1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #ifndef __XFS_DA_FORMAT_H__
8 #define __XFS_DA_FORMAT_H__
9 
10 /*
11  * This structure is common to both leaf nodes and non-leaf nodes in the Btree.
12  *
13  * It is used to manage a doubly linked list of all blocks at the same
14  * level in the Btree, and to identify which type of block this is.
15  */
16 #define XFS_DA_NODE_MAGIC	0xfebe	/* magic number: non-leaf blocks */
17 #define XFS_ATTR_LEAF_MAGIC	0xfbee	/* magic number: attribute leaf blks */
18 #define XFS_DIR2_LEAF1_MAGIC	0xd2f1	/* magic number: v2 dirlf single blks */
19 #define XFS_DIR2_LEAFN_MAGIC	0xd2ff	/* magic number: v2 dirlf multi blks */
20 
21 typedef struct xfs_da_blkinfo {
22 	__be32		forw;			/* previous block in list */
23 	__be32		back;			/* following block in list */
24 	__be16		magic;			/* validity check on block */
25 	__be16		pad;			/* unused */
26 } xfs_da_blkinfo_t;
27 
28 /*
29  * CRC enabled directory structure types
30  *
31  * The headers change size for the additional verification information, but
32  * otherwise the tree layouts and contents are unchanged. Hence the da btree
33  * code can use the struct xfs_da_blkinfo for manipulating the tree links and
34  * magic numbers without modification for both v2 and v3 nodes.
35  */
36 #define XFS_DA3_NODE_MAGIC	0x3ebe	/* magic number: non-leaf blocks */
37 #define XFS_ATTR3_LEAF_MAGIC	0x3bee	/* magic number: attribute leaf blks */
38 #define XFS_DIR3_LEAF1_MAGIC	0x3df1	/* magic number: v3 dirlf single blks */
39 #define XFS_DIR3_LEAFN_MAGIC	0x3dff	/* magic number: v3 dirlf multi blks */
40 
41 struct xfs_da3_blkinfo {
42 	/*
43 	 * the node link manipulation code relies on the fact that the first
44 	 * element of this structure is the struct xfs_da_blkinfo so it can
45 	 * ignore the differences in the rest of the structures.
46 	 */
47 	struct xfs_da_blkinfo	hdr;
48 	__be32			crc;	/* CRC of block */
49 	__be64			blkno;	/* first block of the buffer */
50 	__be64			lsn;	/* sequence number of last write */
51 	uuid_t			uuid;	/* filesystem we belong to */
52 	__be64			owner;	/* inode that owns the block */
53 };
54 
55 /*
56  * This is the structure of the root and intermediate nodes in the Btree.
57  * The leaf nodes are defined above.
58  *
59  * Entries are not packed.
60  *
61  * Since we have duplicate keys, use a binary search but always follow
62  * all match in the block, not just the first match found.
63  */
64 #define XFS_DA_NODE_MAXDEPTH	5	/* max depth of Btree */
65 
66 typedef struct xfs_da_node_hdr {
67 	struct xfs_da_blkinfo	info;	/* block type, links, etc. */
68 	__be16			__count; /* count of active entries */
69 	__be16			__level; /* level above leaves (leaf == 0) */
70 } xfs_da_node_hdr_t;
71 
72 struct xfs_da3_node_hdr {
73 	struct xfs_da3_blkinfo	info;	/* block type, links, etc. */
74 	__be16			__count; /* count of active entries */
75 	__be16			__level; /* level above leaves (leaf == 0) */
76 	__be32			__pad32;
77 };
78 
79 #define XFS_DA3_NODE_CRC_OFF	(offsetof(struct xfs_da3_node_hdr, info.crc))
80 
81 typedef struct xfs_da_node_entry {
82 	__be32	hashval;	/* hash value for this descendant */
83 	__be32	before;		/* Btree block before this key */
84 } xfs_da_node_entry_t;
85 
86 typedef struct xfs_da_intnode {
87 	struct xfs_da_node_hdr	hdr;
88 	struct xfs_da_node_entry __btree[];
89 } xfs_da_intnode_t;
90 
91 struct xfs_da3_intnode {
92 	struct xfs_da3_node_hdr	hdr;
93 	struct xfs_da_node_entry __btree[];
94 };
95 
96 /*
97  * Directory version 2.
98  *
99  * There are 4 possible formats:
100  *  - shortform - embedded into the inode
101  *  - single block - data with embedded leaf at the end
102  *  - multiple data blocks, single leaf+freeindex block
103  *  - data blocks, node and leaf blocks (btree), freeindex blocks
104  *
105  * Note: many node blocks structures and constants are shared with the attr
106  * code and defined in xfs_da_btree.h.
107  */
108 
109 #define	XFS_DIR2_BLOCK_MAGIC	0x58443242	/* XD2B: single block dirs */
110 #define	XFS_DIR2_DATA_MAGIC	0x58443244	/* XD2D: multiblock dirs */
111 #define	XFS_DIR2_FREE_MAGIC	0x58443246	/* XD2F: free index blocks */
112 
113 /*
114  * Directory Version 3 With CRCs.
115  *
116  * The tree formats are the same as for version 2 directories.  The difference
117  * is in the block header and dirent formats. In many cases the v3 structures
118  * use v2 definitions as they are no different and this makes code sharing much
119  * easier.
120  *
121  * Also, the xfs_dir3_*() functions handle both v2 and v3 formats - if the
122  * format is v2 then they switch to the existing v2 code, or the format is v3
123  * they implement the v3 functionality. This means the existing dir2 is a mix of
124  * xfs_dir2/xfs_dir3 calls and functions. The xfs_dir3 functions are called
125  * where there is a difference in the formats, otherwise the code is unchanged.
126  *
127  * Where it is possible, the code decides what to do based on the magic numbers
128  * in the blocks rather than feature bits in the superblock. This means the code
129  * is as independent of the external XFS code as possible as doesn't require
130  * passing struct xfs_mount pointers into places where it isn't really
131  * necessary.
132  *
133  * Version 3 includes:
134  *
135  *	- a larger block header for CRC and identification purposes and so the
136  *	offsets of all the structures inside the blocks are different.
137  *
138  *	- new magic numbers to be able to detect the v2/v3 types on the fly.
139  */
140 
141 #define	XFS_DIR3_BLOCK_MAGIC	0x58444233	/* XDB3: single block dirs */
142 #define	XFS_DIR3_DATA_MAGIC	0x58444433	/* XDD3: multiblock dirs */
143 #define	XFS_DIR3_FREE_MAGIC	0x58444633	/* XDF3: free index blocks */
144 
145 /*
146  * Dirents in version 3 directories have a file type field. Additions to this
147  * list are an on-disk format change, requiring feature bits. Valid values
148  * are as follows:
149  */
150 #define XFS_DIR3_FT_UNKNOWN		0
151 #define XFS_DIR3_FT_REG_FILE		1
152 #define XFS_DIR3_FT_DIR			2
153 #define XFS_DIR3_FT_CHRDEV		3
154 #define XFS_DIR3_FT_BLKDEV		4
155 #define XFS_DIR3_FT_FIFO		5
156 #define XFS_DIR3_FT_SOCK		6
157 #define XFS_DIR3_FT_SYMLINK		7
158 #define XFS_DIR3_FT_WHT			8
159 
160 #define XFS_DIR3_FT_MAX			9
161 
162 /*
163  * Byte offset in data block and shortform entry.
164  */
165 typedef uint16_t	xfs_dir2_data_off_t;
166 #define	NULLDATAOFF	0xffffU
167 typedef uint		xfs_dir2_data_aoff_t;	/* argument form */
168 
169 /*
170  * Offset in data space of a data entry.
171  */
172 typedef uint32_t	xfs_dir2_dataptr_t;
173 #define	XFS_DIR2_MAX_DATAPTR	((xfs_dir2_dataptr_t)0xffffffff)
174 #define	XFS_DIR2_NULL_DATAPTR	((xfs_dir2_dataptr_t)0)
175 
176 /*
177  * Byte offset in a directory.
178  */
179 typedef	xfs_off_t	xfs_dir2_off_t;
180 
181 /*
182  * Directory block number (logical dirblk in file)
183  */
184 typedef uint32_t	xfs_dir2_db_t;
185 
186 #define XFS_INO32_SIZE	4
187 #define XFS_INO64_SIZE	8
188 #define XFS_INO64_DIFF	(XFS_INO64_SIZE - XFS_INO32_SIZE)
189 
190 #define	XFS_DIR2_MAX_SHORT_INUM	((xfs_ino_t)0xffffffffULL)
191 
192 /*
193  * Directory layout when stored internal to an inode.
194  *
195  * Small directories are packed as tightly as possible so as to fit into the
196  * literal area of the inode.  These "shortform" directories consist of a
197  * single xfs_dir2_sf_hdr header followed by zero or more xfs_dir2_sf_entry
198  * structures.  Due the different inode number storage size and the variable
199  * length name field in the xfs_dir2_sf_entry all these structure are
200  * variable length, and the accessors in this file should be used to iterate
201  * over them.
202  */
203 typedef struct xfs_dir2_sf_hdr {
204 	uint8_t			count;		/* count of entries */
205 	uint8_t			i8count;	/* count of 8-byte inode #s */
206 	uint8_t			parent[8];	/* parent dir inode number */
207 } __packed xfs_dir2_sf_hdr_t;
208 
209 typedef struct xfs_dir2_sf_entry {
210 	__u8			namelen;	/* actual name length */
211 	__u8			offset[2];	/* saved offset */
212 	__u8			name[];		/* name, variable size */
213 	/*
214 	 * A single byte containing the file type field follows the inode
215 	 * number for version 3 directory entries.
216 	 *
217 	 * A 64-bit or 32-bit inode number follows here, at a variable offset
218 	 * after the name.
219 	 */
220 } __packed xfs_dir2_sf_entry_t;
221 
xfs_dir2_sf_hdr_size(int i8count)222 static inline int xfs_dir2_sf_hdr_size(int i8count)
223 {
224 	return sizeof(struct xfs_dir2_sf_hdr) -
225 		(i8count == 0) * XFS_INO64_DIFF;
226 }
227 
228 static inline xfs_dir2_data_aoff_t
xfs_dir2_sf_get_offset(xfs_dir2_sf_entry_t * sfep)229 xfs_dir2_sf_get_offset(xfs_dir2_sf_entry_t *sfep)
230 {
231 	return get_unaligned_be16(sfep->offset);
232 }
233 
234 static inline void
xfs_dir2_sf_put_offset(xfs_dir2_sf_entry_t * sfep,xfs_dir2_data_aoff_t off)235 xfs_dir2_sf_put_offset(xfs_dir2_sf_entry_t *sfep, xfs_dir2_data_aoff_t off)
236 {
237 	put_unaligned_be16(off, sfep->offset);
238 }
239 
240 static inline struct xfs_dir2_sf_entry *
xfs_dir2_sf_firstentry(struct xfs_dir2_sf_hdr * hdr)241 xfs_dir2_sf_firstentry(struct xfs_dir2_sf_hdr *hdr)
242 {
243 	return (struct xfs_dir2_sf_entry *)
244 		((char *)hdr + xfs_dir2_sf_hdr_size(hdr->i8count));
245 }
246 
247 /*
248  * Data block structures.
249  *
250  * A pure data block looks like the following drawing on disk:
251  *
252  *    +-------------------------------------------------+
253  *    | xfs_dir2_data_hdr_t                             |
254  *    +-------------------------------------------------+
255  *    | xfs_dir2_data_entry_t OR xfs_dir2_data_unused_t |
256  *    | xfs_dir2_data_entry_t OR xfs_dir2_data_unused_t |
257  *    | xfs_dir2_data_entry_t OR xfs_dir2_data_unused_t |
258  *    | ...                                             |
259  *    +-------------------------------------------------+
260  *    | unused space                                    |
261  *    +-------------------------------------------------+
262  *
263  * As all the entries are variable size structures the accessors below should
264  * be used to iterate over them.
265  *
266  * In addition to the pure data blocks for the data and node formats,
267  * most structures are also used for the combined data/freespace "block"
268  * format below.
269  */
270 
271 #define	XFS_DIR2_DATA_ALIGN_LOG	3		/* i.e., 8 bytes */
272 #define	XFS_DIR2_DATA_ALIGN	(1 << XFS_DIR2_DATA_ALIGN_LOG)
273 #define	XFS_DIR2_DATA_FREE_TAG	0xffff
274 #define	XFS_DIR2_DATA_FD_COUNT	3
275 
276 /*
277  * Directory address space divided into sections,
278  * spaces separated by 32GB.
279  */
280 #define	XFS_DIR2_MAX_SPACES	3
281 #define	XFS_DIR2_SPACE_SIZE	(1ULL << (32 + XFS_DIR2_DATA_ALIGN_LOG))
282 #define	XFS_DIR2_DATA_SPACE	0
283 #define	XFS_DIR2_DATA_OFFSET	(XFS_DIR2_DATA_SPACE * XFS_DIR2_SPACE_SIZE)
284 
285 /*
286  * Describe a free area in the data block.
287  *
288  * The freespace will be formatted as a xfs_dir2_data_unused_t.
289  */
290 typedef struct xfs_dir2_data_free {
291 	__be16			offset;		/* start of freespace */
292 	__be16			length;		/* length of freespace */
293 } xfs_dir2_data_free_t;
294 
295 /*
296  * Header for the data blocks.
297  *
298  * The code knows that XFS_DIR2_DATA_FD_COUNT is 3.
299  */
300 typedef struct xfs_dir2_data_hdr {
301 	__be32			magic;		/* XFS_DIR2_DATA_MAGIC or */
302 						/* XFS_DIR2_BLOCK_MAGIC */
303 	xfs_dir2_data_free_t	bestfree[XFS_DIR2_DATA_FD_COUNT];
304 } xfs_dir2_data_hdr_t;
305 
306 /*
307  * define a structure for all the verification fields we are adding to the
308  * directory block structures. This will be used in several structures.
309  * The magic number must be the first entry to align with all the dir2
310  * structures so we determine how to decode them just by the magic number.
311  */
312 struct xfs_dir3_blk_hdr {
313 	__be32			magic;	/* magic number */
314 	__be32			crc;	/* CRC of block */
315 	__be64			blkno;	/* first block of the buffer */
316 	__be64			lsn;	/* sequence number of last write */
317 	uuid_t			uuid;	/* filesystem we belong to */
318 	__be64			owner;	/* inode that owns the block */
319 };
320 
321 struct xfs_dir3_data_hdr {
322 	struct xfs_dir3_blk_hdr	hdr;
323 	xfs_dir2_data_free_t	best_free[XFS_DIR2_DATA_FD_COUNT];
324 	__be32			pad;	/* 64 bit alignment */
325 };
326 
327 #define XFS_DIR3_DATA_CRC_OFF  offsetof(struct xfs_dir3_data_hdr, hdr.crc)
328 
329 /*
330  * Active entry in a data block.
331  *
332  * Aligned to 8 bytes.  After the variable length name field there is a
333  * 2 byte tag field, which can be accessed using xfs_dir3_data_entry_tag_p.
334  *
335  * For dir3 structures, there is file type field between the name and the tag.
336  * This can only be manipulated by helper functions. It is packed hard against
337  * the end of the name so any padding for rounding is between the file type and
338  * the tag.
339  */
340 typedef struct xfs_dir2_data_entry {
341 	__be64			inumber;	/* inode number */
342 	__u8			namelen;	/* name length */
343 	__u8			name[];		/* name bytes, no null */
344      /* __u8			filetype; */	/* type of inode we point to */
345      /*	__be16                  tag; */		/* starting offset of us */
346 } xfs_dir2_data_entry_t;
347 
348 /*
349  * Unused entry in a data block.
350  *
351  * Aligned to 8 bytes.  Tag appears as the last 2 bytes and must be accessed
352  * using xfs_dir2_data_unused_tag_p.
353  */
354 typedef struct xfs_dir2_data_unused {
355 	__be16			freetag;	/* XFS_DIR2_DATA_FREE_TAG */
356 	__be16			length;		/* total free length */
357 						/* variable offset */
358 	__be16			tag;		/* starting offset of us */
359 } xfs_dir2_data_unused_t;
360 
361 /*
362  * Pointer to a freespace's tag word.
363  */
364 static inline __be16 *
xfs_dir2_data_unused_tag_p(struct xfs_dir2_data_unused * dup)365 xfs_dir2_data_unused_tag_p(struct xfs_dir2_data_unused *dup)
366 {
367 	return (__be16 *)((char *)dup +
368 			be16_to_cpu(dup->length) - sizeof(__be16));
369 }
370 
371 /*
372  * Leaf block structures.
373  *
374  * A pure leaf block looks like the following drawing on disk:
375  *
376  *    +---------------------------+
377  *    | xfs_dir2_leaf_hdr_t       |
378  *    +---------------------------+
379  *    | xfs_dir2_leaf_entry_t     |
380  *    | xfs_dir2_leaf_entry_t     |
381  *    | xfs_dir2_leaf_entry_t     |
382  *    | xfs_dir2_leaf_entry_t     |
383  *    | ...                       |
384  *    +---------------------------+
385  *    | xfs_dir2_data_off_t       |
386  *    | xfs_dir2_data_off_t       |
387  *    | xfs_dir2_data_off_t       |
388  *    | ...                       |
389  *    +---------------------------+
390  *    | xfs_dir2_leaf_tail_t      |
391  *    +---------------------------+
392  *
393  * The xfs_dir2_data_off_t members (bests) and tail are at the end of the block
394  * for single-leaf (magic = XFS_DIR2_LEAF1_MAGIC) blocks only, but not present
395  * for directories with separate leaf nodes and free space blocks
396  * (magic = XFS_DIR2_LEAFN_MAGIC).
397  *
398  * As all the entries are variable size structures the accessors below should
399  * be used to iterate over them.
400  */
401 
402 /*
403  * Offset of the leaf/node space.  First block in this space
404  * is the btree root.
405  */
406 #define	XFS_DIR2_LEAF_SPACE	1
407 #define	XFS_DIR2_LEAF_OFFSET	(XFS_DIR2_LEAF_SPACE * XFS_DIR2_SPACE_SIZE)
408 
409 /*
410  * Leaf block header.
411  */
412 typedef struct xfs_dir2_leaf_hdr {
413 	xfs_da_blkinfo_t	info;		/* header for da routines */
414 	__be16			count;		/* count of entries */
415 	__be16			stale;		/* count of stale entries */
416 } xfs_dir2_leaf_hdr_t;
417 
418 struct xfs_dir3_leaf_hdr {
419 	struct xfs_da3_blkinfo	info;		/* header for da routines */
420 	__be16			count;		/* count of entries */
421 	__be16			stale;		/* count of stale entries */
422 	__be32			pad;		/* 64 bit alignment */
423 };
424 
425 /*
426  * Leaf block entry.
427  */
428 typedef struct xfs_dir2_leaf_entry {
429 	__be32			hashval;	/* hash value of name */
430 	__be32			address;	/* address of data entry */
431 } xfs_dir2_leaf_entry_t;
432 
433 /*
434  * Leaf block tail.
435  */
436 typedef struct xfs_dir2_leaf_tail {
437 	__be32			bestcount;
438 } xfs_dir2_leaf_tail_t;
439 
440 /*
441  * Leaf block.
442  */
443 typedef struct xfs_dir2_leaf {
444 	xfs_dir2_leaf_hdr_t	hdr;			/* leaf header */
445 	xfs_dir2_leaf_entry_t	__ents[];		/* entries */
446 } xfs_dir2_leaf_t;
447 
448 struct xfs_dir3_leaf {
449 	struct xfs_dir3_leaf_hdr	hdr;		/* leaf header */
450 	struct xfs_dir2_leaf_entry	__ents[];	/* entries */
451 };
452 
453 #define XFS_DIR3_LEAF_CRC_OFF  offsetof(struct xfs_dir3_leaf_hdr, info.crc)
454 
455 /*
456  * Get address of the bests array in the single-leaf block.
457  */
458 static inline __be16 *
xfs_dir2_leaf_bests_p(struct xfs_dir2_leaf_tail * ltp)459 xfs_dir2_leaf_bests_p(struct xfs_dir2_leaf_tail *ltp)
460 {
461 	return (__be16 *)ltp - be32_to_cpu(ltp->bestcount);
462 }
463 
464 /*
465  * Free space block definitions for the node format.
466  */
467 
468 /*
469  * Offset of the freespace index.
470  */
471 #define	XFS_DIR2_FREE_SPACE	2
472 #define	XFS_DIR2_FREE_OFFSET	(XFS_DIR2_FREE_SPACE * XFS_DIR2_SPACE_SIZE)
473 
474 typedef	struct xfs_dir2_free_hdr {
475 	__be32			magic;		/* XFS_DIR2_FREE_MAGIC */
476 	__be32			firstdb;	/* db of first entry */
477 	__be32			nvalid;		/* count of valid entries */
478 	__be32			nused;		/* count of used entries */
479 } xfs_dir2_free_hdr_t;
480 
481 typedef struct xfs_dir2_free {
482 	xfs_dir2_free_hdr_t	hdr;		/* block header */
483 	__be16			bests[];	/* best free counts */
484 						/* unused entries are -1 */
485 } xfs_dir2_free_t;
486 
487 struct xfs_dir3_free_hdr {
488 	struct xfs_dir3_blk_hdr	hdr;
489 	__be32			firstdb;	/* db of first entry */
490 	__be32			nvalid;		/* count of valid entries */
491 	__be32			nused;		/* count of used entries */
492 	__be32			pad;		/* 64 bit alignment */
493 };
494 
495 struct xfs_dir3_free {
496 	struct xfs_dir3_free_hdr hdr;
497 	__be16			bests[];	/* best free counts */
498 						/* unused entries are -1 */
499 };
500 
501 #define XFS_DIR3_FREE_CRC_OFF  offsetof(struct xfs_dir3_free, hdr.hdr.crc)
502 
503 /*
504  * Single block format.
505  *
506  * The single block format looks like the following drawing on disk:
507  *
508  *    +-------------------------------------------------+
509  *    | xfs_dir2_data_hdr_t                             |
510  *    +-------------------------------------------------+
511  *    | xfs_dir2_data_entry_t OR xfs_dir2_data_unused_t |
512  *    | xfs_dir2_data_entry_t OR xfs_dir2_data_unused_t |
513  *    | xfs_dir2_data_entry_t OR xfs_dir2_data_unused_t :
514  *    | ...                                             |
515  *    +-------------------------------------------------+
516  *    | unused space                                    |
517  *    +-------------------------------------------------+
518  *    | ...                                             |
519  *    | xfs_dir2_leaf_entry_t                           |
520  *    | xfs_dir2_leaf_entry_t                           |
521  *    +-------------------------------------------------+
522  *    | xfs_dir2_block_tail_t                           |
523  *    +-------------------------------------------------+
524  *
525  * As all the entries are variable size structures the accessors below should
526  * be used to iterate over them.
527  */
528 
529 typedef struct xfs_dir2_block_tail {
530 	__be32		count;			/* count of leaf entries */
531 	__be32		stale;			/* count of stale lf entries */
532 } xfs_dir2_block_tail_t;
533 
534 /*
535  * Pointer to the leaf entries embedded in a data block (1-block format)
536  */
537 static inline struct xfs_dir2_leaf_entry *
xfs_dir2_block_leaf_p(struct xfs_dir2_block_tail * btp)538 xfs_dir2_block_leaf_p(struct xfs_dir2_block_tail *btp)
539 {
540 	return ((struct xfs_dir2_leaf_entry *)btp) - be32_to_cpu(btp->count);
541 }
542 
543 
544 /*
545  * Attribute storage layout
546  *
547  * Attribute lists are structured around Btrees where all the data
548  * elements are in the leaf nodes.  Attribute names are hashed into an int,
549  * then that int is used as the index into the Btree.  Since the hashval
550  * of an attribute name may not be unique, we may have duplicate keys.  The
551  * internal links in the Btree are logical block offsets into the file.
552  *
553  * Struct leaf_entry's are packed from the top.  Name/values grow from the
554  * bottom but are not packed.  The freemap contains run-length-encoded entries
555  * for the free bytes after the leaf_entry's, but only the N largest such,
556  * smaller runs are dropped.  When the freemap doesn't show enough space
557  * for an allocation, we compact the name/value area and try again.  If we
558  * still don't have enough space, then we have to split the block.  The
559  * name/value structs (both local and remote versions) must be 32bit aligned.
560  *
561  * Since we have duplicate hash keys, for each key that matches, compare
562  * the actual name string.  The root and intermediate node search always
563  * takes the first-in-the-block key match found, so we should only have
564  * to work "forw"ard.  If none matches, continue with the "forw"ard leaf
565  * nodes until the hash key changes or the attribute name is found.
566  *
567  * We store the fact that an attribute is a ROOT/USER/SECURE attribute in
568  * the leaf_entry.  The namespaces are independent only because we also look
569  * at the namespace bit when we are looking for a matching attribute name.
570  *
571  * We also store an "incomplete" bit in the leaf_entry.  It shows that an
572  * attribute is in the middle of being created and should not be shown to
573  * the user if we crash during the time that the bit is set.  We clear the
574  * bit when we have finished setting up the attribute.  We do this because
575  * we cannot create some large attributes inside a single transaction, and we
576  * need some indication that we weren't finished if we crash in the middle.
577  */
578 #define XFS_ATTR_LEAF_MAPSIZE	3	/* how many freespace slots */
579 
580 /*
581  * Entries are packed toward the top as tight as possible.
582  */
583 struct xfs_attr_shortform {
584 	struct xfs_attr_sf_hdr {	/* constant-structure header block */
585 		__be16	totsize;	/* total bytes in shortform list */
586 		__u8	count;	/* count of active entries */
587 		__u8	padding;
588 	} hdr;
589 	struct xfs_attr_sf_entry {
590 		uint8_t namelen;	/* actual length of name (no NULL) */
591 		uint8_t valuelen;	/* actual length of value (no NULL) */
592 		uint8_t flags;	/* flags bits (see xfs_attr_leaf.h) */
593 		uint8_t nameval[];	/* name & value bytes concatenated */
594 	} list[];			/* variable sized array */
595 };
596 
597 typedef struct xfs_attr_leaf_map {	/* RLE map of free bytes */
598 	__be16	base;			  /* base of free region */
599 	__be16	size;			  /* length of free region */
600 } xfs_attr_leaf_map_t;
601 
602 typedef struct xfs_attr_leaf_hdr {	/* constant-structure header block */
603 	xfs_da_blkinfo_t info;		/* block type, links, etc. */
604 	__be16	count;			/* count of active leaf_entry's */
605 	__be16	usedbytes;		/* num bytes of names/values stored */
606 	__be16	firstused;		/* first used byte in name area */
607 	__u8	holes;			/* != 0 if blk needs compaction */
608 	__u8	pad1;
609 	xfs_attr_leaf_map_t freemap[XFS_ATTR_LEAF_MAPSIZE];
610 					/* N largest free regions */
611 } xfs_attr_leaf_hdr_t;
612 
613 typedef struct xfs_attr_leaf_entry {	/* sorted on key, not name */
614 	__be32	hashval;		/* hash value of name */
615 	__be16	nameidx;		/* index into buffer of name/value */
616 	__u8	flags;			/* LOCAL/ROOT/SECURE/INCOMPLETE flag */
617 	__u8	pad2;			/* unused pad byte */
618 } xfs_attr_leaf_entry_t;
619 
620 typedef struct xfs_attr_leaf_name_local {
621 	__be16	valuelen;		/* number of bytes in value */
622 	__u8	namelen;		/* length of name bytes */
623 	/*
624 	 * In Linux 6.5 this flex array was converted from nameval[1] to
625 	 * nameval[].  Be very careful here about extra padding at the end;
626 	 * see xfs_attr_leaf_entsize_local() for details.
627 	 */
628 	__u8	nameval[];		/* name/value bytes */
629 } xfs_attr_leaf_name_local_t;
630 
631 typedef struct xfs_attr_leaf_name_remote {
632 	__be32	valueblk;		/* block number of value bytes */
633 	__be32	valuelen;		/* number of bytes in value */
634 	__u8	namelen;		/* length of name bytes */
635 	/*
636 	 * In Linux 6.5 this flex array was converted from name[1] to name[].
637 	 * Be very careful here about extra padding at the end; see
638 	 * xfs_attr_leaf_entsize_remote() for details.
639 	 */
640 	__u8	name[];			/* name bytes */
641 } xfs_attr_leaf_name_remote_t;
642 
643 typedef struct xfs_attr_leafblock {
644 	xfs_attr_leaf_hdr_t	hdr;	/* constant-structure header block */
645 	xfs_attr_leaf_entry_t	entries[];	/* sorted on key, not name */
646 	/*
647 	 * The rest of the block contains the following structures after the
648 	 * leaf entries, growing from the bottom up. The variables are never
649 	 * referenced and definining them can actually make gcc optimize away
650 	 * accesses to the 'entries' array above index 0 so don't do that.
651 	 *
652 	 * xfs_attr_leaf_name_local_t namelist;
653 	 * xfs_attr_leaf_name_remote_t valuelist;
654 	 */
655 } xfs_attr_leafblock_t;
656 
657 /*
658  * CRC enabled leaf structures. Called "version 3" structures to match the
659  * version number of the directory and dablk structures for this feature, and
660  * attr2 is already taken by the variable inode attribute fork size feature.
661  */
662 struct xfs_attr3_leaf_hdr {
663 	struct xfs_da3_blkinfo	info;
664 	__be16			count;
665 	__be16			usedbytes;
666 	__be16			firstused;
667 	__u8			holes;
668 	__u8			pad1;
669 	struct xfs_attr_leaf_map freemap[XFS_ATTR_LEAF_MAPSIZE];
670 	__be32			pad2;		/* 64 bit alignment */
671 };
672 
673 #define XFS_ATTR3_LEAF_CRC_OFF	(offsetof(struct xfs_attr3_leaf_hdr, info.crc))
674 
675 struct xfs_attr3_leafblock {
676 	struct xfs_attr3_leaf_hdr	hdr;
677 	struct xfs_attr_leaf_entry	entries[];
678 
679 	/*
680 	 * The rest of the block contains the following structures after the
681 	 * leaf entries, growing from the bottom up. The variables are never
682 	 * referenced, the locations accessed purely from helper functions.
683 	 *
684 	 * struct xfs_attr_leaf_name_local
685 	 * struct xfs_attr_leaf_name_remote
686 	 */
687 };
688 
689 /*
690  * Special value to represent fs block size in the leaf header firstused field.
691  * Only used when block size overflows the 2-bytes available on disk.
692  */
693 #define XFS_ATTR3_LEAF_NULLOFF	0
694 
695 /*
696  * Flags used in the leaf_entry[i].flags field.
697  */
698 #define	XFS_ATTR_LOCAL_BIT	0	/* attr is stored locally */
699 #define	XFS_ATTR_ROOT_BIT	1	/* limit access to trusted attrs */
700 #define	XFS_ATTR_SECURE_BIT	2	/* limit access to secure attrs */
701 #define	XFS_ATTR_INCOMPLETE_BIT	7	/* attr in middle of create/delete */
702 #define XFS_ATTR_LOCAL		(1u << XFS_ATTR_LOCAL_BIT)
703 #define XFS_ATTR_ROOT		(1u << XFS_ATTR_ROOT_BIT)
704 #define XFS_ATTR_SECURE		(1u << XFS_ATTR_SECURE_BIT)
705 #define XFS_ATTR_INCOMPLETE	(1u << XFS_ATTR_INCOMPLETE_BIT)
706 #define XFS_ATTR_NSP_ONDISK_MASK	(XFS_ATTR_ROOT | XFS_ATTR_SECURE)
707 
708 /*
709  * Alignment for namelist and valuelist entries (since they are mixed
710  * there can be only one alignment value)
711  */
712 #define	XFS_ATTR_LEAF_NAME_ALIGN	((uint)sizeof(xfs_dablk_t))
713 
714 static inline int
xfs_attr3_leaf_hdr_size(struct xfs_attr_leafblock * leafp)715 xfs_attr3_leaf_hdr_size(struct xfs_attr_leafblock *leafp)
716 {
717 	if (leafp->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC))
718 		return sizeof(struct xfs_attr3_leaf_hdr);
719 	return sizeof(struct xfs_attr_leaf_hdr);
720 }
721 
722 static inline struct xfs_attr_leaf_entry *
xfs_attr3_leaf_entryp(xfs_attr_leafblock_t * leafp)723 xfs_attr3_leaf_entryp(xfs_attr_leafblock_t *leafp)
724 {
725 	if (leafp->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC))
726 		return &((struct xfs_attr3_leafblock *)leafp)->entries[0];
727 	return &leafp->entries[0];
728 }
729 
730 /*
731  * Cast typed pointers for "local" and "remote" name/value structs.
732  */
733 static inline char *
xfs_attr3_leaf_name(xfs_attr_leafblock_t * leafp,int idx)734 xfs_attr3_leaf_name(xfs_attr_leafblock_t *leafp, int idx)
735 {
736 	struct xfs_attr_leaf_entry *entries = xfs_attr3_leaf_entryp(leafp);
737 
738 	return &((char *)leafp)[be16_to_cpu(entries[idx].nameidx)];
739 }
740 
741 static inline xfs_attr_leaf_name_remote_t *
xfs_attr3_leaf_name_remote(xfs_attr_leafblock_t * leafp,int idx)742 xfs_attr3_leaf_name_remote(xfs_attr_leafblock_t *leafp, int idx)
743 {
744 	return (xfs_attr_leaf_name_remote_t *)xfs_attr3_leaf_name(leafp, idx);
745 }
746 
747 static inline xfs_attr_leaf_name_local_t *
xfs_attr3_leaf_name_local(xfs_attr_leafblock_t * leafp,int idx)748 xfs_attr3_leaf_name_local(xfs_attr_leafblock_t *leafp, int idx)
749 {
750 	return (xfs_attr_leaf_name_local_t *)xfs_attr3_leaf_name(leafp, idx);
751 }
752 
753 /*
754  * Calculate total bytes used (including trailing pad for alignment) for
755  * a "local" name/value structure, a "remote" name/value structure, and
756  * a pointer which might be either.
757  */
xfs_attr_leaf_entsize_remote(int nlen)758 static inline int xfs_attr_leaf_entsize_remote(int nlen)
759 {
760 	/*
761 	 * Prior to Linux 6.5, struct xfs_attr_leaf_name_remote ended with
762 	 * name[1], which was used as a flexarray.  The layout of this struct
763 	 * is 9 bytes of fixed-length fields followed by a __u8 flex array at
764 	 * offset 9.
765 	 *
766 	 * On most architectures, struct xfs_attr_leaf_name_remote had two
767 	 * bytes of implicit padding at the end of the struct to make the
768 	 * struct length 12.  After converting name[1] to name[], there are
769 	 * three implicit padding bytes and the struct size remains 12.
770 	 * However, there are compiler configurations that do not add implicit
771 	 * padding at all (m68k) and have been broken for years.
772 	 *
773 	 * This entsize computation historically added (the xattr name length)
774 	 * to (the padded struct length - 1) and rounded that sum up to the
775 	 * nearest multiple of 4 (NAME_ALIGN).  IOWs, round_up(11 + nlen, 4).
776 	 * This is encoded in the ondisk format, so we cannot change this.
777 	 *
778 	 * Compute the entsize from offsetof of the flexarray and manually
779 	 * adding bytes for the implicit padding.
780 	 */
781 	const size_t remotesize =
782 			offsetof(struct xfs_attr_leaf_name_remote, name) + 2;
783 
784 	return round_up(remotesize + nlen, XFS_ATTR_LEAF_NAME_ALIGN);
785 }
786 
xfs_attr_leaf_entsize_local(int nlen,int vlen)787 static inline int xfs_attr_leaf_entsize_local(int nlen, int vlen)
788 {
789 	/*
790 	 * Prior to Linux 6.5, struct xfs_attr_leaf_name_local ended with
791 	 * nameval[1], which was used as a flexarray.  The layout of this
792 	 * struct is 3 bytes of fixed-length fields followed by a __u8 flex
793 	 * array at offset 3.
794 	 *
795 	 * struct xfs_attr_leaf_name_local had zero bytes of implicit padding
796 	 * at the end of the struct to make the struct length 4.  On most
797 	 * architectures, after converting nameval[1] to nameval[], there is
798 	 * one implicit padding byte and the struct size remains 4.  However,
799 	 * there are compiler configurations that do not add implicit padding
800 	 * at all (m68k) and would break.
801 	 *
802 	 * This entsize computation historically added (the xattr name and
803 	 * value length) to (the padded struct length - 1) and rounded that sum
804 	 * up to the nearest multiple of 4 (NAME_ALIGN).  IOWs, the formula is
805 	 * round_up(3 + nlen + vlen, 4).  This is encoded in the ondisk format,
806 	 * so we cannot change this.
807 	 *
808 	 * Compute the entsize from offsetof of the flexarray and manually
809 	 * adding bytes for the implicit padding.
810 	 */
811 	const size_t localsize =
812 			offsetof(struct xfs_attr_leaf_name_local, nameval);
813 
814 	return round_up(localsize + nlen + vlen, XFS_ATTR_LEAF_NAME_ALIGN);
815 }
816 
xfs_attr_leaf_entsize_local_max(int bsize)817 static inline int xfs_attr_leaf_entsize_local_max(int bsize)
818 {
819 	return (((bsize) >> 1) + ((bsize) >> 2));
820 }
821 
822 
823 
824 /*
825  * Remote attribute block format definition
826  *
827  * There is one of these headers per filesystem block in a remote attribute.
828  * This is done to ensure there is a 1:1 mapping between the attribute value
829  * length and the number of blocks needed to store the attribute. This makes the
830  * verification of a buffer a little more complex, but greatly simplifies the
831  * allocation, reading and writing of these attributes as we don't have to guess
832  * the number of blocks needed to store the attribute data.
833  */
834 #define XFS_ATTR3_RMT_MAGIC	0x5841524d	/* XARM */
835 
836 struct xfs_attr3_rmt_hdr {
837 	__be32	rm_magic;
838 	__be32	rm_offset;
839 	__be32	rm_bytes;
840 	__be32	rm_crc;
841 	uuid_t	rm_uuid;
842 	__be64	rm_owner;
843 	__be64	rm_blkno;
844 	__be64	rm_lsn;
845 };
846 
847 #define XFS_ATTR3_RMT_CRC_OFF	offsetof(struct xfs_attr3_rmt_hdr, rm_crc)
848 
849 #define XFS_ATTR3_RMT_BUF_SPACE(mp, bufsize)	\
850 	((bufsize) - (xfs_has_crc((mp)) ? \
851 			sizeof(struct xfs_attr3_rmt_hdr) : 0))
852 
853 /* Number of bytes in a directory block. */
xfs_dir2_dirblock_bytes(struct xfs_sb * sbp)854 static inline unsigned int xfs_dir2_dirblock_bytes(struct xfs_sb *sbp)
855 {
856 	return 1 << (sbp->sb_blocklog + sbp->sb_dirblklog);
857 }
858 
859 xfs_failaddr_t xfs_da3_blkinfo_verify(struct xfs_buf *bp,
860 				      struct xfs_da3_blkinfo *hdr3);
861 
862 #endif /* __XFS_DA_FORMAT_H__ */
863