1 /*
2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #ifndef __XFS_AG_H__
19 #define	__XFS_AG_H__
20 
21 /*
22  * Allocation group header
23  * This is divided into three structures, placed in sequential 512-byte
24  * buffers after a copy of the superblock (also in a 512-byte buffer).
25  */
26 
27 struct xfs_buf;
28 struct xfs_mount;
29 struct xfs_trans;
30 
31 #define	XFS_AGF_MAGIC	0x58414746	/* 'XAGF' */
32 #define	XFS_AGI_MAGIC	0x58414749	/* 'XAGI' */
33 #define	XFS_AGF_VERSION	1
34 #define	XFS_AGI_VERSION	1
35 
36 #define	XFS_AGF_GOOD_VERSION(v)	((v) == XFS_AGF_VERSION)
37 #define	XFS_AGI_GOOD_VERSION(v)	((v) == XFS_AGI_VERSION)
38 
39 /*
40  * Btree number 0 is bno, 1 is cnt.  This value gives the size of the
41  * arrays below.
42  */
43 #define	XFS_BTNUM_AGF	((int)XFS_BTNUM_CNTi + 1)
44 
45 /*
46  * The second word of agf_levels in the first a.g. overlaps the EFS
47  * superblock's magic number.  Since the magic numbers valid for EFS
48  * are > 64k, our value cannot be confused for an EFS superblock's.
49  */
50 
51 typedef struct xfs_agf {
52 	/*
53 	 * Common allocation group header information
54 	 */
55 	__be32		agf_magicnum;	/* magic number == XFS_AGF_MAGIC */
56 	__be32		agf_versionnum;	/* header version == XFS_AGF_VERSION */
57 	__be32		agf_seqno;	/* sequence # starting from 0 */
58 	__be32		agf_length;	/* size in blocks of a.g. */
59 	/*
60 	 * Freespace information
61 	 */
62 	__be32		agf_roots[XFS_BTNUM_AGF];	/* root blocks */
63 	__be32		agf_spare0;	/* spare field */
64 	__be32		agf_levels[XFS_BTNUM_AGF];	/* btree levels */
65 	__be32		agf_spare1;	/* spare field */
66 	__be32		agf_flfirst;	/* first freelist block's index */
67 	__be32		agf_fllast;	/* last freelist block's index */
68 	__be32		agf_flcount;	/* count of blocks in freelist */
69 	__be32		agf_freeblks;	/* total free blocks */
70 	__be32		agf_longest;	/* longest free space */
71 	__be32		agf_btreeblks;	/* # of blocks held in AGF btrees */
72 } xfs_agf_t;
73 
74 #define	XFS_AGF_MAGICNUM	0x00000001
75 #define	XFS_AGF_VERSIONNUM	0x00000002
76 #define	XFS_AGF_SEQNO		0x00000004
77 #define	XFS_AGF_LENGTH		0x00000008
78 #define	XFS_AGF_ROOTS		0x00000010
79 #define	XFS_AGF_LEVELS		0x00000020
80 #define	XFS_AGF_FLFIRST		0x00000040
81 #define	XFS_AGF_FLLAST		0x00000080
82 #define	XFS_AGF_FLCOUNT		0x00000100
83 #define	XFS_AGF_FREEBLKS	0x00000200
84 #define	XFS_AGF_LONGEST		0x00000400
85 #define	XFS_AGF_BTREEBLKS	0x00000800
86 #define	XFS_AGF_NUM_BITS	12
87 #define	XFS_AGF_ALL_BITS	((1 << XFS_AGF_NUM_BITS) - 1)
88 
89 #define XFS_AGF_FLAGS \
90 	{ XFS_AGF_MAGICNUM,	"MAGICNUM" }, \
91 	{ XFS_AGF_VERSIONNUM,	"VERSIONNUM" }, \
92 	{ XFS_AGF_SEQNO,	"SEQNO" }, \
93 	{ XFS_AGF_LENGTH,	"LENGTH" }, \
94 	{ XFS_AGF_ROOTS,	"ROOTS" }, \
95 	{ XFS_AGF_LEVELS,	"LEVELS" }, \
96 	{ XFS_AGF_FLFIRST,	"FLFIRST" }, \
97 	{ XFS_AGF_FLLAST,	"FLLAST" }, \
98 	{ XFS_AGF_FLCOUNT,	"FLCOUNT" }, \
99 	{ XFS_AGF_FREEBLKS,	"FREEBLKS" }, \
100 	{ XFS_AGF_LONGEST,	"LONGEST" }, \
101 	{ XFS_AGF_BTREEBLKS,	"BTREEBLKS" }
102 
103 /* disk block (xfs_daddr_t) in the AG */
104 #define XFS_AGF_DADDR(mp)	((xfs_daddr_t)(1 << (mp)->m_sectbb_log))
105 #define	XFS_AGF_BLOCK(mp)	XFS_HDR_BLOCK(mp, XFS_AGF_DADDR(mp))
106 #define	XFS_BUF_TO_AGF(bp)	((xfs_agf_t *)XFS_BUF_PTR(bp))
107 
108 extern int xfs_read_agf(struct xfs_mount *mp, struct xfs_trans *tp,
109 			xfs_agnumber_t agno, int flags, struct xfs_buf **bpp);
110 
111 /*
112  * Size of the unlinked inode hash table in the agi.
113  */
114 #define	XFS_AGI_UNLINKED_BUCKETS	64
115 
116 typedef struct xfs_agi {
117 	/*
118 	 * Common allocation group header information
119 	 */
120 	__be32		agi_magicnum;	/* magic number == XFS_AGI_MAGIC */
121 	__be32		agi_versionnum;	/* header version == XFS_AGI_VERSION */
122 	__be32		agi_seqno;	/* sequence # starting from 0 */
123 	__be32		agi_length;	/* size in blocks of a.g. */
124 	/*
125 	 * Inode information
126 	 * Inodes are mapped by interpreting the inode number, so no
127 	 * mapping data is needed here.
128 	 */
129 	__be32		agi_count;	/* count of allocated inodes */
130 	__be32		agi_root;	/* root of inode btree */
131 	__be32		agi_level;	/* levels in inode btree */
132 	__be32		agi_freecount;	/* number of free inodes */
133 	__be32		agi_newino;	/* new inode just allocated */
134 	__be32		agi_dirino;	/* last directory inode chunk */
135 	/*
136 	 * Hash table of inodes which have been unlinked but are
137 	 * still being referenced.
138 	 */
139 	__be32		agi_unlinked[XFS_AGI_UNLINKED_BUCKETS];
140 } xfs_agi_t;
141 
142 #define	XFS_AGI_MAGICNUM	0x00000001
143 #define	XFS_AGI_VERSIONNUM	0x00000002
144 #define	XFS_AGI_SEQNO		0x00000004
145 #define	XFS_AGI_LENGTH		0x00000008
146 #define	XFS_AGI_COUNT		0x00000010
147 #define	XFS_AGI_ROOT		0x00000020
148 #define	XFS_AGI_LEVEL		0x00000040
149 #define	XFS_AGI_FREECOUNT	0x00000080
150 #define	XFS_AGI_NEWINO		0x00000100
151 #define	XFS_AGI_DIRINO		0x00000200
152 #define	XFS_AGI_UNLINKED	0x00000400
153 #define	XFS_AGI_NUM_BITS	11
154 #define	XFS_AGI_ALL_BITS	((1 << XFS_AGI_NUM_BITS) - 1)
155 
156 /* disk block (xfs_daddr_t) in the AG */
157 #define XFS_AGI_DADDR(mp)	((xfs_daddr_t)(2 << (mp)->m_sectbb_log))
158 #define	XFS_AGI_BLOCK(mp)	XFS_HDR_BLOCK(mp, XFS_AGI_DADDR(mp))
159 #define	XFS_BUF_TO_AGI(bp)	((xfs_agi_t *)XFS_BUF_PTR(bp))
160 
161 extern int xfs_read_agi(struct xfs_mount *mp, struct xfs_trans *tp,
162 				xfs_agnumber_t agno, struct xfs_buf **bpp);
163 
164 /*
165  * The third a.g. block contains the a.g. freelist, an array
166  * of block pointers to blocks owned by the allocation btree code.
167  */
168 #define XFS_AGFL_DADDR(mp)	((xfs_daddr_t)(3 << (mp)->m_sectbb_log))
169 #define	XFS_AGFL_BLOCK(mp)	XFS_HDR_BLOCK(mp, XFS_AGFL_DADDR(mp))
170 #define XFS_AGFL_SIZE(mp)	((mp)->m_sb.sb_sectsize / sizeof(xfs_agblock_t))
171 #define	XFS_BUF_TO_AGFL(bp)	((xfs_agfl_t *)XFS_BUF_PTR(bp))
172 
173 typedef struct xfs_agfl {
174 	__be32		agfl_bno[1];	/* actually XFS_AGFL_SIZE(mp) */
175 } xfs_agfl_t;
176 
177 /*
178  * Busy block/extent entry.  Indexed by a rbtree in perag to mark blocks that
179  * have been freed but whose transactions aren't committed to disk yet.
180  *
181  * Note that we use the transaction ID to record the transaction, not the
182  * transaction structure itself. See xfs_alloc_busy_insert() for details.
183  */
184 struct xfs_busy_extent {
185 	struct rb_node	rb_node;	/* ag by-bno indexed search tree */
186 	struct list_head list;		/* transaction busy extent list */
187 	xfs_agnumber_t	agno;
188 	xfs_agblock_t	bno;
189 	xfs_extlen_t	length;
190 	xlog_tid_t	tid;		/* transaction that created this */
191 };
192 
193 /*
194  * Per-ag incore structure, copies of information in agf and agi,
195  * to improve the performance of allocation group selection.
196  */
197 #define XFS_PAGB_NUM_SLOTS	128
198 
199 typedef struct xfs_perag {
200 	struct xfs_mount *pag_mount;	/* owner filesystem */
201 	xfs_agnumber_t	pag_agno;	/* AG this structure belongs to */
202 	atomic_t	pag_ref;	/* perag reference count */
203 	char		pagf_init;	/* this agf's entry is initialized */
204 	char		pagi_init;	/* this agi's entry is initialized */
205 	char		pagf_metadata;	/* the agf is preferred to be metadata */
206 	char		pagi_inodeok;	/* The agi is ok for inodes */
207 	__uint8_t	pagf_levels[XFS_BTNUM_AGF];
208 					/* # of levels in bno & cnt btree */
209 	__uint32_t	pagf_flcount;	/* count of blocks in freelist */
210 	xfs_extlen_t	pagf_freeblks;	/* total free blocks */
211 	xfs_extlen_t	pagf_longest;	/* longest free space */
212 	__uint32_t	pagf_btreeblks;	/* # of blocks held in AGF btrees */
213 	xfs_agino_t	pagi_freecount;	/* number of free inodes */
214 	xfs_agino_t	pagi_count;	/* number of allocated inodes */
215 
216 	/*
217 	 * Inode allocation search lookup optimisation.
218 	 * If the pagino matches, the search for new inodes
219 	 * doesn't need to search the near ones again straight away
220 	 */
221 	xfs_agino_t	pagl_pagino;
222 	xfs_agino_t	pagl_leftrec;
223 	xfs_agino_t	pagl_rightrec;
224 #ifdef __KERNEL__
225 	spinlock_t	pagb_lock;	/* lock for pagb_tree */
226 	struct rb_root	pagb_tree;	/* ordered tree of busy extents */
227 
228 	atomic_t        pagf_fstrms;    /* # of filestreams active in this AG */
229 
230 	spinlock_t	pag_ici_lock;	/* incore inode cache lock */
231 	struct radix_tree_root pag_ici_root;	/* incore inode cache root */
232 	int		pag_ici_reclaimable;	/* reclaimable inodes */
233 	struct mutex	pag_ici_reclaim_lock;	/* serialisation point */
234 	unsigned long	pag_ici_reclaim_cursor;	/* reclaim restart point */
235 
236 	/* buffer cache index */
237 	spinlock_t	pag_buf_lock;	/* lock for pag_buf_tree */
238 	struct rb_root	pag_buf_tree;	/* ordered tree of active buffers */
239 
240 	/* for rcu-safe freeing */
241 	struct rcu_head	rcu_head;
242 #endif
243 	int		pagb_count;	/* pagb slots in use */
244 } xfs_perag_t;
245 
246 /*
247  * tags for inode radix tree
248  */
249 #define XFS_ICI_NO_TAG		(-1)	/* special flag for an untagged lookup
250 					   in xfs_inode_ag_iterator */
251 #define XFS_ICI_RECLAIM_TAG	0	/* inode is to be reclaimed */
252 
253 #define	XFS_AG_MAXLEVELS(mp)		((mp)->m_ag_maxlevels)
254 #define	XFS_MIN_FREELIST_RAW(bl,cl,mp)	\
255 	(MIN(bl + 1, XFS_AG_MAXLEVELS(mp)) + MIN(cl + 1, XFS_AG_MAXLEVELS(mp)))
256 #define	XFS_MIN_FREELIST(a,mp)		\
257 	(XFS_MIN_FREELIST_RAW(		\
258 		be32_to_cpu((a)->agf_levels[XFS_BTNUM_BNOi]), \
259 		be32_to_cpu((a)->agf_levels[XFS_BTNUM_CNTi]), mp))
260 #define	XFS_MIN_FREELIST_PAG(pag,mp)	\
261 	(XFS_MIN_FREELIST_RAW(		\
262 		(unsigned int)(pag)->pagf_levels[XFS_BTNUM_BNOi], \
263 		(unsigned int)(pag)->pagf_levels[XFS_BTNUM_CNTi], mp))
264 
265 #define XFS_AGB_TO_FSB(mp,agno,agbno)	\
266 	(((xfs_fsblock_t)(agno) << (mp)->m_sb.sb_agblklog) | (agbno))
267 #define	XFS_FSB_TO_AGNO(mp,fsbno)	\
268 	((xfs_agnumber_t)((fsbno) >> (mp)->m_sb.sb_agblklog))
269 #define	XFS_FSB_TO_AGBNO(mp,fsbno)	\
270 	((xfs_agblock_t)((fsbno) & xfs_mask32lo((mp)->m_sb.sb_agblklog)))
271 #define	XFS_AGB_TO_DADDR(mp,agno,agbno)	\
272 	((xfs_daddr_t)XFS_FSB_TO_BB(mp, \
273 		(xfs_fsblock_t)(agno) * (mp)->m_sb.sb_agblocks + (agbno)))
274 #define	XFS_AG_DADDR(mp,agno,d)		(XFS_AGB_TO_DADDR(mp, agno, 0) + (d))
275 
276 /*
277  * For checking for bad ranges of xfs_daddr_t's, covering multiple
278  * allocation groups or a single xfs_daddr_t that's a superblock copy.
279  */
280 #define	XFS_AG_CHECK_DADDR(mp,d,len)	\
281 	((len) == 1 ? \
282 	    ASSERT((d) == XFS_SB_DADDR || \
283 		   xfs_daddr_to_agbno(mp, d) != XFS_SB_DADDR) : \
284 	    ASSERT(xfs_daddr_to_agno(mp, d) == \
285 		   xfs_daddr_to_agno(mp, (d) + (len) - 1)))
286 
287 #endif	/* __XFS_AG_H__ */
288