/src/postgres/src/include/access/slru.h
Line | Count | Source |
1 | | /*------------------------------------------------------------------------- |
2 | | * |
3 | | * slru.h |
4 | | * Simple LRU buffering for transaction status logfiles |
5 | | * |
6 | | * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
7 | | * Portions Copyright (c) 1994, Regents of the University of California |
8 | | * |
9 | | * src/include/access/slru.h |
10 | | * |
11 | | *------------------------------------------------------------------------- |
12 | | */ |
13 | | #ifndef SLRU_H |
14 | | #define SLRU_H |
15 | | |
16 | | #include "access/xlogdefs.h" |
17 | | #include "storage/lwlock.h" |
18 | | #include "storage/sync.h" |
19 | | |
20 | | /* |
21 | | * To avoid overflowing internal arithmetic and the size_t data type, the |
22 | | * number of buffers must not exceed this number. |
23 | | */ |
24 | | #define SLRU_MAX_ALLOWED_BUFFERS ((1024 * 1024 * 1024) / BLCKSZ) |
25 | | |
26 | | /* |
27 | | * Define SLRU segment size. A page is the same BLCKSZ as is used everywhere |
28 | | * else in Postgres. The segment size can be chosen somewhat arbitrarily; |
29 | | * we make it 32 pages by default, or 256Kb, i.e. 1M transactions for CLOG |
30 | | * or 64K transactions for SUBTRANS. |
31 | | * |
32 | | * Note: because TransactionIds are 32 bits and wrap around at 0xFFFFFFFF, |
33 | | * page numbering also wraps around at 0xFFFFFFFF/xxxx_XACTS_PER_PAGE (where |
34 | | * xxxx is CLOG or SUBTRANS, respectively), and segment numbering at |
35 | | * 0xFFFFFFFF/xxxx_XACTS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need |
36 | | * take no explicit notice of that fact in slru.c, except when comparing |
37 | | * segment and page numbers in SimpleLruTruncate (see PagePrecedes()). |
38 | | */ |
39 | 0 | #define SLRU_PAGES_PER_SEGMENT 32 |
40 | | |
41 | | /* |
42 | | * Page status codes. Note that these do not include the "dirty" bit. |
43 | | * page_dirty can be true only in the VALID or WRITE_IN_PROGRESS states; |
44 | | * in the latter case it implies that the page has been re-dirtied since |
45 | | * the write started. |
46 | | */ |
47 | | typedef enum |
48 | | { |
49 | | SLRU_PAGE_EMPTY, /* buffer is not in use */ |
50 | | SLRU_PAGE_READ_IN_PROGRESS, /* page is being read in */ |
51 | | SLRU_PAGE_VALID, /* page is valid and not being written */ |
52 | | SLRU_PAGE_WRITE_IN_PROGRESS, /* page is being written out */ |
53 | | } SlruPageStatus; |
54 | | |
55 | | /* |
56 | | * Shared-memory state |
57 | | * |
58 | | * SLRU bank locks are used to protect access to the other fields, except |
59 | | * latest_page_number, which uses atomics; see comment in slru.c. |
60 | | */ |
61 | | typedef struct SlruSharedData |
62 | | { |
63 | | /* Number of buffers managed by this SLRU structure */ |
64 | | int num_slots; |
65 | | |
66 | | /* |
67 | | * Arrays holding info for each buffer slot. Page number is undefined |
68 | | * when status is EMPTY, as is page_lru_count. |
69 | | */ |
70 | | char **page_buffer; |
71 | | SlruPageStatus *page_status; |
72 | | bool *page_dirty; |
73 | | int64 *page_number; |
74 | | int *page_lru_count; |
75 | | |
76 | | /* The buffer_locks protects the I/O on each buffer slots */ |
77 | | LWLockPadded *buffer_locks; |
78 | | |
79 | | /* Locks to protect the in memory buffer slot access in SLRU bank. */ |
80 | | LWLockPadded *bank_locks; |
81 | | |
82 | | /*---------- |
83 | | * A bank-wise LRU counter is maintained because we do a victim buffer |
84 | | * search within a bank. Furthermore, manipulating an individual bank |
85 | | * counter avoids frequent cache invalidation since we update it every time |
86 | | * we access the page. |
87 | | * |
88 | | * We mark a page "most recently used" by setting |
89 | | * page_lru_count[slotno] = ++bank_cur_lru_count[bankno]; |
90 | | * The oldest page in the bank is therefore the one with the highest value |
91 | | * of |
92 | | * bank_cur_lru_count[bankno] - page_lru_count[slotno] |
93 | | * The counts will eventually wrap around, but this calculation still |
94 | | * works as long as no page's age exceeds INT_MAX counts. |
95 | | *---------- |
96 | | */ |
97 | | int *bank_cur_lru_count; |
98 | | |
99 | | /* |
100 | | * Optional array of WAL flush LSNs associated with entries in the SLRU |
101 | | * pages. If not zero/NULL, we must flush WAL before writing pages (true |
102 | | * for pg_xact, false for everything else). group_lsn[] has |
103 | | * lsn_groups_per_page entries per buffer slot, each containing the |
104 | | * highest LSN known for a contiguous group of SLRU entries on that slot's |
105 | | * page. |
106 | | */ |
107 | | XLogRecPtr *group_lsn; |
108 | | int lsn_groups_per_page; |
109 | | |
110 | | /* |
111 | | * latest_page_number is the page number of the current end of the log; |
112 | | * this is not critical data, since we use it only to avoid swapping out |
113 | | * the latest page. |
114 | | */ |
115 | | pg_atomic_uint64 latest_page_number; |
116 | | |
117 | | /* SLRU's index for statistics purposes (might not be unique) */ |
118 | | int slru_stats_idx; |
119 | | } SlruSharedData; |
120 | | |
121 | | typedef SlruSharedData *SlruShared; |
122 | | |
123 | | /* |
124 | | * SlruCtlData is an unshared structure that points to the active information |
125 | | * in shared memory. |
126 | | */ |
127 | | typedef struct SlruCtlData |
128 | | { |
129 | | SlruShared shared; |
130 | | |
131 | | /* Number of banks in this SLRU. */ |
132 | | uint16 nbanks; |
133 | | |
134 | | /* |
135 | | * If true, use long segment file names. Otherwise, use short file names. |
136 | | * |
137 | | * For details about the file name format, see SlruFileName(). |
138 | | */ |
139 | | bool long_segment_names; |
140 | | |
141 | | /* |
142 | | * Which sync handler function to use when handing sync requests over to |
143 | | * the checkpointer. SYNC_HANDLER_NONE to disable fsync (eg pg_notify). |
144 | | */ |
145 | | SyncRequestHandler sync_handler; |
146 | | |
147 | | /* |
148 | | * Decide whether a page is "older" for truncation and as a hint for |
149 | | * evicting pages in LRU order. Return true if every entry of the first |
150 | | * argument is older than every entry of the second argument. Note that |
151 | | * !PagePrecedes(a,b) && !PagePrecedes(b,a) need not imply a==b; it also |
152 | | * arises when some entries are older and some are not. For SLRUs using |
153 | | * SimpleLruTruncate(), this must use modular arithmetic. (For others, |
154 | | * the behavior of this callback has no functional implications.) Use |
155 | | * SlruPagePrecedesUnitTests() in SLRUs meeting its criteria. |
156 | | */ |
157 | | bool (*PagePrecedes) (int64, int64); |
158 | | |
159 | | /* |
160 | | * Dir is set during SimpleLruInit and does not change thereafter. Since |
161 | | * it's always the same, it doesn't need to be in shared memory. |
162 | | */ |
163 | | char Dir[64]; |
164 | | } SlruCtlData; |
165 | | |
166 | | typedef SlruCtlData *SlruCtl; |
167 | | |
168 | | /* |
169 | | * Get the SLRU bank lock for given SlruCtl and the pageno. |
170 | | * |
171 | | * This lock needs to be acquired to access the slru buffer slots in the |
172 | | * respective bank. |
173 | | */ |
174 | | static inline LWLock * |
175 | | SimpleLruGetBankLock(SlruCtl ctl, int64 pageno) |
176 | 0 | { |
177 | 0 | int bankno; |
178 | |
|
179 | 0 | bankno = pageno % ctl->nbanks; |
180 | 0 | return &(ctl->shared->bank_locks[bankno].lock); |
181 | 0 | } Unexecuted instantiation: clog.c:SimpleLruGetBankLock Unexecuted instantiation: commit_ts.c:SimpleLruGetBankLock Unexecuted instantiation: multixact.c:SimpleLruGetBankLock Unexecuted instantiation: slru.c:SimpleLruGetBankLock Unexecuted instantiation: subtrans.c:SimpleLruGetBankLock Unexecuted instantiation: async.c:SimpleLruGetBankLock Unexecuted instantiation: predicate.c:SimpleLruGetBankLock Unexecuted instantiation: guc_tables.c:SimpleLruGetBankLock |
182 | | |
183 | | extern Size SimpleLruShmemSize(int nslots, int nlsns); |
184 | | extern int SimpleLruAutotuneBuffers(int divisor, int max); |
185 | | extern void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns, |
186 | | const char *subdir, int buffer_tranche_id, |
187 | | int bank_tranche_id, SyncRequestHandler sync_handler, |
188 | | bool long_segment_names); |
189 | | extern int SimpleLruZeroPage(SlruCtl ctl, int64 pageno); |
190 | | extern void SimpleLruZeroAndWritePage(SlruCtl ctl, int64 pageno); |
191 | | extern int SimpleLruReadPage(SlruCtl ctl, int64 pageno, bool write_ok, |
192 | | TransactionId xid); |
193 | | extern int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int64 pageno, |
194 | | TransactionId xid); |
195 | | extern void SimpleLruWritePage(SlruCtl ctl, int slotno); |
196 | | extern void SimpleLruWriteAll(SlruCtl ctl, bool allow_redirtied); |
197 | | #ifdef USE_ASSERT_CHECKING |
198 | | extern void SlruPagePrecedesUnitTests(SlruCtl ctl, int per_page); |
199 | | #else |
200 | 0 | #define SlruPagePrecedesUnitTests(ctl, per_page) do {} while (0) |
201 | | #endif |
202 | | extern void SimpleLruTruncate(SlruCtl ctl, int64 cutoffPage); |
203 | | extern bool SimpleLruDoesPhysicalPageExist(SlruCtl ctl, int64 pageno); |
204 | | |
205 | | typedef bool (*SlruScanCallback) (SlruCtl ctl, char *filename, int64 segpage, |
206 | | void *data); |
207 | | extern bool SlruScanDirectory(SlruCtl ctl, SlruScanCallback callback, void *data); |
208 | | extern void SlruDeleteSegment(SlruCtl ctl, int64 segno); |
209 | | |
210 | | extern int SlruSyncFileTag(SlruCtl ctl, const FileTag *ftag, char *path); |
211 | | |
212 | | /* SlruScanDirectory public callbacks */ |
213 | | extern bool SlruScanDirCbReportPresence(SlruCtl ctl, char *filename, |
214 | | int64 segpage, void *data); |
215 | | extern bool SlruScanDirCbDeleteAll(SlruCtl ctl, char *filename, int64 segpage, |
216 | | void *data); |
217 | | extern bool check_slru_buffers(const char *name, int *newval); |
218 | | |
219 | | #endif /* SLRU_H */ |