| 1 | |
| 2 | |
| 3 | * linux/fs/ext4/namei.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/namei.c
|
| 13 | *
|
| 14 | * Copyright (C) 1991, 1992 Linus Torvalds
|
| 15 | *
|
| 16 | * Big-endian to little-endian byte-swapping/bitmaps by
|
| 17 | * David S. Miller ([email protected]), 1995
|
| 18 | * Directory entry file type support and forward compatibility hooks
|
| 19 | * for B-tree directories by Theodore Ts'o ([email protected]), 1998
|
| 20 | * Hash Tree Directory indexing (c)
|
| 21 | * Daniel Phillips, 2001
|
| 22 | * Hash Tree Directory indexing porting
|
| 23 | * Christopher Li, 2002
|
| 24 | * Hash Tree Directory indexing cleanup
|
| 25 | * Theodore Ts'o, 2002
|
| 26 |
|
| 27 |
|
| 28 | #include <linux/fs.h>
|
| 29 | #include <linux/pagemap.h>
|
| 30 | #include <linux/time.h>
|
| 31 | #include <linux/fcntl.h>
|
| 32 | #include <linux/stat.h>
|
| 33 | #include <linux/string.h>
|
| 34 | #include <linux/quotaops.h>
|
| 35 | #include <linux/buffer_head.h>
|
| 36 | #include <linux/bio.h>
|
| 37 | #include <linux/iversion.h>
|
| 38 | #include <linux/unicode.h>
|
| 39 | #include "ext4.h"
|
| 40 | #include "ext4_jbd2.h"
|
| 41 |
|
| 42 | #include "xattr.h"
|
| 43 | #include "acl.h"
|
| 44 |
|
| 45 | #include <trace/events/ext4.h>
|
| 46 | |
| 47 | * define how far ahead to read directories while searching them.
|
| 48 |
|
| 49 | #define NAMEI_RA_CHUNKS 2
|
| 50 | #define NAMEI_RA_BLOCKS 4
|
| 51 | #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
|
| 52 |
|
| 53 | static struct buffer_head *ext4_append(handle_t *handle,
|
| 54 | struct inode *inode,
|
| 55 | ext4_lblk_t *block)
|
| 56 | {
|
| 57 | struct ext4_map_blocks map;
|
| 58 | struct buffer_head *bh;
|
| 59 | int err;
|
| 60 |
|
| 61 | if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
|
| 62 | ((inode->i_size >> 10) >=
|
| 63 | EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
|
| 64 | return ERR_PTR(-ENOSPC);
|
| 65 |
|
| 66 | *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
|
| 67 | map.m_lblk = *block;
|
| 68 | map.m_len = 1;
|
| 69 |
|
| 70 | |
| 71 | * We're appending new directory block. Make sure the block is not
|
| 72 | * allocated yet, otherwise we will end up corrupting the
|
| 73 | * directory.
|
| 74 |
|
| 75 | err = ext4_map_blocks(NULL, inode, &map, 0);
|
| 76 | if (err < 0)
|
| 77 | return ERR_PTR(err);
|
| 78 | if (err) {
|
| 79 | EXT4_ERROR_INODE(inode, "Logical block already allocated");
|
| 80 | return ERR_PTR(-EFSCORRUPTED);
|
| 81 | }
|
| 82 |
|
| 83 | bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
|
| 84 | if (IS_ERR(bh))
|
| 85 | return bh;
|
| 86 | inode->i_size += inode->i_sb->s_blocksize;
|
| 87 | EXT4_I(inode)->i_disksize = inode->i_size;
|
| 88 | err = ext4_mark_inode_dirty(handle, inode);
|
| 89 | if (err)
|
| 90 | goto out;
|
| 91 | BUFFER_TRACE(bh, "get_write_access");
|
| 92 | err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
|
| 93 | EXT4_JTR_NONE);
|
| 94 | if (err)
|
| 95 | goto out;
|
| 96 | return bh;
|
| 97 |
|
| 98 | out:
|
| 99 | brelse(bh);
|
| 100 | ext4_std_error(inode->i_sb, err);
|
| 101 | return ERR_PTR(err);
|
| 102 | }
|
| 103 |
|
| 104 | static int ext4_dx_csum_verify(struct inode *inode,
|
| 105 | struct ext4_dir_entry *dirent);
|
| 106 |
|
| 107 | |
| 108 | * Hints to ext4_read_dirblock regarding whether we expect a directory
|
| 109 | * block being read to be an index block, or a block containing
|
| 110 | * directory entries (and if the latter, whether it was found via a
|
| 111 | * logical block in an htree index block). This is used to control
|
| 112 | * what sort of sanity checkinig ext4_read_dirblock() will do on the
|
| 113 | * directory block read from the storage device. EITHER will means
|
| 114 | * the caller doesn't know what kind of directory block will be read,
|
| 115 | * so no specific verification will be done.
|
| 116 |
|
| 117 | typedef enum {
|
| 118 | EITHER, INDEX, DIRENT, DIRENT_HTREE
|
| 119 | } dirblock_type_t;
|
| 120 |
|
| 121 | #define ext4_read_dirblock(inode, block, type) \
|
| 122 | __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
|
| 123 |
|
| 124 | static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
|
| 125 | ext4_lblk_t block,
|
| 126 | dirblock_type_t type,
|
| 127 | const char *func,
|
| 128 | unsigned int line)
|
| 129 | {
|
| 130 | struct buffer_head *bh;
|
| 131 | struct ext4_dir_entry *dirent;
|
| 132 | int is_dx_block = 0;
|
| 133 |
|
| 134 | if (block >= inode->i_size >> inode->i_blkbits) {
|
| 135 | ext4_error_inode(inode, func, line, block,
|
| 136 | "Attempting to read directory block (%u) that is past i_size (%llu)",
|
| 137 | block, inode->i_size);
|
| 138 | return ERR_PTR(-EFSCORRUPTED);
|
| 139 | }
|
| 140 |
|
| 141 | if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
|
| 142 | bh = ERR_PTR(-EIO);
|
| 143 | else
|
| 144 | bh = ext4_bread(NULL, inode, block, 0);
|
| 145 | if (IS_ERR(bh)) {
|
| 146 | __ext4_warning(inode->i_sb, func, line,
|
| 147 | "inode #%llu: lblock %lu: comm %s: "
|
| 148 | "error %pe reading directory block",
|
| 149 | inode->i_ino, (unsigned long)block,
|
| 150 | current->comm, bh);
|
| 151 |
|
| 152 | return bh;
|
| 153 | }
|
| 154 |
|
| 155 | if (!bh && (type == INDEX || type == DIRENT_HTREE || block == 0)) {
|
| 156 | ext4_error_inode(inode, func, line, block,
|
| 157 | "Directory hole found for htree %s block %u",
|
| 158 | (type == INDEX) ? "index" : "leaf", block);
|
| 159 | return ERR_PTR(-EFSCORRUPTED);
|
| 160 | }
|
| 161 | if (!bh)
|
| 162 | return NULL;
|
| 163 | dirent = (struct ext4_dir_entry *) bh->b_data;
|
| 164 |
|
| 165 | if (is_dx(inode)) {
|
| 166 | if (block == 0)
|
| 167 | is_dx_block = 1;
|
| 168 | else if (ext4_rec_len_from_disk(dirent->rec_len,
|
| 169 | inode->i_sb->s_blocksize) ==
|
| 170 | inode->i_sb->s_blocksize)
|
| 171 | is_dx_block = 1;
|
| 172 | }
|
| 173 | if (!is_dx_block && type == INDEX) {
|
| 174 | ext4_error_inode(inode, func, line, block,
|
| 175 | "directory leaf block found instead of index block");
|
| 176 | brelse(bh);
|
| 177 | return ERR_PTR(-EFSCORRUPTED);
|
| 178 | }
|
| 179 | if (!ext4_has_feature_metadata_csum(inode->i_sb) ||
|
| 180 | buffer_verified(bh))
|
| 181 | return bh;
|
| 182 |
|
| 183 | |
| 184 | * An empty leaf block can get mistaken for a index block; for
|
| 185 | * this reason, we can only check the index checksum when the
|
| 186 | * caller is sure it should be an index block.
|
| 187 |
|
| 188 | if (is_dx_block && type == INDEX) {
|
| 189 | if (ext4_dx_csum_verify(inode, dirent) &&
|
| 190 | !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
|
| 191 | set_buffer_verified(bh);
|
| 192 | else {
|
| 193 | ext4_error_inode_err(inode, func, line, block,
|
| 194 | EFSBADCRC,
|
| 195 | "Directory index failed checksum");
|
| 196 | brelse(bh);
|
| 197 | return ERR_PTR(-EFSBADCRC);
|
| 198 | }
|
| 199 | }
|
| 200 | if (!is_dx_block) {
|
| 201 | if (ext4_dirblock_csum_verify(inode, bh) &&
|
| 202 | !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
|
| 203 | set_buffer_verified(bh);
|
| 204 | else {
|
| 205 | ext4_error_inode_err(inode, func, line, block,
|
| 206 | EFSBADCRC,
|
| 207 | "Directory block failed checksum");
|
| 208 | brelse(bh);
|
| 209 | return ERR_PTR(-EFSBADCRC);
|
| 210 | }
|
| 211 | }
|
| 212 | return bh;
|
| 213 | }
|
| 214 |
|
| 215 | #ifdef DX_DEBUG
|
| 216 | #define dxtrace(command) command
|
| 217 | #else
|
| 218 | #define dxtrace(command)
|
| 219 | #endif
|
| 220 |
|
| 221 | struct fake_dirent
|
| 222 | {
|
| 223 | __le32 inode;
|
| 224 | __le16 rec_len;
|
| 225 | u8 name_len;
|
| 226 | u8 file_type;
|
| 227 | };
|
| 228 |
|
| 229 | struct dx_countlimit
|
| 230 | {
|
| 231 | __le16 limit;
|
| 232 | __le16 count;
|
| 233 | };
|
| 234 |
|
| 235 | struct dx_entry
|
| 236 | {
|
| 237 | __le32 hash;
|
| 238 | __le32 block;
|
| 239 | };
|
| 240 |
|
| 241 | |
| 242 | * dx_root_info is laid out so that if it should somehow get overlaid by a
|
| 243 | * dirent the two low bits of the hash version will be zero. Therefore, the
|
| 244 | * hash version mod 4 should never be 0. Sincerely, the paranoia department.
|
| 245 |
|
| 246 |
|
| 247 | struct dx_root
|
| 248 | {
|
| 249 | struct fake_dirent dot;
|
| 250 | char dot_name[4];
|
| 251 | struct fake_dirent dotdot;
|
| 252 | char dotdot_name[4];
|
| 253 | struct dx_root_info
|
| 254 | {
|
| 255 | __le32 reserved_zero;
|
| 256 | u8 hash_version;
|
| 257 | u8 info_length;
|
| 258 | u8 indirect_levels;
|
| 259 | u8 unused_flags;
|
| 260 | }
|
| 261 | info;
|
| 262 | struct dx_entry entries[];
|
| 263 | };
|
| 264 |
|
| 265 | struct dx_node
|
| 266 | {
|
| 267 | struct fake_dirent fake;
|
| 268 | struct dx_entry entries[];
|
| 269 | };
|
| 270 |
|
| 271 |
|
| 272 | struct dx_frame
|
| 273 | {
|
| 274 | struct buffer_head *bh;
|
| 275 | struct dx_entry *entries;
|
| 276 | struct dx_entry *at;
|
| 277 | };
|
| 278 |
|
| 279 | struct dx_map_entry
|
| 280 | {
|
| 281 | u32 hash;
|
| 282 | u16 offs;
|
| 283 | u16 size;
|
| 284 | };
|
| 285 |
|
| 286 | |
| 287 | * This goes at the end of each htree block.
|
| 288 |
|
| 289 | struct dx_tail {
|
| 290 | u32 dt_reserved;
|
| 291 | __le32 dt_checksum;
|
| 292 | };
|
| 293 |
|
| 294 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
|
| 295 | struct ext4_filename *fname,
|
| 296 | struct ext4_dir_entry_2 **res_dir);
|
| 297 | static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
|
| 298 | struct inode *dir, struct inode *inode);
|
| 299 |
|
| 300 |
|
| 301 | void ext4_initialize_dirent_tail(struct buffer_head *bh,
|
| 302 | unsigned int blocksize)
|
| 303 | {
|
| 304 | struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
|
| 305 |
|
| 306 | memset(t, 0, sizeof(struct ext4_dir_entry_tail));
|
| 307 | t->det_rec_len = ext4_rec_len_to_disk(
|
| 308 | sizeof(struct ext4_dir_entry_tail), blocksize);
|
| 309 | t->det_reserved_ft = EXT4_FT_DIR_CSUM;
|
| 310 | }
|
| 311 |
|
| 312 |
|
| 313 | static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
|
| 314 | struct buffer_head *bh)
|
| 315 | {
|
| 316 | struct ext4_dir_entry_tail *t;
|
| 317 | int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
|
| 318 |
|
| 319 | #ifdef PARANOID
|
| 320 | struct ext4_dir_entry *d, *top;
|
| 321 |
|
| 322 | d = (struct ext4_dir_entry *)bh->b_data;
|
| 323 | top = (struct ext4_dir_entry *)(bh->b_data +
|
| 324 | (blocksize - sizeof(struct ext4_dir_entry_tail)));
|
| 325 | while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
|
| 326 | d = (struct ext4_dir_entry *)(((void *)d) +
|
| 327 | ext4_rec_len_from_disk(d->rec_len, blocksize));
|
| 328 |
|
| 329 | if (d != top)
|
| 330 | return NULL;
|
| 331 |
|
| 332 | t = (struct ext4_dir_entry_tail *)d;
|
| 333 | #else
|
| 334 | t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
|
| 335 | #endif
|
| 336 |
|
| 337 | if (t->det_reserved_zero1 ||
|
| 338 | (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
|
| 339 | sizeof(struct ext4_dir_entry_tail)) ||
|
| 340 | t->det_reserved_zero2 ||
|
| 341 | t->det_reserved_ft != EXT4_FT_DIR_CSUM)
|
| 342 | return NULL;
|
| 343 |
|
| 344 | return t;
|
| 345 | }
|
| 346 |
|
| 347 | static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
|
| 348 | {
|
| 349 | struct ext4_inode_info *ei = EXT4_I(inode);
|
| 350 | __u32 csum;
|
| 351 |
|
| 352 | csum = ext4_chksum(ei->i_csum_seed, (__u8 *)dirent, size);
|
| 353 | return cpu_to_le32(csum);
|
| 354 | }
|
| 355 |
|
| 356 | #define warn_no_space_for_csum(inode) \
|
| 357 | __warn_no_space_for_csum((inode), __func__, __LINE__)
|
| 358 |
|
| 359 | static void __warn_no_space_for_csum(struct inode *inode, const char *func,
|
| 360 | unsigned int line)
|
| 361 | {
|
| 362 | __ext4_warning_inode(inode, func, line,
|
| 363 | "No space for directory leaf checksum. Please run e2fsck -D.");
|
| 364 | }
|
| 365 |
|
| 366 | int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
|
| 367 | {
|
| 368 | struct ext4_dir_entry_tail *t;
|
| 369 |
|
| 370 | if (!ext4_has_feature_metadata_csum(inode->i_sb))
|
| 371 | return 1;
|
| 372 |
|
| 373 | t = get_dirent_tail(inode, bh);
|
| 374 | if (!t) {
|
| 375 | warn_no_space_for_csum(inode);
|
| 376 | return 0;
|
| 377 | }
|
| 378 |
|
| 379 | if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
|
| 380 | (char *)t - bh->b_data))
|
| 381 | return 0;
|
| 382 |
|
| 383 | return 1;
|
| 384 | }
|
| 385 |
|
| 386 | static void ext4_dirblock_csum_set(struct inode *inode,
|
| 387 | struct buffer_head *bh)
|
| 388 | {
|
| 389 | struct ext4_dir_entry_tail *t;
|
| 390 |
|
| 391 | if (!ext4_has_feature_metadata_csum(inode->i_sb))
|
| 392 | return;
|
| 393 |
|
| 394 | t = get_dirent_tail(inode, bh);
|
| 395 | if (!t) {
|
| 396 | warn_no_space_for_csum(inode);
|
| 397 | return;
|
| 398 | }
|
| 399 |
|
| 400 | t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
|
| 401 | (char *)t - bh->b_data);
|
| 402 | }
|
| 403 |
|
| 404 | int ext4_handle_dirty_dirblock(handle_t *handle,
|
| 405 | struct inode *inode,
|
| 406 | struct buffer_head *bh)
|
| 407 | {
|
| 408 | ext4_dirblock_csum_set(inode, bh);
|
| 409 | return ext4_handle_dirty_metadata(handle, inode, bh);
|
| 410 | }
|
| 411 |
|
| 412 | static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
|
| 413 | struct ext4_dir_entry *dirent,
|
| 414 | int *offset)
|
| 415 | {
|
| 416 | struct ext4_dir_entry *dp;
|
| 417 | struct dx_root_info *root;
|
| 418 | int count_offset;
|
| 419 | int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
|
| 420 | unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
|
| 421 |
|
| 422 | if (rlen == blocksize)
|
| 423 | count_offset = 8;
|
| 424 | else if (rlen == 12) {
|
| 425 | dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
|
| 426 | if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
|
| 427 | return NULL;
|
| 428 | root = (struct dx_root_info *)(((void *)dp + 12));
|
| 429 | if (root->reserved_zero ||
|
| 430 | root->info_length != sizeof(struct dx_root_info))
|
| 431 | return NULL;
|
| 432 | count_offset = 32;
|
| 433 | } else
|
| 434 | return NULL;
|
| 435 |
|
| 436 | if (offset)
|
| 437 | *offset = count_offset;
|
| 438 | return (struct dx_countlimit *)(((void *)dirent) + count_offset);
|
| 439 | }
|
| 440 |
|
| 441 | static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
|
| 442 | int count_offset, int count, struct dx_tail *t)
|
| 443 | {
|
| 444 | struct ext4_inode_info *ei = EXT4_I(inode);
|
| 445 | __u32 csum;
|
| 446 | int size;
|
| 447 | __u32 dummy_csum = 0;
|
| 448 | int offset = offsetof(struct dx_tail, dt_checksum);
|
| 449 |
|
| 450 | size = count_offset + (count * sizeof(struct dx_entry));
|
| 451 | csum = ext4_chksum(ei->i_csum_seed, (__u8 *)dirent, size);
|
| 452 | csum = ext4_chksum(csum, (__u8 *)t, offset);
|
| 453 | csum = ext4_chksum(csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
|
| 454 |
|
| 455 | return cpu_to_le32(csum);
|
| 456 | }
|
| 457 |
|
| 458 | static int ext4_dx_csum_verify(struct inode *inode,
|
| 459 | struct ext4_dir_entry *dirent)
|
| 460 | {
|
| 461 | struct dx_countlimit *c;
|
| 462 | struct dx_tail *t;
|
| 463 | int count_offset, limit, count;
|
| 464 |
|
| 465 | if (!ext4_has_feature_metadata_csum(inode->i_sb))
|
| 466 | return 1;
|
| 467 |
|
| 468 | c = get_dx_countlimit(inode, dirent, &count_offset);
|
| 469 | if (!c) {
|
| 470 | EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
|
| 471 | return 0;
|
| 472 | }
|
| 473 | limit = le16_to_cpu(c->limit);
|
| 474 | count = le16_to_cpu(c->count);
|
| 475 | if (count_offset + (limit * sizeof(struct dx_entry)) >
|
| 476 | EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
|
| 477 | warn_no_space_for_csum(inode);
|
| 478 | return 0;
|
| 479 | }
|
| 480 | t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
|
| 481 |
|
| 482 | if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
|
| 483 | count, t))
|
| 484 | return 0;
|
| 485 | return 1;
|
| 486 | }
|
| 487 |
|
| 488 | static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
|
| 489 | {
|
| 490 | struct dx_countlimit *c;
|
| 491 | struct dx_tail *t;
|
| 492 | int count_offset, limit, count;
|
| 493 |
|
| 494 | if (!ext4_has_feature_metadata_csum(inode->i_sb))
|
| 495 | return;
|
| 496 |
|
| 497 | c = get_dx_countlimit(inode, dirent, &count_offset);
|
| 498 | if (!c) {
|
| 499 | EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
|
| 500 | return;
|
| 501 | }
|
| 502 | limit = le16_to_cpu(c->limit);
|
| 503 | count = le16_to_cpu(c->count);
|
| 504 | if (count_offset + (limit * sizeof(struct dx_entry)) >
|
| 505 | EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
|
| 506 | warn_no_space_for_csum(inode);
|
| 507 | return;
|
| 508 | }
|
| 509 | t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
|
| 510 |
|
| 511 | t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
|
| 512 | }
|
| 513 |
|
| 514 | static inline int ext4_handle_dirty_dx_node(handle_t *handle,
|
| 515 | struct inode *inode,
|
| 516 | struct buffer_head *bh)
|
| 517 | {
|
| 518 | ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
|
| 519 | return ext4_handle_dirty_metadata(handle, inode, bh);
|
| 520 | }
|
| 521 |
|
| 522 | |
| 523 | * p is at least 6 bytes before the end of page
|
| 524 |
|
| 525 | static inline struct ext4_dir_entry_2 *
|
| 526 | ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
|
| 527 | {
|
| 528 | return (struct ext4_dir_entry_2 *)((char *)p +
|
| 529 | ext4_rec_len_from_disk(p->rec_len, blocksize));
|
| 530 | }
|
| 531 |
|
| 532 | |
| 533 | * Future: use high four bits of block for coalesce-on-delete flags
|
| 534 | * Mask them off for now.
|
| 535 |
|
| 536 |
|
| 537 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
|
| 538 | {
|
| 539 | return le32_to_cpu(entry->block) & 0x0fffffff;
|
| 540 | }
|
| 541 |
|
| 542 | static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
|
| 543 | {
|
| 544 | entry->block = cpu_to_le32(value);
|
| 545 | }
|
| 546 |
|
| 547 | static inline unsigned dx_get_hash(struct dx_entry *entry)
|
| 548 | {
|
| 549 | return le32_to_cpu(entry->hash);
|
| 550 | }
|
| 551 |
|
| 552 | static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
|
| 553 | {
|
| 554 | entry->hash = cpu_to_le32(value);
|
| 555 | }
|
| 556 |
|
| 557 | static inline unsigned dx_get_count(struct dx_entry *entries)
|
| 558 | {
|
| 559 | return le16_to_cpu(((struct dx_countlimit *) entries)->count);
|
| 560 | }
|
| 561 |
|
| 562 | static inline unsigned dx_get_limit(struct dx_entry *entries)
|
| 563 | {
|
| 564 | return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
|
| 565 | }
|
| 566 |
|
| 567 | static inline void dx_set_count(struct dx_entry *entries, unsigned value)
|
| 568 | {
|
| 569 | ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
|
| 570 | }
|
| 571 |
|
| 572 | static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
|
| 573 | {
|
| 574 | ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
|
| 575 | }
|
| 576 |
|
| 577 | static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
|
| 578 | {
|
| 579 | unsigned int entry_space = dir->i_sb->s_blocksize -
|
| 580 | ext4_dir_rec_len(1, NULL) -
|
| 581 | ext4_dir_rec_len(2, NULL) - infosize;
|
| 582 |
|
| 583 | if (ext4_has_feature_metadata_csum(dir->i_sb))
|
| 584 | entry_space -= sizeof(struct dx_tail);
|
| 585 | return entry_space / sizeof(struct dx_entry);
|
| 586 | }
|
| 587 |
|
| 588 | static inline unsigned dx_node_limit(struct inode *dir)
|
| 589 | {
|
| 590 | unsigned int entry_space = dir->i_sb->s_blocksize -
|
| 591 | ext4_dir_rec_len(0, dir);
|
| 592 |
|
| 593 | if (ext4_has_feature_metadata_csum(dir->i_sb))
|
| 594 | entry_space -= sizeof(struct dx_tail);
|
| 595 | return entry_space / sizeof(struct dx_entry);
|
| 596 | }
|
| 597 |
|
| 598 | |
| 599 | * Debug
|
| 600 |
|
| 601 | #ifdef DX_DEBUG
|
| 602 | static void dx_show_index(char * label, struct dx_entry *entries)
|
| 603 | {
|
| 604 | int i, n = dx_get_count (entries);
|
| 605 | printk(KERN_DEBUG "%s index", label);
|
| 606 | for (i = 0; i < n; i++) {
|
| 607 | printk(KERN_CONT " %x->%lu",
|
| 608 | i ? dx_get_hash(entries + i) : 0,
|
| 609 | (unsigned long)dx_get_block(entries + i));
|
| 610 | }
|
| 611 | printk(KERN_CONT "\n");
|
| 612 | }
|
| 613 |
|
| 614 | struct stats
|
| 615 | {
|
| 616 | unsigned names;
|
| 617 | unsigned space;
|
| 618 | unsigned bcount;
|
| 619 | };
|
| 620 |
|
| 621 | static struct stats dx_show_leaf(struct inode *dir,
|
| 622 | struct dx_hash_info *hinfo,
|
| 623 | struct ext4_dir_entry_2 *de,
|
| 624 | int size, int show_names)
|
| 625 | {
|
| 626 | unsigned names = 0, space = 0;
|
| 627 | char *base = (char *) de;
|
| 628 | struct dx_hash_info h = *hinfo;
|
| 629 |
|
| 630 | printk("names: ");
|
| 631 | while ((char *) de < base + size)
|
| 632 | {
|
| 633 | if (de->inode)
|
| 634 | {
|
| 635 | if (show_names)
|
| 636 | {
|
| 637 | #ifdef CONFIG_FS_ENCRYPTION
|
| 638 | int len;
|
| 639 | char *name;
|
| 640 | struct fscrypt_str fname_crypto_str =
|
| 641 | FSTR_INIT(NULL, 0);
|
| 642 | int res = 0;
|
| 643 |
|
| 644 | name = de->name;
|
| 645 | len = de->name_len;
|
| 646 | if (!IS_ENCRYPTED(dir)) {
|
| 647 |
|
| 648 | (void) ext4fs_dirhash(dir, de->name,
|
| 649 | de->name_len, &h);
|
| 650 | printk("%.*s:(U)%x.%u ", len,
|
| 651 | name, h.hash,
|
| 652 | (unsigned) ((char *) de
|
| 653 | - base));
|
| 654 | } else {
|
| 655 | struct fscrypt_str de_name =
|
| 656 | FSTR_INIT(name, len);
|
| 657 |
|
| 658 |
|
| 659 | res = fscrypt_fname_alloc_buffer(
|
| 660 | len, &fname_crypto_str);
|
| 661 | if (res)
|
| 662 | printk(KERN_WARNING "Error "
|
| 663 | "allocating crypto "
|
| 664 | "buffer--skipping "
|
| 665 | "crypto\n");
|
| 666 | res = fscrypt_fname_disk_to_usr(dir,
|
| 667 | 0, 0, &de_name,
|
| 668 | &fname_crypto_str);
|
| 669 | if (res) {
|
| 670 | printk(KERN_WARNING "Error "
|
| 671 | "converting filename "
|
| 672 | "from disk to usr"
|
| 673 | "\n");
|
| 674 | name = "??";
|
| 675 | len = 2;
|
| 676 | } else {
|
| 677 | name = fname_crypto_str.name;
|
| 678 | len = fname_crypto_str.len;
|
| 679 | }
|
| 680 | if (IS_CASEFOLDED(dir))
|
| 681 | h.hash = EXT4_DIRENT_HASH(de);
|
| 682 | else
|
| 683 | (void) ext4fs_dirhash(dir,
|
| 684 | de->name,
|
| 685 | de->name_len, &h);
|
| 686 | printk("%.*s:(E)%x.%u ", len, name,
|
| 687 | h.hash, (unsigned) ((char *) de
|
| 688 | - base));
|
| 689 | fscrypt_fname_free_buffer(
|
| 690 | &fname_crypto_str);
|
| 691 | }
|
| 692 | #else
|
| 693 | int len = de->name_len;
|
| 694 | char *name = de->name;
|
| 695 | (void) ext4fs_dirhash(dir, de->name,
|
| 696 | de->name_len, &h);
|
| 697 | printk("%.*s:%x.%u ", len, name, h.hash,
|
| 698 | (unsigned) ((char *) de - base));
|
| 699 | #endif
|
| 700 | }
|
| 701 | space += ext4_dir_rec_len(de->name_len, dir);
|
| 702 | names++;
|
| 703 | }
|
| 704 | de = ext4_next_entry(de, size);
|
| 705 | }
|
| 706 | printk(KERN_CONT "(%i)\n", names);
|
| 707 | return (struct stats) { names, space, 1 };
|
| 708 | }
|
| 709 |
|
| 710 | struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
|
| 711 | struct dx_entry *entries, int levels)
|
| 712 | {
|
| 713 | unsigned blocksize = dir->i_sb->s_blocksize;
|
| 714 | unsigned count = dx_get_count(entries), names = 0, space = 0, i;
|
| 715 | unsigned bcount = 0;
|
| 716 | struct buffer_head *bh;
|
| 717 | printk("%i indexed blocks...\n", count);
|
| 718 | for (i = 0; i < count; i++, entries++)
|
| 719 | {
|
| 720 | ext4_lblk_t block = dx_get_block(entries);
|
| 721 | ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
|
| 722 | u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
|
| 723 | struct stats stats;
|
| 724 | printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
|
| 725 | bh = ext4_bread(NULL,dir, block, 0);
|
| 726 | if (IS_ERR_OR_NULL(bh))
|
| 727 | continue;
|
| 728 | stats = levels?
|
| 729 | dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
|
| 730 | dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
|
| 731 | bh->b_data, blocksize, 0);
|
| 732 | names += stats.names;
|
| 733 | space += stats.space;
|
| 734 | bcount += stats.bcount;
|
| 735 | brelse(bh);
|
| 736 | }
|
| 737 | if (bcount)
|
| 738 | printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
|
| 739 | levels ? "" : " ", names, space/bcount,
|
| 740 | (space/bcount)*100/blocksize);
|
| 741 | return (struct stats) { names, space, bcount};
|
| 742 | }
|
| 743 |
|
| 744 | |
| 745 | * Linear search cross check
|
| 746 |
|
| 747 | static inline void htree_rep_invariant_check(struct dx_entry *at,
|
| 748 | struct dx_entry *target,
|
| 749 | u32 hash, unsigned int n)
|
| 750 | {
|
| 751 | while (n--) {
|
| 752 | dxtrace(printk(KERN_CONT ","));
|
| 753 | if (dx_get_hash(++at) > hash) {
|
| 754 | at--;
|
| 755 | break;
|
| 756 | }
|
| 757 | }
|
| 758 | ASSERT(at == target - 1);
|
| 759 | }
|
| 760 | #else
|
| 761 | static inline void htree_rep_invariant_check(struct dx_entry *at,
|
| 762 | struct dx_entry *target,
|
| 763 | u32 hash, unsigned int n)
|
| 764 | {
|
| 765 | }
|
| 766 | #endif
|
| 767 |
|
| 768 | |
| 769 | * Probe for a directory leaf block to search.
|
| 770 | *
|
| 771 | * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
|
| 772 | * error in the directory index, and the caller should fall back to
|
| 773 | * searching the directory normally. The callers of dx_probe **MUST**
|
| 774 | * check for this error code, and make sure it never gets reflected
|
| 775 | * back to userspace.
|
| 776 |
|
| 777 | static struct dx_frame *
|
| 778 | dx_probe(struct ext4_filename *fname, struct inode *dir,
|
| 779 | struct dx_hash_info *hinfo, struct dx_frame *frame_in)
|
| 780 | {
|
| 781 | unsigned count, indirect, level, i;
|
| 782 | struct dx_entry *at, *entries, *p, *q, *m;
|
| 783 | struct dx_root *root;
|
| 784 | struct dx_frame *frame = frame_in;
|
| 785 | struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
|
| 786 | u32 hash;
|
| 787 | ext4_lblk_t block;
|
| 788 | ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
|
| 789 |
|
| 790 | memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
|
| 791 | frame->bh = ext4_read_dirblock(dir, 0, INDEX);
|
| 792 | if (IS_ERR(frame->bh))
|
| 793 | return (struct dx_frame *) frame->bh;
|
| 794 |
|
| 795 | root = (struct dx_root *) frame->bh->b_data;
|
| 796 | if (root->info.hash_version != DX_HASH_TEA &&
|
| 797 | root->info.hash_version != DX_HASH_HALF_MD4 &&
|
| 798 | root->info.hash_version != DX_HASH_LEGACY &&
|
| 799 | root->info.hash_version != DX_HASH_SIPHASH) {
|
| 800 | ext4_warning_inode(dir, "Unrecognised inode hash code %u",
|
| 801 | root->info.hash_version);
|
| 802 | goto fail;
|
| 803 | }
|
| 804 | if (ext4_hash_in_dirent(dir)) {
|
| 805 | if (root->info.hash_version != DX_HASH_SIPHASH) {
|
| 806 | ext4_warning_inode(dir,
|
| 807 | "Hash in dirent, but hash is not SIPHASH");
|
| 808 | goto fail;
|
| 809 | }
|
| 810 | } else {
|
| 811 | if (root->info.hash_version == DX_HASH_SIPHASH) {
|
| 812 | ext4_warning_inode(dir,
|
| 813 | "Hash code is SIPHASH, but hash not in dirent");
|
| 814 | goto fail;
|
| 815 | }
|
| 816 | }
|
| 817 | if (fname)
|
| 818 | hinfo = &fname->hinfo;
|
| 819 | hinfo->hash_version = root->info.hash_version;
|
| 820 | if (hinfo->hash_version <= DX_HASH_TEA)
|
| 821 | hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
|
| 822 | hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
|
| 823 |
|
| 824 | if (fname && fname_name(fname) &&
|
| 825 | !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
|
| 826 | int ret = ext4fs_dirhash(dir, fname_name(fname),
|
| 827 | fname_len(fname), hinfo);
|
| 828 | if (ret < 0) {
|
| 829 | ret_err = ERR_PTR(ret);
|
| 830 | goto fail;
|
| 831 | }
|
| 832 | }
|
| 833 | hash = hinfo->hash;
|
| 834 |
|
| 835 | if (root->info.unused_flags & 1) {
|
| 836 | ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
|
| 837 | root->info.unused_flags);
|
| 838 | goto fail;
|
| 839 | }
|
| 840 |
|
| 841 | indirect = root->info.indirect_levels;
|
| 842 | if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
|
| 843 | ext4_warning(dir->i_sb,
|
| 844 | "Directory (ino: %llu) htree depth %#06x exceed"
|
| 845 | "supported value", dir->i_ino,
|
| 846 | ext4_dir_htree_level(dir->i_sb));
|
| 847 | if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
|
| 848 | ext4_warning(dir->i_sb, "Enable large directory "
|
| 849 | "feature to access it");
|
| 850 | }
|
| 851 | goto fail;
|
| 852 | }
|
| 853 |
|
| 854 | entries = (struct dx_entry *)(((char *)&root->info) +
|
| 855 | root->info.info_length);
|
| 856 |
|
| 857 | if (dx_get_limit(entries) != dx_root_limit(dir,
|
| 858 | root->info.info_length)) {
|
| 859 | ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
|
| 860 | dx_get_limit(entries),
|
| 861 | dx_root_limit(dir, root->info.info_length));
|
| 862 | goto fail;
|
| 863 | }
|
| 864 |
|
| 865 | dxtrace(printk("Look up %x", hash));
|
| 866 | level = 0;
|
| 867 | blocks[0] = 0;
|
| 868 | while (1) {
|
| 869 | count = dx_get_count(entries);
|
| 870 | if (!count || count > dx_get_limit(entries)) {
|
| 871 | ext4_warning_inode(dir,
|
| 872 | "dx entry: count %u beyond limit %u",
|
| 873 | count, dx_get_limit(entries));
|
| 874 | goto fail;
|
| 875 | }
|
| 876 |
|
| 877 | p = entries + 1;
|
| 878 | q = entries + count - 1;
|
| 879 | while (p <= q) {
|
| 880 | m = p + (q - p) / 2;
|
| 881 | dxtrace(printk(KERN_CONT "."));
|
| 882 | if (dx_get_hash(m) > hash)
|
| 883 | q = m - 1;
|
| 884 | else
|
| 885 | p = m + 1;
|
| 886 | }
|
| 887 |
|
| 888 | htree_rep_invariant_check(entries, p, hash, count - 1);
|
| 889 |
|
| 890 | at = p - 1;
|
| 891 | dxtrace(printk(KERN_CONT " %x->%u\n",
|
| 892 | at == entries ? 0 : dx_get_hash(at),
|
| 893 | dx_get_block(at)));
|
| 894 | frame->entries = entries;
|
| 895 | frame->at = at;
|
| 896 |
|
| 897 | block = dx_get_block(at);
|
| 898 | for (i = 0; i <= level; i++) {
|
| 899 | if (blocks[i] == block) {
|
| 900 | ext4_warning_inode(dir,
|
| 901 | "dx entry: tree cycle block %u points back to block %u",
|
| 902 | blocks[level], block);
|
| 903 | goto fail;
|
| 904 | }
|
| 905 | }
|
| 906 | if (++level > indirect)
|
| 907 | return frame;
|
| 908 | blocks[level] = block;
|
| 909 | frame++;
|
| 910 | frame->bh = ext4_read_dirblock(dir, block, INDEX);
|
| 911 | if (IS_ERR(frame->bh)) {
|
| 912 | ret_err = (struct dx_frame *) frame->bh;
|
| 913 | frame->bh = NULL;
|
| 914 | goto fail;
|
| 915 | }
|
| 916 |
|
| 917 | entries = ((struct dx_node *) frame->bh->b_data)->entries;
|
| 918 |
|
| 919 | if (dx_get_limit(entries) != dx_node_limit(dir)) {
|
| 920 | ext4_warning_inode(dir,
|
| 921 | "dx entry: limit %u != node limit %u",
|
| 922 | dx_get_limit(entries), dx_node_limit(dir));
|
| 923 | goto fail;
|
| 924 | }
|
| 925 | }
|
| 926 | fail:
|
| 927 | while (frame >= frame_in) {
|
| 928 | brelse(frame->bh);
|
| 929 | frame--;
|
| 930 | }
|
| 931 |
|
| 932 | if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
|
| 933 | ext4_warning_inode(dir,
|
| 934 | "Corrupt directory, running e2fsck is recommended");
|
| 935 | return ret_err;
|
| 936 | }
|
| 937 |
|
| 938 | static void dx_release(struct dx_frame *frames)
|
| 939 | {
|
| 940 | struct dx_root_info *info;
|
| 941 | int i;
|
| 942 | unsigned int indirect_levels;
|
| 943 |
|
| 944 | if (frames[0].bh == NULL)
|
| 945 | return;
|
| 946 |
|
| 947 | info = &((struct dx_root *)frames[0].bh->b_data)->info;
|
| 948 |
|
| 949 | indirect_levels = info->indirect_levels;
|
| 950 | for (i = 0; i <= indirect_levels; i++) {
|
| 951 | if (frames[i].bh == NULL)
|
| 952 | break;
|
| 953 | brelse(frames[i].bh);
|
| 954 | frames[i].bh = NULL;
|
| 955 | }
|
| 956 | }
|
| 957 |
|
| 958 | |
| 959 | * This function increments the frame pointer to search the next leaf
|
| 960 | * block, and reads in the necessary intervening nodes if the search
|
| 961 | * should be necessary. Whether or not the search is necessary is
|
| 962 | * controlled by the hash parameter. If the hash value is even, then
|
| 963 | * the search is only continued if the next block starts with that
|
| 964 | * hash value. This is used if we are searching for a specific file.
|
| 965 | *
|
| 966 | * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
|
| 967 | *
|
| 968 | * This function returns 1 if the caller should continue to search,
|
| 969 | * or 0 if it should not. If there is an error reading one of the
|
| 970 | * index blocks, it will a negative error code.
|
| 971 | *
|
| 972 | * If start_hash is non-null, it will be filled in with the starting
|
| 973 | * hash of the next page.
|
| 974 |
|
| 975 | static int ext4_htree_next_block(struct inode *dir, __u32 hash,
|
| 976 | struct dx_frame *frame,
|
| 977 | struct dx_frame *frames,
|
| 978 | __u32 *start_hash)
|
| 979 | {
|
| 980 | struct dx_frame *p;
|
| 981 | struct buffer_head *bh;
|
| 982 | int num_frames = 0;
|
| 983 | __u32 bhash;
|
| 984 |
|
| 985 | p = frame;
|
| 986 | |
| 987 | * Find the next leaf page by incrementing the frame pointer.
|
| 988 | * If we run out of entries in the interior node, loop around and
|
| 989 | * increment pointer in the parent node. When we break out of
|
| 990 | * this loop, num_frames indicates the number of interior
|
| 991 | * nodes need to be read.
|
| 992 |
|
| 993 | while (1) {
|
| 994 | if (++(p->at) < p->entries + dx_get_count(p->entries))
|
| 995 | break;
|
| 996 | if (p == frames)
|
| 997 | return 0;
|
| 998 | num_frames++;
|
| 999 | p--;
|
| 1000 | }
|
| 1001 |
|
| 1002 | |
| 1003 | * If the hash is 1, then continue only if the next page has a
|
| 1004 | * continuation hash of any value. This is used for readdir
|
| 1005 | * handling. Otherwise, check to see if the hash matches the
|
| 1006 | * desired continuation hash. If it doesn't, return since
|
| 1007 | * there's no point to read in the successive index pages.
|
| 1008 |
|
| 1009 | bhash = dx_get_hash(p->at);
|
| 1010 | if (start_hash)
|
| 1011 | *start_hash = bhash;
|
| 1012 | if ((hash & 1) == 0) {
|
| 1013 | if ((bhash & ~1) != hash)
|
| 1014 | return 0;
|
| 1015 | }
|
| 1016 | |
| 1017 | * If the hash is HASH_NB_ALWAYS, we always go to the next
|
| 1018 | * block so no check is necessary
|
| 1019 |
|
| 1020 | while (num_frames--) {
|
| 1021 | bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
|
| 1022 | if (IS_ERR(bh))
|
| 1023 | return PTR_ERR(bh);
|
| 1024 | p++;
|
| 1025 | brelse(p->bh);
|
| 1026 | p->bh = bh;
|
| 1027 | p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
|
| 1028 | }
|
| 1029 | return 1;
|
| 1030 | }
|
| 1031 |
|
| 1032 |
|
| 1033 | |
| 1034 | * This function fills a red-black tree with information from a
|
| 1035 | * directory block. It returns the number directory entries loaded
|
| 1036 | * into the tree. If there is an error it is returned in err.
|
| 1037 |
|
| 1038 | static int htree_dirblock_to_tree(struct file *dir_file,
|
| 1039 | struct inode *dir, ext4_lblk_t block,
|
| 1040 | struct dx_hash_info *hinfo,
|
| 1041 | __u32 start_hash, __u32 start_minor_hash)
|
| 1042 | {
|
| 1043 | struct buffer_head *bh;
|
| 1044 | struct ext4_dir_entry_2 *de, *top;
|
| 1045 | int err = 0, count = 0;
|
| 1046 | struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
|
| 1047 | int csum = ext4_has_feature_metadata_csum(dir->i_sb);
|
| 1048 |
|
| 1049 | dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
|
| 1050 | (unsigned long)block));
|
| 1051 | bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
|
| 1052 | if (IS_ERR(bh))
|
| 1053 | return PTR_ERR(bh);
|
| 1054 |
|
| 1055 | de = (struct ext4_dir_entry_2 *) bh->b_data;
|
| 1056 |
|
| 1057 | top = (struct ext4_dir_entry_2 *) ((char *) de +
|
| 1058 | dir->i_sb->s_blocksize -
|
| 1059 | ext4_dir_rec_len(0,
|
| 1060 | csum ? NULL : dir));
|
| 1061 |
|
| 1062 | if (IS_ENCRYPTED(dir)) {
|
| 1063 | err = fscrypt_prepare_readdir(dir);
|
| 1064 | if (err < 0) {
|
| 1065 | brelse(bh);
|
| 1066 | return err;
|
| 1067 | }
|
| 1068 | err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
|
| 1069 | &fname_crypto_str);
|
| 1070 | if (err < 0) {
|
| 1071 | brelse(bh);
|
| 1072 | return err;
|
| 1073 | }
|
| 1074 | }
|
| 1075 |
|
| 1076 | for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
|
| 1077 | if (ext4_check_dir_entry(dir, NULL, de, bh,
|
| 1078 | bh->b_data, bh->b_size,
|
| 1079 | EXT4_LBLK_TO_B(dir, block)
|
| 1080 | + ((char *)de - bh->b_data))) {
|
| 1081 |
|
| 1082 | break;
|
| 1083 | }
|
| 1084 | if (ext4_hash_in_dirent(dir)) {
|
| 1085 | if (de->name_len && de->inode) {
|
| 1086 | hinfo->hash = EXT4_DIRENT_HASH(de);
|
| 1087 | hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
|
| 1088 | } else {
|
| 1089 | hinfo->hash = 0;
|
| 1090 | hinfo->minor_hash = 0;
|
| 1091 | }
|
| 1092 | } else {
|
| 1093 | err = ext4fs_dirhash(dir, de->name,
|
| 1094 | de->name_len, hinfo);
|
| 1095 | if (err < 0) {
|
| 1096 | count = err;
|
| 1097 | goto errout;
|
| 1098 | }
|
| 1099 | }
|
| 1100 | if ((hinfo->hash < start_hash) ||
|
| 1101 | ((hinfo->hash == start_hash) &&
|
| 1102 | (hinfo->minor_hash < start_minor_hash)))
|
| 1103 | continue;
|
| 1104 | if (de->inode == 0)
|
| 1105 | continue;
|
| 1106 | if (!IS_ENCRYPTED(dir)) {
|
| 1107 | tmp_str.name = de->name;
|
| 1108 | tmp_str.len = de->name_len;
|
| 1109 | err = ext4_htree_store_dirent(dir_file,
|
| 1110 | hinfo->hash, hinfo->minor_hash, de,
|
| 1111 | &tmp_str);
|
| 1112 | } else {
|
| 1113 | int save_len = fname_crypto_str.len;
|
| 1114 | struct fscrypt_str de_name = FSTR_INIT(de->name,
|
| 1115 | de->name_len);
|
| 1116 |
|
| 1117 |
|
| 1118 | err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
|
| 1119 | hinfo->minor_hash, &de_name,
|
| 1120 | &fname_crypto_str);
|
| 1121 | if (err) {
|
| 1122 | count = err;
|
| 1123 | goto errout;
|
| 1124 | }
|
| 1125 | err = ext4_htree_store_dirent(dir_file,
|
| 1126 | hinfo->hash, hinfo->minor_hash, de,
|
| 1127 | &fname_crypto_str);
|
| 1128 | fname_crypto_str.len = save_len;
|
| 1129 | }
|
| 1130 | if (err != 0) {
|
| 1131 | count = err;
|
| 1132 | goto errout;
|
| 1133 | }
|
| 1134 | count++;
|
| 1135 | }
|
| 1136 | errout:
|
| 1137 | brelse(bh);
|
| 1138 | fscrypt_fname_free_buffer(&fname_crypto_str);
|
| 1139 | return count;
|
| 1140 | }
|
| 1141 |
|
| 1142 |
|
| 1143 | |
| 1144 | * This function fills a red-black tree with information from a
|
| 1145 | * directory. We start scanning the directory in hash order, starting
|
| 1146 | * at start_hash and start_minor_hash.
|
| 1147 | *
|
| 1148 | * This function returns the number of entries inserted into the tree,
|
| 1149 | * or a negative error code.
|
| 1150 |
|
| 1151 | int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
|
| 1152 | __u32 start_minor_hash, __u32 *next_hash)
|
| 1153 | {
|
| 1154 | struct dx_hash_info hinfo;
|
| 1155 | struct ext4_dir_entry_2 *de;
|
| 1156 | struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
|
| 1157 | struct inode *dir;
|
| 1158 | ext4_lblk_t block;
|
| 1159 | int count = 0;
|
| 1160 | int ret, err;
|
| 1161 | __u32 hashval;
|
| 1162 | struct fscrypt_str tmp_str;
|
| 1163 |
|
| 1164 | dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
|
| 1165 | start_hash, start_minor_hash));
|
| 1166 | dir = file_inode(dir_file);
|
| 1167 | if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
|
| 1168 | if (ext4_hash_in_dirent(dir))
|
| 1169 | hinfo.hash_version = DX_HASH_SIPHASH;
|
| 1170 | else
|
| 1171 | hinfo.hash_version =
|
| 1172 | EXT4_SB(dir->i_sb)->s_def_hash_version;
|
| 1173 | if (hinfo.hash_version <= DX_HASH_TEA)
|
| 1174 | hinfo.hash_version +=
|
| 1175 | EXT4_SB(dir->i_sb)->s_hash_unsigned;
|
| 1176 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
|
| 1177 | if (ext4_has_inline_data(dir)) {
|
| 1178 | int has_inline_data = 1;
|
| 1179 | count = ext4_inlinedir_to_tree(dir_file, dir, 0,
|
| 1180 | &hinfo, start_hash,
|
| 1181 | start_minor_hash,
|
| 1182 | &has_inline_data);
|
| 1183 | if (has_inline_data) {
|
| 1184 | *next_hash = ~0;
|
| 1185 | return count;
|
| 1186 | }
|
| 1187 | }
|
| 1188 | count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
|
| 1189 | start_hash, start_minor_hash);
|
| 1190 | *next_hash = ~0;
|
| 1191 | return count;
|
| 1192 | }
|
| 1193 | hinfo.hash = start_hash;
|
| 1194 | hinfo.minor_hash = 0;
|
| 1195 | frame = dx_probe(NULL, dir, &hinfo, frames);
|
| 1196 | if (IS_ERR(frame))
|
| 1197 | return PTR_ERR(frame);
|
| 1198 |
|
| 1199 |
|
| 1200 | if (!start_hash && !start_minor_hash) {
|
| 1201 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
|
| 1202 | tmp_str.name = de->name;
|
| 1203 | tmp_str.len = de->name_len;
|
| 1204 | err = ext4_htree_store_dirent(dir_file, 0, 0,
|
| 1205 | de, &tmp_str);
|
| 1206 | if (err != 0)
|
| 1207 | goto errout;
|
| 1208 | count++;
|
| 1209 | }
|
| 1210 | if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
|
| 1211 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
|
| 1212 | de = ext4_next_entry(de, dir->i_sb->s_blocksize);
|
| 1213 | tmp_str.name = de->name;
|
| 1214 | tmp_str.len = de->name_len;
|
| 1215 | err = ext4_htree_store_dirent(dir_file, 2, 0,
|
| 1216 | de, &tmp_str);
|
| 1217 | if (err != 0)
|
| 1218 | goto errout;
|
| 1219 | count++;
|
| 1220 | }
|
| 1221 |
|
| 1222 | while (1) {
|
| 1223 | if (fatal_signal_pending(current)) {
|
| 1224 | err = -ERESTARTSYS;
|
| 1225 | goto errout;
|
| 1226 | }
|
| 1227 | cond_resched();
|
| 1228 | block = dx_get_block(frame->at);
|
| 1229 | ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
|
| 1230 | start_hash, start_minor_hash);
|
| 1231 | if (ret < 0) {
|
| 1232 | err = ret;
|
| 1233 | goto errout;
|
| 1234 | }
|
| 1235 | count += ret;
|
| 1236 | hashval = ~0;
|
| 1237 | ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
|
| 1238 | frame, frames, &hashval);
|
| 1239 | *next_hash = hashval;
|
| 1240 | if (ret < 0) {
|
| 1241 | err = ret;
|
| 1242 | goto errout;
|
| 1243 | }
|
| 1244 | |
| 1245 | * Stop if: (a) there are no more entries, or
|
| 1246 | * (b) we have inserted at least one entry and the
|
| 1247 | * next hash value is not a continuation
|
| 1248 |
|
| 1249 | if ((ret == 0) ||
|
| 1250 | (count && ((hashval & 1) == 0)))
|
| 1251 | break;
|
| 1252 | }
|
| 1253 | dx_release(frames);
|
| 1254 | dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
|
| 1255 | "next hash: %x\n", count, *next_hash));
|
| 1256 | return count;
|
| 1257 | errout:
|
| 1258 | dx_release(frames);
|
| 1259 | return (err);
|
| 1260 | }
|
| 1261 |
|
| 1262 | static inline int search_dirblock(struct buffer_head *bh,
|
| 1263 | struct inode *dir,
|
| 1264 | struct ext4_filename *fname,
|
| 1265 | unsigned int offset,
|
| 1266 | struct ext4_dir_entry_2 **res_dir)
|
| 1267 | {
|
| 1268 | return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
|
| 1269 | fname, offset, res_dir);
|
| 1270 | }
|
| 1271 |
|
| 1272 | |
| 1273 | * Directory block splitting, compacting
|
| 1274 |
|
| 1275 |
|
| 1276 | |
| 1277 | * Create map of hash values, offsets, and sizes, stored at end of block.
|
| 1278 | * Returns number of entries mapped.
|
| 1279 |
|
| 1280 | static int dx_make_map(struct inode *dir, struct buffer_head *bh,
|
| 1281 | struct dx_hash_info *hinfo,
|
| 1282 | struct dx_map_entry *map_tail)
|
| 1283 | {
|
| 1284 | int count = 0;
|
| 1285 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
|
| 1286 | unsigned int buflen = bh->b_size;
|
| 1287 | char *base = bh->b_data;
|
| 1288 | struct dx_hash_info h = *hinfo;
|
| 1289 | int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
|
| 1290 |
|
| 1291 | if (ext4_has_feature_metadata_csum(dir->i_sb))
|
| 1292 | buflen -= sizeof(struct ext4_dir_entry_tail);
|
| 1293 |
|
| 1294 | while ((char *) de < base + buflen) {
|
| 1295 | if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
|
| 1296 | ((char *)de) - base))
|
| 1297 | return -EFSCORRUPTED;
|
| 1298 | if (de->name_len && de->inode) {
|
| 1299 | if (ext4_hash_in_dirent(dir))
|
| 1300 | h.hash = EXT4_DIRENT_HASH(de);
|
| 1301 | else {
|
| 1302 | int err = ext4fs_dirhash(dir, de->name,
|
| 1303 | de->name_len, &h);
|
| 1304 | if (err < 0)
|
| 1305 | return err;
|
| 1306 | }
|
| 1307 | map_tail--;
|
| 1308 | map_tail->hash = h.hash;
|
| 1309 | map_tail->offs = ((char *) de - base)>>2;
|
| 1310 | map_tail->size = ext4_rec_len_from_disk(de->rec_len,
|
| 1311 | blocksize);
|
| 1312 | count++;
|
| 1313 | cond_resched();
|
| 1314 | }
|
| 1315 | de = ext4_next_entry(de, blocksize);
|
| 1316 | }
|
| 1317 | return count;
|
| 1318 | }
|
| 1319 |
|
| 1320 |
|
| 1321 | static void dx_sort_map (struct dx_map_entry *map, unsigned count)
|
| 1322 | {
|
| 1323 | struct dx_map_entry *p, *q, *top = map + count - 1;
|
| 1324 | int more;
|
| 1325 |
|
| 1326 | while (count > 2) {
|
| 1327 | count = count*10/13;
|
| 1328 | if (count - 9 < 2)
|
| 1329 | count = 11;
|
| 1330 | for (p = top, q = p - count; q >= map; p--, q--)
|
| 1331 | if (p->hash < q->hash)
|
| 1332 | swap(*p, *q);
|
| 1333 | }
|
| 1334 |
|
| 1335 | do {
|
| 1336 | more = 0;
|
| 1337 | q = top;
|
| 1338 | while (q-- > map) {
|
| 1339 | if (q[1].hash >= q[0].hash)
|
| 1340 | continue;
|
| 1341 | swap(*(q+1), *q);
|
| 1342 | more = 1;
|
| 1343 | }
|
| 1344 | } while(more);
|
| 1345 | }
|
| 1346 |
|
| 1347 | static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
|
| 1348 | {
|
| 1349 | struct dx_entry *entries = frame->entries;
|
| 1350 | struct dx_entry *old = frame->at, *new = old + 1;
|
| 1351 | int count = dx_get_count(entries);
|
| 1352 |
|
| 1353 | ASSERT(count < dx_get_limit(entries));
|
| 1354 | ASSERT(old < entries + count);
|
| 1355 | memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
|
| 1356 | dx_set_hash(new, hash);
|
| 1357 | dx_set_block(new, block);
|
| 1358 | dx_set_count(entries, count + 1);
|
| 1359 | }
|
| 1360 |
|
| 1361 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 1362 | int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
|
| 1363 | struct ext4_filename *name)
|
| 1364 | {
|
| 1365 | struct qstr *cf_name = &name->cf_name;
|
| 1366 | unsigned char *buf;
|
| 1367 | struct dx_hash_info *hinfo = &name->hinfo;
|
| 1368 | int len;
|
| 1369 |
|
| 1370 | if (!IS_CASEFOLDED(dir) ||
|
| 1371 | (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
|
| 1372 | cf_name->name = NULL;
|
| 1373 | return 0;
|
| 1374 | }
|
| 1375 |
|
| 1376 | buf = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
|
| 1377 | if (!buf)
|
| 1378 | return -ENOMEM;
|
| 1379 |
|
| 1380 | len = utf8_casefold(dir->i_sb->s_encoding, iname, buf, EXT4_NAME_LEN);
|
| 1381 | if (len <= 0) {
|
| 1382 | kfree(buf);
|
| 1383 | buf = NULL;
|
| 1384 | }
|
| 1385 | cf_name->name = buf;
|
| 1386 | cf_name->len = (unsigned) len;
|
| 1387 |
|
| 1388 | if (!IS_ENCRYPTED(dir))
|
| 1389 | return 0;
|
| 1390 |
|
| 1391 | hinfo->hash_version = DX_HASH_SIPHASH;
|
| 1392 | hinfo->seed = NULL;
|
| 1393 | if (cf_name->name)
|
| 1394 | return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
|
| 1395 | else
|
| 1396 | return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
|
| 1397 | }
|
| 1398 | #endif
|
| 1399 |
|
| 1400 | |
| 1401 | * Test whether a directory entry matches the filename being searched for.
|
| 1402 | *
|
| 1403 | * Return: %true if the directory entry matches, otherwise %false.
|
| 1404 |
|
| 1405 | static bool ext4_match(struct inode *parent,
|
| 1406 | const struct ext4_filename *fname,
|
| 1407 | struct ext4_dir_entry_2 *de)
|
| 1408 | {
|
| 1409 | struct fscrypt_name f;
|
| 1410 |
|
| 1411 | if (!de->inode)
|
| 1412 | return false;
|
| 1413 |
|
| 1414 | f.usr_fname = fname->usr_fname;
|
| 1415 | f.disk_name = fname->disk_name;
|
| 1416 | #ifdef CONFIG_FS_ENCRYPTION
|
| 1417 | f.crypto_buf = fname->crypto_buf;
|
| 1418 | #endif
|
| 1419 |
|
| 1420 | #if IS_ENABLED(CONFIG_UNICODE)
|
| 1421 | if (IS_CASEFOLDED(parent) &&
|
| 1422 | (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
|
| 1423 | |
| 1424 | * Just checking IS_ENCRYPTED(parent) below is not
|
| 1425 | * sufficient to decide whether one can use the hash for
|
| 1426 | * skipping the string comparison, because the key might
|
| 1427 | * have been added right after
|
| 1428 | * ext4_fname_setup_ci_filename(). In this case, a hash
|
| 1429 | * mismatch will be a false negative. Therefore, make
|
| 1430 | * sure cf_name was properly initialized before
|
| 1431 | * considering the calculated hash.
|
| 1432 |
|
| 1433 | if (sb_no_casefold_compat_fallback(parent->i_sb) &&
|
| 1434 | IS_ENCRYPTED(parent) && fname->cf_name.name &&
|
| 1435 | (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
|
| 1436 | fname->hinfo.minor_hash != EXT4_DIRENT_MINOR_HASH(de)))
|
| 1437 | return false;
|
| 1438 | |
| 1439 | * Treat comparison errors as not a match. The
|
| 1440 | * only case where it happens is on a disk
|
| 1441 | * corruption or ENOMEM.
|
| 1442 |
|
| 1443 |
|
| 1444 | return generic_ci_match(parent, fname->usr_fname,
|
| 1445 | &fname->cf_name, de->name,
|
| 1446 | de->name_len) > 0;
|
| 1447 | }
|
| 1448 | #endif
|
| 1449 |
|
| 1450 | return fscrypt_match_name(&f, de->name, de->name_len);
|
| 1451 | }
|
| 1452 |
|
| 1453 | |
| 1454 | * Returns 0 if not found, -EFSCORRUPTED on failure, and 1 on success
|
| 1455 |
|
| 1456 | int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
|
| 1457 | struct inode *dir, struct ext4_filename *fname,
|
| 1458 | unsigned int offset, struct ext4_dir_entry_2 **res_dir)
|
| 1459 | {
|
| 1460 | struct ext4_dir_entry_2 * de;
|
| 1461 | char * dlimit;
|
| 1462 | int de_len;
|
| 1463 |
|
| 1464 | de = (struct ext4_dir_entry_2 *)search_buf;
|
| 1465 | dlimit = search_buf + buf_size;
|
| 1466 | while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
|
| 1467 |
|
| 1468 |
|
| 1469 | if (de->name + de->name_len <= dlimit &&
|
| 1470 | (!ext4_hash_in_dirent(dir) ||
|
| 1471 | (char *)de + ext4_dir_rec_len(de->name_len, dir) <= dlimit) &&
|
| 1472 | ext4_match(dir, fname, de)) {
|
| 1473 | |
| 1474 | * a full check
|
| 1475 | if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
|
| 1476 | buf_size, offset))
|
| 1477 | return -EFSCORRUPTED;
|
| 1478 | *res_dir = de;
|
| 1479 | return 1;
|
| 1480 | }
|
| 1481 |
|
| 1482 | de_len = ext4_rec_len_from_disk(de->rec_len,
|
| 1483 | dir->i_sb->s_blocksize);
|
| 1484 | if (de_len <= 0)
|
| 1485 | return -EFSCORRUPTED;
|
| 1486 | offset += de_len;
|
| 1487 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
|
| 1488 | }
|
| 1489 | return 0;
|
| 1490 | }
|
| 1491 |
|
| 1492 | static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
|
| 1493 | struct ext4_dir_entry *de)
|
| 1494 | {
|
| 1495 | struct super_block *sb = dir->i_sb;
|
| 1496 |
|
| 1497 | if (!is_dx(dir))
|
| 1498 | return 0;
|
| 1499 | if (block == 0)
|
| 1500 | return 1;
|
| 1501 | if (de->inode == 0 &&
|
| 1502 | ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
|
| 1503 | sb->s_blocksize)
|
| 1504 | return 1;
|
| 1505 | return 0;
|
| 1506 | }
|
| 1507 |
|
| 1508 | |
| 1509 | * __ext4_find_entry()
|
| 1510 | *
|
| 1511 | * finds an entry in the specified directory with the wanted name. It
|
| 1512 | * returns the cache buffer in which the entry was found, and the entry
|
| 1513 | * itself (as a parameter - res_dir). It does NOT read the inode of the
|
| 1514 | * entry - you'll have to do that yourself if you want to.
|
| 1515 | *
|
| 1516 | * The returned buffer_head has ->b_count elevated. The caller is expected
|
| 1517 | * to brelse() it when appropriate.
|
| 1518 |
|
| 1519 | static struct buffer_head *__ext4_find_entry(struct inode *dir,
|
| 1520 | struct ext4_filename *fname,
|
| 1521 | struct ext4_dir_entry_2 **res_dir,
|
| 1522 | int *inlined)
|
| 1523 | {
|
| 1524 | struct super_block *sb;
|
| 1525 | struct buffer_head *bh_use[NAMEI_RA_SIZE];
|
| 1526 | struct buffer_head *bh, *ret = NULL;
|
| 1527 | ext4_lblk_t start, block;
|
| 1528 | const u8 *name = fname->usr_fname->name;
|
| 1529 | size_t ra_max = 0; |
| 1530 | buffer, bh_use[]
|
| 1531 | size_t ra_ptr = 0; |
| 1532 | buffer
|
| 1533 | ext4_lblk_t nblocks;
|
| 1534 | int i, namelen, retval;
|
| 1535 |
|
| 1536 | *res_dir = NULL;
|
| 1537 | sb = dir->i_sb;
|
| 1538 | namelen = fname->usr_fname->len;
|
| 1539 | if (namelen > EXT4_NAME_LEN)
|
| 1540 | return NULL;
|
| 1541 |
|
| 1542 | if (ext4_has_inline_data(dir)) {
|
| 1543 | int has_inline_data = 1;
|
| 1544 | ret = ext4_find_inline_entry(dir, fname, res_dir,
|
| 1545 | &has_inline_data);
|
| 1546 | if (inlined)
|
| 1547 | *inlined = has_inline_data;
|
| 1548 | if (has_inline_data || IS_ERR(ret))
|
| 1549 | goto cleanup_and_exit;
|
| 1550 | }
|
| 1551 |
|
| 1552 | if ((namelen <= 2) && (name[0] == '.') &&
|
| 1553 | (name[1] == '.' || name[1] == '\0')) {
|
| 1554 | |
| 1555 | * "." or ".." will only be in the first block
|
| 1556 | * NFS may look up ".."; "." should be handled by the VFS
|
| 1557 |
|
| 1558 | block = start = 0;
|
| 1559 | nblocks = 1;
|
| 1560 | goto restart;
|
| 1561 | }
|
| 1562 | if (is_dx(dir)) {
|
| 1563 | ret = ext4_dx_find_entry(dir, fname, res_dir);
|
| 1564 | |
| 1565 | * On success, or if the error was file not found,
|
| 1566 | * return. Otherwise, fall back to doing a search the
|
| 1567 | * old fashioned way.
|
| 1568 |
|
| 1569 | if (IS_ERR(ret) && PTR_ERR(ret) == ERR_BAD_DX_DIR)
|
| 1570 | dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
|
| 1571 | "falling back\n"));
|
| 1572 | else if (!sb_no_casefold_compat_fallback(dir->i_sb) &&
|
| 1573 | *res_dir == NULL && IS_CASEFOLDED(dir))
|
| 1574 | dxtrace(printk(KERN_DEBUG "ext4_find_entry: casefold "
|
| 1575 | "failed, falling back\n"));
|
| 1576 | else
|
| 1577 | goto cleanup_and_exit;
|
| 1578 | ret = NULL;
|
| 1579 | }
|
| 1580 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
|
| 1581 | if (!nblocks) {
|
| 1582 | ret = NULL;
|
| 1583 | goto cleanup_and_exit;
|
| 1584 | }
|
| 1585 | start = EXT4_I(dir)->i_dir_start_lookup;
|
| 1586 | if (start >= nblocks)
|
| 1587 | start = 0;
|
| 1588 | block = start;
|
| 1589 | restart:
|
| 1590 | do {
|
| 1591 | |
| 1592 | * We deal with the read-ahead logic here.
|
| 1593 |
|
| 1594 | cond_resched();
|
| 1595 | if (ra_ptr >= ra_max) {
|
| 1596 |
|
| 1597 | ra_ptr = 0;
|
| 1598 | if (block < start)
|
| 1599 | ra_max = start - block;
|
| 1600 | else
|
| 1601 | ra_max = nblocks - block;
|
| 1602 | ra_max = min(ra_max, ARRAY_SIZE(bh_use));
|
| 1603 | retval = ext4_bread_batch(dir, block, ra_max,
|
| 1604 | false , bh_use);
|
| 1605 | if (retval) {
|
| 1606 | ret = ERR_PTR(retval);
|
| 1607 | ra_max = 0;
|
| 1608 | goto cleanup_and_exit;
|
| 1609 | }
|
| 1610 | }
|
| 1611 | if ((bh = bh_use[ra_ptr++]) == NULL)
|
| 1612 | goto next;
|
| 1613 | wait_on_buffer(bh);
|
| 1614 | if (!buffer_uptodate(bh)) {
|
| 1615 | EXT4_ERROR_INODE_ERR(dir, EIO,
|
| 1616 | "reading directory lblock %lu",
|
| 1617 | (unsigned long) block);
|
| 1618 | brelse(bh);
|
| 1619 | ret = ERR_PTR(-EIO);
|
| 1620 | goto cleanup_and_exit;
|
| 1621 | }
|
| 1622 | if (!buffer_verified(bh) &&
|
| 1623 | !is_dx_internal_node(dir, block,
|
| 1624 | (struct ext4_dir_entry *)bh->b_data) &&
|
| 1625 | !ext4_dirblock_csum_verify(dir, bh)) {
|
| 1626 | EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
|
| 1627 | "checksumming directory "
|
| 1628 | "block %lu", (unsigned long)block);
|
| 1629 | brelse(bh);
|
| 1630 | ret = ERR_PTR(-EFSBADCRC);
|
| 1631 | goto cleanup_and_exit;
|
| 1632 | }
|
| 1633 | set_buffer_verified(bh);
|
| 1634 | i = search_dirblock(bh, dir, fname,
|
| 1635 | EXT4_LBLK_TO_B(dir, block), res_dir);
|
| 1636 | if (i == 1) {
|
| 1637 | EXT4_I(dir)->i_dir_start_lookup = block;
|
| 1638 | ret = bh;
|
| 1639 | goto cleanup_and_exit;
|
| 1640 | } else {
|
| 1641 | brelse(bh);
|
| 1642 | if (i < 0) {
|
| 1643 | ret = ERR_PTR(i);
|
| 1644 | goto cleanup_and_exit;
|
| 1645 | }
|
| 1646 | }
|
| 1647 | next:
|
| 1648 | if (++block >= nblocks)
|
| 1649 | block = 0;
|
| 1650 | } while (block != start);
|
| 1651 |
|
| 1652 | |
| 1653 | * If the directory has grown while we were searching, then
|
| 1654 | * search the last part of the directory before giving up.
|
| 1655 |
|
| 1656 | block = nblocks;
|
| 1657 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
|
| 1658 | if (block < nblocks) {
|
| 1659 | start = 0;
|
| 1660 | goto restart;
|
| 1661 | }
|
| 1662 |
|
| 1663 | cleanup_and_exit:
|
| 1664 |
|
| 1665 | for (; ra_ptr < ra_max; ra_ptr++)
|
| 1666 | brelse(bh_use[ra_ptr]);
|
| 1667 | return ret;
|
| 1668 | }
|
| 1669 |
|
| 1670 | static struct buffer_head *ext4_find_entry(struct inode *dir,
|
| 1671 | const struct qstr *d_name,
|
| 1672 | struct ext4_dir_entry_2 **res_dir,
|
| 1673 | int *inlined)
|
| 1674 | {
|
| 1675 | int err;
|
| 1676 | struct ext4_filename fname;
|
| 1677 | struct buffer_head *bh;
|
| 1678 |
|
| 1679 | err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
|
| 1680 | if (err == -ENOENT)
|
| 1681 | return NULL;
|
| 1682 | if (err)
|
| 1683 | return ERR_PTR(err);
|
| 1684 |
|
| 1685 | bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
|
| 1686 |
|
| 1687 | ext4_fname_free_filename(&fname);
|
| 1688 | return bh;
|
| 1689 | }
|
| 1690 |
|
| 1691 | static struct buffer_head *ext4_lookup_entry(struct inode *dir,
|
| 1692 | struct dentry *dentry,
|
| 1693 | struct ext4_dir_entry_2 **res_dir)
|
| 1694 | {
|
| 1695 | int err;
|
| 1696 | struct ext4_filename fname;
|
| 1697 | struct buffer_head *bh;
|
| 1698 |
|
| 1699 | err = ext4_fname_prepare_lookup(dir, dentry, &fname);
|
| 1700 | if (err == -ENOENT)
|
| 1701 | return NULL;
|
| 1702 | if (err)
|
| 1703 | return ERR_PTR(err);
|
| 1704 |
|
| 1705 | bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
|
| 1706 |
|
| 1707 | ext4_fname_free_filename(&fname);
|
| 1708 | return bh;
|
| 1709 | }
|
| 1710 |
|
| 1711 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
|
| 1712 | struct ext4_filename *fname,
|
| 1713 | struct ext4_dir_entry_2 **res_dir)
|
| 1714 | {
|
| 1715 | struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
|
| 1716 | struct buffer_head *bh;
|
| 1717 | ext4_lblk_t block;
|
| 1718 | int retval;
|
| 1719 |
|
| 1720 | #ifdef CONFIG_FS_ENCRYPTION
|
| 1721 | *res_dir = NULL;
|
| 1722 | #endif
|
| 1723 | frame = dx_probe(fname, dir, NULL, frames);
|
| 1724 | if (IS_ERR(frame))
|
| 1725 | return ERR_CAST(frame);
|
| 1726 | do {
|
| 1727 | block = dx_get_block(frame->at);
|
| 1728 | bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
|
| 1729 | if (IS_ERR(bh))
|
| 1730 | goto errout;
|
| 1731 |
|
| 1732 | retval = search_dirblock(bh, dir, fname,
|
| 1733 | EXT4_LBLK_TO_B(dir, block), res_dir);
|
| 1734 | if (retval == 1)
|
| 1735 | goto success;
|
| 1736 | brelse(bh);
|
| 1737 | if (retval < 0) {
|
| 1738 | bh = ERR_PTR(ERR_BAD_DX_DIR);
|
| 1739 | goto errout;
|
| 1740 | }
|
| 1741 |
|
| 1742 |
|
| 1743 | retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
|
| 1744 | frames, NULL);
|
| 1745 | if (retval < 0) {
|
| 1746 | ext4_warning_inode(dir,
|
| 1747 | "error %d reading directory index block",
|
| 1748 | retval);
|
| 1749 | bh = ERR_PTR(retval);
|
| 1750 | goto errout;
|
| 1751 | }
|
| 1752 | } while (retval == 1);
|
| 1753 |
|
| 1754 | bh = NULL;
|
| 1755 | errout:
|
| 1756 | dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
|
| 1757 | success:
|
| 1758 | dx_release(frames);
|
| 1759 | return bh;
|
| 1760 | }
|
| 1761 |
|
| 1762 | static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
|
| 1763 | {
|
| 1764 | struct inode *inode;
|
| 1765 | struct ext4_dir_entry_2 *de = NULL;
|
| 1766 | struct buffer_head *bh;
|
| 1767 |
|
| 1768 | if (dentry->d_name.len > EXT4_NAME_LEN)
|
| 1769 | return ERR_PTR(-ENAMETOOLONG);
|
| 1770 |
|
| 1771 | bh = ext4_lookup_entry(dir, dentry, &de);
|
| 1772 | if (IS_ERR(bh))
|
| 1773 | return ERR_CAST(bh);
|
| 1774 | inode = NULL;
|
| 1775 | if (bh) {
|
| 1776 | __u32 ino = le32_to_cpu(de->inode);
|
| 1777 | brelse(bh);
|
| 1778 | if (!ext4_valid_inum(dir->i_sb, ino)) {
|
| 1779 | EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
|
| 1780 | return ERR_PTR(-EFSCORRUPTED);
|
| 1781 | }
|
| 1782 | if (unlikely(ino == dir->i_ino)) {
|
| 1783 | EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
|
| 1784 | dentry);
|
| 1785 | return ERR_PTR(-EFSCORRUPTED);
|
| 1786 | }
|
| 1787 | inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
|
| 1788 | if (inode == ERR_PTR(-ESTALE)) {
|
| 1789 | EXT4_ERROR_INODE(dir,
|
| 1790 | "deleted inode referenced: %u",
|
| 1791 | ino);
|
| 1792 | return ERR_PTR(-EFSCORRUPTED);
|
| 1793 | }
|
| 1794 | if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
|
| 1795 | (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
|
| 1796 | !fscrypt_has_permitted_context(dir, inode)) {
|
| 1797 | ext4_warning(inode->i_sb,
|
| 1798 | "Inconsistent encryption contexts: %llu/%llu",
|
| 1799 | dir->i_ino, inode->i_ino);
|
| 1800 | iput(inode);
|
| 1801 | return ERR_PTR(-EPERM);
|
| 1802 | }
|
| 1803 | }
|
| 1804 |
|
| 1805 | if (IS_ENABLED(CONFIG_UNICODE) && !inode && IS_CASEFOLDED(dir)) {
|
| 1806 | |
| 1807 | * for negative dentries in the encoding case as
|
| 1808 | * well. For now, prevent the negative dentry
|
| 1809 | * from being cached.
|
| 1810 |
|
| 1811 | return NULL;
|
| 1812 | }
|
| 1813 |
|
| 1814 | return d_splice_alias(inode, dentry);
|
| 1815 | }
|
| 1816 |
|
| 1817 |
|
| 1818 | struct dentry *ext4_get_parent(struct dentry *child)
|
| 1819 | {
|
| 1820 | __u32 ino;
|
| 1821 | struct ext4_dir_entry_2 * de = NULL;
|
| 1822 | struct buffer_head *bh;
|
| 1823 |
|
| 1824 | bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
|
| 1825 | if (IS_ERR(bh))
|
| 1826 | return ERR_CAST(bh);
|
| 1827 | if (!bh)
|
| 1828 | return ERR_PTR(-ENOENT);
|
| 1829 | ino = le32_to_cpu(de->inode);
|
| 1830 | brelse(bh);
|
| 1831 |
|
| 1832 | if (!ext4_valid_inum(child->d_sb, ino)) {
|
| 1833 | EXT4_ERROR_INODE(d_inode(child),
|
| 1834 | "bad parent inode number: %u", ino);
|
| 1835 | return ERR_PTR(-EFSCORRUPTED);
|
| 1836 | }
|
| 1837 |
|
| 1838 | return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
|
| 1839 | }
|
| 1840 |
|
| 1841 | |
| 1842 | * Move count entries from end of map between two memory locations.
|
| 1843 | * Returns pointer to last entry moved.
|
| 1844 |
|
| 1845 | static struct ext4_dir_entry_2 *
|
| 1846 | dx_move_dirents(struct inode *dir, char *from, char *to,
|
| 1847 | struct dx_map_entry *map, int count,
|
| 1848 | unsigned blocksize)
|
| 1849 | {
|
| 1850 | unsigned rec_len = 0;
|
| 1851 |
|
| 1852 | while (count--) {
|
| 1853 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
|
| 1854 | (from + (map->offs<<2));
|
| 1855 | rec_len = ext4_dir_rec_len(de->name_len, dir);
|
| 1856 |
|
| 1857 | memcpy (to, de, rec_len);
|
| 1858 | ((struct ext4_dir_entry_2 *) to)->rec_len =
|
| 1859 | ext4_rec_len_to_disk(rec_len, blocksize);
|
| 1860 |
|
| 1861 |
|
| 1862 | de->inode = 0;
|
| 1863 | memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
|
| 1864 | blocksize) -
|
| 1865 | offsetof(struct ext4_dir_entry_2,
|
| 1866 | name_len));
|
| 1867 |
|
| 1868 | map++;
|
| 1869 | to += rec_len;
|
| 1870 | }
|
| 1871 | return (struct ext4_dir_entry_2 *) (to - rec_len);
|
| 1872 | }
|
| 1873 |
|
| 1874 | |
| 1875 | * Compact each dir entry in the range to the minimal rec_len.
|
| 1876 | * Returns pointer to last entry in range.
|
| 1877 |
|
| 1878 | static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
|
| 1879 | unsigned int blocksize)
|
| 1880 | {
|
| 1881 | struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
|
| 1882 | unsigned rec_len = 0;
|
| 1883 |
|
| 1884 | prev = to = de;
|
| 1885 | while ((char*)de < base + blocksize) {
|
| 1886 | next = ext4_next_entry(de, blocksize);
|
| 1887 | if (de->inode && de->name_len) {
|
| 1888 | rec_len = ext4_dir_rec_len(de->name_len, dir);
|
| 1889 | if (de > to)
|
| 1890 | memmove(to, de, rec_len);
|
| 1891 | to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
|
| 1892 | prev = to;
|
| 1893 | to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
|
| 1894 | }
|
| 1895 | de = next;
|
| 1896 | }
|
| 1897 | return prev;
|
| 1898 | }
|
| 1899 |
|
| 1900 | |
| 1901 | * Split a full leaf block to make room for a new dir entry.
|
| 1902 | * Allocate a new block, and move entries so that they are approx. equally full.
|
| 1903 | * Returns pointer to de in block into which the new entry will be inserted.
|
| 1904 |
|
| 1905 | static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
|
| 1906 | struct buffer_head **bh,struct dx_frame *frame,
|
| 1907 | struct dx_hash_info *hinfo)
|
| 1908 | {
|
| 1909 | unsigned blocksize = dir->i_sb->s_blocksize;
|
| 1910 | unsigned continued;
|
| 1911 | int count;
|
| 1912 | struct buffer_head *bh2;
|
| 1913 | ext4_lblk_t newblock;
|
| 1914 | u32 hash2;
|
| 1915 | struct dx_map_entry *map;
|
| 1916 | char *data1 = (*bh)->b_data, *data2;
|
| 1917 | unsigned split, move, size;
|
| 1918 | struct ext4_dir_entry_2 *de = NULL, *de2;
|
| 1919 | int csum_size = 0;
|
| 1920 | int err = 0, i;
|
| 1921 |
|
| 1922 | if (ext4_has_feature_metadata_csum(dir->i_sb))
|
| 1923 | csum_size = sizeof(struct ext4_dir_entry_tail);
|
| 1924 |
|
| 1925 | bh2 = ext4_append(handle, dir, &newblock);
|
| 1926 | if (IS_ERR(bh2)) {
|
| 1927 | brelse(*bh);
|
| 1928 | *bh = NULL;
|
| 1929 | return ERR_CAST(bh2);
|
| 1930 | }
|
| 1931 |
|
| 1932 | BUFFER_TRACE(*bh, "get_write_access");
|
| 1933 | err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
|
| 1934 | EXT4_JTR_NONE);
|
| 1935 | if (err)
|
| 1936 | goto journal_error;
|
| 1937 |
|
| 1938 | BUFFER_TRACE(frame->bh, "get_write_access");
|
| 1939 | err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
|
| 1940 | EXT4_JTR_NONE);
|
| 1941 | if (err)
|
| 1942 | goto journal_error;
|
| 1943 |
|
| 1944 | data2 = bh2->b_data;
|
| 1945 |
|
| 1946 |
|
| 1947 | map = (struct dx_map_entry *) (data2 + blocksize);
|
| 1948 | count = dx_make_map(dir, *bh, hinfo, map);
|
| 1949 | if (count < 0) {
|
| 1950 | err = count;
|
| 1951 | goto journal_error;
|
| 1952 | }
|
| 1953 | map -= count;
|
| 1954 | dx_sort_map(map, count);
|
| 1955 |
|
| 1956 | size = 0;
|
| 1957 | move = 0;
|
| 1958 | for (i = count-1; i >= 0; i--) {
|
| 1959 |
|
| 1960 | if (size + map[i].size/2 > blocksize/2)
|
| 1961 | break;
|
| 1962 | size += map[i].size;
|
| 1963 | move++;
|
| 1964 | }
|
| 1965 | |
| 1966 | * map index at which we will split
|
| 1967 | *
|
| 1968 | * If the sum of active entries didn't exceed half the block size, just
|
| 1969 | * split it in half by count; each resulting block will have at least
|
| 1970 | * half the space free.
|
| 1971 |
|
| 1972 | if (i >= 0)
|
| 1973 | split = count - move;
|
| 1974 | else
|
| 1975 | split = count/2;
|
| 1976 |
|
| 1977 | if (WARN_ON_ONCE(split == 0)) {
|
| 1978 |
|
| 1979 | ext4_error_inode_block(dir, (*bh)->b_blocknr, 0,
|
| 1980 | "bad indexed directory? hash=%08x:%08x count=%d move=%u",
|
| 1981 | hinfo->hash, hinfo->minor_hash, count, move);
|
| 1982 | err = -EFSCORRUPTED;
|
| 1983 | goto out;
|
| 1984 | }
|
| 1985 |
|
| 1986 | hash2 = map[split].hash;
|
| 1987 | continued = hash2 == map[split - 1].hash;
|
| 1988 | dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
|
| 1989 | (unsigned long)dx_get_block(frame->at),
|
| 1990 | hash2, split, count-split));
|
| 1991 |
|
| 1992 |
|
| 1993 | de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
|
| 1994 | blocksize);
|
| 1995 | de = dx_pack_dirents(dir, data1, blocksize);
|
| 1996 | de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
|
| 1997 | (char *) de,
|
| 1998 | blocksize);
|
| 1999 | de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
|
| 2000 | (char *) de2,
|
| 2001 | blocksize);
|
| 2002 | if (csum_size) {
|
| 2003 | ext4_initialize_dirent_tail(*bh, blocksize);
|
| 2004 | ext4_initialize_dirent_tail(bh2, blocksize);
|
| 2005 | }
|
| 2006 |
|
| 2007 | dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
|
| 2008 | blocksize, 1));
|
| 2009 | dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
|
| 2010 | blocksize, 1));
|
| 2011 |
|
| 2012 |
|
| 2013 | if (hinfo->hash >= hash2) {
|
| 2014 | swap(*bh, bh2);
|
| 2015 | de = de2;
|
| 2016 | }
|
| 2017 | dx_insert_block(frame, hash2 + continued, newblock);
|
| 2018 | err = ext4_handle_dirty_dirblock(handle, dir, bh2);
|
| 2019 | if (err)
|
| 2020 | goto journal_error;
|
| 2021 | err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
|
| 2022 | if (err)
|
| 2023 | goto journal_error;
|
| 2024 | brelse(bh2);
|
| 2025 | dxtrace(dx_show_index("frame", frame->entries));
|
| 2026 | return de;
|
| 2027 |
|
| 2028 | journal_error:
|
| 2029 | ext4_std_error(dir->i_sb, err);
|
| 2030 | out:
|
| 2031 | brelse(*bh);
|
| 2032 | brelse(bh2);
|
| 2033 | *bh = NULL;
|
| 2034 | return ERR_PTR(err);
|
| 2035 | }
|
| 2036 |
|
| 2037 | int ext4_find_dest_de(struct inode *dir, struct buffer_head *bh,
|
| 2038 | void *buf, int buf_size,
|
| 2039 | struct ext4_filename *fname,
|
| 2040 | struct ext4_dir_entry_2 **dest_de)
|
| 2041 | {
|
| 2042 | struct ext4_dir_entry_2 *de;
|
| 2043 | unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
|
| 2044 | int nlen, rlen;
|
| 2045 | unsigned int offset = 0;
|
| 2046 | char *top;
|
| 2047 |
|
| 2048 | de = buf;
|
| 2049 | top = buf + buf_size - reclen;
|
| 2050 | while ((char *) de <= top) {
|
| 2051 | if (ext4_check_dir_entry(dir, NULL, de, bh,
|
| 2052 | buf, buf_size, offset))
|
| 2053 | return -EFSCORRUPTED;
|
| 2054 | if (ext4_match(dir, fname, de))
|
| 2055 | return -EEXIST;
|
| 2056 | nlen = ext4_dir_rec_len(de->name_len, dir);
|
| 2057 | rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
|
| 2058 | if ((de->inode ? rlen - nlen : rlen) >= reclen)
|
| 2059 | break;
|
| 2060 | de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
|
| 2061 | offset += rlen;
|
| 2062 | }
|
| 2063 | if ((char *) de > top)
|
| 2064 | return -ENOSPC;
|
| 2065 |
|
| 2066 | *dest_de = de;
|
| 2067 | return 0;
|
| 2068 | }
|
| 2069 |
|
| 2070 | void ext4_insert_dentry(struct inode *dir,
|
| 2071 | struct inode *inode,
|
| 2072 | struct ext4_dir_entry_2 *de,
|
| 2073 | int buf_size,
|
| 2074 | struct ext4_filename *fname)
|
| 2075 | {
|
| 2076 |
|
| 2077 | int nlen, rlen;
|
| 2078 |
|
| 2079 | nlen = ext4_dir_rec_len(de->name_len, dir);
|
| 2080 | rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
|
| 2081 | if (de->inode) {
|
| 2082 | struct ext4_dir_entry_2 *de1 =
|
| 2083 | (struct ext4_dir_entry_2 *)((char *)de + nlen);
|
| 2084 | de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
|
| 2085 | de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
|
| 2086 | de = de1;
|
| 2087 | }
|
| 2088 | de->file_type = EXT4_FT_UNKNOWN;
|
| 2089 | de->inode = cpu_to_le32(inode->i_ino);
|
| 2090 | ext4_set_de_type(inode->i_sb, de, inode->i_mode);
|
| 2091 | de->name_len = fname_len(fname);
|
| 2092 | memcpy(de->name, fname_name(fname), fname_len(fname));
|
| 2093 | if (ext4_hash_in_dirent(dir)) {
|
| 2094 | struct dx_hash_info *hinfo = &fname->hinfo;
|
| 2095 |
|
| 2096 | EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
|
| 2097 | EXT4_DIRENT_HASHES(de)->minor_hash =
|
| 2098 | cpu_to_le32(hinfo->minor_hash);
|
| 2099 | }
|
| 2100 | }
|
| 2101 |
|
| 2102 | |
| 2103 | * Add a new entry into a directory (leaf) block. If de is non-NULL,
|
| 2104 | * it points to a directory entry which is guaranteed to be large
|
| 2105 | * enough for new directory entry. If de is NULL, then
|
| 2106 | * add_dirent_to_buf will attempt search the directory block for
|
| 2107 | * space. It will return -ENOSPC if no space is available, and -EIO
|
| 2108 | * and -EEXIST if directory entry already exists.
|
| 2109 |
|
| 2110 | static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
|
| 2111 | struct inode *dir,
|
| 2112 | struct inode *inode, struct ext4_dir_entry_2 *de,
|
| 2113 | struct buffer_head *bh)
|
| 2114 | {
|
| 2115 | unsigned int blocksize = dir->i_sb->s_blocksize;
|
| 2116 | int csum_size = 0;
|
| 2117 | int err, err2;
|
| 2118 |
|
| 2119 | if (ext4_has_feature_metadata_csum(inode->i_sb))
|
| 2120 | csum_size = sizeof(struct ext4_dir_entry_tail);
|
| 2121 |
|
| 2122 | if (!de) {
|
| 2123 | err = ext4_find_dest_de(dir, bh, bh->b_data,
|
| 2124 | blocksize - csum_size, fname, &de);
|
| 2125 | if (err)
|
| 2126 | return err;
|
| 2127 | }
|
| 2128 | BUFFER_TRACE(bh, "get_write_access");
|
| 2129 | err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
|
| 2130 | EXT4_JTR_NONE);
|
| 2131 | if (err) {
|
| 2132 | ext4_std_error(dir->i_sb, err);
|
| 2133 | return err;
|
| 2134 | }
|
| 2135 |
|
| 2136 |
|
| 2137 | ext4_insert_dentry(dir, inode, de, blocksize, fname);
|
| 2138 |
|
| 2139 | |
| 2140 | * XXX shouldn't update any times until successful
|
| 2141 | * completion of syscall, but too many callers depend
|
| 2142 | * on this.
|
| 2143 | *
|
| 2144 | * XXX similarly, too many callers depend on
|
| 2145 | * ext4_new_inode() setting the times, but error
|
| 2146 | * recovery deletes the inode, so the worst that can
|
| 2147 | * happen is that the times are slightly out of date
|
| 2148 | * and/or different from the directory change time.
|
| 2149 |
|
| 2150 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
|
| 2151 | ext4_update_dx_flag(dir);
|
| 2152 | inode_inc_iversion(dir);
|
| 2153 | err2 = ext4_mark_inode_dirty(handle, dir);
|
| 2154 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
|
| 2155 | err = ext4_handle_dirty_dirblock(handle, dir, bh);
|
| 2156 | if (err)
|
| 2157 | ext4_std_error(dir->i_sb, err);
|
| 2158 | return err ? err : err2;
|
| 2159 | }
|
| 2160 |
|
| 2161 | static bool ext4_check_dx_root(struct inode *dir, struct dx_root *root)
|
| 2162 | {
|
| 2163 | struct fake_dirent *fde;
|
| 2164 | const char *error_msg;
|
| 2165 | unsigned int rlen;
|
| 2166 | unsigned int blocksize = dir->i_sb->s_blocksize;
|
| 2167 | char *blockend = (char *)root + dir->i_sb->s_blocksize;
|
| 2168 |
|
| 2169 | fde = &root->dot;
|
| 2170 | if (unlikely(fde->name_len != 1)) {
|
| 2171 | error_msg = "invalid name_len for '.'";
|
| 2172 | goto corrupted;
|
| 2173 | }
|
| 2174 | if (unlikely(strncmp(root->dot_name, ".", fde->name_len))) {
|
| 2175 | error_msg = "invalid name for '.'";
|
| 2176 | goto corrupted;
|
| 2177 | }
|
| 2178 | rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
|
| 2179 | if (unlikely((char *)fde + rlen >= blockend)) {
|
| 2180 | error_msg = "invalid rec_len for '.'";
|
| 2181 | goto corrupted;
|
| 2182 | }
|
| 2183 |
|
| 2184 | fde = &root->dotdot;
|
| 2185 | if (unlikely(fde->name_len != 2)) {
|
| 2186 | error_msg = "invalid name_len for '..'";
|
| 2187 | goto corrupted;
|
| 2188 | }
|
| 2189 | if (unlikely(strncmp(root->dotdot_name, "..", fde->name_len))) {
|
| 2190 | error_msg = "invalid name for '..'";
|
| 2191 | goto corrupted;
|
| 2192 | }
|
| 2193 | rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
|
| 2194 | if (unlikely((char *)fde + rlen >= blockend)) {
|
| 2195 | error_msg = "invalid rec_len for '..'";
|
| 2196 | goto corrupted;
|
| 2197 | }
|
| 2198 |
|
| 2199 | return true;
|
| 2200 |
|
| 2201 | corrupted:
|
| 2202 | EXT4_ERROR_INODE(dir, "Corrupt dir, %s, running e2fsck is recommended",
|
| 2203 | error_msg);
|
| 2204 | return false;
|
| 2205 | }
|
| 2206 |
|
| 2207 | |
| 2208 | * This converts a one block unindexed directory to a 3 block indexed
|
| 2209 | * directory, and adds the dentry to the indexed directory.
|
| 2210 |
|
| 2211 | static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
|
| 2212 | struct inode *dir,
|
| 2213 | struct inode *inode, struct buffer_head *bh)
|
| 2214 | {
|
| 2215 | struct buffer_head *bh2;
|
| 2216 | struct dx_root *root;
|
| 2217 | struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
|
| 2218 | struct dx_entry *entries;
|
| 2219 | struct ext4_dir_entry_2 *de, *de2;
|
| 2220 | char *data2, *top;
|
| 2221 | unsigned len;
|
| 2222 | int retval;
|
| 2223 | unsigned blocksize;
|
| 2224 | ext4_lblk_t block;
|
| 2225 | struct fake_dirent *fde;
|
| 2226 | int csum_size = 0;
|
| 2227 |
|
| 2228 | if (ext4_has_feature_metadata_csum(inode->i_sb))
|
| 2229 | csum_size = sizeof(struct ext4_dir_entry_tail);
|
| 2230 |
|
| 2231 | blocksize = dir->i_sb->s_blocksize;
|
| 2232 | dxtrace(printk(KERN_DEBUG "Creating index: inode %llu\n", dir->i_ino));
|
| 2233 | BUFFER_TRACE(bh, "get_write_access");
|
| 2234 | retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
|
| 2235 | EXT4_JTR_NONE);
|
| 2236 | if (retval) {
|
| 2237 | ext4_std_error(dir->i_sb, retval);
|
| 2238 | brelse(bh);
|
| 2239 | return retval;
|
| 2240 | }
|
| 2241 |
|
| 2242 | root = (struct dx_root *) bh->b_data;
|
| 2243 | if (!ext4_check_dx_root(dir, root)) {
|
| 2244 | brelse(bh);
|
| 2245 | return -EFSCORRUPTED;
|
| 2246 | }
|
| 2247 |
|
| 2248 |
|
| 2249 | fde = &root->dotdot;
|
| 2250 | de = (struct ext4_dir_entry_2 *)((char *)fde +
|
| 2251 | ext4_rec_len_from_disk(fde->rec_len, blocksize));
|
| 2252 | len = ((char *) root) + (blocksize - csum_size) - (char *) de;
|
| 2253 |
|
| 2254 |
|
| 2255 | bh2 = ext4_append(handle, dir, &block);
|
| 2256 | if (IS_ERR(bh2)) {
|
| 2257 | brelse(bh);
|
| 2258 | return PTR_ERR(bh2);
|
| 2259 | }
|
| 2260 | ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
|
| 2261 | data2 = bh2->b_data;
|
| 2262 |
|
| 2263 | memcpy(data2, de, len);
|
| 2264 | memset(de, 0, len);
|
| 2265 | de = (struct ext4_dir_entry_2 *) data2;
|
| 2266 | top = data2 + len;
|
| 2267 | while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
|
| 2268 | if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
|
| 2269 | (char *)de - data2)) {
|
| 2270 | brelse(bh2);
|
| 2271 | brelse(bh);
|
| 2272 | return -EFSCORRUPTED;
|
| 2273 | }
|
| 2274 | de = de2;
|
| 2275 | }
|
| 2276 | de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
|
| 2277 | (char *) de, blocksize);
|
| 2278 |
|
| 2279 | if (csum_size)
|
| 2280 | ext4_initialize_dirent_tail(bh2, blocksize);
|
| 2281 |
|
| 2282 |
|
| 2283 | de = (struct ext4_dir_entry_2 *) (&root->dotdot);
|
| 2284 | de->rec_len = ext4_rec_len_to_disk(
|
| 2285 | blocksize - ext4_dir_rec_len(2, NULL), blocksize);
|
| 2286 | memset (&root->info, 0, sizeof(root->info));
|
| 2287 | root->info.info_length = sizeof(root->info);
|
| 2288 | if (ext4_hash_in_dirent(dir))
|
| 2289 | root->info.hash_version = DX_HASH_SIPHASH;
|
| 2290 | else
|
| 2291 | root->info.hash_version =
|
| 2292 | EXT4_SB(dir->i_sb)->s_def_hash_version;
|
| 2293 |
|
| 2294 | entries = root->entries;
|
| 2295 | dx_set_block(entries, 1);
|
| 2296 | dx_set_count(entries, 1);
|
| 2297 | dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
|
| 2298 |
|
| 2299 |
|
| 2300 | fname->hinfo.hash_version = root->info.hash_version;
|
| 2301 | if (fname->hinfo.hash_version <= DX_HASH_TEA)
|
| 2302 | fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
|
| 2303 | fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
|
| 2304 |
|
| 2305 |
|
| 2306 | if (!ext4_hash_in_dirent(dir)) {
|
| 2307 | int err = ext4fs_dirhash(dir, fname_name(fname),
|
| 2308 | fname_len(fname), &fname->hinfo);
|
| 2309 | if (err < 0) {
|
| 2310 | brelse(bh2);
|
| 2311 | brelse(bh);
|
| 2312 | return err;
|
| 2313 | }
|
| 2314 | }
|
| 2315 | memset(frames, 0, sizeof(frames));
|
| 2316 | frame = frames;
|
| 2317 | frame->entries = entries;
|
| 2318 | frame->at = entries;
|
| 2319 | frame->bh = bh;
|
| 2320 |
|
| 2321 | retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
|
| 2322 | if (retval)
|
| 2323 | goto out_frames;
|
| 2324 | retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
|
| 2325 | if (retval)
|
| 2326 | goto out_frames;
|
| 2327 |
|
| 2328 | de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
|
| 2329 | if (IS_ERR(de)) {
|
| 2330 | retval = PTR_ERR(de);
|
| 2331 | goto out_frames;
|
| 2332 | }
|
| 2333 |
|
| 2334 | retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
|
| 2335 | out_frames:
|
| 2336 | |
| 2337 | * Even if the block split failed, we have to properly write
|
| 2338 | * out all the changes we did so far. Otherwise we can end up
|
| 2339 | * with corrupted filesystem.
|
| 2340 |
|
| 2341 | if (retval)
|
| 2342 | ext4_mark_inode_dirty(handle, dir);
|
| 2343 | dx_release(frames);
|
| 2344 | brelse(bh2);
|
| 2345 | return retval;
|
| 2346 | }
|
| 2347 |
|
| 2348 | |
| 2349 | * ext4_add_entry()
|
| 2350 | *
|
| 2351 | * adds a file entry to the specified directory, using the same
|
| 2352 | * semantics as ext4_find_entry(). It returns NULL if it failed.
|
| 2353 | *
|
| 2354 | * NOTE!! The inode part of 'de' is left at 0 - which means you
|
| 2355 | * may not sleep between calling this and putting something into
|
| 2356 | * the entry, as someone else might have used it while you slept.
|
| 2357 |
|
| 2358 | static int __ext4_add_entry(handle_t *handle, struct inode *dir,
|
| 2359 | const struct qstr *d_name,
|
| 2360 | struct inode *inode)
|
| 2361 | {
|
| 2362 | struct buffer_head *bh = NULL;
|
| 2363 | struct ext4_dir_entry_2 *de;
|
| 2364 | struct super_block *sb;
|
| 2365 | struct ext4_filename fname;
|
| 2366 | int retval;
|
| 2367 | int dx_fallback=0;
|
| 2368 | unsigned blocksize;
|
| 2369 | ext4_lblk_t block, blocks;
|
| 2370 | int csum_size = 0;
|
| 2371 |
|
| 2372 | if (ext4_has_feature_metadata_csum(inode->i_sb))
|
| 2373 | csum_size = sizeof(struct ext4_dir_entry_tail);
|
| 2374 |
|
| 2375 | sb = dir->i_sb;
|
| 2376 | blocksize = sb->s_blocksize;
|
| 2377 |
|
| 2378 | if (!generic_ci_validate_strict_name(dir, d_name))
|
| 2379 | return -EINVAL;
|
| 2380 |
|
| 2381 | retval = ext4_fname_setup_filename(dir, d_name, 0, &fname);
|
| 2382 | if (retval)
|
| 2383 | return retval;
|
| 2384 |
|
| 2385 | if (ext4_has_inline_data(dir)) {
|
| 2386 | retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
|
| 2387 | if (retval < 0)
|
| 2388 | goto out;
|
| 2389 | if (retval == 1) {
|
| 2390 | retval = 0;
|
| 2391 | goto out;
|
| 2392 | }
|
| 2393 | }
|
| 2394 |
|
| 2395 | if (is_dx(dir)) {
|
| 2396 | retval = ext4_dx_add_entry(handle, &fname, dir, inode);
|
| 2397 | if (!retval || (retval != ERR_BAD_DX_DIR))
|
| 2398 | goto out;
|
| 2399 |
|
| 2400 | if (ext4_has_feature_metadata_csum(sb)) {
|
| 2401 | EXT4_ERROR_INODE(dir,
|
| 2402 | "Directory has corrupted htree index.");
|
| 2403 | retval = -EFSCORRUPTED;
|
| 2404 | goto out;
|
| 2405 | }
|
| 2406 | ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
|
| 2407 | dx_fallback++;
|
| 2408 | retval = ext4_mark_inode_dirty(handle, dir);
|
| 2409 | if (unlikely(retval))
|
| 2410 | goto out;
|
| 2411 | }
|
| 2412 | blocks = dir->i_size >> sb->s_blocksize_bits;
|
| 2413 | for (block = 0; block < blocks; block++) {
|
| 2414 | bh = ext4_read_dirblock(dir, block, DIRENT);
|
| 2415 | if (bh == NULL) {
|
| 2416 | bh = ext4_bread(handle, dir, block,
|
| 2417 | EXT4_GET_BLOCKS_CREATE);
|
| 2418 | goto add_to_new_block;
|
| 2419 | }
|
| 2420 | if (IS_ERR(bh)) {
|
| 2421 | retval = PTR_ERR(bh);
|
| 2422 | bh = NULL;
|
| 2423 | goto out;
|
| 2424 | }
|
| 2425 | retval = add_dirent_to_buf(handle, &fname, dir, inode,
|
| 2426 | NULL, bh);
|
| 2427 | if (retval != -ENOSPC)
|
| 2428 | goto out;
|
| 2429 |
|
| 2430 | if (blocks == 1 && !dx_fallback &&
|
| 2431 | ext4_has_feature_dir_index(sb)) {
|
| 2432 | retval = make_indexed_dir(handle, &fname, dir,
|
| 2433 | inode, bh);
|
| 2434 | bh = NULL;
|
| 2435 | goto out;
|
| 2436 | }
|
| 2437 | brelse(bh);
|
| 2438 | }
|
| 2439 | bh = ext4_append(handle, dir, &block);
|
| 2440 | add_to_new_block:
|
| 2441 | if (IS_ERR(bh)) {
|
| 2442 | retval = PTR_ERR(bh);
|
| 2443 | bh = NULL;
|
| 2444 | goto out;
|
| 2445 | }
|
| 2446 | de = (struct ext4_dir_entry_2 *) bh->b_data;
|
| 2447 | de->inode = 0;
|
| 2448 | de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
|
| 2449 |
|
| 2450 | if (csum_size)
|
| 2451 | ext4_initialize_dirent_tail(bh, blocksize);
|
| 2452 |
|
| 2453 | retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
|
| 2454 | out:
|
| 2455 | ext4_fname_free_filename(&fname);
|
| 2456 | brelse(bh);
|
| 2457 | if (retval == 0)
|
| 2458 | ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
|
| 2459 | return retval;
|
| 2460 | }
|
| 2461 |
|
| 2462 | static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
|
| 2463 | struct inode *inode)
|
| 2464 | {
|
| 2465 | struct inode *dir = d_inode(dentry->d_parent);
|
| 2466 |
|
| 2467 | if (fscrypt_is_nokey_name(dentry))
|
| 2468 | return -ENOKEY;
|
| 2469 | return __ext4_add_entry(handle, dir, &dentry->d_name, inode);
|
| 2470 | }
|
| 2471 |
|
| 2472 | |
| 2473 | * Returns 0 for success, or a negative error value
|
| 2474 |
|
| 2475 | static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
|
| 2476 | struct inode *dir, struct inode *inode)
|
| 2477 | {
|
| 2478 | struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
|
| 2479 | struct dx_entry *entries, *at;
|
| 2480 | struct buffer_head *bh;
|
| 2481 | struct super_block *sb = dir->i_sb;
|
| 2482 | struct ext4_dir_entry_2 *de;
|
| 2483 | int restart;
|
| 2484 | int err;
|
| 2485 |
|
| 2486 | again:
|
| 2487 | restart = 0;
|
| 2488 | frame = dx_probe(fname, dir, NULL, frames);
|
| 2489 | if (IS_ERR(frame))
|
| 2490 | return PTR_ERR(frame);
|
| 2491 | entries = frame->entries;
|
| 2492 | at = frame->at;
|
| 2493 | bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
|
| 2494 | if (IS_ERR(bh)) {
|
| 2495 | err = PTR_ERR(bh);
|
| 2496 | bh = NULL;
|
| 2497 | goto cleanup;
|
| 2498 | }
|
| 2499 |
|
| 2500 | BUFFER_TRACE(bh, "get_write_access");
|
| 2501 | err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
|
| 2502 | if (err)
|
| 2503 | goto journal_error;
|
| 2504 |
|
| 2505 | err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
|
| 2506 | if (err != -ENOSPC)
|
| 2507 | goto cleanup;
|
| 2508 |
|
| 2509 | err = 0;
|
| 2510 |
|
| 2511 | dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
|
| 2512 | dx_get_count(entries), dx_get_limit(entries)));
|
| 2513 |
|
| 2514 | if (dx_get_count(entries) == dx_get_limit(entries)) {
|
| 2515 | ext4_lblk_t newblock;
|
| 2516 | int levels = frame - frames + 1;
|
| 2517 | unsigned int icount;
|
| 2518 | int add_level = 1;
|
| 2519 | struct dx_entry *entries2;
|
| 2520 | struct dx_node *node2;
|
| 2521 | struct buffer_head *bh2;
|
| 2522 |
|
| 2523 | while (frame > frames) {
|
| 2524 | if (dx_get_count((frame - 1)->entries) <
|
| 2525 | dx_get_limit((frame - 1)->entries)) {
|
| 2526 | add_level = 0;
|
| 2527 | break;
|
| 2528 | }
|
| 2529 | frame--;
|
| 2530 | at = frame->at;
|
| 2531 | entries = frame->entries;
|
| 2532 | restart = 1;
|
| 2533 | }
|
| 2534 | if (add_level && levels == ext4_dir_htree_level(sb)) {
|
| 2535 | ext4_warning(sb, "Directory (ino: %llu) index full, "
|
| 2536 | "reach max htree level :%d",
|
| 2537 | dir->i_ino, levels);
|
| 2538 | if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
|
| 2539 | ext4_warning(sb, "Large directory feature is "
|
| 2540 | "not enabled on this "
|
| 2541 | "filesystem");
|
| 2542 | }
|
| 2543 | err = -ENOSPC;
|
| 2544 | goto cleanup;
|
| 2545 | }
|
| 2546 | icount = dx_get_count(entries);
|
| 2547 | bh2 = ext4_append(handle, dir, &newblock);
|
| 2548 | if (IS_ERR(bh2)) {
|
| 2549 | err = PTR_ERR(bh2);
|
| 2550 | goto cleanup;
|
| 2551 | }
|
| 2552 | node2 = (struct dx_node *)(bh2->b_data);
|
| 2553 | entries2 = node2->entries;
|
| 2554 | memset(&node2->fake, 0, sizeof(struct fake_dirent));
|
| 2555 | node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
|
| 2556 | sb->s_blocksize);
|
| 2557 | BUFFER_TRACE(frame->bh, "get_write_access");
|
| 2558 | err = ext4_journal_get_write_access(handle, sb, frame->bh,
|
| 2559 | EXT4_JTR_NONE);
|
| 2560 | if (err) {
|
| 2561 | brelse(bh2);
|
| 2562 | goto journal_error;
|
| 2563 | }
|
| 2564 | if (!add_level) {
|
| 2565 | unsigned icount1 = icount/2, icount2 = icount - icount1;
|
| 2566 | unsigned hash2 = dx_get_hash(entries + icount1);
|
| 2567 | dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
|
| 2568 | icount1, icount2));
|
| 2569 |
|
| 2570 | BUFFER_TRACE(frame->bh, "get_write_access");
|
| 2571 | err = ext4_journal_get_write_access(handle, sb,
|
| 2572 | (frame - 1)->bh,
|
| 2573 | EXT4_JTR_NONE);
|
| 2574 | if (err) {
|
| 2575 | brelse(bh2);
|
| 2576 | goto journal_error;
|
| 2577 | }
|
| 2578 |
|
| 2579 | memcpy((char *) entries2, (char *) (entries + icount1),
|
| 2580 | icount2 * sizeof(struct dx_entry));
|
| 2581 | dx_set_count(entries, icount1);
|
| 2582 | dx_set_count(entries2, icount2);
|
| 2583 | dx_set_limit(entries2, dx_node_limit(dir));
|
| 2584 |
|
| 2585 |
|
| 2586 | if (at - entries >= icount1) {
|
| 2587 | frame->at = at - entries - icount1 + entries2;
|
| 2588 | frame->entries = entries = entries2;
|
| 2589 | swap(frame->bh, bh2);
|
| 2590 | }
|
| 2591 | dx_insert_block((frame - 1), hash2, newblock);
|
| 2592 | dxtrace(dx_show_index("node", frame->entries));
|
| 2593 | dxtrace(dx_show_index("node",
|
| 2594 | ((struct dx_node *) bh2->b_data)->entries));
|
| 2595 | err = ext4_handle_dirty_dx_node(handle, dir, bh2);
|
| 2596 | if (err) {
|
| 2597 | brelse(bh2);
|
| 2598 | goto journal_error;
|
| 2599 | }
|
| 2600 | brelse (bh2);
|
| 2601 | err = ext4_handle_dirty_dx_node(handle, dir,
|
| 2602 | (frame - 1)->bh);
|
| 2603 | if (err)
|
| 2604 | goto journal_error;
|
| 2605 | err = ext4_handle_dirty_dx_node(handle, dir,
|
| 2606 | frame->bh);
|
| 2607 | if (restart || err)
|
| 2608 | goto journal_error;
|
| 2609 | } else {
|
| 2610 | struct dx_root *dxroot;
|
| 2611 | memcpy((char *) entries2, (char *) entries,
|
| 2612 | icount * sizeof(struct dx_entry));
|
| 2613 | dx_set_limit(entries2, dx_node_limit(dir));
|
| 2614 |
|
| 2615 |
|
| 2616 | dx_set_count(entries, 1);
|
| 2617 | dx_set_block(entries + 0, newblock);
|
| 2618 | dxroot = (struct dx_root *)frames[0].bh->b_data;
|
| 2619 | dxroot->info.indirect_levels += 1;
|
| 2620 | dxtrace(printk(KERN_DEBUG
|
| 2621 | "Creating %d level index...\n",
|
| 2622 | dxroot->info.indirect_levels));
|
| 2623 | err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
|
| 2624 | if (err) {
|
| 2625 | brelse(bh2);
|
| 2626 | goto journal_error;
|
| 2627 | }
|
| 2628 | err = ext4_handle_dirty_dx_node(handle, dir, bh2);
|
| 2629 | brelse(bh2);
|
| 2630 | restart = 1;
|
| 2631 | goto journal_error;
|
| 2632 | }
|
| 2633 | }
|
| 2634 | de = do_split(handle, dir, &bh, frame, &fname->hinfo);
|
| 2635 | if (IS_ERR(de)) {
|
| 2636 | err = PTR_ERR(de);
|
| 2637 | goto cleanup;
|
| 2638 | }
|
| 2639 | err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
|
| 2640 | goto cleanup;
|
| 2641 |
|
| 2642 | journal_error:
|
| 2643 | ext4_std_error(dir->i_sb, err);
|
| 2644 | cleanup:
|
| 2645 | brelse(bh);
|
| 2646 | dx_release(frames);
|
| 2647 | |
| 2648 | * repeat dx_probe() to find out valid htree-path
|
| 2649 |
|
| 2650 | if (restart && err == 0)
|
| 2651 | goto again;
|
| 2652 | return err;
|
| 2653 | }
|
| 2654 |
|
| 2655 | |
| 2656 | * ext4_generic_delete_entry deletes a directory entry by merging it
|
| 2657 | * with the previous entry
|
| 2658 |
|
| 2659 | int ext4_generic_delete_entry(struct inode *dir,
|
| 2660 | struct ext4_dir_entry_2 *de_del,
|
| 2661 | struct buffer_head *bh,
|
| 2662 | void *entry_buf,
|
| 2663 | int buf_size,
|
| 2664 | int csum_size)
|
| 2665 | {
|
| 2666 | struct ext4_dir_entry_2 *de, *pde;
|
| 2667 | unsigned int blocksize = dir->i_sb->s_blocksize;
|
| 2668 | int i;
|
| 2669 |
|
| 2670 | i = 0;
|
| 2671 | pde = NULL;
|
| 2672 | de = entry_buf;
|
| 2673 | while (i < buf_size - csum_size) {
|
| 2674 | if (ext4_check_dir_entry(dir, NULL, de, bh,
|
| 2675 | entry_buf, buf_size, i))
|
| 2676 | return -EFSCORRUPTED;
|
| 2677 | if (de == de_del) {
|
| 2678 | if (pde) {
|
| 2679 | pde->rec_len = ext4_rec_len_to_disk(
|
| 2680 | ext4_rec_len_from_disk(pde->rec_len,
|
| 2681 | blocksize) +
|
| 2682 | ext4_rec_len_from_disk(de->rec_len,
|
| 2683 | blocksize),
|
| 2684 | blocksize);
|
| 2685 |
|
| 2686 |
|
| 2687 | memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
|
| 2688 | blocksize));
|
| 2689 | } else {
|
| 2690 |
|
| 2691 | de->inode = 0;
|
| 2692 | memset(&de->name_len, 0,
|
| 2693 | ext4_rec_len_from_disk(de->rec_len,
|
| 2694 | blocksize) -
|
| 2695 | offsetof(struct ext4_dir_entry_2,
|
| 2696 | name_len));
|
| 2697 | }
|
| 2698 |
|
| 2699 | inode_inc_iversion(dir);
|
| 2700 | return 0;
|
| 2701 | }
|
| 2702 | i += ext4_rec_len_from_disk(de->rec_len, blocksize);
|
| 2703 | pde = de;
|
| 2704 | de = ext4_next_entry(de, blocksize);
|
| 2705 | }
|
| 2706 | return -ENOENT;
|
| 2707 | }
|
| 2708 |
|
| 2709 | static int ext4_delete_entry(handle_t *handle,
|
| 2710 | struct inode *dir,
|
| 2711 | struct ext4_dir_entry_2 *de_del,
|
| 2712 | struct buffer_head *bh)
|
| 2713 | {
|
| 2714 | int err, csum_size = 0;
|
| 2715 |
|
| 2716 | if (ext4_has_inline_data(dir)) {
|
| 2717 | int has_inline_data = 1;
|
| 2718 | err = ext4_delete_inline_entry(handle, dir, de_del, bh,
|
| 2719 | &has_inline_data);
|
| 2720 | if (has_inline_data)
|
| 2721 | return err;
|
| 2722 | }
|
| 2723 |
|
| 2724 | if (ext4_has_feature_metadata_csum(dir->i_sb))
|
| 2725 | csum_size = sizeof(struct ext4_dir_entry_tail);
|
| 2726 |
|
| 2727 | BUFFER_TRACE(bh, "get_write_access");
|
| 2728 | err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
|
| 2729 | EXT4_JTR_NONE);
|
| 2730 | if (unlikely(err))
|
| 2731 | goto out;
|
| 2732 |
|
| 2733 | err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
|
| 2734 | dir->i_sb->s_blocksize, csum_size);
|
| 2735 | if (err)
|
| 2736 | goto out;
|
| 2737 |
|
| 2738 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
|
| 2739 | err = ext4_handle_dirty_dirblock(handle, dir, bh);
|
| 2740 | if (unlikely(err))
|
| 2741 | goto out;
|
| 2742 |
|
| 2743 | return 0;
|
| 2744 | out:
|
| 2745 | if (err != -ENOENT)
|
| 2746 | ext4_std_error(dir->i_sb, err);
|
| 2747 | return err;
|
| 2748 | }
|
| 2749 |
|
| 2750 | |
| 2751 | * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
|
| 2752 | * since this indicates that nlinks count was previously 1 to avoid overflowing
|
| 2753 | * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
|
| 2754 | * that subdirectory link counts are not being maintained accurately.
|
| 2755 | *
|
| 2756 | * The caller has already checked for i_nlink overflow in case the DIR_LINK
|
| 2757 | * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
|
| 2758 | * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
|
| 2759 | * on regular files) and to avoid creating huge/slow non-HTREE directories.
|
| 2760 |
|
| 2761 | static void ext4_inc_count(struct inode *inode)
|
| 2762 | {
|
| 2763 | inc_nlink(inode);
|
| 2764 | if (is_dx(inode) &&
|
| 2765 | (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
|
| 2766 | set_nlink(inode, 1);
|
| 2767 | }
|
| 2768 |
|
| 2769 | |
| 2770 | * If a directory had nlink == 1, then we should let it be 1. This indicates
|
| 2771 | * directory has >EXT4_LINK_MAX subdirs.
|
| 2772 |
|
| 2773 | static void ext4_dec_count(struct inode *inode)
|
| 2774 | {
|
| 2775 | if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
|
| 2776 | drop_nlink(inode);
|
| 2777 | }
|
| 2778 |
|
| 2779 |
|
| 2780 | |
| 2781 | * Add non-directory inode to a directory. On success, the inode reference is
|
| 2782 | * consumed by dentry is instantiation. This is also indicated by clearing of
|
| 2783 | * *inodep pointer. On failure, the caller is responsible for dropping the
|
| 2784 | * inode reference in the safe context.
|
| 2785 |
|
| 2786 | static int ext4_add_nondir(handle_t *handle,
|
| 2787 | struct dentry *dentry, struct inode **inodep)
|
| 2788 | {
|
| 2789 | struct inode *dir = d_inode(dentry->d_parent);
|
| 2790 | struct inode *inode = *inodep;
|
| 2791 | int err = ext4_add_entry(handle, dentry, inode);
|
| 2792 | if (!err) {
|
| 2793 | err = ext4_mark_inode_dirty(handle, inode);
|
| 2794 | if (IS_DIRSYNC(dir))
|
| 2795 | ext4_handle_sync(handle);
|
| 2796 | d_instantiate_new(dentry, inode);
|
| 2797 | *inodep = NULL;
|
| 2798 | return err;
|
| 2799 | }
|
| 2800 | drop_nlink(inode);
|
| 2801 | ext4_mark_inode_dirty(handle, inode);
|
| 2802 | ext4_orphan_add(handle, inode);
|
| 2803 | unlock_new_inode(inode);
|
| 2804 | return err;
|
| 2805 | }
|
| 2806 |
|
| 2807 | |
| 2808 | * By the time this is called, we already have created
|
| 2809 | * the directory cache entry for the new file, but it
|
| 2810 | * is so far negative - it has no inode.
|
| 2811 | *
|
| 2812 | * If the create succeeds, we fill in the inode information
|
| 2813 | * with d_instantiate().
|
| 2814 |
|
| 2815 | static int ext4_create(const struct mnt_idmap *idmap, struct inode *dir,
|
| 2816 | struct dentry *dentry, umode_t mode)
|
| 2817 | {
|
| 2818 | handle_t *handle;
|
| 2819 | struct inode *inode;
|
| 2820 | int err, credits, retries = 0;
|
| 2821 |
|
| 2822 | err = dquot_initialize(dir);
|
| 2823 | if (err)
|
| 2824 | return err;
|
| 2825 |
|
| 2826 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
|
| 2827 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
|
| 2828 | retry:
|
| 2829 | inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
|
| 2830 | 0, NULL, EXT4_HT_DIR, credits);
|
| 2831 | handle = ext4_journal_current_handle();
|
| 2832 | err = PTR_ERR(inode);
|
| 2833 | if (!IS_ERR(inode)) {
|
| 2834 | inode->i_op = &ext4_file_inode_operations;
|
| 2835 | inode->i_fop = &ext4_file_operations;
|
| 2836 | ext4_set_aops(inode);
|
| 2837 | err = ext4_add_nondir(handle, dentry, &inode);
|
| 2838 | if (!err)
|
| 2839 | ext4_fc_track_create(handle, dentry);
|
| 2840 | }
|
| 2841 | if (handle)
|
| 2842 | ext4_journal_stop(handle);
|
| 2843 | if (!IS_ERR_OR_NULL(inode))
|
| 2844 | iput(inode);
|
| 2845 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
|
| 2846 | goto retry;
|
| 2847 | return err;
|
| 2848 | }
|
| 2849 |
|
| 2850 | static int ext4_mknod(const struct mnt_idmap *idmap, struct inode *dir,
|
| 2851 | struct dentry *dentry, umode_t mode, dev_t rdev)
|
| 2852 | {
|
| 2853 | handle_t *handle;
|
| 2854 | struct inode *inode;
|
| 2855 | int err, credits, retries = 0;
|
| 2856 |
|
| 2857 | err = dquot_initialize(dir);
|
| 2858 | if (err)
|
| 2859 | return err;
|
| 2860 |
|
| 2861 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
|
| 2862 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
|
| 2863 | retry:
|
| 2864 | inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
|
| 2865 | 0, NULL, EXT4_HT_DIR, credits);
|
| 2866 | handle = ext4_journal_current_handle();
|
| 2867 | err = PTR_ERR(inode);
|
| 2868 | if (!IS_ERR(inode)) {
|
| 2869 | init_special_inode(inode, inode->i_mode, rdev);
|
| 2870 | inode->i_op = &ext4_special_inode_operations;
|
| 2871 | err = ext4_add_nondir(handle, dentry, &inode);
|
| 2872 | if (!err)
|
| 2873 | ext4_fc_track_create(handle, dentry);
|
| 2874 | }
|
| 2875 | if (handle)
|
| 2876 | ext4_journal_stop(handle);
|
| 2877 | if (!IS_ERR_OR_NULL(inode))
|
| 2878 | iput(inode);
|
| 2879 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
|
| 2880 | goto retry;
|
| 2881 | return err;
|
| 2882 | }
|
| 2883 |
|
| 2884 | static int ext4_tmpfile(const struct mnt_idmap *idmap, struct inode *dir,
|
| 2885 | struct file *file, umode_t mode)
|
| 2886 | {
|
| 2887 | handle_t *handle;
|
| 2888 | struct inode *inode;
|
| 2889 | int err, retries = 0;
|
| 2890 |
|
| 2891 | err = dquot_initialize(dir);
|
| 2892 | if (err)
|
| 2893 | return err;
|
| 2894 |
|
| 2895 | retry:
|
| 2896 | inode = ext4_new_inode_start_handle(idmap, dir, mode,
|
| 2897 | NULL, 0, NULL,
|
| 2898 | EXT4_HT_DIR,
|
| 2899 | EXT4_MAXQUOTAS_TRANS_BLOCKS(dir->i_sb) +
|
| 2900 | 4 + EXT4_XATTR_TRANS_BLOCKS);
|
| 2901 | handle = ext4_journal_current_handle();
|
| 2902 | err = PTR_ERR(inode);
|
| 2903 | if (!IS_ERR(inode)) {
|
| 2904 | inode->i_op = &ext4_file_inode_operations;
|
| 2905 | inode->i_fop = &ext4_file_operations;
|
| 2906 | ext4_set_aops(inode);
|
| 2907 | d_tmpfile(file, inode);
|
| 2908 | err = ext4_orphan_add(handle, inode);
|
| 2909 | if (err)
|
| 2910 | goto err_unlock_inode;
|
| 2911 | mark_inode_dirty(inode);
|
| 2912 | unlock_new_inode(inode);
|
| 2913 | }
|
| 2914 | if (handle)
|
| 2915 | ext4_journal_stop(handle);
|
| 2916 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
|
| 2917 | goto retry;
|
| 2918 | return finish_open_simple(file, err);
|
| 2919 | err_unlock_inode:
|
| 2920 | ext4_journal_stop(handle);
|
| 2921 | unlock_new_inode(inode);
|
| 2922 | return err;
|
| 2923 | }
|
| 2924 |
|
| 2925 | int ext4_init_dirblock(handle_t *handle, struct inode *inode,
|
| 2926 | struct buffer_head *bh, unsigned int parent_ino,
|
| 2927 | void *inline_buf, int inline_size)
|
| 2928 | {
|
| 2929 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) bh->b_data;
|
| 2930 | size_t blocksize = bh->b_size;
|
| 2931 | int csum_size = 0, header_size;
|
| 2932 |
|
| 2933 | if (ext4_has_feature_metadata_csum(inode->i_sb))
|
| 2934 | csum_size = sizeof(struct ext4_dir_entry_tail);
|
| 2935 |
|
| 2936 | de->inode = cpu_to_le32(inode->i_ino);
|
| 2937 | de->name_len = 1;
|
| 2938 | de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
|
| 2939 | blocksize);
|
| 2940 | memcpy(de->name, ".", 2);
|
| 2941 | ext4_set_de_type(inode->i_sb, de, S_IFDIR);
|
| 2942 |
|
| 2943 | de = ext4_next_entry(de, blocksize);
|
| 2944 | de->inode = cpu_to_le32(parent_ino);
|
| 2945 | de->name_len = 2;
|
| 2946 | memcpy(de->name, "..", 3);
|
| 2947 | ext4_set_de_type(inode->i_sb, de, S_IFDIR);
|
| 2948 | if (inline_buf) {
|
| 2949 | de->rec_len = ext4_rec_len_to_disk(
|
| 2950 | ext4_dir_rec_len(de->name_len, NULL),
|
| 2951 | blocksize);
|
| 2952 | de = ext4_next_entry(de, blocksize);
|
| 2953 | header_size = (char *)de - bh->b_data;
|
| 2954 | memcpy((void *)de, inline_buf, inline_size);
|
| 2955 | ext4_update_final_de(bh->b_data, inline_size + header_size,
|
| 2956 | blocksize - csum_size);
|
| 2957 | } else {
|
| 2958 | de->rec_len = ext4_rec_len_to_disk(blocksize -
|
| 2959 | (csum_size + ext4_dir_rec_len(1, NULL)),
|
| 2960 | blocksize);
|
| 2961 | }
|
| 2962 |
|
| 2963 | if (csum_size)
|
| 2964 | ext4_initialize_dirent_tail(bh, blocksize);
|
| 2965 | BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
|
| 2966 | set_buffer_uptodate(bh);
|
| 2967 | set_buffer_verified(bh);
|
| 2968 | return ext4_handle_dirty_dirblock(handle, inode, bh);
|
| 2969 | }
|
| 2970 |
|
| 2971 | int ext4_init_new_dir(handle_t *handle, struct inode *dir,
|
| 2972 | struct inode *inode)
|
| 2973 | {
|
| 2974 | struct buffer_head *dir_block = NULL;
|
| 2975 | ext4_lblk_t block = 0;
|
| 2976 | int err;
|
| 2977 |
|
| 2978 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
|
| 2979 | err = ext4_try_create_inline_dir(handle, dir, inode);
|
| 2980 | if (err < 0 && err != -ENOSPC)
|
| 2981 | goto out;
|
| 2982 | if (!err)
|
| 2983 | goto out;
|
| 2984 | }
|
| 2985 |
|
| 2986 | set_nlink(inode, 2);
|
| 2987 | inode->i_size = 0;
|
| 2988 | dir_block = ext4_append(handle, inode, &block);
|
| 2989 | if (IS_ERR(dir_block))
|
| 2990 | return PTR_ERR(dir_block);
|
| 2991 | err = ext4_init_dirblock(handle, inode, dir_block, dir->i_ino, NULL, 0);
|
| 2992 | out:
|
| 2993 | brelse(dir_block);
|
| 2994 | return err;
|
| 2995 | }
|
| 2996 |
|
| 2997 | static struct dentry *ext4_mkdir(const struct mnt_idmap *idmap, struct inode *dir,
|
| 2998 | struct dentry *dentry, umode_t mode)
|
| 2999 | {
|
| 3000 | handle_t *handle;
|
| 3001 | struct inode *inode;
|
| 3002 | int err, err2 = 0, credits, retries = 0;
|
| 3003 |
|
| 3004 | if (EXT4_DIR_LINK_MAX(dir))
|
| 3005 | return ERR_PTR(-EMLINK);
|
| 3006 |
|
| 3007 | err = dquot_initialize(dir);
|
| 3008 | if (err)
|
| 3009 | return ERR_PTR(err);
|
| 3010 |
|
| 3011 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
|
| 3012 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
|
| 3013 | retry:
|
| 3014 | inode = ext4_new_inode_start_handle(idmap, dir, mode,
|
| 3015 | &dentry->d_name,
|
| 3016 | 0, NULL, EXT4_HT_DIR, credits);
|
| 3017 | handle = ext4_journal_current_handle();
|
| 3018 | err = PTR_ERR(inode);
|
| 3019 | if (IS_ERR(inode))
|
| 3020 | goto out_stop;
|
| 3021 |
|
| 3022 | inode->i_op = &ext4_dir_inode_operations;
|
| 3023 | inode->i_fop = &ext4_dir_operations;
|
| 3024 | err = ext4_init_new_dir(handle, dir, inode);
|
| 3025 | if (err)
|
| 3026 | goto out_clear_inode;
|
| 3027 | err = ext4_mark_inode_dirty(handle, inode);
|
| 3028 | if (!err)
|
| 3029 | err = ext4_add_entry(handle, dentry, inode);
|
| 3030 | if (err) {
|
| 3031 | out_clear_inode:
|
| 3032 | clear_nlink(inode);
|
| 3033 | ext4_orphan_add(handle, inode);
|
| 3034 | unlock_new_inode(inode);
|
| 3035 | err2 = ext4_mark_inode_dirty(handle, inode);
|
| 3036 | if (unlikely(err2))
|
| 3037 | err = err2;
|
| 3038 | ext4_journal_stop(handle);
|
| 3039 | iput(inode);
|
| 3040 | goto out_retry;
|
| 3041 | }
|
| 3042 | ext4_inc_count(dir);
|
| 3043 |
|
| 3044 | ext4_update_dx_flag(dir);
|
| 3045 | err = ext4_mark_inode_dirty(handle, dir);
|
| 3046 | if (err)
|
| 3047 | goto out_clear_inode;
|
| 3048 | d_instantiate_new(dentry, inode);
|
| 3049 | ext4_fc_track_create(handle, dentry);
|
| 3050 | if (IS_DIRSYNC(dir))
|
| 3051 | ext4_handle_sync(handle);
|
| 3052 |
|
| 3053 | out_stop:
|
| 3054 | if (handle)
|
| 3055 | ext4_journal_stop(handle);
|
| 3056 | out_retry:
|
| 3057 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
|
| 3058 | goto retry;
|
| 3059 | return err ? ERR_PTR(err) : NULL;
|
| 3060 | }
|
| 3061 |
|
| 3062 | |
| 3063 | * routine to check that the specified directory is empty (for rmdir)
|
| 3064 |
|
| 3065 | bool ext4_empty_dir(struct inode *inode)
|
| 3066 | {
|
| 3067 | unsigned int offset;
|
| 3068 | struct buffer_head *bh;
|
| 3069 | struct ext4_dir_entry_2 *de;
|
| 3070 | struct super_block *sb;
|
| 3071 |
|
| 3072 | if (ext4_has_inline_data(inode)) {
|
| 3073 | int has_inline_data = 1;
|
| 3074 | int ret;
|
| 3075 |
|
| 3076 | ret = empty_inline_dir(inode, &has_inline_data);
|
| 3077 | if (has_inline_data)
|
| 3078 | return ret;
|
| 3079 | }
|
| 3080 |
|
| 3081 | sb = inode->i_sb;
|
| 3082 | if (inode->i_size < ext4_dir_rec_len(1, NULL) +
|
| 3083 | ext4_dir_rec_len(2, NULL)) {
|
| 3084 | EXT4_ERROR_INODE(inode, "invalid size");
|
| 3085 | return false;
|
| 3086 | }
|
| 3087 | bh = ext4_read_dirblock(inode, 0, EITHER);
|
| 3088 | if (IS_ERR(bh))
|
| 3089 | return false;
|
| 3090 |
|
| 3091 | de = (struct ext4_dir_entry_2 *) bh->b_data;
|
| 3092 | if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
|
| 3093 | 0) ||
|
| 3094 | le32_to_cpu(de->inode) != inode->i_ino || de->name_len != 1 ||
|
| 3095 | de->name[0] != '.') {
|
| 3096 | ext4_warning_inode(inode, "directory missing '.'");
|
| 3097 | brelse(bh);
|
| 3098 | return false;
|
| 3099 | }
|
| 3100 | offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
|
| 3101 | de = ext4_next_entry(de, sb->s_blocksize);
|
| 3102 | if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
|
| 3103 | offset) ||
|
| 3104 | le32_to_cpu(de->inode) == 0 || de->name_len != 2 ||
|
| 3105 | de->name[0] != '.' || de->name[1] != '.') {
|
| 3106 | ext4_warning_inode(inode, "directory missing '..'");
|
| 3107 | brelse(bh);
|
| 3108 | return false;
|
| 3109 | }
|
| 3110 | offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
|
| 3111 | while (offset < inode->i_size) {
|
| 3112 | if (!(offset & (sb->s_blocksize - 1))) {
|
| 3113 | unsigned int lblock;
|
| 3114 | brelse(bh);
|
| 3115 | lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
|
| 3116 | bh = ext4_read_dirblock(inode, lblock, EITHER);
|
| 3117 | if (bh == NULL) {
|
| 3118 | offset += sb->s_blocksize;
|
| 3119 | continue;
|
| 3120 | }
|
| 3121 | if (IS_ERR(bh))
|
| 3122 | return false;
|
| 3123 | }
|
| 3124 | de = (struct ext4_dir_entry_2 *) (bh->b_data +
|
| 3125 | (offset & (sb->s_blocksize - 1)));
|
| 3126 | if (ext4_check_dir_entry(inode, NULL, de, bh,
|
| 3127 | bh->b_data, bh->b_size, offset) ||
|
| 3128 | le32_to_cpu(de->inode)) {
|
| 3129 | brelse(bh);
|
| 3130 | return false;
|
| 3131 | }
|
| 3132 | offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
|
| 3133 | }
|
| 3134 | brelse(bh);
|
| 3135 | return true;
|
| 3136 | }
|
| 3137 |
|
| 3138 | static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
|
| 3139 | {
|
| 3140 | int retval;
|
| 3141 | struct inode *inode;
|
| 3142 | struct buffer_head *bh;
|
| 3143 | struct ext4_dir_entry_2 *de = NULL;
|
| 3144 | handle_t *handle = NULL;
|
| 3145 |
|
| 3146 | retval = ext4_emergency_state(dir->i_sb);
|
| 3147 | if (unlikely(retval))
|
| 3148 | return retval;
|
| 3149 |
|
| 3150 | |
| 3151 | * separate transaction
|
| 3152 | retval = dquot_initialize(dir);
|
| 3153 | if (retval)
|
| 3154 | return retval;
|
| 3155 | retval = dquot_initialize(d_inode(dentry));
|
| 3156 | if (retval)
|
| 3157 | return retval;
|
| 3158 |
|
| 3159 | retval = -ENOENT;
|
| 3160 | bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
|
| 3161 | if (IS_ERR(bh))
|
| 3162 | return PTR_ERR(bh);
|
| 3163 | if (!bh)
|
| 3164 | goto end_rmdir;
|
| 3165 |
|
| 3166 | inode = d_inode(dentry);
|
| 3167 |
|
| 3168 | retval = -EFSCORRUPTED;
|
| 3169 | if (le32_to_cpu(de->inode) != inode->i_ino)
|
| 3170 | goto end_rmdir;
|
| 3171 |
|
| 3172 | retval = -ENOTEMPTY;
|
| 3173 | if (!ext4_empty_dir(inode))
|
| 3174 | goto end_rmdir;
|
| 3175 |
|
| 3176 | handle = ext4_journal_start(dir, EXT4_HT_DIR,
|
| 3177 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
|
| 3178 | if (IS_ERR(handle)) {
|
| 3179 | retval = PTR_ERR(handle);
|
| 3180 | handle = NULL;
|
| 3181 | goto end_rmdir;
|
| 3182 | }
|
| 3183 |
|
| 3184 | if (IS_DIRSYNC(dir))
|
| 3185 | ext4_handle_sync(handle);
|
| 3186 |
|
| 3187 | retval = ext4_delete_entry(handle, dir, de, bh);
|
| 3188 | if (retval)
|
| 3189 | goto end_rmdir;
|
| 3190 | if (!EXT4_DIR_LINK_EMPTY(inode))
|
| 3191 | ext4_warning_inode(inode,
|
| 3192 | "empty directory '%.*s' has too many links (%u)",
|
| 3193 | dentry->d_name.len, dentry->d_name.name,
|
| 3194 | inode->i_nlink);
|
| 3195 | inode_inc_iversion(inode);
|
| 3196 | clear_nlink(inode);
|
| 3197 | |
| 3198 | * zero will ensure that the right thing happens during any
|
| 3199 | * recovery.
|
| 3200 | inode->i_size = 0;
|
| 3201 | ext4_orphan_add(handle, inode);
|
| 3202 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
|
| 3203 | inode_set_ctime_current(inode);
|
| 3204 | retval = ext4_mark_inode_dirty(handle, inode);
|
| 3205 | if (retval)
|
| 3206 | goto end_rmdir;
|
| 3207 | ext4_dec_count(dir);
|
| 3208 | ext4_update_dx_flag(dir);
|
| 3209 | ext4_fc_track_unlink(handle, dentry);
|
| 3210 | retval = ext4_mark_inode_dirty(handle, dir);
|
| 3211 |
|
| 3212 | |
| 3213 | * Case-insensitiveness. Eventually we'll want avoid
|
| 3214 | * invalidating the dentries here, alongside with returning the
|
| 3215 | * negative dentries at ext4_lookup(), when it is better
|
| 3216 | * supported by the VFS for the CI case.
|
| 3217 |
|
| 3218 | if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
|
| 3219 | d_invalidate(dentry);
|
| 3220 |
|
| 3221 | end_rmdir:
|
| 3222 | brelse(bh);
|
| 3223 | if (handle)
|
| 3224 | ext4_journal_stop(handle);
|
| 3225 | return retval;
|
| 3226 | }
|
| 3227 |
|
| 3228 | int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
|
| 3229 | struct inode *inode,
|
| 3230 | struct dentry *dentry )
|
| 3231 | {
|
| 3232 | int retval = -ENOENT;
|
| 3233 | struct buffer_head *bh;
|
| 3234 | struct ext4_dir_entry_2 *de = NULL;
|
| 3235 | handle_t *handle;
|
| 3236 | int skip_remove_dentry = 0;
|
| 3237 |
|
| 3238 | |
| 3239 | * Keep this outside the transaction; it may have to set up the
|
| 3240 | * directory's encryption key, which isn't GFP_NOFS-safe.
|
| 3241 |
|
| 3242 | bh = ext4_find_entry(dir, d_name, &de, NULL);
|
| 3243 | if (IS_ERR(bh))
|
| 3244 | return PTR_ERR(bh);
|
| 3245 |
|
| 3246 | if (!bh)
|
| 3247 | return -ENOENT;
|
| 3248 |
|
| 3249 | if (le32_to_cpu(de->inode) != inode->i_ino) {
|
| 3250 | |
| 3251 | * It's okay if we find dont find dentry which matches
|
| 3252 | * the inode. That's because it might have gotten
|
| 3253 | * renamed to a different inode number
|
| 3254 |
|
| 3255 | if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
|
| 3256 | skip_remove_dentry = 1;
|
| 3257 | else
|
| 3258 | goto out_bh;
|
| 3259 | }
|
| 3260 |
|
| 3261 | handle = ext4_journal_start(dir, EXT4_HT_DIR,
|
| 3262 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
|
| 3263 | if (IS_ERR(handle)) {
|
| 3264 | retval = PTR_ERR(handle);
|
| 3265 | goto out_bh;
|
| 3266 | }
|
| 3267 |
|
| 3268 | if (IS_DIRSYNC(dir))
|
| 3269 | ext4_handle_sync(handle);
|
| 3270 |
|
| 3271 | if (!skip_remove_dentry) {
|
| 3272 | retval = ext4_delete_entry(handle, dir, de, bh);
|
| 3273 | if (retval)
|
| 3274 | goto out_handle;
|
| 3275 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
|
| 3276 | ext4_update_dx_flag(dir);
|
| 3277 | retval = ext4_mark_inode_dirty(handle, dir);
|
| 3278 | if (retval)
|
| 3279 | goto out_handle;
|
| 3280 | } else {
|
| 3281 | retval = 0;
|
| 3282 | }
|
| 3283 | if (inode->i_nlink == 0)
|
| 3284 | ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
|
| 3285 | d_name->len, d_name->name);
|
| 3286 | else
|
| 3287 | drop_nlink(inode);
|
| 3288 | if (!inode->i_nlink)
|
| 3289 | ext4_orphan_add(handle, inode);
|
| 3290 | inode_set_ctime_current(inode);
|
| 3291 | retval = ext4_mark_inode_dirty(handle, inode);
|
| 3292 | if (dentry && !retval)
|
| 3293 | ext4_fc_track_unlink(handle, dentry);
|
| 3294 | out_handle:
|
| 3295 | ext4_journal_stop(handle);
|
| 3296 | out_bh:
|
| 3297 | brelse(bh);
|
| 3298 | return retval;
|
| 3299 | }
|
| 3300 |
|
| 3301 | static int ext4_unlink(struct inode *dir, struct dentry *dentry)
|
| 3302 | {
|
| 3303 | int retval;
|
| 3304 |
|
| 3305 | retval = ext4_emergency_state(dir->i_sb);
|
| 3306 | if (unlikely(retval))
|
| 3307 | return retval;
|
| 3308 |
|
| 3309 | trace_ext4_unlink_enter(dir, dentry);
|
| 3310 | |
| 3311 | * Initialize quotas before so that eventual writes go
|
| 3312 | * in separate transaction
|
| 3313 |
|
| 3314 | retval = dquot_initialize(dir);
|
| 3315 | if (retval)
|
| 3316 | goto out_trace;
|
| 3317 | retval = dquot_initialize(d_inode(dentry));
|
| 3318 | if (retval)
|
| 3319 | goto out_trace;
|
| 3320 |
|
| 3321 | retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
|
| 3322 |
|
| 3323 | |
| 3324 | * Case-insensitiveness. Eventually we'll want avoid
|
| 3325 | * invalidating the dentries here, alongside with returning the
|
| 3326 | * negative dentries at ext4_lookup(), when it is better
|
| 3327 | * supported by the VFS for the CI case.
|
| 3328 |
|
| 3329 | if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
|
| 3330 | d_invalidate(dentry);
|
| 3331 |
|
| 3332 | out_trace:
|
| 3333 | trace_ext4_unlink_exit(dentry, retval);
|
| 3334 | return retval;
|
| 3335 | }
|
| 3336 |
|
| 3337 | static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
|
| 3338 | struct fscrypt_str *disk_link)
|
| 3339 | {
|
| 3340 | struct buffer_head *bh;
|
| 3341 | char *kaddr;
|
| 3342 | int err = 0;
|
| 3343 |
|
| 3344 | bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
|
| 3345 | if (IS_ERR(bh))
|
| 3346 | return PTR_ERR(bh);
|
| 3347 |
|
| 3348 | BUFFER_TRACE(bh, "get_write_access");
|
| 3349 | err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
|
| 3350 | if (err)
|
| 3351 | goto out;
|
| 3352 |
|
| 3353 | kaddr = (char *)bh->b_data;
|
| 3354 | memcpy(kaddr, disk_link->name, disk_link->len);
|
| 3355 | inode->i_size = disk_link->len - 1;
|
| 3356 | EXT4_I(inode)->i_disksize = inode->i_size;
|
| 3357 | err = ext4_handle_dirty_metadata(handle, inode, bh);
|
| 3358 | out:
|
| 3359 | brelse(bh);
|
| 3360 | return err;
|
| 3361 | }
|
| 3362 |
|
| 3363 | static int ext4_symlink(const struct mnt_idmap *idmap, struct inode *dir,
|
| 3364 | struct dentry *dentry, const char *symname)
|
| 3365 | {
|
| 3366 | handle_t *handle;
|
| 3367 | struct inode *inode;
|
| 3368 | int err, len = strlen(symname);
|
| 3369 | int credits;
|
| 3370 | struct fscrypt_str disk_link;
|
| 3371 | int retries = 0;
|
| 3372 |
|
| 3373 | err = ext4_emergency_state(dir->i_sb);
|
| 3374 | if (unlikely(err))
|
| 3375 | return err;
|
| 3376 |
|
| 3377 | err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
|
| 3378 | &disk_link);
|
| 3379 | if (err)
|
| 3380 | return err;
|
| 3381 |
|
| 3382 | err = dquot_initialize(dir);
|
| 3383 | if (err)
|
| 3384 | return err;
|
| 3385 |
|
| 3386 | |
| 3387 | * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
|
| 3388 | * directory. +3 for inode, inode bitmap, group descriptor allocation.
|
| 3389 | * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
|
| 3390 | * modification.
|
| 3391 |
|
| 3392 | credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
|
| 3393 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
|
| 3394 | retry:
|
| 3395 | inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
|
| 3396 | &dentry->d_name, 0, NULL,
|
| 3397 | EXT4_HT_DIR, credits);
|
| 3398 | handle = ext4_journal_current_handle();
|
| 3399 | if (IS_ERR(inode)) {
|
| 3400 | if (handle)
|
| 3401 | ext4_journal_stop(handle);
|
| 3402 | err = PTR_ERR(inode);
|
| 3403 | goto out_retry;
|
| 3404 | }
|
| 3405 |
|
| 3406 | if (IS_ENCRYPTED(inode)) {
|
| 3407 | err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
|
| 3408 | if (err)
|
| 3409 | goto err_drop_inode;
|
| 3410 | inode->i_op = &ext4_encrypted_symlink_inode_operations;
|
| 3411 | } else {
|
| 3412 | if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
|
| 3413 | inode->i_op = &ext4_symlink_inode_operations;
|
| 3414 | } else {
|
| 3415 | inode->i_op = &ext4_fast_symlink_inode_operations;
|
| 3416 | }
|
| 3417 | }
|
| 3418 |
|
| 3419 | if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
|
| 3420 |
|
| 3421 | err = ext4_init_symlink_block(handle, inode, &disk_link);
|
| 3422 | if (err)
|
| 3423 | goto err_drop_inode;
|
| 3424 | } else {
|
| 3425 |
|
| 3426 | ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
|
| 3427 | memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
|
| 3428 | disk_link.len);
|
| 3429 | inode->i_size = disk_link.len - 1;
|
| 3430 | EXT4_I(inode)->i_disksize = inode->i_size;
|
| 3431 | if (!IS_ENCRYPTED(inode))
|
| 3432 | inode_set_cached_link(inode, (char *)&EXT4_I(inode)->i_data,
|
| 3433 | inode->i_size);
|
| 3434 | }
|
| 3435 | err = ext4_add_nondir(handle, dentry, &inode);
|
| 3436 | if (handle)
|
| 3437 | ext4_journal_stop(handle);
|
| 3438 | iput(inode);
|
| 3439 | goto out_retry;
|
| 3440 |
|
| 3441 | err_drop_inode:
|
| 3442 | clear_nlink(inode);
|
| 3443 | ext4_mark_inode_dirty(handle, inode);
|
| 3444 | ext4_orphan_add(handle, inode);
|
| 3445 | unlock_new_inode(inode);
|
| 3446 | if (handle)
|
| 3447 | ext4_journal_stop(handle);
|
| 3448 | iput(inode);
|
| 3449 | out_retry:
|
| 3450 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
|
| 3451 | goto retry;
|
| 3452 | if (disk_link.name != (unsigned char *)symname)
|
| 3453 | kfree(disk_link.name);
|
| 3454 | return err;
|
| 3455 | }
|
| 3456 |
|
| 3457 | int __ext4_link(struct inode *dir, struct inode *inode,
|
| 3458 | const struct qstr *d_name, struct dentry *dentry)
|
| 3459 | {
|
| 3460 | handle_t *handle;
|
| 3461 | int err, retries = 0;
|
| 3462 | retry:
|
| 3463 | handle = ext4_journal_start(dir, EXT4_HT_DIR,
|
| 3464 | (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
|
| 3465 | EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
|
| 3466 | if (IS_ERR(handle))
|
| 3467 | return PTR_ERR(handle);
|
| 3468 |
|
| 3469 | if (IS_DIRSYNC(dir))
|
| 3470 | ext4_handle_sync(handle);
|
| 3471 |
|
| 3472 | inode_set_ctime_current(inode);
|
| 3473 | ext4_inc_count(inode);
|
| 3474 |
|
| 3475 | err = __ext4_add_entry(handle, dir, d_name, inode);
|
| 3476 | if (!err) {
|
| 3477 | err = ext4_mark_inode_dirty(handle, inode);
|
| 3478 | |
| 3479 | * linked the first time
|
| 3480 |
|
| 3481 | if (inode->i_nlink == 1)
|
| 3482 | ext4_orphan_del(handle, inode);
|
| 3483 | if (dentry)
|
| 3484 | ext4_fc_track_link(handle, inode, dentry);
|
| 3485 | } else {
|
| 3486 | drop_nlink(inode);
|
| 3487 | }
|
| 3488 | ext4_journal_stop(handle);
|
| 3489 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
|
| 3490 | goto retry;
|
| 3491 | return err;
|
| 3492 | }
|
| 3493 |
|
| 3494 | static int ext4_link(struct dentry *old_dentry,
|
| 3495 | struct inode *dir, struct dentry *dentry)
|
| 3496 | {
|
| 3497 | struct inode *inode = d_inode(old_dentry);
|
| 3498 | int err;
|
| 3499 |
|
| 3500 | if (inode->i_nlink >= EXT4_LINK_MAX)
|
| 3501 | return -EMLINK;
|
| 3502 |
|
| 3503 | err = fscrypt_prepare_link(old_dentry, dir, dentry);
|
| 3504 | if (err)
|
| 3505 | return err;
|
| 3506 |
|
| 3507 | if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
|
| 3508 | (!projid_eq(EXT4_I(dir)->i_projid,
|
| 3509 | EXT4_I(old_dentry->d_inode)->i_projid)))
|
| 3510 | return -EXDEV;
|
| 3511 |
|
| 3512 | err = dquot_initialize(dir);
|
| 3513 | if (err)
|
| 3514 | return err;
|
| 3515 | err = __ext4_link(dir, inode, &dentry->d_name, dentry);
|
| 3516 | if (!err) {
|
| 3517 | ihold(inode);
|
| 3518 | d_instantiate(dentry, inode);
|
| 3519 | }
|
| 3520 | return err;
|
| 3521 | }
|
| 3522 | |
| 3523 | * Try to find buffer head where contains the parent block.
|
| 3524 | * It should be the inode block if it is inlined or the 1st block
|
| 3525 | * if it is a normal dir.
|
| 3526 |
|
| 3527 | static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
|
| 3528 | struct inode *inode,
|
| 3529 | int *retval,
|
| 3530 | struct ext4_dir_entry_2 **parent_de,
|
| 3531 | int *inlined)
|
| 3532 | {
|
| 3533 | struct buffer_head *bh;
|
| 3534 |
|
| 3535 | if (!ext4_has_inline_data(inode)) {
|
| 3536 | struct ext4_dir_entry_2 *de;
|
| 3537 | unsigned int offset;
|
| 3538 |
|
| 3539 | bh = ext4_read_dirblock(inode, 0, EITHER);
|
| 3540 | if (IS_ERR(bh)) {
|
| 3541 | *retval = PTR_ERR(bh);
|
| 3542 | return NULL;
|
| 3543 | }
|
| 3544 |
|
| 3545 | de = (struct ext4_dir_entry_2 *) bh->b_data;
|
| 3546 | if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
|
| 3547 | bh->b_size, 0) ||
|
| 3548 | le32_to_cpu(de->inode) != inode->i_ino ||
|
| 3549 | de->name_len != 1 || de->name[0] != '.') {
|
| 3550 | EXT4_ERROR_INODE(inode, "directory missing '.'");
|
| 3551 | brelse(bh);
|
| 3552 | *retval = -EFSCORRUPTED;
|
| 3553 | return NULL;
|
| 3554 | }
|
| 3555 | offset = ext4_rec_len_from_disk(de->rec_len,
|
| 3556 | inode->i_sb->s_blocksize);
|
| 3557 | de = ext4_next_entry(de, inode->i_sb->s_blocksize);
|
| 3558 | if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
|
| 3559 | bh->b_size, offset) ||
|
| 3560 | le32_to_cpu(de->inode) == 0 || de->name_len != 2 ||
|
| 3561 | de->name[0] != '.' || de->name[1] != '.') {
|
| 3562 | EXT4_ERROR_INODE(inode, "directory missing '..'");
|
| 3563 | brelse(bh);
|
| 3564 | *retval = -EFSCORRUPTED;
|
| 3565 | return NULL;
|
| 3566 | }
|
| 3567 | *parent_de = de;
|
| 3568 |
|
| 3569 | return bh;
|
| 3570 | }
|
| 3571 |
|
| 3572 | *inlined = 1;
|
| 3573 | return ext4_get_first_inline_block(inode, parent_de, retval);
|
| 3574 | }
|
| 3575 |
|
| 3576 | struct ext4_renament {
|
| 3577 | struct inode *dir;
|
| 3578 | struct dentry *dentry;
|
| 3579 | struct inode *inode;
|
| 3580 | bool is_dir;
|
| 3581 | int dir_nlink_delta;
|
| 3582 |
|
| 3583 |
|
| 3584 | struct buffer_head *bh;
|
| 3585 | struct ext4_dir_entry_2 *de;
|
| 3586 | int inlined;
|
| 3587 |
|
| 3588 |
|
| 3589 | struct buffer_head *dir_bh;
|
| 3590 | struct ext4_dir_entry_2 *parent_de;
|
| 3591 | int dir_inlined;
|
| 3592 | };
|
| 3593 |
|
| 3594 | static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
|
| 3595 | {
|
| 3596 | int retval;
|
| 3597 |
|
| 3598 | ent->is_dir = true;
|
| 3599 | if (!is_cross)
|
| 3600 | return 0;
|
| 3601 |
|
| 3602 | ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
|
| 3603 | &retval, &ent->parent_de,
|
| 3604 | &ent->dir_inlined);
|
| 3605 | if (!ent->dir_bh)
|
| 3606 | return retval;
|
| 3607 | if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
|
| 3608 | return -EFSCORRUPTED;
|
| 3609 | BUFFER_TRACE(ent->dir_bh, "get_write_access");
|
| 3610 | return ext4_journal_get_write_access(handle, ent->dir->i_sb,
|
| 3611 | ent->dir_bh, EXT4_JTR_NONE);
|
| 3612 | }
|
| 3613 |
|
| 3614 | static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
|
| 3615 | unsigned dir_ino)
|
| 3616 | {
|
| 3617 | int retval;
|
| 3618 |
|
| 3619 | if (!ent->dir_bh)
|
| 3620 | return 0;
|
| 3621 |
|
| 3622 | ent->parent_de->inode = cpu_to_le32(dir_ino);
|
| 3623 | BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
|
| 3624 | if (!ent->dir_inlined) {
|
| 3625 | if (is_dx(ent->inode)) {
|
| 3626 | retval = ext4_handle_dirty_dx_node(handle,
|
| 3627 | ent->inode,
|
| 3628 | ent->dir_bh);
|
| 3629 | } else {
|
| 3630 | retval = ext4_handle_dirty_dirblock(handle, ent->inode,
|
| 3631 | ent->dir_bh);
|
| 3632 | }
|
| 3633 | } else {
|
| 3634 | retval = ext4_mark_inode_dirty(handle, ent->inode);
|
| 3635 | }
|
| 3636 | if (retval) {
|
| 3637 | ext4_std_error(ent->dir->i_sb, retval);
|
| 3638 | return retval;
|
| 3639 | }
|
| 3640 | return 0;
|
| 3641 | }
|
| 3642 |
|
| 3643 | static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
|
| 3644 | unsigned ino, unsigned file_type)
|
| 3645 | {
|
| 3646 | int retval, retval2;
|
| 3647 |
|
| 3648 | BUFFER_TRACE(ent->bh, "get write access");
|
| 3649 | retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
|
| 3650 | EXT4_JTR_NONE);
|
| 3651 | if (retval)
|
| 3652 | return retval;
|
| 3653 | ent->de->inode = cpu_to_le32(ino);
|
| 3654 | if (ext4_has_feature_filetype(ent->dir->i_sb))
|
| 3655 | ent->de->file_type = file_type;
|
| 3656 | inode_inc_iversion(ent->dir);
|
| 3657 | inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
|
| 3658 | retval = ext4_mark_inode_dirty(handle, ent->dir);
|
| 3659 | BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
|
| 3660 | if (!ent->inlined) {
|
| 3661 | retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
|
| 3662 | if (unlikely(retval2)) {
|
| 3663 | ext4_std_error(ent->dir->i_sb, retval2);
|
| 3664 | return retval2;
|
| 3665 | }
|
| 3666 | }
|
| 3667 | return retval;
|
| 3668 | }
|
| 3669 |
|
| 3670 | static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
|
| 3671 | unsigned ino, unsigned file_type)
|
| 3672 | {
|
| 3673 | struct ext4_renament old = *ent;
|
| 3674 | int retval = 0;
|
| 3675 |
|
| 3676 | |
| 3677 | * old->de could have moved from under us during make indexed dir,
|
| 3678 | * so the old->de may no longer valid and need to find it again
|
| 3679 | * before reset old inode info.
|
| 3680 |
|
| 3681 | old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
|
| 3682 | &old.inlined);
|
| 3683 | if (IS_ERR(old.bh))
|
| 3684 | retval = PTR_ERR(old.bh);
|
| 3685 | if (!old.bh)
|
| 3686 | retval = -ENOENT;
|
| 3687 | if (retval) {
|
| 3688 | ext4_std_error(old.dir->i_sb, retval);
|
| 3689 | return;
|
| 3690 | }
|
| 3691 |
|
| 3692 | ext4_setent(handle, &old, ino, file_type);
|
| 3693 | brelse(old.bh);
|
| 3694 | }
|
| 3695 |
|
| 3696 | static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
|
| 3697 | const struct qstr *d_name)
|
| 3698 | {
|
| 3699 | int retval = -ENOENT;
|
| 3700 | struct buffer_head *bh;
|
| 3701 | struct ext4_dir_entry_2 *de = NULL;
|
| 3702 |
|
| 3703 | bh = ext4_find_entry(dir, d_name, &de, NULL);
|
| 3704 | if (IS_ERR(bh))
|
| 3705 | return PTR_ERR(bh);
|
| 3706 | if (bh) {
|
| 3707 | retval = ext4_delete_entry(handle, dir, de, bh);
|
| 3708 | brelse(bh);
|
| 3709 | }
|
| 3710 | return retval;
|
| 3711 | }
|
| 3712 |
|
| 3713 | static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
|
| 3714 | int force_reread)
|
| 3715 | {
|
| 3716 | int retval;
|
| 3717 | |
| 3718 | * ent->de could have moved from under us during htree split, so make
|
| 3719 | * sure that we are deleting the right entry. We might also be pointing
|
| 3720 | * to a stale entry in the unused part of ent->bh so just checking inum
|
| 3721 | * and the name isn't enough.
|
| 3722 |
|
| 3723 | if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
|
| 3724 | ent->de->name_len != ent->dentry->d_name.len ||
|
| 3725 | strncmp(ent->de->name, ent->dentry->d_name.name,
|
| 3726 | ent->de->name_len) ||
|
| 3727 | force_reread) {
|
| 3728 | retval = ext4_find_delete_entry(handle, ent->dir,
|
| 3729 | &ent->dentry->d_name);
|
| 3730 | } else {
|
| 3731 | retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
|
| 3732 | if (retval == -ENOENT) {
|
| 3733 | retval = ext4_find_delete_entry(handle, ent->dir,
|
| 3734 | &ent->dentry->d_name);
|
| 3735 | }
|
| 3736 | }
|
| 3737 |
|
| 3738 | if (retval) {
|
| 3739 | ext4_warning_inode(ent->dir,
|
| 3740 | "Deleting old file: nlink %d, error=%d",
|
| 3741 | ent->dir->i_nlink, retval);
|
| 3742 | }
|
| 3743 | }
|
| 3744 |
|
| 3745 | static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
|
| 3746 | {
|
| 3747 | if (ent->dir_nlink_delta) {
|
| 3748 | if (ent->dir_nlink_delta == -1)
|
| 3749 | ext4_dec_count(ent->dir);
|
| 3750 | else
|
| 3751 | ext4_inc_count(ent->dir);
|
| 3752 | ext4_mark_inode_dirty(handle, ent->dir);
|
| 3753 | }
|
| 3754 | }
|
| 3755 |
|
| 3756 | static struct inode *ext4_whiteout_for_rename(const struct mnt_idmap *idmap,
|
| 3757 | struct ext4_renament *ent,
|
| 3758 | int credits, handle_t **h)
|
| 3759 | {
|
| 3760 | struct inode *wh;
|
| 3761 | handle_t *handle;
|
| 3762 | int retries = 0;
|
| 3763 |
|
| 3764 | |
| 3765 | * for inode block, sb block, group summaries,
|
| 3766 | * and inode bitmap
|
| 3767 |
|
| 3768 | credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
|
| 3769 | EXT4_XATTR_TRANS_BLOCKS + 4);
|
| 3770 | retry:
|
| 3771 | wh = ext4_new_inode_start_handle(idmap, ent->dir,
|
| 3772 | S_IFCHR | WHITEOUT_MODE,
|
| 3773 | &ent->dentry->d_name, 0, NULL,
|
| 3774 | EXT4_HT_DIR, credits);
|
| 3775 |
|
| 3776 | handle = ext4_journal_current_handle();
|
| 3777 | if (IS_ERR(wh)) {
|
| 3778 | if (handle)
|
| 3779 | ext4_journal_stop(handle);
|
| 3780 | if (PTR_ERR(wh) == -ENOSPC &&
|
| 3781 | ext4_should_retry_alloc(ent->dir->i_sb, &retries))
|
| 3782 | goto retry;
|
| 3783 | } else {
|
| 3784 | *h = handle;
|
| 3785 | init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
|
| 3786 | wh->i_op = &ext4_special_inode_operations;
|
| 3787 | }
|
| 3788 | return wh;
|
| 3789 | }
|
| 3790 |
|
| 3791 | |
| 3792 | * Anybody can rename anything with this: the permission checks are left to the
|
| 3793 | * higher-level routines.
|
| 3794 | *
|
| 3795 | * n.b. old_{dentry,inode) refers to the source dentry/inode
|
| 3796 | * while new_{dentry,inode) refers to the destination dentry/inode
|
| 3797 | * This comes from rename(const char *oldpath, const char *newpath)
|
| 3798 |
|
| 3799 | static int ext4_rename(const struct mnt_idmap *idmap, struct inode *old_dir,
|
| 3800 | struct dentry *old_dentry, struct inode *new_dir,
|
| 3801 | struct dentry *new_dentry, unsigned int flags)
|
| 3802 | {
|
| 3803 | handle_t *handle = NULL;
|
| 3804 | struct ext4_renament old = {
|
| 3805 | .dir = old_dir,
|
| 3806 | .dentry = old_dentry,
|
| 3807 | .inode = d_inode(old_dentry),
|
| 3808 | };
|
| 3809 | struct ext4_renament new = {
|
| 3810 | .dir = new_dir,
|
| 3811 | .dentry = new_dentry,
|
| 3812 | .inode = d_inode(new_dentry),
|
| 3813 | };
|
| 3814 | int force_reread;
|
| 3815 | int retval;
|
| 3816 | struct inode *whiteout = NULL;
|
| 3817 | int credits;
|
| 3818 | u8 old_file_type;
|
| 3819 |
|
| 3820 | if (new.inode && new.inode->i_nlink == 0) {
|
| 3821 | EXT4_ERROR_INODE(new.inode,
|
| 3822 | "target of rename is already freed");
|
| 3823 | return -EFSCORRUPTED;
|
| 3824 | }
|
| 3825 |
|
| 3826 | if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
|
| 3827 | (!projid_eq(EXT4_I(new_dir)->i_projid,
|
| 3828 | EXT4_I(old_dentry->d_inode)->i_projid)))
|
| 3829 | return -EXDEV;
|
| 3830 |
|
| 3831 | retval = dquot_initialize(old.dir);
|
| 3832 | if (retval)
|
| 3833 | return retval;
|
| 3834 | retval = dquot_initialize(old.inode);
|
| 3835 | if (retval)
|
| 3836 | return retval;
|
| 3837 | retval = dquot_initialize(new.dir);
|
| 3838 | if (retval)
|
| 3839 | return retval;
|
| 3840 |
|
| 3841 | |
| 3842 | * in separate transaction
|
| 3843 | if (new.inode) {
|
| 3844 | retval = dquot_initialize(new.inode);
|
| 3845 | if (retval)
|
| 3846 | return retval;
|
| 3847 | }
|
| 3848 |
|
| 3849 | old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
|
| 3850 | &old.inlined);
|
| 3851 | if (IS_ERR(old.bh))
|
| 3852 | return PTR_ERR(old.bh);
|
| 3853 |
|
| 3854 | |
| 3855 | * Check for inode number is _not_ due to possible IO errors.
|
| 3856 | * We might rmdir the source, keep it as pwd of some process
|
| 3857 | * and merrily kill the link to whatever was created under the
|
| 3858 | * same name. Goodbye sticky bit ;-<
|
| 3859 |
|
| 3860 | retval = -ENOENT;
|
| 3861 | if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
|
| 3862 | goto release_bh;
|
| 3863 |
|
| 3864 | new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
|
| 3865 | &new.de, &new.inlined);
|
| 3866 | if (IS_ERR(new.bh)) {
|
| 3867 | retval = PTR_ERR(new.bh);
|
| 3868 | new.bh = NULL;
|
| 3869 | goto release_bh;
|
| 3870 | }
|
| 3871 | if (new.bh) {
|
| 3872 | if (!new.inode) {
|
| 3873 | brelse(new.bh);
|
| 3874 | new.bh = NULL;
|
| 3875 | }
|
| 3876 | }
|
| 3877 | if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
|
| 3878 | ext4_alloc_da_blocks(old.inode);
|
| 3879 |
|
| 3880 | credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
|
| 3881 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
|
| 3882 | if (!(flags & RENAME_WHITEOUT)) {
|
| 3883 | handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
|
| 3884 | if (IS_ERR(handle)) {
|
| 3885 | retval = PTR_ERR(handle);
|
| 3886 | goto release_bh;
|
| 3887 | }
|
| 3888 | } else {
|
| 3889 | whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
|
| 3890 | if (IS_ERR(whiteout)) {
|
| 3891 | retval = PTR_ERR(whiteout);
|
| 3892 | goto release_bh;
|
| 3893 | }
|
| 3894 | }
|
| 3895 |
|
| 3896 | old_file_type = old.de->file_type;
|
| 3897 | if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
|
| 3898 | ext4_handle_sync(handle);
|
| 3899 |
|
| 3900 | if (S_ISDIR(old.inode->i_mode)) {
|
| 3901 | if (new.inode) {
|
| 3902 | retval = -ENOTEMPTY;
|
| 3903 | if (!ext4_empty_dir(new.inode))
|
| 3904 | goto end_rename;
|
| 3905 | } else {
|
| 3906 | retval = -EMLINK;
|
| 3907 | if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
|
| 3908 | goto end_rename;
|
| 3909 | }
|
| 3910 | retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
|
| 3911 | if (retval)
|
| 3912 | goto end_rename;
|
| 3913 | }
|
| 3914 | |
| 3915 | * If we're renaming a file within an inline_data dir and adding or
|
| 3916 | * setting the new dirent causes a conversion from inline_data to
|
| 3917 | * extents/blockmap, we need to force the dirent delete code to
|
| 3918 | * re-read the directory, or else we end up trying to delete a dirent
|
| 3919 | * from what is now the extent tree root (or a block map).
|
| 3920 |
|
| 3921 | force_reread = (new.dir->i_ino == old.dir->i_ino &&
|
| 3922 | ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
|
| 3923 |
|
| 3924 | if (whiteout) {
|
| 3925 | |
| 3926 | * Do this before adding a new entry, so the old entry is sure
|
| 3927 | * to be still pointing to the valid old entry.
|
| 3928 |
|
| 3929 | retval = ext4_setent(handle, &old, whiteout->i_ino,
|
| 3930 | EXT4_FT_CHRDEV);
|
| 3931 | if (retval)
|
| 3932 | goto end_rename;
|
| 3933 | retval = ext4_mark_inode_dirty(handle, whiteout);
|
| 3934 | if (unlikely(retval))
|
| 3935 | goto end_rename;
|
| 3936 |
|
| 3937 | }
|
| 3938 | if (!new.bh) {
|
| 3939 | retval = ext4_add_entry(handle, new.dentry, old.inode);
|
| 3940 | if (retval)
|
| 3941 | goto end_rename;
|
| 3942 | } else {
|
| 3943 | retval = ext4_setent(handle, &new,
|
| 3944 | old.inode->i_ino, old_file_type);
|
| 3945 | if (retval)
|
| 3946 | goto end_rename;
|
| 3947 | }
|
| 3948 | if (force_reread)
|
| 3949 | force_reread = !ext4_test_inode_flag(new.dir,
|
| 3950 | EXT4_INODE_INLINE_DATA);
|
| 3951 |
|
| 3952 | |
| 3953 | * Like most other Unix systems, set the ctime for inodes on a
|
| 3954 | * rename.
|
| 3955 |
|
| 3956 | inode_set_ctime_current(old.inode);
|
| 3957 | retval = ext4_mark_inode_dirty(handle, old.inode);
|
| 3958 | if (unlikely(retval))
|
| 3959 | goto end_rename;
|
| 3960 |
|
| 3961 | if (!whiteout) {
|
| 3962 | |
| 3963 | * ok, that's it
|
| 3964 |
|
| 3965 | ext4_rename_delete(handle, &old, force_reread);
|
| 3966 | }
|
| 3967 |
|
| 3968 | if (new.inode) {
|
| 3969 | ext4_dec_count(new.inode);
|
| 3970 | inode_set_ctime_current(new.inode);
|
| 3971 | }
|
| 3972 | inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
|
| 3973 | ext4_update_dx_flag(old.dir);
|
| 3974 | if (old.is_dir) {
|
| 3975 | retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
|
| 3976 | if (retval)
|
| 3977 | goto end_rename;
|
| 3978 |
|
| 3979 | ext4_dec_count(old.dir);
|
| 3980 | if (new.inode) {
|
| 3981 | |
| 3982 | * parent, ext4_dec_count() won't work for many-linked
|
| 3983 | * dirs
|
| 3984 | clear_nlink(new.inode);
|
| 3985 | } else {
|
| 3986 | ext4_inc_count(new.dir);
|
| 3987 | ext4_update_dx_flag(new.dir);
|
| 3988 | retval = ext4_mark_inode_dirty(handle, new.dir);
|
| 3989 | if (unlikely(retval))
|
| 3990 | goto end_rename;
|
| 3991 | }
|
| 3992 | }
|
| 3993 | retval = ext4_mark_inode_dirty(handle, old.dir);
|
| 3994 | if (unlikely(retval))
|
| 3995 | goto end_rename;
|
| 3996 |
|
| 3997 | if (old.is_dir) {
|
| 3998 | |
| 3999 | * We disable fast commits here that's because the
|
| 4000 | * replay code is not yet capable of changing dot dot
|
| 4001 | * dirents in directories.
|
| 4002 |
|
| 4003 | ext4_fc_mark_ineligible(old.inode->i_sb,
|
| 4004 | EXT4_FC_REASON_RENAME_DIR, handle);
|
| 4005 | } else {
|
| 4006 | struct super_block *sb = old.inode->i_sb;
|
| 4007 |
|
| 4008 | if (new.inode)
|
| 4009 | ext4_fc_track_unlink(handle, new.dentry);
|
| 4010 | if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
|
| 4011 | !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
|
| 4012 | !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
|
| 4013 | __ext4_fc_track_link(handle, old.inode, new.dentry);
|
| 4014 | __ext4_fc_track_unlink(handle, old.inode, old.dentry);
|
| 4015 | if (whiteout)
|
| 4016 | __ext4_fc_track_create(handle, whiteout,
|
| 4017 | old.dentry);
|
| 4018 | }
|
| 4019 | }
|
| 4020 |
|
| 4021 | if (new.inode) {
|
| 4022 | retval = ext4_mark_inode_dirty(handle, new.inode);
|
| 4023 | if (unlikely(retval))
|
| 4024 | goto end_rename;
|
| 4025 | if (!new.inode->i_nlink)
|
| 4026 | ext4_orphan_add(handle, new.inode);
|
| 4027 | }
|
| 4028 | retval = 0;
|
| 4029 |
|
| 4030 | end_rename:
|
| 4031 | if (whiteout) {
|
| 4032 | if (retval) {
|
| 4033 | ext4_resetent(handle, &old,
|
| 4034 | old.inode->i_ino, old_file_type);
|
| 4035 | drop_nlink(whiteout);
|
| 4036 | ext4_mark_inode_dirty(handle, whiteout);
|
| 4037 | ext4_orphan_add(handle, whiteout);
|
| 4038 | }
|
| 4039 | unlock_new_inode(whiteout);
|
| 4040 | ext4_journal_stop(handle);
|
| 4041 | iput(whiteout);
|
| 4042 | } else {
|
| 4043 | ext4_journal_stop(handle);
|
| 4044 | }
|
| 4045 | release_bh:
|
| 4046 | brelse(old.dir_bh);
|
| 4047 | brelse(old.bh);
|
| 4048 | brelse(new.bh);
|
| 4049 |
|
| 4050 | return retval;
|
| 4051 | }
|
| 4052 |
|
| 4053 | static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
|
| 4054 | struct inode *new_dir, struct dentry *new_dentry)
|
| 4055 | {
|
| 4056 | handle_t *handle = NULL;
|
| 4057 | struct ext4_renament old = {
|
| 4058 | .dir = old_dir,
|
| 4059 | .dentry = old_dentry,
|
| 4060 | .inode = d_inode(old_dentry),
|
| 4061 | };
|
| 4062 | struct ext4_renament new = {
|
| 4063 | .dir = new_dir,
|
| 4064 | .dentry = new_dentry,
|
| 4065 | .inode = d_inode(new_dentry),
|
| 4066 | };
|
| 4067 | u8 new_file_type;
|
| 4068 | int retval;
|
| 4069 |
|
| 4070 | if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
|
| 4071 | !projid_eq(EXT4_I(new_dir)->i_projid,
|
| 4072 | EXT4_I(old_dentry->d_inode)->i_projid)) ||
|
| 4073 | (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
|
| 4074 | !projid_eq(EXT4_I(old_dir)->i_projid,
|
| 4075 | EXT4_I(new_dentry->d_inode)->i_projid)))
|
| 4076 | return -EXDEV;
|
| 4077 |
|
| 4078 | retval = dquot_initialize(old.dir);
|
| 4079 | if (retval)
|
| 4080 | return retval;
|
| 4081 | retval = dquot_initialize(new.dir);
|
| 4082 | if (retval)
|
| 4083 | return retval;
|
| 4084 |
|
| 4085 | old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
|
| 4086 | &old.de, &old.inlined);
|
| 4087 | if (IS_ERR(old.bh))
|
| 4088 | return PTR_ERR(old.bh);
|
| 4089 | |
| 4090 | * Check for inode number is _not_ due to possible IO errors.
|
| 4091 | * We might rmdir the source, keep it as pwd of some process
|
| 4092 | * and merrily kill the link to whatever was created under the
|
| 4093 | * same name. Goodbye sticky bit ;-<
|
| 4094 |
|
| 4095 | retval = -ENOENT;
|
| 4096 | if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
|
| 4097 | goto end_rename;
|
| 4098 |
|
| 4099 | new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
|
| 4100 | &new.de, &new.inlined);
|
| 4101 | if (IS_ERR(new.bh)) {
|
| 4102 | retval = PTR_ERR(new.bh);
|
| 4103 | new.bh = NULL;
|
| 4104 | goto end_rename;
|
| 4105 | }
|
| 4106 |
|
| 4107 |
|
| 4108 | if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
|
| 4109 | goto end_rename;
|
| 4110 |
|
| 4111 | handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
|
| 4112 | (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
|
| 4113 | 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
|
| 4114 | if (IS_ERR(handle)) {
|
| 4115 | retval = PTR_ERR(handle);
|
| 4116 | handle = NULL;
|
| 4117 | goto end_rename;
|
| 4118 | }
|
| 4119 |
|
| 4120 | if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
|
| 4121 | ext4_handle_sync(handle);
|
| 4122 |
|
| 4123 | if (S_ISDIR(old.inode->i_mode)) {
|
| 4124 | retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
|
| 4125 | if (retval)
|
| 4126 | goto end_rename;
|
| 4127 | }
|
| 4128 | if (S_ISDIR(new.inode->i_mode)) {
|
| 4129 | retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
|
| 4130 | if (retval)
|
| 4131 | goto end_rename;
|
| 4132 | }
|
| 4133 |
|
| 4134 | |
| 4135 | * Other than the special case of overwriting a directory, parents'
|
| 4136 | * nlink only needs to be modified if this is a cross directory rename.
|
| 4137 |
|
| 4138 | if (old.dir != new.dir && old.is_dir != new.is_dir) {
|
| 4139 | old.dir_nlink_delta = old.is_dir ? -1 : 1;
|
| 4140 | new.dir_nlink_delta = -old.dir_nlink_delta;
|
| 4141 | retval = -EMLINK;
|
| 4142 | if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
|
| 4143 | (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
|
| 4144 | goto end_rename;
|
| 4145 | }
|
| 4146 |
|
| 4147 | new_file_type = new.de->file_type;
|
| 4148 | retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
|
| 4149 | if (retval)
|
| 4150 | goto end_rename;
|
| 4151 |
|
| 4152 | retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
|
| 4153 | if (retval)
|
| 4154 | goto end_rename;
|
| 4155 |
|
| 4156 | |
| 4157 | * Like most other Unix systems, set the ctime for inodes on a
|
| 4158 | * rename.
|
| 4159 |
|
| 4160 | inode_set_ctime_current(old.inode);
|
| 4161 | inode_set_ctime_current(new.inode);
|
| 4162 | retval = ext4_mark_inode_dirty(handle, old.inode);
|
| 4163 | if (unlikely(retval))
|
| 4164 | goto end_rename;
|
| 4165 | retval = ext4_mark_inode_dirty(handle, new.inode);
|
| 4166 | if (unlikely(retval))
|
| 4167 | goto end_rename;
|
| 4168 | ext4_fc_mark_ineligible(new.inode->i_sb,
|
| 4169 | EXT4_FC_REASON_CROSS_RENAME, handle);
|
| 4170 | if (old.dir_bh) {
|
| 4171 | retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
|
| 4172 | if (retval)
|
| 4173 | goto end_rename;
|
| 4174 | }
|
| 4175 | if (new.dir_bh) {
|
| 4176 | retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
|
| 4177 | if (retval)
|
| 4178 | goto end_rename;
|
| 4179 | }
|
| 4180 | ext4_update_dir_count(handle, &old);
|
| 4181 | ext4_update_dir_count(handle, &new);
|
| 4182 | retval = 0;
|
| 4183 |
|
| 4184 | end_rename:
|
| 4185 | brelse(old.dir_bh);
|
| 4186 | brelse(new.dir_bh);
|
| 4187 | brelse(old.bh);
|
| 4188 | brelse(new.bh);
|
| 4189 | if (handle)
|
| 4190 | ext4_journal_stop(handle);
|
| 4191 | return retval;
|
| 4192 | }
|
| 4193 |
|
| 4194 | static int ext4_rename2(const struct mnt_idmap *idmap,
|
| 4195 | struct inode *old_dir, struct dentry *old_dentry,
|
| 4196 | struct inode *new_dir, struct dentry *new_dentry,
|
| 4197 | unsigned int flags)
|
| 4198 | {
|
| 4199 | int err;
|
| 4200 |
|
| 4201 | err = ext4_emergency_state(old_dir->i_sb);
|
| 4202 | if (unlikely(err))
|
| 4203 | return err;
|
| 4204 |
|
| 4205 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
|
| 4206 | return -EINVAL;
|
| 4207 |
|
| 4208 | err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
|
| 4209 | flags);
|
| 4210 | if (err)
|
| 4211 | return err;
|
| 4212 |
|
| 4213 | if (flags & RENAME_EXCHANGE) {
|
| 4214 | return ext4_cross_rename(old_dir, old_dentry,
|
| 4215 | new_dir, new_dentry);
|
| 4216 | }
|
| 4217 |
|
| 4218 | return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
|
| 4219 | }
|
| 4220 |
|
| 4221 | |
| 4222 | * directories can handle most operations...
|
| 4223 |
|
| 4224 | const struct inode_operations ext4_dir_inode_operations = {
|
| 4225 | .create = ext4_create,
|
| 4226 | .lookup = ext4_lookup,
|
| 4227 | .link = ext4_link,
|
| 4228 | .unlink = ext4_unlink,
|
| 4229 | .symlink = ext4_symlink,
|
| 4230 | .mkdir = ext4_mkdir,
|
| 4231 | .rmdir = ext4_rmdir,
|
| 4232 | .mknod = ext4_mknod,
|
| 4233 | .tmpfile = ext4_tmpfile,
|
| 4234 | .rename = ext4_rename2,
|
| 4235 | .setattr = ext4_setattr,
|
| 4236 | .getattr = ext4_getattr,
|
| 4237 | .listxattr = ext4_listxattr,
|
| 4238 | .get_inode_acl = ext4_get_acl,
|
| 4239 | .set_acl = ext4_set_acl,
|
| 4240 | .fiemap = ext4_fiemap,
|
| 4241 | .fileattr_get = ext4_fileattr_get,
|
| 4242 | .fileattr_set = ext4_fileattr_set,
|
| 4243 | };
|
| 4244 |
|
| 4245 | const struct inode_operations ext4_special_inode_operations = {
|
| 4246 | .setattr = ext4_setattr,
|
| 4247 | .getattr = ext4_getattr,
|
| 4248 | .listxattr = ext4_listxattr,
|
| 4249 | .get_inode_acl = ext4_get_acl,
|
| 4250 | .set_acl = ext4_set_acl,
|
| 4251 | };
|