147e4937aSGao Xiang // SPDX-License-Identifier: GPL-2.0-only 247e4937aSGao Xiang /* 347e4937aSGao Xiang * Copyright (C) 2017-2018 HUAWEI, Inc. 4592e7cd0SAlexander A. Klimov * https://www.huawei.com/ 5c5aa903aSGao Xiang * Copyright (C) 2021, Alibaba Cloud 647e4937aSGao Xiang */ 747e4937aSGao Xiang #include "internal.h" 847e4937aSGao Xiang #include <linux/prefetch.h> 95375e7c8SJeffle Xu #include <linux/sched/mm.h> 1006252e9cSGao Xiang #include <linux/dax.h> 1147e4937aSGao Xiang #include <trace/events/erofs.h> 1247e4937aSGao Xiang 13fdf80a47SGao Xiang void erofs_unmap_metabuf(struct erofs_buf *buf) 14fdf80a47SGao Xiang { 15fdf80a47SGao Xiang if (buf->kmap_type == EROFS_KMAP) 16fdf80a47SGao Xiang kunmap(buf->page); 17fdf80a47SGao Xiang else if (buf->kmap_type == EROFS_KMAP_ATOMIC) 18fdf80a47SGao Xiang kunmap_atomic(buf->base); 19fdf80a47SGao Xiang buf->base = NULL; 20fdf80a47SGao Xiang buf->kmap_type = EROFS_NO_KMAP; 21fdf80a47SGao Xiang } 22fdf80a47SGao Xiang 23fdf80a47SGao Xiang void erofs_put_metabuf(struct erofs_buf *buf) 24fdf80a47SGao Xiang { 25fdf80a47SGao Xiang if (!buf->page) 26fdf80a47SGao Xiang return; 27fdf80a47SGao Xiang erofs_unmap_metabuf(buf); 28fdf80a47SGao Xiang put_page(buf->page); 29fdf80a47SGao Xiang buf->page = NULL; 30fdf80a47SGao Xiang } 31fdf80a47SGao Xiang 32fe5de585SGao Xiang void *erofs_bread(struct erofs_buf *buf, struct inode *inode, 33fdf80a47SGao Xiang erofs_blk_t blkaddr, enum erofs_kmap_type type) 34fdf80a47SGao Xiang { 35fe5de585SGao Xiang struct address_space *const mapping = inode->i_mapping; 36fdf80a47SGao Xiang erofs_off_t offset = blknr_to_addr(blkaddr); 37fdf80a47SGao Xiang pgoff_t index = offset >> PAGE_SHIFT; 38fdf80a47SGao Xiang struct page *page = buf->page; 395375e7c8SJeffle Xu struct folio *folio; 405375e7c8SJeffle Xu unsigned int nofs_flag; 41fdf80a47SGao Xiang 42fdf80a47SGao Xiang if (!page || page->index != index) { 43fdf80a47SGao Xiang erofs_put_metabuf(buf); 445375e7c8SJeffle Xu 455375e7c8SJeffle Xu nofs_flag = memalloc_nofs_save(); 465375e7c8SJeffle Xu folio = read_cache_folio(mapping, index, NULL, NULL); 475375e7c8SJeffle Xu memalloc_nofs_restore(nofs_flag); 485375e7c8SJeffle Xu if (IS_ERR(folio)) 495375e7c8SJeffle Xu return folio; 505375e7c8SJeffle Xu 51fdf80a47SGao Xiang /* should already be PageUptodate, no need to lock page */ 525375e7c8SJeffle Xu page = folio_file_page(folio, index); 53fdf80a47SGao Xiang buf->page = page; 54fdf80a47SGao Xiang } 55fdf80a47SGao Xiang if (buf->kmap_type == EROFS_NO_KMAP) { 56fdf80a47SGao Xiang if (type == EROFS_KMAP) 57fdf80a47SGao Xiang buf->base = kmap(page); 58fdf80a47SGao Xiang else if (type == EROFS_KMAP_ATOMIC) 59fdf80a47SGao Xiang buf->base = kmap_atomic(page); 60fdf80a47SGao Xiang buf->kmap_type = type; 61fdf80a47SGao Xiang } else if (buf->kmap_type != type) { 62fdf80a47SGao Xiang DBG_BUGON(1); 63fdf80a47SGao Xiang return ERR_PTR(-EFAULT); 64fdf80a47SGao Xiang } 65fdf80a47SGao Xiang if (type == EROFS_NO_KMAP) 66fdf80a47SGao Xiang return NULL; 67fdf80a47SGao Xiang return buf->base + (offset & ~PAGE_MASK); 68fdf80a47SGao Xiang } 69fdf80a47SGao Xiang 70fe5de585SGao Xiang void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb, 71fe5de585SGao Xiang erofs_blk_t blkaddr, enum erofs_kmap_type type) 72fe5de585SGao Xiang { 735375e7c8SJeffle Xu if (erofs_is_fscache_mode(sb)) 745375e7c8SJeffle Xu return erofs_bread(buf, EROFS_SB(sb)->s_fscache->inode, 755375e7c8SJeffle Xu blkaddr, type); 765375e7c8SJeffle Xu 77fe5de585SGao Xiang return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type); 78fe5de585SGao Xiang } 79fe5de585SGao Xiang 8047e4937aSGao Xiang static int erofs_map_blocks_flatmode(struct inode *inode, 8147e4937aSGao Xiang struct erofs_map_blocks *map, 8247e4937aSGao Xiang int flags) 8347e4937aSGao Xiang { 8447e4937aSGao Xiang erofs_blk_t nblocks, lastblk; 8547e4937aSGao Xiang u64 offset = map->m_la; 86a5876e24SGao Xiang struct erofs_inode *vi = EROFS_I(inode); 878a765682SGao Xiang bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); 8847e4937aSGao Xiang 89fdf80a47SGao Xiang nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ); 908a765682SGao Xiang lastblk = nblocks - tailendpacking; 9147e4937aSGao Xiang 9247e4937aSGao Xiang /* there is no hole in flatmode */ 9347e4937aSGao Xiang map->m_flags = EROFS_MAP_MAPPED; 9447e4937aSGao Xiang if (offset < blknr_to_addr(lastblk)) { 9547e4937aSGao Xiang map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la; 9647e4937aSGao Xiang map->m_plen = blknr_to_addr(lastblk) - offset; 978a765682SGao Xiang } else if (tailendpacking) { 9847e4937aSGao Xiang /* 2 - inode inline B: inode, [xattrs], inline last blk... */ 9947e4937aSGao Xiang struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb); 10047e4937aSGao Xiang 10147e4937aSGao Xiang map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize + 10247e4937aSGao Xiang vi->xattr_isize + erofs_blkoff(map->m_la); 10347e4937aSGao Xiang map->m_plen = inode->i_size - offset; 10447e4937aSGao Xiang 105469407a3SGao Xiang /* inline data should be located in the same meta block */ 106469407a3SGao Xiang if (erofs_blkoff(map->m_pa) + map->m_plen > EROFS_BLKSIZ) { 1074f761fa2SGao Xiang erofs_err(inode->i_sb, 1084f761fa2SGao Xiang "inline data cross block boundary @ nid %llu", 10947e4937aSGao Xiang vi->nid); 11047e4937aSGao Xiang DBG_BUGON(1); 111469407a3SGao Xiang return -EFSCORRUPTED; 11247e4937aSGao Xiang } 11347e4937aSGao Xiang map->m_flags |= EROFS_MAP_META; 11447e4937aSGao Xiang } else { 1154f761fa2SGao Xiang erofs_err(inode->i_sb, 1164f761fa2SGao Xiang "internal error @ nid: %llu (size %llu), m_la 0x%llx", 11747e4937aSGao Xiang vi->nid, inode->i_size, map->m_la); 11847e4937aSGao Xiang DBG_BUGON(1); 119469407a3SGao Xiang return -EIO; 12047e4937aSGao Xiang } 121469407a3SGao Xiang return 0; 12247e4937aSGao Xiang } 12347e4937aSGao Xiang 12494d78946SJeffle Xu int erofs_map_blocks(struct inode *inode, 125c5aa903aSGao Xiang struct erofs_map_blocks *map, int flags) 126c5aa903aSGao Xiang { 127c5aa903aSGao Xiang struct super_block *sb = inode->i_sb; 128c5aa903aSGao Xiang struct erofs_inode *vi = EROFS_I(inode); 129c5aa903aSGao Xiang struct erofs_inode_chunk_index *idx; 130fdf80a47SGao Xiang struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 131c5aa903aSGao Xiang u64 chunknr; 132c5aa903aSGao Xiang unsigned int unit; 133c5aa903aSGao Xiang erofs_off_t pos; 134fdf80a47SGao Xiang void *kaddr; 135c5aa903aSGao Xiang int err = 0; 136c5aa903aSGao Xiang 137469407a3SGao Xiang trace_erofs_map_blocks_enter(inode, map, flags); 138dfeab2e9SGao Xiang map->m_deviceid = 0; 139c5aa903aSGao Xiang if (map->m_la >= inode->i_size) { 140c5aa903aSGao Xiang /* leave out-of-bound access unmapped */ 141c5aa903aSGao Xiang map->m_flags = 0; 142c5aa903aSGao Xiang map->m_plen = 0; 143c5aa903aSGao Xiang goto out; 144c5aa903aSGao Xiang } 145c5aa903aSGao Xiang 146469407a3SGao Xiang if (vi->datalayout != EROFS_INODE_CHUNK_BASED) { 147469407a3SGao Xiang err = erofs_map_blocks_flatmode(inode, map, flags); 148469407a3SGao Xiang goto out; 149469407a3SGao Xiang } 150c5aa903aSGao Xiang 151c5aa903aSGao Xiang if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) 152c5aa903aSGao Xiang unit = sizeof(*idx); /* chunk index */ 153c5aa903aSGao Xiang else 154c5aa903aSGao Xiang unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */ 155c5aa903aSGao Xiang 156c5aa903aSGao Xiang chunknr = map->m_la >> vi->chunkbits; 157c5aa903aSGao Xiang pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize + 158c5aa903aSGao Xiang vi->xattr_isize, unit) + unit * chunknr; 159c5aa903aSGao Xiang 160fdf80a47SGao Xiang kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP); 161fdf80a47SGao Xiang if (IS_ERR(kaddr)) { 162fdf80a47SGao Xiang err = PTR_ERR(kaddr); 163469407a3SGao Xiang goto out; 164469407a3SGao Xiang } 165c5aa903aSGao Xiang map->m_la = chunknr << vi->chunkbits; 166c5aa903aSGao Xiang map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits, 167c5aa903aSGao Xiang roundup(inode->i_size - map->m_la, EROFS_BLKSIZ)); 168c5aa903aSGao Xiang 169c5aa903aSGao Xiang /* handle block map */ 170c5aa903aSGao Xiang if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) { 171fdf80a47SGao Xiang __le32 *blkaddr = kaddr + erofs_blkoff(pos); 172c5aa903aSGao Xiang 173c5aa903aSGao Xiang if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) { 174c5aa903aSGao Xiang map->m_flags = 0; 175c5aa903aSGao Xiang } else { 176c5aa903aSGao Xiang map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr)); 177c5aa903aSGao Xiang map->m_flags = EROFS_MAP_MAPPED; 178c5aa903aSGao Xiang } 179c5aa903aSGao Xiang goto out_unlock; 180c5aa903aSGao Xiang } 181c5aa903aSGao Xiang /* parse chunk indexes */ 182fdf80a47SGao Xiang idx = kaddr + erofs_blkoff(pos); 183c5aa903aSGao Xiang switch (le32_to_cpu(idx->blkaddr)) { 184c5aa903aSGao Xiang case EROFS_NULL_ADDR: 185c5aa903aSGao Xiang map->m_flags = 0; 186c5aa903aSGao Xiang break; 187c5aa903aSGao Xiang default: 188dfeab2e9SGao Xiang map->m_deviceid = le16_to_cpu(idx->device_id) & 189dfeab2e9SGao Xiang EROFS_SB(sb)->device_id_mask; 190c5aa903aSGao Xiang map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr)); 191c5aa903aSGao Xiang map->m_flags = EROFS_MAP_MAPPED; 192c5aa903aSGao Xiang break; 193c5aa903aSGao Xiang } 194c5aa903aSGao Xiang out_unlock: 195fdf80a47SGao Xiang erofs_put_metabuf(&buf); 196c5aa903aSGao Xiang out: 197469407a3SGao Xiang if (!err) 198c5aa903aSGao Xiang map->m_llen = map->m_plen; 199469407a3SGao Xiang trace_erofs_map_blocks_exit(inode, map, flags, 0); 200c5aa903aSGao Xiang return err; 201c5aa903aSGao Xiang } 202c5aa903aSGao Xiang 203dfeab2e9SGao Xiang int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map) 204dfeab2e9SGao Xiang { 205dfeab2e9SGao Xiang struct erofs_dev_context *devs = EROFS_SB(sb)->devs; 206dfeab2e9SGao Xiang struct erofs_device_info *dif; 207dfeab2e9SGao Xiang int id; 208dfeab2e9SGao Xiang 209dfeab2e9SGao Xiang /* primary device by default */ 210dfeab2e9SGao Xiang map->m_bdev = sb->s_bdev; 211dfeab2e9SGao Xiang map->m_daxdev = EROFS_SB(sb)->dax_dev; 212de205114SChristoph Hellwig map->m_dax_part_off = EROFS_SB(sb)->dax_part_off; 213955b478eSJeffle Xu map->m_fscache = EROFS_SB(sb)->s_fscache; 214dfeab2e9SGao Xiang 215dfeab2e9SGao Xiang if (map->m_deviceid) { 216dfeab2e9SGao Xiang down_read(&devs->rwsem); 217dfeab2e9SGao Xiang dif = idr_find(&devs->tree, map->m_deviceid - 1); 218dfeab2e9SGao Xiang if (!dif) { 219dfeab2e9SGao Xiang up_read(&devs->rwsem); 220dfeab2e9SGao Xiang return -ENODEV; 221dfeab2e9SGao Xiang } 222dfeab2e9SGao Xiang map->m_bdev = dif->bdev; 223dfeab2e9SGao Xiang map->m_daxdev = dif->dax_dev; 224de205114SChristoph Hellwig map->m_dax_part_off = dif->dax_part_off; 225955b478eSJeffle Xu map->m_fscache = dif->fscache; 226dfeab2e9SGao Xiang up_read(&devs->rwsem); 227dfeab2e9SGao Xiang } else if (devs->extra_devices) { 228dfeab2e9SGao Xiang down_read(&devs->rwsem); 229dfeab2e9SGao Xiang idr_for_each_entry(&devs->tree, dif, id) { 230dfeab2e9SGao Xiang erofs_off_t startoff, length; 231dfeab2e9SGao Xiang 232dfeab2e9SGao Xiang if (!dif->mapped_blkaddr) 233dfeab2e9SGao Xiang continue; 234dfeab2e9SGao Xiang startoff = blknr_to_addr(dif->mapped_blkaddr); 235dfeab2e9SGao Xiang length = blknr_to_addr(dif->blocks); 236dfeab2e9SGao Xiang 237dfeab2e9SGao Xiang if (map->m_pa >= startoff && 238dfeab2e9SGao Xiang map->m_pa < startoff + length) { 239dfeab2e9SGao Xiang map->m_pa -= startoff; 240dfeab2e9SGao Xiang map->m_bdev = dif->bdev; 241dfeab2e9SGao Xiang map->m_daxdev = dif->dax_dev; 242de205114SChristoph Hellwig map->m_dax_part_off = dif->dax_part_off; 243955b478eSJeffle Xu map->m_fscache = dif->fscache; 244dfeab2e9SGao Xiang break; 245dfeab2e9SGao Xiang } 246dfeab2e9SGao Xiang } 247dfeab2e9SGao Xiang up_read(&devs->rwsem); 248dfeab2e9SGao Xiang } 249dfeab2e9SGao Xiang return 0; 250dfeab2e9SGao Xiang } 251dfeab2e9SGao Xiang 252a08e67a0SHuang Jianan static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 253a08e67a0SHuang Jianan unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 254a08e67a0SHuang Jianan { 255a08e67a0SHuang Jianan int ret; 256a08e67a0SHuang Jianan struct erofs_map_blocks map; 257dfeab2e9SGao Xiang struct erofs_map_dev mdev; 258a08e67a0SHuang Jianan 259a08e67a0SHuang Jianan map.m_la = offset; 260a08e67a0SHuang Jianan map.m_llen = length; 261a08e67a0SHuang Jianan 262c5aa903aSGao Xiang ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); 263a08e67a0SHuang Jianan if (ret < 0) 264a08e67a0SHuang Jianan return ret; 265a08e67a0SHuang Jianan 266dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 267dfeab2e9SGao Xiang .m_deviceid = map.m_deviceid, 268dfeab2e9SGao Xiang .m_pa = map.m_pa, 269dfeab2e9SGao Xiang }; 270dfeab2e9SGao Xiang ret = erofs_map_dev(inode->i_sb, &mdev); 271dfeab2e9SGao Xiang if (ret) 272dfeab2e9SGao Xiang return ret; 273dfeab2e9SGao Xiang 274a08e67a0SHuang Jianan iomap->offset = map.m_la; 275e33f42b2SGao Xiang if (flags & IOMAP_DAX) 276de205114SChristoph Hellwig iomap->dax_dev = mdev.m_daxdev; 277e33f42b2SGao Xiang else 278de205114SChristoph Hellwig iomap->bdev = mdev.m_bdev; 279a08e67a0SHuang Jianan iomap->length = map.m_llen; 280a08e67a0SHuang Jianan iomap->flags = 0; 281771c994eSGao Xiang iomap->private = NULL; 282a08e67a0SHuang Jianan 283a08e67a0SHuang Jianan if (!(map.m_flags & EROFS_MAP_MAPPED)) { 284a08e67a0SHuang Jianan iomap->type = IOMAP_HOLE; 285a08e67a0SHuang Jianan iomap->addr = IOMAP_NULL_ADDR; 286a08e67a0SHuang Jianan if (!iomap->length) 287a08e67a0SHuang Jianan iomap->length = length; 288a08e67a0SHuang Jianan return 0; 289a08e67a0SHuang Jianan } 290a08e67a0SHuang Jianan 291a08e67a0SHuang Jianan if (map.m_flags & EROFS_MAP_META) { 292fdf80a47SGao Xiang void *ptr; 293fdf80a47SGao Xiang struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 294771c994eSGao Xiang 295771c994eSGao Xiang iomap->type = IOMAP_INLINE; 296fdf80a47SGao Xiang ptr = erofs_read_metabuf(&buf, inode->i_sb, 297fdf80a47SGao Xiang erofs_blknr(mdev.m_pa), EROFS_KMAP); 298fdf80a47SGao Xiang if (IS_ERR(ptr)) 299fdf80a47SGao Xiang return PTR_ERR(ptr); 300fdf80a47SGao Xiang iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa); 301fdf80a47SGao Xiang iomap->private = buf.base; 302771c994eSGao Xiang } else { 303a08e67a0SHuang Jianan iomap->type = IOMAP_MAPPED; 304dfeab2e9SGao Xiang iomap->addr = mdev.m_pa; 305e33f42b2SGao Xiang if (flags & IOMAP_DAX) 306e33f42b2SGao Xiang iomap->addr += mdev.m_dax_part_off; 307771c994eSGao Xiang } 308a08e67a0SHuang Jianan return 0; 309a08e67a0SHuang Jianan } 310a08e67a0SHuang Jianan 311771c994eSGao Xiang static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length, 312771c994eSGao Xiang ssize_t written, unsigned int flags, struct iomap *iomap) 313771c994eSGao Xiang { 314fdf80a47SGao Xiang void *ptr = iomap->private; 315771c994eSGao Xiang 316fdf80a47SGao Xiang if (ptr) { 317fdf80a47SGao Xiang struct erofs_buf buf = { 318fdf80a47SGao Xiang .page = kmap_to_page(ptr), 319fdf80a47SGao Xiang .base = ptr, 320fdf80a47SGao Xiang .kmap_type = EROFS_KMAP, 321fdf80a47SGao Xiang }; 322fdf80a47SGao Xiang 323771c994eSGao Xiang DBG_BUGON(iomap->type != IOMAP_INLINE); 324fdf80a47SGao Xiang erofs_put_metabuf(&buf); 325771c994eSGao Xiang } else { 326771c994eSGao Xiang DBG_BUGON(iomap->type == IOMAP_INLINE); 327771c994eSGao Xiang } 328771c994eSGao Xiang return written; 329771c994eSGao Xiang } 330771c994eSGao Xiang 331a08e67a0SHuang Jianan static const struct iomap_ops erofs_iomap_ops = { 332a08e67a0SHuang Jianan .iomap_begin = erofs_iomap_begin, 333771c994eSGao Xiang .iomap_end = erofs_iomap_end, 334a08e67a0SHuang Jianan }; 335a08e67a0SHuang Jianan 336eadcd6b5SGao Xiang int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 337eadcd6b5SGao Xiang u64 start, u64 len) 338eadcd6b5SGao Xiang { 339eadcd6b5SGao Xiang if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 340eadcd6b5SGao Xiang #ifdef CONFIG_EROFS_FS_ZIP 341eadcd6b5SGao Xiang return iomap_fiemap(inode, fieinfo, start, len, 342eadcd6b5SGao Xiang &z_erofs_iomap_report_ops); 343eadcd6b5SGao Xiang #else 344eadcd6b5SGao Xiang return -EOPNOTSUPP; 345eadcd6b5SGao Xiang #endif 346eadcd6b5SGao Xiang } 347eadcd6b5SGao Xiang return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops); 348eadcd6b5SGao Xiang } 349eadcd6b5SGao Xiang 350771c994eSGao Xiang /* 351771c994eSGao Xiang * since we dont have write or truncate flows, so no inode 352771c994eSGao Xiang * locking needs to be held at the moment. 353771c994eSGao Xiang */ 3547479c505SMatthew Wilcox (Oracle) static int erofs_read_folio(struct file *file, struct folio *folio) 355771c994eSGao Xiang { 3567479c505SMatthew Wilcox (Oracle) return iomap_read_folio(folio, &erofs_iomap_ops); 357771c994eSGao Xiang } 358771c994eSGao Xiang 359771c994eSGao Xiang static void erofs_readahead(struct readahead_control *rac) 360771c994eSGao Xiang { 361771c994eSGao Xiang return iomap_readahead(rac, &erofs_iomap_ops); 362771c994eSGao Xiang } 363771c994eSGao Xiang 364771c994eSGao Xiang static sector_t erofs_bmap(struct address_space *mapping, sector_t block) 365771c994eSGao Xiang { 366771c994eSGao Xiang return iomap_bmap(mapping, block, &erofs_iomap_ops); 367771c994eSGao Xiang } 368771c994eSGao Xiang 369a08e67a0SHuang Jianan static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 370a08e67a0SHuang Jianan { 371de8a801aSGao Xiang struct inode *inode = file_inode(iocb->ki_filp); 372de8a801aSGao Xiang 373a08e67a0SHuang Jianan /* no need taking (shared) inode lock since it's a ro filesystem */ 374a08e67a0SHuang Jianan if (!iov_iter_count(to)) 375a08e67a0SHuang Jianan return 0; 376a08e67a0SHuang Jianan 37706252e9cSGao Xiang #ifdef CONFIG_FS_DAX 378de8a801aSGao Xiang if (IS_DAX(inode)) 37906252e9cSGao Xiang return dax_iomap_rw(iocb, to, &erofs_iomap_ops); 38006252e9cSGao Xiang #endif 381a08e67a0SHuang Jianan if (iocb->ki_flags & IOCB_DIRECT) { 382de8a801aSGao Xiang struct block_device *bdev = inode->i_sb->s_bdev; 383de8a801aSGao Xiang unsigned int blksize_mask; 384a08e67a0SHuang Jianan 385de8a801aSGao Xiang if (bdev) 386de8a801aSGao Xiang blksize_mask = bdev_logical_block_size(bdev) - 1; 387de8a801aSGao Xiang else 388de8a801aSGao Xiang blksize_mask = (1 << inode->i_blkbits) - 1; 389de8a801aSGao Xiang 390de8a801aSGao Xiang if ((iocb->ki_pos | iov_iter_count(to) | 391de8a801aSGao Xiang iov_iter_alignment(to)) & blksize_mask) 392de8a801aSGao Xiang return -EINVAL; 393de8a801aSGao Xiang 394a08e67a0SHuang Jianan return iomap_dio_rw(iocb, to, &erofs_iomap_ops, 395786f847fSChristoph Hellwig NULL, 0, NULL, 0); 396a08e67a0SHuang Jianan } 397a08e67a0SHuang Jianan return filemap_read(iocb, to, 0); 398a08e67a0SHuang Jianan } 399a08e67a0SHuang Jianan 40047e4937aSGao Xiang /* for uncompressed (aligned) files and raw access for other files */ 40147e4937aSGao Xiang const struct address_space_operations erofs_raw_access_aops = { 4027479c505SMatthew Wilcox (Oracle) .read_folio = erofs_read_folio, 403771c994eSGao Xiang .readahead = erofs_readahead, 40447e4937aSGao Xiang .bmap = erofs_bmap, 405a08e67a0SHuang Jianan .direct_IO = noop_direct_IO, 406*ce529cc2SJingbo Xu .release_folio = iomap_release_folio, 407*ce529cc2SJingbo Xu .invalidate_folio = iomap_invalidate_folio, 408a08e67a0SHuang Jianan }; 409a08e67a0SHuang Jianan 41006252e9cSGao Xiang #ifdef CONFIG_FS_DAX 41106252e9cSGao Xiang static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf, 41206252e9cSGao Xiang enum page_entry_size pe_size) 41306252e9cSGao Xiang { 41406252e9cSGao Xiang return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops); 41506252e9cSGao Xiang } 41606252e9cSGao Xiang 41706252e9cSGao Xiang static vm_fault_t erofs_dax_fault(struct vm_fault *vmf) 41806252e9cSGao Xiang { 41906252e9cSGao Xiang return erofs_dax_huge_fault(vmf, PE_SIZE_PTE); 42006252e9cSGao Xiang } 42106252e9cSGao Xiang 42206252e9cSGao Xiang static const struct vm_operations_struct erofs_dax_vm_ops = { 42306252e9cSGao Xiang .fault = erofs_dax_fault, 42406252e9cSGao Xiang .huge_fault = erofs_dax_huge_fault, 42506252e9cSGao Xiang }; 42606252e9cSGao Xiang 42706252e9cSGao Xiang static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma) 42806252e9cSGao Xiang { 42906252e9cSGao Xiang if (!IS_DAX(file_inode(file))) 43006252e9cSGao Xiang return generic_file_readonly_mmap(file, vma); 43106252e9cSGao Xiang 43206252e9cSGao Xiang if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 43306252e9cSGao Xiang return -EINVAL; 43406252e9cSGao Xiang 43506252e9cSGao Xiang vma->vm_ops = &erofs_dax_vm_ops; 43606252e9cSGao Xiang vma->vm_flags |= VM_HUGEPAGE; 43706252e9cSGao Xiang return 0; 43806252e9cSGao Xiang } 43906252e9cSGao Xiang #else 44006252e9cSGao Xiang #define erofs_file_mmap generic_file_readonly_mmap 44106252e9cSGao Xiang #endif 44206252e9cSGao Xiang 443a08e67a0SHuang Jianan const struct file_operations erofs_file_fops = { 444a08e67a0SHuang Jianan .llseek = generic_file_llseek, 445a08e67a0SHuang Jianan .read_iter = erofs_file_read_iter, 44606252e9cSGao Xiang .mmap = erofs_file_mmap, 447a08e67a0SHuang Jianan .splice_read = generic_file_splice_read, 44847e4937aSGao Xiang }; 449