2 fs/ext4/dir.c | 26 +++++++++---
3 fs/ext4/ext4.h | 70 ++++++++++++++++++++++++++++++++-
4 fs/ext4/namei.c | 117 ++++++++++++++++++++++++++++++++++++++++----------------
5 3 files changed, 170 insertions(+), 43 deletions(-)
9 @@ -53,11 +53,18 @@ const struct file_operations ext4_dir_op
11 static unsigned char get_dtype(struct super_block *sb, int filetype)
13 + int fl_index = filetype & EXT4_FT_MASK;
15 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
16 - (filetype >= EXT4_FT_MAX))
17 + (fl_index >= EXT4_FT_MAX))
20 - return (ext4_filetype_table[filetype]);
21 + if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA))
22 + return (ext4_filetype_table[fl_index]);
24 + return (ext4_filetype_table[fl_index]) |
25 + (filetype & EXT4_DIRENT_LUFID);
30 @@ -75,11 +82,11 @@ int __ext4_check_dir_entry(const char *f
31 const int rlen = ext4_rec_len_from_disk(de->rec_len,
32 dir->i_sb->s_blocksize);
34 - if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
35 + if (unlikely(rlen < __EXT4_DIR_REC_LEN(1)))
36 error_msg = "rec_len is smaller than minimal";
37 else if (unlikely(rlen % 4 != 0))
38 error_msg = "rec_len % 4 != 0";
39 - else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
40 + else if (unlikely(rlen < EXT4_DIR_REC_LEN(de)))
41 error_msg = "rec_len is too small for name_len";
42 else if (unlikely(((char *) de - bh->b_data) + rlen >
43 dir->i_sb->s_blocksize))
44 @@ -196,7 +203,7 @@ revalidate:
45 * failure will be detected in the
46 * dirent test below. */
47 if (ext4_rec_len_from_disk(de->rec_len,
48 - sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
49 + sb->s_blocksize) < __EXT4_DIR_REC_LEN(1))
51 i += ext4_rec_len_from_disk(de->rec_len,
53 @@ -359,12 +366,17 @@ int ext4_htree_store_dirent(struct file
54 struct fname *fname, *new_fn;
55 struct dir_private_info *info;
59 info = dir_file->private_data;
60 p = &info->root.rb_node;
62 /* Create and allocate the fname structure */
63 - len = sizeof(struct fname) + dirent->name_len + 1;
64 + if (dirent->file_type & EXT4_DIRENT_LUFID)
65 + extra_data = ext4_get_dirent_data_len(dirent);
67 + len = sizeof(struct fname) + dirent->name_len + extra_data;
69 new_fn = kzalloc(len, GFP_KERNEL);
72 @@ -373,7 +385,7 @@ int ext4_htree_store_dirent(struct file
73 new_fn->inode = le32_to_cpu(dirent->inode);
74 new_fn->name_len = dirent->name_len;
75 new_fn->file_type = dirent->file_type;
76 - memcpy(new_fn->name, dirent->name, dirent->name_len);
77 + memcpy(new_fn->name, dirent->name, dirent->name_len + extra_data);
78 new_fn->name[dirent->name_len] = 0;
83 @@ -1414,7 +1414,9 @@ static inline void ext4_clear_state_flag
84 EXT4_FEATURE_INCOMPAT_EXTENTS| \
85 EXT4_FEATURE_INCOMPAT_64BIT| \
86 EXT4_FEATURE_INCOMPAT_FLEX_BG| \
87 - EXT4_FEATURE_INCOMPAT_MMP)
88 + EXT4_FEATURE_INCOMPAT_MMP| \
89 + EXT4_FEATURE_INCOMPAT_DIRDATA)
91 #define EXT4_FEATURE_RO_COMPAT_SUPP (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
92 EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
93 EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
94 @@ -1501,6 +1503,43 @@ struct ext4_dir_entry_2 {
95 #define EXT4_FT_SYMLINK 7
98 +#define EXT4_FT_MASK 0xf
100 +#if EXT4_FT_MAX > EXT4_FT_MASK
101 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
105 + * d_type has 4 unused bits, so it can hold four types data. these different
106 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
107 + * stored, in flag order, after file-name in ext4 dirent.
110 + * this flag is added to d_type if ext4 dirent has extra data after
111 + * filename. this data length is variable and length is stored in first byte
112 + * of data. data start after filename NUL byte.
113 + * This is used by Lustre FS.
115 +#define EXT4_DIRENT_LUFID 0x10
117 +#define EXT4_LUFID_MAGIC 0xAD200907UL
118 +struct ext4_dentry_param {
119 + __u32 edp_magic; /* EXT4_LUFID_MAGIC */
120 + char edp_len; /* size of edp_data in bytes */
121 + char edp_data[0]; /* packed array of data */
122 +} __attribute__((packed));
124 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
125 + struct ext4_dentry_param* p)
128 + if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA))
130 + if (p && p->edp_magic == EXT4_LUFID_MAGIC)
131 + return &p->edp_len;
137 * EXT4_DIR_PAD defines the directory entries boundaries
138 @@ -1509,8 +1548,11 @@ struct ext4_dir_entry_2 {
140 #define EXT4_DIR_PAD 4
141 #define EXT4_DIR_ROUND (EXT4_DIR_PAD - 1)
142 -#define EXT4_DIR_REC_LEN(name_len) (((name_len) + 8 + EXT4_DIR_ROUND) & \
143 +#define __EXT4_DIR_REC_LEN(name_len) (((name_len) + 8 + EXT4_DIR_ROUND) & \
145 +#define EXT4_DIR_REC_LEN(de) (__EXT4_DIR_REC_LEN(de->name_len +\
146 + ext4_get_dirent_data_len(de)))
148 #define EXT4_MAX_REC_LEN ((1<<16)-1)
151 @@ -1908,7 +1950,7 @@ extern struct buffer_head * ext4_find_en
152 struct ext4_dir_entry_2 ** res_dir);
153 #define ll_ext4_find_entry(inode, dentry, res_dir) ext4_find_entry(inode, &(dentry)->d_name, res_dir)
154 extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
155 - struct inode *inode);
156 + struct inode *inode, const void *, const void *);
157 extern struct buffer_head *ext4_append(handle_t *handle,
159 ext4_lblk_t *block, int *err);
160 @@ -2308,6 +2350,28 @@ static inline void set_bitmap_uptodate(s
161 extern wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
162 extern struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
165 + * Compute the total directory entry data length.
166 + * This includes the filename and an implicit NUL terminator (always present),
167 + * and optional extensions. Each extension has a bit set in the high 4 bits of
168 + * de->file_type, and the extension length is the first byte in each entry.
170 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
172 + char *len = de->name + de->name_len + 1 /* NUL terminator */;
174 + __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
176 + while (extra_data_flags) {
177 + if (extra_data_flags & 1) {
178 + dlen += *len + (dlen == 0);
181 + extra_data_flags >>= 1;
186 #endif /* __KERNEL__ */
189 --- a/fs/ext4/namei.c
190 +++ b/fs/ext4/namei.c
191 @@ -170,7 +170,8 @@ static unsigned dx_get_count(struct dx_e
192 static unsigned dx_get_limit(struct dx_entry *entries);
193 static void dx_set_count(struct dx_entry *entries, unsigned value);
194 static void dx_set_limit(struct dx_entry *entries, unsigned value);
195 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
196 +static inline unsigned dx_root_limit(__u32 blocksize,
197 + struct ext4_dir_entry_2 *dot_de, unsigned infosize);
198 static unsigned dx_node_limit(struct inode *dir);
199 static struct dx_frame *dx_probe(const struct qstr *d_name,
201 @@ -213,11 +214,12 @@ ext4_next_entry(struct ext4_dir_entry_2
203 struct dx_root_info * dx_get_dx_info(struct ext4_dir_entry_2 *de)
205 - /* get dotdot first */
206 - de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
207 + BUG_ON(de->name_len != 1);
208 + /* get dotdot first */
209 + de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
211 - /* dx root info is after dotdot entry */
212 - de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
213 + /* dx root info is after dotdot entry */
214 + de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
216 return (struct dx_root_info *) de;
218 @@ -262,16 +264,23 @@ static inline void dx_set_limit(struct d
219 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
222 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
223 +static inline unsigned dx_root_limit(__u32 blocksize,
224 + struct ext4_dir_entry_2 *dot_de, unsigned infosize)
226 - unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
227 - EXT4_DIR_REC_LEN(2) - infosize;
228 + struct ext4_dir_entry_2 *dotdot_de;
229 + unsigned entry_space;
231 + BUG_ON(dot_de->name_len != 1);
232 + dotdot_de = ext4_next_entry(dot_de, blocksize);
233 + entry_space = blocksize - EXT4_DIR_REC_LEN(dot_de) -
234 + EXT4_DIR_REC_LEN(dotdot_de) - infosize;
236 return entry_space / sizeof(struct dx_entry);
239 static inline unsigned dx_node_limit(struct inode *dir)
241 - unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
242 + unsigned entry_space = dir->i_sb->s_blocksize - __EXT4_DIR_REC_LEN(0);
243 return entry_space / sizeof(struct dx_entry);
246 @@ -318,7 +327,7 @@ static struct stats dx_show_leaf(struct
247 printk(":%x.%u ", h.hash,
248 ((char *) de - base));
250 - space += EXT4_DIR_REC_LEN(de->name_len);
251 + space += EXT4_DIR_REC_LEN(de);
254 de = ext4_next_entry(de, size);
255 @@ -420,7 +429,8 @@ dx_probe(const struct qstr *d_name, stru
257 entries = (struct dx_entry *) (((char *)info) + info->info_length);
259 - if (dx_get_limit(entries) != dx_root_limit(dir,
260 + if (dx_get_limit(entries) != dx_root_limit(dir->i_sb->s_blocksize,
261 + (struct ext4_dir_entry_2*)bh->b_data,
262 info->info_length)) {
263 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
265 @@ -609,7 +619,7 @@ static int htree_dirblock_to_tree(struct
266 de = (struct ext4_dir_entry_2 *) bh->b_data;
267 top = (struct ext4_dir_entry_2 *) ((char *) de +
268 dir->i_sb->s_blocksize -
269 - EXT4_DIR_REC_LEN(0));
270 + __EXT4_DIR_REC_LEN(0));
271 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
272 if (ext4_check_dir_entry(dir, NULL, de, bh,
273 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
274 @@ -1172,7 +1182,7 @@ dx_move_dirents(char *from, char *to, st
276 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
277 (from + (map->offs<<2));
278 - rec_len = EXT4_DIR_REC_LEN(de->name_len);
279 + rec_len = EXT4_DIR_REC_LEN(de);
280 memcpy (to, de, rec_len);
281 ((struct ext4_dir_entry_2 *) to)->rec_len =
282 ext4_rec_len_to_disk(rec_len, blocksize);
283 @@ -1196,7 +1206,7 @@ static struct ext4_dir_entry_2* dx_pack_
284 while ((char*)de < base + blocksize) {
285 next = ext4_next_entry(de, blocksize);
286 if (de->inode && de->name_len) {
287 - rec_len = EXT4_DIR_REC_LEN(de->name_len);
288 + rec_len = EXT4_DIR_REC_LEN(de);
290 memmove(to, de, rec_len);
291 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
292 @@ -1326,10 +1336,16 @@ static int add_dirent_to_buf(handle_t *h
293 unsigned int offset = 0;
294 unsigned int blocksize = dir->i_sb->s_blocksize;
295 unsigned short reclen;
296 - int nlen, rlen, err;
297 + int nlen, rlen, err, dlen = 0;
298 + unsigned char *data;
301 - reclen = EXT4_DIR_REC_LEN(namelen);
302 + data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
305 + dlen = (*data) + 1;
307 + reclen = __EXT4_DIR_REC_LEN(namelen + dlen);
309 de = (struct ext4_dir_entry_2 *)bh->b_data;
310 top = bh->b_data + blocksize - reclen;
311 @@ -1338,7 +1354,7 @@ static int add_dirent_to_buf(handle_t *h
313 if (ext4_match(namelen, name, de))
315 - nlen = EXT4_DIR_REC_LEN(de->name_len);
316 + nlen = EXT4_DIR_REC_LEN(de);
317 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
318 if ((de->inode? rlen - nlen: rlen) >= reclen)
320 @@ -1356,7 +1372,7 @@ static int add_dirent_to_buf(handle_t *h
323 /* By now the buffer is marked for journaling */
324 - nlen = EXT4_DIR_REC_LEN(de->name_len);
325 + nlen = EXT4_DIR_REC_LEN(de);
326 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
328 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
329 @@ -1372,6 +1388,12 @@ static int add_dirent_to_buf(handle_t *h
331 de->name_len = namelen;
332 memcpy(de->name, name, namelen);
334 + de->name[namelen] = 0;
335 + memcpy(&de->name[namelen + 1], data, *(char *) data);
336 + de->file_type |= EXT4_DIRENT_LUFID;
340 * XXX shouldn't update any times until successful
341 * completion of syscall, but too many callers depend
342 @@ -1468,7 +1490,8 @@ static int make_indexed_dir(handle_t *ha
344 dx_set_block(entries, 1);
345 dx_set_count(entries, 1);
346 - dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
347 + dx_set_limit(entries, dx_root_limit(dir->i_sb->s_blocksize,
348 + dot_de, sizeof(*dx_info)));
350 /* Initialize as for dx_probe */
351 hinfo.hash_version = dx_info->hash_version;
352 @@ -1511,6 +1534,8 @@ static int ext4_update_dotdot(handle_t *
353 struct buffer_head * dir_block;
354 struct ext4_dir_entry_2 * de;
355 int len, journal = 0, err = 0;
360 return PTR_ERR(handle);
361 @@ -1526,19 +1551,24 @@ static int ext4_update_dotdot(handle_t *
362 /* the first item must be "." */
363 assert(de->name_len == 1 && de->name[0] == '.');
364 len = le16_to_cpu(de->rec_len);
365 - assert(len >= EXT4_DIR_REC_LEN(1));
366 - if (len > EXT4_DIR_REC_LEN(1)) {
367 + assert(len >= __EXT4_DIR_REC_LEN(1));
368 + if (len > __EXT4_DIR_REC_LEN(1)) {
369 BUFFER_TRACE(dir_block, "get_write_access");
370 err = ext4_journal_get_write_access(handle, dir_block);
375 - de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
376 + de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
379 - len -= EXT4_DIR_REC_LEN(1);
380 - assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
381 + len -= EXT4_DIR_REC_LEN(de);
382 + data = ext4_dentry_get_data(dir->i_sb,
383 + (struct ext4_dentry_param *) dentry->d_fsdata);
386 + assert(len == 0 || len >= __EXT4_DIR_REC_LEN(2 + dlen));
388 de = (struct ext4_dir_entry_2 *)
389 ((char *) de + le16_to_cpu(de->rec_len));
391 @@ -1552,10 +1582,15 @@ static int ext4_update_dotdot(handle_t *
393 de->rec_len = cpu_to_le16(len);
395 - assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
396 + assert(le16_to_cpu(de->rec_len) >= __EXT4_DIR_REC_LEN(2));
398 strcpy (de->name, "..");
399 - ext4_set_de_type(dir->i_sb, de, S_IFDIR);
400 + if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
402 + memcpy(&de->name[2 + 1], data, *data);
403 + ext4_set_de_type(dir->i_sb, de, S_IFDIR);
404 + de->file_type |= EXT4_DIRENT_LUFID;
409 @@ -1994,12 +2029,13 @@ retry:
410 /* Initialize @inode as a subdirectory of @dir, and add the
411 * "." and ".." entries into the first directory block. */
412 int ext4_add_dot_dotdot(handle_t *handle, struct inode * dir,
413 - struct inode *inode)
414 + struct inode *inode,
415 + const void *data1, const void *data2)
417 struct buffer_head *dir_block;
418 struct ext4_dir_entry_2 *de;
419 unsigned int blocksize = dir->i_sb->s_blocksize;
421 + int err = 0, dot_reclen;
424 return PTR_ERR(handle);
425 @@ -2020,17 +2056,32 @@ int ext4_add_dot_dotdot(handle_t *handle
426 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
427 de->inode = cpu_to_le32(inode->i_ino);
429 - de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
431 strcpy(de->name, ".");
432 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
433 + /* get packed fid data */
434 + data1 = ext4_dentry_get_data(dir->i_sb,
435 + (struct ext4_dentry_param *) data1);
438 + memcpy(&de->name[2], data1, *(char *) data1);
439 + de->file_type |= EXT4_DIRENT_LUFID;
441 + de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
442 + dot_reclen = cpu_to_le16(de->rec_len);
443 de = ext4_next_entry(de, blocksize);
444 de->inode = cpu_to_le32(dir->i_ino);
445 - de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
446 + de->rec_len = ext4_rec_len_to_disk(blocksize - dot_reclen,
449 strcpy(de->name, "..");
450 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
451 + data2 = ext4_dentry_get_data(dir->i_sb,
452 + (struct ext4_dentry_param *) data2);
455 + memcpy(&de->name[3], data2, *(char *) data2);
456 + de->file_type |= EXT4_DIRENT_LUFID;
459 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
460 err = ext4_handle_dirty_metadata(handle, inode, dir_block);
461 @@ -2070,7 +2121,7 @@ retry:
465 - err = ext4_add_dot_dotdot(handle, dir, inode);
466 + err = ext4_add_dot_dotdot(handle, dir, inode, NULL, NULL);
468 goto out_clear_inode;
469 err = ext4_add_entry(handle, dentry, inode);
470 @@ -2108,7 +2159,7 @@ static int empty_dir(struct inode *inode
474 - if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
475 + if (inode->i_size < __EXT4_DIR_REC_LEN(1) + __EXT4_DIR_REC_LEN(2) ||
476 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
478 EXT4_ERROR_INODE(inode,