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 void erofs_unmap_metabuf(struct erofs_buf *buf) 13 { 14 if (buf->kmap_type == EROFS_KMAP) 15 kunmap(buf->page); 16 else if (buf->kmap_type == EROFS_KMAP_ATOMIC) 17 kunmap_atomic(buf->base); 18 buf->base = NULL; 19 buf->kmap_type = EROFS_NO_KMAP; 20 } 21 22 void erofs_put_metabuf(struct erofs_buf *buf) 23 { 24 if (!buf->page) 25 return; 26 erofs_unmap_metabuf(buf); 27 put_page(buf->page); 28 buf->page = NULL; 29 } 30 31 void *erofs_bread(struct erofs_buf *buf, struct inode *inode, 32 erofs_blk_t blkaddr, enum erofs_kmap_type type) 33 { 34 struct address_space *const mapping = inode->i_mapping; 35 erofs_off_t offset = blknr_to_addr(blkaddr); 36 pgoff_t index = offset >> PAGE_SHIFT; 37 struct page *page = buf->page; 38 39 if (!page || page->index != index) { 40 erofs_put_metabuf(buf); 41 page = read_cache_page_gfp(mapping, index, 42 mapping_gfp_constraint(mapping, ~__GFP_FS)); 43 if (IS_ERR(page)) 44 return page; 45 /* should already be PageUptodate, no need to lock page */ 46 buf->page = page; 47 } 48 if (buf->kmap_type == EROFS_NO_KMAP) { 49 if (type == EROFS_KMAP) 50 buf->base = kmap(page); 51 else if (type == EROFS_KMAP_ATOMIC) 52 buf->base = kmap_atomic(page); 53 buf->kmap_type = type; 54 } else if (buf->kmap_type != type) { 55 DBG_BUGON(1); 56 return ERR_PTR(-EFAULT); 57 } 58 if (type == EROFS_NO_KMAP) 59 return NULL; 60 return buf->base + (offset & ~PAGE_MASK); 61 } 62 63 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb, 64 erofs_blk_t blkaddr, enum erofs_kmap_type type) 65 { 66 return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type); 67 } 68 69 static int erofs_map_blocks_flatmode(struct inode *inode, 70 struct erofs_map_blocks *map, 71 int flags) 72 { 73 erofs_blk_t nblocks, lastblk; 74 u64 offset = map->m_la; 75 struct erofs_inode *vi = EROFS_I(inode); 76 bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); 77 78 nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ); 79 lastblk = nblocks - tailendpacking; 80 81 /* there is no hole in flatmode */ 82 map->m_flags = EROFS_MAP_MAPPED; 83 if (offset < blknr_to_addr(lastblk)) { 84 map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la; 85 map->m_plen = blknr_to_addr(lastblk) - offset; 86 } else if (tailendpacking) { 87 /* 2 - inode inline B: inode, [xattrs], inline last blk... */ 88 struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb); 89 90 map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize + 91 vi->xattr_isize + erofs_blkoff(map->m_la); 92 map->m_plen = inode->i_size - offset; 93 94 /* inline data should be located in the same meta block */ 95 if (erofs_blkoff(map->m_pa) + map->m_plen > EROFS_BLKSIZ) { 96 erofs_err(inode->i_sb, 97 "inline data cross block boundary @ nid %llu", 98 vi->nid); 99 DBG_BUGON(1); 100 return -EFSCORRUPTED; 101 } 102 map->m_flags |= EROFS_MAP_META; 103 } else { 104 erofs_err(inode->i_sb, 105 "internal error @ nid: %llu (size %llu), m_la 0x%llx", 106 vi->nid, inode->i_size, map->m_la); 107 DBG_BUGON(1); 108 return -EIO; 109 } 110 return 0; 111 } 112 113 static int erofs_map_blocks(struct inode *inode, 114 struct erofs_map_blocks *map, int flags) 115 { 116 struct super_block *sb = inode->i_sb; 117 struct erofs_inode *vi = EROFS_I(inode); 118 struct erofs_inode_chunk_index *idx; 119 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 120 u64 chunknr; 121 unsigned int unit; 122 erofs_off_t pos; 123 void *kaddr; 124 int err = 0; 125 126 trace_erofs_map_blocks_enter(inode, map, flags); 127 map->m_deviceid = 0; 128 if (map->m_la >= inode->i_size) { 129 /* leave out-of-bound access unmapped */ 130 map->m_flags = 0; 131 map->m_plen = 0; 132 goto out; 133 } 134 135 if (vi->datalayout != EROFS_INODE_CHUNK_BASED) { 136 err = erofs_map_blocks_flatmode(inode, map, flags); 137 goto out; 138 } 139 140 if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) 141 unit = sizeof(*idx); /* chunk index */ 142 else 143 unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */ 144 145 chunknr = map->m_la >> vi->chunkbits; 146 pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize + 147 vi->xattr_isize, unit) + unit * chunknr; 148 149 kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP); 150 if (IS_ERR(kaddr)) { 151 err = PTR_ERR(kaddr); 152 goto out; 153 } 154 map->m_la = chunknr << vi->chunkbits; 155 map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits, 156 roundup(inode->i_size - map->m_la, EROFS_BLKSIZ)); 157 158 /* handle block map */ 159 if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) { 160 __le32 *blkaddr = kaddr + erofs_blkoff(pos); 161 162 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) { 163 map->m_flags = 0; 164 } else { 165 map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr)); 166 map->m_flags = EROFS_MAP_MAPPED; 167 } 168 goto out_unlock; 169 } 170 /* parse chunk indexes */ 171 idx = kaddr + erofs_blkoff(pos); 172 switch (le32_to_cpu(idx->blkaddr)) { 173 case EROFS_NULL_ADDR: 174 map->m_flags = 0; 175 break; 176 default: 177 map->m_deviceid = le16_to_cpu(idx->device_id) & 178 EROFS_SB(sb)->device_id_mask; 179 map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr)); 180 map->m_flags = EROFS_MAP_MAPPED; 181 break; 182 } 183 out_unlock: 184 erofs_put_metabuf(&buf); 185 out: 186 if (!err) 187 map->m_llen = map->m_plen; 188 trace_erofs_map_blocks_exit(inode, map, flags, 0); 189 return err; 190 } 191 192 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map) 193 { 194 struct erofs_dev_context *devs = EROFS_SB(sb)->devs; 195 struct erofs_device_info *dif; 196 int id; 197 198 /* primary device by default */ 199 map->m_bdev = sb->s_bdev; 200 map->m_daxdev = EROFS_SB(sb)->dax_dev; 201 map->m_dax_part_off = EROFS_SB(sb)->dax_part_off; 202 203 if (map->m_deviceid) { 204 down_read(&devs->rwsem); 205 dif = idr_find(&devs->tree, map->m_deviceid - 1); 206 if (!dif) { 207 up_read(&devs->rwsem); 208 return -ENODEV; 209 } 210 map->m_bdev = dif->bdev; 211 map->m_daxdev = dif->dax_dev; 212 map->m_dax_part_off = dif->dax_part_off; 213 up_read(&devs->rwsem); 214 } else if (devs->extra_devices) { 215 down_read(&devs->rwsem); 216 idr_for_each_entry(&devs->tree, dif, id) { 217 erofs_off_t startoff, length; 218 219 if (!dif->mapped_blkaddr) 220 continue; 221 startoff = blknr_to_addr(dif->mapped_blkaddr); 222 length = blknr_to_addr(dif->blocks); 223 224 if (map->m_pa >= startoff && 225 map->m_pa < startoff + length) { 226 map->m_pa -= startoff; 227 map->m_bdev = dif->bdev; 228 map->m_daxdev = dif->dax_dev; 229 map->m_dax_part_off = dif->dax_part_off; 230 break; 231 } 232 } 233 up_read(&devs->rwsem); 234 } 235 return 0; 236 } 237 238 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 239 unsigned int flags, struct iomap *iomap, struct iomap *srcmap) 240 { 241 int ret; 242 struct erofs_map_blocks map; 243 struct erofs_map_dev mdev; 244 245 map.m_la = offset; 246 map.m_llen = length; 247 248 ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); 249 if (ret < 0) 250 return ret; 251 252 mdev = (struct erofs_map_dev) { 253 .m_deviceid = map.m_deviceid, 254 .m_pa = map.m_pa, 255 }; 256 ret = erofs_map_dev(inode->i_sb, &mdev); 257 if (ret) 258 return ret; 259 260 iomap->offset = map.m_la; 261 if (flags & IOMAP_DAX) 262 iomap->dax_dev = mdev.m_daxdev; 263 else 264 iomap->bdev = mdev.m_bdev; 265 iomap->length = map.m_llen; 266 iomap->flags = 0; 267 iomap->private = NULL; 268 269 if (!(map.m_flags & EROFS_MAP_MAPPED)) { 270 iomap->type = IOMAP_HOLE; 271 iomap->addr = IOMAP_NULL_ADDR; 272 if (!iomap->length) 273 iomap->length = length; 274 return 0; 275 } 276 277 if (map.m_flags & EROFS_MAP_META) { 278 void *ptr; 279 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 280 281 iomap->type = IOMAP_INLINE; 282 ptr = erofs_read_metabuf(&buf, inode->i_sb, 283 erofs_blknr(mdev.m_pa), EROFS_KMAP); 284 if (IS_ERR(ptr)) 285 return PTR_ERR(ptr); 286 iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa); 287 iomap->private = buf.base; 288 } else { 289 iomap->type = IOMAP_MAPPED; 290 iomap->addr = mdev.m_pa; 291 if (flags & IOMAP_DAX) 292 iomap->addr += mdev.m_dax_part_off; 293 } 294 return 0; 295 } 296 297 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length, 298 ssize_t written, unsigned int flags, struct iomap *iomap) 299 { 300 void *ptr = iomap->private; 301 302 if (ptr) { 303 struct erofs_buf buf = { 304 .page = kmap_to_page(ptr), 305 .base = ptr, 306 .kmap_type = EROFS_KMAP, 307 }; 308 309 DBG_BUGON(iomap->type != IOMAP_INLINE); 310 erofs_put_metabuf(&buf); 311 } else { 312 DBG_BUGON(iomap->type == IOMAP_INLINE); 313 } 314 return written; 315 } 316 317 static const struct iomap_ops erofs_iomap_ops = { 318 .iomap_begin = erofs_iomap_begin, 319 .iomap_end = erofs_iomap_end, 320 }; 321 322 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 323 u64 start, u64 len) 324 { 325 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 326 #ifdef CONFIG_EROFS_FS_ZIP 327 return iomap_fiemap(inode, fieinfo, start, len, 328 &z_erofs_iomap_report_ops); 329 #else 330 return -EOPNOTSUPP; 331 #endif 332 } 333 return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops); 334 } 335 336 /* 337 * since we dont have write or truncate flows, so no inode 338 * locking needs to be held at the moment. 339 */ 340 static int erofs_readpage(struct file *file, struct page *page) 341 { 342 return iomap_readpage(page, &erofs_iomap_ops); 343 } 344 345 static void erofs_readahead(struct readahead_control *rac) 346 { 347 return iomap_readahead(rac, &erofs_iomap_ops); 348 } 349 350 static sector_t erofs_bmap(struct address_space *mapping, sector_t block) 351 { 352 return iomap_bmap(mapping, block, &erofs_iomap_ops); 353 } 354 355 static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to) 356 { 357 struct inode *inode = file_inode(iocb->ki_filp); 358 loff_t align = iocb->ki_pos | iov_iter_count(to) | 359 iov_iter_alignment(to); 360 struct block_device *bdev = inode->i_sb->s_bdev; 361 unsigned int blksize_mask; 362 363 if (bdev) 364 blksize_mask = (1 << ilog2(bdev_logical_block_size(bdev))) - 1; 365 else 366 blksize_mask = (1 << inode->i_blkbits) - 1; 367 368 if (align & blksize_mask) 369 return -EINVAL; 370 return 0; 371 } 372 373 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 374 { 375 /* no need taking (shared) inode lock since it's a ro filesystem */ 376 if (!iov_iter_count(to)) 377 return 0; 378 379 #ifdef CONFIG_FS_DAX 380 if (IS_DAX(iocb->ki_filp->f_mapping->host)) 381 return dax_iomap_rw(iocb, to, &erofs_iomap_ops); 382 #endif 383 if (iocb->ki_flags & IOCB_DIRECT) { 384 int err = erofs_prepare_dio(iocb, to); 385 386 if (!err) 387 return iomap_dio_rw(iocb, to, &erofs_iomap_ops, 388 NULL, 0, 0); 389 if (err < 0) 390 return err; 391 } 392 return filemap_read(iocb, to, 0); 393 } 394 395 /* for uncompressed (aligned) files and raw access for other files */ 396 const struct address_space_operations erofs_raw_access_aops = { 397 .readpage = erofs_readpage, 398 .readahead = erofs_readahead, 399 .bmap = erofs_bmap, 400 .direct_IO = noop_direct_IO, 401 }; 402 403 #ifdef CONFIG_FS_DAX 404 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf, 405 enum page_entry_size pe_size) 406 { 407 return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops); 408 } 409 410 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf) 411 { 412 return erofs_dax_huge_fault(vmf, PE_SIZE_PTE); 413 } 414 415 static const struct vm_operations_struct erofs_dax_vm_ops = { 416 .fault = erofs_dax_fault, 417 .huge_fault = erofs_dax_huge_fault, 418 }; 419 420 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma) 421 { 422 if (!IS_DAX(file_inode(file))) 423 return generic_file_readonly_mmap(file, vma); 424 425 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 426 return -EINVAL; 427 428 vma->vm_ops = &erofs_dax_vm_ops; 429 vma->vm_flags |= VM_HUGEPAGE; 430 return 0; 431 } 432 #else 433 #define erofs_file_mmap generic_file_readonly_mmap 434 #endif 435 436 const struct file_operations erofs_file_fops = { 437 .llseek = generic_file_llseek, 438 .read_iter = erofs_file_read_iter, 439 .mmap = erofs_file_mmap, 440 .splice_read = generic_file_splice_read, 441 }; 442