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/sched/mm.h> 10 #include <linux/dax.h> 11 #include <trace/events/erofs.h> 12 13 void erofs_unmap_metabuf(struct erofs_buf *buf) 14 { 15 if (buf->kmap_type == EROFS_KMAP) 16 kunmap_local(buf->base); 17 buf->base = NULL; 18 buf->kmap_type = EROFS_NO_KMAP; 19 } 20 21 void erofs_put_metabuf(struct erofs_buf *buf) 22 { 23 if (!buf->page) 24 return; 25 erofs_unmap_metabuf(buf); 26 put_page(buf->page); 27 buf->page = NULL; 28 } 29 30 /* 31 * Derive the block size from inode->i_blkbits to make compatible with 32 * anonymous inode in fscache mode. 33 */ 34 void *erofs_bread(struct erofs_buf *buf, struct inode *inode, 35 erofs_blk_t blkaddr, enum erofs_kmap_type type) 36 { 37 erofs_off_t offset = (erofs_off_t)blkaddr << inode->i_blkbits; 38 struct address_space *const mapping = inode->i_mapping; 39 pgoff_t index = offset >> PAGE_SHIFT; 40 struct page *page = buf->page; 41 struct folio *folio; 42 unsigned int nofs_flag; 43 44 if (!page || page->index != index) { 45 erofs_put_metabuf(buf); 46 47 nofs_flag = memalloc_nofs_save(); 48 folio = read_cache_folio(mapping, index, NULL, NULL); 49 memalloc_nofs_restore(nofs_flag); 50 if (IS_ERR(folio)) 51 return folio; 52 53 /* should already be PageUptodate, no need to lock page */ 54 page = folio_file_page(folio, index); 55 buf->page = page; 56 } 57 if (buf->kmap_type == EROFS_NO_KMAP) { 58 if (type == EROFS_KMAP) 59 buf->base = kmap_local_page(page); 60 buf->kmap_type = type; 61 } else if (buf->kmap_type != type) { 62 DBG_BUGON(1); 63 return ERR_PTR(-EFAULT); 64 } 65 if (type == EROFS_NO_KMAP) 66 return NULL; 67 return buf->base + (offset & ~PAGE_MASK); 68 } 69 70 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb, 71 erofs_blk_t blkaddr, enum erofs_kmap_type type) 72 { 73 if (erofs_is_fscache_mode(sb)) 74 return erofs_bread(buf, EROFS_SB(sb)->s_fscache->inode, 75 blkaddr, type); 76 77 return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type); 78 } 79 80 static int erofs_map_blocks_flatmode(struct inode *inode, 81 struct erofs_map_blocks *map) 82 { 83 erofs_blk_t nblocks, lastblk; 84 u64 offset = map->m_la; 85 struct erofs_inode *vi = EROFS_I(inode); 86 struct super_block *sb = inode->i_sb; 87 bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); 88 89 nblocks = erofs_iblks(inode); 90 lastblk = nblocks - tailendpacking; 91 92 /* there is no hole in flatmode */ 93 map->m_flags = EROFS_MAP_MAPPED; 94 if (offset < erofs_pos(sb, lastblk)) { 95 map->m_pa = erofs_pos(sb, vi->raw_blkaddr) + map->m_la; 96 map->m_plen = erofs_pos(sb, lastblk) - offset; 97 } else if (tailendpacking) { 98 map->m_pa = erofs_iloc(inode) + vi->inode_isize + 99 vi->xattr_isize + erofs_blkoff(sb, offset); 100 map->m_plen = inode->i_size - offset; 101 102 /* inline data should be located in the same meta block */ 103 if (erofs_blkoff(sb, map->m_pa) + map->m_plen > sb->s_blocksize) { 104 erofs_err(sb, "inline data cross block boundary @ nid %llu", 105 vi->nid); 106 DBG_BUGON(1); 107 return -EFSCORRUPTED; 108 } 109 map->m_flags |= EROFS_MAP_META; 110 } else { 111 erofs_err(sb, "internal error @ nid: %llu (size %llu), m_la 0x%llx", 112 vi->nid, inode->i_size, map->m_la); 113 DBG_BUGON(1); 114 return -EIO; 115 } 116 return 0; 117 } 118 119 int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map) 120 { 121 struct super_block *sb = inode->i_sb; 122 struct erofs_inode *vi = EROFS_I(inode); 123 struct erofs_inode_chunk_index *idx; 124 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 125 u64 chunknr; 126 unsigned int unit; 127 erofs_off_t pos; 128 void *kaddr; 129 int err = 0; 130 131 trace_erofs_map_blocks_enter(inode, map, 0); 132 map->m_deviceid = 0; 133 if (map->m_la >= inode->i_size) { 134 /* leave out-of-bound access unmapped */ 135 map->m_flags = 0; 136 map->m_plen = 0; 137 goto out; 138 } 139 140 if (vi->datalayout != EROFS_INODE_CHUNK_BASED) { 141 err = erofs_map_blocks_flatmode(inode, map); 142 goto out; 143 } 144 145 if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) 146 unit = sizeof(*idx); /* chunk index */ 147 else 148 unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */ 149 150 chunknr = map->m_la >> vi->chunkbits; 151 pos = ALIGN(erofs_iloc(inode) + vi->inode_isize + 152 vi->xattr_isize, unit) + unit * chunknr; 153 154 kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(sb, pos), EROFS_KMAP); 155 if (IS_ERR(kaddr)) { 156 err = PTR_ERR(kaddr); 157 goto out; 158 } 159 map->m_la = chunknr << vi->chunkbits; 160 map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits, 161 round_up(inode->i_size - map->m_la, sb->s_blocksize)); 162 163 /* handle block map */ 164 if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) { 165 __le32 *blkaddr = kaddr + erofs_blkoff(sb, pos); 166 167 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) { 168 map->m_flags = 0; 169 } else { 170 map->m_pa = erofs_pos(sb, le32_to_cpu(*blkaddr)); 171 map->m_flags = EROFS_MAP_MAPPED; 172 } 173 goto out_unlock; 174 } 175 /* parse chunk indexes */ 176 idx = kaddr + erofs_blkoff(sb, pos); 177 switch (le32_to_cpu(idx->blkaddr)) { 178 case EROFS_NULL_ADDR: 179 map->m_flags = 0; 180 break; 181 default: 182 map->m_deviceid = le16_to_cpu(idx->device_id) & 183 EROFS_SB(sb)->device_id_mask; 184 map->m_pa = erofs_pos(sb, le32_to_cpu(idx->blkaddr)); 185 map->m_flags = EROFS_MAP_MAPPED; 186 break; 187 } 188 out_unlock: 189 erofs_put_metabuf(&buf); 190 out: 191 if (!err) 192 map->m_llen = map->m_plen; 193 trace_erofs_map_blocks_exit(inode, map, 0, err); 194 return err; 195 } 196 197 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map) 198 { 199 struct erofs_dev_context *devs = EROFS_SB(sb)->devs; 200 struct erofs_device_info *dif; 201 int id; 202 203 map->m_bdev = sb->s_bdev; 204 map->m_daxdev = EROFS_SB(sb)->dax_dev; 205 map->m_dax_part_off = EROFS_SB(sb)->dax_part_off; 206 map->m_fscache = EROFS_SB(sb)->s_fscache; 207 208 if (map->m_deviceid) { 209 down_read(&devs->rwsem); 210 dif = idr_find(&devs->tree, map->m_deviceid - 1); 211 if (!dif) { 212 up_read(&devs->rwsem); 213 return -ENODEV; 214 } 215 if (devs->flatdev) { 216 map->m_pa += erofs_pos(sb, dif->mapped_blkaddr); 217 up_read(&devs->rwsem); 218 return 0; 219 } 220 map->m_bdev = dif->bdev; 221 map->m_daxdev = dif->dax_dev; 222 map->m_dax_part_off = dif->dax_part_off; 223 map->m_fscache = dif->fscache; 224 up_read(&devs->rwsem); 225 } else if (devs->extra_devices && !devs->flatdev) { 226 down_read(&devs->rwsem); 227 idr_for_each_entry(&devs->tree, dif, id) { 228 erofs_off_t startoff, length; 229 230 if (!dif->mapped_blkaddr) 231 continue; 232 startoff = erofs_pos(sb, dif->mapped_blkaddr); 233 length = erofs_pos(sb, dif->blocks); 234 235 if (map->m_pa >= startoff && 236 map->m_pa < startoff + length) { 237 map->m_pa -= startoff; 238 map->m_bdev = dif->bdev; 239 map->m_daxdev = dif->dax_dev; 240 map->m_dax_part_off = dif->dax_part_off; 241 map->m_fscache = dif->fscache; 242 break; 243 } 244 } 245 up_read(&devs->rwsem); 246 } 247 return 0; 248 } 249 250 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 251 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 252 { 253 int ret; 254 struct super_block *sb = inode->i_sb; 255 struct erofs_map_blocks map; 256 struct erofs_map_dev mdev; 257 258 map.m_la = offset; 259 map.m_llen = length; 260 261 ret = erofs_map_blocks(inode, &map); 262 if (ret < 0) 263 return ret; 264 265 mdev = (struct erofs_map_dev) { 266 .m_deviceid = map.m_deviceid, 267 .m_pa = map.m_pa, 268 }; 269 ret = erofs_map_dev(sb, &mdev); 270 if (ret) 271 return ret; 272 273 iomap->offset = map.m_la; 274 if (flags & IOMAP_DAX) 275 iomap->dax_dev = mdev.m_daxdev; 276 else 277 iomap->bdev = mdev.m_bdev; 278 iomap->length = map.m_llen; 279 iomap->flags = 0; 280 iomap->private = NULL; 281 282 if (!(map.m_flags & EROFS_MAP_MAPPED)) { 283 iomap->type = IOMAP_HOLE; 284 iomap->addr = IOMAP_NULL_ADDR; 285 if (!iomap->length) 286 iomap->length = length; 287 return 0; 288 } 289 290 if (map.m_flags & EROFS_MAP_META) { 291 void *ptr; 292 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 293 294 iomap->type = IOMAP_INLINE; 295 ptr = erofs_read_metabuf(&buf, sb, 296 erofs_blknr(sb, mdev.m_pa), EROFS_KMAP); 297 if (IS_ERR(ptr)) 298 return PTR_ERR(ptr); 299 iomap->inline_data = ptr + erofs_blkoff(sb, mdev.m_pa); 300 iomap->private = buf.base; 301 } else { 302 iomap->type = IOMAP_MAPPED; 303 iomap->addr = mdev.m_pa; 304 if (flags & IOMAP_DAX) 305 iomap->addr += mdev.m_dax_part_off; 306 } 307 return 0; 308 } 309 310 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length, 311 ssize_t written, unsigned int flags, struct iomap *iomap) 312 { 313 void *ptr = iomap->private; 314 315 if (ptr) { 316 struct erofs_buf buf = { 317 .page = kmap_to_page(ptr), 318 .base = ptr, 319 .kmap_type = EROFS_KMAP, 320 }; 321 322 DBG_BUGON(iomap->type != IOMAP_INLINE); 323 erofs_put_metabuf(&buf); 324 } else { 325 DBG_BUGON(iomap->type == IOMAP_INLINE); 326 } 327 return written; 328 } 329 330 static const struct iomap_ops erofs_iomap_ops = { 331 .iomap_begin = erofs_iomap_begin, 332 .iomap_end = erofs_iomap_end, 333 }; 334 335 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 336 u64 start, u64 len) 337 { 338 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 339 #ifdef CONFIG_EROFS_FS_ZIP 340 return iomap_fiemap(inode, fieinfo, start, len, 341 &z_erofs_iomap_report_ops); 342 #else 343 return -EOPNOTSUPP; 344 #endif 345 } 346 return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops); 347 } 348 349 /* 350 * since we dont have write or truncate flows, so no inode 351 * locking needs to be held at the moment. 352 */ 353 static int erofs_read_folio(struct file *file, struct folio *folio) 354 { 355 return iomap_read_folio(folio, &erofs_iomap_ops); 356 } 357 358 static void erofs_readahead(struct readahead_control *rac) 359 { 360 return iomap_readahead(rac, &erofs_iomap_ops); 361 } 362 363 static sector_t erofs_bmap(struct address_space *mapping, sector_t block) 364 { 365 return iomap_bmap(mapping, block, &erofs_iomap_ops); 366 } 367 368 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 369 { 370 struct inode *inode = file_inode(iocb->ki_filp); 371 372 /* no need taking (shared) inode lock since it's a ro filesystem */ 373 if (!iov_iter_count(to)) 374 return 0; 375 376 #ifdef CONFIG_FS_DAX 377 if (IS_DAX(inode)) 378 return dax_iomap_rw(iocb, to, &erofs_iomap_ops); 379 #endif 380 if (iocb->ki_flags & IOCB_DIRECT) { 381 struct block_device *bdev = inode->i_sb->s_bdev; 382 unsigned int blksize_mask; 383 384 if (bdev) 385 blksize_mask = bdev_logical_block_size(bdev) - 1; 386 else 387 blksize_mask = i_blocksize(inode) - 1; 388 389 if ((iocb->ki_pos | iov_iter_count(to) | 390 iov_iter_alignment(to)) & blksize_mask) 391 return -EINVAL; 392 393 return iomap_dio_rw(iocb, to, &erofs_iomap_ops, 394 NULL, 0, NULL, 0); 395 } 396 return filemap_read(iocb, to, 0); 397 } 398 399 /* for uncompressed (aligned) files and raw access for other files */ 400 const struct address_space_operations erofs_raw_access_aops = { 401 .read_folio = erofs_read_folio, 402 .readahead = erofs_readahead, 403 .bmap = erofs_bmap, 404 .direct_IO = noop_direct_IO, 405 .release_folio = iomap_release_folio, 406 .invalidate_folio = iomap_invalidate_folio, 407 }; 408 409 #ifdef CONFIG_FS_DAX 410 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf, 411 enum page_entry_size pe_size) 412 { 413 return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops); 414 } 415 416 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf) 417 { 418 return erofs_dax_huge_fault(vmf, PE_SIZE_PTE); 419 } 420 421 static const struct vm_operations_struct erofs_dax_vm_ops = { 422 .fault = erofs_dax_fault, 423 .huge_fault = erofs_dax_huge_fault, 424 }; 425 426 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma) 427 { 428 if (!IS_DAX(file_inode(file))) 429 return generic_file_readonly_mmap(file, vma); 430 431 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 432 return -EINVAL; 433 434 vma->vm_ops = &erofs_dax_vm_ops; 435 vm_flags_set(vma, VM_HUGEPAGE); 436 return 0; 437 } 438 #else 439 #define erofs_file_mmap generic_file_readonly_mmap 440 #endif 441 442 const struct file_operations erofs_file_fops = { 443 .llseek = generic_file_llseek, 444 .read_iter = erofs_file_read_iter, 445 .mmap = erofs_file_mmap, 446 .splice_read = generic_file_splice_read, 447 }; 448