSGao Xiang enum { 130647e4937aSGao Xiang JQ_BYPASS, 130747e4937aSGao Xiang JQ_SUBMIT, 130847e4937aSGao Xiang NR_JOBQUEUES, 130947e4937aSGao Xiang }; 131047e4937aSGao Xiang 131147e4937aSGao Xiang static void *jobqueueset_init(struct super_block *sb, 1312a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[], 1313a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, bool *fg) 131447e4937aSGao Xiang { 131547e4937aSGao Xiang /* 131647e4937aSGao Xiang * if managed cache is enabled, bypass jobqueue is needed, 131747e4937aSGao Xiang * no need to read from device for all pclusters in this queue. 131847e4937aSGao Xiang */ 1319a4b1fab1SGao Xiang q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1320a4b1fab1SGao Xiang q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 132147e4937aSGao Xiang 1322a4b1fab1SGao Xiang return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 132347e4937aSGao Xiang } 132447e4937aSGao Xiang 132547e4937aSGao Xiang static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 132647e4937aSGao Xiang z_erofs_next_pcluster_t qtail[], 132747e4937aSGao Xiang z_erofs_next_pcluster_t owned_head) 132847e4937aSGao Xiang { 132947e4937aSGao Xiang z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 133047e4937aSGao Xiang z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 133147e4937aSGao Xiang 133247e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 133347e4937aSGao Xiang if (owned_head == Z_EROFS_PCLUSTER_TAIL) 133447e4937aSGao Xiang owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 133547e4937aSGao Xiang 133647e4937aSGao Xiang WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 133747e4937aSGao Xiang 133847e4937aSGao Xiang WRITE_ONCE(*submit_qtail, owned_head); 133947e4937aSGao Xiang WRITE_ONCE(*bypass_qtail, &pcl->next); 134047e4937aSGao Xiang 134147e4937aSGao Xiang qtail[JQ_BYPASS] = &pcl->next; 134247e4937aSGao Xiang } 134347e4937aSGao Xiang 13447865827cSGao Xiang static void z_erofs_decompressqueue_endio(struct bio *bio) 13457865827cSGao Xiang { 13467865827cSGao Xiang tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 13477865827cSGao Xiang struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 13487865827cSGao Xiang blk_status_t err = bio->bi_status; 13497865827cSGao Xiang struct bio_vec *bvec; 13507865827cSGao Xiang struct bvec_iter_all iter_all; 13517865827cSGao Xiang 13527865827cSGao Xiang bio_for_each_segment_all(bvec, bio, iter_all) { 13537865827cSGao Xiang struct page *page = bvec->bv_page; 13547865827cSGao Xiang 13557865827cSGao Xiang DBG_BUGON(PageUptodate(page)); 13567865827cSGao Xiang DBG_BUGON(z_erofs_page_is_invalidated(page)); 13577865827cSGao Xiang 13587865827cSGao Xiang if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 13597865827cSGao Xiang if (!err) 13607865827cSGao Xiang SetPageUptodate(page); 13617865827cSGao Xiang unlock_page(page); 13627865827cSGao Xiang } 13637865827cSGao Xiang } 1364*67148551SGao Xiang if (err) 1365*67148551SGao Xiang q->eio = true; 13667865827cSGao Xiang z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 13677865827cSGao Xiang bio_put(bio); 13687865827cSGao Xiang } 13697865827cSGao Xiang 137083a386c0SGao Xiang static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1371eaa9172aSGao Xiang struct page **pagepool, 1372a4b1fab1SGao Xiang struct z_erofs_decompressqueue *fgq, 1373a4b1fab1SGao Xiang bool *force_fg) 137447e4937aSGao Xiang { 137583a386c0SGao Xiang struct super_block *sb = f->inode->i_sb; 137683a386c0SGao Xiang struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 137747e4937aSGao Xiang z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1378a4b1fab1SGao Xiang struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 137947e4937aSGao Xiang void *bi_private; 13805c6dcc57SGao Xiang z_erofs_next_pcluster_t owned_head = f->owned_head; 1381dfeab2e9SGao Xiang /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 13823f649ab7SKees Cook pgoff_t last_index; 1383dfeab2e9SGao Xiang struct block_device *last_bdev; 13841e4a2955SGao Xiang unsigned int nr_bios = 0; 13851e4a2955SGao Xiang struct bio *bio = NULL; 138647e4937aSGao Xiang 1387a4b1fab1SGao Xiang bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1388a4b1fab1SGao Xiang qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1389a4b1fab1SGao Xiang qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 139047e4937aSGao Xiang 139147e4937aSGao Xiang /* by default, all need io submission */ 139247e4937aSGao Xiang q[JQ_SUBMIT]->head = owned_head; 139347e4937aSGao Xiang 139447e4937aSGao Xiang do { 1395dfeab2e9SGao Xiang struct erofs_map_dev mdev; 139647e4937aSGao Xiang struct z_erofs_pcluster *pcl; 13971e4a2955SGao Xiang pgoff_t cur, end; 13981e4a2955SGao Xiang unsigned int i = 0; 13991e4a2955SGao Xiang bool bypass = true; 140047e4937aSGao Xiang 140147e4937aSGao Xiang /* no possible 'owned_head' equals the following */ 140247e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 140347e4937aSGao Xiang DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 140447e4937aSGao Xiang 140547e4937aSGao Xiang pcl = container_of(owned_head, struct z_erofs_pcluster, next); 140647e4937aSGao Xiang 1407cecf864dSYue Hu /* close the main owned chain at first */ 1408cecf864dSYue Hu owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1409cecf864dSYue Hu Z_EROFS_PCLUSTER_TAIL_CLOSED); 1410cecf864dSYue Hu if (z_erofs_is_inline_pcluster(pcl)) { 1411cecf864dSYue Hu move_to_bypass_jobqueue(pcl, qtail, owned_head); 1412cecf864dSYue Hu continue; 1413cecf864dSYue Hu } 1414cecf864dSYue Hu 1415dfeab2e9SGao Xiang /* no device id here, thus it will always succeed */ 1416dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 1417dfeab2e9SGao Xiang .m_pa = blknr_to_addr(pcl->obj.index), 1418dfeab2e9SGao Xiang }; 1419dfeab2e9SGao Xiang (void)erofs_map_dev(sb, &mdev); 1420dfeab2e9SGao Xiang 1421dfeab2e9SGao Xiang cur = erofs_blknr(mdev.m_pa); 14229f6cc76eSGao Xiang end = cur + pcl->pclusterpages; 142347e4937aSGao Xiang 14241e4a2955SGao Xiang do { 14251e4a2955SGao Xiang struct page *page; 142647e4937aSGao Xiang 14271e4a2955SGao Xiang page = pickup_page_for_submission(pcl, i++, pagepool, 142883a386c0SGao Xiang mc); 14291e4a2955SGao Xiang if (!page) 14301e4a2955SGao Xiang continue; 143147e4937aSGao Xiang 1432dfeab2e9SGao Xiang if (bio && (cur != last_index + 1 || 1433dfeab2e9SGao Xiang last_bdev != mdev.m_bdev)) { 143447e4937aSGao Xiang submit_bio_retry: 143594e4e153SGao Xiang submit_bio(bio); 143647e4937aSGao Xiang bio = NULL; 143747e4937aSGao Xiang } 143847e4937aSGao Xiang 143947e4937aSGao Xiang if (!bio) { 144007888c66SChristoph Hellwig bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 144107888c66SChristoph Hellwig REQ_OP_READ, GFP_NOIO); 14420c638f70SGao Xiang bio->bi_end_io = z_erofs_decompressqueue_endio; 1443dfeab2e9SGao Xiang 1444dfeab2e9SGao Xiang last_bdev = mdev.m_bdev; 14451e4a2955SGao Xiang bio->bi_iter.bi_sector = (sector_t)cur << 1446a5c0b780SGao Xiang LOG_SECTORS_PER_BLOCK; 1447a5c0b780SGao Xiang bio->bi_private = bi_private; 14486ea5aad3SGao Xiang if (f->readahead) 14496ea5aad3SGao Xiang bio->bi_opf |= REQ_RAHEAD; 145047e4937aSGao Xiang ++nr_bios; 145147e4937aSGao Xiang } 145247e4937aSGao Xiang 14536c3e485eSGao Xiang if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 145447e4937aSGao Xiang goto submit_bio_retry; 145547e4937aSGao Xiang 14561e4a2955SGao Xiang last_index = cur; 14571e4a2955SGao Xiang bypass = false; 14581e4a2955SGao Xiang } while (++cur < end); 145947e4937aSGao Xiang 14601e4a2955SGao Xiang if (!bypass) 146147e4937aSGao Xiang qtail[JQ_SUBMIT] = &pcl->next; 146247e4937aSGao Xiang else 146347e4937aSGao Xiang move_to_bypass_jobqueue(pcl, qtail, owned_head); 146447e4937aSGao Xiang } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 146547e4937aSGao Xiang 146647e4937aSGao Xiang if (bio) 146794e4e153SGao Xiang submit_bio(bio); 146847e4937aSGao Xiang 1469587a67b7SGao Xiang /* 1470587a67b7SGao Xiang * although background is preferred, no one is pending for submission. 1471587a67b7SGao Xiang * don't issue workqueue for decompression but drop it directly instead. 1472587a67b7SGao Xiang */ 1473587a67b7SGao Xiang if (!*force_fg && !nr_bios) { 1474587a67b7SGao Xiang kvfree(q[JQ_SUBMIT]); 14751e4a2955SGao Xiang return; 1476587a67b7SGao Xiang } 1477a4b1fab1SGao Xiang z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 147847e4937aSGao Xiang } 147947e4937aSGao Xiang 148083a386c0SGao Xiang static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1481eaa9172aSGao Xiang struct page **pagepool, bool force_fg) 148247e4937aSGao Xiang { 1483a4b1fab1SGao Xiang struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 148447e4937aSGao Xiang 14855c6dcc57SGao Xiang if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 148647e4937aSGao Xiang return; 148783a386c0SGao Xiang z_erofs_submit_queue(f, pagepool, io, &force_fg); 148847e4937aSGao Xiang 14890c638f70SGao Xiang /* handle bypass queue (no i/o pclusters) immediately */ 14900c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 149147e4937aSGao Xiang 149247e4937aSGao Xiang if (!force_fg) 149347e4937aSGao Xiang return; 149447e4937aSGao Xiang 149547e4937aSGao Xiang /* wait until all bios are completed */ 149660b30050SHongyu Jin wait_for_completion_io(&io[JQ_SUBMIT].u.done); 149747e4937aSGao Xiang 14980c638f70SGao Xiang /* handle synchronous decompress queue in the caller context */ 14990c638f70SGao Xiang z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 150047e4937aSGao Xiang } 150147e4937aSGao Xiang 150238629291SGao Xiang /* 150338629291SGao Xiang * Since partial uptodate is still unimplemented for now, we have to use 150438629291SGao Xiang * approximate readmore strategies as a start. 150538629291SGao Xiang */ 150638629291SGao Xiang static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 150738629291SGao Xiang struct readahead_control *rac, 150838629291SGao Xiang erofs_off_t end, 1509eaa9172aSGao Xiang struct page **pagepool, 151038629291SGao Xiang bool backmost) 151138629291SGao Xiang { 151238629291SGao Xiang struct inode *inode = f->inode; 151338629291SGao Xiang struct erofs_map_blocks *map = &f->map; 151438629291SGao Xiang erofs_off_t cur; 151538629291SGao Xiang int err; 151638629291SGao Xiang 151738629291SGao Xiang if (backmost) { 151838629291SGao Xiang map->m_la = end; 1519622ceaddSGao Xiang err = z_erofs_map_blocks_iter(inode, map, 1520622ceaddSGao Xiang EROFS_GET_BLOCKS_READMORE); 152138629291SGao Xiang if (err) 152238629291SGao Xiang return; 152338629291SGao Xiang 152438629291SGao Xiang /* expend ra for the trailing edge if readahead */ 152538629291SGao Xiang if (rac) { 152638629291SGao Xiang loff_t newstart = readahead_pos(rac); 152738629291SGao Xiang 152838629291SGao Xiang cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 152938629291SGao Xiang readahead_expand(rac, newstart, cur - newstart); 153038629291SGao Xiang return; 153138629291SGao Xiang } 153238629291SGao Xiang end = round_up(end, PAGE_SIZE); 153338629291SGao Xiang } else { 153438629291SGao Xiang end = round_up(map->m_la, PAGE_SIZE); 153538629291SGao Xiang 153638629291SGao Xiang if (!map->m_llen) 153738629291SGao Xiang return; 153838629291SGao Xiang } 153938629291SGao Xiang 154038629291SGao Xiang cur = map->m_la + map->m_llen - 1; 154138629291SGao Xiang while (cur >= end) { 154238629291SGao Xiang pgoff_t index = cur >> PAGE_SHIFT; 154338629291SGao Xiang struct page *page; 154438629291SGao Xiang 154538629291SGao Xiang page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1546aa793b46SGao Xiang if (page) { 154738629291SGao Xiang if (PageUptodate(page)) { 154838629291SGao Xiang unlock_page(page); 1549aa793b46SGao Xiang } else { 155038629291SGao Xiang err = z_erofs_do_read_page(f, page, pagepool); 155138629291SGao Xiang if (err) 155238629291SGao Xiang erofs_err(inode->i_sb, 155338629291SGao Xiang "readmore error at page %lu @ nid %llu", 155438629291SGao Xiang index, EROFS_I(inode)->nid); 1555aa793b46SGao Xiang } 155638629291SGao Xiang put_page(page); 1557aa793b46SGao Xiang } 1558aa793b46SGao Xiang 155938629291SGao Xiang if (cur < PAGE_SIZE) 156038629291SGao Xiang break; 156138629291SGao Xiang cur = (index << PAGE_SHIFT) - 1; 156238629291SGao Xiang } 156338629291SGao Xiang } 156438629291SGao Xiang 1565a2e20a25SMatthew Wilcox (Oracle) static int z_erofs_read_folio(struct file *file, struct folio *folio) 156647e4937aSGao Xiang { 1567a2e20a25SMatthew Wilcox (Oracle) struct page *page = &folio->page; 156847e4937aSGao Xiang struct inode *const inode = page->mapping->host; 156940452ffcSHuang Jianan struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 157047e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1571eaa9172aSGao Xiang struct page *pagepool = NULL; 157247e4937aSGao Xiang int err; 157347e4937aSGao Xiang 157447e4937aSGao Xiang trace_erofs_readpage(page, false); 157547e4937aSGao Xiang f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 157647e4937aSGao Xiang 157738629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 157838629291SGao Xiang &pagepool, true); 15791825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 158038629291SGao Xiang z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 158138629291SGao Xiang 15825c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 158347e4937aSGao Xiang 158447e4937aSGao Xiang /* if some compressed cluster ready, need submit them anyway */ 158583a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 158640452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, 0)); 158747e4937aSGao Xiang 158847e4937aSGao Xiang if (err) 15894f761fa2SGao Xiang erofs_err(inode->i_sb, "failed to read, err [%d]", err); 159047e4937aSGao Xiang 159109c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1592eaa9172aSGao Xiang erofs_release_pages(&pagepool); 159347e4937aSGao Xiang return err; 159447e4937aSGao Xiang } 159547e4937aSGao Xiang 15960615090cSMatthew Wilcox (Oracle) static void z_erofs_readahead(struct readahead_control *rac) 159747e4937aSGao Xiang { 15980615090cSMatthew Wilcox (Oracle) struct inode *const inode = rac->mapping->host; 159947e4937aSGao Xiang struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 160047e4937aSGao Xiang struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1601eaa9172aSGao Xiang struct page *pagepool = NULL, *head = NULL, *page; 160238629291SGao Xiang unsigned int nr_pages; 160347e4937aSGao Xiang 16046ea5aad3SGao Xiang f.readahead = true; 16050615090cSMatthew Wilcox (Oracle) f.headoffset = readahead_pos(rac); 160647e4937aSGao Xiang 160738629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, f.headoffset + 160838629291SGao Xiang readahead_length(rac) - 1, &pagepool, true); 160938629291SGao Xiang nr_pages = readahead_count(rac); 161038629291SGao Xiang trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 161138629291SGao Xiang 16120615090cSMatthew Wilcox (Oracle) while ((page = readahead_page(rac))) { 161347e4937aSGao Xiang set_page_private(page, (unsigned long)head); 161447e4937aSGao Xiang head = page; 161547e4937aSGao Xiang } 161647e4937aSGao Xiang 161747e4937aSGao Xiang while (head) { 161847e4937aSGao Xiang struct page *page = head; 161947e4937aSGao Xiang int err; 162047e4937aSGao Xiang 162147e4937aSGao Xiang /* traversal in reverse order */ 162247e4937aSGao Xiang head = (void *)page_private(page); 162347e4937aSGao Xiang 16241825c8d7SGao Xiang err = z_erofs_do_read_page(&f, page, &pagepool); 1625a5876e24SGao Xiang if (err) 16264f761fa2SGao Xiang erofs_err(inode->i_sb, 16274f761fa2SGao Xiang "readahead error at page %lu @ nid %llu", 16284f761fa2SGao Xiang page->index, EROFS_I(inode)->nid); 162947e4937aSGao Xiang put_page(page); 163047e4937aSGao Xiang } 163138629291SGao Xiang z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 16325c6dcc57SGao Xiang (void)z_erofs_collector_end(&f); 163347e4937aSGao Xiang 163483a386c0SGao Xiang z_erofs_runqueue(&f, &pagepool, 163540452ffcSHuang Jianan z_erofs_get_sync_decompress_policy(sbi, nr_pages)); 163609c54379SGao Xiang erofs_put_metabuf(&f.map.buf); 1637eaa9172aSGao Xiang erofs_release_pages(&pagepool); 163847e4937aSGao Xiang } 163947e4937aSGao Xiang 16400c638f70SGao Xiang const struct address_space_operations z_erofs_aops = { 1641a2e20a25SMatthew Wilcox (Oracle) .read_folio = z_erofs_read_folio, 16420615090cSMatthew Wilcox (Oracle) .readahead = z_erofs_readahead, 164347e4937aSGao Xiang }; 1644