1 this patch implements feature which allows ext4 fs users (e.g. Lustre)
2 to store data in ext4 dirent.
3 data is stored in ext4 dirent after file-name, this space is accounted
4 in de->rec_len. flag EXT4_DIRENT_LUFID added to d_type if extra data
7 make use of dentry->d_fsdata to pass fid to ext4. so no
8 changes in ext4_add_entry() interface required.
10 Index: linux-stage/fs/ext4/dir.c
11 ===================================================================
12 --- linux-stage.orig/fs/ext4/dir.c
13 +++ linux-stage/fs/ext4/dir.c
14 @@ -53,11 +53,18 @@ const struct file_operations ext4_dir_op
16 static unsigned char get_dtype(struct super_block *sb, int filetype)
18 + int fl_index = filetype & EXT4_FT_MASK;
20 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
21 - (filetype >= EXT4_FT_MAX))
22 + (fl_index >= EXT4_FT_MAX))
25 - return (ext4_filetype_table[filetype]);
26 + if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA))
27 + return (ext4_filetype_table[fl_index]);
29 + return (ext4_filetype_table[fl_index]) |
30 + (filetype & EXT4_DIRENT_LUFID);
35 @@ -70,11 +77,11 @@ int ext4_check_dir_entry(const char *fun
36 const int rlen = ext4_rec_len_from_disk(de->rec_len,
37 dir->i_sb->s_blocksize);
39 - if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
40 + if (unlikely(rlen < __EXT4_DIR_REC_LEN(1)))
41 error_msg = "rec_len is smaller than minimal";
42 else if (unlikely(rlen % 4 != 0))
43 error_msg = "rec_len % 4 != 0";
44 - else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
45 + else if (unlikely(rlen < EXT4_DIR_REC_LEN(de)))
46 error_msg = "rec_len is too small for name_len";
47 else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
48 error_msg = "directory entry across blocks";
49 @@ -182,7 +189,7 @@ revalidate:
50 * failure will be detected in the
51 * dirent test below. */
52 if (ext4_rec_len_from_disk(de->rec_len,
53 - sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
54 + sb->s_blocksize) < __EXT4_DIR_REC_LEN(1))
56 i += ext4_rec_len_from_disk(de->rec_len,
58 @@ -345,12 +352,17 @@ int ext4_htree_store_dirent(struct file
59 struct fname *fname, *new_fn;
60 struct dir_private_info *info;
64 info = (struct dir_private_info *) dir_file->private_data;
65 p = &info->root.rb_node;
67 /* Create and allocate the fname structure */
68 - len = sizeof(struct fname) + dirent->name_len + 1;
69 + if (dirent->file_type & EXT4_DIRENT_LUFID)
70 + extra_data = ext4_get_dirent_data_len(dirent);
72 + len = sizeof(struct fname) + dirent->name_len + extra_data;
74 new_fn = kzalloc(len, GFP_KERNEL);
77 @@ -359,7 +371,7 @@ int ext4_htree_store_dirent(struct file
78 new_fn->inode = le32_to_cpu(dirent->inode);
79 new_fn->name_len = dirent->name_len;
80 new_fn->file_type = dirent->file_type;
81 - memcpy(new_fn->name, dirent->name, dirent->name_len);
82 + memcpy(new_fn->name, dirent->name, dirent->name_len + extra_data);
83 new_fn->name[dirent->name_len] = 0;
86 Index: linux-stage/fs/ext4/ext4.h
87 ===================================================================
88 --- linux-stage.orig/fs/ext4/ext4.h
89 +++ linux-stage/fs/ext4/ext4.h
90 @@ -1267,6 +1267,7 @@ EXT4_INODE_BIT_FNS(state, state_flags)
91 #define EXT4_FEATURE_INCOMPAT_64BIT 0x0080
92 #define EXT4_FEATURE_INCOMPAT_MMP 0x0100
93 #define EXT4_FEATURE_INCOMPAT_FLEX_BG 0x0200
94 +#define EXT4_FEATURE_INCOMPAT_DIRDATA 0x1000
96 #define EXT4_FEATURE_COMPAT_SUPP EXT2_FEATURE_COMPAT_EXT_ATTR
97 #define EXT4_FEATURE_INCOMPAT_SUPP (EXT4_FEATURE_INCOMPAT_FILETYPE| \
98 @@ -1275,7 +1276,9 @@ EXT4_INODE_BIT_FNS(state, state_flags)
99 EXT4_FEATURE_INCOMPAT_EXTENTS| \
100 EXT4_FEATURE_INCOMPAT_64BIT| \
101 EXT4_FEATURE_INCOMPAT_FLEX_BG| \
102 - EXT4_FEATURE_INCOMPAT_MMP)
103 + EXT4_FEATURE_INCOMPAT_MMP| \
104 + EXT4_FEATURE_INCOMPAT_DIRDATA)
106 #define EXT4_FEATURE_RO_COMPAT_SUPP (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
107 EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
108 EXT4_FEATURE_RO_COMPAT_GDT_CSUM| \
109 @@ -1357,6 +1360,43 @@ struct ext4_dir_entry_2 {
110 #define EXT4_FT_SYMLINK 7
112 #define EXT4_FT_MAX 8
113 +#define EXT4_FT_MASK 0xf
115 +#if EXT4_FT_MAX > EXT4_FT_MASK
116 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
120 + * d_type has 4 unused bits, so it can hold four types data. these different
121 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
122 + * stored, in flag order, after file-name in ext4 dirent.
125 + * this flag is added to d_type if ext4 dirent has extra data after
126 + * filename. this data length is variable and length is stored in first byte
127 + * of data. data start after filename NUL byte.
128 + * This is used by Lustre FS.
130 +#define EXT4_DIRENT_LUFID 0x10
132 +#define EXT4_LUFID_MAGIC 0xAD200907UL
133 +struct ext4_dentry_param {
134 + __u32 edp_magic; /* EXT4_LUFID_MAGIC */
135 + char edp_len; /* size of edp_data in bytes */
136 + char edp_data[0]; /* packed array of data */
137 +} __attribute__((packed));
139 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
140 + struct ext4_dentry_param* p)
143 + if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA))
145 + if (p && p->edp_magic == EXT4_LUFID_MAGIC)
146 + return &p->edp_len;
152 * EXT4_DIR_PAD defines the directory entries boundaries
153 @@ -1365,8 +1405,11 @@ struct ext4_dir_entry_2 {
155 #define EXT4_DIR_PAD 4
156 #define EXT4_DIR_ROUND (EXT4_DIR_PAD - 1)
157 -#define EXT4_DIR_REC_LEN(name_len) (((name_len) + 8 + EXT4_DIR_ROUND) & \
158 +#define __EXT4_DIR_REC_LEN(name_len) (((name_len) + 8 + EXT4_DIR_ROUND) & \
160 +#define EXT4_DIR_REC_LEN(de) (__EXT4_DIR_REC_LEN(de->name_len +\
161 + ext4_get_dirent_data_len(de)))
163 #define EXT4_MAX_REC_LEN ((1<<16)-1)
165 static inline unsigned int
166 @@ -1726,7 +1769,7 @@ extern struct buffer_head * ext4_find_en
167 struct ext4_dir_entry_2 ** res_dir);
168 #define ll_ext4_find_entry(inode, dentry, res_dir) ext4_find_entry(inode, &(dentry)->d_name, res_dir)
169 extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
170 - struct inode *inode);
171 + struct inode *inode, const void *, const void *);
172 extern struct buffer_head *ext4_append(handle_t *handle,
174 ext4_lblk_t *block, int *err);
175 @@ -2069,6 +2112,28 @@ extern wait_queue_head_t aio_wq[];
176 #define to_aio_wq(v) (&aio_wq[((unsigned long)v) % WQ_HASH_SZ])
177 extern void ext4_aio_wait(struct inode *inode);
180 + * Compute the total directory entry data length.
181 + * This includes the filename and an implicit NUL terminator (always present),
182 + * and optional extensions. Each extension has a bit set in the high 4 bits of
183 + * de->file_type, and the extension length is the first byte in each entry.
185 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
187 + char *len = de->name + de->name_len + 1 /* NUL terminator */;
189 + __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
191 + while (extra_data_flags) {
192 + if (extra_data_flags & 1) {
193 + dlen += *len + (dlen == 0);
196 + extra_data_flags >>= 1;
201 #endif /* __KERNEL__ */
204 Index: linux-stage/fs/ext4/namei.c
205 ===================================================================
206 --- linux-stage.orig/fs/ext4/namei.c
207 +++ linux-stage/fs/ext4/namei.c
208 @@ -169,7 +169,8 @@ static unsigned dx_get_count(struct dx_e
209 static unsigned dx_get_limit(struct dx_entry *entries);
210 static void dx_set_count(struct dx_entry *entries, unsigned value);
211 static void dx_set_limit(struct dx_entry *entries, unsigned value);
212 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
213 +static inline unsigned dx_root_limit(__u32 blocksize,
214 + struct ext4_dir_entry_2 *dot_de, unsigned infosize);
215 static unsigned dx_node_limit(struct inode *dir);
216 static struct dx_frame *dx_probe(const struct qstr *d_name,
218 @@ -212,11 +213,12 @@ ext4_next_entry(struct ext4_dir_entry_2
220 struct dx_root_info * dx_get_dx_info(struct ext4_dir_entry_2 *de)
222 - /* get dotdot first */
223 - de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
224 + BUG_ON(de->name_len != 1);
225 + /* get dotdot first */
226 + de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
228 - /* dx root info is after dotdot entry */
229 - de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
230 + /* dx root info is after dotdot entry */
231 + de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
233 return (struct dx_root_info *) de;
235 @@ -261,16 +263,23 @@ static inline void dx_set_limit(struct d
236 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
239 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
240 +static inline unsigned dx_root_limit(__u32 blocksize,
241 + struct ext4_dir_entry_2 *dot_de, unsigned infosize)
243 - unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
244 - EXT4_DIR_REC_LEN(2) - infosize;
245 + struct ext4_dir_entry_2 *dotdot_de;
246 + unsigned entry_space;
248 + BUG_ON(dot_de->name_len != 1);
249 + dotdot_de = ext4_next_entry(dot_de, blocksize);
250 + entry_space = blocksize - EXT4_DIR_REC_LEN(dot_de) -
251 + EXT4_DIR_REC_LEN(dotdot_de) - infosize;
253 return entry_space / sizeof(struct dx_entry);
256 static inline unsigned dx_node_limit(struct inode *dir)
258 - unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
259 + unsigned entry_space = dir->i_sb->s_blocksize - __EXT4_DIR_REC_LEN(0);
260 return entry_space / sizeof(struct dx_entry);
263 @@ -317,7 +326,7 @@ static struct stats dx_show_leaf(struct
264 printk(":%x.%u ", h.hash,
265 ((char *) de - base));
267 - space += EXT4_DIR_REC_LEN(de->name_len);
268 + space += EXT4_DIR_REC_LEN(de);
271 de = ext4_next_entry(de, size);
272 @@ -419,7 +428,8 @@ dx_probe(const struct qstr *d_name, stru
274 entries = (struct dx_entry *) (((char *)info) + info->info_length);
276 - if (dx_get_limit(entries) != dx_root_limit(dir,
277 + if (dx_get_limit(entries) != dx_root_limit(dir->i_sb->s_blocksize,
278 + (struct ext4_dir_entry_2*)bh->b_data,
279 info->info_length)) {
280 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
282 @@ -608,7 +618,7 @@ static int htree_dirblock_to_tree(struct
283 de = (struct ext4_dir_entry_2 *) bh->b_data;
284 top = (struct ext4_dir_entry_2 *) ((char *) de +
285 dir->i_sb->s_blocksize -
286 - EXT4_DIR_REC_LEN(0));
287 + __EXT4_DIR_REC_LEN(0));
288 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
289 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
290 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
291 @@ -1020,7 +1030,7 @@ static struct buffer_head * ext4_dx_find
293 de = (struct ext4_dir_entry_2 *) bh->b_data;
294 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
295 - EXT4_DIR_REC_LEN(0));
296 + __EXT4_DIR_REC_LEN(0));
297 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
298 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
299 + ((char *) de - bh->b_data);
300 @@ -1181,7 +1191,7 @@ dx_move_dirents(char *from, char *to, st
302 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
303 (from + (map->offs<<2));
304 - rec_len = EXT4_DIR_REC_LEN(de->name_len);
305 + rec_len = EXT4_DIR_REC_LEN(de);
306 memcpy (to, de, rec_len);
307 ((struct ext4_dir_entry_2 *) to)->rec_len =
308 ext4_rec_len_to_disk(rec_len, blocksize);
309 @@ -1205,7 +1215,7 @@ static struct ext4_dir_entry_2* dx_pack_
310 while ((char*)de < base + blocksize) {
311 next = ext4_next_entry(de, blocksize);
312 if (de->inode && de->name_len) {
313 - rec_len = EXT4_DIR_REC_LEN(de->name_len);
314 + rec_len = EXT4_DIR_REC_LEN(de);
316 memmove(to, de, rec_len);
317 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
318 @@ -1335,10 +1345,16 @@ static int add_dirent_to_buf(handle_t *h
319 unsigned int offset = 0;
320 unsigned int blocksize = dir->i_sb->s_blocksize;
321 unsigned short reclen;
322 - int nlen, rlen, err;
323 + int nlen, rlen, err, dlen = 0;
324 + unsigned char *data;
327 - reclen = EXT4_DIR_REC_LEN(namelen);
328 + data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
331 + dlen = (*data) + 1;
333 + reclen = __EXT4_DIR_REC_LEN(namelen + dlen);
335 de = (struct ext4_dir_entry_2 *)bh->b_data;
336 top = bh->b_data + blocksize - reclen;
337 @@ -1348,7 +1364,7 @@ static int add_dirent_to_buf(handle_t *h
339 if (ext4_match(namelen, name, de))
341 - nlen = EXT4_DIR_REC_LEN(de->name_len);
342 + nlen = EXT4_DIR_REC_LEN(de);
343 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
344 if ((de->inode? rlen - nlen: rlen) >= reclen)
346 @@ -1366,7 +1382,7 @@ static int add_dirent_to_buf(handle_t *h
349 /* By now the buffer is marked for journaling */
350 - nlen = EXT4_DIR_REC_LEN(de->name_len);
351 + nlen = EXT4_DIR_REC_LEN(de);
352 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
354 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
355 @@ -1382,6 +1398,12 @@ static int add_dirent_to_buf(handle_t *h
357 de->name_len = namelen;
358 memcpy(de->name, name, namelen);
360 + de->name[namelen] = 0;
361 + memcpy(&de->name[namelen + 1], data, *(char *) data);
362 + de->file_type |= EXT4_DIRENT_LUFID;
366 * XXX shouldn't update any times until successful
367 * completion of syscall, but too many callers depend
368 @@ -1480,7 +1502,8 @@ static int make_indexed_dir(handle_t *ha
370 dx_set_block(entries, 1);
371 dx_set_count(entries, 1);
372 - dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
373 + dx_set_limit(entries, dx_root_limit(dir->i_sb->s_blocksize,
374 + dot_de, sizeof(*dx_info)));
376 /* Initialize as for dx_probe */
377 hinfo.hash_version = dx_info->hash_version;
378 @@ -1511,6 +1534,8 @@ static int ext4_update_dotdot(handle_t *
379 struct buffer_head * dir_block;
380 struct ext4_dir_entry_2 * de;
381 int len, journal = 0, err = 0;
386 return PTR_ERR(handle);
387 @@ -1526,19 +1551,24 @@ static int ext4_update_dotdot(handle_t *
388 /* the first item must be "." */
389 assert(de->name_len == 1 && de->name[0] == '.');
390 len = le16_to_cpu(de->rec_len);
391 - assert(len >= EXT4_DIR_REC_LEN(1));
392 - if (len > EXT4_DIR_REC_LEN(1)) {
393 + assert(len >= __EXT4_DIR_REC_LEN(1));
394 + if (len > __EXT4_DIR_REC_LEN(1)) {
395 BUFFER_TRACE(dir_block, "get_write_access");
396 err = ext4_journal_get_write_access(handle, dir_block);
401 - de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
402 + de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
405 - len -= EXT4_DIR_REC_LEN(1);
406 - assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
407 + len -= EXT4_DIR_REC_LEN(de);
408 + data = ext4_dentry_get_data(dir->i_sb,
409 + (struct ext4_dentry_param *) dentry->d_fsdata);
412 + assert(len == 0 || len >= __EXT4_DIR_REC_LEN(2 + dlen));
414 de = (struct ext4_dir_entry_2 *)
415 ((char *) de + le16_to_cpu(de->rec_len));
417 @@ -1552,10 +1582,15 @@ static int ext4_update_dotdot(handle_t *
419 de->rec_len = cpu_to_le16(len);
421 - assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
422 + assert(le16_to_cpu(de->rec_len) >= __EXT4_DIR_REC_LEN(2));
424 strcpy (de->name, "..");
425 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
428 + memcpy(&de->name[2 + 1], data, dlen);
429 + de->file_type |= EXT4_DIRENT_LUFID;
434 @@ -1978,12 +2013,13 @@ retry:
435 /* Initialize @inode as a subdirectory of @dir, and add the
436 * "." and ".." entries into the first directory block. */
437 int ext4_add_dot_dotdot(handle_t *handle, struct inode * dir,
438 - struct inode *inode)
439 + struct inode *inode,
440 + const void *data1, const void *data2)
442 struct buffer_head * dir_block;
443 struct ext4_dir_entry_2 * de;
444 unsigned int blocksize = dir->i_sb->s_blocksize;
446 + int err = 0, dot_reclen;
449 return PTR_ERR(handle);
450 @@ -1996,28 +2032,40 @@ int ext4_add_dot_dotdot(handle_t *handle
451 inode->i_fop = &ext4_dir_operations;
452 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
453 dir_block = ext4_bread(handle, inode, 0, 1, &err);
455 - clear_nlink(inode);
456 - ext4_mark_inode_dirty(handle, inode);
462 BUFFER_TRACE(dir_block, "get_write_access");
463 ext4_journal_get_write_access(handle, dir_block);
464 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
465 de->inode = cpu_to_le32(inode->i_ino);
467 - de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
469 strcpy(de->name, ".");
470 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
471 + /* get packed fid data*/
472 + data1 = ext4_dentry_get_data(dir->i_sb,
473 + (struct ext4_dentry_param *) data1);
476 + memcpy(&de->name[2], data1, *(char *) data1);
477 + de->file_type |= EXT4_DIRENT_LUFID;
479 + de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
480 + dot_reclen = cpu_to_le16(de->rec_len);
481 de = ext4_next_entry(de, blocksize);
482 de->inode = cpu_to_le32(dir->i_ino);
483 - de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
484 + de->rec_len = ext4_rec_len_to_disk(blocksize - dot_reclen,
487 strcpy(de->name, "..");
488 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
489 + data2 = ext4_dentry_get_data(dir->i_sb,
490 + (struct ext4_dentry_param *) data2);
493 + memcpy(&de->name[3], data2, *(char *) data2);
494 + de->file_type |= EXT4_DIRENT_LUFID;
497 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
498 ext4_handle_dirty_metadata(handle, dir, dir_block);
499 @@ -2054,9 +2102,12 @@ retry:
503 - err = ext4_add_dot_dotdot(handle, dir, inode);
504 + err = ext4_add_dot_dotdot(handle, dir, inode, NULL, NULL);
506 + clear_nlink(inode);
507 unlock_new_inode(inode);
508 + ext4_mark_inode_dirty(handle, inode);
513 @@ -2092,7 +2143,7 @@ static int empty_dir(struct inode *inode
517 - if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
518 + if (inode->i_size < __EXT4_DIR_REC_LEN(1) + __EXT4_DIR_REC_LEN(2) ||
519 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
521 ext4_error(inode->i_sb,