| 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 "xattr.h" |
| 8 | #include <trace/events/erofs.h> |
| 9 | |
| 10 | static int erofs_fill_symlink(struct inode *inode, void *kaddr, |
| 11 | unsigned int m_pofs) |
| 12 | { |
| 13 | struct erofs_inode *vi = EROFS_I(inode); |
| 14 | loff_t off; |
| 15 | |
| 16 | m_pofs += vi->xattr_isize; |
| 17 | /* check if it cannot be handled with fast symlink scheme */ |
| 18 | if (vi->datalayout != EROFS_INODE_FLAT_INLINE || |
| 19 | check_add_overflow(m_pofs, inode->i_size, &off) || |
| 20 | off > i_blocksize(node: inode)) |
| 21 | return 0; |
| 22 | |
| 23 | inode->i_link = kmemdup_nul(s: kaddr + m_pofs, len: inode->i_size, GFP_KERNEL); |
| 24 | return inode->i_link ? 0 : -ENOMEM; |
| 25 | } |
| 26 | |
| 27 | static int erofs_read_inode(struct inode *inode) |
| 28 | { |
| 29 | struct super_block *sb = inode->i_sb; |
| 30 | erofs_blk_t blkaddr = erofs_blknr(sb, erofs_iloc(inode)); |
| 31 | unsigned int ofs = erofs_blkoff(sb, erofs_iloc(inode)); |
| 32 | struct erofs_buf buf = __EROFS_BUF_INITIALIZER; |
| 33 | struct erofs_sb_info *sbi = EROFS_SB(sb); |
| 34 | erofs_blk_t addrmask = BIT_ULL(48) - 1; |
| 35 | struct erofs_inode *vi = EROFS_I(inode); |
| 36 | struct erofs_inode_extended *die, copied; |
| 37 | struct erofs_inode_compact *dic; |
| 38 | unsigned int ifmt; |
| 39 | void *ptr; |
| 40 | int err = 0; |
| 41 | |
| 42 | ptr = erofs_read_metabuf(buf: &buf, sb, erofs_pos(sb, blkaddr), need_kmap: true); |
| 43 | if (IS_ERR(ptr)) { |
| 44 | err = PTR_ERR(ptr); |
| 45 | erofs_err(sb, "failed to get inode (nid: %llu) page, err %d" , |
| 46 | vi->nid, err); |
| 47 | goto err_out; |
| 48 | } |
| 49 | |
| 50 | dic = ptr + ofs; |
| 51 | ifmt = le16_to_cpu(dic->i_format); |
| 52 | if (ifmt & ~EROFS_I_ALL) { |
| 53 | erofs_err(sb, "unsupported i_format %u of nid %llu" , |
| 54 | ifmt, vi->nid); |
| 55 | err = -EOPNOTSUPP; |
| 56 | goto err_out; |
| 57 | } |
| 58 | |
| 59 | vi->datalayout = erofs_inode_datalayout(ifmt); |
| 60 | if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) { |
| 61 | erofs_err(sb, "unsupported datalayout %u of nid %llu" , |
| 62 | vi->datalayout, vi->nid); |
| 63 | err = -EOPNOTSUPP; |
| 64 | goto err_out; |
| 65 | } |
| 66 | |
| 67 | switch (erofs_inode_version(ifmt)) { |
| 68 | case EROFS_INODE_LAYOUT_EXTENDED: |
| 69 | vi->inode_isize = sizeof(struct erofs_inode_extended); |
| 70 | /* check if the extended inode acrosses block boundary */ |
| 71 | if (ofs + vi->inode_isize <= sb->s_blocksize) { |
| 72 | ofs += vi->inode_isize; |
| 73 | die = (struct erofs_inode_extended *)dic; |
| 74 | copied.i_u = die->i_u; |
| 75 | copied.i_nb = die->i_nb; |
| 76 | } else { |
| 77 | const unsigned int gotten = sb->s_blocksize - ofs; |
| 78 | |
| 79 | memcpy(&copied, dic, gotten); |
| 80 | ptr = erofs_read_metabuf(buf: &buf, sb, |
| 81 | erofs_pos(sb, blkaddr + 1), need_kmap: true); |
| 82 | if (IS_ERR(ptr)) { |
| 83 | err = PTR_ERR(ptr); |
| 84 | erofs_err(sb, "failed to get inode payload block (nid: %llu), err %d" , |
| 85 | vi->nid, err); |
| 86 | goto err_out; |
| 87 | } |
| 88 | ofs = vi->inode_isize - gotten; |
| 89 | memcpy((u8 *)&copied + gotten, ptr, ofs); |
| 90 | die = &copied; |
| 91 | } |
| 92 | vi->xattr_isize = erofs_xattr_ibody_size(i_xattr_icount: die->i_xattr_icount); |
| 93 | |
| 94 | inode->i_mode = le16_to_cpu(die->i_mode); |
| 95 | i_uid_write(inode, le32_to_cpu(die->i_uid)); |
| 96 | i_gid_write(inode, le32_to_cpu(die->i_gid)); |
| 97 | set_nlink(inode, le32_to_cpu(die->i_nlink)); |
| 98 | inode_set_mtime(inode, le64_to_cpu(die->i_mtime), |
| 99 | le32_to_cpu(die->i_mtime_nsec)); |
| 100 | |
| 101 | inode->i_size = le64_to_cpu(die->i_size); |
| 102 | break; |
| 103 | case EROFS_INODE_LAYOUT_COMPACT: |
| 104 | vi->inode_isize = sizeof(struct erofs_inode_compact); |
| 105 | ofs += vi->inode_isize; |
| 106 | vi->xattr_isize = erofs_xattr_ibody_size(i_xattr_icount: dic->i_xattr_icount); |
| 107 | |
| 108 | inode->i_mode = le16_to_cpu(dic->i_mode); |
| 109 | copied.i_u = dic->i_u; |
| 110 | i_uid_write(inode, le16_to_cpu(dic->i_uid)); |
| 111 | i_gid_write(inode, le16_to_cpu(dic->i_gid)); |
| 112 | if (!S_ISDIR(inode->i_mode) && |
| 113 | ((ifmt >> EROFS_I_NLINK_1_BIT) & 1)) { |
| 114 | set_nlink(inode, nlink: 1); |
| 115 | copied.i_nb = dic->i_nb; |
| 116 | } else { |
| 117 | set_nlink(inode, le16_to_cpu(dic->i_nb.nlink)); |
| 118 | copied.i_nb.startblk_hi = 0; |
| 119 | addrmask = BIT_ULL(32) - 1; |
| 120 | } |
| 121 | inode_set_mtime(inode, sec: sbi->epoch + le32_to_cpu(dic->i_mtime), |
| 122 | nsec: sbi->fixed_nsec); |
| 123 | |
| 124 | inode->i_size = le32_to_cpu(dic->i_size); |
| 125 | break; |
| 126 | default: |
| 127 | erofs_err(sb, "unsupported on-disk inode version %u of nid %llu" , |
| 128 | erofs_inode_version(ifmt), vi->nid); |
| 129 | err = -EOPNOTSUPP; |
| 130 | goto err_out; |
| 131 | } |
| 132 | |
| 133 | if (unlikely(inode->i_size < 0)) { |
| 134 | erofs_err(sb, "negative i_size @ nid %llu" , vi->nid); |
| 135 | err = -EFSCORRUPTED; |
| 136 | goto err_out; |
| 137 | } |
| 138 | switch (inode->i_mode & S_IFMT) { |
| 139 | case S_IFDIR: |
| 140 | vi->dot_omitted = (ifmt >> EROFS_I_DOT_OMITTED_BIT) & 1; |
| 141 | fallthrough; |
| 142 | case S_IFREG: |
| 143 | case S_IFLNK: |
| 144 | vi->startblk = le32_to_cpu(copied.i_u.startblk_lo) | |
| 145 | ((u64)le16_to_cpu(copied.i_nb.startblk_hi) << 32); |
| 146 | if (vi->datalayout == EROFS_INODE_FLAT_PLAIN && |
| 147 | !((vi->startblk ^ EROFS_NULL_ADDR) & addrmask)) |
| 148 | vi->startblk = EROFS_NULL_ADDR; |
| 149 | |
| 150 | if(S_ISLNK(inode->i_mode)) { |
| 151 | err = erofs_fill_symlink(inode, kaddr: ptr, m_pofs: ofs); |
| 152 | if (err) |
| 153 | goto err_out; |
| 154 | } |
| 155 | break; |
| 156 | case S_IFCHR: |
| 157 | case S_IFBLK: |
| 158 | inode->i_rdev = new_decode_dev(le32_to_cpu(copied.i_u.rdev)); |
| 159 | break; |
| 160 | case S_IFIFO: |
| 161 | case S_IFSOCK: |
| 162 | inode->i_rdev = 0; |
| 163 | break; |
| 164 | default: |
| 165 | erofs_err(sb, "bogus i_mode (%o) @ nid %llu" , inode->i_mode, |
| 166 | vi->nid); |
| 167 | err = -EFSCORRUPTED; |
| 168 | goto err_out; |
| 169 | } |
| 170 | |
| 171 | if (erofs_inode_is_data_compressed(datamode: vi->datalayout)) |
| 172 | inode->i_blocks = le32_to_cpu(copied.i_u.blocks_lo) << |
| 173 | (sb->s_blocksize_bits - 9); |
| 174 | else |
| 175 | inode->i_blocks = round_up(inode->i_size, sb->s_blocksize) >> 9; |
| 176 | |
| 177 | if (vi->datalayout == EROFS_INODE_CHUNK_BASED) { |
| 178 | /* fill chunked inode summary info */ |
| 179 | vi->chunkformat = le16_to_cpu(copied.i_u.c.format); |
| 180 | if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) { |
| 181 | erofs_err(sb, "unsupported chunk format %x of nid %llu" , |
| 182 | vi->chunkformat, vi->nid); |
| 183 | err = -EOPNOTSUPP; |
| 184 | goto err_out; |
| 185 | } |
| 186 | vi->chunkbits = sb->s_blocksize_bits + |
| 187 | (vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK); |
| 188 | } |
| 189 | inode_set_atime_to_ts(inode, |
| 190 | ts: inode_set_ctime_to_ts(inode, ts: inode_get_mtime(inode))); |
| 191 | |
| 192 | inode->i_flags &= ~S_DAX; |
| 193 | if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) && |
| 194 | (vi->datalayout == EROFS_INODE_FLAT_PLAIN || |
| 195 | vi->datalayout == EROFS_INODE_CHUNK_BASED)) |
| 196 | inode->i_flags |= S_DAX; |
| 197 | err_out: |
| 198 | erofs_put_metabuf(buf: &buf); |
| 199 | return err; |
| 200 | } |
| 201 | |
| 202 | static int erofs_fill_inode(struct inode *inode) |
| 203 | { |
| 204 | struct erofs_inode *vi = EROFS_I(inode); |
| 205 | int err; |
| 206 | |
| 207 | trace_erofs_fill_inode(inode); |
| 208 | err = erofs_read_inode(inode); |
| 209 | if (err) |
| 210 | return err; |
| 211 | |
| 212 | switch (inode->i_mode & S_IFMT) { |
| 213 | case S_IFREG: |
| 214 | inode->i_op = &erofs_generic_iops; |
| 215 | if (erofs_inode_is_data_compressed(datamode: vi->datalayout)) |
| 216 | inode->i_fop = &generic_ro_fops; |
| 217 | else |
| 218 | inode->i_fop = &erofs_file_fops; |
| 219 | break; |
| 220 | case S_IFDIR: |
| 221 | inode->i_op = &erofs_dir_iops; |
| 222 | inode->i_fop = &erofs_dir_fops; |
| 223 | inode_nohighmem(inode); |
| 224 | break; |
| 225 | case S_IFLNK: |
| 226 | if (inode->i_link) |
| 227 | inode->i_op = &erofs_fast_symlink_iops; |
| 228 | else |
| 229 | inode->i_op = &erofs_symlink_iops; |
| 230 | inode_nohighmem(inode); |
| 231 | break; |
| 232 | default: |
| 233 | inode->i_op = &erofs_generic_iops; |
| 234 | init_special_inode(inode, inode->i_mode, inode->i_rdev); |
| 235 | return 0; |
| 236 | } |
| 237 | |
| 238 | mapping_set_large_folios(mapping: inode->i_mapping); |
| 239 | if (erofs_inode_is_data_compressed(datamode: vi->datalayout)) { |
| 240 | #ifdef CONFIG_EROFS_FS_ZIP |
| 241 | DO_ONCE_LITE_IF(inode->i_blkbits != PAGE_SHIFT, |
| 242 | erofs_info, inode->i_sb, |
| 243 | "EXPERIMENTAL EROFS subpage compressed block support in use. Use at your own risk!" ); |
| 244 | inode->i_mapping->a_ops = &z_erofs_aops; |
| 245 | #else |
| 246 | err = -EOPNOTSUPP; |
| 247 | #endif |
| 248 | } else { |
| 249 | inode->i_mapping->a_ops = &erofs_aops; |
| 250 | #ifdef CONFIG_EROFS_FS_ONDEMAND |
| 251 | if (erofs_is_fscache_mode(sb: inode->i_sb)) |
| 252 | inode->i_mapping->a_ops = &erofs_fscache_access_aops; |
| 253 | #endif |
| 254 | #ifdef CONFIG_EROFS_FS_BACKED_BY_FILE |
| 255 | if (erofs_is_fileio_mode(EROFS_SB(inode->i_sb))) |
| 256 | inode->i_mapping->a_ops = &erofs_fileio_aops; |
| 257 | #endif |
| 258 | } |
| 259 | |
| 260 | return err; |
| 261 | } |
| 262 | |
| 263 | /* |
| 264 | * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down |
| 265 | * so that it will fit. |
| 266 | */ |
| 267 | static ino_t erofs_squash_ino(erofs_nid_t nid) |
| 268 | { |
| 269 | ino_t ino = (ino_t)nid; |
| 270 | |
| 271 | if (sizeof(ino_t) < sizeof(erofs_nid_t)) |
| 272 | ino ^= nid >> (sizeof(erofs_nid_t) - sizeof(ino_t)) * 8; |
| 273 | return ino; |
| 274 | } |
| 275 | |
| 276 | static int erofs_iget5_eq(struct inode *inode, void *opaque) |
| 277 | { |
| 278 | return EROFS_I(inode)->nid == *(erofs_nid_t *)opaque; |
| 279 | } |
| 280 | |
| 281 | static int erofs_iget5_set(struct inode *inode, void *opaque) |
| 282 | { |
| 283 | const erofs_nid_t nid = *(erofs_nid_t *)opaque; |
| 284 | |
| 285 | inode->i_ino = erofs_squash_ino(nid); |
| 286 | EROFS_I(inode)->nid = nid; |
| 287 | return 0; |
| 288 | } |
| 289 | |
| 290 | struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid) |
| 291 | { |
| 292 | struct inode *inode; |
| 293 | |
| 294 | inode = iget5_locked(sb, erofs_squash_ino(nid), test: erofs_iget5_eq, |
| 295 | set: erofs_iget5_set, &nid); |
| 296 | if (!inode) |
| 297 | return ERR_PTR(error: -ENOMEM); |
| 298 | |
| 299 | if (inode->i_state & I_NEW) { |
| 300 | int err = erofs_fill_inode(inode); |
| 301 | |
| 302 | if (err) { |
| 303 | iget_failed(inode); |
| 304 | return ERR_PTR(error: err); |
| 305 | } |
| 306 | unlock_new_inode(inode); |
| 307 | } |
| 308 | return inode; |
| 309 | } |
| 310 | |
| 311 | int erofs_getattr(struct mnt_idmap *idmap, const struct path *path, |
| 312 | struct kstat *stat, u32 request_mask, |
| 313 | unsigned int query_flags) |
| 314 | { |
| 315 | struct inode *const inode = d_inode(dentry: path->dentry); |
| 316 | struct block_device *bdev = inode->i_sb->s_bdev; |
| 317 | bool compressed = |
| 318 | erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout); |
| 319 | |
| 320 | if (compressed) |
| 321 | stat->attributes |= STATX_ATTR_COMPRESSED; |
| 322 | stat->attributes |= STATX_ATTR_IMMUTABLE; |
| 323 | stat->attributes_mask |= (STATX_ATTR_COMPRESSED | |
| 324 | STATX_ATTR_IMMUTABLE); |
| 325 | |
| 326 | /* |
| 327 | * Return the DIO alignment restrictions if requested. |
| 328 | * |
| 329 | * In EROFS, STATX_DIOALIGN is only supported in bdev-based mode |
| 330 | * and uncompressed inodes, otherwise we report no DIO support. |
| 331 | */ |
| 332 | if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) { |
| 333 | stat->result_mask |= STATX_DIOALIGN; |
| 334 | if (bdev && !compressed) { |
| 335 | stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; |
| 336 | stat->dio_offset_align = bdev_logical_block_size(bdev); |
| 337 | } |
| 338 | } |
| 339 | generic_fillattr(idmap, request_mask, inode, stat); |
| 340 | return 0; |
| 341 | } |
| 342 | |
| 343 | const struct inode_operations erofs_generic_iops = { |
| 344 | .getattr = erofs_getattr, |
| 345 | .listxattr = erofs_listxattr, |
| 346 | .get_inode_acl = erofs_get_acl, |
| 347 | .fiemap = erofs_fiemap, |
| 348 | }; |
| 349 | |
| 350 | const struct inode_operations erofs_symlink_iops = { |
| 351 | .get_link = page_get_link, |
| 352 | .getattr = erofs_getattr, |
| 353 | .listxattr = erofs_listxattr, |
| 354 | .get_inode_acl = erofs_get_acl, |
| 355 | }; |
| 356 | |
| 357 | const struct inode_operations erofs_fast_symlink_iops = { |
| 358 | .get_link = simple_get_link, |
| 359 | .getattr = erofs_getattr, |
| 360 | .listxattr = erofs_listxattr, |
| 361 | .get_inode_acl = erofs_get_acl, |
| 362 | }; |
| 363 | |