1 | /***************************************************************************** |
2 | |
3 | Copyright (c) 1995, 2016, Oracle and/or its affiliates. All Rights Reserved. |
4 | Copyright (c) 2013, 2018, MariaDB Corporation. |
5 | |
6 | This program is free software; you can redistribute it and/or modify it under |
7 | the terms of the GNU General Public License as published by the Free Software |
8 | Foundation; version 2 of the License. |
9 | |
10 | This program is distributed in the hope that it will be useful, but WITHOUT |
11 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS |
12 | FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
13 | |
14 | You should have received a copy of the GNU General Public License along with |
15 | this program; if not, write to the Free Software Foundation, Inc., |
16 | 51 Franklin Street, Suite 500, Boston, MA 02110-1335 USA |
17 | |
18 | *****************************************************************************/ |
19 | |
20 | /**************************************************//** |
21 | @file include/fsp0fsp.h |
22 | File space management |
23 | |
24 | Created 12/18/1995 Heikki Tuuri |
25 | *******************************************************/ |
26 | |
27 | #ifndef fsp0fsp_h |
28 | #define fsp0fsp_h |
29 | |
30 | #include "univ.i" |
31 | |
32 | #include "fsp0types.h" |
33 | |
34 | #ifndef UNIV_INNOCHECKSUM |
35 | |
36 | #include "fsp0space.h" |
37 | #include "fut0lst.h" |
38 | #include "mtr0mtr.h" |
39 | #include "page0types.h" |
40 | #include "rem0types.h" |
41 | #include "ut0byte.h" |
42 | |
43 | #endif /* !UNIV_INNOCHECKSUM */ |
44 | #include "fsp0types.h" |
45 | |
46 | /** @return the PAGE_SSIZE flags for the current innodb_page_size */ |
47 | #define FSP_FLAGS_PAGE_SSIZE() \ |
48 | ((srv_page_size == UNIV_PAGE_SIZE_ORIG) ? \ |
49 | 0U : (srv_page_size_shift - UNIV_ZIP_SIZE_SHIFT_MIN + 1) \ |
50 | << FSP_FLAGS_POS_PAGE_SSIZE) |
51 | |
52 | /* @defgroup Compatibility macros for MariaDB 10.1.0 through 10.1.20; |
53 | see the table in fsp0types.h @{ */ |
54 | /** Zero relative shift position of the PAGE_COMPRESSION field */ |
55 | #define FSP_FLAGS_POS_PAGE_COMPRESSION_MARIADB101 \ |
56 | (FSP_FLAGS_POS_ATOMIC_BLOBS \ |
57 | + FSP_FLAGS_WIDTH_ATOMIC_BLOBS) |
58 | /** Zero relative shift position of the PAGE_COMPRESSION_LEVEL field */ |
59 | #define FSP_FLAGS_POS_PAGE_COMPRESSION_LEVEL_MARIADB101 \ |
60 | (FSP_FLAGS_POS_PAGE_COMPRESSION_MARIADB101 + 1) |
61 | /** Zero relative shift position of the ATOMIC_WRITES field */ |
62 | #define FSP_FLAGS_POS_ATOMIC_WRITES_MARIADB101 \ |
63 | (FSP_FLAGS_POS_PAGE_COMPRESSION_LEVEL_MARIADB101 + 4) |
64 | /** Zero relative shift position of the PAGE_SSIZE field */ |
65 | #define FSP_FLAGS_POS_PAGE_SSIZE_MARIADB101 \ |
66 | (FSP_FLAGS_POS_ATOMIC_WRITES_MARIADB101 + 2) |
67 | |
68 | /** Bit mask of the PAGE_COMPRESSION field */ |
69 | #define FSP_FLAGS_MASK_PAGE_COMPRESSION_MARIADB101 \ |
70 | (1U << FSP_FLAGS_POS_PAGE_COMPRESSION_MARIADB101) |
71 | /** Bit mask of the PAGE_COMPRESSION_LEVEL field */ |
72 | #define FSP_FLAGS_MASK_PAGE_COMPRESSION_LEVEL_MARIADB101 \ |
73 | (15U << FSP_FLAGS_POS_PAGE_COMPRESSION_LEVEL_MARIADB101) |
74 | /** Bit mask of the ATOMIC_WRITES field */ |
75 | #define FSP_FLAGS_MASK_ATOMIC_WRITES_MARIADB101 \ |
76 | (3U << FSP_FLAGS_POS_ATOMIC_WRITES_MARIADB101) |
77 | /** Bit mask of the PAGE_SSIZE field */ |
78 | #define FSP_FLAGS_MASK_PAGE_SSIZE_MARIADB101 \ |
79 | (15U << FSP_FLAGS_POS_PAGE_SSIZE_MARIADB101) |
80 | |
81 | /** Return the value of the PAGE_COMPRESSION field */ |
82 | #define FSP_FLAGS_GET_PAGE_COMPRESSION_MARIADB101(flags) \ |
83 | ((flags & FSP_FLAGS_MASK_PAGE_COMPRESSION_MARIADB101) \ |
84 | >> FSP_FLAGS_POS_PAGE_COMPRESSION_MARIADB101) |
85 | /** Return the value of the PAGE_COMPRESSION_LEVEL field */ |
86 | #define FSP_FLAGS_GET_PAGE_COMPRESSION_LEVEL_MARIADB101(flags) \ |
87 | ((flags & FSP_FLAGS_MASK_PAGE_COMPRESSION_LEVEL_MARIADB101) \ |
88 | >> FSP_FLAGS_POS_PAGE_COMPRESSION_LEVEL_MARIADB101) |
89 | /** Return the value of the PAGE_SSIZE field */ |
90 | #define FSP_FLAGS_GET_PAGE_SSIZE_MARIADB101(flags) \ |
91 | ((flags & FSP_FLAGS_MASK_PAGE_SSIZE_MARIADB101) \ |
92 | >> FSP_FLAGS_POS_PAGE_SSIZE_MARIADB101) |
93 | |
94 | /* @} */ |
95 | |
96 | /* @defgroup Tablespace Header Constants (moved from fsp0fsp.c) @{ */ |
97 | |
98 | /** Offset of the space header within a file page */ |
99 | #define FIL_PAGE_DATA |
100 | |
101 | /* The data structures in files are defined just as byte strings in C */ |
102 | typedef byte ; |
103 | typedef byte xdes_t; |
104 | |
105 | /* SPACE HEADER |
106 | ============ |
107 | |
108 | File space header data structure: this data structure is contained in the |
109 | first page of a space. The space for this header is reserved in every extent |
110 | descriptor page, but used only in the first. */ |
111 | |
112 | /*-------------------------------------*/ |
113 | #define FSP_SPACE_ID 0 /* space id */ |
114 | #define FSP_NOT_USED 4 /* this field contained a value up to |
115 | which we know that the modifications |
116 | in the database have been flushed to |
117 | the file space; not used now */ |
118 | #define FSP_SIZE 8 /* Current size of the space in |
119 | pages */ |
120 | #define FSP_FREE_LIMIT 12 /* Minimum page number for which the |
121 | free list has not been initialized: |
122 | the pages >= this limit are, by |
123 | definition, free; note that in a |
124 | single-table tablespace where size |
125 | < 64 pages, this number is 64, i.e., |
126 | we have initialized the space |
127 | about the first extent, but have not |
128 | physically allocated those pages to the |
129 | file */ |
130 | #define FSP_SPACE_FLAGS 16 /* fsp_space_t.flags, similar to |
131 | dict_table_t::flags */ |
132 | #define FSP_FRAG_N_USED 20 /* number of used pages in the |
133 | FSP_FREE_FRAG list */ |
134 | #define FSP_FREE 24 /* list of free extents */ |
135 | #define FSP_FREE_FRAG (24 + FLST_BASE_NODE_SIZE) |
136 | /* list of partially free extents not |
137 | belonging to any segment */ |
138 | #define FSP_FULL_FRAG (24 + 2 * FLST_BASE_NODE_SIZE) |
139 | /* list of full extents not belonging |
140 | to any segment */ |
141 | #define FSP_SEG_ID (24 + 3 * FLST_BASE_NODE_SIZE) |
142 | /* 8 bytes which give the first unused |
143 | segment id */ |
144 | #define FSP_SEG_INODES_FULL (32 + 3 * FLST_BASE_NODE_SIZE) |
145 | /* list of pages containing segment |
146 | headers, where all the segment inode |
147 | slots are reserved */ |
148 | #define FSP_SEG_INODES_FREE (32 + 4 * FLST_BASE_NODE_SIZE) |
149 | /* list of pages containing segment |
150 | headers, where not all the segment |
151 | header slots are reserved */ |
152 | /*-------------------------------------*/ |
153 | /* File space header size */ |
154 | #define (32 + 5 * FLST_BASE_NODE_SIZE) |
155 | |
156 | #define FSP_FREE_ADD 4 /* this many free extents are added |
157 | to the free list from above |
158 | FSP_FREE_LIMIT at a time */ |
159 | /* @} */ |
160 | |
161 | /* @defgroup File Segment Inode Constants (moved from fsp0fsp.c) @{ */ |
162 | |
163 | /* FILE SEGMENT INODE |
164 | ================== |
165 | |
166 | Segment inode which is created for each segment in a tablespace. NOTE: in |
167 | purge we assume that a segment having only one currently used page can be |
168 | freed in a few steps, so that the freeing cannot fill the file buffer with |
169 | bufferfixed file pages. */ |
170 | |
171 | typedef byte fseg_inode_t; |
172 | |
173 | #define FSEG_INODE_PAGE_NODE FSEG_PAGE_DATA |
174 | /* the list node for linking |
175 | segment inode pages */ |
176 | |
177 | #define FSEG_ARR_OFFSET (FSEG_PAGE_DATA + FLST_NODE_SIZE) |
178 | /*-------------------------------------*/ |
179 | #define FSEG_ID 0 /* 8 bytes of segment id: if this is 0, |
180 | it means that the header is unused */ |
181 | #define FSEG_NOT_FULL_N_USED 8 |
182 | /* number of used segment pages in |
183 | the FSEG_NOT_FULL list */ |
184 | #define FSEG_FREE 12 |
185 | /* list of free extents of this |
186 | segment */ |
187 | #define FSEG_NOT_FULL (12 + FLST_BASE_NODE_SIZE) |
188 | /* list of partially free extents */ |
189 | #define FSEG_FULL (12 + 2 * FLST_BASE_NODE_SIZE) |
190 | /* list of full extents */ |
191 | #define FSEG_MAGIC_N (12 + 3 * FLST_BASE_NODE_SIZE) |
192 | /* magic number used in debugging */ |
193 | #define FSEG_FRAG_ARR (16 + 3 * FLST_BASE_NODE_SIZE) |
194 | /* array of individual pages |
195 | belonging to this segment in fsp |
196 | fragment extent lists */ |
197 | #define FSEG_FRAG_ARR_N_SLOTS (FSP_EXTENT_SIZE / 2) |
198 | /* number of slots in the array for |
199 | the fragment pages */ |
200 | #define FSEG_FRAG_SLOT_SIZE 4 /* a fragment page slot contains its |
201 | page number within space, FIL_NULL |
202 | means that the slot is not in use */ |
203 | /*-------------------------------------*/ |
204 | #define FSEG_INODE_SIZE \ |
205 | (16 + 3 * FLST_BASE_NODE_SIZE \ |
206 | + FSEG_FRAG_ARR_N_SLOTS * FSEG_FRAG_SLOT_SIZE) |
207 | |
208 | #define FSP_SEG_INODES_PER_PAGE(page_size) \ |
209 | ((page_size.physical() - FSEG_ARR_OFFSET - 10) / FSEG_INODE_SIZE) |
210 | /* Number of segment inodes which fit on a |
211 | single page */ |
212 | |
213 | #define FSEG_MAGIC_N_VALUE 97937874 |
214 | |
215 | #define FSEG_FILLFACTOR 8 /* If this value is x, then if |
216 | the number of unused but reserved |
217 | pages in a segment is less than |
218 | reserved pages * 1/x, and there are |
219 | at least FSEG_FRAG_LIMIT used pages, |
220 | then we allow a new empty extent to |
221 | be added to the segment in |
222 | fseg_alloc_free_page. Otherwise, we |
223 | use unused pages of the segment. */ |
224 | |
225 | #define FSEG_FRAG_LIMIT FSEG_FRAG_ARR_N_SLOTS |
226 | /* If the segment has >= this many |
227 | used pages, it may be expanded by |
228 | allocating extents to the segment; |
229 | until that only individual fragment |
230 | pages are allocated from the space */ |
231 | |
232 | #define FSEG_FREE_LIST_LIMIT 40 /* If the reserved size of a segment |
233 | is at least this many extents, we |
234 | allow extents to be put to the free |
235 | list of the extent: at most |
236 | FSEG_FREE_LIST_MAX_LEN many */ |
237 | #define FSEG_FREE_LIST_MAX_LEN 4 |
238 | /* @} */ |
239 | |
240 | /* @defgroup Extent Descriptor Constants (moved from fsp0fsp.c) @{ */ |
241 | |
242 | /* EXTENT DESCRIPTOR |
243 | ================= |
244 | |
245 | File extent descriptor data structure: contains bits to tell which pages in |
246 | the extent are free and which contain old tuple version to clean. */ |
247 | |
248 | /*-------------------------------------*/ |
249 | #define XDES_ID 0 /* The identifier of the segment |
250 | to which this extent belongs */ |
251 | #define XDES_FLST_NODE 8 /* The list node data structure |
252 | for the descriptors */ |
253 | #define XDES_STATE (FLST_NODE_SIZE + 8) |
254 | /* contains state information |
255 | of the extent */ |
256 | #define XDES_BITMAP (FLST_NODE_SIZE + 12) |
257 | /* Descriptor bitmap of the pages |
258 | in the extent */ |
259 | /*-------------------------------------*/ |
260 | |
261 | #define XDES_BITS_PER_PAGE 2 /* How many bits are there per page */ |
262 | #define XDES_FREE_BIT 0 /* Index of the bit which tells if |
263 | the page is free */ |
264 | #define XDES_CLEAN_BIT 1 /* NOTE: currently not used! |
265 | Index of the bit which tells if |
266 | there are old versions of tuples |
267 | on the page */ |
268 | /* States of a descriptor */ |
269 | #define XDES_FREE 1 /* extent is in free list of space */ |
270 | #define XDES_FREE_FRAG 2 /* extent is in free fragment list of |
271 | space */ |
272 | #define XDES_FULL_FRAG 3 /* extent is in full fragment list of |
273 | space */ |
274 | #define XDES_FSEG 4 /* extent belongs to a segment */ |
275 | |
276 | /** File extent data structure size in bytes. */ |
277 | #define XDES_SIZE \ |
278 | (XDES_BITMAP \ |
279 | + UT_BITS_IN_BYTES(FSP_EXTENT_SIZE * XDES_BITS_PER_PAGE)) |
280 | |
281 | /** File extent data structure size in bytes for MAX page size. */ |
282 | #define XDES_SIZE_MAX \ |
283 | (XDES_BITMAP \ |
284 | + UT_BITS_IN_BYTES(FSP_EXTENT_SIZE_MAX * XDES_BITS_PER_PAGE)) |
285 | |
286 | /** File extent data structure size in bytes for MIN page size. */ |
287 | #define XDES_SIZE_MIN \ |
288 | (XDES_BITMAP \ |
289 | + UT_BITS_IN_BYTES(FSP_EXTENT_SIZE_MIN * XDES_BITS_PER_PAGE)) |
290 | |
291 | /** Offset of the descriptor array on a descriptor page */ |
292 | #define XDES_ARR_OFFSET (FSP_HEADER_OFFSET + FSP_HEADER_SIZE) |
293 | |
294 | #ifndef UNIV_INNOCHECKSUM |
295 | /* @} */ |
296 | |
297 | /** Calculate the number of pages to extend a datafile. |
298 | We extend single-table tablespaces first one extent at a time, |
299 | but 4 at a time for bigger tablespaces. It is not enough to extend always |
300 | by one extent, because we need to add at least one extent to FSP_FREE. |
301 | A single extent descriptor page will track many extents. And the extent |
302 | that uses its extent descriptor page is put onto the FSP_FREE_FRAG list. |
303 | Extents that do not use their extent descriptor page are added to FSP_FREE. |
304 | The physical page size is used to determine how many extents are tracked |
305 | on one extent descriptor page. See xdes_calc_descriptor_page(). |
306 | @param[in] page_size page_size of the datafile |
307 | @param[in] size current number of pages in the datafile |
308 | @return number of pages to extend the file. */ |
309 | ulint |
310 | fsp_get_pages_to_extend_ibd( |
311 | const page_size_t& page_size, |
312 | ulint size); |
313 | |
314 | /** Calculate the number of physical pages in an extent for this file. |
315 | @param[in] page_size page_size of the datafile |
316 | @return number of pages in an extent for this file. */ |
317 | UNIV_INLINE |
318 | ulint |
319 | fsp_get_extent_size_in_pages(const page_size_t& page_size) |
320 | { |
321 | return (FSP_EXTENT_SIZE << srv_page_size_shift) / page_size.physical(); |
322 | } |
323 | |
324 | /**********************************************************************//** |
325 | Reads the space id from the first page of a tablespace. |
326 | @return space id, ULINT UNDEFINED if error */ |
327 | ulint |
328 | fsp_header_get_space_id( |
329 | /*====================*/ |
330 | const page_t* page); /*!< in: first page of a tablespace */ |
331 | |
332 | /** Read a tablespace header field. |
333 | @param[in] page first page of a tablespace |
334 | @param[in] field the header field |
335 | @return the contents of the header field */ |
336 | inline |
337 | ulint |
338 | fsp_header_get_field(const page_t* page, ulint field) |
339 | { |
340 | return(mach_read_from_4(FSP_HEADER_OFFSET + field + page)); |
341 | } |
342 | |
343 | /** Read the flags from the tablespace header page. |
344 | @param[in] page first page of a tablespace |
345 | @return the contents of FSP_SPACE_FLAGS */ |
346 | inline |
347 | ulint |
348 | fsp_header_get_flags(const page_t* page) |
349 | { |
350 | return(fsp_header_get_field(page, FSP_SPACE_FLAGS)); |
351 | } |
352 | |
353 | /** Get the byte offset of encryption information in page 0. |
354 | @param[in] ps page size |
355 | @return byte offset relative to FSP_HEADER_OFFSET */ |
356 | inline MY_ATTRIBUTE((pure, warn_unused_result)) |
357 | ulint |
358 | fsp_header_get_encryption_offset(const page_size_t& ps) |
359 | { |
360 | return XDES_ARR_OFFSET + XDES_SIZE * ps.physical() / FSP_EXTENT_SIZE; |
361 | } |
362 | |
363 | /** Check the encryption key from the first page of a tablespace. |
364 | @param[in] fsp_flags tablespace flags |
365 | @param[in] page first page of a tablespace |
366 | @return true if success */ |
367 | bool |
368 | fsp_header_check_encryption_key( |
369 | ulint fsp_flags, |
370 | page_t* page); |
371 | |
372 | /**********************************************************************//** |
373 | Writes the space id and flags to a tablespace header. The flags contain |
374 | row type, physical/compressed page size, and logical/uncompressed page |
375 | size of the tablespace. */ |
376 | void |
377 | fsp_header_init_fields( |
378 | /*===================*/ |
379 | page_t* page, /*!< in/out: first page in the space */ |
380 | ulint space_id, /*!< in: space id */ |
381 | ulint flags); /*!< in: tablespace flags (FSP_SPACE_FLAGS): |
382 | 0, or table->flags if newer than COMPACT */ |
383 | /** Initialize a tablespace header. |
384 | @param[in,out] space tablespace |
385 | @param[in] size current size in blocks |
386 | @param[in,out] mtr mini-transaction */ |
387 | void fsp_header_init(fil_space_t* space, ulint size, mtr_t* mtr) |
388 | MY_ATTRIBUTE((nonnull)); |
389 | |
390 | /**********************************************************************//** |
391 | Creates a new segment. |
392 | @return the block where the segment header is placed, x-latched, NULL |
393 | if could not create segment because of lack of space */ |
394 | buf_block_t* |
395 | fseg_create( |
396 | fil_space_t* space, /*!< in,out: tablespace */ |
397 | ulint page, /*!< in: page where the segment header is placed: if |
398 | this is != 0, the page must belong to another segment, |
399 | if this is 0, a new page will be allocated and it |
400 | will belong to the created segment */ |
401 | ulint byte_offset, /*!< in: byte offset of the created segment header |
402 | on the page */ |
403 | mtr_t* mtr, |
404 | bool has_done_reservation = false); /*!< in: whether the caller |
405 | has already done the reservation for the pages with |
406 | fsp_reserve_free_extents (at least 2 extents: one for |
407 | the inode and the other for the segment) then there is |
408 | no need to do the check for this individual |
409 | operation */ |
410 | |
411 | /**********************************************************************//** |
412 | Calculates the number of pages reserved by a segment, and how many pages are |
413 | currently used. |
414 | @return number of reserved pages */ |
415 | ulint |
416 | fseg_n_reserved_pages( |
417 | /*==================*/ |
418 | fseg_header_t* header, /*!< in: segment header */ |
419 | ulint* used, /*!< out: number of pages used (<= reserved) */ |
420 | mtr_t* mtr); /*!< in/out: mini-transaction */ |
421 | /**********************************************************************//** |
422 | Allocates a single free page from a segment. This function implements |
423 | the intelligent allocation strategy which tries to minimize |
424 | file space fragmentation. |
425 | @param[in,out] seg_header segment header |
426 | @param[in] hint hint of which page would be desirable |
427 | @param[in] direction if the new page is needed because |
428 | of an index page split, and records are |
429 | inserted there in order, into which |
430 | direction they go alphabetically: FSP_DOWN, |
431 | FSP_UP, FSP_NO_DIR |
432 | @param[in,out] mtr mini-transaction |
433 | @return X-latched block, or NULL if no page could be allocated */ |
434 | #define fseg_alloc_free_page(seg_header, hint, direction, mtr) \ |
435 | fseg_alloc_free_page_general(seg_header, hint, direction, \ |
436 | FALSE, mtr, mtr) |
437 | /**********************************************************************//** |
438 | Allocates a single free page from a segment. This function implements |
439 | the intelligent allocation strategy which tries to minimize file space |
440 | fragmentation. |
441 | @retval NULL if no page could be allocated |
442 | @retval block, rw_lock_x_lock_count(&block->lock) == 1 if allocation succeeded |
443 | (init_mtr == mtr, or the page was not previously freed in mtr) |
444 | @retval block (not allocated or initialized) otherwise */ |
445 | buf_block_t* |
446 | fseg_alloc_free_page_general( |
447 | /*=========================*/ |
448 | fseg_header_t* seg_header,/*!< in/out: segment header */ |
449 | ulint hint, /*!< in: hint of which page would be |
450 | desirable */ |
451 | byte direction,/*!< in: if the new page is needed because |
452 | of an index page split, and records are |
453 | inserted there in order, into which |
454 | direction they go alphabetically: FSP_DOWN, |
455 | FSP_UP, FSP_NO_DIR */ |
456 | ibool has_done_reservation, /*!< in: TRUE if the caller has |
457 | already done the reservation for the page |
458 | with fsp_reserve_free_extents, then there |
459 | is no need to do the check for this individual |
460 | page */ |
461 | mtr_t* mtr, /*!< in/out: mini-transaction */ |
462 | mtr_t* init_mtr)/*!< in/out: mtr or another mini-transaction |
463 | in which the page should be initialized. |
464 | If init_mtr!=mtr, but the page is already |
465 | latched in mtr, do not initialize the page. */ |
466 | MY_ATTRIBUTE((warn_unused_result, nonnull)); |
467 | |
468 | /** Reserves free pages from a tablespace. All mini-transactions which may |
469 | use several pages from the tablespace should call this function beforehand |
470 | and reserve enough free extents so that they certainly will be able |
471 | to do their operation, like a B-tree page split, fully. Reservations |
472 | must be released with function fil_space_t::release_free_extents()! |
473 | |
474 | The alloc_type below has the following meaning: FSP_NORMAL means an |
475 | operation which will probably result in more space usage, like an |
476 | insert in a B-tree; FSP_UNDO means allocation to undo logs: if we are |
477 | deleting rows, then this allocation will in the long run result in |
478 | less space usage (after a purge); FSP_CLEANING means allocation done |
479 | in a physical record delete (like in a purge) or other cleaning operation |
480 | which will result in less space usage in the long run. We prefer the latter |
481 | two types of allocation: when space is scarce, FSP_NORMAL allocations |
482 | will not succeed, but the latter two allocations will succeed, if possible. |
483 | The purpose is to avoid dead end where the database is full but the |
484 | user cannot free any space because these freeing operations temporarily |
485 | reserve some space. |
486 | |
487 | Single-table tablespaces whose size is < FSP_EXTENT_SIZE pages are a special |
488 | case. In this function we would liberally reserve several extents for |
489 | every page split or merge in a B-tree. But we do not want to waste disk space |
490 | if the table only occupies < FSP_EXTENT_SIZE pages. That is why we apply |
491 | different rules in that special case, just ensuring that there are n_pages |
492 | free pages available. |
493 | |
494 | @param[out] n_reserved number of extents actually reserved; if we |
495 | return true and the tablespace size is < |
496 | FSP_EXTENT_SIZE pages, then this can be 0, |
497 | otherwise it is n_ext |
498 | @param[in,out] space tablespace |
499 | @param[in] n_ext number of extents to reserve |
500 | @param[in] alloc_type page reservation type (FSP_BLOB, etc) |
501 | @param[in,out] mtr the mini transaction |
502 | @param[in] n_pages for small tablespaces (tablespace size is |
503 | less than FSP_EXTENT_SIZE), number of free |
504 | pages to reserve. |
505 | @return true if we were able to make the reservation */ |
506 | bool |
507 | fsp_reserve_free_extents( |
508 | ulint* n_reserved, |
509 | fil_space_t* space, |
510 | ulint n_ext, |
511 | fsp_reserve_t alloc_type, |
512 | mtr_t* mtr, |
513 | ulint n_pages = 2); |
514 | |
515 | /**********************************************************************//** |
516 | Frees a single page of a segment. */ |
517 | void |
518 | fseg_free_page_func( |
519 | fseg_header_t* seg_header, /*!< in: segment header */ |
520 | ulint space_id, /*!< in: space id */ |
521 | ulint page, /*!< in: page offset */ |
522 | #ifdef BTR_CUR_HASH_ADAPT |
523 | bool ahi, /*!< in: whether we may need to drop |
524 | the adaptive hash index */ |
525 | #endif /* BTR_CUR_HASH_ADAPT */ |
526 | mtr_t* mtr); /*!< in/out: mini-transaction */ |
527 | #ifdef BTR_CUR_HASH_ADAPT |
528 | # define fseg_free_page(header, space_id, page, ahi, mtr) \ |
529 | fseg_free_page_func(header, space_id, page, ahi, mtr) |
530 | #else /* BTR_CUR_HASH_ADAPT */ |
531 | # define fseg_free_page(header, space_id, page, ahi, mtr) \ |
532 | fseg_free_page_func(header, space_id, page, mtr) |
533 | #endif /* BTR_CUR_HASH_ADAPT */ |
534 | /** Determine whether a page is free. |
535 | @param[in,out] space tablespace |
536 | @param[in] page page number |
537 | @return whether the page is marked as free */ |
538 | bool |
539 | fseg_page_is_free(fil_space_t* space, unsigned page) |
540 | MY_ATTRIBUTE((nonnull, warn_unused_result)); |
541 | /**********************************************************************//** |
542 | Frees part of a segment. This function can be used to free a segment |
543 | by repeatedly calling this function in different mini-transactions. |
544 | Doing the freeing in a single mini-transaction might result in |
545 | too big a mini-transaction. |
546 | @return TRUE if freeing completed */ |
547 | ibool |
548 | fseg_free_step_func( |
549 | fseg_header_t* header, /*!< in, own: segment header; NOTE: if the header |
550 | resides on the first page of the frag list |
551 | of the segment, this pointer becomes obsolete |
552 | after the last freeing step */ |
553 | #ifdef BTR_CUR_HASH_ADAPT |
554 | bool ahi, /*!< in: whether we may need to drop |
555 | the adaptive hash index */ |
556 | #endif /* BTR_CUR_HASH_ADAPT */ |
557 | mtr_t* mtr) /*!< in/out: mini-transaction */ |
558 | MY_ATTRIBUTE((warn_unused_result)); |
559 | #ifdef BTR_CUR_HASH_ADAPT |
560 | # define fseg_free_step(header, ahi, mtr) fseg_free_step_func(header, ahi, mtr) |
561 | #else /* BTR_CUR_HASH_ADAPT */ |
562 | # define fseg_free_step(header, ahi, mtr) fseg_free_step_func(header, mtr) |
563 | #endif /* BTR_CUR_HASH_ADAPT */ |
564 | /**********************************************************************//** |
565 | Frees part of a segment. Differs from fseg_free_step because this function |
566 | leaves the header page unfreed. |
567 | @return TRUE if freeing completed, except the header page */ |
568 | ibool |
569 | fseg_free_step_not_header_func( |
570 | fseg_header_t* header, /*!< in: segment header which must reside on |
571 | the first fragment page of the segment */ |
572 | #ifdef BTR_CUR_HASH_ADAPT |
573 | bool ahi, /*!< in: whether we may need to drop |
574 | the adaptive hash index */ |
575 | #endif /* BTR_CUR_HASH_ADAPT */ |
576 | mtr_t* mtr) /*!< in/out: mini-transaction */ |
577 | MY_ATTRIBUTE((warn_unused_result)); |
578 | #ifdef BTR_CUR_HASH_ADAPT |
579 | # define fseg_free_step_not_header(header, ahi, mtr) \ |
580 | fseg_free_step_not_header_func(header, ahi, mtr) |
581 | #else /* BTR_CUR_HASH_ADAPT */ |
582 | # define fseg_free_step_not_header(header, ahi, mtr) \ |
583 | fseg_free_step_not_header_func(header, mtr) |
584 | #endif /* BTR_CUR_HASH_ADAPT */ |
585 | |
586 | /** Checks if a page address is an extent descriptor page address. |
587 | @param[in] page_id page id |
588 | @param[in] page_size page size |
589 | @return TRUE if a descriptor page */ |
590 | UNIV_INLINE |
591 | ibool |
592 | fsp_descr_page( |
593 | const page_id_t& page_id, |
594 | const page_size_t& page_size); |
595 | |
596 | /***********************************************************//** |
597 | Parses a redo log record of a file page init. |
598 | @return end of log record or NULL */ |
599 | byte* |
600 | fsp_parse_init_file_page( |
601 | /*=====================*/ |
602 | byte* ptr, /*!< in: buffer */ |
603 | byte* end_ptr, /*!< in: buffer end */ |
604 | buf_block_t* block); /*!< in: block or NULL */ |
605 | #ifdef UNIV_BTR_PRINT |
606 | /*******************************************************************//** |
607 | Writes info of a segment. */ |
608 | void |
609 | fseg_print( |
610 | /*=======*/ |
611 | fseg_header_t* header, /*!< in: segment header */ |
612 | mtr_t* mtr); /*!< in/out: mini-transaction */ |
613 | #endif /* UNIV_BTR_PRINT */ |
614 | |
615 | /** Convert FSP_SPACE_FLAGS from the buggy MariaDB 10.1.0..10.1.20 format. |
616 | @param[in] flags the contents of FSP_SPACE_FLAGS |
617 | @return the flags corrected from the buggy MariaDB 10.1 format |
618 | @retval ULINT_UNDEFINED if the flags are not in the buggy 10.1 format */ |
619 | MY_ATTRIBUTE((warn_unused_result, const)) |
620 | UNIV_INLINE |
621 | ulint |
622 | fsp_flags_convert_from_101(ulint flags) |
623 | { |
624 | DBUG_EXECUTE_IF("fsp_flags_is_valid_failure" , |
625 | return(ULINT_UNDEFINED);); |
626 | if (flags == 0) { |
627 | return(flags); |
628 | } |
629 | |
630 | if (flags >> 18) { |
631 | /* The most significant FSP_SPACE_FLAGS bit that was ever set |
632 | by MariaDB 10.1.0 to 10.1.20 was bit 17 (misplaced DATA_DIR flag). |
633 | The flags must be less than 1<<18 in order to be valid. */ |
634 | return(ULINT_UNDEFINED); |
635 | } |
636 | |
637 | if ((flags & (FSP_FLAGS_MASK_POST_ANTELOPE | FSP_FLAGS_MASK_ATOMIC_BLOBS)) |
638 | == FSP_FLAGS_MASK_ATOMIC_BLOBS) { |
639 | /* If the "atomic blobs" flag (indicating |
640 | ROW_FORMAT=DYNAMIC or ROW_FORMAT=COMPRESSED) flag |
641 | is set, then the "post Antelope" (ROW_FORMAT!=REDUNDANT) flag |
642 | must also be set. */ |
643 | return(ULINT_UNDEFINED); |
644 | } |
645 | |
646 | /* Bits 6..10 denote compression in MariaDB 10.1.0 to 10.1.20. |
647 | They must be either 0b00000 or 0b00011 through 0b10011. |
648 | In correct versions, these bits would be |
649 | 0bd0sss where d is the DATA_DIR flag (garbage bit) and |
650 | sss is the PAGE_SSIZE (3, 4, 6, or 7). |
651 | |
652 | NOTE: MariaDB 10.1.0 to 10.1.20 can misinterpret |
653 | uncompressed data files with innodb_page_size=4k or 64k as |
654 | compressed innodb_page_size=16k files. Below is an exhaustive |
655 | state space analysis. |
656 | |
657 | -0by1zzz: impossible (the bit 4 must be clean; see above) |
658 | -0b101xx: DATA_DIR, innodb_page_size>4k: invalid (COMPRESSION_LEVEL>9) |
659 | +0bx0011: innodb_page_size=4k: |
660 | !!! Misinterpreted as COMPRESSION_LEVEL=9 or 1, COMPRESSION=1. |
661 | -0bx0010: impossible, because sss must be 0b011 or 0b1xx |
662 | -0bx0001: impossible, because sss must be 0b011 or 0b1xx |
663 | -0b10000: DATA_DIR, innodb_page_size=16: |
664 | invalid (COMPRESSION_LEVEL=8 but COMPRESSION=0) |
665 | +0b00111: no DATA_DIR, innodb_page_size=64k: |
666 | !!! Misinterpreted as COMPRESSION_LEVEL=3, COMPRESSION=1. |
667 | -0b00101: impossible, because sss must be 0 for 16k, not 0b101 |
668 | -0b001x0: no DATA_DIR, innodb_page_size=32k or 8k: |
669 | invalid (COMPRESSION_LEVEL=3 but COMPRESSION=0) |
670 | +0b00000: innodb_page_size=16k (looks like COMPRESSION=0) |
671 | ??? Could actually be compressed; see PAGE_SSIZE below */ |
672 | const ulint level = FSP_FLAGS_GET_PAGE_COMPRESSION_LEVEL_MARIADB101( |
673 | flags); |
674 | if (FSP_FLAGS_GET_PAGE_COMPRESSION_MARIADB101(flags) != (level != 0) |
675 | || level > 9) { |
676 | /* The compression flags are not in the buggy MariaDB |
677 | 10.1 format. */ |
678 | return(ULINT_UNDEFINED); |
679 | } |
680 | if (!(~flags & FSP_FLAGS_MASK_ATOMIC_WRITES_MARIADB101)) { |
681 | /* The ATOMIC_WRITES flags cannot be 0b11. |
682 | (The bits 11..12 should actually never be 0b11, |
683 | because in MySQL they would be SHARED|TEMPORARY.) */ |
684 | return(ULINT_UNDEFINED); |
685 | } |
686 | |
687 | /* Bits 13..16 are the wrong position for PAGE_SSIZE, and they |
688 | should contain one of the values 3,4,6,7, that is, be of the form |
689 | 0b0011 or 0b01xx (except 0b0110). |
690 | In correct versions, these bits should be 0bc0se |
691 | where c is the MariaDB COMPRESSED flag |
692 | and e is the MySQL 5.7 ENCRYPTION flag |
693 | and s is the MySQL 8.0 SDI flag. MariaDB can only support s=0, e=0. |
694 | |
695 | Compressed innodb_page_size=16k tables with correct FSP_SPACE_FLAGS |
696 | will be properly rejected by older MariaDB 10.1.x because they |
697 | would read as PAGE_SSIZE>=8 which is not valid. */ |
698 | |
699 | const ulint ssize = FSP_FLAGS_GET_PAGE_SSIZE_MARIADB101(flags); |
700 | if (ssize == 1 || ssize == 2 || ssize == 5 || ssize & 8) { |
701 | /* the page_size is not between 4k and 64k; |
702 | 16k should be encoded as 0, not 5 */ |
703 | return(ULINT_UNDEFINED); |
704 | } |
705 | const ulint zssize = FSP_FLAGS_GET_ZIP_SSIZE(flags); |
706 | if (zssize == 0) { |
707 | /* not ROW_FORMAT=COMPRESSED */ |
708 | } else if (zssize > (ssize ? ssize : 5)) { |
709 | /* invalid KEY_BLOCK_SIZE */ |
710 | return(ULINT_UNDEFINED); |
711 | } else if (~flags & (FSP_FLAGS_MASK_POST_ANTELOPE |
712 | | FSP_FLAGS_MASK_ATOMIC_BLOBS)) { |
713 | /* both these flags should be set for |
714 | ROW_FORMAT=COMPRESSED */ |
715 | return(ULINT_UNDEFINED); |
716 | } |
717 | |
718 | flags = ((flags & 0x3f) | ssize << FSP_FLAGS_POS_PAGE_SSIZE |
719 | | FSP_FLAGS_GET_PAGE_COMPRESSION_MARIADB101(flags) |
720 | << FSP_FLAGS_POS_PAGE_COMPRESSION); |
721 | ut_ad(fsp_flags_is_valid(flags, false)); |
722 | return(flags); |
723 | } |
724 | |
725 | /** Compare tablespace flags. |
726 | @param[in] expected expected flags from dict_tf_to_fsp_flags() |
727 | @param[in] actual flags read from FSP_SPACE_FLAGS |
728 | @return whether the flags match */ |
729 | MY_ATTRIBUTE((warn_unused_result)) |
730 | UNIV_INLINE |
731 | bool |
732 | fsp_flags_match(ulint expected, ulint actual) |
733 | { |
734 | expected &= ~FSP_FLAGS_MEM_MASK; |
735 | ut_ad(fsp_flags_is_valid(expected, false)); |
736 | |
737 | if (actual == expected) { |
738 | return(true); |
739 | } |
740 | |
741 | actual = fsp_flags_convert_from_101(actual); |
742 | return(actual == expected); |
743 | } |
744 | |
745 | /** Calculates the descriptor index within a descriptor page. |
746 | @param[in] page_size page size |
747 | @param[in] offset page offset |
748 | @return descriptor index */ |
749 | UNIV_INLINE |
750 | ulint |
751 | xdes_calc_descriptor_index( |
752 | const page_size_t& page_size, |
753 | ulint offset); |
754 | |
755 | /**********************************************************************//** |
756 | Gets a descriptor bit of a page. |
757 | @return TRUE if free */ |
758 | UNIV_INLINE |
759 | ibool |
760 | xdes_get_bit( |
761 | /*=========*/ |
762 | const xdes_t* descr, /*!< in: descriptor */ |
763 | ulint bit, /*!< in: XDES_FREE_BIT or XDES_CLEAN_BIT */ |
764 | ulint offset);/*!< in: page offset within extent: |
765 | 0 ... FSP_EXTENT_SIZE - 1 */ |
766 | |
767 | /** Calculates the page where the descriptor of a page resides. |
768 | @param[in] page_size page size |
769 | @param[in] offset page offset |
770 | @return descriptor page offset */ |
771 | UNIV_INLINE |
772 | ulint |
773 | xdes_calc_descriptor_page( |
774 | const page_size_t& page_size, |
775 | ulint offset); |
776 | |
777 | #endif /* UNIV_INNOCHECKSUM */ |
778 | |
779 | #include "fsp0fsp.ic" |
780 | |
781 | #endif |
782 | |