1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2017-2018 HUAWEI, Inc. 4 * https://www.huawei.com/ 5 * Copyright (C) 2021, Alibaba Cloud 6 */ 7 #include "internal.h" 8 #include <linux/prefetch.h> 9 #include <linux/dax.h> 10 #include <trace/events/erofs.h> 11 12 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr) 13 { 14 struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping; 15 struct page *page; 16 17 page = read_cache_page_gfp(mapping, blkaddr, 18 mapping_gfp_constraint(mapping, ~__GFP_FS)); 19 /* should already be PageUptodate */ 20 if (!IS_ERR(page)) 21 lock_page(page); 22 return page; 23 } 24 25 static int erofs_map_blocks_flatmode(struct inode *inode, 26 struct erofs_map_blocks *map, 27 int flags) 28 { 29 int err = 0; 30 erofs_blk_t nblocks, lastblk; 31 u64 offset = map->m_la; 32 struct erofs_inode *vi = EROFS_I(inode); 33 bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); 34 35 trace_erofs_map_blocks_flatmode_enter(inode, map, flags); 36 37 nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 38 lastblk = nblocks - tailendpacking; 39 40 /* there is no hole in flatmode */ 41 map->m_flags = EROFS_MAP_MAPPED; 42 43 if (offset < blknr_to_addr(lastblk)) { 44 map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la; 45 map->m_plen = blknr_to_addr(lastblk) - offset; 46 } else if (tailendpacking) { 47 /* 2 - inode inline B: inode, [xattrs], inline last blk... */ 48 struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb); 49 50 map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize + 51 vi->xattr_isize + erofs_blkoff(map->m_la); 52 map->m_plen = inode->i_size - offset; 53 54 /* inline data should be located in one meta block */ 55 if (erofs_blkoff(map->m_pa) + map->m_plen > PAGE_SIZE) { 56 erofs_err(inode->i_sb, 57 "inline data cross block boundary @ nid %llu", 58 vi->nid); 59 DBG_BUGON(1); 60 err = -EFSCORRUPTED; 61 goto err_out; 62 } 63 64 map->m_flags |= EROFS_MAP_META; 65 } else { 66 erofs_err(inode->i_sb, 67 "internal error @ nid: %llu (size %llu), m_la 0x%llx", 68 vi->nid, inode->i_size, map->m_la); 69 DBG_BUGON(1); 70 err = -EIO; 71 goto err_out; 72 } 73 74 map->m_llen = map->m_plen; 75 err_out: 76 trace_erofs_map_blocks_flatmode_exit(inode, map, flags, 0); 77 return err; 78 } 79 80 static int erofs_map_blocks(struct inode *inode, 81 struct erofs_map_blocks *map, int flags) 82 { 83 struct super_block *sb = inode->i_sb; 84 struct erofs_inode *vi = EROFS_I(inode); 85 struct erofs_inode_chunk_index *idx; 86 struct page *page; 87 u64 chunknr; 88 unsigned int unit; 89 erofs_off_t pos; 90 int err = 0; 91 92 map->m_deviceid = 0; 93 if (map->m_la >= inode->i_size) { 94 /* leave out-of-bound access unmapped */ 95 map->m_flags = 0; 96 map->m_plen = 0; 97 goto out; 98 } 99 100 if (vi->datalayout != EROFS_INODE_CHUNK_BASED) 101 return erofs_map_blocks_flatmode(inode, map, flags); 102 103 if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) 104 unit = sizeof(*idx); /* chunk index */ 105 else 106 unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */ 107 108 chunknr = map->m_la >> vi->chunkbits; 109 pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize + 110 vi->xattr_isize, unit) + unit * chunknr; 111 112 page = erofs_get_meta_page(inode->i_sb, erofs_blknr(pos)); 113 if (IS_ERR(page)) 114 return PTR_ERR(page); 115 116 map->m_la = chunknr << vi->chunkbits; 117 map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits, 118 roundup(inode->i_size - map->m_la, EROFS_BLKSIZ)); 119 120 /* handle block map */ 121 if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) { 122 __le32 *blkaddr = page_address(page) + erofs_blkoff(pos); 123 124 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) { 125 map->m_flags = 0; 126 } else { 127 map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr)); 128 map->m_flags = EROFS_MAP_MAPPED; 129 } 130 goto out_unlock; 131 } 132 /* parse chunk indexes */ 133 idx = page_address(page) + erofs_blkoff(pos); 134 switch (le32_to_cpu(idx->blkaddr)) { 135 case EROFS_NULL_ADDR: 136 map->m_flags = 0; 137 break; 138 default: 139 map->m_deviceid = le16_to_cpu(idx->device_id) & 140 EROFS_SB(sb)->device_id_mask; 141 map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr)); 142 map->m_flags = EROFS_MAP_MAPPED; 143 break; 144 } 145 out_unlock: 146 unlock_page(page); 147 put_page(page); 148 out: 149 map->m_llen = map->m_plen; 150 return err; 151 } 152 153 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map) 154 { 155 struct erofs_dev_context *devs = EROFS_SB(sb)->devs; 156 struct erofs_device_info *dif; 157 int id; 158 159 /* primary device by default */ 160 map->m_bdev = sb->s_bdev; 161 map->m_daxdev = EROFS_SB(sb)->dax_dev; 162 map->m_dax_part_off = EROFS_SB(sb)->dax_part_off; 163 164 if (map->m_deviceid) { 165 down_read(&devs->rwsem); 166 dif = idr_find(&devs->tree, map->m_deviceid - 1); 167 if (!dif) { 168 up_read(&devs->rwsem); 169 return -ENODEV; 170 } 171 map->m_bdev = dif->bdev; 172 map->m_daxdev = dif->dax_dev; 173 map->m_dax_part_off = dif->dax_part_off; 174 up_read(&devs->rwsem); 175 } else if (devs->extra_devices) { 176 down_read(&devs->rwsem); 177 idr_for_each_entry(&devs->tree, dif, id) { 178 erofs_off_t startoff, length; 179 180 if (!dif->mapped_blkaddr) 181 continue; 182 startoff = blknr_to_addr(dif->mapped_blkaddr); 183 length = blknr_to_addr(dif->blocks); 184 185 if (map->m_pa >= startoff && 186 map->m_pa < startoff + length) { 187 map->m_pa -= startoff; 188 map->m_bdev = dif->bdev; 189 map->m_daxdev = dif->dax_dev; 190 map->m_dax_part_off = dif->dax_part_off; 191 break; 192 } 193 } 194 up_read(&devs->rwsem); 195 } 196 return 0; 197 } 198 199 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 200 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 201 { 202 int ret; 203 struct erofs_map_blocks map; 204 struct erofs_map_dev mdev; 205 206 map.m_la = offset; 207 map.m_llen = length; 208 209 ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); 210 if (ret < 0) 211 return ret; 212 213 mdev = (struct erofs_map_dev) { 214 .m_deviceid = map.m_deviceid, 215 .m_pa = map.m_pa, 216 }; 217 ret = erofs_map_dev(inode->i_sb, &mdev); 218 if (ret) 219 return ret; 220 221 iomap->offset = map.m_la; 222 if (flags & IOMAP_DAX) { 223 iomap->dax_dev = mdev.m_daxdev; 224 iomap->offset += mdev.m_dax_part_off; 225 } else { 226 iomap->bdev = mdev.m_bdev; 227 } 228 iomap->length = map.m_llen; 229 iomap->flags = 0; 230 iomap->private = NULL; 231 232 if (!(map.m_flags & EROFS_MAP_MAPPED)) { 233 iomap->type = IOMAP_HOLE; 234 iomap->addr = IOMAP_NULL_ADDR; 235 if (!iomap->length) 236 iomap->length = length; 237 return 0; 238 } 239 240 if (map.m_flags & EROFS_MAP_META) { 241 struct page *ipage; 242 243 iomap->type = IOMAP_INLINE; 244 ipage = erofs_get_meta_page(inode->i_sb, 245 erofs_blknr(mdev.m_pa)); 246 if (IS_ERR(ipage)) 247 return PTR_ERR(ipage); 248 iomap->inline_data = page_address(ipage) + 249 erofs_blkoff(mdev.m_pa); 250 iomap->private = ipage; 251 } else { 252 iomap->type = IOMAP_MAPPED; 253 iomap->addr = mdev.m_pa; 254 } 255 return 0; 256 } 257 258 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length, 259 ssize_t written, unsigned int flags, struct iomap *iomap) 260 { 261 struct page *ipage = iomap->private; 262 263 if (ipage) { 264 DBG_BUGON(iomap->type != IOMAP_INLINE); 265 unlock_page(ipage); 266 put_page(ipage); 267 } else { 268 DBG_BUGON(iomap->type == IOMAP_INLINE); 269 } 270 return written; 271 } 272 273 static const struct iomap_ops erofs_iomap_ops = { 274 .iomap_begin = erofs_iomap_begin, 275 .iomap_end = erofs_iomap_end, 276 }; 277 278 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 279 u64 start, u64 len) 280 { 281 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 282 #ifdef CONFIG_EROFS_FS_ZIP 283 return iomap_fiemap(inode, fieinfo, start, len, 284 &z_erofs_iomap_report_ops); 285 #else 286 return -EOPNOTSUPP; 287 #endif 288 } 289 return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops); 290 } 291 292 /* 293 * since we dont have write or truncate flows, so no inode 294 * locking needs to be held at the moment. 295 */ 296 static int erofs_readpage(struct file *file, struct page *page) 297 { 298 return iomap_readpage(page, &erofs_iomap_ops); 299 } 300 301 static void erofs_readahead(struct readahead_control *rac) 302 { 303 return iomap_readahead(rac, &erofs_iomap_ops); 304 } 305 306 static sector_t erofs_bmap(struct address_space *mapping, sector_t block) 307 { 308 return iomap_bmap(mapping, block, &erofs_iomap_ops); 309 } 310 311 static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to) 312 { 313 struct inode *inode = file_inode(iocb->ki_filp); 314 loff_t align = iocb->ki_pos | iov_iter_count(to) | 315 iov_iter_alignment(to); 316 struct block_device *bdev = inode->i_sb->s_bdev; 317 unsigned int blksize_mask; 318 319 if (bdev) 320 blksize_mask = (1 << ilog2(bdev_logical_block_size(bdev))) - 1; 321 else 322 blksize_mask = (1 << inode->i_blkbits) - 1; 323 324 if (align & blksize_mask) 325 return -EINVAL; 326 return 0; 327 } 328 329 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 330 { 331 /* no need taking (shared) inode lock since it's a ro filesystem */ 332 if (!iov_iter_count(to)) 333 return 0; 334 335 #ifdef CONFIG_FS_DAX 336 if (IS_DAX(iocb->ki_filp->f_mapping->host)) 337 return dax_iomap_rw(iocb, to, &erofs_iomap_ops); 338 #endif 339 if (iocb->ki_flags & IOCB_DIRECT) { 340 int err = erofs_prepare_dio(iocb, to); 341 342 if (!err) 343 return iomap_dio_rw(iocb, to, &erofs_iomap_ops, 344 NULL, 0, 0); 345 if (err < 0) 346 return err; 347 } 348 return filemap_read(iocb, to, 0); 349 } 350 351 /* for uncompressed (aligned) files and raw access for other files */ 352 const struct address_space_operations erofs_raw_access_aops = { 353 .readpage = erofs_readpage, 354 .readahead = erofs_readahead, 355 .bmap = erofs_bmap, 356 .direct_IO = noop_direct_IO, 357 }; 358 359 #ifdef CONFIG_FS_DAX 360 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf, 361 enum page_entry_size pe_size) 362 { 363 return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops); 364 } 365 366 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf) 367 { 368 return erofs_dax_huge_fault(vmf, PE_SIZE_PTE); 369 } 370 371 static const struct vm_operations_struct erofs_dax_vm_ops = { 372 .fault = erofs_dax_fault, 373 .huge_fault = erofs_dax_huge_fault, 374 }; 375 376 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma) 377 { 378 if (!IS_DAX(file_inode(file))) 379 return generic_file_readonly_mmap(file, vma); 380 381 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 382 return -EINVAL; 383 384 vma->vm_ops = &erofs_dax_vm_ops; 385 vma->vm_flags |= VM_HUGEPAGE; 386 return 0; 387 } 388 #else 389 #define erofs_file_mmap generic_file_readonly_mmap 390 #endif 391 392 const struct file_operations erofs_file_fops = { 393 .llseek = generic_file_llseek, 394 .read_iter = erofs_file_read_iter, 395 .mmap = erofs_file_mmap, 396 .splice_read = generic_file_splice_read, 397 }; 398