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> 906252e9cSGao Xiang #include <linux/dax.h> 1047e4937aSGao Xiang #include <trace/events/erofs.h> 1147e4937aSGao Xiang 12fdf80a47SGao Xiang void erofs_unmap_metabuf(struct erofs_buf *buf) 13fdf80a47SGao Xiang { 14fdf80a47SGao Xiang if (buf->kmap_type == EROFS_KMAP) 15fdf80a47SGao Xiang kunmap(buf->page); 16fdf80a47SGao Xiang else if (buf->kmap_type == EROFS_KMAP_ATOMIC) 17fdf80a47SGao Xiang kunmap_atomic(buf->base); 18fdf80a47SGao Xiang buf->base = NULL; 19fdf80a47SGao Xiang buf->kmap_type = EROFS_NO_KMAP; 20fdf80a47SGao Xiang } 21fdf80a47SGao Xiang 22fdf80a47SGao Xiang void erofs_put_metabuf(struct erofs_buf *buf) 23fdf80a47SGao Xiang { 24fdf80a47SGao Xiang if (!buf->page) 25fdf80a47SGao Xiang return; 26fdf80a47SGao Xiang erofs_unmap_metabuf(buf); 27fdf80a47SGao Xiang put_page(buf->page); 28fdf80a47SGao Xiang buf->page = NULL; 29fdf80a47SGao Xiang } 30fdf80a47SGao Xiang 31*fe5de585SGao Xiang void *erofs_bread(struct erofs_buf *buf, struct inode *inode, 32fdf80a47SGao Xiang erofs_blk_t blkaddr, enum erofs_kmap_type type) 33fdf80a47SGao Xiang { 34*fe5de585SGao Xiang struct address_space *const mapping = inode->i_mapping; 35fdf80a47SGao Xiang erofs_off_t offset = blknr_to_addr(blkaddr); 36fdf80a47SGao Xiang pgoff_t index = offset >> PAGE_SHIFT; 37fdf80a47SGao Xiang struct page *page = buf->page; 38fdf80a47SGao Xiang 39fdf80a47SGao Xiang if (!page || page->index != index) { 40fdf80a47SGao Xiang erofs_put_metabuf(buf); 41fdf80a47SGao Xiang page = read_cache_page_gfp(mapping, index, 42fdf80a47SGao Xiang mapping_gfp_constraint(mapping, ~__GFP_FS)); 43fdf80a47SGao Xiang if (IS_ERR(page)) 44fdf80a47SGao Xiang return page; 45fdf80a47SGao Xiang /* should already be PageUptodate, no need to lock page */ 46fdf80a47SGao Xiang buf->page = page; 47fdf80a47SGao Xiang } 48fdf80a47SGao Xiang if (buf->kmap_type == EROFS_NO_KMAP) { 49fdf80a47SGao Xiang if (type == EROFS_KMAP) 50fdf80a47SGao Xiang buf->base = kmap(page); 51fdf80a47SGao Xiang else if (type == EROFS_KMAP_ATOMIC) 52fdf80a47SGao Xiang buf->base = kmap_atomic(page); 53fdf80a47SGao Xiang buf->kmap_type = type; 54fdf80a47SGao Xiang } else if (buf->kmap_type != type) { 55fdf80a47SGao Xiang DBG_BUGON(1); 56fdf80a47SGao Xiang return ERR_PTR(-EFAULT); 57fdf80a47SGao Xiang } 58fdf80a47SGao Xiang if (type == EROFS_NO_KMAP) 59fdf80a47SGao Xiang return NULL; 60fdf80a47SGao Xiang return buf->base + (offset & ~PAGE_MASK); 61fdf80a47SGao Xiang } 62fdf80a47SGao Xiang 63*fe5de585SGao Xiang void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb, 64*fe5de585SGao Xiang erofs_blk_t blkaddr, enum erofs_kmap_type type) 65*fe5de585SGao Xiang { 66*fe5de585SGao Xiang return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type); 67*fe5de585SGao Xiang } 68*fe5de585SGao Xiang 6947e4937aSGao Xiang static int erofs_map_blocks_flatmode(struct inode *inode, 7047e4937aSGao Xiang struct erofs_map_blocks *map, 7147e4937aSGao Xiang int flags) 7247e4937aSGao Xiang { 7347e4937aSGao Xiang erofs_blk_t nblocks, lastblk; 7447e4937aSGao Xiang u64 offset = map->m_la; 75a5876e24SGao Xiang struct erofs_inode *vi = EROFS_I(inode); 768a765682SGao Xiang bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); 7747e4937aSGao Xiang 78fdf80a47SGao Xiang nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ); 798a765682SGao Xiang lastblk = nblocks - tailendpacking; 8047e4937aSGao Xiang 8147e4937aSGao Xiang /* there is no hole in flatmode */ 8247e4937aSGao Xiang map->m_flags = EROFS_MAP_MAPPED; 8347e4937aSGao Xiang if (offset < blknr_to_addr(lastblk)) { 8447e4937aSGao Xiang map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la; 8547e4937aSGao Xiang map->m_plen = blknr_to_addr(lastblk) - offset; 868a765682SGao Xiang } else if (tailendpacking) { 8747e4937aSGao Xiang /* 2 - inode inline B: inode, [xattrs], inline last blk... */ 8847e4937aSGao Xiang struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb); 8947e4937aSGao Xiang 9047e4937aSGao Xiang map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize + 9147e4937aSGao Xiang vi->xattr_isize + erofs_blkoff(map->m_la); 9247e4937aSGao Xiang map->m_plen = inode->i_size - offset; 9347e4937aSGao Xiang 94469407a3SGao Xiang /* inline data should be located in the same meta block */ 95469407a3SGao Xiang if (erofs_blkoff(map->m_pa) + map->m_plen > EROFS_BLKSIZ) { 964f761fa2SGao Xiang erofs_err(inode->i_sb, 974f761fa2SGao Xiang "inline data cross block boundary @ nid %llu", 9847e4937aSGao Xiang vi->nid); 9947e4937aSGao Xiang DBG_BUGON(1); 100469407a3SGao Xiang return -EFSCORRUPTED; 10147e4937aSGao Xiang } 10247e4937aSGao Xiang map->m_flags |= EROFS_MAP_META; 10347e4937aSGao Xiang } else { 1044f761fa2SGao Xiang erofs_err(inode->i_sb, 1054f761fa2SGao Xiang "internal error @ nid: %llu (size %llu), m_la 0x%llx", 10647e4937aSGao Xiang vi->nid, inode->i_size, map->m_la); 10747e4937aSGao Xiang DBG_BUGON(1); 108469407a3SGao Xiang return -EIO; 10947e4937aSGao Xiang } 110469407a3SGao Xiang return 0; 11147e4937aSGao Xiang } 11247e4937aSGao Xiang 113c5aa903aSGao Xiang static int erofs_map_blocks(struct inode *inode, 114c5aa903aSGao Xiang struct erofs_map_blocks *map, int flags) 115c5aa903aSGao Xiang { 116c5aa903aSGao Xiang struct super_block *sb = inode->i_sb; 117c5aa903aSGao Xiang struct erofs_inode *vi = EROFS_I(inode); 118c5aa903aSGao Xiang struct erofs_inode_chunk_index *idx; 119fdf80a47SGao Xiang struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 120c5aa903aSGao Xiang u64 chunknr; 121c5aa903aSGao Xiang unsigned int unit; 122c5aa903aSGao Xiang erofs_off_t pos; 123fdf80a47SGao Xiang void *kaddr; 124c5aa903aSGao Xiang int err = 0; 125c5aa903aSGao Xiang 126469407a3SGao Xiang trace_erofs_map_blocks_enter(inode, map, flags); 127dfeab2e9SGao Xiang map->m_deviceid = 0; 128c5aa903aSGao Xiang if (map->m_la >= inode->i_size) { 129c5aa903aSGao Xiang /* leave out-of-bound access unmapped */ 130c5aa903aSGao Xiang map->m_flags = 0; 131c5aa903aSGao Xiang map->m_plen = 0; 132c5aa903aSGao Xiang goto out; 133c5aa903aSGao Xiang } 134c5aa903aSGao Xiang 135469407a3SGao Xiang if (vi->datalayout != EROFS_INODE_CHUNK_BASED) { 136469407a3SGao Xiang err = erofs_map_blocks_flatmode(inode, map, flags); 137469407a3SGao Xiang goto out; 138469407a3SGao Xiang } 139c5aa903aSGao Xiang 140c5aa903aSGao Xiang if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) 141c5aa903aSGao Xiang unit = sizeof(*idx); /* chunk index */ 142c5aa903aSGao Xiang else 143c5aa903aSGao Xiang unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */ 144c5aa903aSGao Xiang 145c5aa903aSGao Xiang chunknr = map->m_la >> vi->chunkbits; 146c5aa903aSGao Xiang pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize + 147c5aa903aSGao Xiang vi->xattr_isize, unit) + unit * chunknr; 148c5aa903aSGao Xiang 149fdf80a47SGao Xiang kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP); 150fdf80a47SGao Xiang if (IS_ERR(kaddr)) { 151fdf80a47SGao Xiang err = PTR_ERR(kaddr); 152469407a3SGao Xiang goto out; 153469407a3SGao Xiang } 154c5aa903aSGao Xiang map->m_la = chunknr << vi->chunkbits; 155c5aa903aSGao Xiang map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits, 156c5aa903aSGao Xiang roundup(inode->i_size - map->m_la, EROFS_BLKSIZ)); 157c5aa903aSGao Xiang 158c5aa903aSGao Xiang /* handle block map */ 159c5aa903aSGao Xiang if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) { 160fdf80a47SGao Xiang __le32 *blkaddr = kaddr + erofs_blkoff(pos); 161c5aa903aSGao Xiang 162c5aa903aSGao Xiang if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) { 163c5aa903aSGao Xiang map->m_flags = 0; 164c5aa903aSGao Xiang } else { 165c5aa903aSGao Xiang map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr)); 166c5aa903aSGao Xiang map->m_flags = EROFS_MAP_MAPPED; 167c5aa903aSGao Xiang } 168c5aa903aSGao Xiang goto out_unlock; 169c5aa903aSGao Xiang } 170c5aa903aSGao Xiang /* parse chunk indexes */ 171fdf80a47SGao Xiang idx = kaddr + erofs_blkoff(pos); 172c5aa903aSGao Xiang switch (le32_to_cpu(idx->blkaddr)) { 173c5aa903aSGao Xiang case EROFS_NULL_ADDR: 174c5aa903aSGao Xiang map->m_flags = 0; 175c5aa903aSGao Xiang break; 176c5aa903aSGao Xiang default: 177dfeab2e9SGao Xiang map->m_deviceid = le16_to_cpu(idx->device_id) & 178dfeab2e9SGao Xiang EROFS_SB(sb)->device_id_mask; 179c5aa903aSGao Xiang map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr)); 180c5aa903aSGao Xiang map->m_flags = EROFS_MAP_MAPPED; 181c5aa903aSGao Xiang break; 182c5aa903aSGao Xiang } 183c5aa903aSGao Xiang out_unlock: 184fdf80a47SGao Xiang erofs_put_metabuf(&buf); 185c5aa903aSGao Xiang out: 186469407a3SGao Xiang if (!err) 187c5aa903aSGao Xiang map->m_llen = map->m_plen; 188469407a3SGao Xiang trace_erofs_map_blocks_exit(inode, map, flags, 0); 189c5aa903aSGao Xiang return err; 190c5aa903aSGao Xiang } 191c5aa903aSGao Xiang 192dfeab2e9SGao Xiang int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map) 193dfeab2e9SGao Xiang { 194dfeab2e9SGao Xiang struct erofs_dev_context *devs = EROFS_SB(sb)->devs; 195dfeab2e9SGao Xiang struct erofs_device_info *dif; 196dfeab2e9SGao Xiang int id; 197dfeab2e9SGao Xiang 198dfeab2e9SGao Xiang /* primary device by default */ 199dfeab2e9SGao Xiang map->m_bdev = sb->s_bdev; 200dfeab2e9SGao Xiang map->m_daxdev = EROFS_SB(sb)->dax_dev; 201de205114SChristoph Hellwig map->m_dax_part_off = EROFS_SB(sb)->dax_part_off; 202dfeab2e9SGao Xiang 203dfeab2e9SGao Xiang if (map->m_deviceid) { 204dfeab2e9SGao Xiang down_read(&devs->rwsem); 205dfeab2e9SGao Xiang dif = idr_find(&devs->tree, map->m_deviceid - 1); 206dfeab2e9SGao Xiang if (!dif) { 207dfeab2e9SGao Xiang up_read(&devs->rwsem); 208dfeab2e9SGao Xiang return -ENODEV; 209dfeab2e9SGao Xiang } 210dfeab2e9SGao Xiang map->m_bdev = dif->bdev; 211dfeab2e9SGao Xiang map->m_daxdev = dif->dax_dev; 212de205114SChristoph Hellwig map->m_dax_part_off = dif->dax_part_off; 213dfeab2e9SGao Xiang up_read(&devs->rwsem); 214dfeab2e9SGao Xiang } else if (devs->extra_devices) { 215dfeab2e9SGao Xiang down_read(&devs->rwsem); 216dfeab2e9SGao Xiang idr_for_each_entry(&devs->tree, dif, id) { 217dfeab2e9SGao Xiang erofs_off_t startoff, length; 218dfeab2e9SGao Xiang 219dfeab2e9SGao Xiang if (!dif->mapped_blkaddr) 220dfeab2e9SGao Xiang continue; 221dfeab2e9SGao Xiang startoff = blknr_to_addr(dif->mapped_blkaddr); 222dfeab2e9SGao Xiang length = blknr_to_addr(dif->blocks); 223dfeab2e9SGao Xiang 224dfeab2e9SGao Xiang if (map->m_pa >= startoff && 225dfeab2e9SGao Xiang map->m_pa < startoff + length) { 226dfeab2e9SGao Xiang map->m_pa -= startoff; 227dfeab2e9SGao Xiang map->m_bdev = dif->bdev; 228dfeab2e9SGao Xiang map->m_daxdev = dif->dax_dev; 229de205114SChristoph Hellwig map->m_dax_part_off = dif->dax_part_off; 230dfeab2e9SGao Xiang break; 231dfeab2e9SGao Xiang } 232dfeab2e9SGao Xiang } 233dfeab2e9SGao Xiang up_read(&devs->rwsem); 234dfeab2e9SGao Xiang } 235dfeab2e9SGao Xiang return 0; 236dfeab2e9SGao Xiang } 237dfeab2e9SGao Xiang 238a08e67a0SHuang Jianan static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 239a08e67a0SHuang Jianan unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 240a08e67a0SHuang Jianan { 241a08e67a0SHuang Jianan int ret; 242a08e67a0SHuang Jianan struct erofs_map_blocks map; 243dfeab2e9SGao Xiang struct erofs_map_dev mdev; 244a08e67a0SHuang Jianan 245a08e67a0SHuang Jianan map.m_la = offset; 246a08e67a0SHuang Jianan map.m_llen = length; 247a08e67a0SHuang Jianan 248c5aa903aSGao Xiang ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); 249a08e67a0SHuang Jianan if (ret < 0) 250a08e67a0SHuang Jianan return ret; 251a08e67a0SHuang Jianan 252dfeab2e9SGao Xiang mdev = (struct erofs_map_dev) { 253dfeab2e9SGao Xiang .m_deviceid = map.m_deviceid, 254dfeab2e9SGao Xiang .m_pa = map.m_pa, 255dfeab2e9SGao Xiang }; 256dfeab2e9SGao Xiang ret = erofs_map_dev(inode->i_sb, &mdev); 257dfeab2e9SGao Xiang if (ret) 258dfeab2e9SGao Xiang return ret; 259dfeab2e9SGao Xiang 260a08e67a0SHuang Jianan iomap->offset = map.m_la; 261e33f42b2SGao Xiang if (flags & IOMAP_DAX) 262de205114SChristoph Hellwig iomap->dax_dev = mdev.m_daxdev; 263e33f42b2SGao Xiang else 264de205114SChristoph Hellwig iomap->bdev = mdev.m_bdev; 265a08e67a0SHuang Jianan iomap->length = map.m_llen; 266a08e67a0SHuang Jianan iomap->flags = 0; 267771c994eSGao Xiang iomap->private = NULL; 268a08e67a0SHuang Jianan 269a08e67a0SHuang Jianan if (!(map.m_flags & EROFS_MAP_MAPPED)) { 270a08e67a0SHuang Jianan iomap->type = IOMAP_HOLE; 271a08e67a0SHuang Jianan iomap->addr = IOMAP_NULL_ADDR; 272a08e67a0SHuang Jianan if (!iomap->length) 273a08e67a0SHuang Jianan iomap->length = length; 274a08e67a0SHuang Jianan return 0; 275a08e67a0SHuang Jianan } 276a08e67a0SHuang Jianan 277a08e67a0SHuang Jianan if (map.m_flags & EROFS_MAP_META) { 278fdf80a47SGao Xiang void *ptr; 279fdf80a47SGao Xiang struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 280771c994eSGao Xiang 281771c994eSGao Xiang iomap->type = IOMAP_INLINE; 282fdf80a47SGao Xiang ptr = erofs_read_metabuf(&buf, inode->i_sb, 283fdf80a47SGao Xiang erofs_blknr(mdev.m_pa), EROFS_KMAP); 284fdf80a47SGao Xiang if (IS_ERR(ptr)) 285fdf80a47SGao Xiang return PTR_ERR(ptr); 286fdf80a47SGao Xiang iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa); 287fdf80a47SGao Xiang iomap->private = buf.base; 288771c994eSGao Xiang } else { 289a08e67a0SHuang Jianan iomap->type = IOMAP_MAPPED; 290dfeab2e9SGao Xiang iomap->addr = mdev.m_pa; 291e33f42b2SGao Xiang if (flags & IOMAP_DAX) 292e33f42b2SGao Xiang iomap->addr += mdev.m_dax_part_off; 293771c994eSGao Xiang } 294a08e67a0SHuang Jianan return 0; 295a08e67a0SHuang Jianan } 296a08e67a0SHuang Jianan 297771c994eSGao Xiang static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length, 298771c994eSGao Xiang ssize_t written, unsigned int flags, struct iomap *iomap) 299771c994eSGao Xiang { 300fdf80a47SGao Xiang void *ptr = iomap->private; 301771c994eSGao Xiang 302fdf80a47SGao Xiang if (ptr) { 303fdf80a47SGao Xiang struct erofs_buf buf = { 304fdf80a47SGao Xiang .page = kmap_to_page(ptr), 305fdf80a47SGao Xiang .base = ptr, 306fdf80a47SGao Xiang .kmap_type = EROFS_KMAP, 307fdf80a47SGao Xiang }; 308fdf80a47SGao Xiang 309771c994eSGao Xiang DBG_BUGON(iomap->type != IOMAP_INLINE); 310fdf80a47SGao Xiang erofs_put_metabuf(&buf); 311771c994eSGao Xiang } else { 312771c994eSGao Xiang DBG_BUGON(iomap->type == IOMAP_INLINE); 313771c994eSGao Xiang } 314771c994eSGao Xiang return written; 315771c994eSGao Xiang } 316771c994eSGao Xiang 317a08e67a0SHuang Jianan static const struct iomap_ops erofs_iomap_ops = { 318a08e67a0SHuang Jianan .iomap_begin = erofs_iomap_begin, 319771c994eSGao Xiang .iomap_end = erofs_iomap_end, 320a08e67a0SHuang Jianan }; 321a08e67a0SHuang Jianan 322eadcd6b5SGao Xiang int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 323eadcd6b5SGao Xiang u64 start, u64 len) 324eadcd6b5SGao Xiang { 325eadcd6b5SGao Xiang if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 326eadcd6b5SGao Xiang #ifdef CONFIG_EROFS_FS_ZIP 327eadcd6b5SGao Xiang return iomap_fiemap(inode, fieinfo, start, len, 328eadcd6b5SGao Xiang &z_erofs_iomap_report_ops); 329eadcd6b5SGao Xiang #else 330eadcd6b5SGao Xiang return -EOPNOTSUPP; 331eadcd6b5SGao Xiang #endif 332eadcd6b5SGao Xiang } 333eadcd6b5SGao Xiang return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops); 334eadcd6b5SGao Xiang } 335eadcd6b5SGao Xiang 336771c994eSGao Xiang /* 337771c994eSGao Xiang * since we dont have write or truncate flows, so no inode 338771c994eSGao Xiang * locking needs to be held at the moment. 339771c994eSGao Xiang */ 340771c994eSGao Xiang static int erofs_readpage(struct file *file, struct page *page) 341771c994eSGao Xiang { 342771c994eSGao Xiang return iomap_readpage(page, &erofs_iomap_ops); 343771c994eSGao Xiang } 344771c994eSGao Xiang 345771c994eSGao Xiang static void erofs_readahead(struct readahead_control *rac) 346771c994eSGao Xiang { 347771c994eSGao Xiang return iomap_readahead(rac, &erofs_iomap_ops); 348771c994eSGao Xiang } 349771c994eSGao Xiang 350771c994eSGao Xiang static sector_t erofs_bmap(struct address_space *mapping, sector_t block) 351771c994eSGao Xiang { 352771c994eSGao Xiang return iomap_bmap(mapping, block, &erofs_iomap_ops); 353771c994eSGao Xiang } 354771c994eSGao Xiang 355a08e67a0SHuang Jianan static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to) 356a08e67a0SHuang Jianan { 357a08e67a0SHuang Jianan struct inode *inode = file_inode(iocb->ki_filp); 358a08e67a0SHuang Jianan loff_t align = iocb->ki_pos | iov_iter_count(to) | 359a08e67a0SHuang Jianan iov_iter_alignment(to); 360a08e67a0SHuang Jianan struct block_device *bdev = inode->i_sb->s_bdev; 361a08e67a0SHuang Jianan unsigned int blksize_mask; 362a08e67a0SHuang Jianan 363a08e67a0SHuang Jianan if (bdev) 364a08e67a0SHuang Jianan blksize_mask = (1 << ilog2(bdev_logical_block_size(bdev))) - 1; 365a08e67a0SHuang Jianan else 366a08e67a0SHuang Jianan blksize_mask = (1 << inode->i_blkbits) - 1; 367a08e67a0SHuang Jianan 368a08e67a0SHuang Jianan if (align & blksize_mask) 369a08e67a0SHuang Jianan return -EINVAL; 370a08e67a0SHuang Jianan return 0; 371a08e67a0SHuang Jianan } 372a08e67a0SHuang Jianan 373a08e67a0SHuang Jianan static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 374a08e67a0SHuang Jianan { 375a08e67a0SHuang Jianan /* no need taking (shared) inode lock since it's a ro filesystem */ 376a08e67a0SHuang Jianan if (!iov_iter_count(to)) 377a08e67a0SHuang Jianan return 0; 378a08e67a0SHuang Jianan 37906252e9cSGao Xiang #ifdef CONFIG_FS_DAX 38006252e9cSGao Xiang if (IS_DAX(iocb->ki_filp->f_mapping->host)) 38106252e9cSGao Xiang return dax_iomap_rw(iocb, to, &erofs_iomap_ops); 38206252e9cSGao Xiang #endif 383a08e67a0SHuang Jianan if (iocb->ki_flags & IOCB_DIRECT) { 384a08e67a0SHuang Jianan int err = erofs_prepare_dio(iocb, to); 385a08e67a0SHuang Jianan 386a08e67a0SHuang Jianan if (!err) 387a08e67a0SHuang Jianan return iomap_dio_rw(iocb, to, &erofs_iomap_ops, 3884fdccaa0SAndreas Gruenbacher NULL, 0, 0); 389a08e67a0SHuang Jianan if (err < 0) 390a08e67a0SHuang Jianan return err; 391a08e67a0SHuang Jianan } 392a08e67a0SHuang Jianan return filemap_read(iocb, to, 0); 393a08e67a0SHuang Jianan } 394a08e67a0SHuang Jianan 39547e4937aSGao Xiang /* for uncompressed (aligned) files and raw access for other files */ 39647e4937aSGao Xiang const struct address_space_operations erofs_raw_access_aops = { 397771c994eSGao Xiang .readpage = erofs_readpage, 398771c994eSGao Xiang .readahead = erofs_readahead, 39947e4937aSGao Xiang .bmap = erofs_bmap, 400a08e67a0SHuang Jianan .direct_IO = noop_direct_IO, 401a08e67a0SHuang Jianan }; 402a08e67a0SHuang Jianan 40306252e9cSGao Xiang #ifdef CONFIG_FS_DAX 40406252e9cSGao Xiang static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf, 40506252e9cSGao Xiang enum page_entry_size pe_size) 40606252e9cSGao Xiang { 40706252e9cSGao Xiang return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops); 40806252e9cSGao Xiang } 40906252e9cSGao Xiang 41006252e9cSGao Xiang static vm_fault_t erofs_dax_fault(struct vm_fault *vmf) 41106252e9cSGao Xiang { 41206252e9cSGao Xiang return erofs_dax_huge_fault(vmf, PE_SIZE_PTE); 41306252e9cSGao Xiang } 41406252e9cSGao Xiang 41506252e9cSGao Xiang static const struct vm_operations_struct erofs_dax_vm_ops = { 41606252e9cSGao Xiang .fault = erofs_dax_fault, 41706252e9cSGao Xiang .huge_fault = erofs_dax_huge_fault, 41806252e9cSGao Xiang }; 41906252e9cSGao Xiang 42006252e9cSGao Xiang static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma) 42106252e9cSGao Xiang { 42206252e9cSGao Xiang if (!IS_DAX(file_inode(file))) 42306252e9cSGao Xiang return generic_file_readonly_mmap(file, vma); 42406252e9cSGao Xiang 42506252e9cSGao Xiang if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 42606252e9cSGao Xiang return -EINVAL; 42706252e9cSGao Xiang 42806252e9cSGao Xiang vma->vm_ops = &erofs_dax_vm_ops; 42906252e9cSGao Xiang vma->vm_flags |= VM_HUGEPAGE; 43006252e9cSGao Xiang return 0; 43106252e9cSGao Xiang } 43206252e9cSGao Xiang #else 43306252e9cSGao Xiang #define erofs_file_mmap generic_file_readonly_mmap 43406252e9cSGao Xiang #endif 43506252e9cSGao Xiang 436a08e67a0SHuang Jianan const struct file_operations erofs_file_fops = { 437a08e67a0SHuang Jianan .llseek = generic_file_llseek, 438a08e67a0SHuang Jianan .read_iter = erofs_file_read_iter, 43906252e9cSGao Xiang .mmap = erofs_file_mmap, 440a08e67a0SHuang Jianan .splice_read = generic_file_splice_read, 44147e4937aSGao Xiang }; 442