diff options
| author | unknown <heikki@hundin.mysql.fi> | 2003-06-15 01:04:28 +0300 |
|---|---|---|
| committer | unknown <heikki@hundin.mysql.fi> | 2003-06-15 01:04:28 +0300 |
| commit | 8fe7e029c211dc3856fd3a8274f5fbbc95f12e3b (patch) | |
| tree | 20d5e35fb0abb5c95cf3a833e58bcd4bf0a1c585 | |
| parent | 9adfaf1b2f682871fa7577de1e32f199a2955e79 (diff) | |
| download | mariadb-git-8fe7e029c211dc3856fd3a8274f5fbbc95f12e3b.tar.gz | |
Many files:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
sql/ha_innodb.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
sql/handler.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
sql/sql_lex.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
sql/ha_innodb.cc:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
sql/handler.cc:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
sql/sql_parse.cc:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
sql/sql_yacc.yy:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
include/my_base.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
mysql-test/t/innodb.test:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
mysql-test/r/innodb.result:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/btr/btr0cur.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/btr/btr0pcur.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/buf/buf0buf.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/buf/buf0flu.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/data/data0data.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/data/data0type.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/dict/dict0boot.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/dict/dict0crea.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/dict/dict0dict.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/dict/dict0load.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/dict/dict0mem.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/fil/fil0fil.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/fsp/fsp0fsp.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/ha/ha0ha.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/ibuf/ibuf0ibuf.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/btr0cur.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/buf0buf.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/data0data.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/data0type.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/db0err.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/dict0dict.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/dict0mem.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/fil0fil.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/lock0lock.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/os0file.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/page0page.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/rem0cmp.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/row0row.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/row0sel.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/row0upd.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/srv0srv.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/data0type.ic:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/trx0roll.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/trx0trx.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/trx0types.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/ut0dbg.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/ut0mem.h:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/row0mysql.ic:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/include/trx0sys.ic:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/lock/lock0lock.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/log/log0log.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/log/log0recv.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/mem/mem0pool.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/os/os0file.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/os/os0thread.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/page/page0cur.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/page/page0page.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/pars/pars0opt.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/pars/pars0pars.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/rem/rem0cmp.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/row/row0ins.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/row/row0mysql.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/row/row0row.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/row/row0sel.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/row/row0umod.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/row/row0upd.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/row/row0vers.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/srv/srv0srv.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/srv/srv0start.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/trx/trx0rec.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/trx/trx0roll.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/trx/trx0sys.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/trx/trx0trx.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/ut/ut0mem.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
innobase/ut/ut0ut.c:
Merge InnoDB-4.0.14: SAVEPOINT now implemented; InnoDB now accepts also column prefix keys; crashing bug in ON UPDATE CASCADE fixed; page checksum formula fixed
75 files changed, 2290 insertions, 778 deletions
diff --git a/include/my_base.h b/include/my_base.h index cd04ab971db..91a248cd401 100644 --- a/include/my_base.h +++ b/include/my_base.h @@ -255,6 +255,7 @@ enum ha_base_keytype { #define HA_ERR_CANNOT_ADD_FOREIGN 150 /* Cannot add a foreign key constr. */ #define HA_ERR_NO_REFERENCED_ROW 151 /* Cannot add a child row */ #define HA_ERR_ROW_IS_REFERENCED 152 /* Cannot delete a parent row */ +#define HA_ERR_NO_SAVEPOINT 153 /* No savepoint with that name */ /* Other constants */ diff --git a/innobase/btr/btr0cur.c b/innobase/btr/btr0cur.c index e61dcf4ecee..8402993e971 100644 --- a/innobase/btr/btr0cur.c +++ b/innobase/btr/btr0cur.c @@ -1364,7 +1364,8 @@ btr_cur_update_sec_rec_in_place( } /***************************************************************** -Updates a record when the update causes no size changes in its fields. */ +Updates a record when the update causes no size changes in its fields. +We assume here that the ordering fields of the record do not change. */ ulint btr_cur_update_in_place( @@ -1455,7 +1456,8 @@ btr_cur_update_in_place( Tries to update a record on a page in an index tree. It is assumed that mtr holds an x-latch on the page. The operation does not succeed if there is too little space on the page or if the update would result in too empty a page, -so that tree compression is recommended. */ +so that tree compression is recommended. We assume here that the ordering +fields of the record do not change. */ ulint btr_cur_optimistic_update( @@ -1507,10 +1509,11 @@ btr_cur_optimistic_update( ut_ad(mtr_memo_contains(mtr, buf_block_align(page), MTR_MEMO_PAGE_X_FIX)); - if (!row_upd_changes_field_size(rec, index, update)) { + if (!row_upd_changes_field_size_or_external(rec, index, update)) { - /* The simplest and most common case: the update does not - change the size of any field */ + /* The simplest and the most common case: the update does not + change the size of any field and none of the updated fields is + externally stored in rec or update */ return(btr_cur_update_in_place(flags, cursor, update, cmpl_info, thr, mtr)); @@ -1539,7 +1542,7 @@ btr_cur_optimistic_update( new_entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap); - row_upd_clust_index_replace_new_col_vals(new_entry, update); + row_upd_index_replace_new_col_vals(new_entry, index, update, NULL); old_rec_size = rec_get_size(rec); new_rec_size = rec_get_converted_size(new_entry); @@ -1669,54 +1672,13 @@ btr_cur_pess_upd_restore_supremum( lock_rec_reset_and_inherit_gap_locks(page_get_supremum_rec(prev_page), rec); } - -/*************************************************************** -Replaces and copies the data in the new column values stored in the -update vector to the clustered index entry given. */ -static -void -btr_cur_copy_new_col_vals( -/*======================*/ - dtuple_t* entry, /* in/out: index entry where replaced */ - upd_t* update, /* in: update vector */ - mem_heap_t* heap) /* in: heap where data is copied */ -{ - upd_field_t* upd_field; - dfield_t* dfield; - dfield_t* new_val; - ulint field_no; - byte* data; - ulint i; - - dtuple_set_info_bits(entry, update->info_bits); - - for (i = 0; i < upd_get_n_fields(update); i++) { - - upd_field = upd_get_nth_field(update, i); - - field_no = upd_field->field_no; - - dfield = dtuple_get_nth_field(entry, field_no); - - new_val = &(upd_field->new_val); - - if (new_val->len == UNIV_SQL_NULL) { - data = NULL; - } else { - data = mem_heap_alloc(heap, new_val->len); - - ut_memcpy(data, new_val->data, new_val->len); - } - - dfield_set_data(dfield, data, new_val->len); - } -} /***************************************************************** Performs an update of a record on a page of a tree. It is assumed that mtr holds an x-latch on the tree and on the cursor page. If the update is made on the leaf level, to avoid deadlocks, mtr must also -own x-latches to brothers of page, if those brothers exist. */ +own x-latches to brothers of page, if those brothers exist. We assume +here that the ordering fields of the record do not change. */ ulint btr_cur_pessimistic_update( @@ -1813,7 +1775,7 @@ btr_cur_pessimistic_update( new_entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap); - btr_cur_copy_new_col_vals(new_entry, update, heap); + row_upd_index_replace_new_col_vals(new_entry, index, update, heap); if (!(flags & BTR_KEEP_SYS_FLAG)) { row_upd_index_entry_sys_field(new_entry, index, DATA_ROLL_PTR, @@ -3369,8 +3331,8 @@ btr_free_externally_stored_field( page_no = mach_read_from_4(data + local_len + BTR_EXTERN_PAGE_NO); - offset = mach_read_from_4(data + local_len + BTR_EXTERN_OFFSET); - + offset = mach_read_from_4(data + local_len + + BTR_EXTERN_OFFSET); extern_len = mach_read_from_4(data + local_len + BTR_EXTERN_LEN + 4); diff --git a/innobase/btr/btr0pcur.c b/innobase/btr/btr0pcur.c index 7b817d8263d..63e7763ef87 100644 --- a/innobase/btr/btr0pcur.c +++ b/innobase/btr/btr0pcur.c @@ -364,6 +364,8 @@ btr_pcur_move_to_next_page( btr_leaf_page_release(page, cursor->latch_mode, mtr); page_cur_set_before_first(next_page, btr_pcur_get_page_cur(cursor)); + + page_check_dir(next_page); } /************************************************************* diff --git a/innobase/buf/buf0buf.c b/innobase/buf/buf0buf.c index e000d862403..246a60a61cd 100644 --- a/innobase/buf/buf0buf.c +++ b/innobase/buf/buf0buf.c @@ -209,12 +209,12 @@ ibool buf_debug_prints = FALSE; /* If this is set TRUE, /************************************************************************ Calculates a page checksum which is stored to the page when it is written -to a file. Note that we must be careful to calculate the same value -on 32-bit and 64-bit architectures. */ +to a file. Note that we must be careful to calculate the same value on +32-bit and 64-bit architectures. */ ulint -buf_calc_page_checksum( -/*===================*/ +buf_calc_page_new_checksum( +/*=======================*/ /* out: checksum */ byte* page) /* in: buffer page */ { @@ -222,12 +222,39 @@ buf_calc_page_checksum( /* Since the fields FIL_PAGE_FILE_FLUSH_LSN and ..._ARCH_LOG_NO are written outside the buffer pool to the first pages of data - files, we have to skip them in page checksum calculation */ + files, we have to skip them in the page checksum calculation. + We must also skip the field FIL_PAGE_SPACE_OR_CHKSUM where the + checksum is stored, and also the last 8 bytes of page because + there we store the old formula checksum. */ + + checksum = ut_fold_binary(page + FIL_PAGE_OFFSET, + FIL_PAGE_FILE_FLUSH_LSN - FIL_PAGE_OFFSET) + + ut_fold_binary(page + FIL_PAGE_DATA, + UNIV_PAGE_SIZE - FIL_PAGE_DATA + - FIL_PAGE_END_LSN_OLD_CHKSUM); + checksum = checksum & 0xFFFFFFFF; + + return(checksum); +} + +/************************************************************************ +In versions < 4.0.14 and < 4.1.1 there was a bug that the checksum only +looked at the first few bytes of the page. This calculates that old +checksum. +NOTE: we must first store the new formula checksum to +FIL_PAGE_SPACE_OR_CHKSUM before calculating and storing this old checksum +because this takes that field as an input! */ + +ulint +buf_calc_page_old_checksum( +/*=======================*/ + /* out: checksum */ + byte* page) /* in: buffer page */ +{ + ulint checksum; checksum = ut_fold_binary(page, FIL_PAGE_FILE_FLUSH_LSN); - + ut_fold_binary(page + FIL_PAGE_DATA, - UNIV_PAGE_SIZE - FIL_PAGE_DATA - - FIL_PAGE_END_LSN); + checksum = checksum & 0xFFFFFFFF; return(checksum); @@ -243,27 +270,47 @@ buf_page_is_corrupted( byte* read_buf) /* in: a database page */ { ulint checksum; + ulint old_checksum; + ulint checksum_field; + ulint old_checksum_field; + + if (mach_read_from_4(read_buf + FIL_PAGE_LSN + 4) + != mach_read_from_4(read_buf + UNIV_PAGE_SIZE + - FIL_PAGE_END_LSN_OLD_CHKSUM + 4)) { + + /* Stored log sequence numbers at the start and the end + of page do not match */ + + return(TRUE); + } - checksum = buf_calc_page_checksum(read_buf); + old_checksum = buf_calc_page_old_checksum(read_buf); - /* Note that InnoDB initializes empty pages to zero, and - early versions of InnoDB did not store page checksum to - the 4 most significant bytes of the page lsn field at the - end of a page: */ + old_checksum_field = mach_read_from_4(read_buf + UNIV_PAGE_SIZE + - FIL_PAGE_END_LSN_OLD_CHKSUM); + + /* There are 2 valid formulas for old_checksum_field: + 1. Very old versions of InnoDB only stored 8 byte lsn to the start + and the end of the page. + 2. Newer InnoDB versions store the old formula checksum there. */ - if ((mach_read_from_4(read_buf + FIL_PAGE_LSN + 4) - != mach_read_from_4(read_buf + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN + 4)) - || (checksum != mach_read_from_4(read_buf - + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN) - && mach_read_from_4(read_buf + FIL_PAGE_LSN) - != mach_read_from_4(read_buf - + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN))) { + if (old_checksum_field != mach_read_from_4(read_buf + FIL_PAGE_LSN) + && old_checksum_field != old_checksum) { + return(TRUE); } + checksum = buf_calc_page_new_checksum(read_buf); + checksum_field = mach_read_from_4(read_buf + FIL_PAGE_SPACE_OR_CHKSUM); + + /* InnoDB versions < 4.0.14 and < 4.1.1 stored the space id + (always equal to 0), to FIL_PAGE_SPACE_SPACE_OR_CHKSUM */ + + if (checksum_field != 0 && checksum_field != checksum) { + + return(TRUE); + } + return(FALSE); } @@ -277,6 +324,7 @@ buf_page_print( { dict_index_t* index; ulint checksum; + ulint old_checksum; char* buf; buf = mem_alloc(4 * UNIV_PAGE_SIZE); @@ -291,19 +339,23 @@ buf_page_print( mem_free(buf); - checksum = buf_calc_page_checksum(read_buf); + checksum = buf_calc_page_new_checksum(read_buf); + old_checksum = buf_calc_page_old_checksum(read_buf); ut_print_timestamp(stderr); - fprintf(stderr, " InnoDB: Page checksum %lu stored checksum %lu\n", - checksum, mach_read_from_4(read_buf - + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN)); + fprintf(stderr, +" InnoDB: Page checksum %lu, prior-to-4.0.14-form checksum %lu\n" +"InnoDB: stored checksum %lu, prior-to-4.0.14-form stored checksum %lu\n", + checksum, old_checksum, + mach_read_from_4(read_buf + FIL_PAGE_SPACE_OR_CHKSUM), + mach_read_from_4(read_buf + UNIV_PAGE_SIZE + - FIL_PAGE_END_LSN_OLD_CHKSUM)); fprintf(stderr, "InnoDB: Page lsn %lu %lu, low 4 bytes of lsn at page end %lu\n", mach_read_from_4(read_buf + FIL_PAGE_LSN), mach_read_from_4(read_buf + FIL_PAGE_LSN + 4), mach_read_from_4(read_buf + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN + 4)); + - FIL_PAGE_END_LSN_OLD_CHKSUM + 4)); if (mach_read_from_2(read_buf + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_INSERT) { fprintf(stderr, diff --git a/innobase/buf/buf0flu.c b/innobase/buf/buf0flu.c index 735966c28c5..c0999ee4841 100644 --- a/innobase/buf/buf0flu.c +++ b/innobase/buf/buf0flu.c @@ -361,21 +361,29 @@ buf_flush_init_for_writing( ulint space, /* in: space id */ ulint page_no) /* in: page number */ { - /* Write the newest modification lsn to the page */ - mach_write_to_8(page + FIL_PAGE_LSN, newest_lsn); + UT_NOT_USED(space); - mach_write_to_8(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN, newest_lsn); + /* Write the newest modification lsn to the page header and trailer */ + mach_write_to_8(page + FIL_PAGE_LSN, newest_lsn); - /* Write to the page the space id and page number */ + mach_write_to_8(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM, + newest_lsn); + /* Write the page number */ - mach_write_to_4(page + FIL_PAGE_SPACE, space); mach_write_to_4(page + FIL_PAGE_OFFSET, page_no); + /* Store the new formula checksum */ + + mach_write_to_4(page + FIL_PAGE_SPACE_OR_CHKSUM, + buf_calc_page_new_checksum(page)); + /* We overwrite the first 4 bytes of the end lsn field to store - a page checksum */ + the old formula checksum. Since it depends also on the field + FIL_PAGE_SPACE_OR_CHKSUM, it has to be calculated after storing the + new formula checksum. */ - mach_write_to_4(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN, - buf_calc_page_checksum(page)); + mach_write_to_4(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM, + buf_calc_page_old_checksum(page)); } /************************************************************************ diff --git a/innobase/data/data0data.c b/innobase/data/data0data.c index 8ab5acb4da7..f2f94cc47ce 100644 --- a/innobase/data/data0data.c +++ b/innobase/data/data0data.c @@ -584,8 +584,7 @@ dtuple_convert_big_rec( * sizeof(big_rec_field_t)); /* Decide which fields to shorten: the algorithm is to look for - the longest field which does not occur in the ordering part - of any index on the table */ + the longest field whose type is DATA_BLOB */ n_fields = 0; @@ -610,12 +609,9 @@ dtuple_convert_big_rec( } } - /* Skip over fields which are ordering in some index */ - - if (!is_externally_stored && - dict_field_get_col( - dict_index_get_nth_field(index, i)) - ->ord_part == 0) { + if (!is_externally_stored + && dict_index_get_nth_type(index, i)->mtype + == DATA_BLOB) { dfield = dtuple_get_nth_field(entry, i); @@ -629,9 +625,13 @@ dtuple_convert_big_rec( } } - if (longest < BTR_EXTERN_FIELD_REF_SIZE + 10 - + REC_1BYTE_OFFS_LIMIT) { + /* We do not store externally fields which are smaller than + DICT_MAX_COL_PREFIX_LEN */ + + ut_a(DICT_MAX_COL_PREFIX_LEN > REC_1BYTE_OFFS_LIMIT); + if (longest < BTR_EXTERN_FIELD_REF_SIZE + 10 + + DICT_MAX_COL_PREFIX_LEN) { /* Cannot shorten more */ mem_heap_free(heap); @@ -644,13 +644,19 @@ dtuple_convert_big_rec( drop below 128 which is the limit for the 2-byte offset storage format in a physical record. This we accomplish by storing 128 bytes of data in entry - itself, and only the remaining part to big rec vec. */ + itself, and only the remaining part to big rec vec. + + We store the first bytes locally to the record. Then + we can calculate all ordering fields in all indexes + from locally stored data. */ dfield = dtuple_get_nth_field(entry, longest_i); vector->fields[n_fields].field_no = longest_i; + ut_a(dfield->len > DICT_MAX_COL_PREFIX_LEN); + vector->fields[n_fields].len = dfield->len - - REC_1BYTE_OFFS_LIMIT; + - DICT_MAX_COL_PREFIX_LEN; vector->fields[n_fields].data = mem_heap_alloc(heap, vector->fields[n_fields].len); diff --git a/innobase/data/data0type.c b/innobase/data/data0type.c index 5d0ddf3e887..df430f06bcb 100644 --- a/innobase/data/data0type.c +++ b/innobase/data/data0type.c @@ -85,8 +85,6 @@ dtype_print( printf("DATA_MIX_ID"); } else if (prtype == DATA_ENGLISH) { printf("DATA_ENGLISH"); - } else if (prtype == DATA_FINNISH) { - printf("DATA_FINNISH"); } else { printf("prtype %lu", mtype); } diff --git a/innobase/dict/dict0boot.c b/innobase/dict/dict0boot.c index 374c567c3ca..0bf2ace3324 100644 --- a/innobase/dict/dict0boot.c +++ b/innobase/dict/dict0boot.c @@ -276,7 +276,7 @@ dict_boot(void) DICT_HDR_SPACE, DICT_UNIQUE | DICT_CLUSTERED, 1); - dict_mem_index_add_field(index, (char *) "NAME", 0); + dict_mem_index_add_field(index, (char *) "NAME", 0, 0); index->page_no = mtr_read_ulint(dict_hdr + DICT_HDR_TABLES, MLOG_4BYTES, &mtr); @@ -287,7 +287,7 @@ dict_boot(void) index = dict_mem_index_create((char *) "SYS_TABLES", (char *) "ID_IND", DICT_HDR_SPACE, DICT_UNIQUE, 1); - dict_mem_index_add_field(index, (char *) "ID", 0); + dict_mem_index_add_field(index, (char *) "ID", 0, 0); index->page_no = mtr_read_ulint(dict_hdr + DICT_HDR_TABLE_IDS, MLOG_4BYTES, &mtr); @@ -313,8 +313,8 @@ dict_boot(void) (char *) "CLUST_IND", DICT_HDR_SPACE, DICT_UNIQUE | DICT_CLUSTERED, 2); - dict_mem_index_add_field(index, (char *) "TABLE_ID", 0); - dict_mem_index_add_field(index, (char *) "POS", 0); + dict_mem_index_add_field(index, (char *) "TABLE_ID", 0, 0); + dict_mem_index_add_field(index, (char *) "POS", 0, 0); index->page_no = mtr_read_ulint(dict_hdr + DICT_HDR_COLUMNS, MLOG_4BYTES, &mtr); @@ -343,8 +343,8 @@ dict_boot(void) (char *) "CLUST_IND", DICT_HDR_SPACE, DICT_UNIQUE | DICT_CLUSTERED, 2); - dict_mem_index_add_field(index, (char *) "TABLE_ID", 0); - dict_mem_index_add_field(index, (char *) "ID", 0); + dict_mem_index_add_field(index, (char *) "TABLE_ID", 0, 0); + dict_mem_index_add_field(index, (char *) "ID", 0, 0); index->page_no = mtr_read_ulint(dict_hdr + DICT_HDR_INDEXES, MLOG_4BYTES, &mtr); @@ -365,8 +365,8 @@ dict_boot(void) (char *) "CLUST_IND", DICT_HDR_SPACE, DICT_UNIQUE | DICT_CLUSTERED, 2); - dict_mem_index_add_field(index, (char *) "INDEX_ID", 0); - dict_mem_index_add_field(index, (char *) "POS", 0); + dict_mem_index_add_field(index, (char *) "INDEX_ID", 0, 0); + dict_mem_index_add_field(index, (char *) "POS", 0, 0); index->page_no = mtr_read_ulint(dict_hdr + DICT_HDR_FIELDS, MLOG_4BYTES, &mtr); diff --git a/innobase/dict/dict0crea.c b/innobase/dict/dict0crea.c index 3619ac02f4d..9139e589a0a 100644 --- a/innobase/dict/dict0crea.c +++ b/innobase/dict/dict0crea.c @@ -337,7 +337,7 @@ dict_create_index_for_cluster_step( for (i = 0; i < table->n_cols; i++) { col = dict_table_get_nth_col(table, i); - dict_mem_index_add_field(index, col->name, 0); + dict_mem_index_add_field(index, col->name, 0, 0); } (node->cluster)->index = index; @@ -450,9 +450,17 @@ dict_create_sys_fields_tuple( dict_field_t* field; dfield_t* dfield; byte* ptr; + ibool index_contains_column_prefix_field = FALSE; + ulint j; ut_ad(index && heap); + for (j = 0; j < index->n_fields; j++) { + if (dict_index_get_nth_field(index, j)->prefix_len > 0) { + index_contains_column_prefix_field = TRUE; + } + } + field = dict_index_get_nth_field(index, i); sys_fields = dict_sys->sys_fields; @@ -466,11 +474,25 @@ dict_create_sys_fields_tuple( mach_write_to_8(ptr, index->id); dfield_set_data(dfield, ptr, 8); - /* 1: POS ----------------------------*/ + /* 1: POS + PREFIX LENGTH ----------------------------*/ + dfield = dtuple_get_nth_field(entry, 1); ptr = mem_heap_alloc(heap, 4); - mach_write_to_4(ptr, i); + + if (index_contains_column_prefix_field) { + /* If there are column prefix fields in the index, then + we store the number of the field to the 2 HIGH bytes + and the prefix length to the 2 low bytes, */ + + mach_write_to_4(ptr, (i << 16) + field->prefix_len); + } else { + /* Else we store the number of the field to the 2 LOW bytes. + This is to keep the storage format compatible with + InnoDB versions < 4.0.14. */ + + mach_write_to_4(ptr, i); + } dfield_set_data(dfield, ptr, 4); /* 4: COL_NAME -------------------------*/ diff --git a/innobase/dict/dict0dict.c b/innobase/dict/dict0dict.c index c11a5f76d94..2fc05b1923f 100644 --- a/innobase/dict/dict0dict.c +++ b/innobase/dict/dict0dict.c @@ -88,15 +88,6 @@ dict_index_remove_from_cache( dict_table_t* table, /* in: table */ dict_index_t* index); /* in, own: index */ /*********************************************************************** -Adds a column to index. */ -UNIV_INLINE -void -dict_index_add_col( -/*===============*/ - dict_index_t* index, /* in: index */ - dict_col_t* col, /* in: column */ - ulint order); /* in: order criterion */ -/*********************************************************************** Copies fields contained in index2 to index1. */ static void @@ -482,8 +473,9 @@ dict_index_get_nth_col_pos( ut_ad(index); ut_ad(index->magic_n == DICT_INDEX_MAGIC_N); + col = dict_table_get_nth_col(index->table, n); + if (index->type & DICT_CLUSTERED) { - col = dict_table_get_nth_col(index->table, n); return(col->clust_pos); } @@ -492,9 +484,8 @@ dict_index_get_nth_col_pos( for (pos = 0; pos < n_fields; pos++) { field = dict_index_get_nth_field(index, pos); - col = field->col; - if (dict_col_get_no(col) == n) { + if (col == field->col && field->prefix_len == 0) { return(pos); } @@ -502,7 +493,46 @@ dict_index_get_nth_col_pos( return(ULINT_UNDEFINED); } + +/************************************************************************ +Looks for a matching field in an index. The column and the prefix len have +to be the same. */ + +ulint +dict_index_get_nth_field_pos( +/*=========================*/ + /* out: position in internal representation + of the index; if not contained, returns + ULINT_UNDEFINED */ + dict_index_t* index, /* in: index from which to search */ + dict_index_t* index2, /* in: index */ + ulint n) /* in: field number in index2 */ +{ + dict_field_t* field; + dict_field_t* field2; + ulint n_fields; + ulint pos; + ut_ad(index); + ut_ad(index->magic_n == DICT_INDEX_MAGIC_N); + + field2 = dict_index_get_nth_field(index2, n); + + n_fields = dict_index_get_n_fields(index); + + for (pos = 0; pos < n_fields; pos++) { + field = dict_index_get_nth_field(index, pos); + + if (field->col == field2->col + && field->prefix_len == field2->prefix_len) { + + return(pos); + } + } + + return(ULINT_UNDEFINED); +} + /************************************************************************** Returns a table object, based on table id, and memoryfixes it. */ @@ -622,8 +652,7 @@ dict_table_get( } /************************************************************************** -Returns a table object and increments MySQL open handle count on the table. -*/ +Returns a table object and increments MySQL open handle count on the table. */ dict_table_t* dict_table_get_and_increment_handle_count( @@ -732,11 +761,12 @@ dict_table_add_to_cache( } /* Add table to hash table of tables */ - HASH_INSERT(dict_table_t, name_hash, dict_sys->table_hash, fold, table); + HASH_INSERT(dict_table_t, name_hash, dict_sys->table_hash, fold, + table); /* Add table to hash table of tables based on table id */ HASH_INSERT(dict_table_t, id_hash, dict_sys->table_id_hash, id_fold, - table); + table); /* Add table to LRU list of tables */ UT_LIST_ADD_FIRST(table_LRU, dict_sys->table_LRU, table); @@ -828,7 +858,7 @@ dict_table_rename_in_cache( /* Remove table from the hash tables of tables */ HASH_DELETE(dict_table_t, name_hash, dict_sys->table_hash, - ut_fold_string(table->name), table); + ut_fold_string(table->name), table); name_buf = mem_heap_alloc(table->heap, ut_strlen(new_name) + 1); @@ -837,7 +867,8 @@ dict_table_rename_in_cache( table->name = name_buf; /* Add table to hash table of tables */ - HASH_INSERT(dict_table_t, name_hash, dict_sys->table_hash, fold, table); + HASH_INSERT(dict_table_t, name_hash, dict_sys->table_hash, fold, + table); dict_sys->size += (mem_heap_get_size(table->heap) - old_size); @@ -1128,7 +1159,6 @@ dict_index_add_to_cache( ulint n_ord; ibool success; ulint i; - ulint j; ut_ad(index); ut_ad(mutex_own(&(dict_sys->mutex))); @@ -1158,28 +1188,6 @@ dict_index_add_to_cache( return(FALSE); } - - /* Check that the same column does not appear twice in the index. - InnoDB assumes this in its algorithms, e.g., update of an index - entry */ - - for (i = 0; i < dict_index_get_n_fields(index); i++) { - - for (j = 0; j < i; j++) { - if (dict_index_get_nth_field(index, j)->col - == dict_index_get_nth_field(index, i)->col) { - - ut_print_timestamp(stderr); - - fprintf(stderr, -" InnoDB: Error: column %s appears twice in index %s of table %s\n" -"InnoDB: This is not allowed in InnoDB.\n" -"InnoDB: UPDATE can cause such an index to become corrupt in InnoDB.\n", - dict_index_get_nth_field(index, i)->col->name, - index->name, table->name); - } - } - } /* Build the cache internal representation of the index, containing also the added system fields */ @@ -1223,8 +1231,8 @@ dict_index_add_to_cache( cluster = dict_table_get_low(table->cluster_name); - tree = dict_index_get_tree(UT_LIST_GET_FIRST(cluster->indexes)); - + tree = dict_index_get_tree( + UT_LIST_GET_FIRST(cluster->indexes)); new_index->tree = tree; new_index->page_no = tree->page; } else { @@ -1352,13 +1360,14 @@ UNIV_INLINE void dict_index_add_col( /*===============*/ - dict_index_t* index, /* in: index */ - dict_col_t* col, /* in: column */ - ulint order) /* in: order criterion */ + dict_index_t* index, /* in: index */ + dict_col_t* col, /* in: column */ + ulint order, /* in: order criterion */ + ulint prefix_len) /* in: column prefix length */ { dict_field_t* field; - dict_mem_index_add_field(index, col->name, order); + dict_mem_index_add_field(index, col->name, order, prefix_len); field = dict_index_get_nth_field(index, index->n_def - 1); @@ -1384,7 +1393,8 @@ dict_index_copy( for (i = start; i < end; i++) { field = dict_index_get_nth_field(index2, i); - dict_index_add_col(index1, field->col, field->order); + dict_index_add_col(index1, field->col, field->order, + field->prefix_len); } } @@ -1487,7 +1497,7 @@ dict_index_build_internal_clust( /* Add the mix id column */ dict_index_add_col(new_index, - dict_table_get_sys_col(table, DATA_MIX_ID), 0); + dict_table_get_sys_col(table, DATA_MIX_ID), 0, 0); /* Copy the rest of fields */ dict_index_copy(new_index, index, table->mix_len, @@ -1525,14 +1535,15 @@ dict_index_build_internal_clust( if (!(index->type & DICT_UNIQUE)) { dict_index_add_col(new_index, - dict_table_get_sys_col(table, DATA_ROW_ID), 0); + dict_table_get_sys_col(table, DATA_ROW_ID), 0, 0); trx_id_pos++; } dict_index_add_col(new_index, - dict_table_get_sys_col(table, DATA_TRX_ID), 0); + dict_table_get_sys_col(table, DATA_TRX_ID), 0, 0); + dict_index_add_col(new_index, - dict_table_get_sys_col(table, DATA_ROLL_PTR), 0); + dict_table_get_sys_col(table, DATA_ROLL_PTR), 0, 0); for (i = 0; i < trx_id_pos; i++) { @@ -1561,7 +1572,14 @@ dict_index_build_internal_clust( for (i = 0; i < new_index->n_def; i++) { field = dict_index_get_nth_field(new_index, i); - (field->col)->aux = 0; + + /* If there is only a prefix of the column in the index + field, do not mark the column as contained in the index */ + + if (field->prefix_len == 0) { + + field->col->aux = 0; + } } /* Add to new_index non-system columns of table not yet included @@ -1572,7 +1590,7 @@ dict_index_build_internal_clust( ut_ad(col->type.mtype != DATA_SYS); if (col->aux == ULINT_UNDEFINED) { - dict_index_add_col(new_index, col, 0); + dict_index_add_col(new_index, col, 0, 0); } } @@ -1584,7 +1602,11 @@ dict_index_build_internal_clust( for (i = 0; i < new_index->n_def; i++) { field = dict_index_get_nth_field(new_index, i); - (field->col)->clust_pos = i; + + if (field->prefix_len == 0) { + + field->col->clust_pos = i; + } } new_index->cached = TRUE; @@ -1646,25 +1668,33 @@ dict_index_build_internal_non_clust( for (i = 0; i < clust_index->n_uniq; i++) { field = dict_index_get_nth_field(clust_index, i); - (field->col)->aux = ULINT_UNDEFINED; + field->col->aux = ULINT_UNDEFINED; } /* Mark with 0 table columns already contained in new_index */ for (i = 0; i < new_index->n_def; i++) { field = dict_index_get_nth_field(new_index, i); - (field->col)->aux = 0; + + /* If there is only a prefix of the column in the index + field, do not mark the column as contained in the index */ + + if (field->prefix_len == 0) { + + field->col->aux = 0; + } } - /* Add to new_index columns necessary to determine the clustered + /* Add to new_index the columns necessary to determine the clustered index entry uniquely */ for (i = 0; i < clust_index->n_uniq; i++) { field = dict_index_get_nth_field(clust_index, i); - if ((field->col)->aux == ULINT_UNDEFINED) { - dict_index_add_col(new_index, field->col, 0); + if (field->col->aux == ULINT_UNDEFINED) { + dict_index_add_col(new_index, field->col, 0, + field->prefix_len); } } @@ -1787,6 +1817,14 @@ dict_foreign_find_index( for (i = 0; i < n_cols; i++) { col_name = dict_index_get_nth_field(index, i) ->col->name; + if (dict_index_get_nth_field(index, i) + ->prefix_len != 0) { + /* We do not accept column prefix + indexes here */ + + break; + } + if (ut_strlen(columns[i]) != ut_strlen(col_name) || 0 != ut_cmp_in_lower_case(columns[i], @@ -3776,6 +3814,10 @@ dict_field_print_low( ut_ad(mutex_own(&(dict_sys->mutex))); printf(" %s", field->name); + + if (field->prefix_len != 0) { + printf("(%lu)", field->prefix_len); + } } /************************************************************************** diff --git a/innobase/dict/dict0load.c b/innobase/dict/dict0load.c index 8f39605e493..48c445fa0c9 100644 --- a/innobase/dict/dict0load.c +++ b/innobase/dict/dict0load.c @@ -301,6 +301,8 @@ dict_load_fields( dtuple_t* tuple; dfield_t* dfield; char* col_name; + ulint pos_and_prefix_len; + ulint prefix_len; rec_t* rec; byte* field; ulint len; @@ -345,8 +347,28 @@ dict_load_fields( ut_a(ut_memcmp(buf, field, len) == 0); field = rec_get_nth_field(rec, 1, &len); - ut_ad(len == 4); - ut_a(i == mach_read_from_4(field)); + ut_a(len == 4); + + /* The next field stores the field position in the index + and a possible column prefix length if the index field + does not contain the whole column. The storage format is + like this: if there is at least one prefix field in the index, + then the HIGH 2 bytes contain the field number (== i) and the + low 2 bytes the prefix length for the field. Otherwise the + field number (== i) is contained in the 2 LOW bytes. */ + + pos_and_prefix_len = mach_read_from_4(field); + + ut_a((pos_and_prefix_len & 0xFFFF) == i + || (pos_and_prefix_len & 0xFFFF0000) == (i << 16)); + + if ((i == 0 && pos_and_prefix_len > 0) + || (pos_and_prefix_len & 0xFFFF0000) > 0) { + + prefix_len = pos_and_prefix_len & 0xFFFF; + } else { + prefix_len = 0; + } ut_a(0 == ut_strcmp((char*) "COL_NAME", dict_field_get_col( @@ -359,7 +381,7 @@ dict_load_fields( ut_memcpy(col_name, field, len); col_name[len] = '\0'; - dict_mem_index_add_field(index, col_name, 0); + dict_mem_index_add_field(index, col_name, 0, prefix_len); btr_pcur_move_to_next_user_rec(&pcur, &mtr); } diff --git a/innobase/dict/dict0mem.c b/innobase/dict/dict0mem.c index e5918c6aeb6..56efc0a0117 100644 --- a/innobase/dict/dict0mem.c +++ b/innobase/dict/dict0mem.c @@ -266,10 +266,13 @@ by the column name may be released only after publishing the index. */ void dict_mem_index_add_field( /*=====================*/ - dict_index_t* index, /* in: index */ - char* name, /* in: column name */ - ulint order) /* in: order criterion; 0 means an ascending - order */ + dict_index_t* index, /* in: index */ + char* name, /* in: column name */ + ulint order, /* in: order criterion; 0 means an + ascending order */ + ulint prefix_len) /* in: 0 or the column prefix length + in a MySQL index like + INDEX (textcol(25)) */ { dict_field_t* field; @@ -282,6 +285,8 @@ dict_mem_index_add_field( field->name = name; field->order = order; + + field->prefix_len = prefix_len; } /************************************************************************** diff --git a/innobase/fil/fil0fil.c b/innobase/fil/fil0fil.c index 98980f6c337..a8dc357749c 100644 --- a/innobase/fil/fil0fil.c +++ b/innobase/fil/fil0fil.c @@ -632,7 +632,7 @@ fil_space_create( /* Spaces with an odd id number are reserved to replicate spaces used in log debugging */ - ut_a((purpose == FIL_LOG) || (id % 2 == 0)); + ut_anp((purpose == FIL_LOG) || (id % 2 == 0)); #endif mutex_enter(&(system->mutex)); @@ -1202,8 +1202,8 @@ loop: /* Do aio */ - ut_a(byte_offset % OS_FILE_LOG_BLOCK_SIZE == 0); - ut_a((len % OS_FILE_LOG_BLOCK_SIZE) == 0); + ut_anp(byte_offset % OS_FILE_LOG_BLOCK_SIZE == 0); + ut_anp((len % OS_FILE_LOG_BLOCK_SIZE) == 0); /* Queue the aio request */ ret = os_aio(type, mode | wake_later, node->name, node->handle, buf, diff --git a/innobase/fsp/fsp0fsp.c b/innobase/fsp/fsp0fsp.c index ee48288b875..b6941d80e90 100644 --- a/innobase/fsp/fsp0fsp.c +++ b/innobase/fsp/fsp0fsp.c @@ -778,7 +778,7 @@ fsp_init_file_page_low( page[i] = 0xFF; } #endif - mach_write_to_8(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN, + mach_write_to_8(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM, ut_dulint_zero); mach_write_to_8(page + FIL_PAGE_LSN, ut_dulint_zero); } @@ -2875,7 +2875,7 @@ fseg_free_step( freed yet */ ut_a(descr); - ut_a(xdes_get_bit(descr, XDES_FREE_BIT, buf_frame_get_page_no(header) + ut_anp(xdes_get_bit(descr, XDES_FREE_BIT, buf_frame_get_page_no(header) % FSP_EXTENT_SIZE, mtr) == FALSE); inode = fseg_inode_get(header, mtr); diff --git a/innobase/ha/ha0ha.c b/innobase/ha/ha0ha.c index 4489b25ec2b..eb28e15215d 100644 --- a/innobase/ha/ha0ha.c +++ b/innobase/ha/ha0ha.c @@ -293,11 +293,13 @@ ha_print_info( hash_table_t* table) /* in: hash table */ { hash_cell_t* cell; -/* ha_node_t* node; */ - ulint nodes = 0; - ulint cells = 0; +/* + ha_node_t* node; ulint len = 0; ulint max_len = 0; + ulint nodes = 0; +*/ + ulint cells = 0; ulint n_bufs; ulint i; diff --git a/innobase/ibuf/ibuf0ibuf.c b/innobase/ibuf/ibuf0ibuf.c index 187afa17047..c07756ab308 100644 --- a/innobase/ibuf/ibuf0ibuf.c +++ b/innobase/ibuf/ibuf0ibuf.c @@ -170,7 +170,7 @@ dropped! So, there seems to be no problem. */ /********************************************************************** Validates the ibuf data structures when the caller owns ibuf_mutex. */ -static + ibool ibuf_validate_low(void); /*===================*/ @@ -484,8 +484,8 @@ ibuf_data_init_for_space( index = dict_mem_index_create(buf, (char *) "CLUST_IND", space, DICT_CLUSTERED | DICT_UNIVERSAL | DICT_IBUF,2); - dict_mem_index_add_field(index, (char *) "PAGE_NO", 0); - dict_mem_index_add_field(index, (char *) "TYPES", 0); + dict_mem_index_add_field(index, (char *) "PAGE_NO", 0, 0); + dict_mem_index_add_field(index, (char *) "TYPES", 0, 0); index->page_no = FSP_IBUF_TREE_ROOT_PAGE_NO; @@ -2727,7 +2727,7 @@ reset_bit: /********************************************************************** Validates the ibuf data structures when the caller owns ibuf_mutex. */ -static + ibool ibuf_validate_low(void) /*===================*/ diff --git a/innobase/include/btr0cur.h b/innobase/include/btr0cur.h index 1d17c0e952d..506877333c3 100644 --- a/innobase/include/btr0cur.h +++ b/innobase/include/btr0cur.h @@ -690,7 +690,13 @@ and sleep this many microseconds in between */ #define BTR_CUR_RETRY_DELETE_N_TIMES 100 #define BTR_CUR_RETRY_SLEEP_TIME 50000 -/* The reference in a field of which data is stored on a different page */ +/* The reference in a field for which data is stored on a different page. +The reference is at the end of the 'locally' stored part of the field. +'Locally' means storage in the index record. +We store locally a long enough prefix of each column so that we can determine +the ordering parts of each index record without looking into the externally +stored part. */ + /*--------------------------------------*/ #define BTR_EXTERN_SPACE_ID 0 /* space id where stored */ #define BTR_EXTERN_PAGE_NO 4 /* page no where stored */ diff --git a/innobase/include/buf0buf.h b/innobase/include/buf0buf.h index e4d3671586d..2963efd6396 100644 --- a/innobase/include/buf0buf.h +++ b/innobase/include/buf0buf.h @@ -364,11 +364,24 @@ to a file. Note that we must be careful to calculate the same value on 32-bit and 64-bit architectures. */ ulint -buf_calc_page_checksum( -/*===================*/ +buf_calc_page_new_checksum( +/*=======================*/ /* out: checksum */ byte* page); /* in: buffer page */ /************************************************************************ +In versions < 4.0.14 and < 4.1.1 there was a bug that the checksum only +looked at the first few bytes of the page. This calculates that old +checksum. +NOTE: we must first store the new formula checksum to +FIL_PAGE_SPACE_OR_CHKSUM before calculating and storing this old checksum +because this takes that field as an input! */ + +ulint +buf_calc_page_old_checksum( +/*=======================*/ + /* out: checksum */ + byte* page); /* in: buffer page */ +/************************************************************************ Checks if a page is corrupt. */ ibool diff --git a/innobase/include/data0data.h b/innobase/include/data0data.h index e0fb06e5018..889d148d3fe 100644 --- a/innobase/include/data0data.h +++ b/innobase/include/data0data.h @@ -453,8 +453,6 @@ struct dfield_struct{ void* data; /* pointer to data */ ulint len; /* data length; UNIV_SQL_NULL if SQL null; */ dtype_t type; /* type of data */ - ulint col_no; /* when building index entries, the column - number can be stored here */ }; struct dtuple_struct { diff --git a/innobase/include/data0type.h b/innobase/include/data0type.h index b53a70a8909..5e28f657f0c 100644 --- a/innobase/include/data0type.h +++ b/innobase/include/data0type.h @@ -18,14 +18,16 @@ typedef struct dtype_struct dtype_t; data type */ extern dtype_t* dtype_binary; -/* Data main types of SQL data; NOTE! character data types requiring -collation transformation must have the smallest codes! All codes must be -less than 256! */ +/* Data main types of SQL data */ #define DATA_VARCHAR 1 /* character varying */ #define DATA_CHAR 2 /* fixed length character */ #define DATA_FIXBINARY 3 /* binary string of fixed length */ #define DATA_BINARY 4 /* binary string */ -#define DATA_BLOB 5 /* binary large object */ +#define DATA_BLOB 5 /* binary large object, or a TEXT type; if + prtype & DATA_NONLATIN1 != 0 the data must + be compared by MySQL as a whole field; if + prtype & DATA_BINARY_TYPE == 0, then this is + actually a TEXT column */ #define DATA_INT 6 /* integer: can be any size 1 - 8 bytes */ #define DATA_SYS_CHILD 7 /* address of the child page in node pointer */ #define DATA_SYS 8 /* system column */ @@ -34,35 +36,55 @@ binary strings */ #define DATA_FLOAT 9 #define DATA_DOUBLE 10 #define DATA_DECIMAL 11 /* decimal number stored as an ASCII string */ -#define DATA_VARMYSQL 12 /* data types for which comparisons must be */ -#define DATA_MYSQL 13 /* made by MySQL */ -#define DATA_ERROR 111 /* error value */ -#define DATA_MTYPE_MAX 255 +#define DATA_VARMYSQL 12 /* non-latin1 varying length char */ +#define DATA_MYSQL 13 /* non-latin1 fixed length char */ +#define DATA_MTYPE_MAX 63 /* dtype_store_for_order_and_null_size() + requires the values are <= 63 */ /*-------------------------------------------*/ -/* Precise data types for system columns; NOTE: the values must run -from 0 up in the order given! All codes must be less than 256! */ +/* In the lowest byte in the precise type we store the MySQL type code +(not applicable for system columns). */ + +#define DATA_ENGLISH 4 /* English language character string: this + is a relic from pre-MySQL time and only used + for InnoDB's own system tables */ +#define DATA_ERROR 111 /* another relic from pre-MySQL time */ + +#define DATA_MYSQL_TYPE_MASK 255 /* AND with this mask to extract the MySQL + type from the precise type */ + +/* Precise data types for system columns and the length of those columns; +NOTE: the values must run from 0 up in the order given! All codes must +be less than 256 */ #define DATA_ROW_ID 0 /* row id: a dulint */ #define DATA_ROW_ID_LEN 6 /* stored length for row id */ + #define DATA_TRX_ID 1 /* transaction id: 6 bytes */ #define DATA_TRX_ID_LEN 6 + #define DATA_ROLL_PTR 2 /* rollback data pointer: 7 bytes */ #define DATA_ROLL_PTR_LEN 7 + #define DATA_MIX_ID 3 /* mixed index label: a dulint, stored in a row in a compressed form */ #define DATA_MIX_ID_LEN 9 /* maximum stored length for mix id (in a compressed dulint form) */ #define DATA_N_SYS_COLS 4 /* number of system columns defined above */ +/*-------------------------------------------*/ +/* Flags ORed to the precise data type */ #define DATA_NOT_NULL 256 /* this is ORed to the precise type when the column is declared as NOT NULL */ #define DATA_UNSIGNED 512 /* this id ORed to the precise type when we have an unsigned integer type */ +#define DATA_BINARY_TYPE 1024 /* if the data type is a binary character + string, this is ORed to the precise type: + this only holds for tables created with + >= MySQL-4.0.14 */ +#define DATA_NONLATIN1 2048 /* if the data type is a character string + of a non-latin1 type, this is ORed to the + precise type: this only holds for tables + created with >= MySQL-4.0.14 */ /*-------------------------------------------*/ -/* Precise types of a char or varchar data. All codes must be less than 256! */ -#define DATA_ENGLISH 4 /* English language character string */ -#define DATA_FINNISH 5 /* Finnish */ -#define DATA_PRTYPE_MAX 255 - /* This many bytes we need to store the type information affecting the alphabetical order for a single field and decide the storage size of an SQL null*/ @@ -123,7 +145,7 @@ dtype_get_pad_char( /*===============*/ /* out: padding character code, or ULINT_UNDEFINED if no padding specified */ - dtype_t* type); /* in: typeumn */ + dtype_t* type); /* in: type */ /*************************************************************************** Returns the size of a fixed size data type, 0 if not a fixed size type. */ UNIV_INLINE @@ -150,24 +172,24 @@ dtype_is_fixed_size( /* out: TRUE if fixed size */ dtype_t* type); /* in: type */ /************************************************************************** -Stores to a type the information which determines its alphabetical -ordering. */ +Stores for a type the information which determines its alphabetical ordering +and the storage size of an SQL NULL value. */ UNIV_INLINE void dtype_store_for_order_and_null_size( /*================================*/ byte* buf, /* in: buffer for DATA_ORDER_NULL_TYPE_BUF_SIZE - bytes */ + bytes where we store the info */ dtype_t* type); /* in: type struct */ /************************************************************************** -Reads of a type the stored information which determines its alphabetical -ordering. */ +Reads to a type the stored information which determines its alphabetical +ordering and the storage size of an SQL NULL value. */ UNIV_INLINE void dtype_read_for_order_and_null_size( /*===============================*/ dtype_t* type, /* in: type struct */ - byte* buf); /* in: buffer for type order info */ + byte* buf); /* in: buffer for the stored order info */ /************************************************************************* Validates a data type structure. */ diff --git a/innobase/include/data0type.ic b/innobase/include/data0type.ic index d82d976d076..ddd0b0ae8cc 100644 --- a/innobase/include/data0type.ic +++ b/innobase/include/data0type.ic @@ -110,7 +110,9 @@ dtype_get_pad_char( if (type->mtype == DATA_CHAR || type->mtype == DATA_VARCHAR || type->mtype == DATA_BINARY - || type->mtype == DATA_FIXBINARY) { + || type->mtype == DATA_FIXBINARY + || type->mtype == DATA_MYSQL + || type->mtype == DATA_VARMYSQL) { /* Space is the padding character for all char and binary strings */ @@ -124,39 +126,56 @@ dtype_get_pad_char( } /************************************************************************** -Stores to a type the information which determines its alphabetical -ordering. */ +Stores for a type the information which determines its alphabetical ordering +and the storage size of an SQL NULL value. */ UNIV_INLINE void dtype_store_for_order_and_null_size( /*================================*/ byte* buf, /* in: buffer for DATA_ORDER_NULL_TYPE_BUF_SIZE - bytes */ + bytes where we store the info */ dtype_t* type) /* in: type struct */ { ut_ad(4 == DATA_ORDER_NULL_TYPE_BUF_SIZE); buf[0] = (byte)(type->mtype & 0xFF); + + if (type->prtype & DATA_BINARY_TYPE) { + buf[0] = buf[0] | 128; + } + + if (type->prtype & DATA_NONLATIN1) { + buf[0] = buf[0] | 64; + } + buf[1] = (byte)(type->prtype & 0xFF); mach_write_to_2(buf + 2, type->len & 0xFFFF); } /************************************************************************** -Reads of a type the stored information which determines its alphabetical -ordering. */ +Reads to a type the stored information which determines its alphabetical +ordering and the storage size of an SQL NULL value. */ UNIV_INLINE void dtype_read_for_order_and_null_size( /*===============================*/ dtype_t* type, /* in: type struct */ - byte* buf) /* in: buffer for type order info */ + byte* buf) /* in: buffer for stored type order info */ { ut_ad(4 == DATA_ORDER_NULL_TYPE_BUF_SIZE); - type->mtype = buf[0]; + type->mtype = buf[0] & 63; type->prtype = buf[1]; + if (buf[0] & 128) { + type->prtype = type->prtype | DATA_BINARY_TYPE; + } + + if (buf[0] & 64) { + type->prtype = type->prtype | DATA_NONLATIN1; + } + type->len = mach_read_from_2(buf + 2); } diff --git a/innobase/include/db0err.h b/innobase/include/db0err.h index ab7d0caa35c..854b9794c00 100644 --- a/innobase/include/db0err.h +++ b/innobase/include/db0err.h @@ -44,8 +44,10 @@ Created 5/24/1996 Heikki Tuuri #define DB_CORRUPTION 39 /* data structure corruption noticed */ #define DB_COL_APPEARS_TWICE_IN_INDEX 40 /* InnoDB cannot handle an index where same column appears twice */ -#define DB_CANNOT_DROP_CONSTRAINT 40 /* dropping a foreign key constraint +#define DB_CANNOT_DROP_CONSTRAINT 41 /* dropping a foreign key constraint from a table failed */ +#define DB_NO_SAVEPOINT 42 /* no savepoint exists with the given + name */ /* The following are partial failure codes */ #define DB_FAIL 1000 diff --git a/innobase/include/dict0dict.h b/innobase/include/dict0dict.h index 97486a7c2f6..e88c6a52bcb 100644 --- a/innobase/include/dict0dict.h +++ b/innobase/include/dict0dict.h @@ -569,6 +569,19 @@ dict_index_get_nth_col_pos( dict_index_t* index, /* in: index */ ulint n); /* in: column number */ /************************************************************************ +Looks for a matching field in an index. The column and the prefix len has +to be the same. */ + +ulint +dict_index_get_nth_field_pos( +/*=========================*/ + /* out: position in internal representation + of the index; if not contained, returns + ULINT_UNDEFINED */ + dict_index_t* index, /* in: index from which to search */ + dict_index_t* index2, /* in: index */ + ulint n); /* in: field number in index2 */ +/************************************************************************ Looks for column n position in the clustered index. */ ulint diff --git a/innobase/include/dict0mem.h b/innobase/include/dict0mem.h index 0798541cfe0..03dc913a7c9 100644 --- a/innobase/include/dict0mem.h +++ b/innobase/include/dict0mem.h @@ -111,10 +111,13 @@ by the column name may be released only after publishing the index. */ void dict_mem_index_add_field( /*=====================*/ - dict_index_t* index, /* in: index */ - char* name, /* in: column name */ - ulint order); /* in: order criterion; 0 means an ascending - order */ + dict_index_t* index, /* in: index */ + char* name, /* in: column name */ + ulint order, /* in: order criterion; 0 means an + ascending order */ + ulint prefix_len); /* in: 0 or the column prefix length + in a MySQL index like + INDEX (textcol(25)) */ /************************************************************************** Frees an index memory object. */ @@ -158,12 +161,18 @@ struct dict_col_struct{ in some of the functions below */ }; +#define DICT_MAX_COL_PREFIX_LEN 512 + /* Data structure for a field in an index */ struct dict_field_struct{ - dict_col_t* col; /* pointer to the table column */ - char* name; /* name of the column */ - ulint order; /* flags for ordering this field: - DICT_DESCEND, ... */ + dict_col_t* col; /* pointer to the table column */ + char* name; /* name of the column */ + ulint order; /* flags for ordering this field: + DICT_DESCEND, ... */ + ulint prefix_len; /* 0 or the length of the column + prefix in a MySQL index of type, e.g., + INDEX (textcol(25)); must be smaller + than DICT_MAX_COL_PREFIX_LEN */ }; /* Data structure for an index tree */ diff --git a/innobase/include/fil0fil.h b/innobase/include/fil0fil.h index 23ef0304b2d..4f78fdb2fd7 100644 --- a/innobase/include/fil0fil.h +++ b/innobase/include/fil0fil.h @@ -43,7 +43,10 @@ struct fil_addr_struct{ extern fil_addr_t fil_addr_null; /* The byte offsets on a file page for various variables */ -#define FIL_PAGE_SPACE 0 /* space id the page belongs to */ +#define FIL_PAGE_SPACE_OR_CHKSUM 0 /* in < MySQL-4.0.14 space id the + page belongs to (== 0) but in later + versions the 'new' checksum of the + page */ #define FIL_PAGE_OFFSET 4 /* page offset inside space */ #define FIL_PAGE_PREV 8 /* if there is a 'natural' predecessor of the page, its offset */ @@ -64,7 +67,7 @@ extern fil_addr_t fil_addr_null; #define FIL_PAGE_DATA 38 /* start of the data on the page */ /* File page trailer */ -#define FIL_PAGE_END_LSN 8 /* the low 4 bytes of this are used +#define FIL_PAGE_END_LSN_OLD_CHKSUM 8 /* the low 4 bytes of this are used to store the page checksum, the last 4 bytes should be identical to the last 4 bytes of FIL_PAGE_LSN */ diff --git a/innobase/include/lock0lock.h b/innobase/include/lock0lock.h index d3b3d55d015..5608ba020b7 100644 --- a/innobase/include/lock0lock.h +++ b/innobase/include/lock0lock.h @@ -450,6 +450,18 @@ lock_rec_get_mutex_for_addr( ulint space, /* in: space id */ ulint page_no);/* in: page number */ /************************************************************************* +Checks that a transaction id is sensible, i.e., not in the future. */ + +ibool +lock_check_trx_id_sanity( +/*=====================*/ + /* out: TRUE if ok */ + dulint trx_id, /* in: trx id */ + rec_t* rec, /* in: user record */ + dict_index_t* index, /* in: clustered index */ + ibool has_kernel_mutex);/* in: TRUE if the caller owns the + kernel mutex */ +/************************************************************************* Validates the lock queue on a single record. */ ibool diff --git a/innobase/include/os0file.h b/innobase/include/os0file.h index 86f27a2d3eb..1ec8b71d069 100644 --- a/innobase/include/os0file.h +++ b/innobase/include/os0file.h @@ -146,6 +146,21 @@ os_file_create_simple( ulint access_type,/* in: OS_FILE_READ_ONLY or OS_FILE_READ_WRITE */ ibool* success);/* out: TRUE if succeed, FALSE if error */ /******************************************************************** +A simple function to open or create a file. */ + +os_file_t +os_file_create_simple_no_error_handling( +/*====================================*/ + /* out, own: handle to the file, not defined if error, + error number can be retrieved with os_get_last_error */ + char* name, /* in: name of the file or path as a null-terminated + string */ + ulint create_mode,/* in: OS_FILE_OPEN if an existing file is opened + (if does not exist, error), or OS_FILE_CREATE if a new + file is created (if exists, error) */ + ulint access_type,/* in: OS_FILE_READ_ONLY or OS_FILE_READ_WRITE */ + ibool* success);/* out: TRUE if succeed, FALSE if error */ +/******************************************************************** Opens an existing file or creates a new. */ os_file_t @@ -173,6 +188,14 @@ os_file_close( /* out: TRUE if success */ os_file_t file); /* in, own: handle to a file */ /*************************************************************************** +Closes a file handle. */ + +ibool +os_file_close_no_error_handling( +/*============================*/ + /* out: TRUE if success */ + os_file_t file); /* in, own: handle to a file */ +/*************************************************************************** Gets a file size. */ ibool diff --git a/innobase/include/page0page.h b/innobase/include/page0page.h index b5e33af5bc0..04f771c3abd 100644 --- a/innobase/include/page0page.h +++ b/innobase/include/page0page.h @@ -666,6 +666,15 @@ page_rec_validate( /* out: TRUE if ok */ rec_t* rec); /* in: record on the page */ /******************************************************************* +Checks that the first directory slot points to the infimum record and +the last to the supremum. This function is intended to track if the +bug fixed in 4.0.14 has caused corruption to users' databases. */ + +void +page_check_dir( +/*===========*/ + page_t* page); /* in: index page */ +/******************************************************************* This function checks the consistency of an index page when we do not know the index. This is also resilient so that this should never crash even if the page is total garbage. */ diff --git a/innobase/include/rem0cmp.h b/innobase/include/rem0cmp.h index 6f2a99fc8c2..712e263350e 100644 --- a/innobase/include/rem0cmp.h +++ b/innobase/include/rem0cmp.h @@ -42,6 +42,22 @@ cmp_data_data( buffer) */ ulint len2); /* in: data field length or UNIV_SQL_NULL */ /***************************************************************** +This function is used to compare two data fields for which we know the +data type. */ + +int +cmp_data_data_slow( +/*===============*/ + /* out: 1, 0, -1, if data1 is greater, equal, + less than data2, respectively */ + dtype_t* cur_type,/* in: data type of the fields */ + byte* data1, /* in: data field (== a pointer to a memory + buffer) */ + ulint len1, /* in: data field length or UNIV_SQL_NULL */ + byte* data2, /* in: data field (== a pointer to a memory + buffer) */ + ulint len2); /* in: data field length or UNIV_SQL_NULL */ +/***************************************************************** This function is used to compare two dfields where at least the first has its data type field set. */ UNIV_INLINE diff --git a/innobase/include/row0mysql.ic b/innobase/include/row0mysql.ic index e9d493da8b5..4ecd66e06ec 100644 --- a/innobase/include/row0mysql.ic +++ b/innobase/include/row0mysql.ic @@ -58,7 +58,8 @@ row_mysql_store_col_in_innobase_format( /*===================================*/ dfield_t* dfield, /* in/out: dfield */ byte* buf, /* in/out: buffer for the converted - value */ + value; this must be at least col_len + long! */ byte* mysql_data, /* in: MySQL column value, not SQL NULL; NOTE that dfield may also get a pointer to mysql_data, @@ -96,7 +97,6 @@ row_mysql_store_col_in_innobase_format( while (col_len > 0 && ptr[col_len - 1] == ' ') { col_len--; } - } else if (type == DATA_BLOB) { ptr = row_mysql_read_blob_ref(&col_len, mysql_data, col_len); } diff --git a/innobase/include/row0row.h b/innobase/include/row0row.h index 09a79e19fd7..d1befbbbad3 100644 --- a/innobase/include/row0row.h +++ b/innobase/include/row0row.h @@ -86,9 +86,10 @@ dtuple_t* row_build( /*======*/ /* out, own: row built; see the NOTE below! */ - ulint type, /* in: ROW_COPY_DATA, or ROW_COPY_POINTERS: - the former copies also the data fields to - heap as the latter only places pointers to + ulint type, /* in: ROW_COPY_POINTERS, ROW_COPY_DATA, or + ROW_COPY_ALSO_EXTERNALS, + the two last copy also the data fields to + heap as the first only places pointers to data fields on the index page, and thus is more efficient */ dict_index_t* index, /* in: clustered index */ diff --git a/innobase/include/row0sel.h b/innobase/include/row0sel.h index aa2da6fe5f6..5ef7ff9399a 100644 --- a/innobase/include/row0sel.h +++ b/innobase/include/row0sel.h @@ -87,9 +87,11 @@ row_printf_step( /* out: query thread to run next or NULL */ que_thr_t* thr); /* in: query thread */ /******************************************************************** -Converts a key value stored in MySQL format to an Innobase dtuple. -The last field of the key value may be just a prefix of a fixed length -field: hence the parameter key_len. */ +Converts a key value stored in MySQL format to an Innobase dtuple. The last +field of the key value may be just a prefix of a fixed length field: hence +the parameter key_len. But currently we do not allow search keys where the +last field is only a prefix of the full key field len and print a warning if +such appears. */ void row_sel_convert_mysql_key_to_innobase( @@ -100,6 +102,7 @@ row_sel_convert_mysql_key_to_innobase( to index! */ byte* buf, /* in: buffer to use in field conversions */ + ulint buf_len, /* in: buffer length */ dict_index_t* index, /* in: index of the key value */ byte* key_ptr, /* in: MySQL key value */ ulint key_len); /* in: MySQL key value length */ diff --git a/innobase/include/row0upd.h b/innobase/include/row0upd.h index 273ec6074eb..473c55c7ef9 100644 --- a/innobase/include/row0upd.h +++ b/innobase/include/row0upd.h @@ -114,13 +114,15 @@ row_upd_index_write_log( closed within this function */ mtr_t* mtr); /* in: mtr into whose log to write */ /*************************************************************** -Returns TRUE if row update changes size of some field in index. */ +Returns TRUE if row update changes size of some field in index or if some +field to be updated is stored externally in rec or update. */ ibool -row_upd_changes_field_size( -/*=======================*/ +row_upd_changes_field_size_or_external( +/*===================================*/ /* out: TRUE if the update changes the size of - some field in index */ + some field in index or the field is external + in rec or update */ rec_t* rec, /* in: record in clustered index */ dict_index_t* index, /* in: clustered index */ upd_t* update);/* in: update vector */ @@ -175,16 +177,10 @@ row_upd_index_replace_new_col_vals( dtuple_t* entry, /* in/out: index entry where replaced */ dict_index_t* index, /* in: index; NOTE that may also be a non-clustered index */ - upd_t* update); /* in: update vector */ -/*************************************************************** -Replaces the new column values stored in the update vector to the -clustered index entry given. */ - -void -row_upd_clust_index_replace_new_col_vals( -/*=====================================*/ - dtuple_t* entry, /* in/out: index entry where replaced */ - upd_t* update); /* in: update vector */ + upd_t* update, /* in: update vector */ + mem_heap_t* heap); /* in: memory heap to which we allocate and + copy the new values, set this as NULL if you + do not want allocation */ /*************************************************************** Checks if an update vector changes an ordering field of an index record. This function is fast if the update vector is short or the number of ordering @@ -358,9 +354,9 @@ struct upd_node_struct{ externally in the clustered index record of row */ ulint n_ext_vec;/* number of fields in ext_vec */ - mem_heap_t* heap; /* memory heap used as auxiliary storage for - row; this must be emptied after a successful - update if node->row != NULL */ + mem_heap_t* heap; /* memory heap used as auxiliary storage; + this must be emptied after a successful + update */ /*----------------------*/ sym_node_t* table_sym;/* table node in symbol table */ que_node_t* col_assign_list; diff --git a/innobase/include/srv0srv.h b/innobase/include/srv0srv.h index 24e692dedab..1e54d7bfc35 100644 --- a/innobase/include/srv0srv.h +++ b/innobase/include/srv0srv.h @@ -153,6 +153,7 @@ extern mutex_t* kernel_mutex_temp;/* mutex protecting the server, trx structs, /* Array of English strings describing the current state of an i/o handler thread */ extern char* srv_io_thread_op_info[]; +extern char* srv_io_thread_function[]; typedef struct srv_sys_struct srv_sys_t; diff --git a/innobase/include/trx0roll.h b/innobase/include/trx0roll.h index 820af4cd014..0d7126c9c57 100644 --- a/innobase/include/trx0roll.h +++ b/innobase/include/trx0roll.h @@ -177,6 +177,55 @@ trx_general_rollback_for_mysql( ibool partial,/* in: TRUE if partial rollback requested */ trx_savept_t* savept);/* in: pointer to savepoint undo number, if partial rollback requested */ +/*********************************************************************** +Rolls back a transaction back to a named savepoint. Modifications after the +savepoint are undone but InnoDB does NOT release the corresponding locks +which are stored in memory. If a lock is 'implicit', that is, a new inserted +row holds a lock where the lock information is carried by the trx id stored in +the row, these locks are naturally released in the rollback. Savepoints which +were set after this savepoint are deleted. */ + +ulint +trx_rollback_to_savepoint_for_mysql( +/*================================*/ + /* out: if no savepoint + of the name found then + DB_NO_SAVEPOINT, + otherwise DB_SUCCESS */ + trx_t* trx, /* in: transaction handle */ + char* savepoint_name, /* in: savepoint name */ + ib_longlong* mysql_binlog_cache_pos);/* out: the MySQL binlog cache + position corresponding to this + savepoint; MySQL needs this + information to remove the + binlog entries of the queries + executed after the savepoint */ +/*********************************************************************** +Creates a named savepoint. If the transaction is not yet started, starts it. +If there is already a savepoint of the same name, this call erases that old +savepoint and replaces it with a new. Savepoints are deleted in a transaction +commit or rollback. */ + +ulint +trx_savepoint_for_mysql( +/*====================*/ + /* out: always DB_SUCCESS */ + trx_t* trx, /* in: transaction handle */ + char* savepoint_name, /* in: savepoint name */ + ib_longlong binlog_cache_pos); /* in: MySQL binlog cache + position corresponding to this + connection at the time of the + savepoint */ +/*********************************************************************** +Frees savepoint structs. */ + +void +trx_roll_savepoints_free( +/*=====================*/ + trx_t* trx, /* in: transaction handle */ + trx_named_savept_t* savep); /* in: free all savepoints > this one; + if this is NULL, free all savepoints + of trx */ extern sess_t* trx_dummy_sess; @@ -207,6 +256,21 @@ struct roll_node_struct{ case of a partial rollback */ }; +/* A savepoint set with SQL's "SAVEPOINT savepoint_id" command */ +struct trx_named_savept_struct{ + char* name; /* savepoint name */ + trx_savept_t savept; /* the undo number corresponding to + the savepoint */ + ib_longlong mysql_binlog_cache_pos; + /* the MySQL binlog cache position + corresponding to this savepoint, not + defined if the MySQL binlogging is not + enabled */ + UT_LIST_NODE_T(trx_named_savept_t) + trx_savepoints; /* the list of savepoints of a + transaction */ +}; + /* Rollback node states */ #define ROLL_NODE_SEND 1 #define ROLL_NODE_WAIT 2 diff --git a/innobase/include/trx0sys.ic b/innobase/include/trx0sys.ic index ada2d8cb19c..343e6d7c2fa 100644 --- a/innobase/include/trx0sys.ic +++ b/innobase/include/trx0sys.ic @@ -296,6 +296,16 @@ trx_is_active( return(FALSE); } + if (ut_dulint_cmp(trx_id, trx_sys->max_trx_id) >= 0) { + + /* There must be corruption: we return TRUE because this + function is only called by lock_clust_rec_some_has_impl() + and row_vers_impl_x_locked_off_kernel() and they have + diagnostic prints in this case */ + + return(TRUE); + } + trx = trx_get_on_id(trx_id); if (trx && (trx->conc_state == TRX_ACTIVE)) { diff --git a/innobase/include/trx0trx.h b/innobase/include/trx0trx.h index 39229923375..6b08b674db8 100644 --- a/innobase/include/trx0trx.h +++ b/innobase/include/trx0trx.h @@ -381,7 +381,8 @@ struct trx_struct{ replication slave, we have here the master binlog name up to which replication has processed; otherwise - this is a pointer to a null character */ + this is a pointer to a null + character */ ib_longlong mysql_master_log_pos; /* if the database server is a MySQL replication slave, this is the @@ -501,6 +502,10 @@ struct trx_struct{ mem_heap_t* read_view_heap; /* memory heap for the read view */ read_view_t* read_view; /* consistent read view or NULL */ /*------------------------------*/ + UT_LIST_BASE_NODE_T(trx_named_savept_t) + trx_savepoints; /* savepoints set with SAVEPOINT ..., + oldest first */ + /*------------------------------*/ mutex_t undo_mutex; /* mutex protecting the fields in this section (down to undo_no_arr), EXCEPT last_sql_stat_start, which can be diff --git a/innobase/include/trx0types.h b/innobase/include/trx0types.h index b8befe7172f..2965eb4451f 100644 --- a/innobase/include/trx0types.h +++ b/innobase/include/trx0types.h @@ -24,6 +24,7 @@ typedef struct trx_undo_inf_struct trx_undo_inf_t; typedef struct trx_purge_struct trx_purge_t; typedef struct roll_node_struct roll_node_t; typedef struct commit_node_struct commit_node_t; +typedef struct trx_named_savept_struct trx_named_savept_t; /* Transaction savepoint */ typedef struct trx_savept_struct trx_savept_t; diff --git a/innobase/include/ut0dbg.h b/innobase/include/ut0dbg.h index e99dc8c09d6..802557099fc 100644 --- a/innobase/include/ut0dbg.h +++ b/innobase/include/ut0dbg.h @@ -50,6 +50,37 @@ extern ulint* ut_dbg_null_ptr; }\ } +/* This can be used if there are % characters in the assertion formula: +if we try to printf the formula gcc would complain of illegal print +format characters */ +#define ut_anp(EXPR)\ +{\ + ulint dbg_i;\ +\ + if (!((ulint)(EXPR) + ut_dbg_zero)) {\ + ut_print_timestamp(stderr);\ + fprintf(stderr,\ + " InnoDB: Assertion failure in thread %lu in file %s line %lu\n",\ + os_thread_pf(os_thread_get_curr_id()), IB__FILE__,\ + (ulint)__LINE__);\ + fprintf(stderr,\ + "\nInnoDB: We intentionally generate a memory trap.\n");\ + fprintf(stderr,\ + "InnoDB: Send a detailed bug report to mysql@lists.mysql.com\n");\ + ut_dbg_stop_threads = TRUE;\ + dbg_i = *(ut_dbg_null_ptr);\ + if (dbg_i) {\ + ut_dbg_null_ptr = NULL;\ + }\ + }\ + if (ut_dbg_stop_threads) {\ + fprintf(stderr,\ + "InnoDB: Thread %lu stopped in file %s line %lu\n",\ + os_thread_pf(os_thread_get_curr_id()), IB__FILE__, (ulint)__LINE__);\ + os_thread_sleep(1000000000);\ + }\ +} + #define ut_error {\ ulint dbg_i;\ ut_print_timestamp(stderr);\ diff --git a/innobase/include/ut0mem.h b/innobase/include/ut0mem.h index 09e0d800685..4e8566eba1b 100644 --- a/innobase/include/ut0mem.h +++ b/innobase/include/ut0mem.h @@ -57,7 +57,7 @@ ut_free( /*====*/ void* ptr); /* in, own: memory block */ /************************************************************************** -Frees all allocated memory not freed yet. */ +Frees in shutdown all allocated memory not freed yet. */ void ut_free_all_mem(void); diff --git a/innobase/lock/lock0lock.c b/innobase/lock/lock0lock.c index 4bb1d243ed4..fecb1f95c68 100644 --- a/innobase/lock/lock0lock.c +++ b/innobase/lock/lock0lock.c @@ -356,7 +356,7 @@ lock_mutex_enter_kernel(void) } /************************************************************************* -Releses the kernel mutex. This function is used in this module to allow +Releases the kernel mutex. This function is used in this module to allow monitoring the contention degree on the kernel mutex caused by the lock operations. */ UNIV_INLINE @@ -515,6 +515,53 @@ lock_rec_mutex_own_all(void) #endif /************************************************************************* +Checks that a transaction id is sensible, i.e., not in the future. */ + +ibool +lock_check_trx_id_sanity( +/*=====================*/ + /* out: TRUE if ok */ + dulint trx_id, /* in: trx id */ + rec_t* rec, /* in: user record */ + dict_index_t* index, /* in: clustered index */ + ibool has_kernel_mutex)/* in: TRUE if the caller owns the + kernel mutex */ +{ + char err_buf[500]; + ibool is_ok = TRUE; + + if (!has_kernel_mutex) { + mutex_enter(&kernel_mutex); + } + + /* A sanity check: the trx_id in rec must be smaller than the global + trx id counter */ + + if (ut_dulint_cmp(trx_id, trx_sys->max_trx_id) >= 0) { + rec_sprintf(err_buf, 400, rec); + ut_print_timestamp(stderr); + fprintf(stderr, +"InnoDB: Error: transaction id associated with record\n%s\n" +"InnoDB: in table %s index %s\n" +"InnoDB: is %lu %lu which is higher than the global trx id counter %lu %lu!\n" +"InnoDB: The table is corrupt. You have to do dump + drop + reimport.\n", + err_buf, index->table_name, index->name, + ut_dulint_get_high(trx_id), + ut_dulint_get_low(trx_id), + ut_dulint_get_high(trx_sys->max_trx_id), + ut_dulint_get_low(trx_sys->max_trx_id)); + + is_ok = FALSE; + } + + if (!has_kernel_mutex) { + mutex_exit(&kernel_mutex); + } + + return(is_ok); +} + +/************************************************************************* Checks that a record is seen in a consistent read. */ ibool @@ -539,6 +586,15 @@ lock_clust_rec_cons_read_sees( return(TRUE); } + if (!lock_check_trx_id_sanity(trx_id, rec, index, FALSE)) { + /* Trying to get the 'history' of a corrupt record is bound + to fail: let us try to use the record itself in the query */ + fprintf(stderr, +"InnoDB: We try to access the corrupt record in the query anyway.\n"); + + return(TRUE); + } + return(FALSE); } @@ -562,7 +618,9 @@ lock_sec_rec_cons_read_sees( read_view_t* view) /* in: consistent read view */ { dulint max_trx_id; - + + UT_NOT_USED(index); + ut_ad(!(index->type & DICT_CLUSTERED)); ut_ad(page_rec_is_user_rec(rec)); @@ -575,6 +633,16 @@ lock_sec_rec_cons_read_sees( if (ut_dulint_cmp(max_trx_id, view->up_limit_id) >= 0) { + if (!lock_check_trx_id_sanity(max_trx_id, rec, index, FALSE)) { + /* Trying to get the 'history' of a corrupt record is + bound to fail: let us try to use the record itself in + the query */ + fprintf(stderr, +"InnoDB: We try to access the corrupt record in the query anyway.\n"); + + return(TRUE); + } + return(FALSE); } @@ -1569,6 +1637,15 @@ lock_sec_rec_some_has_impl_off_kernel( /* Ok, in this case it is possible that some transaction has an implicit x-lock. We have to look in the clustered index. */ + if (!lock_check_trx_id_sanity(page_get_max_trx_id(page), rec, index, + TRUE)) { + buf_page_print(page); + + /* The page is corrupt: try to avoid a crash by returning + NULL */ + return(NULL); + } + return(row_vers_impl_x_locked_off_kernel(rec, index)); } @@ -2565,7 +2642,7 @@ lock_move_rec_list_start( ulint heap_no; ulint type_mode; - ut_ad(new_page); + ut_a(new_page); lock_mutex_enter_kernel(); @@ -3028,7 +3105,7 @@ lock_deadlock_recursive( we return LOCK_VICTIM_IS_START */ { lock_t* lock; - ulint bit_no; + ulint bit_no = ULINT_UNDEFINED; trx_t* lock_trx; char* err_buf; ulint ret; @@ -3067,6 +3144,7 @@ lock_deadlock_recursive( lock = UT_LIST_GET_PREV(un_member.tab_lock.locks, lock); } else { ut_ad(lock_get_type(lock) == LOCK_REC); + ut_a(bit_no != ULINT_UNDEFINED); lock = lock_rec_get_prev(lock, bit_no); } diff --git a/innobase/log/log0log.c b/innobase/log/log0log.c index e15812e03af..b5ce1a3d97b 100644 --- a/innobase/log/log0log.c +++ b/innobase/log/log0log.c @@ -375,7 +375,7 @@ log_pad_current_log_block(void) log_close(); log_release(); - ut_ad((ut_dulint_get_low(lsn) % OS_FILE_LOG_BLOCK_SIZE) + ut_anp((ut_dulint_get_low(lsn) % OS_FILE_LOG_BLOCK_SIZE) == LOG_BLOCK_HDR_SIZE); } @@ -1070,8 +1070,8 @@ log_group_write_buf( ulint i; ut_ad(mutex_own(&(log_sys->mutex))); - ut_ad(len % OS_FILE_LOG_BLOCK_SIZE == 0); - ut_ad(ut_dulint_get_low(start_lsn) % OS_FILE_LOG_BLOCK_SIZE == 0); + ut_anp(len % OS_FILE_LOG_BLOCK_SIZE == 0); + ut_anp(ut_dulint_get_low(start_lsn) % OS_FILE_LOG_BLOCK_SIZE == 0); if (new_data_offset == 0) { write_header = TRUE; @@ -2123,11 +2123,11 @@ log_group_archive( start_lsn = log_sys->archived_lsn; - ut_ad(ut_dulint_get_low(start_lsn) % OS_FILE_LOG_BLOCK_SIZE == 0); + ut_anp(ut_dulint_get_low(start_lsn) % OS_FILE_LOG_BLOCK_SIZE == 0); end_lsn = log_sys->next_archived_lsn; - ut_ad(ut_dulint_get_low(end_lsn) % OS_FILE_LOG_BLOCK_SIZE == 0); + ut_anp(ut_dulint_get_low(end_lsn) % OS_FILE_LOG_BLOCK_SIZE == 0); buf = log_sys->archive_buf; @@ -2234,7 +2234,7 @@ loop: group->next_archived_file_no = group->archived_file_no + n_files; group->next_archived_offset = next_offset % group->file_size; - ut_ad(group->next_archived_offset % OS_FILE_LOG_BLOCK_SIZE == 0); + ut_anp(group->next_archived_offset % OS_FILE_LOG_BLOCK_SIZE == 0); } /********************************************************* @@ -2429,8 +2429,8 @@ loop: start_lsn = log_sys->archived_lsn; if (calc_new_limit) { - ut_ad(log_sys->archive_buf_size % OS_FILE_LOG_BLOCK_SIZE == 0); - + ut_anp(log_sys->archive_buf_size % OS_FILE_LOG_BLOCK_SIZE + == 0); limit_lsn = ut_dulint_add(start_lsn, log_sys->archive_buf_size); @@ -2916,6 +2916,7 @@ loop: mutex_enter(&kernel_mutex); + /* Check that there are no longer transactions */ if (trx_n_mysql_transactions > 0 || UT_LIST_GET_LEN(trx_sys->trx_list) > 0) { @@ -2924,6 +2925,8 @@ loop: goto loop; } + /* Check that the master thread is suspended */ + if (srv_n_threads_active[SRV_MASTER] != 0) { mutex_exit(&kernel_mutex); @@ -2952,7 +2955,6 @@ loop: } log_archive_all(); - log_make_checkpoint_at(ut_dulint_max, TRUE); mutex_enter(&(log_sys->mutex)); @@ -2961,8 +2963,9 @@ loop: if (ut_dulint_cmp(lsn, log_sys->last_checkpoint_lsn) != 0 || (srv_log_archive_on - && ut_dulint_cmp(lsn, - ut_dulint_add(log_sys->archived_lsn, LOG_BLOCK_HDR_SIZE)) != 0)) { + && ut_dulint_cmp(lsn, + ut_dulint_add(log_sys->archived_lsn, LOG_BLOCK_HDR_SIZE)) + != 0)) { mutex_exit(&(log_sys->mutex)); @@ -2981,10 +2984,22 @@ loop: mutex_exit(&(log_sys->mutex)); + mutex_enter(&kernel_mutex); + /* Check that the master thread has stayed suspended */ + if (srv_n_threads_active[SRV_MASTER] != 0) { + fprintf(stderr, +"InnoDB: Warning: the master thread woke up during shutdown\n"); + + mutex_exit(&kernel_mutex); + + goto loop; + } + mutex_exit(&kernel_mutex); + fil_flush_file_spaces(FIL_TABLESPACE); fil_flush_file_spaces(FIL_LOG); - /* The following fil_write_... will pass the buffer pool: therefore + /* The next fil_write_... will pass the buffer pool: therefore it is essential that the buffer pool has been completely flushed to disk! */ @@ -2993,12 +3008,14 @@ loop: goto loop; } + /* The lock timeout thread should now have exited */ + if (srv_lock_timeout_and_monitor_active) { goto loop; } - /* We now suspend also the InnoDB error monitor thread */ + /* We now let also the InnoDB error monitor thread to exit */ srv_shutdown_state = SRV_SHUTDOWN_LAST_PHASE; @@ -3008,6 +3025,7 @@ loop: } /* Make some checks that the server really is quiet */ + ut_a(srv_n_threads_active[SRV_MASTER] == 0); ut_a(buf_all_freed()); ut_a(0 == ut_dulint_cmp(lsn, log_sys->lsn)); @@ -3016,6 +3034,7 @@ loop: fil_flush_file_spaces(FIL_TABLESPACE); /* Make some checks that the server really is quiet */ + ut_a(srv_n_threads_active[SRV_MASTER] == 0); ut_a(buf_all_freed()); ut_a(0 == ut_dulint_cmp(lsn, log_sys->lsn)); } diff --git a/innobase/log/log0recv.c b/innobase/log/log0recv.c index 47833214d15..8e5fe819afb 100644 --- a/innobase/log/log0recv.c +++ b/innobase/log/log0recv.c @@ -973,7 +973,7 @@ recv_recover_page( ulint space, /* in: space id */ ulint page_no) /* in: page number */ { - buf_block_t* block; + buf_block_t* block = NULL; recv_addr_t* recv_addr; recv_t* recv; byte* buf; @@ -1085,7 +1085,7 @@ recv_recover_page( page_lsn = page_newest_lsn; mach_write_to_8(page + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN, ut_dulint_zero); + - FIL_PAGE_END_LSN_OLD_CHKSUM, ut_dulint_zero); mach_write_to_8(page + FIL_PAGE_LSN, ut_dulint_zero); } @@ -1107,7 +1107,7 @@ recv_recover_page( recv_parse_or_apply_log_rec_body(recv->type, buf, buf + recv->len, page, &mtr); mach_write_to_8(page + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN, + - FIL_PAGE_END_LSN_OLD_CHKSUM, ut_dulint_add(recv->start_lsn, recv->len)); mach_write_to_8(page + FIL_PAGE_LSN, @@ -1132,6 +1132,8 @@ recv_recover_page( mutex_exit(&(recv_sys->mutex)); if (!recover_backup && modification_to_page) { + ut_a(block); + buf_flush_recv_note_modification(block, start_lsn, end_lsn); } @@ -1339,6 +1341,7 @@ loop: mutex_exit(&(recv_sys->mutex)); } +#ifdef UNIV_HOTBACKUP /*********************************************************************** Applies log records in the hash table to a backup. */ @@ -1520,8 +1523,8 @@ recv_check_identical( for (i = 0; i < len; i++) { if (str1[i] != str2[i]) { - fprintf(stderr, "Strings do not match at offset %lu\n", i); - + fprintf(stderr, + "Strings do not match at offset %lu\n", i); ut_print_buf(str1 + i, 16); fprintf(stderr, "\n"); ut_print_buf(str2 + i, 16); @@ -1654,6 +1657,7 @@ recv_compare_spaces_low( recv_compare_spaces(space1, space2, n_pages); } +#endif /*********************************************************************** Tries to parse a single log record and returns its length. */ diff --git a/innobase/mem/mem0pool.c b/innobase/mem/mem0pool.c index 382e505b63f..b004a8c4df7 100644 --- a/innobase/mem/mem0pool.c +++ b/innobase/mem/mem0pool.c @@ -99,6 +99,12 @@ mem_pool_t* mem_comm_pool = NULL; ulint mem_out_of_mem_err_msg_count = 0; +/* We use this counter to check that the mem pool mutex does not leak; +this is to track a strange assertion failure reported at +mysql@lists.mysql.com */ + +ulint mem_n_threads_inside = 0; + /************************************************************************ Reserves the mem pool mutex. */ @@ -328,6 +334,9 @@ mem_area_alloc( n = ut_2_log(ut_max(size + MEM_AREA_EXTRA_SIZE, MEM_AREA_MIN_SIZE)); mutex_enter(&(pool->mutex)); + mem_n_threads_inside++; + + ut_a(mem_n_threads_inside == 1); area = UT_LIST_GET_FIRST(pool->free_list[n]); @@ -338,6 +347,7 @@ mem_area_alloc( /* Out of memory in memory pool: we try to allocate from the operating system with the regular malloc: */ + mem_n_threads_inside--; mutex_exit(&(pool->mutex)); return(ut_malloc(size)); @@ -353,6 +363,16 @@ mem_area_alloc( n); mem_analyze_corruption((byte*)area); + + /* Try to analyze a strange assertion failure reported at + mysql@lists.mysql.com where the free bit IS 1 in the + hex dump above */ + + if (mem_area_get_free(area)) { + fprintf(stderr, +"InnoDB: Probably a race condition because now the area is marked free!\n"); + } + ut_a(0); } @@ -374,6 +394,7 @@ mem_area_alloc( pool->reserved += mem_area_get_size(area); + mem_n_threads_inside--; mutex_exit(&(pool->mutex)); ut_ad(mem_pool_validate(pool)); @@ -495,6 +516,9 @@ mem_area_free( n = ut_2_log(size); mutex_enter(&(pool->mutex)); + mem_n_threads_inside++; + + ut_a(mem_n_threads_inside == 1); if (buddy && mem_area_get_free(buddy) && (size == mem_area_get_size(buddy))) { @@ -518,6 +542,7 @@ mem_area_free( pool->reserved += ut_2_exp(n); + mem_n_threads_inside--; mutex_exit(&(pool->mutex)); mem_area_free(new_ptr, pool); @@ -533,6 +558,7 @@ mem_area_free( pool->reserved -= size; } + mem_n_threads_inside--; mutex_exit(&(pool->mutex)); ut_ad(mem_pool_validate(pool)); @@ -577,7 +603,7 @@ mem_pool_validate( } } - ut_a(free + pool->reserved == pool->size + ut_anp(free + pool->reserved == pool->size - (pool->size % MEM_AREA_MIN_SIZE)); mutex_exit(&(pool->mutex)); diff --git a/innobase/os/os0file.c b/innobase/os/os0file.c index 2f32b9347dc..612bd534fd1 100644 --- a/innobase/os/os0file.c +++ b/innobase/os/os0file.c @@ -60,6 +60,7 @@ struct os_aio_slot_struct{ ulint pos; /* index of the slot in the aio array */ ibool reserved; /* TRUE if this slot is reserved */ + time_t reservation_time;/* time when reserved */ ulint len; /* length of the block to read or write */ byte* buf; /* buffer used in i/o */ @@ -147,6 +148,12 @@ time_t os_last_printout; ibool os_has_said_disk_full = FALSE; +/* The mutex protecting the following counts of pending pread and pwrite +operations */ +os_mutex_t os_file_count_mutex; +ulint os_file_n_pending_preads = 0; +ulint os_file_n_pending_pwrites = 0; + /*************************************************************************** Gets the operating system version. Currently works only on Windows. */ @@ -364,6 +371,8 @@ os_io_init_simple(void) { ulint i; + os_file_count_mutex = os_mutex_create(NULL); + for (i = 0; i < OS_FILE_N_SEEK_MUTEXES; i++) { os_file_seek_mutexes[i] = os_mutex_create(NULL); } @@ -415,9 +424,8 @@ try_again: file = CreateFile(name, access, - FILE_SHARE_READ | FILE_SHARE_WRITE, - /* file can be read and written - also by other processes */ + FILE_SHARE_READ,/* file can be read also by other + processes */ NULL, /* default security attributes */ create_flag, attributes, @@ -481,6 +489,101 @@ try_again: return(file); #endif } + +/******************************************************************** +A simple function to open or create a file. */ + +os_file_t +os_file_create_simple_no_error_handling( +/*====================================*/ + /* out, own: handle to the file, not defined if error, + error number can be retrieved with os_get_last_error */ + char* name, /* in: name of the file or path as a null-terminated + string */ + ulint create_mode,/* in: OS_FILE_OPEN if an existing file is opened + (if does not exist, error), or OS_FILE_CREATE if a new + file is created (if exists, error) */ + ulint access_type,/* in: OS_FILE_READ_ONLY or OS_FILE_READ_WRITE */ + ibool* success)/* out: TRUE if succeed, FALSE if error */ +{ +#ifdef __WIN__ + os_file_t file; + DWORD create_flag; + DWORD access; + DWORD attributes = 0; + + ut_a(name); + + if (create_mode == OS_FILE_OPEN) { + create_flag = OPEN_EXISTING; + } else if (create_mode == OS_FILE_CREATE) { + create_flag = CREATE_NEW; + } else { + create_flag = 0; + ut_error; + } + + if (access_type == OS_FILE_READ_ONLY) { + access = GENERIC_READ; + } else if (access_type == OS_FILE_READ_WRITE) { + access = GENERIC_READ | GENERIC_WRITE; + } else { + access = 0; + ut_error; + } + + file = CreateFile(name, + access, + FILE_SHARE_READ,/* file can be read also by other + processes */ + NULL, /* default security attributes */ + create_flag, + attributes, + NULL); /* no template file */ + + if (file == INVALID_HANDLE_VALUE) { + *success = FALSE; + } else { + *success = TRUE; + } + + return(file); +#else + os_file_t file; + int create_flag; + + ut_a(name); + + if (create_mode == OS_FILE_OPEN) { + if (access_type == OS_FILE_READ_ONLY) { + create_flag = O_RDONLY; + } else { + create_flag = O_RDWR; + } + } else if (create_mode == OS_FILE_CREATE) { + create_flag = O_RDWR | O_CREAT | O_EXCL; + } else { + create_flag = 0; + ut_error; + } + + if (create_mode == OS_FILE_CREATE) { + file = open(name, create_flag, S_IRUSR | S_IWUSR + | S_IRGRP | S_IWGRP); + } else { + file = open(name, create_flag); + } + + if (file == -1) { + *success = FALSE; + } else { + *success = TRUE; + } + + return(file); +#endif +} + /******************************************************************** Opens an existing file or creates a new. */ @@ -566,9 +669,14 @@ try_again: file = CreateFile(name, GENERIC_READ | GENERIC_WRITE, /* read and write access */ - FILE_SHARE_READ | FILE_SHARE_WRITE, - /* file can be read and written - also by other processes */ + FILE_SHARE_READ,/* File can be read also by other + processes; we must give the read + permission because of ibbackup. We do + not give the write permission to + others because if one would succeed to + start 2 instances of mysqld on the + SAME files, that could cause severe + database corruption! */ NULL, /* default security attributes */ create_flag, attributes, @@ -677,6 +785,41 @@ os_file_close( } /*************************************************************************** +Closes a file handle. */ + +ibool +os_file_close_no_error_handling( +/*============================*/ + /* out: TRUE if success */ + os_file_t file) /* in, own: handle to a file */ +{ +#ifdef __WIN__ + BOOL ret; + + ut_a(file); + + ret = CloseHandle(file); + + if (ret) { + return(TRUE); + } + + return(FALSE); +#else + int ret; + + ret = close(file); + + if (ret == -1) { + + return(FALSE); + } + + return(TRUE); +#endif +} + +/*************************************************************************** Gets a file size. */ ibool @@ -896,6 +1039,7 @@ os_file_pread( offset */ { off_t offs; + ssize_t n_bytes; ut_a((offset & 0xFFFFFFFF) == offset); @@ -917,7 +1061,17 @@ os_file_pread( os_n_file_reads++; #ifdef HAVE_PREAD - return(pread(file, buf, n, offs)); + os_mutex_enter(os_file_count_mutex); + os_file_n_pending_preads++; + os_mutex_exit(os_file_count_mutex); + + n_bytes = pread(file, buf, n, offs); + + os_mutex_enter(os_file_count_mutex); + os_file_n_pending_preads--; + os_mutex_exit(os_file_count_mutex); + + return(n_bytes); #else { ssize_t ret; @@ -982,8 +1136,16 @@ os_file_pwrite( os_n_file_writes++; #ifdef HAVE_PWRITE + os_mutex_enter(os_file_count_mutex); + os_file_n_pending_pwrites++; + os_mutex_exit(os_file_count_mutex); + ret = pwrite(file, buf, n, offs); + os_mutex_enter(os_file_count_mutex); + os_file_n_pending_pwrites--; + os_mutex_exit(os_file_count_mutex); + if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC && srv_unix_file_flush_method != SRV_UNIX_NOSYNC && !os_do_not_call_flush_at_each_write) { @@ -1372,19 +1534,35 @@ os_aio_init( os_io_init_simple(); + for (i = 0; i < n_segments; i++) { + srv_io_thread_op_info[i] = (char*)"not started yet"; + } + n_per_seg = n / n_segments; n_write_segs = (n_segments - 2) / 2; n_read_segs = n_segments - 2 - n_write_segs; /* printf("Array n per seg %lu\n", n_per_seg); */ + os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1); + + srv_io_thread_function[0] = (char*)"insert buffer thread"; + + os_aio_log_array = os_aio_array_create(n_per_seg, 1); + + srv_io_thread_function[1] = (char*)"log thread"; + os_aio_read_array = os_aio_array_create(n_read_segs * n_per_seg, n_read_segs); + for (i = 2; i < 2 + n_read_segs; i++) { + srv_io_thread_function[i] = (char*)"read thread"; + } + os_aio_write_array = os_aio_array_create(n_write_segs * n_per_seg, n_write_segs); - os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1); - - os_aio_log_array = os_aio_array_create(n_per_seg, 1); + for (i = 2 + n_read_segs; i < n_segments; i++) { + srv_io_thread_function[i] = (char*)"write thread"; + } os_aio_sync_array = os_aio_array_create(n_slots_sync, 1); @@ -1677,6 +1855,7 @@ loop: } slot->reserved = TRUE; + slot->reservation_time = time(NULL); slot->message1 = message1; slot->message2 = message2; slot->file = file; @@ -2249,6 +2428,8 @@ os_aio_simulated_handle( ulint total_len; ulint offs; ulint lowest_offset; + ulint biggest_age; + ulint age; byte* combined_buf; byte* combined_buf2= 0; /* Remove warning */ ibool ret; @@ -2301,22 +2482,55 @@ restart: n_consecutive = 0; - /* Look for an i/o request at the lowest offset in the array - (we ignore the high 32 bits of the offset in these heuristics) */ + /* If there are at least 2 seconds old requests, then pick the oldest + one to prevent starvation. If several requests have the same age, + then pick the one at the lowest offset. */ + biggest_age = 0; lowest_offset = ULINT_MAX; - + for (i = 0; i < n; i++) { slot = os_aio_array_get_nth_slot(array, i + segment * n); - if (slot->reserved && slot->offset < lowest_offset) { + if (slot->reserved) { + age = (ulint)difftime(time(NULL), + slot->reservation_time); - /* Found an i/o request */ - consecutive_ios[0] = slot; + if ((age >= 2 && age > biggest_age) + || (age >= 2 && age == biggest_age + && slot->offset < lowest_offset)) { + + /* Found an i/o request */ + consecutive_ios[0] = slot; + + n_consecutive = 1; + + biggest_age = age; + lowest_offset = slot->offset; + } + } + } + + if (n_consecutive == 0) { + /* There were no old requests. Look for an i/o request at the + lowest offset in the array (we ignore the high 32 bits of the + offset in these heuristics) */ + + lowest_offset = ULINT_MAX; + + for (i = 0; i < n; i++) { + slot = os_aio_array_get_nth_slot(array, + i + segment * n); - n_consecutive = 1; + if (slot->reserved && slot->offset < lowest_offset) { - lowest_offset = slot->offset; + /* Found an i/o request */ + consecutive_ios[0] = slot; + + n_consecutive = 1; + + lowest_offset = slot->offset; + } } } @@ -2422,7 +2636,7 @@ consecutive_loop: + FIL_PAGE_LSN + 4) != mach_read_from_4(combined_buf + len2 + UNIV_PAGE_SIZE - - FIL_PAGE_END_LSN + 4)) { + - FIL_PAGE_END_LSN_OLD_CHKSUM + 4)) { ut_print_timestamp(stderr); fprintf(stderr, " InnoDB: ERROR: The page to be written seems corrupt!\n"); @@ -2583,14 +2797,15 @@ os_aio_print( double avg_bytes_read; ulint i; - if (buf_end - buf < 1000) { + if (buf_end - buf < 1200) { return; } for (i = 0; i < srv_n_file_io_threads; i++) { - buf += sprintf(buf, "I/O thread %lu state: %s\n", i, - srv_io_thread_op_info[i]); + buf += sprintf(buf, "I/O thread %lu state: %s (%s)\n", i, + srv_io_thread_op_info[i], + srv_io_thread_function[i]); } buf += sprintf(buf, "Pending normal aio reads:"); @@ -2665,6 +2880,12 @@ loop: "%lu OS file reads, %lu OS file writes, %lu OS fsyncs\n", os_n_file_reads, os_n_file_writes, os_n_fsyncs); + if (os_file_n_pending_preads != 0 || os_file_n_pending_pwrites != 0) { + buf += sprintf(buf, + "%lu pending preads, %lu pending pwrites\n", + os_file_n_pending_preads, os_file_n_pending_pwrites); + } + if (os_n_file_reads == os_n_file_reads_old) { avg_bytes_read = 0.0; } else { diff --git a/innobase/os/os0thread.c b/innobase/os/os0thread.c index 9af98760ad1..1252cc5e4b7 100644 --- a/innobase/os/os0thread.c +++ b/innobase/os/os0thread.c @@ -187,8 +187,8 @@ os_thread_exit( is cast as a DWORD */ { #ifdef UNIV_DEBUG_THREAD_CREATION - printf("A thread exits.\n"); - printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id())); + printf("Thread exits, id %lu\n", + os_thread_pf(os_thread_get_curr_id())); #endif os_mutex_enter(os_sync_mutex); os_thread_count--; diff --git a/innobase/page/page0cur.c b/innobase/page/page0cur.c index d3a40668c4b..7e2fc19c00f 100644 --- a/innobase/page/page0cur.c +++ b/innobase/page/page0cur.c @@ -14,6 +14,7 @@ Created 10/4/1994 Heikki Tuuri #include "rem0cmp.h" #include "mtr0log.h" #include "log0recv.h" +#include "rem0cmp.h" ulint page_cur_short_succ = 0; @@ -218,6 +219,8 @@ page_cur_search_with_match( || (mode == PAGE_CUR_G) || (mode == PAGE_CUR_GE) || (mode == PAGE_CUR_LE_OR_EXTENDS) || (mode == PAGE_CUR_DBG)); + page_check_dir(page); + #ifdef PAGE_CUR_ADAPT if ((page_header_get_field(page, PAGE_LEVEL) == 0) && (mode == PAGE_CUR_LE) @@ -595,6 +598,7 @@ page_cur_parse_insert_rec( rec_t* cursor_rec; byte buf1[1024]; byte* buf; + byte* ptr2 = ptr; ulint info_bits = 0; /* remove warning */ page_cur_t cursor; @@ -697,7 +701,20 @@ page_cur_parse_insert_rec( /* Build the inserted record to buf */ - ut_a(mismatch_index < UNIV_PAGE_SIZE); + if (mismatch_index >= UNIV_PAGE_SIZE) { + printf("Is short %lu, info_bits %lu, offset %lu, o_offset %lu\n" + "mismatch index %lu, end_seg_len %lu\n" + "parsed len %lu\n", + is_short, info_bits, offset, origin_offset, + mismatch_index, end_seg_len, (ulint)(ptr - ptr2)); + + printf("Dump of 300 bytes of log:\n"); + ut_print_buf(ptr2, 300); + + buf_page_print(page); + + ut_a(0); + } ut_memcpy(buf, rec_get_start(cursor_rec), mismatch_index); ut_memcpy(buf + mismatch_index, ptr, end_seg_len); diff --git a/innobase/page/page0page.c b/innobase/page/page0page.c index 7d240bdd5b0..ef5dad60c08 100644 --- a/innobase/page/page0page.c +++ b/innobase/page/page0page.c @@ -353,7 +353,7 @@ page_create( infimum_rec = rec_convert_dtuple_to_rec(heap_top, tuple); - ut_ad(infimum_rec == page + PAGE_INFIMUM); + ut_a(infimum_rec == page + PAGE_INFIMUM); rec_set_n_owned(infimum_rec, 1); rec_set_heap_no(infimum_rec, 0); @@ -370,7 +370,7 @@ page_create( supremum_rec = rec_convert_dtuple_to_rec(heap_top, tuple); - ut_ad(supremum_rec == page + PAGE_SUPREMUM); + ut_a(supremum_rec == page + PAGE_SUPREMUM); rec_set_n_owned(supremum_rec, 1); rec_set_heap_no(supremum_rec, 1); @@ -389,6 +389,8 @@ page_create( page_header_set_ptr(page, PAGE_FREE, NULL); page_header_set_field(page, PAGE_GARBAGE, 0); page_header_set_ptr(page, PAGE_LAST_INSERT, NULL); + page_header_set_field(page, PAGE_DIRECTION, PAGE_NO_DIRECTION); + page_header_set_field(page, PAGE_N_DIRECTION, 0); page_header_set_field(page, PAGE_N_RECS, 0); page_set_max_trx_id(page, ut_dulint_zero); @@ -402,17 +404,22 @@ page_create( slot = page_dir_get_nth_slot(page, 1); page_dir_slot_set_rec(slot, supremum_rec); - /* Set next pointers in infimum and supremum */ + /* Set the next pointers in infimum and supremum */ rec_set_next_offs(infimum_rec, (ulint)(supremum_rec - page)); rec_set_next_offs(supremum_rec, 0); +#ifdef notdefined + /* Disable the use of page_template: there is a race condition here: + while one thread is creating page_template, another one can start + using it before the memcpy completes! */ + if (page_template == NULL) { page_template = mem_alloc(UNIV_PAGE_SIZE); ut_memcpy(page_template, page, UNIV_PAGE_SIZE); } - +#endif return(page); } @@ -439,6 +446,9 @@ page_copy_rec_list_end_no_locks( page_cur_move_to_next(&cur1); } + /* Track a memory corruption bug in Windows */ + ut_a(mach_read_from_2(new_page + UNIV_PAGE_SIZE - 10) == PAGE_INFIMUM); + page_cur_set_before_first(new_page, &cur2); /* Copy records from the original page to the new page */ @@ -449,6 +459,8 @@ page_copy_rec_list_end_no_locks( ut_a( page_cur_rec_insert(&cur2, page_cur_get_rec(&cur1), mtr)); + ut_a(mach_read_from_2(new_page + UNIV_PAGE_SIZE - 10) + == PAGE_INFIMUM); page_cur_move_to_next(&cur1); page_cur_move_to_next(&cur2); } @@ -1315,6 +1327,37 @@ page_rec_validate( return(TRUE); } + +/******************************************************************* +Checks that the first directory slot points to the infimum record and +the last to the supremum. This function is intended to track if the +bug fixed in 4.0.14 has caused corruption to users' databases. */ + +void +page_check_dir( +/*===========*/ + page_t* page) /* in: index page */ +{ + ulint n_slots; + + n_slots = page_dir_get_n_slots(page); + + if (page_dir_slot_get_rec(page_dir_get_nth_slot(page, 0)) + != page_get_infimum_rec(page)) { + + fprintf(stderr, +"InnoDB: Page directory corruption: supremum not pointed to\n"); + buf_page_print(page); + } + + if (page_dir_slot_get_rec(page_dir_get_nth_slot(page, n_slots - 1)) + != page_get_supremum_rec(page)) { + + fprintf(stderr, +"InnoDB: Page directory corruption: supremum not pointed to\n"); + buf_page_print(page); + } +} /******************************************************************* This function checks the consistency of an index page when we do not @@ -1598,7 +1641,8 @@ page_validate( "InnoDB: previous record %s\n", err_buf); rec_sprintf(err_buf, 900, rec); - fprintf(stderr, "InnoDB: record %s\n", err_buf); + fprintf(stderr, + "InnoDB: record %s\n", err_buf); goto func_exit; } diff --git a/innobase/pars/pars0opt.c b/innobase/pars/pars0opt.c index 91083e6fa16..4faf83b47a3 100644 --- a/innobase/pars/pars0opt.c +++ b/innobase/pars/pars0opt.c @@ -1058,7 +1058,6 @@ opt_clust_access( dfield_t* dfield; mem_heap_t* heap; ulint n_fields; - ulint col_no; ulint pos; ulint i; @@ -1093,8 +1092,7 @@ opt_clust_access( plan->clust_map = mem_heap_alloc(heap, n_fields * sizeof(ulint)); for (i = 0; i < n_fields; i++) { - col_no = dict_index_get_nth_col_no(clust_index, i); - pos = dict_index_get_nth_col_pos(index, col_no); + pos = dict_index_get_nth_field_pos(index, clust_index, i); *(plan->clust_map + i) = pos; @@ -1109,7 +1107,8 @@ opt_clust_access( dfield = dtuple_get_nth_field(plan->clust_ref, table->mix_len); - dfield_set_data(dfield, mem_heap_alloc(heap, table->mix_id_len), + dfield_set_data(dfield, mem_heap_alloc(heap, + table->mix_id_len), table->mix_id_len); ut_memcpy(dfield_get_data(dfield), table->mix_id_buf, table->mix_id_len); diff --git a/innobase/pars/pars0pars.c b/innobase/pars/pars0pars.c index 664f498ef3e..3e43b6ae262 100644 --- a/innobase/pars/pars0pars.c +++ b/innobase/pars/pars0pars.c @@ -244,13 +244,11 @@ pars_resolve_func_data_type( /* Inherit the data type from the first argument (which must not be the SQL null literal whose type is DATA_ERROR) */ - ut_a(dtype_get_mtype(que_node_get_data_type(arg)) - != DATA_ERROR); dtype_copy(que_node_get_data_type(node), que_node_get_data_type(arg)); - ut_a(dtype_get_mtype(que_node_get_data_type(node)) == DATA_INT); - + ut_a(dtype_get_mtype(que_node_get_data_type(node)) + == DATA_INT); } else if (func == PARS_COUNT_TOKEN) { ut_a(arg); dtype_set(que_node_get_data_type(node), DATA_INT, 0, 4, 0); @@ -1596,7 +1594,7 @@ pars_create_index( column = column_list; while (column) { - dict_mem_index_add_field(index, column->name, 0); + dict_mem_index_add_field(index, column->name, 0, 0); column->resolved = TRUE; column->token_type = SYM_COLUMN; diff --git a/innobase/rem/rem0cmp.c b/innobase/rem/rem0cmp.c index e9740d7ea78..2e18e68ec43 100644 --- a/innobase/rem/rem0cmp.c +++ b/innobase/rem/rem0cmp.c @@ -38,7 +38,7 @@ Used in debug checking of cmp_dtuple_... . This function is used to compare a data tuple to a physical record. If dtuple has n fields then rec must have either m >= n fields, or it must differ from dtuple in some of the m fields rec has. */ -static + int cmp_debug_dtuple_rec_with_match( /*============================*/ @@ -50,9 +50,10 @@ cmp_debug_dtuple_rec_with_match( dtuple in some of the common fields, or which has an equal number or more fields than dtuple */ - ulint* matched_fields);/* in/out: number of already completely - matched fields; when function returns, - contains the value for current comparison */ + ulint* matched_fields);/* in/out: number of already + completely matched fields; when function + returns, contains the value for current + comparison */ /***************************************************************** This function is used to compare two data fields for which the data type is such that we must use MySQL code to compare them. The prototype here @@ -79,17 +80,12 @@ UNIV_INLINE ulint cmp_collate( /*========*/ - /* out: collation order position */ - dtype_t* type __attribute__((unused)) , /* in: type */ - ulint code) /* in: code of a character stored in database - record */ -{ - ut_ad((type->mtype == DATA_CHAR) || (type->mtype == DATA_VARCHAR)); - + /* out: collation order position */ + ulint code) /* in: code of a character stored in database record */ +{ return((ulint) srv_latin1_ordering[code]); } - /***************************************************************** Returns TRUE if two types are equal for comparison purposes. */ @@ -118,7 +114,8 @@ cmp_types_are_equal( if (type1->mtype == DATA_INT && (type1->prtype & DATA_UNSIGNED) - != (type2->prtype & DATA_UNSIGNED)) { + != (type2->prtype & DATA_UNSIGNED)) { + /* The storage format of an unsigned integer is different from a signed integer: in a signed integer we OR 0x8000... to the value of positive integers. */ @@ -131,12 +128,17 @@ cmp_types_are_equal( return(FALSE); } + if (type1->mtype == DATA_BLOB && (type1->prtype & DATA_BINARY_TYPE) + != (type2->prtype & DATA_BINARY_TYPE)) { + return(FALSE); + } + return(TRUE); } /***************************************************************** -Innobase uses this function is to compare two data fields for which the -data type is such that we must compare whole fields. */ +Innobase uses this function to compare two data fields for which the data type +is such that we must compare whole fields or call MySQL to do the comparison */ static int cmp_whole_field( @@ -239,8 +241,34 @@ cmp_whole_field( return(0); case DATA_VARMYSQL: case DATA_MYSQL: + case DATA_BLOB: + if (data_type == DATA_BLOB + && 0 != (type->prtype & DATA_BINARY_TYPE)) { + + ut_print_timestamp(stderr); + fprintf(stderr, +" InnoDB: Error: comparing a binary BLOB with a character set sensitive\n" +"InnoDB: comparison!\n"); + } + + /* MySQL does not pad the ends of strings with spaces in a + comparison. That would cause a foreign key check to fail for + non-latin1 character sets if we have different length columns. + To prevent that we remove trailing spaces here before doing + the comparison. NOTE that if we in the future map more MySQL + types to DATA_MYSQL or DATA_VARMYSQL, we have to change this + code. */ + + while (a_length > 0 && a[a_length - 1] == ' ') { + a_length--; + } + + while (b_length > 0 && b[b_length - 1] == ' ') { + b_length--; + } + return(innobase_mysql_cmp( - (int)(type->prtype & ~DATA_NOT_NULL), + (int)(type->prtype & DATA_MYSQL_TYPE_MASK), a, a_length, b, b_length)); default: fprintf(stderr, @@ -291,7 +319,10 @@ cmp_data_data_slow( return(1); } - if (cur_type->mtype >= DATA_FLOAT) { + if (cur_type->mtype >= DATA_FLOAT + || (cur_type->mtype == DATA_BLOB + && (cur_type->prtype & DATA_NONLATIN1))) { + return(cmp_whole_field(cur_type, data1, len1, data2, len2)); } @@ -334,9 +365,12 @@ cmp_data_data_slow( goto next_byte; } - if (cur_type->mtype <= DATA_CHAR) { - data1_byte = cmp_collate(cur_type, data1_byte); - data2_byte = cmp_collate(cur_type, data2_byte); + if (cur_type->mtype <= DATA_CHAR + || (cur_type->mtype == DATA_BLOB + && 0 == (cur_type->prtype & DATA_BINARY_TYPE))) { + + data1_byte = cmp_collate(data1_byte); + data2_byte = cmp_collate(data2_byte); } if (data1_byte > data2_byte) { @@ -487,7 +521,9 @@ cmp_dtuple_rec_with_match( } } - if (cur_type->mtype >= DATA_FLOAT) { + if (cur_type->mtype >= DATA_FLOAT + || (cur_type->mtype == DATA_BLOB + && (cur_type->prtype & DATA_NONLATIN1))) { ret = cmp_whole_field(cur_type, dfield_get_data(dtuple_field), dtuple_f_len, @@ -547,10 +583,13 @@ cmp_dtuple_rec_with_match( goto next_byte; } - if (cur_type->mtype <= DATA_CHAR) { - rec_byte = cmp_collate(cur_type, rec_byte); - dtuple_byte = cmp_collate(cur_type, - dtuple_byte); + if (cur_type->mtype <= DATA_CHAR + || (cur_type->mtype == DATA_BLOB + && 0 == + (cur_type->prtype & DATA_BINARY_TYPE))) { + + rec_byte = cmp_collate(rec_byte); + dtuple_byte = cmp_collate(dtuple_byte); } if (dtuple_byte > rec_byte) { @@ -583,8 +622,8 @@ order_resolved: matched_fields)); ut_ad(*matched_fields == cur_field); /* In the debug version, the above cmp_debug_... sets - *matched_fields to a value */ - *matched_fields = cur_field; + *matched_fields to a value */ + *matched_fields = cur_field; *matched_bytes = cur_bytes; return(ret); @@ -804,7 +843,10 @@ cmp_rec_rec_with_match( } } - if (cur_type->mtype >= DATA_FLOAT) { + if (cur_type->mtype >= DATA_FLOAT + || (cur_type->mtype == DATA_BLOB + && (cur_type->prtype & DATA_NONLATIN1))) { + ret = cmp_whole_field(cur_type, rec1_b_ptr, rec1_f_len, rec2_b_ptr, rec2_f_len); @@ -861,9 +903,13 @@ cmp_rec_rec_with_match( goto next_byte; } - if (cur_type->mtype <= DATA_CHAR) { - rec1_byte = cmp_collate(cur_type, rec1_byte); - rec2_byte = cmp_collate(cur_type, rec2_byte); + if (cur_type->mtype <= DATA_CHAR + || (cur_type->mtype == DATA_BLOB + && 0 == + (cur_type->prtype & DATA_BINARY_TYPE))) { + + rec1_byte = cmp_collate(rec1_byte); + rec2_byte = cmp_collate(rec2_byte); } if (rec1_byte < rec2_byte) { @@ -906,7 +952,7 @@ This function is used to compare a data tuple to a physical record. If dtuple has n fields then rec must have either m >= n fields, or it must differ from dtuple in some of the m fields rec has. If encounters an externally stored field, returns 0. */ -static + int cmp_debug_dtuple_rec_with_match( /*============================*/ @@ -918,9 +964,10 @@ cmp_debug_dtuple_rec_with_match( dtuple in some of the common fields, or which has an equal number or more fields than dtuple */ - ulint* matched_fields) /* in/out: number of already completely - matched fields; when function returns, - contains the value for current comparison */ + ulint* matched_fields) /* in/out: number of already + completely matched fields; when function + returns, contains the value for current + comparison */ { dtype_t* cur_type; /* pointer to type of the current field in dtuple */ diff --git a/innobase/row/row0ins.c b/innobase/row/row0ins.c index e96c08a715b..23da0b9b93c 100644 --- a/innobase/row/row0ins.c +++ b/innobase/row/row0ins.c @@ -217,8 +217,8 @@ ins_node_set_new_row( } /*********************************************************************** -Does an insert operation by updating a delete marked existing record -in the index. This situation can occur if the delete marked record is +Does an insert operation by updating a delete-marked existing record +in the index. This situation can occur if the delete-marked record is kept in the index for consistent reads. */ static ulint @@ -240,9 +240,9 @@ row_ins_sec_index_entry_by_modify( ut_ad((cursor->index->type & DICT_CLUSTERED) == 0); ut_ad(rec_get_deleted_flag(rec)); - /* We know that in the ordering entry and rec are identified. - But in their binary form there may be differences if there - are char fields in them. Therefore we have to calculate the + /* We know that in the alphabetical ordering, entry and rec are + identical. But in their binary form there may be differences if + there are char fields in them. Therefore we have to calculate the difference and do an update-in-place if necessary. */ heap = mem_heap_create(1024); @@ -305,8 +305,8 @@ row_ins_clust_index_entry_by_modify( /* Try optimistic updating of the record, keeping changes within the page */ - err = btr_cur_optimistic_update(0, cursor, update, 0, thr, mtr); - + err = btr_cur_optimistic_update(0, cursor, update, 0, thr, + mtr); if (err == DB_OVERFLOW || err == DB_UNDERFLOW) { err = DB_FAIL; } @@ -364,11 +364,17 @@ row_ins_cascade_calc_update_vec( /* out: number of fields in the calculated update vector; the value can also be 0 if no foreign key - fields changed */ + fields changed; the returned value + is ULINT_UNDEFINED if the column + type in the child table is too short + to fit the new value in the parent + table: that means the update fails */ upd_node_t* node, /* in: update node of the parent table */ - dict_foreign_t* foreign) /* in: foreign key constraint whose + dict_foreign_t* foreign, /* in: foreign key constraint whose type is != 0 */ + mem_heap_t* heap) /* in: memory heap to use as + temporary storage */ { upd_node_t* cascade = node->cascade_node; dict_table_t* table = foreign->foreign_table; @@ -381,14 +387,16 @@ row_ins_cascade_calc_update_vec( upd_field_t* parent_ufield; ulint n_fields_updated; ulint parent_field_no; + dtype_t* type; ulint i; ulint j; ut_a(node && foreign && cascade && table && index); /* Calculate the appropriate update vector which will set the fields - in the child index record to the same value as the referenced index - record will get in the update. */ + in the child index record to the same value (possibly padded with + spaces if the column is a fixed length CHAR or FIXBINARY column) as + the referenced index record will get in the update. */ parent_table = node->table; ut_a(parent_table == foreign->referenced_table); @@ -424,7 +432,56 @@ row_ins_cascade_calc_update_vec( dict_table_get_nth_col_pos(table, dict_index_get_nth_col_no(index, i)); ufield->exp = NULL; + ufield->new_val = parent_ufield->new_val; + + type = dict_index_get_nth_type(index, i); + + /* Do not allow a NOT NULL column to be + updated as NULL */ + + if (ufield->new_val.len == UNIV_SQL_NULL + && (type->prtype & DATA_NOT_NULL)) { + + return(ULINT_UNDEFINED); + } + + /* If the new value would not fit in the + column, do not allow the update */ + + if (ufield->new_val.len != UNIV_SQL_NULL + && ufield->new_val.len + > dtype_get_len(type)) { + + return(ULINT_UNDEFINED); + } + + /* If the parent column type has a different + length than the child column type, we may + need to pad with spaces the new value of the + child column */ + + if (dtype_is_fixed_size(type) + && ufield->new_val.len != UNIV_SQL_NULL + && ufield->new_val.len + < dtype_get_fixed_size(type)) { + + ufield->new_val.data = + mem_heap_alloc(heap, + dtype_get_fixed_size(type)); + ufield->new_val.len = + dtype_get_fixed_size(type); + ut_a(dtype_get_pad_char(type) + != ULINT_UNDEFINED); + + memset(ufield->new_val.data, + (byte)dtype_get_pad_char(type), + dtype_get_fixed_size(type)); + ut_memcpy(ufield->new_val.data, + parent_ufield->new_val.data, + parent_ufield->new_val.len); + } + ufield->extern_storage = FALSE; n_fields_updated++; @@ -570,9 +627,11 @@ row_ins_foreign_check_on_constraint( dict_index_t* clust_index; dtuple_t* ref; mem_heap_t* tmp_heap; + mem_heap_t* upd_vec_heap = NULL; rec_t* rec; rec_t* clust_rec; upd_t* update; + ulint n_to_update; ulint err; ulint i; char* ptr; @@ -597,8 +656,10 @@ row_ins_foreign_check_on_constraint( *ptr = '\0'; /* We call a function in ha_innodb.cc */ +#ifndef UNIV_HOTBACKUP innobase_invalidate_query_cache(thr_get_trx(thr), table_name_buf, ut_strlen(table->name) + 1); +#endif node = thr->run_node; if (node->is_delete && 0 == (foreign->type & @@ -828,7 +889,21 @@ row_ins_foreign_check_on_constraint( /* Build the appropriate update vector which sets changing foreign->n_fields first fields in rec to new values */ - row_ins_cascade_calc_update_vec(node, foreign); + upd_vec_heap = mem_heap_create(256); + + n_to_update = row_ins_cascade_calc_update_vec(node, foreign, + upd_vec_heap); + if (n_to_update == ULINT_UNDEFINED) { + err = DB_ROW_IS_REFERENCED; + + row_ins_foreign_report_err( +(char*)"Trying a cascaded update where the updated value in the child\n" +"table would not fit in the length of the column, or the value would\n" +"be NULL and the column is declared as not NULL in the child table,", + thr, foreign, btr_pcur_get_rec(pcur), entry); + + goto nonstandard_exit_func; + } if (cascade->update->n_fields == 0) { @@ -867,10 +942,18 @@ row_ins_foreign_check_on_constraint( btr_pcur_restore_position(BTR_SEARCH_LEAF, pcur, mtr); + if (upd_vec_heap) { + mem_heap_free(upd_vec_heap); + } + return(err); nonstandard_exit_func: + if (upd_vec_heap) { + mem_heap_free(upd_vec_heap); + } + btr_pcur_store_position(pcur, mtr); mtr_commit(mtr); @@ -1275,6 +1358,11 @@ row_ins_unique_report_err( dtuple_t* entry, /* in: index entry to insert in the index */ dict_index_t* index) /* in: index */ { + UT_NOT_USED(thr); + UT_NOT_USED(rec); + UT_NOT_USED(entry); + UT_NOT_USED(index); + #ifdef notdefined /* Disable reporting to test if the slowdown of REPLACE in 4.0.13 was caused by this! */ @@ -1816,7 +1904,7 @@ row_ins_index_entry( /* Try first optimistic descent to the B-tree */ err = row_ins_index_entry_low(BTR_MODIFY_LEAF, index, entry, - ext_vec, n_ext_vec, thr); + ext_vec, n_ext_vec, thr); if (err != DB_FAIL) { return(err); @@ -1832,13 +1920,15 @@ row_ins_index_entry( /*************************************************************** Sets the values of the dtuple fields in entry from the values of appropriate columns in row. */ -UNIV_INLINE +static void row_ins_index_entry_set_vals( /*=========================*/ + dict_index_t* index, /* in: index */ dtuple_t* entry, /* in: index entry to make */ dtuple_t* row) /* in: row */ { + dict_field_t* ind_field; dfield_t* field; dfield_t* row_field; ulint n_fields; @@ -1850,11 +1940,21 @@ row_ins_index_entry_set_vals( for (i = 0; i < n_fields; i++) { field = dtuple_get_nth_field(entry, i); + ind_field = dict_index_get_nth_field(index, i); + + row_field = dtuple_get_nth_field(row, ind_field->col->ind); - row_field = dtuple_get_nth_field(row, field->col_no); + /* Check column prefix indexes */ + if (ind_field->prefix_len > 0 + && dfield_get_len(row_field) != UNIV_SQL_NULL + && dfield_get_len(row_field) > ind_field->prefix_len) { + + field->len = ind_field->prefix_len; + } else { + field->len = row_field->len; + } field->data = row_field->data; - field->len = row_field->len; } } @@ -1873,7 +1973,7 @@ row_ins_index_entry_step( ut_ad(dtuple_check_typed(node->row)); - row_ins_index_entry_set_vals(node->entry, node->row); + row_ins_index_entry_set_vals(node->index, node->entry, node->row); ut_ad(dtuple_check_typed(node->entry)); diff --git a/innobase/row/row0mysql.c b/innobase/row/row0mysql.c index 428e4d568f3..61ba9111b91 100644 --- a/innobase/row/row0mysql.c +++ b/innobase/row/row0mysql.c @@ -76,9 +76,6 @@ row_mysql_store_blob_ref( also to set the NULL bit in the MySQL record header! */ { - ulint sum = 0; - ulint i; - /* MySQL might assume the field is set to zero except the length and the pointer fields */ @@ -93,22 +90,6 @@ row_mysql_store_blob_ref( ut_a(col_len - 8 > 2 || len < 256 * 256); ut_a(col_len - 8 > 3 || len < 256 * 256 * 256); - /* We try to track an elusive bug which probably was fixed - May 9, 2002, but better be sure: we probe the data buffer - to make sure it is in valid allocated memory */ - - for (i = 0; i < len; i++) { - - sum += (ulint)(data + i); - } - - /* The variable below is identically false, we just fool the - compiler to not optimize away our loop */ - if (row_mysql_identically_false) { - - printf("Sum %lu\n", sum); - } - mach_write_to_n_little_endian(dest, col_len - 8, len); ut_memcpy(dest + col_len - 8, (byte*)&data, sizeof(byte*)); @@ -952,7 +933,8 @@ row_update_for_mysql( if (prebuilt->pcur->btr_cur.index == clust_index) { btr_pcur_copy_stored_position(node->pcur, prebuilt->pcur); } else { - btr_pcur_copy_stored_position(node->pcur, prebuilt->clust_pcur); + btr_pcur_copy_stored_position(node->pcur, + prebuilt->clust_pcur); } ut_a(node->pcur->rel_pos == BTR_PCUR_ON); @@ -1477,8 +1459,7 @@ row_create_index_for_mysql( ulint namelen; ulint keywordlen; ulint err; - ulint i; - ulint j; + ulint i, j; ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_EX)); ut_ad(mutex_own(&(dict_sys->mutex))); @@ -1486,23 +1467,9 @@ row_create_index_for_mysql( trx->op_info = (char *) "creating index"; - trx_start_if_not_started(trx); - - namelen = ut_strlen(index->table_name); - - keywordlen = ut_strlen("_recover_innodb_tmp_table"); - - if (namelen >= keywordlen - && 0 == ut_memcmp( - index->table_name + namelen - keywordlen, - (char*)"_recover_innodb_tmp_table", keywordlen)) { - - return(DB_SUCCESS); - } - /* Check that the same column does not appear twice in the index. - InnoDB assumes this in its algorithms, e.g., update of an index - entry */ + Starting from 4.0.14 InnoDB should be able to cope with that, but + safer not to allow them. */ for (i = 0; i < dict_index_get_n_fields(index); i++) { for (j = 0; j < i; j++) { @@ -1525,6 +1492,20 @@ row_create_index_for_mysql( } } + trx_start_if_not_started(trx); + + namelen = ut_strlen(index->table_name); + + keywordlen = ut_strlen("_recover_innodb_tmp_table"); + + if (namelen >= keywordlen + && 0 == ut_memcmp( + index->table_name + namelen - keywordlen, + (char*)"_recover_innodb_tmp_table", keywordlen)) { + + return(DB_SUCCESS); + } + heap = mem_heap_create(512); trx->dict_operation = TRUE; @@ -1542,6 +1523,7 @@ row_create_index_for_mysql( que_graph_free((que_t*) que_node_get_parent(thr)); error_handling: + if (err != DB_SUCCESS) { /* We have special error handling here */ @@ -2541,7 +2523,7 @@ loop: prev_entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap); - ret = row_search_for_mysql(buf, PAGE_CUR_G, prebuilt, 0, ROW_SEL_NEXT); + ret = row_search_for_mysql(buf, PAGE_CUR_G, prebuilt, 0, ROW_SEL_NEXT); goto loop; } diff --git a/innobase/row/row0row.c b/innobase/row/row0row.c index 40a775143f4..6c0c6c04cd5 100644 --- a/innobase/row/row0row.c +++ b/innobase/row/row0row.c @@ -136,7 +136,14 @@ row_build_index_entry( dfield2 = dtuple_get_nth_field(row, dict_col_get_no(col)); dfield_copy(dfield, dfield2); - dfield->col_no = dict_col_get_no(col); + + /* If a column prefix index, take only the prefix */ + if (ind_field->prefix_len > 0 + && dfield_get_len(dfield2) != UNIV_SQL_NULL + && dfield_get_len(dfield2) > ind_field->prefix_len) { + + dfield_set_len(dfield, ind_field->prefix_len); + } } ut_ad(dtuple_check_typed(entry)); @@ -146,8 +153,7 @@ row_build_index_entry( /*********************************************************************** An inverse function to dict_row_build_index_entry. Builds a row from a -record in a clustered index. NOTE that externally stored (often big) -fields are always copied to heap. */ +record in a clustered index. */ dtuple_t* row_build( @@ -172,6 +178,7 @@ row_build( { dtuple_t* row; dict_table_t* table; + dict_field_t* ind_field; dict_col_t* col; dfield_t* dfield; ulint n_fields; @@ -204,19 +211,24 @@ row_build( dict_table_copy_types(row, table); for (i = 0; i < n_fields; i++) { + ind_field = dict_index_get_nth_field(index, i); - col = dict_field_get_col(dict_index_get_nth_field(index, i)); - dfield = dtuple_get_nth_field(row, dict_col_get_no(col)); - field = rec_get_nth_field(rec, i, &len); + if (ind_field->prefix_len == 0) { - if (type == ROW_COPY_ALSO_EXTERNALS - && rec_get_nth_field_extern_bit(rec, i)) { + col = dict_field_get_col(ind_field); + dfield = dtuple_get_nth_field(row, + dict_col_get_no(col)); + field = rec_get_nth_field(rec, i, &len); - field = btr_rec_copy_externally_stored_field(rec, - i, &len, heap); - } + if (type == ROW_COPY_ALSO_EXTERNALS + && rec_get_nth_field_extern_bit(rec, i)) { - dfield_set_data(dfield, field, len); + field = btr_rec_copy_externally_stored_field( + rec, i, &len, heap); + } + + dfield_set_data(dfield, field, len); + } } ut_ad(dtuple_check_typed(row)); @@ -371,7 +383,6 @@ row_build_row_ref( dict_table_t* table; dict_index_t* clust_index; dfield_t* dfield; - dict_col_t* col; dtuple_t* ref; byte* field; ulint len; @@ -403,24 +414,13 @@ row_build_row_ref( for (i = 0; i < ref_len; i++) { dfield = dtuple_get_nth_field(ref, i); - col = dict_field_get_col( - dict_index_get_nth_field(clust_index, i)); - pos = dict_index_get_nth_col_pos(index, dict_col_get_no(col)); + pos = dict_index_get_nth_field_pos(index, clust_index, i); - if (pos != ULINT_UNDEFINED) { - field = rec_get_nth_field(rec, pos, &len); + ut_a(pos != ULINT_UNDEFINED); + + field = rec_get_nth_field(rec, pos, &len); - dfield_set_data(dfield, field, len); - } else { - ut_ad(table->type == DICT_TABLE_CLUSTER_MEMBER); - ut_ad(i == table->mix_len); - - dfield_set_data(dfield, - mem_heap_alloc(heap, table->mix_id_len), - table->mix_id_len); - ut_memcpy(dfield_get_data(dfield), table->mix_id_buf, - table->mix_id_len); - } + dfield_set_data(dfield, field, len); } ut_ad(dtuple_check_typed(ref)); @@ -448,7 +448,6 @@ row_build_row_ref_in_tuple( dict_table_t* table; dict_index_t* clust_index; dfield_t* dfield; - dict_col_t* col; byte* field; ulint len; ulint ref_len; @@ -483,19 +482,13 @@ row_build_row_ref_in_tuple( for (i = 0; i < ref_len; i++) { dfield = dtuple_get_nth_field(ref, i); - col = dict_field_get_col( - dict_index_get_nth_field(clust_index, i)); - pos = dict_index_get_nth_col_pos(index, dict_col_get_no(col)); + pos = dict_index_get_nth_field_pos(index, clust_index, i); - if (pos != ULINT_UNDEFINED) { - field = rec_get_nth_field(rec, pos, &len); + ut_a(pos != ULINT_UNDEFINED); + + field = rec_get_nth_field(rec, pos, &len); - dfield_set_data(dfield, field, len); - } else { - ut_ad(table->type == DICT_TABLE_CLUSTER_MEMBER); - ut_ad(i == table->mix_len); - ut_a(0); - } + dfield_set_data(dfield, field, len); } ut_ad(dtuple_check_typed(ref)); @@ -517,6 +510,7 @@ row_build_row_ref_from_row( directly into data of this row */ { dict_index_t* clust_index; + dict_field_t* field; dfield_t* dfield; dfield_t* dfield2; dict_col_t* col; @@ -533,13 +527,21 @@ row_build_row_ref_from_row( for (i = 0; i < ref_len; i++) { dfield = dtuple_get_nth_field(ref, i); - - col = dict_field_get_col( - dict_index_get_nth_field(clust_index, i)); - + + field = dict_index_get_nth_field(clust_index, i); + + col = dict_field_get_col(field); + dfield2 = dtuple_get_nth_field(row, dict_col_get_no(col)); dfield_copy(dfield, dfield2); + + if (field->prefix_len > 0 + && dfield->len != UNIV_SQL_NULL + && dfield->len > field->prefix_len) { + + dfield->len = field->prefix_len; + } } ut_ad(dtuple_check_typed(ref)); diff --git a/innobase/row/row0sel.c b/innobase/row/row0sel.c index 4732472d805..114ebf870b0 100644 --- a/innobase/row/row0sel.c +++ b/innobase/row/row0sel.c @@ -65,41 +65,50 @@ row_sel_sec_rec_is_for_clust_rec( rec_t* sec_rec, /* in: secondary index record */ dict_index_t* sec_index, /* in: secondary index */ rec_t* clust_rec, /* in: clustered index record */ - dict_index_t* clust_index __attribute__((unused))) - /* in: clustered index */ + dict_index_t* clust_index) /* in: clustered index */ { - dict_col_t* col; - byte* sec_field; - ulint sec_len; - byte* clust_field; - ulint clust_len; - ulint n; - ulint i; + dict_field_t* ifield; + dict_col_t* col; + byte* sec_field; + ulint sec_len; + byte* clust_field; + ulint clust_len; + ulint n; + ulint i; - n = dict_index_get_n_ordering_defined_by_user(sec_index); + UT_NOT_USED(clust_index); - for (i = 0; i < n; i++) { - col = dict_field_get_col( - dict_index_get_nth_field(sec_index, i)); + n = dict_index_get_n_ordering_defined_by_user(sec_index); - clust_field = rec_get_nth_field(clust_rec, - dict_col_get_clust_pos(col), - &clust_len); - sec_field = rec_get_nth_field(sec_rec, i, &sec_len); + for (i = 0; i < n; i++) { + ifield = dict_index_get_nth_field(sec_index, i); + col = dict_field_get_col(ifield); + + clust_field = rec_get_nth_field(clust_rec, + dict_col_get_clust_pos(col), + &clust_len); + sec_field = rec_get_nth_field(sec_rec, i, &sec_len); - if (sec_len != clust_len) { + if (ifield->prefix_len > 0 + && clust_len != UNIV_SQL_NULL + && clust_len > ifield->prefix_len) { - return(FALSE); + clust_len = ifield->prefix_len; } - if (0 != cmp_data_data(dict_col_get_type(col), - clust_field, clust_len, - sec_field, sec_len)) { - return(FALSE); - } - } + if (sec_len != clust_len) { - return(TRUE); + return(FALSE); + } + + if (0 != cmp_data_data(dict_col_get_type(col), + clust_field, clust_len, + sec_field, sec_len)) { + return(FALSE); + } + } + + return(TRUE); } /************************************************************************* @@ -606,7 +615,7 @@ row_sel_get_clust_rec( /* Try to place a lock on the index record */ err = lock_clust_rec_read_check_and_lock(0, clust_rec, index, - node->row_lock_mode, LOCK_ORDINARY, thr); + node->row_lock_mode, LOCK_ORDINARY, thr); if (err != DB_SUCCESS) { return(err); @@ -656,7 +665,7 @@ row_sel_get_clust_rec( *out_rec = clust_rec; return(DB_SUCCESS); - } + } } /* Fetch the columns needed in test conditions */ @@ -1850,9 +1859,11 @@ row_printf_step( } /******************************************************************** -Converts a key value stored in MySQL format to an Innobase dtuple. -The last field of the key value may be just a prefix of a fixed length -field: hence the parameter key_len. */ +Converts a key value stored in MySQL format to an Innobase dtuple. The last +field of the key value may be just a prefix of a fixed length field: hence +the parameter key_len. But currently we do not allow search keys where the +last field is only a prefix of the full key field len and print a warning if +such appears. */ void row_sel_convert_mysql_key_to_innobase( @@ -1863,17 +1874,24 @@ row_sel_convert_mysql_key_to_innobase( to index! */ byte* buf, /* in: buffer to use in field conversions */ + ulint buf_len, /* in: buffer length */ dict_index_t* index, /* in: index of the key value */ byte* key_ptr, /* in: MySQL key value */ ulint key_len) /* in: MySQL key value length */ { + byte* original_buf = buf; + dict_field_t* field; dfield_t* dfield; - ulint offset; - ulint len; + ulint data_offset; + ulint data_len; + ulint data_field_len; + ibool is_null; byte* key_end; ulint n_fields = 0; + ulint type; - UT_NOT_USED(index); + /* For documentation of the key value storage format in MySQL, see + ha_innobase::store_key_val_for_row() in ha_innodb.cc. */ key_end = key_ptr + key_len; @@ -1882,11 +1900,14 @@ row_sel_convert_mysql_key_to_innobase( dtuple_set_n_fields(tuple, ULINT_MAX); dfield = dtuple_get_nth_field(tuple, 0); + field = dict_index_get_nth_field(index, 0); if (dfield_get_type(dfield)->mtype == DATA_SYS) { - /* A special case: we are looking for a position in a - generated clustered index: the first and the only - ordering column is ROW_ID */ + /* A special case: we are looking for a position in the + generated clustered index which InnoDB automatically added + to a table with no primary key: the first and the only + ordering column is ROW_ID which InnoDB stored to the key_ptr + buffer. */ ut_a(key_len == DATA_ROW_ID_LEN); @@ -1897,70 +1918,114 @@ row_sel_convert_mysql_key_to_innobase( return; } - while (key_ptr < key_end) { - offset = 0; - len = dfield_get_type(dfield)->len; + while (key_ptr < key_end) { - n_fields++; + ut_a(dict_col_get_type(field->col)->mtype + == dfield_get_type(dfield)->mtype); + + data_offset = 0; + is_null = FALSE; if (!(dfield_get_type(dfield)->prtype & DATA_NOT_NULL)) { /* The first byte in the field tells if this is an SQL NULL value */ - offset = 1; + data_offset = 1; - if (*key_ptr != 0) { + if (*key_ptr != 0) { dfield_set_data(dfield, NULL, UNIV_SQL_NULL); - goto next_part; + is_null = TRUE; } } - row_mysql_store_col_in_innobase_format( - dfield, buf, key_ptr + offset, len, - dfield_get_type(dfield)->mtype, + type = dfield_get_type(dfield)->mtype; + + /* Calculate data length and data field total length */ + + if (type == DATA_BLOB) { + /* The key field is a column prefix of a BLOB or + TEXT type column */ + + ut_a(field->prefix_len > 0); + + /* MySQL stores the actual data length to the first 2 + bytes after the optional SQL NULL marker byte. The + storage format is little-endian. */ + + /* There are no key fields > 255 bytes currently in + MySQL */ + if (key_ptr[data_offset + 1] != 0) { + ut_print_timestamp(stderr); + fprintf(stderr, +" InnoDB: Error: BLOB or TEXT prefix > 255 bytes in query to table %s\n", + index->table_name); + } + + data_len = key_ptr[data_offset]; + data_field_len = data_offset + 2 + field->prefix_len; + data_offset += 2; + + type = DATA_CHAR; /* now that we know the length, we + store the column value like it would + be a fixed char field */ + } else if (field->prefix_len > 0) { + data_len = field->prefix_len; + data_field_len = data_offset + data_len; + } else { + data_len = dfield_get_type(dfield)->len; + data_field_len = data_offset + data_len; + } + + /* Storing may use at most data_len bytes of buf */ + + if (!is_null) { + row_mysql_store_col_in_innobase_format( + dfield, buf, key_ptr + data_offset, + data_len, type, dfield_get_type(dfield)->prtype & DATA_UNSIGNED); - next_part: - key_ptr += (offset + len); + buf += data_len; + } + + key_ptr += data_field_len; if (key_ptr > key_end) { - /* The last field in key was not a complete - field but a prefix of it. + /* The last field in key was not a complete key field + but a prefix of it. - Print a warning about this! HA_READ_PREFIX_LAST - does not currently work in InnoDB with partial-field - key value prefixes. Since MySQL currently uses a - padding trick to calculate LIKE 'abc%' type queries - there should never be partial-field prefixes - in searches. */ + Print a warning about this! HA_READ_PREFIX_LAST does + not currently work in InnoDB with partial-field key + value prefixes. Since MySQL currently uses a padding + trick to calculate LIKE 'abc%' type queries there + should never be partial-field prefixes in searches. */ ut_print_timestamp(stderr); fprintf(stderr, " InnoDB: Warning: using a partial-field key prefix in search\n"); - ut_ad(dfield_get_len(dfield) != UNIV_SQL_NULL); - - dfield_set_data(dfield, buf, - len - (ulint)(key_ptr - key_end)); + if (!is_null) { + dfield->len -= (ulint)(key_ptr - key_end); + } } - buf += len; - + n_fields++; + field++; dfield++; } - /* We set the length of tuple to n_fields: we assume that - the memory area allocated for it is big enough (usually - bigger than n_fields). */ + ut_a(buf <= original_buf + buf_len); + + /* We set the length of tuple to n_fields: we assume that the memory + area allocated for it is big enough (usually bigger than n_fields). */ dtuple_set_n_fields(tuple, n_fields); } /****************************************************************** Stores the row id to the prebuilt struct. */ -UNIV_INLINE +static void row_sel_store_row_id_to_prebuilt( /*=============================*/ @@ -1970,11 +2035,22 @@ row_sel_store_row_id_to_prebuilt( { byte* data; ulint len; + char err_buf[1000]; data = rec_get_nth_field(index_rec, dict_index_get_sys_col_pos(index, DATA_ROW_ID), &len); - ut_a(len == DATA_ROW_ID_LEN); + if (len != DATA_ROW_ID_LEN) { + rec_sprintf(err_buf, 900, index_rec); + + fprintf(stderr, +"InnoDB: Error: Row id field is wrong length %lu in table %s index %s\n" +"InnoDB: Field number %lu, record:\n%s\n", + len, index->table_name, index->name, + dict_index_get_sys_col_pos(index, DATA_ROW_ID), + err_buf); + ut_a(0); + } ut_memcpy(prebuilt->row_id, data, len); } @@ -3021,7 +3097,7 @@ rec_loop: if (prebuilt->select_lock_type != LOCK_NONE && set_also_gap_locks) { - /* Try to place a lock on the index record */ + /* Try to place a lock on the index record */ err = sel_set_rec_lock(rec, index, prebuilt->select_lock_type, diff --git a/innobase/row/row0umod.c b/innobase/row/row0umod.c index b84e55ca643..b22e494f891 100644 --- a/innobase/row/row0umod.c +++ b/innobase/row/row0umod.c @@ -428,7 +428,8 @@ row_undo_mod_del_unmark_sec( found = row_search_index_entry(index, entry, BTR_MODIFY_LEAF, &pcur, &mtr); if (!found) { - fprintf(stderr, "InnoDB: error in sec index entry del undo in\n" + fprintf(stderr, + "InnoDB: error in sec index entry del undo in\n" "InnoDB: index %s table %s\n", index->name, index->table->name); dtuple_sprintf(err_buf, 900, entry); @@ -570,7 +571,7 @@ row_undo_mod_upd_exist_sec( the row */ row_upd_index_replace_new_col_vals(entry, index, - node->update); + node->update, NULL); row_undo_mod_del_unmark_sec(node, thr, index, entry); } diff --git a/innobase/row/row0upd.c b/innobase/row/row0upd.c index 5fce1c1861b..db68479509d 100644 --- a/innobase/row/row0upd.c +++ b/innobase/row/row0upd.c @@ -72,8 +72,9 @@ searched delete is obviously to keep the x-latch for several steps of query graph execution. */ /*************************************************************** -Checks if an update vector changes some of the first fields of an index -record. */ +Checks if an update vector changes some of the first ordering fields of an +index record. This is only used in foreign key checks and we can assume +that index does not contain column prefixes. */ static ibool row_upd_changes_first_fields( @@ -234,7 +235,8 @@ row_upd_check_references_constraints( if (err != DB_SUCCESS) { if (got_s_lock) { - row_mysql_unfreeze_data_dictionary(trx); + row_mysql_unfreeze_data_dictionary( + trx); } mem_heap_free(heap); @@ -350,14 +352,15 @@ row_upd_index_entry_sys_field( } /*************************************************************** -Returns TRUE if row update changes size of some field in index -or if some field to be updated is stored externally in rec or update. */ +Returns TRUE if row update changes size of some field in index or if some +field to be updated is stored externally in rec or update. */ ibool -row_upd_changes_field_size( -/*=======================*/ +row_upd_changes_field_size_or_external( +/*===================================*/ /* out: TRUE if the update changes the size of - some field in index */ + some field in index or the field is external + in rec or update */ rec_t* rec, /* in: record in clustered index */ dict_index_t* index, /* in: clustered index */ upd_t* update) /* in: update vector */ @@ -820,69 +823,55 @@ void row_upd_index_replace_new_col_vals( /*===============================*/ dtuple_t* entry, /* in/out: index entry where replaced */ - dict_index_t* index, /* in: index; NOTE that may also be a + dict_index_t* index, /* in: index; NOTE that this may also be a non-clustered index */ - upd_t* update) /* in: update vector */ + upd_t* update, /* in: update vector */ + mem_heap_t* heap) /* in: memory heap to which we allocate and + copy the new values, set this as NULL if you + do not want allocation */ { + dict_field_t* field; upd_field_t* upd_field; dfield_t* dfield; dfield_t* new_val; - ulint field_no; - dict_index_t* clust_index; + ulint j; ulint i; ut_ad(index); - clust_index = dict_table_get_first_index(index->table); - dtuple_set_info_bits(entry, update->info_bits); - for (i = 0; i < upd_get_n_fields(update); i++) { - - upd_field = upd_get_nth_field(update, i); - - field_no = dict_index_get_nth_col_pos(index, - dict_index_get_nth_col_no(clust_index, - upd_field->field_no)); - if (field_no != ULINT_UNDEFINED) { - dfield = dtuple_get_nth_field(entry, field_no); + for (j = 0; j < dict_index_get_n_fields(index); j++) { - new_val = &(upd_field->new_val); + field = dict_index_get_nth_field(index, j); - dfield_set_data(dfield, new_val->data, new_val->len); - } - } -} - -/*************************************************************** -Replaces the new column values stored in the update vector to the -clustered index entry given. */ - -void -row_upd_clust_index_replace_new_col_vals( -/*=====================================*/ - dtuple_t* entry, /* in/out: index entry where replaced */ - upd_t* update) /* in: update vector */ -{ - upd_field_t* upd_field; - dfield_t* dfield; - dfield_t* new_val; - ulint field_no; - ulint i; + for (i = 0; i < upd_get_n_fields(update); i++) { - dtuple_set_info_bits(entry, update->info_bits); + upd_field = upd_get_nth_field(update, i); - for (i = 0; i < upd_get_n_fields(update); i++) { + if (upd_field->field_no == field->col->clust_pos) { - upd_field = upd_get_nth_field(update, i); + dfield = dtuple_get_nth_field(entry, j); - field_no = upd_field->field_no; + new_val = &(upd_field->new_val); - dfield = dtuple_get_nth_field(entry, field_no); + dfield_set_data(dfield, new_val->data, + new_val->len); + if (heap && new_val->len != UNIV_SQL_NULL) { + dfield->data = mem_heap_alloc(heap, + new_val->len); + ut_memcpy(dfield->data, new_val->data, + new_val->len); + } - new_val = &(upd_field->new_val); + if (field->prefix_len > 0 + && new_val->len != UNIV_SQL_NULL + && new_val->len > field->prefix_len) { - dfield_set_data(dfield, new_val->data, new_val->len); + dfield->len = field->prefix_len; + } + } + } } } @@ -931,9 +920,15 @@ row_upd_changes_ord_field_binary( upd_field = upd_get_nth_field(update, j); + /* Note that if the index field is a column prefix + then it may be that row does not contain an externally + stored part of the column value, and we cannot compare + the datas */ + if (col_pos == upd_field->field_no - && (row == NULL - || !dfield_datas_are_binary_equal( + && (row == NULL + || ind_field->prefix_len > 0 + || !dfield_datas_are_binary_equal( dtuple_get_nth_field(row, col_no), &(upd_field->new_val)))) { return(TRUE); @@ -978,8 +973,9 @@ row_upd_changes_some_index_ord_field_binary( } /*************************************************************** -Checks if an update vector changes some of the first fields of an index -record. */ +Checks if an update vector changes some of the first ordering fields of an +index record. This is only used in foreign key checks and we can assume +that index does not contain column prefixes. */ static ibool row_upd_changes_first_fields( @@ -1013,9 +1009,10 @@ row_upd_changes_first_fields( upd_field = upd_get_nth_field(update, j); if (col_pos == upd_field->field_no - && cmp_dfield_dfield( + && (ind_field->prefix_len > 0 + || 0 != cmp_dfield_dfield( dtuple_get_nth_field(entry, i), - &(upd_field->new_val))) { + &(upd_field->new_val)))) { return(TRUE); } } @@ -1204,7 +1201,7 @@ close_cur: } /* Build a new index entry */ - row_upd_index_replace_new_col_vals(entry, index, node->update); + row_upd_index_replace_new_col_vals(entry, index, node->update, NULL); /* Insert new index entry */ err = row_ins_index_entry(index, entry, NULL, 0, thr); @@ -1317,12 +1314,12 @@ row_upd_clust_rec_by_insert( entry = row_build_index_entry(node->row, index, heap); - row_upd_clust_index_replace_new_col_vals(entry, node->update); + row_upd_index_replace_new_col_vals(entry, index, node->update, NULL); row_upd_index_entry_sys_field(entry, index, DATA_TRX_ID, trx->id); /* If we return from a lock wait, for example, we may have - extern fields marked as not-owned in entry (marked if the + extern fields marked as not-owned in entry (marked in the if-branch above). We must unmark them. */ btr_cur_unmark_dtuple_extern_fields(entry, node->ext_vec, @@ -1702,9 +1699,9 @@ function_exit: /* Do some cleanup */ if (node->row != NULL) { - mem_heap_empty(node->heap); node->row = NULL; node->n_ext_vec = 0; + mem_heap_empty(node->heap); } node->state = UPD_NODE_UPDATE_CLUSTERED; diff --git a/innobase/row/row0vers.c b/innobase/row/row0vers.c index cd8b18e5e12..91aaba40812 100644 --- a/innobase/row/row0vers.c +++ b/innobase/row/row0vers.c @@ -27,6 +27,7 @@ Created 2/6/1997 Heikki Tuuri #include "row0upd.h" #include "rem0cmp.h" #include "read0read.h" +#include "lock0lock.h" /********************************************************************* Finds out if an active transaction has inserted or modified a secondary @@ -111,6 +112,14 @@ row_vers_impl_x_locked_off_kernel( return(NULL); } + if (!lock_check_trx_id_sanity(trx_id, clust_rec, clust_index, TRUE)) { + /* Corruption noticed: try to avoid a crash by returning */ + + mtr_commit(&mtr); + + return(NULL); + } + /* We look up if some earlier version of the clustered index record would require rec to be in a different state (delete marked or unmarked, or not existing). If there is such a version, then rec was @@ -177,7 +186,8 @@ row_vers_impl_x_locked_off_kernel( /* If we get here, we know that the trx_id transaction is still active and it has modified prev_version. Let us check - if prev_version would require rec to be in a different state. */ + if prev_version would require rec to be in a different + state. */ vers_del = rec_get_deleted_flag(prev_version); diff --git a/innobase/srv/srv0srv.c b/innobase/srv/srv0srv.c index 2a93ca966eb..d13d499dd17 100644 --- a/innobase/srv/srv0srv.c +++ b/innobase/srv/srv0srv.c @@ -286,6 +286,7 @@ ulint srv_test_n_mutexes = ULINT_MAX; i/o handler thread */ char* srv_io_thread_op_info[SRV_MAX_N_IO_THREADS]; +char* srv_io_thread_function[SRV_MAX_N_IO_THREADS]; time_t srv_last_monitor_time; @@ -2399,8 +2400,9 @@ srv_sprintf_innodb_monitor( srv_conc_n_threads, srv_conc_n_waiting_threads); #ifdef UNIV_LINUX buf += sprintf(buf, - "Main thread process no %lu, state: %s\n", + "Main thread process no. %lu, id %lu, state: %s\n", srv_main_thread_process_no, + srv_main_thread_id, srv_main_thread_op_info); #else buf += sprintf(buf, @@ -2464,8 +2466,8 @@ srv_lock_timeout_and_monitor_thread( ulint i; #ifdef UNIV_DEBUG_THREAD_CREATION - printf("Lock timeout thread starts\n"); - printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id())); + printf("Lock timeout thread starts, id %lu\n", + os_thread_pf(os_thread_get_curr_id())); #endif UT_NOT_USED(arg); srv_last_monitor_time = time(NULL); @@ -2637,8 +2639,8 @@ srv_error_monitor_thread( UT_NOT_USED(arg); #ifdef UNIV_DEBUG_THREAD_CREATION - printf("Error monitor thread starts\n"); - printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id())); + printf("Error monitor thread starts, id %lu\n", + os_thread_pf(os_thread_get_curr_id())); #endif loop: srv_error_monitor_active = TRUE; @@ -2760,8 +2762,8 @@ srv_master_thread( UT_NOT_USED(arg); #ifdef UNIV_DEBUG_THREAD_CREATION - printf("Master thread starts\n"); - printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id())); + printf("Master thread starts, id %lu\n", + os_thread_pf(os_thread_get_curr_id())); #endif srv_main_thread_process_no = os_proc_get_number(); srv_main_thread_id = os_thread_pf(os_thread_get_curr_id()); diff --git a/innobase/srv/srv0start.c b/innobase/srv/srv0start.c index 1f278d82bc0..964e728b23c 100644 --- a/innobase/srv/srv0start.c +++ b/innobase/srv/srv0start.c @@ -415,8 +415,8 @@ io_handler_thread( segment = *((ulint*)arg); #ifdef UNIV_DEBUG_THREAD_CREATION - printf("Io handler thread %lu starts\n", segment); - printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id())); + printf("Io handler thread %lu starts, id %lu\n", segment, + os_thread_pf(os_thread_get_curr_id())); #endif for (i = 0;; i++) { fil_aio_wait(segment); @@ -1492,7 +1492,9 @@ innobase_shutdown_for_mysql(void) } /* 1. Flush buffer pool to disk, write the current lsn to - the tablespace header(s), and copy all log data to archive */ + the tablespace header(s), and copy all log data to archive. + The step 1 is the real InnoDB shutdown. The remaining steps + just free data structures after the shutdown. */ logs_empty_and_mark_files_at_shutdown(); diff --git a/innobase/trx/trx0rec.c b/innobase/trx/trx0rec.c index 05e179e06a5..9453189d598 100644 --- a/innobase/trx/trx0rec.c +++ b/innobase/trx/trx0rec.c @@ -272,8 +272,8 @@ trx_undo_page_report_insert( mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE, ptr - undo_page); - /* Write the log entry to the REDO log of this change in the UNDO log */ - + /* Write the log entry to the REDO log of this change in the UNDO + log */ trx_undof_page_add_undo_rec_log(undo_page, first_free, ptr - undo_page, mtr); return(first_free); @@ -492,7 +492,8 @@ trx_undo_page_report_modify( /* Reserve 2 bytes for the pointer to the next undo log record */ ptr += 2; - /* Store first some general parameters to the undo log */ + /* Store first some general parameters to the undo log */ + if (update) { if (rec_get_deleted_flag(rec)) { type_cmpl = TRX_UNDO_UPD_DEL_REC; @@ -526,8 +527,7 @@ trx_undo_page_report_modify( /* Store the values of the system columns */ trx_id = dict_index_rec_get_sys_col(index, DATA_TRX_ID, rec); - roll_ptr = dict_index_rec_get_sys_col(index, DATA_ROLL_PTR, rec); - + roll_ptr = dict_index_rec_get_sys_col(index, DATA_ROLL_PTR, rec); len = mach_dulint_write_compressed(ptr, trx_id); ptr += len; @@ -632,7 +632,11 @@ trx_undo_page_report_modify( columns which occur as ordering fields in any index. This info is used in the purge of old versions where we use it to build and search the delete marked index records, to look if we can remove them from the - index tree. */ + index tree. Note that starting from 4.0.14 also externally stored + fields can be ordering in some index. But we always store at least + 384 first bytes locally to the clustered index record, which means + we can construct the column prefix fields in the index from the + stored data. */ if (!update || !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) { @@ -1408,11 +1412,11 @@ trx_undo_prev_version_build( return(DB_ERROR); } - if (row_upd_changes_field_size(rec, index, update)) { - - entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap); + if (row_upd_changes_field_size_or_external(rec, index, update)) { - row_upd_clust_index_replace_new_col_vals(entry, update); + entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, + heap); + row_upd_index_replace_new_col_vals(entry, index, update, heap); buf = mem_heap_alloc(heap, rec_get_converted_size(entry)); diff --git a/innobase/trx/trx0roll.c b/innobase/trx/trx0roll.c index a9f8c5ad22c..7d1b341221c 100644 --- a/innobase/trx/trx0roll.c +++ b/innobase/trx/trx0roll.c @@ -52,6 +52,11 @@ trx_general_rollback_for_mysql( que_thr_t* thr; roll_node_t* roll_node; + /* Tell Innobase server that there might be work for + utility threads: */ + + srv_active_wake_master_thread(); + trx_start_if_not_started(trx); heap = mem_heap_create(512); @@ -89,6 +94,11 @@ trx_general_rollback_for_mysql( ut_a(trx->error_state == DB_SUCCESS); + /* Tell Innobase server that there might be work for + utility threads: */ + + srv_active_wake_master_thread(); + return((int) trx->error_state); } @@ -110,20 +120,8 @@ trx_rollback_for_mysql( trx->op_info = (char *) "rollback"; - /* Tell Innobase server that there might be work for - utility threads: */ - - srv_active_wake_master_thread(); - err = trx_general_rollback_for_mysql(trx, FALSE, NULL); - trx_mark_sql_stat_end(trx); - - /* Tell Innobase server that there might be work for - utility threads: */ - - srv_active_wake_master_thread(); - trx->op_info = (char *) ""; return(err); @@ -147,26 +145,192 @@ trx_rollback_last_sql_stat_for_mysql( trx->op_info = (char *) "rollback of SQL statement"; - /* Tell Innobase server that there might be work for - utility threads: */ - - srv_active_wake_master_thread(); - err = trx_general_rollback_for_mysql(trx, TRUE, &(trx->last_sql_stat_start)); + /* The following call should not be needed, but we play safe: */ trx_mark_sql_stat_end(trx); - /* Tell Innobase server that there might be work for - utility threads: */ + trx->op_info = (char *) ""; + + return(err); +} - srv_active_wake_master_thread(); +/*********************************************************************** +Frees savepoint structs. */ - trx->op_info = (char *) ""; +void +trx_roll_savepoints_free( +/*=====================*/ + trx_t* trx, /* in: transaction handle */ + trx_named_savept_t* savep) /* in: free all savepoints > this one; + if this is NULL, free all savepoints + of trx */ +{ + trx_named_savept_t* next_savep; + + if (savep == NULL) { + savep = UT_LIST_GET_FIRST(trx->trx_savepoints); + } else { + savep = UT_LIST_GET_NEXT(trx_savepoints, savep); + } + while (savep != NULL) { + next_savep = UT_LIST_GET_NEXT(trx_savepoints, savep); + + UT_LIST_REMOVE(trx_savepoints, trx->trx_savepoints, savep); + mem_free(savep->name); + mem_free(savep); + + savep = next_savep; + } +} + +/*********************************************************************** +Rolls back a transaction back to a named savepoint. Modifications after the +savepoint are undone but InnoDB does NOT release the corresponding locks +which are stored in memory. If a lock is 'implicit', that is, a new inserted +row holds a lock where the lock information is carried by the trx id stored in +the row, these locks are naturally released in the rollback. Savepoints which +were set after this savepoint are deleted. */ + +ulint +trx_rollback_to_savepoint_for_mysql( +/*================================*/ + /* out: if no savepoint + of the name found then + DB_NO_SAVEPOINT, + otherwise DB_SUCCESS */ + trx_t* trx, /* in: transaction handle */ + char* savepoint_name, /* in: savepoint name */ + ib_longlong* mysql_binlog_cache_pos) /* out: the MySQL binlog cache + position corresponding to this + savepoint; MySQL needs this + information to remove the + binlog entries of the queries + executed after the savepoint */ +{ + trx_named_savept_t* savep; + ulint err; + + savep = UT_LIST_GET_FIRST(trx->trx_savepoints); + + while (savep != NULL) { + if (0 == ut_strcmp(savep->name, savepoint_name)) { + /* Found */ + break; + } + savep = UT_LIST_GET_NEXT(trx_savepoints, savep); + } + + if (savep == NULL) { + + return(DB_NO_SAVEPOINT); + } + + if (trx->conc_state == TRX_NOT_STARTED) { + ut_print_timestamp(stderr); + fprintf(stderr, +" InnoDB: Error: transaction has a savepoint %s though it is not started\n", + savep->name); + return(DB_ERROR); + } + + /* We can now free all savepoints strictly later than this one */ + + trx_roll_savepoints_free(trx, savep); + + *mysql_binlog_cache_pos = savep->mysql_binlog_cache_pos; + + trx->op_info = (char *) "rollback to a savepoint"; + + err = trx_general_rollback_for_mysql(trx, TRUE, &(savep->savept)); + + /* Store the current undo_no of the transaction so that we know where + to roll back if we have to roll back the next SQL statement: */ + + trx_mark_sql_stat_end(trx); + + trx->op_info = (char *) ""; + return(err); } /*********************************************************************** +Creates a named savepoint. If the transaction is not yet started, starts it. +If there is already a savepoint of the same name, this call erases that old +savepoint and replaces it with a new. Savepoints are deleted in a transaction +commit or rollback. */ + +ulint +trx_savepoint_for_mysql( +/*====================*/ + /* out: always DB_SUCCESS */ + trx_t* trx, /* in: transaction handle */ + char* savepoint_name, /* in: savepoint name */ + ib_longlong binlog_cache_pos) /* in: MySQL binlog cache + position corresponding to this + connection at the time of the + savepoint */ +{ + trx_named_savept_t* savep; + + ut_a(trx); + ut_a(savepoint_name); + + trx_start_if_not_started(trx); + + savep = UT_LIST_GET_FIRST(trx->trx_savepoints); + + while (savep != NULL) { + if (0 == ut_strcmp(savep->name, savepoint_name)) { + /* Found */ + break; + } + savep = UT_LIST_GET_NEXT(trx_savepoints, savep); + } + + if (savep) { + /* There is a savepoint with the same name: free that */ + + UT_LIST_REMOVE(trx_savepoints, trx->trx_savepoints, savep); + + mem_free(savep->name); + mem_free(savep); + } + + /* Create a new savepoint and add it as the last in the list */ + + savep = mem_alloc(sizeof(trx_named_savept_t)); + + savep->name = mem_alloc(1 + ut_strlen(savepoint_name)); + ut_memcpy(savep->name, savepoint_name, 1 + ut_strlen(savepoint_name)); + + savep->savept = trx_savept_take(trx); + + savep->mysql_binlog_cache_pos = binlog_cache_pos; + + UT_LIST_ADD_LAST(trx_savepoints, trx->trx_savepoints, savep); + + return(DB_SUCCESS); +} + +/*********************************************************************** +Returns a transaction savepoint taken at this point in time. */ + +trx_savept_t +trx_savept_take( +/*============*/ + /* out: savepoint */ + trx_t* trx) /* in: transaction */ +{ + trx_savept_t savept; + + savept.least_undo_no = trx->undo_no; + + return(savept); +} + +/*********************************************************************** Rollback or clean up transactions which have no user session. If the transaction already was committed, then we clean up a possible insert undo log. If the transaction was not yet committed, then we roll it back. */ @@ -325,22 +489,6 @@ loop: goto loop; } - -/*********************************************************************** -Returns a transaction savepoint taken at this point in time. */ - -trx_savept_t -trx_savept_take( -/*============*/ - /* out: savepoint */ - trx_t* trx) /* in: transaction */ -{ - trx_savept_t savept; - - savept.least_undo_no = trx->undo_no; - - return(savept); -} /*********************************************************************** Creates an undo number array. */ diff --git a/innobase/trx/trx0sys.c b/innobase/trx/trx0sys.c index 51aad60d3e2..f1b03fff3bd 100644 --- a/innobase/trx/trx0sys.c +++ b/innobase/trx/trx0sys.c @@ -321,8 +321,8 @@ trx_sys_doublewrite_restore_corrupt_pages(void) for (i = 0; i < TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 2; i++) { - space_id = mach_read_from_4(page + FIL_PAGE_SPACE); page_no = mach_read_from_4(page + FIL_PAGE_OFFSET); + space_id = 0; if (!fil_check_adress_in_tablespace(space_id, page_no)) { fprintf(stderr, diff --git a/innobase/trx/trx0trx.c b/innobase/trx/trx0trx.c index d73d6327d76..9233e861784 100644 --- a/innobase/trx/trx0trx.c +++ b/innobase/trx/trx0trx.c @@ -135,6 +135,8 @@ trx_create( trx->lock_heap = mem_heap_create_in_buffer(256); UT_LIST_INIT(trx->trx_locks); + UT_LIST_INIT(trx->trx_savepoints); + trx->dict_operation_lock_mode = 0; trx->has_search_latch = FALSE; trx->search_latch_timeout = BTR_SEA_TIMEOUT; @@ -807,6 +809,9 @@ trx_commit_off_kernel( mutex_enter(&kernel_mutex); } + /* Free savepoints */ + trx_roll_savepoints_free(trx, NULL); + trx->conc_state = TRX_NOT_STARTED; trx->rseg = NULL; trx->undo_no = ut_dulint_zero; diff --git a/innobase/ut/ut0mem.c b/innobase/ut/ut0mem.c index 174ae4cc6bb..f5d207d8bba 100644 --- a/innobase/ut/ut0mem.c +++ b/innobase/ut/ut0mem.c @@ -166,7 +166,7 @@ ut_free( } /************************************************************************** -Frees all allocated memory not freed yet. */ +Frees in shutdown all allocated memory not freed yet. */ void ut_free_all_mem(void) @@ -174,7 +174,7 @@ ut_free_all_mem(void) { ut_mem_block_t* block; - os_fast_mutex_lock(&ut_list_mutex); + os_fast_mutex_free(&ut_list_mutex); while ((block = UT_LIST_GET_FIRST(ut_mem_block_list))) { @@ -187,11 +187,11 @@ ut_free_all_mem(void) free(block); } - os_fast_mutex_unlock(&ut_list_mutex); - - ut_a(ut_total_allocated_memory == 0); - - os_fast_mutex_free(&ut_list_mutex); + if (ut_total_allocated_memory != 0) { + fprintf(stderr, +"InnoDB: Warning: after shutdown total allocated memory is %lu\n", + ut_total_allocated_memory); + } } /************************************************************************** diff --git a/innobase/ut/ut0ut.c b/innobase/ut/ut0ut.c index 95037ec3570..06bfb5c45ba 100644 --- a/innobase/ut/ut0ut.c +++ b/innobase/ut/ut0ut.c @@ -53,6 +53,8 @@ ut_get_high32( ulint a) /* in: ulint */ { #if SIZEOF_LONG == 4 + UT_NOT_USED(a); + return 0; #else return(a >> 32); diff --git a/mysql-test/r/innodb.result b/mysql-test/r/innodb.result index e2dea324ff2..9d1c232e830 100644 --- a/mysql-test/r/innodb.result +++ b/mysql-test/r/innodb.result @@ -787,16 +787,6 @@ id id3 100 2 UNLOCK TABLES; DROP TABLE t1; -create table t1 (a char(20), unique (a(5))) type=innodb; -Incorrect sub part key. The used key part isn't a string, the used length is longer than the key part or the table handler doesn't support unique sub keys -create table t1 (a char(20), index (a(5))) type=innodb; -show create table t1; -Table Create Table -t1 CREATE TABLE `t1` ( - `a` char(20) default NULL, - KEY `a` (`a`) -) TYPE=InnoDB -drop table t1; create temporary table t1 (a int not null auto_increment, primary key(a)) type=innodb; insert into t1 values (NULL),(NULL),(NULL); delete from t1 where a=3; diff --git a/mysql-test/t/innodb.test b/mysql-test/t/innodb.test index dc3c76f1a91..cf203d87c8b 100644 --- a/mysql-test/t/innodb.test +++ b/mysql-test/t/innodb.test @@ -472,15 +472,6 @@ UNLOCK TABLES; DROP TABLE t1; # -# Test prefix key -# ---error 1089 -create table t1 (a char(20), unique (a(5))) type=innodb; -create table t1 (a char(20), index (a(5))) type=innodb; -show create table t1; -drop table t1; - -# # Test using temporary table and auto_increment # diff --git a/sql/ha_innodb.cc b/sql/ha_innodb.cc index fd030fff091..28e61f44111 100644 --- a/sql/ha_innodb.cc +++ b/sql/ha_innodb.cc @@ -130,11 +130,44 @@ static void innobase_print_error(const char* db_errpfx, char* buffer); /* General functions */ /********************************************************************** +Save some CPU by testing the value of srv_thread_concurrency in inline +functions. */ +inline +void +innodb_srv_conc_enter_innodb( +/*=========================*/ + trx_t* trx) /* in: transaction handle */ +{ + if (srv_thread_concurrency >= 500) { + + return; + } + + srv_conc_enter_innodb(trx); +} + +/********************************************************************** +Save some CPU by testing the value of srv_thread_concurrency in inline +functions. */ +inline +void +innodb_srv_conc_exit_innodb( +/*========================*/ + trx_t* trx) /* in: transaction handle */ +{ + if (srv_thread_concurrency >= 500) { + + return; + } + + srv_conc_exit_innodb(trx); +} + +/********************************************************************** Releases possible search latch and InnoDB thread FIFO ticket. These should be released at each SQL statement end, and also when mysqld passes the control to the client. It does no harm to release these also in the middle of an SQL statement. */ -static inline void innobase_release_stat_resources( @@ -183,7 +216,9 @@ innobase_active_small(void) } /************************************************************************ -Converts an InnoDB error code to a MySQL error code. */ +Converts an InnoDB error code to a MySQL error code and also tells to MySQL +about a possible transaction rollback inside InnoDB caused by a lock wait +timeout or a deadlock. */ static int convert_error_code_to_mysql( @@ -206,10 +241,10 @@ convert_error_code_to_mysql( } else if (error == (int) DB_ERROR) { - return(HA_ERR_NO_ACTIVE_RECORD); + return(-1); /* unspecified error */ } else if (error == (int) DB_DEADLOCK) { - /* Since we roll back the whole transaction, we must + /* Since we rolled back the whole transaction, we must tell it also to MySQL so that MySQL knows to empty the cached binlog for this transaction */ @@ -221,11 +256,10 @@ convert_error_code_to_mysql( } else if (error == (int) DB_LOCK_WAIT_TIMEOUT) { - /* Since we roll back the whole transaction, we must + /* Since we rolled back the whole transaction, we must tell it also to MySQL so that MySQL knows to empty the cached binlog for this transaction */ - if (thd) { ha_rollback(thd); } @@ -271,6 +305,9 @@ convert_error_code_to_mysql( } else if (error == (int) DB_CORRUPTION) { return(HA_ERR_CRASHED); + } else if (error == (int) DB_NO_SAVEPOINT) { + + return(HA_ERR_NO_SAVEPOINT); } else { return(-1); // Unknown error } @@ -941,18 +978,23 @@ innobase_commit( DBUG_ENTER("innobase_commit"); DBUG_PRINT("trans", ("ending transaction")); - /* The flag thd->transaction.all.innodb_active_trans is set to 1 - in ::external_lock and ::start_stmt, and it is only set to 0 in - a commit or a rollback. If it is 0 we know there cannot be resources - to be freed and we can return immediately. */ + trx = check_trx_exists(thd); - if (thd->transaction.all.innodb_active_trans == 0) { + /* The flag thd->transaction.all.innodb_active_trans is set to 1 in + ::external_lock, ::start_stmt, and innobase_savepoint, and it is only + set to 0 in a commit or a rollback. If it is 0 we know there cannot be + resources to be freed and we could return immediately. For the time + being we play safe and do the cleanup though there should be nothing + to clean up. */ - DBUG_RETURN(0); + if (thd->transaction.all.innodb_active_trans == 0 + && trx->conc_state != TRX_NOT_STARTED) { + + fprintf(stderr, +"InnoDB: Error: thd->transaction.all.innodb_active_trans == 0\n" +"InnoDB: but trx->conc_state != TRX_NOT_STARTED\n"); } - trx = check_trx_exists(thd); - if (trx_handle != (void*)&innodb_dummy_stmt_trx_handle || (!(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))) { @@ -964,9 +1006,9 @@ innobase_commit( /* If we had reserved the auto-inc lock for some table in this SQL statement we release it now */ - srv_conc_enter_innodb(trx); + innodb_srv_conc_enter_innodb(trx); row_unlock_table_autoinc_for_mysql(trx); - srv_conc_exit_innodb(trx); + innodb_srv_conc_exit_innodb(trx); } /* Store the current undo_no of the transaction so that we know where to roll back if we have to roll back the next @@ -1050,7 +1092,7 @@ innobase_commit_complete( } /********************************************************************* -Rolls back a transaction or the latest SQL statement in an InnoDB database. */ +Rolls back a transaction or the latest SQL statement. */ int innobase_rollback( @@ -1071,25 +1113,64 @@ innobase_rollback( trx = check_trx_exists(thd); if (trx->auto_inc_lock) { - - /* If we had reserved the auto-inc lock for - some table in this SQL statement, we release it now */ - - srv_conc_enter_innodb(trx); + /* If we had reserved the auto-inc lock for some table (if + we come here to roll back the latest SQL statement) we + release it now before a possibly lengthy rollback */ + + innodb_srv_conc_enter_innodb(trx); row_unlock_table_autoinc_for_mysql(trx); - srv_conc_exit_innodb(trx); + innodb_srv_conc_exit_innodb(trx); } - srv_conc_enter_innodb(trx); + innodb_srv_conc_enter_innodb(trx); + + if (trx_handle != (void*)&innodb_dummy_stmt_trx_handle + || (!(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))) { - if (trx_handle != (void*)&innodb_dummy_stmt_trx_handle) { error = trx_rollback_for_mysql(trx); - thd->transaction.all.innodb_active_trans=0; + thd->transaction.all.innodb_active_trans = 0; } else { error = trx_rollback_last_sql_stat_for_mysql(trx); } - srv_conc_exit_innodb(trx); + innodb_srv_conc_exit_innodb(trx); + + /* Release a possible FIFO ticket and search latch */ + innobase_release_stat_resources(trx); + + DBUG_RETURN(convert_error_code_to_mysql(error, NULL)); +} + +/********************************************************************* +Rolls back a transaction to a savepoint. */ + +int +innobase_rollback_to_savepoint( +/*===========================*/ + /* out: 0 if success, HA_ERR_NO_SAVEPOINT if + no savepoint with the given name */ + THD* thd, /* in: handle to the MySQL thread of the user + whose transaction should be rolled back */ + char* savepoint_name, /* in: savepoint name */ + my_off_t* binlog_cache_pos)/* out: position which corresponds to the + savepoint in the binlog cache of this + transaction, not defined if error */ +{ + ib_longlong mysql_binlog_cache_pos; + int error = 0; + trx_t* trx; + + DBUG_ENTER("innobase_rollback_to_savepoint"); + + trx = check_trx_exists(thd); + + innodb_srv_conc_enter_innodb(trx); + + error = trx_rollback_to_savepoint_for_mysql(trx, savepoint_name, + &mysql_binlog_cache_pos); + innodb_srv_conc_exit_innodb(trx); + + *binlog_cache_pos = (my_off_t)mysql_binlog_cache_pos; /* Release a possible FIFO ticket and search latch */ innobase_release_stat_resources(trx); @@ -1098,6 +1179,46 @@ innobase_rollback( } /********************************************************************* +Sets a transaction savepoint. */ + +int +innobase_savepoint( +/*===============*/ + /* out: always 0, that is, always succeeds */ + THD* thd, /* in: handle to the MySQL thread */ + char* savepoint_name, /* in: savepoint name */ + my_off_t binlog_cache_pos)/* in: offset up to which the current + transaction has cached log entries to its + binlog cache, not defined if no transaction + active, or we are in the autocommit state, or + binlogging is not switched on */ +{ + int error = 0; + trx_t* trx; + + DBUG_ENTER("innobase_savepoint"); + + if (!(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))) { + /* In the autocommit state there is no sense to set a + savepoint: we return immediate success */ + DBUG_RETURN(0); + } + + trx = check_trx_exists(thd); + + /* Setting a savepoint starts a transaction inside InnoDB since + it allocates resources for it (memory to store the savepoint name, + for example) */ + + thd->transaction.all.innodb_active_trans = 1; + + error = trx_savepoint_for_mysql(trx, savepoint_name, + (ib_longlong)binlog_cache_pos); + + DBUG_RETURN(convert_error_code_to_mysql(error, NULL)); +} + +/********************************************************************* Frees a possible InnoDB trx object associated with the current THD. */ @@ -1220,7 +1341,6 @@ ha_innobase::open( { dict_table_t* ib_table; int error = 0; - uint buff_len; char norm_name[1000]; DBUG_ENTER("ha_innobase::open"); @@ -1245,11 +1365,11 @@ ha_innobase::open( fields when packed actually became 1 byte longer, when we also stored the string length as the first byte. */ - buff_len = table->reclength + table->max_key_length + upd_and_key_val_buff_len = table->reclength + table->max_key_length + MAX_REF_PARTS * 3; if (!(mysql_byte*) my_multi_malloc(MYF(MY_WME), - &upd_buff, buff_len, - &key_val_buff, buff_len, + &upd_buff, upd_and_key_val_buff_len, + &key_val_buff, upd_and_key_val_buff_len, NullS)) { free_share(share); DBUG_RETURN(1); @@ -1500,6 +1620,10 @@ innobase_mysql_cmp( case FIELD_TYPE_STRING: case FIELD_TYPE_VAR_STRING: + case FIELD_TYPE_TINY_BLOB: + case FIELD_TYPE_MEDIUM_BLOB: + case FIELD_TYPE_BLOB: + case FIELD_TYPE_LONG_BLOB: ret = my_sortncmp((const char*) a, a_length, (const char*) b, b_length); if (ret < 0) { @@ -1526,7 +1650,7 @@ get_innobase_type_from_mysql_type( /* out: DATA_BINARY, DATA_VARCHAR, ... */ Field* field) /* in: MySQL field */ { - /* The following asserts check that MySQL type code fits in + /* The following asserts check that the MySQL type code fits in 8 bits: this is used in ibuf and also when DATA_NOT_NULL is ORed to the type */ @@ -1537,6 +1661,8 @@ get_innobase_type_from_mysql_type( DBUG_ASSERT((ulint)FIELD_TYPE_DECIMAL < 256); switch (field->type()) { + /* NOTE that we only allow string types in DATA_MYSQL + and DATA_VARMYSQL */ case FIELD_TYPE_VAR_STRING: if (field->flags & BINARY_FLAG) { return(DATA_BINARY); @@ -1607,33 +1733,98 @@ ha_innobase::store_key_val_for_row( KEY_PART_INFO* key_part = key_info->key_part; KEY_PART_INFO* end = key_part + key_info->key_parts; char* buff_start = buff; + enum_field_types mysql_type; + Field* field; + ulint blob_len; + byte* blob_data; + ibool is_null; DBUG_ENTER("store_key_val_for_row"); + /* The format for storing a key field in MySQL is the following: + + 1. If the column can be NULL, then in the first byte we put 1 if the + field value is NULL, 0 otherwise. + + 2. If the column is of a BLOB type (it must be a column prefix field + in this case), then we put the length of the data in the field to the + next 2 bytes, in the little-endian format. If the field is SQL NULL, + then these 2 bytes are set to 0. Note that the length of data in the + field is <= column prefix length. + + 3. In a column prefix field, prefix_len next bytes are reserved for + data. In a normal field the max field length next bytes are reserved + for data. For a VARCHAR(n) the max field length is n. If the stored + value is the SQL NULL then these data bytes are set to 0. */ + + /* We have to zero-fill the 'ref' buffer so that MySQL is able to + use a simple memcmp to compare two key values to determine if they + are equal */ + + bzero(buff, ref_length); + for (; key_part != end; key_part++) { + is_null = FALSE; if (key_part->null_bit) { - /* Store 0 if the key part is a NULL part */ - if (record[key_part->null_offset] & key_part->null_bit) { - *buff++ = 1; - continue; - } - - *buff++ = 0; + *buff = 1; + is_null = TRUE; + } else { + *buff = 0; + } + buff++; } - memcpy(buff, record + key_part->offset, key_part->length); - buff += key_part->length; - } + field = key_part->field; + mysql_type = field->type(); + + if (mysql_type == FIELD_TYPE_TINY_BLOB + || mysql_type == FIELD_TYPE_MEDIUM_BLOB + || mysql_type == FIELD_TYPE_BLOB + || mysql_type == FIELD_TYPE_LONG_BLOB) { - /* - We have to zero-fill the 'ref' buffer so that MySQL is able to - use a simple memcmp to compare two key values to determine if they - are equal - */ - bzero(buff, (ref_length- (uint) (buff - buff_start))); + ut_a(key_part->key_part_flag & HA_PART_KEY); + + if (is_null) { + buff += key_part->length + 2; + + continue; + } + + blob_data = row_mysql_read_blob_ref(&blob_len, + (byte*) (record + + (ulint)get_field_offset(table, field)), + (ulint) field->pack_length()); + + ut_a(get_field_offset(table, field) + == key_part->offset); + if (blob_len > key_part->length) { + blob_len = key_part->length; + } + + /* MySQL reserves 2 bytes for the length and the + storage of the number is little-endian */ + + ut_a(blob_len < 256); + *buff = blob_len; + buff += 2; + + memcpy(buff, blob_data, blob_len); + + buff += key_part->length; + } else { + if (is_null) { + buff += key_part->length; + + continue; + } + memcpy(buff, record + key_part->offset, + key_part->length); + buff += key_part->length; + } + } DBUG_RETURN((uint)(buff - buff_start)); } @@ -1905,9 +2096,9 @@ ha_innobase::write_row( The lock is released at each SQL statement's end. */ - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); error = row_lock_table_autoinc_for_mysql(prebuilt); - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); if (error != DB_SUCCESS) { @@ -1918,14 +2109,15 @@ ha_innobase::write_row( dict_table_autoinc_update(prebuilt->table, auto_inc); } else { - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); if (!prebuilt->trx->auto_inc_lock) { error = row_lock_table_autoinc_for_mysql( prebuilt); if (error != DB_SUCCESS) { - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb( + prebuilt->trx); error = convert_error_code_to_mysql( error, user_thd); @@ -1939,7 +2131,7 @@ ha_innobase::write_row( auto_inc = dict_table_autoinc_get(prebuilt->table); incremented_auto_inc_counter = TRUE; - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); /* We can give the new value for MySQL to place in the field */ @@ -1962,11 +2154,11 @@ ha_innobase::write_row( build_template(prebuilt, NULL, table, ROW_MYSQL_WHOLE_ROW); } - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); error = row_insert_for_mysql((byte*) record, prebuilt); - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); if (error != DB_SUCCESS) { /* If the insert did not succeed we restore the value of @@ -2037,7 +2229,6 @@ innobase_convert_and_store_changed_col( while (len > 0 && data[len - 1] == ' ') { len--; } - } else if (col_type == DATA_INT) { /* Store integer data in InnoDB in a big-endian format, sign bit negated, if signed */ @@ -2075,9 +2266,11 @@ calc_row_difference( struct st_table* table, /* in: table in MySQL data dictionary */ mysql_byte* upd_buff, /* in: buffer to use */ + ulint buff_len, /* in: buffer length */ row_prebuilt_t* prebuilt, /* in: InnoDB prebuilt struct */ THD* thd) /* in: user thread */ { + mysql_byte* original_upd_buff = upd_buff; Field* field; uint n_fields; ulint o_len; @@ -2159,12 +2352,13 @@ calc_row_difference( (prebuilt->table->cols + i)->clust_pos; n_changed++; } - ; } uvect->n_fields = n_changed; uvect->info_bits = 0; + ut_a(buf <= (byte*)original_upd_buff + buff_len); + return(0); } @@ -2213,17 +2407,19 @@ ha_innobase::update_row( (uses upd_buff of the handle) */ calc_row_difference(uvect, (mysql_byte*) old_row, new_row, table, - upd_buff, prebuilt, user_thd); + upd_buff, (ulint)upd_and_key_val_buff_len, + prebuilt, user_thd); + /* This is not a delete */ prebuilt->upd_node->is_delete = FALSE; assert(prebuilt->template_type == ROW_MYSQL_WHOLE_ROW); - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); error = row_update_for_mysql((byte*) old_row, prebuilt); - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); error = convert_error_code_to_mysql(error, user_thd); @@ -2267,11 +2463,11 @@ ha_innobase::delete_row( prebuilt->upd_node->is_delete = TRUE; - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); error = row_update_for_mysql((byte*) record, prebuilt); - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); error = convert_error_code_to_mysql(error, user_thd); @@ -2459,10 +2655,11 @@ ha_innobase::index_read( prebuilt->search_tuple */ row_sel_convert_mysql_key_to_innobase(prebuilt->search_tuple, - (byte*) key_val_buff, - index, - (byte*) key_ptr, - (ulint) key_len); + (byte*) key_val_buff, + (ulint)upd_and_key_val_buff_len, + index, + (byte*) key_ptr, + (ulint) key_len); } else { /* We position the cursor to the last or the first entry in the index */ @@ -2484,11 +2681,11 @@ ha_innobase::index_read( last_match_mode = match_mode; - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); ret = row_search_for_mysql((byte*) buf, mode, prebuilt, match_mode, 0); - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); if (ret == DB_SUCCESS) { error = 0; @@ -2632,11 +2829,11 @@ ha_innobase::general_fetch( ut_a(prebuilt->trx == (trx_t*) current_thd->transaction.all.innobase_tid); - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); ret = row_search_for_mysql((byte*)buf, 0, prebuilt, match_mode, direction); - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); if (ret == DB_SUCCESS) { error = 0; @@ -2939,7 +3136,6 @@ ha_innobase::position( } } - /********************************************************************* Creates a table definition to an InnoDB database. */ static @@ -2958,6 +3154,8 @@ create_table_def( ulint col_type; ulint nulls_allowed; ulint unsigned_type; + ulint binary_type; + ulint nonlatin1_type; ulint i; DBUG_ENTER("create_table_def"); @@ -2986,9 +3184,23 @@ create_table_def( unsigned_type = 0; } + if (strcmp(default_charset_info->name, "latin1") != 0) { + nonlatin1_type = DATA_NONLATIN1; + } else { + nonlatin1_type = 0; + } + + if (field->flags & BINARY_FLAG) { + binary_type = DATA_BINARY_TYPE; + nonlatin1_type = 0; + } else { + binary_type = 0; + } + dict_mem_table_add_col(table, (char*) field->field_name, col_type, (ulint)field->type() - | nulls_allowed | unsigned_type, + | nulls_allowed | unsigned_type + | nonlatin1_type | binary_type, field->pack_length(), 0); } @@ -3011,6 +3223,7 @@ create_index( const char* table_name, /* in: table name */ uint key_num) /* in: index number */ { + Field* field; dict_index_t* index; int error; ulint n_fields; @@ -3020,6 +3233,7 @@ create_index( ulint col_type; ulint prefix_len; ulint i; + ulint j; DBUG_ENTER("create_index"); @@ -3046,31 +3260,63 @@ create_index( for (i = 0; i < n_fields; i++) { key_part = key->key_part + i; - if (key_part->length != key_part->field->pack_length()) { + /* (The flag HA_PART_KEY denotes in MySQL a column prefix + field in an index: we only store a specified number of first + bytes of the column to the index field.) The flag does not + seem to be properly set by MySQL. Let us fall back on testing + the length of the key part versus the column. */ + + field = NULL; + for (j = 0; j < form->fields; j++) { + + field = form->field[j]; + + if (strlen(field->field_name) + == strlen(key_part->field->field_name) + && 0 == ut_cmp_in_lower_case( + (char*)field->field_name, + (char*)key_part->field->field_name, + strlen(field->field_name))) { + /* Found the corresponding column */ + + break; + } + } + + ut_a(j < form->fields); + + col_type = get_innobase_type_from_mysql_type(key_part->field); + + if (DATA_BLOB == col_type + || key_part->length < field->pack_length()) { + prefix_len = key_part->length; - col_type = get_innobase_type_from_mysql_type( - key_part->field); if (col_type == DATA_INT || col_type == DATA_FLOAT || col_type == DATA_DOUBLE || col_type == DATA_DECIMAL) { fprintf(stderr, "InnoDB: error: MySQL is trying to create a column prefix index field\n" -"InnoDB: on an inappropriate data type %lu. Table name %s, column name %s.\n", - col_type, table_name, - key_part->field->field_name); +"InnoDB: on an inappropriate data type. Table name %s, column name %s.\n", + table_name, key_part->field->field_name); prefix_len = 0; } } else { prefix_len = 0; - } + } + + if (prefix_len >= DICT_MAX_COL_PREFIX_LEN) { + DBUG_RETURN(-1); + } /* We assume all fields should be sorted in ascending order, hence the '0': */ + dict_mem_index_add_field(index, - (char*) key_part->field->field_name, 0); + (char*) key_part->field->field_name, + 0, prefix_len); } error = row_create_index_for_mysql(index, trx); @@ -3536,6 +3782,8 @@ ha_innobase::records_in_range( table->reclength + table->max_key_length + 100, MYF(MY_WME)); + ulint buff2_len = table->reclength + + table->max_key_length + 100; dtuple_t* range_start; dtuple_t* range_end; ib_longlong n_rows; @@ -3572,12 +3820,15 @@ ha_innobase::records_in_range( dict_index_copy_types(range_end, index, key->key_parts); row_sel_convert_mysql_key_to_innobase( - range_start, (byte*) key_val_buff, index, + range_start, (byte*) key_val_buff, + (ulint)upd_and_key_val_buff_len, + index, (byte*) start_key, (ulint) start_key_len); row_sel_convert_mysql_key_to_innobase( - range_end, (byte*) key_val_buff2, index, + range_end, (byte*) key_val_buff2, + buff2_len, index, (byte*) end_key, (ulint) end_key_len); @@ -3787,8 +4038,32 @@ ha_innobase::info( } for (i = 0; i < table->keys; i++) { + if (index == NULL) { + ut_print_timestamp(stderr); + fprintf(stderr, +" InnoDB: Error: table %s contains less indexes inside InnoDB\n" +"InnoDB: than are defined in the MySQL .frm file. Have you mixed up\n" +"InnoDB: .frm files from different installations? See section\n" +"InnoDB: 15.1 at http://www.innodb.com/ibman.html\n", + ib_table->name); + break; + } + for (j = 0; j < table->key_info[i].key_parts; j++) { + if (j + 1 > index->n_uniq) { + ut_print_timestamp(stderr); + fprintf(stderr, +" InnoDB: Error: index %s of %s has %lu columns unique inside InnoDB\n" +"InnoDB: but MySQL is asking statistics for %lu columns. Have you mixed up\n" +"InnoDB: .frm files from different installations? See section\n" +"InnoDB: 15.1 at http://www.innodb.com/ibman.html\n", + index->name, + ib_table->name, index->n_uniq, + j + 1); + break; + } + if (index->stat_n_diff_key_vals[j + 1] == 0) { rec_per_key = records; @@ -4046,10 +4321,11 @@ ha_innobase::reset(void) } /********************************************************************** -MySQL calls this function at the start of each SQL statement. Inside LOCK -TABLES the ::external_lock method does not work to mark SQL statement -borders. Note also a special case: if a temporary table is created inside -LOCK TABLES, MySQL has not called external_lock() at all on that table. */ +MySQL calls this function at the start of each SQL statement inside LOCK +TABLES. Inside LOCK TABLES the ::external_lock method does not work to +mark SQL statement borders. Note also a special case: if a temporary table +is created inside LOCK TABLES, MySQL has not called external_lock() at all +on that table. */ int ha_innobase::start_stmt( @@ -4448,9 +4724,9 @@ ha_innobase::innobase_read_and_init_auto_inc( return(0); } - srv_conc_enter_innodb(prebuilt->trx); + innodb_srv_conc_enter_innodb(prebuilt->trx); error = row_lock_table_autoinc_for_mysql(prebuilt); - srv_conc_exit_innodb(prebuilt->trx); + innodb_srv_conc_exit_innodb(prebuilt->trx); if (error != DB_SUCCESS) { error = convert_error_code_to_mysql(error, user_thd); diff --git a/sql/ha_innodb.h b/sql/ha_innodb.h index 1a9b1b16c64..bfa2687b87a 100644 --- a/sql/ha_innodb.h +++ b/sql/ha_innodb.h @@ -52,6 +52,9 @@ class ha_innobase: public handler byte* key_val_buff; /* buffer used in converting search key values from MySQL format to Innodb format */ + ulong upd_and_key_val_buff_len; + /* the length of each of the previous + two buffers */ ulong int_table_flags; uint primary_key; uint last_dup_key; @@ -83,13 +86,15 @@ class ha_innobase: public handler public: ha_innobase(TABLE *table): handler(table), int_table_flags(HA_REC_NOT_IN_SEQ | - HA_KEYPOS_TO_RNDPOS | HA_LASTKEY_ORDER | - HA_NULL_KEY | HA_CAN_SQL_HANDLER | + HA_KEYPOS_TO_RNDPOS | + HA_LASTKEY_ORDER | + HA_NULL_KEY | + HA_BLOB_KEY | + HA_CAN_SQL_HANDLER | HA_NOT_EXACT_COUNT | HA_NO_WRITE_DELAYED | HA_PRIMARY_KEY_IN_READ_INDEX | HA_DROP_BEFORE_CREATE | - HA_NO_PREFIX_CHAR_KEYS | HA_TABLE_SCAN_ON_INDEX), last_dup_key((uint) -1), start_of_scan(0) @@ -217,6 +222,14 @@ int innobase_report_binlog_offset_and_commit( int innobase_commit_complete( void* trx_handle); int innobase_rollback(THD *thd, void* trx_handle); +int innobase_rollback_to_savepoint( + THD* thd, + char* savepoint_name, + my_off_t* binlog_cache_pos); +int innobase_savepoint( + THD* thd, + char* savepoint_name, + my_off_t binlog_cache_pos); int innobase_close_connection(THD *thd); int innobase_drop_database(char *path); int innodb_show_status(THD* thd); diff --git a/sql/handler.cc b/sql/handler.cc index cae1777e958..4ea5bc0e9f5 100644 --- a/sql/handler.cc +++ b/sql/handler.cc @@ -379,7 +379,6 @@ int ha_commit_trans(THD *thd, THD_TRANS* trans) trans->innodb_active_trans=0; if (trans == &thd->transaction.all) operation_done= transaction_commited= 1; - } #endif #ifdef HAVE_QUERY_CACHE @@ -447,6 +446,70 @@ int ha_rollback_trans(THD *thd, THD_TRANS *trans) DBUG_RETURN(error); } + +/* +Rolls the current transaction back to a savepoint. +Return value: 0 if success, 1 if there was not a savepoint of the given +name. +*/ + +int ha_rollback_to_savepoint(THD *thd, char *savepoint_name) +{ + my_off_t binlog_cache_pos=0; + bool operation_done=0; + int error=0; + DBUG_ENTER("ha_rollback_to_savepoint"); +#ifdef USING_TRANSACTIONS + if (opt_using_transactions) + { +#ifdef HAVE_INNOBASE_DB + /* + Retrieve the trans_log binlog cache position corresponding to the + savepoint, and if the rollback is successful inside InnoDB reset the write + position in the binlog cache to what it was at the savepoint. + */ + if ((error=innobase_rollback_to_savepoint(thd, savepoint_name, + &binlog_cache_pos))) + { + my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), error); + error=1; + } + else + reinit_io_cache(&thd->transaction.trans_log, WRITE_CACHE, + binlog_cache_pos, 0, 0); + operation_done=1; +#endif + if (operation_done) + statistic_increment(ha_rollback_count,&LOCK_status); + } +#endif /* USING_TRANSACTIONS */ + + DBUG_RETURN(error); +} + + +/* +Sets a transaction savepoint. +Return value: always 0, that is, succeeds always +*/ + +int ha_savepoint(THD *thd, char *savepoint_name) +{ + my_off_t binlog_cache_pos=0; + int error=0; + DBUG_ENTER("ha_savepoint"); +#ifdef USING_TRANSACTIONS + if (opt_using_transactions) + { + binlog_cache_pos=my_b_tell(&thd->transaction.trans_log); +#ifdef HAVE_INNOBASE_DB + innobase_savepoint(thd,savepoint_name, binlog_cache_pos); +#endif + } +#endif /* USING_TRANSACTIONS */ + DBUG_RETURN(error); +} + bool ha_flush_logs() { bool result=0; diff --git a/sql/handler.h b/sql/handler.h index fbad36bffdd..41f6fdc656a 100644 --- a/sql/handler.h +++ b/sql/handler.h @@ -376,6 +376,8 @@ int ha_commit_complete(THD *thd); int ha_release_temporary_latches(THD *thd); int ha_commit_trans(THD *thd, THD_TRANS *trans); int ha_rollback_trans(THD *thd, THD_TRANS *trans); +int ha_rollback_to_savepoint(THD *thd, char *savepoint_name); +int ha_savepoint(THD *thd, char *savepoint_name); int ha_autocommit_or_rollback(THD *thd, int error); void ha_set_spin_retries(uint retries); bool ha_flush_logs(void); diff --git a/sql/sql_lex.h b/sql/sql_lex.h index 7d931399782..0618f04a79b 100644 --- a/sql/sql_lex.h +++ b/sql/sql_lex.h @@ -53,8 +53,9 @@ enum enum_sql_command { SQLCOM_REPAIR, SQLCOM_REPLACE, SQLCOM_REPLACE_SELECT, SQLCOM_CREATE_FUNCTION, SQLCOM_DROP_FUNCTION, SQLCOM_REVOKE,SQLCOM_OPTIMIZE, SQLCOM_CHECK, - SQLCOM_FLUSH, SQLCOM_KILL, SQLCOM_ANALYZE, - SQLCOM_ROLLBACK, SQLCOM_COMMIT, SQLCOM_SAVEPOINT, + SQLCOM_FLUSH, SQLCOM_KILL, SQLCOM_ANALYZE, + SQLCOM_ROLLBACK, SQLCOM_ROLLBACK_TO_SAVEPOINT, + SQLCOM_COMMIT, SQLCOM_SAVEPOINT, SQLCOM_SLAVE_START, SQLCOM_SLAVE_STOP, SQLCOM_BEGIN, SQLCOM_LOAD_MASTER_TABLE, SQLCOM_CHANGE_MASTER, SQLCOM_RENAME_TABLE, SQLCOM_BACKUP_TABLE, SQLCOM_RESTORE_TABLE, diff --git a/sql/sql_parse.cc b/sql/sql_parse.cc index 3a8895ab120..bb06713ec4c 100644 --- a/sql/sql_parse.cc +++ b/sql/sql_parse.cc @@ -2536,8 +2536,22 @@ mysql_execute_command(void) res= -1; thd->options&= ~(ulong) (OPTION_BEGIN | OPTION_STATUS_NO_TRANS_UPDATE); break; + case SQLCOM_ROLLBACK_TO_SAVEPOINT: + if (!ha_rollback_to_savepoint(thd, lex->savepoint_name)) + { + if (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) + send_warning(&thd->net,ER_WARNING_NOT_COMPLETE_ROLLBACK,0); + else + send_ok(&thd->net); + } + else + res= -1; + break; case SQLCOM_SAVEPOINT: - send_ok(&thd->net); + if (!ha_savepoint(thd, lex->savepoint_name)) + send_ok(&thd->net); + else + res= -1; break; default: /* Impossible */ send_ok(&thd->net); diff --git a/sql/sql_yacc.yy b/sql/sql_yacc.yy index 2ef0992cdf7..4a41ad32dcc 100644 --- a/sql/sql_yacc.yy +++ b/sql/sql_yacc.yy @@ -3922,11 +3922,10 @@ rollback: ROLLBACK_SYM { Lex->sql_command = SQLCOM_ROLLBACK; - Lex->savepoint_name = NULL; } | ROLLBACK_SYM TO_SYM SAVEPOINT_SYM ident { - Lex->sql_command = SQLCOM_ROLLBACK; + Lex->sql_command = SQLCOM_ROLLBACK_TO_SAVEPOINT; Lex->savepoint_name = $4.str; }; savepoint: |