Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Copyright (c) 2018 Red Hat, Inc.
4 * All rights reserved.
5 */
6
7#ifndef __LIBXFS_AG_H
8#define __LIBXFS_AG_H 1
9
10struct xfs_mount;
11struct xfs_trans;
12struct xfs_perag;
13
14/*
15 * Per-ag infrastructure
16 */
17
18/* per-AG block reservation data structures*/
19struct xfs_ag_resv {
20 /* number of blocks originally reserved here */
21 xfs_extlen_t ar_orig_reserved;
22 /* number of blocks reserved here */
23 xfs_extlen_t ar_reserved;
24 /* number of blocks originally asked for */
25 xfs_extlen_t ar_asked;
26};
27
28/*
29 * Per-ag incore structure, copies of information in agf and agi, to improve the
30 * performance of allocation group selection.
31 */
32struct xfs_perag {
33 struct xfs_mount *pag_mount; /* owner filesystem */
34 xfs_agnumber_t pag_agno; /* AG this structure belongs to */
35 atomic_t pag_ref; /* passive reference count */
36 atomic_t pag_active_ref; /* active reference count */
37 wait_queue_head_t pag_active_wq;/* woken active_ref falls to zero */
38 unsigned long pag_opstate;
39 uint8_t pagf_bno_level; /* # of levels in bno btree */
40 uint8_t pagf_cnt_level; /* # of levels in cnt btree */
41 uint8_t pagf_rmap_level;/* # of levels in rmap btree */
42 uint32_t pagf_flcount; /* count of blocks in freelist */
43 xfs_extlen_t pagf_freeblks; /* total free blocks */
44 xfs_extlen_t pagf_longest; /* longest free space */
45 uint32_t pagf_btreeblks; /* # of blocks held in AGF btrees */
46 xfs_agino_t pagi_freecount; /* number of free inodes */
47 xfs_agino_t pagi_count; /* number of allocated inodes */
48
49 /*
50 * Inode allocation search lookup optimisation.
51 * If the pagino matches, the search for new inodes
52 * doesn't need to search the near ones again straight away
53 */
54 xfs_agino_t pagl_pagino;
55 xfs_agino_t pagl_leftrec;
56 xfs_agino_t pagl_rightrec;
57
58 int pagb_count; /* pagb slots in use */
59 uint8_t pagf_refcount_level; /* recount btree height */
60
61 /* Blocks reserved for all kinds of metadata. */
62 struct xfs_ag_resv pag_meta_resv;
63 /* Blocks reserved for the reverse mapping btree. */
64 struct xfs_ag_resv pag_rmapbt_resv;
65
66 /* for rcu-safe freeing */
67 struct rcu_head rcu_head;
68
69 /* Precalculated geometry info */
70 xfs_agblock_t block_count;
71 xfs_agblock_t min_block;
72 xfs_agino_t agino_min;
73 xfs_agino_t agino_max;
74
75#ifdef __KERNEL__
76 /* -- kernel only structures below this line -- */
77
78 /*
79 * Bitsets of per-ag metadata that have been checked and/or are sick.
80 * Callers should hold pag_state_lock before accessing this field.
81 */
82 uint16_t pag_checked;
83 uint16_t pag_sick;
84
85#ifdef CONFIG_XFS_ONLINE_REPAIR
86 /*
87 * Alternate btree heights so that online repair won't trip the write
88 * verifiers while rebuilding the AG btrees.
89 */
90 uint8_t pagf_repair_bno_level;
91 uint8_t pagf_repair_cnt_level;
92 uint8_t pagf_repair_refcount_level;
93 uint8_t pagf_repair_rmap_level;
94#endif
95
96 spinlock_t pag_state_lock;
97
98 spinlock_t pagb_lock; /* lock for pagb_tree */
99 struct rb_root pagb_tree; /* ordered tree of busy extents */
100 unsigned int pagb_gen; /* generation count for pagb_tree */
101 wait_queue_head_t pagb_wait; /* woken when pagb_gen changes */
102
103 atomic_t pagf_fstrms; /* # of filestreams active in this AG */
104
105 spinlock_t pag_ici_lock; /* incore inode cache lock */
106 struct radix_tree_root pag_ici_root; /* incore inode cache root */
107 int pag_ici_reclaimable; /* reclaimable inodes */
108 unsigned long pag_ici_reclaim_cursor; /* reclaim restart point */
109
110 struct xfs_buf_cache pag_bcache;
111
112 /* background prealloc block trimming */
113 struct delayed_work pag_blockgc_work;
114
115 /*
116 * We use xfs_drain to track the number of deferred log intent items
117 * that have been queued (but not yet processed) so that waiters (e.g.
118 * scrub) will not lock resources when other threads are in the middle
119 * of processing a chain of intent items only to find momentary
120 * inconsistencies.
121 */
122 struct xfs_defer_drain pag_intents_drain;
123
124 /* Hook to feed rmapbt updates to an active online repair. */
125 struct xfs_hooks pag_rmap_update_hooks;
126#endif /* __KERNEL__ */
127};
128
129/*
130 * Per-AG operational state. These are atomic flag bits.
131 */
132#define XFS_AGSTATE_AGF_INIT 0
133#define XFS_AGSTATE_AGI_INIT 1
134#define XFS_AGSTATE_PREFERS_METADATA 2
135#define XFS_AGSTATE_ALLOWS_INODES 3
136#define XFS_AGSTATE_AGFL_NEEDS_RESET 4
137
138#define __XFS_AG_OPSTATE(name, NAME) \
139static inline bool xfs_perag_ ## name (struct xfs_perag *pag) \
140{ \
141 return test_bit(XFS_AGSTATE_ ## NAME, &pag->pag_opstate); \
142}
143
144__XFS_AG_OPSTATE(initialised_agf, AGF_INIT)
145__XFS_AG_OPSTATE(initialised_agi, AGI_INIT)
146__XFS_AG_OPSTATE(prefers_metadata, PREFERS_METADATA)
147__XFS_AG_OPSTATE(allows_inodes, ALLOWS_INODES)
148__XFS_AG_OPSTATE(agfl_needs_reset, AGFL_NEEDS_RESET)
149
150void xfs_free_unused_perag_range(struct xfs_mount *mp, xfs_agnumber_t agstart,
151 xfs_agnumber_t agend);
152int xfs_initialize_perag(struct xfs_mount *mp, xfs_agnumber_t agcount,
153 xfs_rfsblock_t dcount, xfs_agnumber_t *maxagi);
154int xfs_initialize_perag_data(struct xfs_mount *mp, xfs_agnumber_t agno);
155void xfs_free_perag(struct xfs_mount *mp);
156
157/* Passive AG references */
158struct xfs_perag *xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno);
159struct xfs_perag *xfs_perag_get_tag(struct xfs_mount *mp, xfs_agnumber_t agno,
160 unsigned int tag);
161struct xfs_perag *xfs_perag_hold(struct xfs_perag *pag);
162void xfs_perag_put(struct xfs_perag *pag);
163
164/* Active AG references */
165struct xfs_perag *xfs_perag_grab(struct xfs_mount *, xfs_agnumber_t);
166struct xfs_perag *xfs_perag_grab_tag(struct xfs_mount *, xfs_agnumber_t,
167 int tag);
168void xfs_perag_rele(struct xfs_perag *pag);
169
170/*
171 * Per-ag geometry infomation and validation
172 */
173xfs_agblock_t xfs_ag_block_count(struct xfs_mount *mp, xfs_agnumber_t agno);
174void xfs_agino_range(struct xfs_mount *mp, xfs_agnumber_t agno,
175 xfs_agino_t *first, xfs_agino_t *last);
176
177static inline bool
178xfs_verify_agbno(struct xfs_perag *pag, xfs_agblock_t agbno)
179{
180 if (agbno >= pag->block_count)
181 return false;
182 if (agbno <= pag->min_block)
183 return false;
184 return true;
185}
186
187static inline bool
188xfs_verify_agbext(
189 struct xfs_perag *pag,
190 xfs_agblock_t agbno,
191 xfs_agblock_t len)
192{
193 if (agbno + len <= agbno)
194 return false;
195
196 if (!xfs_verify_agbno(pag, agbno))
197 return false;
198
199 return xfs_verify_agbno(pag, agbno + len - 1);
200}
201
202/*
203 * Verify that an AG inode number pointer neither points outside the AG
204 * nor points at static metadata.
205 */
206static inline bool
207xfs_verify_agino(struct xfs_perag *pag, xfs_agino_t agino)
208{
209 if (agino < pag->agino_min)
210 return false;
211 if (agino > pag->agino_max)
212 return false;
213 return true;
214}
215
216/*
217 * Verify that an AG inode number pointer neither points outside the AG
218 * nor points at static metadata, or is NULLAGINO.
219 */
220static inline bool
221xfs_verify_agino_or_null(struct xfs_perag *pag, xfs_agino_t agino)
222{
223 if (agino == NULLAGINO)
224 return true;
225 return xfs_verify_agino(pag, agino);
226}
227
228static inline bool
229xfs_ag_contains_log(struct xfs_mount *mp, xfs_agnumber_t agno)
230{
231 return mp->m_sb.sb_logstart > 0 &&
232 agno == XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart);
233}
234
235/*
236 * Perag iteration APIs
237 */
238static inline struct xfs_perag *
239xfs_perag_next(
240 struct xfs_perag *pag,
241 xfs_agnumber_t *agno,
242 xfs_agnumber_t end_agno)
243{
244 struct xfs_mount *mp = pag->pag_mount;
245
246 *agno = pag->pag_agno + 1;
247 xfs_perag_rele(pag);
248 while (*agno <= end_agno) {
249 pag = xfs_perag_grab(mp, *agno);
250 if (pag)
251 return pag;
252 (*agno)++;
253 }
254 return NULL;
255}
256
257#define for_each_perag_range(mp, agno, end_agno, pag) \
258 for ((pag) = xfs_perag_grab((mp), (agno)); \
259 (pag) != NULL; \
260 (pag) = xfs_perag_next((pag), &(agno), (end_agno)))
261
262#define for_each_perag_from(mp, agno, pag) \
263 for_each_perag_range((mp), (agno), (mp)->m_sb.sb_agcount - 1, (pag))
264
265#define for_each_perag(mp, agno, pag) \
266 (agno) = 0; \
267 for_each_perag_from((mp), (agno), (pag))
268
269#define for_each_perag_tag(mp, agno, pag, tag) \
270 for ((agno) = 0, (pag) = xfs_perag_grab_tag((mp), 0, (tag)); \
271 (pag) != NULL; \
272 (agno) = (pag)->pag_agno + 1, \
273 xfs_perag_rele(pag), \
274 (pag) = xfs_perag_grab_tag((mp), (agno), (tag)))
275
276static inline struct xfs_perag *
277xfs_perag_next_wrap(
278 struct xfs_perag *pag,
279 xfs_agnumber_t *agno,
280 xfs_agnumber_t stop_agno,
281 xfs_agnumber_t restart_agno,
282 xfs_agnumber_t wrap_agno)
283{
284 struct xfs_mount *mp = pag->pag_mount;
285
286 *agno = pag->pag_agno + 1;
287 xfs_perag_rele(pag);
288 while (*agno != stop_agno) {
289 if (*agno >= wrap_agno) {
290 if (restart_agno >= stop_agno)
291 break;
292 *agno = restart_agno;
293 }
294
295 pag = xfs_perag_grab(mp, *agno);
296 if (pag)
297 return pag;
298 (*agno)++;
299 }
300 return NULL;
301}
302
303/*
304 * Iterate all AGs from start_agno through wrap_agno, then restart_agno through
305 * (start_agno - 1).
306 */
307#define for_each_perag_wrap_range(mp, start_agno, restart_agno, wrap_agno, agno, pag) \
308 for ((agno) = (start_agno), (pag) = xfs_perag_grab((mp), (agno)); \
309 (pag) != NULL; \
310 (pag) = xfs_perag_next_wrap((pag), &(agno), (start_agno), \
311 (restart_agno), (wrap_agno)))
312/*
313 * Iterate all AGs from start_agno through wrap_agno, then 0 through
314 * (start_agno - 1).
315 */
316#define for_each_perag_wrap_at(mp, start_agno, wrap_agno, agno, pag) \
317 for_each_perag_wrap_range((mp), (start_agno), 0, (wrap_agno), (agno), (pag))
318
319/*
320 * Iterate all AGs from start_agno through to the end of the filesystem, then 0
321 * through (start_agno - 1).
322 */
323#define for_each_perag_wrap(mp, start_agno, agno, pag) \
324 for_each_perag_wrap_at((mp), (start_agno), (mp)->m_sb.sb_agcount, \
325 (agno), (pag))
326
327
328struct aghdr_init_data {
329 /* per ag data */
330 xfs_agblock_t agno; /* ag to init */
331 xfs_extlen_t agsize; /* new AG size */
332 struct list_head buffer_list; /* buffer writeback list */
333 xfs_rfsblock_t nfree; /* cumulative new free space */
334
335 /* per header data */
336 xfs_daddr_t daddr; /* header location */
337 size_t numblks; /* size of header */
338 const struct xfs_btree_ops *bc_ops; /* btree ops */
339};
340
341int xfs_ag_init_headers(struct xfs_mount *mp, struct aghdr_init_data *id);
342int xfs_ag_shrink_space(struct xfs_perag *pag, struct xfs_trans **tpp,
343 xfs_extlen_t delta);
344int xfs_ag_extend_space(struct xfs_perag *pag, struct xfs_trans *tp,
345 xfs_extlen_t len);
346int xfs_ag_get_geometry(struct xfs_perag *pag, struct xfs_ag_geometry *ageo);
347
348#endif /* __LIBXFS_AG_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Copyright (c) 2018 Red Hat, Inc.
4 * All rights reserved.
5 */
6
7#ifndef __LIBXFS_AG_H
8#define __LIBXFS_AG_H 1
9
10struct xfs_mount;
11struct xfs_trans;
12struct xfs_perag;
13
14/*
15 * Per-ag infrastructure
16 */
17
18/* per-AG block reservation data structures*/
19struct xfs_ag_resv {
20 /* number of blocks originally reserved here */
21 xfs_extlen_t ar_orig_reserved;
22 /* number of blocks reserved here */
23 xfs_extlen_t ar_reserved;
24 /* number of blocks originally asked for */
25 xfs_extlen_t ar_asked;
26};
27
28/*
29 * Per-ag incore structure, copies of information in agf and agi, to improve the
30 * performance of allocation group selection.
31 */
32struct xfs_perag {
33 struct xfs_mount *pag_mount; /* owner filesystem */
34 xfs_agnumber_t pag_agno; /* AG this structure belongs to */
35 atomic_t pag_ref; /* perag reference count */
36 char pagf_init; /* this agf's entry is initialized */
37 char pagi_init; /* this agi's entry is initialized */
38 char pagf_metadata; /* the agf is preferred to be metadata */
39 char pagi_inodeok; /* The agi is ok for inodes */
40 uint8_t pagf_levels[XFS_BTNUM_AGF];
41 /* # of levels in bno & cnt btree */
42 bool pagf_agflreset; /* agfl requires reset before use */
43 uint32_t pagf_flcount; /* count of blocks in freelist */
44 xfs_extlen_t pagf_freeblks; /* total free blocks */
45 xfs_extlen_t pagf_longest; /* longest free space */
46 uint32_t pagf_btreeblks; /* # of blocks held in AGF btrees */
47 xfs_agino_t pagi_freecount; /* number of free inodes */
48 xfs_agino_t pagi_count; /* number of allocated inodes */
49
50 /*
51 * Inode allocation search lookup optimisation.
52 * If the pagino matches, the search for new inodes
53 * doesn't need to search the near ones again straight away
54 */
55 xfs_agino_t pagl_pagino;
56 xfs_agino_t pagl_leftrec;
57 xfs_agino_t pagl_rightrec;
58
59 int pagb_count; /* pagb slots in use */
60 uint8_t pagf_refcount_level; /* recount btree height */
61
62 /* Blocks reserved for all kinds of metadata. */
63 struct xfs_ag_resv pag_meta_resv;
64 /* Blocks reserved for the reverse mapping btree. */
65 struct xfs_ag_resv pag_rmapbt_resv;
66
67 /* -- kernel only structures below this line -- */
68
69 /*
70 * Bitsets of per-ag metadata that have been checked and/or are sick.
71 * Callers should hold pag_state_lock before accessing this field.
72 */
73 uint16_t pag_checked;
74 uint16_t pag_sick;
75 spinlock_t pag_state_lock;
76
77 spinlock_t pagb_lock; /* lock for pagb_tree */
78 struct rb_root pagb_tree; /* ordered tree of busy extents */
79 unsigned int pagb_gen; /* generation count for pagb_tree */
80 wait_queue_head_t pagb_wait; /* woken when pagb_gen changes */
81
82 atomic_t pagf_fstrms; /* # of filestreams active in this AG */
83
84 spinlock_t pag_ici_lock; /* incore inode cache lock */
85 struct radix_tree_root pag_ici_root; /* incore inode cache root */
86 int pag_ici_reclaimable; /* reclaimable inodes */
87 unsigned long pag_ici_reclaim_cursor; /* reclaim restart point */
88
89 /* buffer cache index */
90 spinlock_t pag_buf_lock; /* lock for pag_buf_hash */
91 struct rhashtable pag_buf_hash;
92
93 /* for rcu-safe freeing */
94 struct rcu_head rcu_head;
95
96 /* background prealloc block trimming */
97 struct delayed_work pag_blockgc_work;
98
99 /*
100 * Unlinked inode information. This incore information reflects
101 * data stored in the AGI, so callers must hold the AGI buffer lock
102 * or have some other means to control concurrency.
103 */
104 struct rhashtable pagi_unlinked_hash;
105};
106
107int xfs_initialize_perag(struct xfs_mount *mp, xfs_agnumber_t agcount,
108 xfs_agnumber_t *maxagi);
109int xfs_initialize_perag_data(struct xfs_mount *mp, xfs_agnumber_t agno);
110void xfs_free_perag(struct xfs_mount *mp);
111
112struct xfs_perag *xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno);
113struct xfs_perag *xfs_perag_get_tag(struct xfs_mount *mp, xfs_agnumber_t agno,
114 unsigned int tag);
115void xfs_perag_put(struct xfs_perag *pag);
116
117/*
118 * Perag iteration APIs
119 *
120 * XXX: for_each_perag_range() usage really needs an iterator to clean up when
121 * we terminate at end_agno because we may have taken a reference to the perag
122 * beyond end_agno. Right now callers have to be careful to catch and clean that
123 * up themselves. This is not necessary for the callers of for_each_perag() and
124 * for_each_perag_from() because they terminate at sb_agcount where there are
125 * no perag structures in tree beyond end_agno.
126 */
127#define for_each_perag_range(mp, next_agno, end_agno, pag) \
128 for ((pag) = xfs_perag_get((mp), (next_agno)); \
129 (pag) != NULL && (next_agno) <= (end_agno); \
130 (next_agno) = (pag)->pag_agno + 1, \
131 xfs_perag_put(pag), \
132 (pag) = xfs_perag_get((mp), (next_agno)))
133
134#define for_each_perag_from(mp, next_agno, pag) \
135 for_each_perag_range((mp), (next_agno), (mp)->m_sb.sb_agcount, (pag))
136
137
138#define for_each_perag(mp, agno, pag) \
139 (agno) = 0; \
140 for_each_perag_from((mp), (agno), (pag))
141
142#define for_each_perag_tag(mp, agno, pag, tag) \
143 for ((agno) = 0, (pag) = xfs_perag_get_tag((mp), 0, (tag)); \
144 (pag) != NULL; \
145 (agno) = (pag)->pag_agno + 1, \
146 xfs_perag_put(pag), \
147 (pag) = xfs_perag_get_tag((mp), (agno), (tag)))
148
149struct aghdr_init_data {
150 /* per ag data */
151 xfs_agblock_t agno; /* ag to init */
152 xfs_extlen_t agsize; /* new AG size */
153 struct list_head buffer_list; /* buffer writeback list */
154 xfs_rfsblock_t nfree; /* cumulative new free space */
155
156 /* per header data */
157 xfs_daddr_t daddr; /* header location */
158 size_t numblks; /* size of header */
159 xfs_btnum_t type; /* type of btree root block */
160};
161
162int xfs_ag_init_headers(struct xfs_mount *mp, struct aghdr_init_data *id);
163int xfs_ag_shrink_space(struct xfs_mount *mp, struct xfs_trans **tpp,
164 xfs_agnumber_t agno, xfs_extlen_t delta);
165int xfs_ag_extend_space(struct xfs_mount *mp, struct xfs_trans *tp,
166 struct aghdr_init_data *id, xfs_extlen_t len);
167int xfs_ag_get_geometry(struct xfs_mount *mp, xfs_agnumber_t agno,
168 struct xfs_ag_geometry *ageo);
169
170#endif /* __LIBXFS_AG_H */