| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* linux/fs/ext4/dir.c |
| 4 |
* |
| 5 |
* Copyright (C) 1992, 1993, 1994, 1995 |
| 6 |
* Remy Card ([email protected]) |
| 7 |
* Laboratoire MASI - Institut Blaise Pascal |
| 8 |
* Universite Pierre et Marie Curie (Paris VI) |
| 9 |
* |
| 10 |
* from |
| 11 |
* |
| 12 |
* linux/fs/minix/dir.c |
| 13 |
* |
| 14 |
* Copyright (C) 1991, 1992 Linus Torvalds |
| 15 |
* |
| 16 |
* ext4 directory handling functions |
| 17 |
* |
| 18 |
* Big-endian to little-endian byte-swapping/bitmaps by |
| 19 |
* David S. Miller ([email protected]), 1995 |
| 20 |
* |
| 21 |
* Hash Tree Directory indexing (c) 2001 Daniel Phillips |
| 22 |
* |
| 23 |
*/ |
| 24 |
|
| 25 |
#include <linux/fs.h> |
| 26 |
#include <linux/buffer_head.h> |
| 27 |
#include <linux/filelock.h> |
| 28 |
#include <linux/slab.h> |
| 29 |
#include <linux/iversion.h> |
| 30 |
#include <linux/unicode.h> |
| 31 |
#include "ext4.h" |
| 32 |
#include "xattr.h" |
| 33 |
|
| 34 |
static int ext4_dx_readdir(struct file *, struct dir_context *); |
| 35 |
|
| 36 |
/** |
| 37 |
* is_dx_dir() - check if a directory is using htree indexing |
| 38 |
* @inode: directory inode |
| 39 |
* |
| 40 |
* Check if the given dir-inode refers to an htree-indexed directory |
| 41 |
* (or a directory which could potentially get converted to use htree |
| 42 |
* indexing). |
| 43 |
* |
| 44 |
* Return 1 if it is a dx dir, 0 if not |
| 45 |
*/ |
| 46 |
static int is_dx_dir(struct inode *inode) |
| 47 |
{ |
| 48 |
struct super_block *sb = inode->i_sb; |
| 49 |
|
| 50 |
if (ext4_has_feature_dir_index(inode->i_sb) && |
| 51 |
((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) || |
| 52 |
((inode->i_size >> sb->s_blocksize_bits) == 1) || |
| 53 |
ext4_has_inline_data(inode))) |
| 54 |
return 1; |
| 55 |
|
| 56 |
return 0; |
| 57 |
} |
| 58 |
|
| 59 |
static bool is_fake_dir_entry(struct ext4_dir_entry_2 *de) |
| 60 |
{ |
| 61 |
/* Check if . or .. , or skip if namelen is 0 */ |
| 62 |
if ((de->name_len > 0) && (de->name_len <= 2) && (de->name[0] == '.') && |
| 63 |
(de->name[1] == '.' || de->name[1] == '\0')) |
| 64 |
return true; |
| 65 |
/* Check if this is a csum entry */ |
| 66 |
if (de->file_type == EXT4_FT_DIR_CSUM) |
| 67 |
return true; |
| 68 |
return false; |
| 69 |
} |
| 70 |
|
| 71 |
/* |
| 72 |
* Return 0 if the directory entry is OK, and 1 if there is a problem |
| 73 |
* |
| 74 |
* Note: this is the opposite of what ext2 and ext3 historically returned... |
| 75 |
* |
| 76 |
* bh passed here can be an inode block or a dir data block, depending |
| 77 |
* on the inode inline data flag. |
| 78 |
*/ |
| 79 |
int __ext4_check_dir_entry(const char *function, unsigned int line, |
| 80 |
struct inode *dir, struct file *filp, |
| 81 |
struct ext4_dir_entry_2 *de, |
| 82 |
struct buffer_head *bh, char *buf, int size, |
| 83 |
unsigned int offset) |
| 84 |
{ |
| 85 |
const char *error_msg = NULL; |
| 86 |
const int rlen = ext4_rec_len_from_disk(de->rec_len, |
| 87 |
dir->i_sb->s_blocksize); |
| 88 |
const int next_offset = ((char *) de - buf) + rlen; |
| 89 |
bool fake = is_fake_dir_entry(de); |
| 90 |
bool has_csum = ext4_has_feature_metadata_csum(dir->i_sb); |
| 91 |
|
| 92 |
if (unlikely(rlen < ext4_dir_rec_len(1, fake ? NULL : dir))) |
| 93 |
error_msg = "rec_len is smaller than minimal"; |
| 94 |
else if (unlikely(rlen % 4 != 0)) |
| 95 |
error_msg = "rec_len % 4 != 0"; |
| 96 |
else if (unlikely(rlen < ext4_dir_rec_len(de->name_len, |
| 97 |
fake ? NULL : dir))) |
| 98 |
error_msg = "rec_len is too small for name_len"; |
| 99 |
else if (unlikely(next_offset > size)) |
| 100 |
error_msg = "directory entry overrun"; |
| 101 |
else if (unlikely(next_offset > size - ext4_dir_rec_len(1, |
| 102 |
has_csum ? NULL : dir) && |
| 103 |
next_offset != size)) |
| 104 |
error_msg = "directory entry too close to block end"; |
| 105 |
else if (unlikely(le32_to_cpu(de->inode) > |
| 106 |
le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) |
| 107 |
error_msg = "inode out of bounds"; |
| 108 |
else if (unlikely(next_offset == size && de->name_len == 1 && |
| 109 |
de->name[0] == '.')) |
| 110 |
error_msg = "'.' directory cannot be the last in data block"; |
| 111 |
else |
| 112 |
return 0; |
| 113 |
|
| 114 |
if (filp) |
| 115 |
ext4_error_file(filp, function, line, bh->b_blocknr, |
| 116 |
"bad entry in directory: %s - offset=%u, " |
| 117 |
"inode=%u, rec_len=%d, size=%d fake=%d", |
| 118 |
error_msg, offset, le32_to_cpu(de->inode), |
| 119 |
rlen, size, fake); |
| 120 |
else |
| 121 |
ext4_error_inode(dir, function, line, bh->b_blocknr, |
| 122 |
"bad entry in directory: %s - offset=%u, " |
| 123 |
"inode=%u, rec_len=%d, size=%d fake=%d", |
| 124 |
error_msg, offset, le32_to_cpu(de->inode), |
| 125 |
rlen, size, fake); |
| 126 |
|
| 127 |
return 1; |
| 128 |
} |
| 129 |
|
| 130 |
static int ext4_readdir(struct file *file, struct dir_context *ctx) |
| 131 |
{ |
| 132 |
unsigned int offset; |
| 133 |
int i; |
| 134 |
struct ext4_dir_entry_2 *de; |
| 135 |
int err; |
| 136 |
struct inode *inode = file_inode(file); |
| 137 |
struct super_block *sb = inode->i_sb; |
| 138 |
struct buffer_head *bh = NULL; |
| 139 |
struct fscrypt_str fstr = FSTR_INIT(NULL, 0); |
| 140 |
struct dir_private_info *info = file->private_data; |
| 141 |
|
| 142 |
err = fscrypt_prepare_readdir(inode); |
| 143 |
if (err) |
| 144 |
return err; |
| 145 |
|
| 146 |
if (is_dx_dir(inode)) { |
| 147 |
err = ext4_dx_readdir(file, ctx); |
| 148 |
if (err != ERR_BAD_DX_DIR) |
| 149 |
return err; |
| 150 |
|
| 151 |
/* Can we just clear INDEX flag to ignore htree information? */ |
| 152 |
if (!ext4_has_feature_metadata_csum(sb)) { |
| 153 |
/* |
| 154 |
* We don't set the inode dirty flag since it's not |
| 155 |
* critical that it gets flushed back to the disk. |
| 156 |
*/ |
| 157 |
ext4_clear_inode_flag(inode, EXT4_INODE_INDEX); |
| 158 |
} |
| 159 |
} |
| 160 |
|
| 161 |
if (ext4_has_inline_data(inode)) { |
| 162 |
int has_inline_data = 1; |
| 163 |
err = ext4_read_inline_dir(file, ctx, |
| 164 |
&has_inline_data); |
| 165 |
if (has_inline_data) |
| 166 |
return err; |
| 167 |
} |
| 168 |
|
| 169 |
if (IS_ENCRYPTED(inode)) { |
| 170 |
err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, &fstr); |
| 171 |
if (err < 0) |
| 172 |
return err; |
| 173 |
} |
| 174 |
|
| 175 |
while (ctx->pos < inode->i_size) { |
| 176 |
struct ext4_map_blocks map; |
| 177 |
|
| 178 |
if (fatal_signal_pending(current)) { |
| 179 |
err = -ERESTARTSYS; |
| 180 |
goto errout; |
| 181 |
} |
| 182 |
cond_resched(); |
| 183 |
offset = ctx->pos & (sb->s_blocksize - 1); |
| 184 |
map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); |
| 185 |
map.m_len = 1; |
| 186 |
err = ext4_map_blocks(NULL, inode, &map, 0); |
| 187 |
if (err == 0) { |
| 188 |
/* m_len should never be zero but let's avoid |
| 189 |
* an infinite loop if it somehow is */ |
| 190 |
if (map.m_len == 0) |
| 191 |
map.m_len = 1; |
| 192 |
ctx->pos += map.m_len * sb->s_blocksize; |
| 193 |
continue; |
| 194 |
} |
| 195 |
if (err > 0) { |
| 196 |
pgoff_t index = map.m_pblk << inode->i_blkbits >> |
| 197 |
PAGE_SHIFT; |
| 198 |
if (!ra_has_index(&file->f_ra, index)) |
| 199 |
page_cache_sync_readahead( |
| 200 |
sb->s_bdev->bd_mapping, |
| 201 |
&file->f_ra, file, index, |
| 202 |
1 << EXT4_SB(sb)->s_min_folio_order); |
| 203 |
file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT; |
| 204 |
bh = ext4_bread(NULL, inode, map.m_lblk, 0); |
| 205 |
if (IS_ERR(bh)) { |
| 206 |
err = PTR_ERR(bh); |
| 207 |
bh = NULL; |
| 208 |
goto errout; |
| 209 |
} |
| 210 |
} |
| 211 |
|
| 212 |
if (!bh) { |
| 213 |
/* corrupt size? Maybe no more blocks to read */ |
| 214 |
if (ctx->pos > inode->i_blocks << 9) |
| 215 |
break; |
| 216 |
ctx->pos += sb->s_blocksize - offset; |
| 217 |
continue; |
| 218 |
} |
| 219 |
|
| 220 |
/* Check the checksum */ |
| 221 |
if (!buffer_verified(bh) && |
| 222 |
!ext4_dirblock_csum_verify(inode, bh)) { |
| 223 |
EXT4_ERROR_FILE(file, 0, "directory fails checksum " |
| 224 |
"at offset %llu", |
| 225 |
(unsigned long long)ctx->pos); |
| 226 |
ctx->pos += sb->s_blocksize - offset; |
| 227 |
brelse(bh); |
| 228 |
bh = NULL; |
| 229 |
continue; |
| 230 |
} |
| 231 |
set_buffer_verified(bh); |
| 232 |
|
| 233 |
/* If the dir block has changed since the last call to |
| 234 |
* readdir(2), then we might be pointing to an invalid |
| 235 |
* dirent right now. Scan from the start of the block |
| 236 |
* to make sure. */ |
| 237 |
if (!inode_eq_iversion(inode, info->cookie)) { |
| 238 |
for (i = 0; i < sb->s_blocksize && i < offset; ) { |
| 239 |
de = (struct ext4_dir_entry_2 *) |
| 240 |
(bh->b_data + i); |
| 241 |
/* It's too expensive to do a full |
| 242 |
* dirent test each time round this |
| 243 |
* loop, but we do have to test at |
| 244 |
* least that it is non-zero. A |
| 245 |
* failure will be detected in the |
| 246 |
* dirent test below. */ |
| 247 |
if (ext4_rec_len_from_disk(de->rec_len, |
| 248 |
sb->s_blocksize) < ext4_dir_rec_len(1, |
| 249 |
inode)) |
| 250 |
break; |
| 251 |
i += ext4_rec_len_from_disk(de->rec_len, |
| 252 |
sb->s_blocksize); |
| 253 |
} |
| 254 |
offset = i; |
| 255 |
ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) |
| 256 |
| offset; |
| 257 |
info->cookie = inode_query_iversion(inode); |
| 258 |
} |
| 259 |
|
| 260 |
while (ctx->pos < inode->i_size |
| 261 |
&& offset < sb->s_blocksize) { |
| 262 |
de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); |
| 263 |
if (ext4_check_dir_entry(inode, file, de, bh, |
| 264 |
bh->b_data, bh->b_size, |
| 265 |
offset)) { |
| 266 |
/* |
| 267 |
* On error, skip to the next block |
| 268 |
*/ |
| 269 |
ctx->pos = (ctx->pos | |
| 270 |
(sb->s_blocksize - 1)) + 1; |
| 271 |
break; |
| 272 |
} |
| 273 |
offset += ext4_rec_len_from_disk(de->rec_len, |
| 274 |
sb->s_blocksize); |
| 275 |
if (le32_to_cpu(de->inode)) { |
| 276 |
if (!IS_ENCRYPTED(inode)) { |
| 277 |
if (!dir_emit(ctx, de->name, |
| 278 |
de->name_len, |
| 279 |
le32_to_cpu(de->inode), |
| 280 |
get_dtype(sb, de->file_type))) |
| 281 |
goto done; |
| 282 |
} else { |
| 283 |
int save_len = fstr.len; |
| 284 |
struct fscrypt_str de_name = |
| 285 |
FSTR_INIT(de->name, |
| 286 |
de->name_len); |
| 287 |
u32 hash; |
| 288 |
u32 minor_hash; |
| 289 |
|
| 290 |
if (IS_CASEFOLDED(inode)) { |
| 291 |
hash = EXT4_DIRENT_HASH(de); |
| 292 |
minor_hash = EXT4_DIRENT_MINOR_HASH(de); |
| 293 |
} else { |
| 294 |
hash = 0; |
| 295 |
minor_hash = 0; |
| 296 |
} |
| 297 |
|
| 298 |
/* Directory is encrypted */ |
| 299 |
err = fscrypt_fname_disk_to_usr(inode, |
| 300 |
hash, minor_hash, &de_name, &fstr); |
| 301 |
de_name = fstr; |
| 302 |
fstr.len = save_len; |
| 303 |
if (err) |
| 304 |
goto errout; |
| 305 |
if (!dir_emit(ctx, |
| 306 |
de_name.name, de_name.len, |
| 307 |
le32_to_cpu(de->inode), |
| 308 |
get_dtype(sb, de->file_type))) |
| 309 |
goto done; |
| 310 |
} |
| 311 |
} |
| 312 |
ctx->pos += ext4_rec_len_from_disk(de->rec_len, |
| 313 |
sb->s_blocksize); |
| 314 |
} |
| 315 |
if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode)) |
| 316 |
goto done; |
| 317 |
brelse(bh); |
| 318 |
bh = NULL; |
| 319 |
} |
| 320 |
done: |
| 321 |
err = 0; |
| 322 |
errout: |
| 323 |
fscrypt_fname_free_buffer(&fstr); |
| 324 |
brelse(bh); |
| 325 |
return err; |
| 326 |
} |
| 327 |
|
| 328 |
static inline int is_32bit_api(void) |
| 329 |
{ |
| 330 |
#ifdef CONFIG_COMPAT |
| 331 |
return in_compat_syscall(); |
| 332 |
#else |
| 333 |
return (BITS_PER_LONG == 32); |
| 334 |
#endif |
| 335 |
} |
| 336 |
|
| 337 |
/* |
| 338 |
* These functions convert from the major/minor hash to an f_pos |
| 339 |
* value for dx directories |
| 340 |
* |
| 341 |
* Upper layer (for example NFS) should specify FMODE_32BITHASH or |
| 342 |
* FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted |
| 343 |
* directly on both 32-bit and 64-bit nodes, under such case, neither |
| 344 |
* FMODE_32BITHASH nor FMODE_64BITHASH is specified. |
| 345 |
*/ |
| 346 |
static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) |
| 347 |
{ |
| 348 |
if ((filp->f_mode & FMODE_32BITHASH) || |
| 349 |
(!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) |
| 350 |
return major >> 1; |
| 351 |
else |
| 352 |
return ((__u64)(major >> 1) << 32) | (__u64)minor; |
| 353 |
} |
| 354 |
|
| 355 |
static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) |
| 356 |
{ |
| 357 |
if ((filp->f_mode & FMODE_32BITHASH) || |
| 358 |
(!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) |
| 359 |
return (pos << 1) & 0xffffffff; |
| 360 |
else |
| 361 |
return ((pos >> 32) << 1) & 0xffffffff; |
| 362 |
} |
| 363 |
|
| 364 |
static inline __u32 pos2min_hash(struct file *filp, loff_t pos) |
| 365 |
{ |
| 366 |
if ((filp->f_mode & FMODE_32BITHASH) || |
| 367 |
(!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) |
| 368 |
return 0; |
| 369 |
else |
| 370 |
return pos & 0xffffffff; |
| 371 |
} |
| 372 |
|
| 373 |
/* |
| 374 |
* Return 32- or 64-bit end-of-file for dx directories |
| 375 |
*/ |
| 376 |
static inline loff_t ext4_get_htree_eof(struct file *filp) |
| 377 |
{ |
| 378 |
if ((filp->f_mode & FMODE_32BITHASH) || |
| 379 |
(!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) |
| 380 |
return EXT4_HTREE_EOF_32BIT; |
| 381 |
else |
| 382 |
return EXT4_HTREE_EOF_64BIT; |
| 383 |
} |
| 384 |
|
| 385 |
|
| 386 |
/* |
| 387 |
* ext4_dir_llseek() calls generic_file_llseek_size to handle htree |
| 388 |
* directories, where the "offset" is in terms of the filename hash |
| 389 |
* value instead of the byte offset. |
| 390 |
* |
| 391 |
* Because we may return a 64-bit hash that is well beyond offset limits, |
| 392 |
* we need to pass the max hash as the maximum allowable offset in |
| 393 |
* the htree directory case. |
| 394 |
* |
| 395 |
* For non-htree, ext4_llseek already chooses the proper max offset. |
| 396 |
*/ |
| 397 |
static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) |
| 398 |
{ |
| 399 |
struct inode *inode = file->f_mapping->host; |
| 400 |
struct dir_private_info *info = file->private_data; |
| 401 |
int dx_dir = is_dx_dir(inode); |
| 402 |
loff_t ret, htree_max = ext4_get_htree_eof(file); |
| 403 |
|
| 404 |
if (likely(dx_dir)) |
| 405 |
ret = generic_file_llseek_size(file, offset, whence, |
| 406 |
htree_max, htree_max); |
| 407 |
else |
| 408 |
ret = ext4_llseek(file, offset, whence); |
| 409 |
info->cookie = inode_peek_iversion(inode) - 1; |
| 410 |
return ret; |
| 411 |
} |
| 412 |
|
| 413 |
/* |
| 414 |
* This structure holds the nodes of the red-black tree used to store |
| 415 |
* the directory entry in hash order. |
| 416 |
*/ |
| 417 |
struct fname { |
| 418 |
__u32 hash; |
| 419 |
__u32 minor_hash; |
| 420 |
struct rb_node rb_hash; |
| 421 |
struct fname *next; |
| 422 |
__u32 inode; |
| 423 |
__u8 name_len; |
| 424 |
__u8 file_type; |
| 425 |
char name[] __counted_by(name_len); |
| 426 |
}; |
| 427 |
|
| 428 |
/* |
| 429 |
* This function implements a non-recursive way of freeing all of the |
| 430 |
* nodes in the red-black tree. |
| 431 |
*/ |
| 432 |
static void free_rb_tree_fname(struct rb_root *root) |
| 433 |
{ |
| 434 |
struct fname *fname, *next; |
| 435 |
|
| 436 |
rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) |
| 437 |
while (fname) { |
| 438 |
struct fname *old = fname; |
| 439 |
fname = fname->next; |
| 440 |
kfree(old); |
| 441 |
} |
| 442 |
|
| 443 |
*root = RB_ROOT; |
| 444 |
} |
| 445 |
|
| 446 |
static void ext4_htree_init_dir_info(struct file *filp, loff_t pos) |
| 447 |
{ |
| 448 |
struct dir_private_info *p = filp->private_data; |
| 449 |
|
| 450 |
if (is_dx_dir(file_inode(filp)) && !p->initialized) { |
| 451 |
p->curr_hash = pos2maj_hash(filp, pos); |
| 452 |
p->curr_minor_hash = pos2min_hash(filp, pos); |
| 453 |
p->initialized = true; |
| 454 |
} |
| 455 |
} |
| 456 |
|
| 457 |
void ext4_htree_free_dir_info(struct dir_private_info *p) |
| 458 |
{ |
| 459 |
free_rb_tree_fname(&p->root); |
| 460 |
kfree(p); |
| 461 |
} |
| 462 |
|
| 463 |
/* |
| 464 |
* Given a directory entry, enter it into the fname rb tree. |
| 465 |
* |
| 466 |
* When filename encryption is enabled, the dirent will hold the |
| 467 |
* encrypted filename, while the htree will hold decrypted filename. |
| 468 |
* The decrypted filename is passed in via ent_name. parameter. |
| 469 |
*/ |
| 470 |
int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, |
| 471 |
__u32 minor_hash, |
| 472 |
struct ext4_dir_entry_2 *dirent, |
| 473 |
struct fscrypt_str *ent_name) |
| 474 |
{ |
| 475 |
struct rb_node **p, *parent = NULL; |
| 476 |
struct fname *fname, *new_fn; |
| 477 |
struct dir_private_info *info; |
| 478 |
|
| 479 |
info = dir_file->private_data; |
| 480 |
p = &info->root.rb_node; |
| 481 |
|
| 482 |
/* Create and allocate the fname structure */ |
| 483 |
new_fn = kzalloc_flex(*new_fn, name, ent_name->len + 1); |
| 484 |
if (!new_fn) |
| 485 |
return -ENOMEM; |
| 486 |
new_fn->hash = hash; |
| 487 |
new_fn->minor_hash = minor_hash; |
| 488 |
new_fn->inode = le32_to_cpu(dirent->inode); |
| 489 |
new_fn->name_len = ent_name->len; |
| 490 |
new_fn->file_type = dirent->file_type; |
| 491 |
memcpy(new_fn->name, ent_name->name, ent_name->len); |
| 492 |
|
| 493 |
while (*p) { |
| 494 |
parent = *p; |
| 495 |
fname = rb_entry(parent, struct fname, rb_hash); |
| 496 |
|
| 497 |
/* |
| 498 |
* If the hash and minor hash match up, then we put |
| 499 |
* them on a linked list. This rarely happens... |
| 500 |
*/ |
| 501 |
if ((new_fn->hash == fname->hash) && |
| 502 |
(new_fn->minor_hash == fname->minor_hash)) { |
| 503 |
new_fn->next = fname->next; |
| 504 |
fname->next = new_fn; |
| 505 |
return 0; |
| 506 |
} |
| 507 |
|
| 508 |
if (new_fn->hash < fname->hash) |
| 509 |
p = &(*p)->rb_left; |
| 510 |
else if (new_fn->hash > fname->hash) |
| 511 |
p = &(*p)->rb_right; |
| 512 |
else if (new_fn->minor_hash < fname->minor_hash) |
| 513 |
p = &(*p)->rb_left; |
| 514 |
else /* if (new_fn->minor_hash > fname->minor_hash) */ |
| 515 |
p = &(*p)->rb_right; |
| 516 |
} |
| 517 |
|
| 518 |
rb_link_node(&new_fn->rb_hash, parent, p); |
| 519 |
rb_insert_color(&new_fn->rb_hash, &info->root); |
| 520 |
return 0; |
| 521 |
} |
| 522 |
|
| 523 |
|
| 524 |
|
| 525 |
/* |
| 526 |
* This is a helper function for ext4_dx_readdir. It calls filldir |
| 527 |
* for all entries on the fname linked list. (Normally there is only |
| 528 |
* one entry on the linked list, unless there are 62 bit hash collisions.) |
| 529 |
*/ |
| 530 |
static int call_filldir(struct file *file, struct dir_context *ctx, |
| 531 |
struct fname *fname) |
| 532 |
{ |
| 533 |
struct dir_private_info *info = file->private_data; |
| 534 |
struct inode *inode = file_inode(file); |
| 535 |
struct super_block *sb = inode->i_sb; |
| 536 |
|
| 537 |
if (!fname) { |
| 538 |
ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " |
| 539 |
"called with null fname?!?", __func__, __LINE__, |
| 540 |
inode->i_ino, current->comm); |
| 541 |
return 0; |
| 542 |
} |
| 543 |
ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); |
| 544 |
while (fname) { |
| 545 |
if (!dir_emit(ctx, fname->name, |
| 546 |
fname->name_len, |
| 547 |
fname->inode, |
| 548 |
get_dtype(sb, fname->file_type))) { |
| 549 |
info->extra_fname = fname; |
| 550 |
return 1; |
| 551 |
} |
| 552 |
fname = fname->next; |
| 553 |
} |
| 554 |
return 0; |
| 555 |
} |
| 556 |
|
| 557 |
static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) |
| 558 |
{ |
| 559 |
struct dir_private_info *info = file->private_data; |
| 560 |
struct inode *inode = file_inode(file); |
| 561 |
struct fname *fname; |
| 562 |
int ret = 0; |
| 563 |
|
| 564 |
ext4_htree_init_dir_info(file, ctx->pos); |
| 565 |
|
| 566 |
if (ctx->pos == ext4_get_htree_eof(file)) |
| 567 |
return 0; /* EOF */ |
| 568 |
|
| 569 |
/* Some one has messed with f_pos; reset the world */ |
| 570 |
if (info->last_pos != ctx->pos) { |
| 571 |
free_rb_tree_fname(&info->root); |
| 572 |
info->curr_node = NULL; |
| 573 |
info->extra_fname = NULL; |
| 574 |
info->curr_hash = pos2maj_hash(file, ctx->pos); |
| 575 |
info->curr_minor_hash = pos2min_hash(file, ctx->pos); |
| 576 |
} |
| 577 |
|
| 578 |
/* |
| 579 |
* If there are any leftover names on the hash collision |
| 580 |
* chain, return them first. |
| 581 |
*/ |
| 582 |
if (info->extra_fname) { |
| 583 |
if (call_filldir(file, ctx, info->extra_fname)) |
| 584 |
goto finished; |
| 585 |
info->extra_fname = NULL; |
| 586 |
goto next_node; |
| 587 |
} else if (!info->curr_node) |
| 588 |
info->curr_node = rb_first(&info->root); |
| 589 |
|
| 590 |
while (1) { |
| 591 |
/* |
| 592 |
* Fill the rbtree if we have no more entries, |
| 593 |
* or the inode has changed since we last read in the |
| 594 |
* cached entries. |
| 595 |
*/ |
| 596 |
if ((!info->curr_node) || |
| 597 |
!inode_eq_iversion(inode, info->cookie)) { |
| 598 |
info->curr_node = NULL; |
| 599 |
free_rb_tree_fname(&info->root); |
| 600 |
info->cookie = inode_query_iversion(inode); |
| 601 |
ret = ext4_htree_fill_tree(file, info->curr_hash, |
| 602 |
info->curr_minor_hash, |
| 603 |
&info->next_hash); |
| 604 |
if (ret < 0) |
| 605 |
goto finished; |
| 606 |
if (ret == 0) { |
| 607 |
ctx->pos = ext4_get_htree_eof(file); |
| 608 |
break; |
| 609 |
} |
| 610 |
info->curr_node = rb_first(&info->root); |
| 611 |
} |
| 612 |
|
| 613 |
fname = rb_entry(info->curr_node, struct fname, rb_hash); |
| 614 |
info->curr_hash = fname->hash; |
| 615 |
info->curr_minor_hash = fname->minor_hash; |
| 616 |
if (call_filldir(file, ctx, fname)) |
| 617 |
break; |
| 618 |
next_node: |
| 619 |
info->curr_node = rb_next(info->curr_node); |
| 620 |
if (info->curr_node) { |
| 621 |
fname = rb_entry(info->curr_node, struct fname, |
| 622 |
rb_hash); |
| 623 |
info->curr_hash = fname->hash; |
| 624 |
info->curr_minor_hash = fname->minor_hash; |
| 625 |
} else { |
| 626 |
if (info->next_hash == ~0) { |
| 627 |
ctx->pos = ext4_get_htree_eof(file); |
| 628 |
break; |
| 629 |
} |
| 630 |
info->curr_hash = info->next_hash; |
| 631 |
info->curr_minor_hash = 0; |
| 632 |
} |
| 633 |
} |
| 634 |
finished: |
| 635 |
info->last_pos = ctx->pos; |
| 636 |
return ret < 0 ? ret : 0; |
| 637 |
} |
| 638 |
|
| 639 |
static int ext4_release_dir(struct inode *inode, struct file *filp) |
| 640 |
{ |
| 641 |
if (filp->private_data) |
| 642 |
ext4_htree_free_dir_info(filp->private_data); |
| 643 |
|
| 644 |
return 0; |
| 645 |
} |
| 646 |
|
| 647 |
int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, |
| 648 |
int buf_size) |
| 649 |
{ |
| 650 |
struct ext4_dir_entry_2 *de; |
| 651 |
int rlen; |
| 652 |
unsigned int offset = 0; |
| 653 |
char *top; |
| 654 |
|
| 655 |
de = buf; |
| 656 |
top = buf + buf_size; |
| 657 |
while ((char *) de < top) { |
| 658 |
if (ext4_check_dir_entry(dir, NULL, de, bh, |
| 659 |
buf, buf_size, offset)) |
| 660 |
return -EFSCORRUPTED; |
| 661 |
rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); |
| 662 |
de = (struct ext4_dir_entry_2 *)((char *)de + rlen); |
| 663 |
offset += rlen; |
| 664 |
} |
| 665 |
if ((char *) de > top) |
| 666 |
return -EFSCORRUPTED; |
| 667 |
|
| 668 |
return 0; |
| 669 |
} |
| 670 |
|
| 671 |
static int ext4_dir_open(struct inode *inode, struct file *file) |
| 672 |
{ |
| 673 |
struct dir_private_info *info; |
| 674 |
|
| 675 |
info = kzalloc_obj(*info); |
| 676 |
if (!info) |
| 677 |
return -ENOMEM; |
| 678 |
file->private_data = info; |
| 679 |
return 0; |
| 680 |
} |
| 681 |
|
| 682 |
const struct file_operations ext4_dir_operations = { |
| 683 |
.open = ext4_dir_open, |
| 684 |
.llseek = ext4_dir_llseek, |
| 685 |
.read = generic_read_dir, |
| 686 |
.iterate_shared = ext4_readdir, |
| 687 |
.unlocked_ioctl = ext4_ioctl, |
| 688 |
#ifdef CONFIG_COMPAT |
| 689 |
.compat_ioctl = ext4_compat_ioctl, |
| 690 |
#endif |
| 691 |
.fsync = ext4_sync_file, |
| 692 |
.release = ext4_release_dir, |
| 693 |
.setlease = generic_setlease, |
| 694 |
}; |