4 ** The author disclaims copyright to this source code. In place of
5 ** a legal notice, here is a blessing:
7 ** May you do good and not evil.
8 ** May you find forgiveness for yourself and forgive others.
9 ** May you share freely, never taking more than you give.
11 /* TODO(shess): THIS MODULE IS STILL EXPERIMENTAL. DO NOT USE IT. */
12 /* Implements a virtual table "recover" which can be used to recover
13 * data from a corrupt table. The table is walked manually, with
14 * corrupt items skipped. Additionally, any errors while reading will
17 * Given a table with this definition:
19 * CREATE TABLE Stuff (
20 * name TEXT PRIMARY KEY,
24 * to recover the data from teh table, you could do something like:
26 * -- Attach another database, the original is not trustworthy.
27 * ATTACH DATABASE '/tmp/db.db' AS rdb;
28 * -- Create a new version of the table.
29 * CREATE TABLE rdb.Stuff (
30 * name TEXT PRIMARY KEY,
33 * -- This will read the original table's data.
34 * CREATE VIRTUAL TABLE temp.recover_Stuff using recover(
36 * name TEXT STRICT NOT NULL, -- only real TEXT data allowed
37 * value TEXT STRICT NOT NULL
39 * -- Corruption means the UNIQUE constraint may no longer hold for
40 * -- Stuff, so either OR REPLACE or OR IGNORE must be used.
41 * INSERT OR REPLACE INTO rdb.Stuff (rowid, name, value )
42 * SELECT rowid, name, value FROM temp.recover_Stuff;
43 * DROP TABLE temp.recover_Stuff;
44 * DETACH DATABASE rdb;
45 * -- Move db.db to replace original db in filesystem.
50 * Given the goal of dealing with corruption, it would not be safe to
51 * create a recovery table in the database being recovered. So
52 * recovery tables must be created in the temp database. They are not
53 * appropriate to persist, in any case. [As a bonus, sqlite_master
54 * tables can be recovered. Perhaps more cute than useful, though.]
56 * The parameters are a specifier for the table to read, and a column
57 * definition for each bit of data stored in that table. The named
58 * table must be convertable to a root page number by reading the
59 * sqlite_master table. Bare table names are assumed to be in
60 * database 0 ("main"), other databases can be specified in db.table
63 * Column definitions are similar to BUT NOT THE SAME AS those
64 * provided to CREATE statements:
65 * column-def: column-name [type-name [STRICT] [NOT NULL]]
66 * type-name: (ANY|ROWID|INTEGER|FLOAT|NUMERIC|TEXT|BLOB)
68 * Only those exact type names are accepted, there is no type
69 * intuition. The only constraints accepted are STRICT (see below)
70 * and NOT NULL. Anything unexpected will cause the create to fail.
72 * ANY is a convenience to indicate that manifest typing is desired.
73 * It is equivalent to not specifying a type at all. The results for
74 * such columns will have the type of the data's storage. The exposed
75 * schema will contain no type for that column.
77 * ROWID is used for columns representing aliases to the rowid
78 * (INTEGER PRIMARY KEY, with or without AUTOINCREMENT), to make the
79 * concept explicit. Such columns are actually stored as NULL, so
80 * they cannot be simply ignored. The exposed schema will be INTEGER
83 * NOT NULL causes rows with a NULL in that column to be skipped. It
84 * also adds NOT NULL to the column in the exposed schema. If the
85 * table has ever had columns added using ALTER TABLE, then those
86 * columns implicitly contain NULL for rows which have not been
87 * updated. [Workaround using COALESCE() in your SELECT statement.]
89 * The created table is read-only, with no indices. Any SELECT will
90 * be a full-table scan, returning each valid row read from the
91 * storage of the backing table. The rowid will be the rowid of the
92 * row from the backing table. "Valid" means:
93 * - The cell metadata for the row is well-formed. Mainly this means that
94 * the cell header info describes a payload of the size indicated by
95 * the cell's payload size.
96 * - The cell does not run off the page.
97 * - The cell does not overlap any other cell on the page.
98 * - The cell contains doesn't contain too many columns.
99 * - The types of the serialized data match the indicated types (see below).
102 * Type affinity versus type storage.
104 * http://www.sqlite.org/datatype3.html describes SQLite's type
105 * affinity system. The system provides for automated coercion of
106 * types in certain cases, transparently enough that many developers
107 * do not realize that it is happening. Importantly, it implies that
108 * the raw data stored in the database may not have the obvious type.
110 * Differences between the stored data types and the expected data
111 * types may be a signal of corruption. This module makes some
112 * allowances for automatic coercion. It is important to be concious
113 * of the difference between the schema exposed by the module, and the
114 * data types read from storage. The following table describes how
115 * the module interprets things:
117 * type schema data STRICT
118 * ---- ------ ---- ------
120 * ROWID INTEGER n/a n/a
121 * INTEGER INTEGER integer integer
122 * FLOAT FLOAT integer or float float
123 * NUMERIC NUMERIC integer, float, or text integer or float
124 * TEXT TEXT text or blob text
125 * BLOB BLOB blob blob
127 * type is the type provided to the recover module, schema is the
128 * schema exposed by the module, data is the acceptable types of data
129 * decoded from storage, and STRICT is a modification of that.
131 * A very loose recovery system might use ANY for all columns, then
132 * use the appropriate sqlite3_column_*() calls to coerce to expected
133 * types. This doesn't provide much protection if a page from a
134 * different table with the same column count is linked into an
135 * inappropriate btree.
137 * A very tight recovery system might use STRICT to enforce typing on
138 * all columns, preferring to skip rows which are valid at the storage
139 * level but don't contain the right types. Note that FLOAT STRICT is
140 * almost certainly not appropriate, since integral values are
141 * transparently stored as integers, when that is more efficient.
143 * Another option is to use ANY for all columns and inspect each
144 * result manually (using sqlite3_column_*). This should only be
145 * necessary in cases where developers have used manifest typing (test
146 * to make sure before you decide that you aren't using manifest
152 * Leaf pages not referenced by interior nodes will not be found.
154 * Leaf pages referenced from interior nodes of other tables will not
157 * Rows referencing invalid overflow pages will be skipped.
159 * SQlite rows have a header which describes how to interpret the rest
160 * of the payload. The header can be valid in cases where the rest of
161 * the record is actually corrupt (in the sense that the data is not
162 * the intended data). This can especially happen WRT overflow pages,
163 * as lack of atomic updates between pages is the primary form of
164 * corruption I have seen in the wild.
166 /* The implementation is via a series of cursors. The cursor
167 * implementations follow the pattern:
169 * // Creates the cursor using various initialization info.
170 * int cursorCreate(...);
172 * // Returns 1 if there is no more data, 0 otherwise.
173 * int cursorEOF(Cursor *pCursor);
175 * // Various accessors can be used if not at EOF.
177 * // Move to the next item.
178 * int cursorNext(Cursor *pCursor);
180 * // Destroy the memory associated with the cursor.
181 * void cursorDestroy(Cursor *pCursor);
183 * References in the following are to sections at
184 * http://www.sqlite.org/fileformat2.html .
186 * RecoverLeafCursor iterates the records in a leaf table node
187 * described in section 1.5 "B-tree Pages". When the node is
188 * exhausted, an interior cursor is used to get the next leaf node,
189 * and iteration continues there.
191 * RecoverInteriorCursor iterates the child pages in an interior table
192 * node described in section 1.5 "B-tree Pages". When the node is
193 * exhausted, a parent interior cursor is used to get the next
194 * interior node at the same level, and iteration continues there.
196 * Together these record the path from the leaf level to the root of
197 * the tree. Iteration happens from the leaves rather than the root
198 * both for efficiency and putting the special case at the front of
199 * the list is easier to implement.
201 * RecoverCursor uses a RecoverLeafCursor to iterate the rows of a
202 * table, returning results via the SQLite virtual table interface.
204 /* TODO(shess): It might be useful to allow DEFAULT in types to
205 * specify what to do for NULL when an ALTER TABLE case comes up.
206 * Unfortunately, simply adding it to the exposed schema and using
207 * sqlite3_result_null() does not cause the default to be generate.
208 * Handling it ourselves seems hard, unfortunately.
216 /* Internal SQLite things that are used:
217 * u32, u64, i64 types.
218 * Btree, Pager, and DbPage structs.
219 * DbPage.pData, .pPager, and .pgno
221 * sqlite3BtreePager() and sqlite3BtreeGetPageSize()
222 * sqlite3PagerAcquire() and sqlite3PagerUnref()
225 #include "sqliteInt.h"
229 #define FNENTRY() fprintf(stderr, "In %s\n", __FUNCTION__)
234 /* Generic constants and helper functions. */
236 static const unsigned char kTableLeafPage = 0x0D;
237 static const unsigned char kTableInteriorPage = 0x05;
239 /* From section 1.5. */
240 static const unsigned kiPageTypeOffset = 0;
241 static const unsigned kiPageFreeBlockOffset = 1;
242 static const unsigned kiPageCellCountOffset = 3;
243 static const unsigned kiPageCellContentOffset = 5;
244 static const unsigned kiPageFragmentedBytesOffset = 7;
245 static const unsigned knPageLeafHeaderBytes = 8;
246 /* Interior pages contain an additional field. */
247 static const unsigned kiPageRightChildOffset = 8;
248 static const unsigned kiPageInteriorHeaderBytes = 12;
250 /* Accepted types are specified by a mask. */
251 #define MASK_ROWID (1<<0)
252 #define MASK_INTEGER (1<<1)
253 #define MASK_FLOAT (1<<2)
254 #define MASK_TEXT (1<<3)
255 #define MASK_BLOB (1<<4)
256 #define MASK_NULL (1<<5)
258 /* Helpers to decode fixed-size fields. */
259 static u32 decodeUnsigned16(const unsigned char *pData){
260 return (pData[0]<<8) + pData[1];
262 static u32 decodeUnsigned32(const unsigned char *pData){
263 return (decodeUnsigned16(pData)<<16) + decodeUnsigned16(pData+2);
265 static i64 decodeSigned(const unsigned char *pData, unsigned nBytes){
266 i64 r = (char)(*pData);
273 /* Derived from vdbeaux.c, sqlite3VdbeSerialGet(), case 7. */
274 /* TODO(shess): Determine if swapMixedEndianFloat() applies. */
275 static double decodeFloat64(const unsigned char *pData){
277 static const u64 t1 = ((u64)0x3ff00000)<<32;
278 static const double r1 = 1.0;
280 assert( sizeof(r1)==sizeof(t2) && memcmp(&r1, &t2, sizeof(r1))==0 );
282 i64 x = decodeSigned(pData, 8);
284 memcpy(&d, &x, sizeof(x));
288 /* Return true if a varint can safely be read from pData/nData. */
289 /* TODO(shess): DbPage points into the middle of a buffer which
290 * contains the page data before DbPage. So code should always be
291 * able to read a small number of varints safely. Consider whether to
294 static int checkVarint(const unsigned char *pData, unsigned nData){
297 /* In the worst case the decoder takes all 8 bits of the 9th byte. */
302 /* Look for a high-bit-clear byte in what's left. */
303 for( i=0; i<nData; ++i ){
304 if( !(pData[i]&0x80) ){
309 /* Cannot decode in the space given. */
313 /* Return 1 if n varints can be read from pData/nData. */
314 static int checkVarints(const unsigned char *pData, unsigned nData,
316 unsigned nCur = 0; /* Byte offset within current varint. */
317 unsigned nFound = 0; /* Number of varints found. */
320 /* In the worst case the decoder takes all 8 bits of the 9th byte. */
325 for( i=0; nFound<n && i<nData; ++i ){
327 if( nCur==9 || !(pData[i]&0x80) ){
336 /* ctype and str[n]casecmp() can be affected by locale (eg, tr_TR).
337 * These versions consider only the ASCII space.
339 /* TODO(shess): It may be reasonable to just remove the need for these
340 * entirely. The module could require "TEXT STRICT NOT NULL", not
341 * "Text Strict Not Null" or whatever the developer felt like typing
342 * that day. Handling corrupt data is a PERFECT place to be pedantic.
344 static int ascii_isspace(char c){
345 /* From fts3_expr.c */
346 return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f';
348 static int ascii_isalnum(int x){
349 /* From fts3_tokenizer1.c */
350 return (x>='0' && x<='9') || (x>='A' && x<='Z') || (x>='a' && x<='z');
352 static int ascii_tolower(int x){
353 /* From fts3_tokenizer1.c */
354 return (x>='A' && x<='Z') ? x-'A'+'a' : x;
356 /* TODO(shess): Consider sqlite3_strnicmp() */
357 static int ascii_strncasecmp(const char *s1, const char *s2, size_t n){
358 const unsigned char *us1 = (const unsigned char *)s1;
359 const unsigned char *us2 = (const unsigned char *)s2;
360 while( *us1 && *us2 && n && ascii_tolower(*us1)==ascii_tolower(*us2) ){
363 return n ? ascii_tolower(*us1)-ascii_tolower(*us2) : 0;
365 static int ascii_strcasecmp(const char *s1, const char *s2){
366 /* If s2 is equal through strlen(s1), will exit while() due to s1's
367 * trailing NUL, and return NUL-s2[strlen(s1)].
369 return ascii_strncasecmp(s1, s2, strlen(s1)+1);
372 /* For some reason I kept making mistakes with offset calculations. */
373 static const unsigned char *PageData(DbPage *pPage, unsigned iOffset){
374 assert( iOffset<=pPage->nPageSize );
375 return (unsigned char *)pPage->pData + iOffset;
378 /* The first page in the file contains a file header in the first 100
379 * bytes. The page's header information comes after that. Note that
380 * the offsets in the page's header information are relative to the
381 * beginning of the page, NOT the end of the page header.
383 static const unsigned char *PageHeader(DbPage *pPage){
384 if( pPage->pgno==1 ){
385 const unsigned nDatabaseHeader = 100;
386 return PageData(pPage, nDatabaseHeader);
388 return PageData(pPage, 0);
392 /* Helper to fetch the pager and page size for the named database. */
393 static int GetPager(sqlite3 *db, const char *zName,
394 Pager **pPager, unsigned *pnPageSize){
397 for( i=0; i<db->nDb; ++i ){
398 if( ascii_strcasecmp(db->aDb[i].zName, zName)==0 ){
399 pBt = db->aDb[i].pBt;
407 *pPager = sqlite3BtreePager(pBt);
408 *pnPageSize = sqlite3BtreeGetPageSize(pBt) - sqlite3BtreeGetReserve(pBt);
412 /* iSerialType is a type read from a record header. See "2.1 Record Format".
415 /* Storage size of iSerialType in bytes. My interpretation of SQLite
416 * documentation is that text and blob fields can have 32-bit length.
417 * Values past 2^31-12 will need more than 32 bits to encode, which is
418 * why iSerialType is u64.
420 static u32 SerialTypeLength(u64 iSerialType){
421 switch( iSerialType ){
422 case 0 : return 0; /* NULL */
423 case 1 : return 1; /* Various integers. */
429 case 7 : return 8; /* 64-bit float. */
430 case 8 : return 0; /* Constant 0. */
431 case 9 : return 0; /* Constant 1. */
432 case 10 : case 11 : assert( !"RESERVED TYPE"); return 0;
434 return (u32)((iSerialType>>1) - 6);
437 /* True if iSerialType refers to a blob. */
438 static int SerialTypeIsBlob(u64 iSerialType){
439 assert( iSerialType>=12 );
440 return (iSerialType%2)==0;
443 /* Returns true if the serialized type represented by iSerialType is
444 * compatible with the given type mask.
446 static int SerialTypeIsCompatible(u64 iSerialType, unsigned char mask){
447 switch( iSerialType ){
448 case 0 : return (mask&MASK_NULL)!=0;
449 case 1 : return (mask&MASK_INTEGER)!=0;
450 case 2 : return (mask&MASK_INTEGER)!=0;
451 case 3 : return (mask&MASK_INTEGER)!=0;
452 case 4 : return (mask&MASK_INTEGER)!=0;
453 case 5 : return (mask&MASK_INTEGER)!=0;
454 case 6 : return (mask&MASK_INTEGER)!=0;
455 case 7 : return (mask&MASK_FLOAT)!=0;
456 case 8 : return (mask&MASK_INTEGER)!=0;
457 case 9 : return (mask&MASK_INTEGER)!=0;
458 case 10 : assert( !"RESERVED TYPE"); return 0;
459 case 11 : assert( !"RESERVED TYPE"); return 0;
461 return (mask&(SerialTypeIsBlob(iSerialType) ? MASK_BLOB : MASK_TEXT));
464 /* Versions of strdup() with return values appropriate for
465 * sqlite3_free(). malloc.c has sqlite3DbStrDup()/NDup(), but those
466 * need sqlite3DbFree(), which seems intrusive.
468 static char *sqlite3_strndup(const char *z, unsigned n){
475 zNew = sqlite3_malloc(n+1);
482 static char *sqlite3_strdup(const char *z){
486 return sqlite3_strndup(z, strlen(z));
489 /* Fetch the page number of zTable in zDb from sqlite_master in zDb,
490 * and put it in *piRootPage.
492 static int getRootPage(sqlite3 *db, const char *zDb, const char *zTable,
494 char *zSql; /* SQL selecting root page of named element. */
498 if( strcmp(zTable, "sqlite_master")==0 ){
503 zSql = sqlite3_mprintf("SELECT rootpage FROM %s.sqlite_master "
504 "WHERE type = 'table' AND tbl_name = %Q",
510 rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0);
516 /* Require a result. */
517 rc = sqlite3_step(pStmt);
518 if( rc==SQLITE_DONE ){
520 }else if( rc==SQLITE_ROW ){
521 *piRootPage = sqlite3_column_int(pStmt, 0);
523 /* Require only one result. */
524 rc = sqlite3_step(pStmt);
525 if( rc==SQLITE_DONE ){
527 }else if( rc==SQLITE_ROW ){
531 sqlite3_finalize(pStmt);
535 static int getEncoding(sqlite3 *db, const char *zDb, int* piEncoding){
538 char *zSql = sqlite3_mprintf("PRAGMA %s.encoding", zDb);
543 rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0);
549 /* Require a result. */
550 rc = sqlite3_step(pStmt);
551 if( rc==SQLITE_DONE ){
552 /* This case should not be possible. */
554 }else if( rc==SQLITE_ROW ){
555 if( sqlite3_column_type(pStmt, 0)==SQLITE_TEXT ){
556 const char* z = (const char *)sqlite3_column_text(pStmt, 0);
557 /* These strings match the literals in pragma.c. */
558 if( !strcmp(z, "UTF-16le") ){
559 *piEncoding = SQLITE_UTF16LE;
560 }else if( !strcmp(z, "UTF-16be") ){
561 *piEncoding = SQLITE_UTF16BE;
562 }else if( !strcmp(z, "UTF-8") ){
563 *piEncoding = SQLITE_UTF8;
565 /* This case should not be possible. */
566 *piEncoding = SQLITE_UTF8;
569 /* This case should not be possible. */
570 *piEncoding = SQLITE_UTF8;
573 /* Require only one result. */
574 rc = sqlite3_step(pStmt);
575 if( rc==SQLITE_DONE ){
577 }else if( rc==SQLITE_ROW ){
578 /* This case should not be possible. */
582 sqlite3_finalize(pStmt);
586 /* Cursor for iterating interior nodes. Interior page cells contain a
587 * child page number and a rowid. The child page contains items left
588 * of the rowid (less than). The rightmost page of the subtree is
589 * stored in the page header.
591 * interiorCursorDestroy - release all resources associated with the
592 * cursor and any parent cursors.
593 * interiorCursorCreate - create a cursor with the given parent and page.
594 * interiorCursorEOF - returns true if neither the cursor nor the
595 * parent cursors can return any more data.
596 * interiorCursorNextPage - fetch the next child page from the cursor.
598 * Logically, interiorCursorNextPage() returns the next child page
599 * number from the page the cursor is currently reading, calling the
600 * parent cursor as necessary to get new pages to read, until done.
601 * SQLITE_ROW if a page is returned, SQLITE_DONE if out of pages,
602 * error otherwise. Unfortunately, if the table is corrupted
603 * unexpected pages can be returned. If any unexpected page is found,
604 * leaf or otherwise, it is returned to the caller for processing,
605 * with the interior cursor left empty. The next call to
606 * interiorCursorNextPage() will recurse to the parent cursor until an
607 * interior page to iterate is returned.
609 * Note that while interiorCursorNextPage() will refuse to follow
610 * loops, it does not keep track of pages returned for purposes of
611 * preventing duplication.
613 * Note that interiorCursorEOF() could return false (not at EOF), and
614 * interiorCursorNextPage() could still return SQLITE_DONE. This
615 * could happen if there are more cells to iterate in an interior
616 * page, but those cells refer to invalid pages.
618 typedef struct RecoverInteriorCursor RecoverInteriorCursor;
619 struct RecoverInteriorCursor {
620 RecoverInteriorCursor *pParent; /* Parent node to this node. */
621 DbPage *pPage; /* Reference to leaf page. */
622 unsigned nPageSize; /* Size of page. */
623 unsigned nChildren; /* Number of children on the page. */
624 unsigned iChild; /* Index of next child to return. */
627 static void interiorCursorDestroy(RecoverInteriorCursor *pCursor){
628 /* Destroy all the cursors to the root. */
630 RecoverInteriorCursor *p = pCursor;
631 pCursor = pCursor->pParent;
634 sqlite3PagerUnref(p->pPage);
638 memset(p, 0xA5, sizeof(*p));
643 /* Internal helper. Reset storage in preparation for iterating pPage. */
644 static void interiorCursorSetPage(RecoverInteriorCursor *pCursor,
646 assert( PageHeader(pPage)[kiPageTypeOffset]==kTableInteriorPage );
648 if( pCursor->pPage ){
649 sqlite3PagerUnref(pCursor->pPage);
650 pCursor->pPage = NULL;
652 pCursor->pPage = pPage;
655 /* A child for each cell, plus one in the header. */
656 /* TODO(shess): Sanity-check the count? Page header plus per-cell
657 * cost of 16-bit offset, 32-bit page number, and one varint
660 pCursor->nChildren = decodeUnsigned16(PageHeader(pPage) +
661 kiPageCellCountOffset) + 1;
664 static int interiorCursorCreate(RecoverInteriorCursor *pParent,
665 DbPage *pPage, int nPageSize,
666 RecoverInteriorCursor **ppCursor){
667 RecoverInteriorCursor *pCursor =
668 sqlite3_malloc(sizeof(RecoverInteriorCursor));
673 memset(pCursor, 0, sizeof(*pCursor));
674 pCursor->pParent = pParent;
675 pCursor->nPageSize = nPageSize;
676 interiorCursorSetPage(pCursor, pPage);
681 /* Internal helper. Return the child page number at iChild. */
682 static unsigned interiorCursorChildPage(RecoverInteriorCursor *pCursor){
683 const unsigned char *pPageHeader; /* Header of the current page. */
684 const unsigned char *pCellOffsets; /* Offset to page's cell offsets. */
685 unsigned iCellOffset; /* Offset of target cell. */
687 assert( pCursor->iChild<pCursor->nChildren );
689 /* Rightmost child is in the header. */
690 pPageHeader = PageHeader(pCursor->pPage);
691 if( pCursor->iChild==pCursor->nChildren-1 ){
692 return decodeUnsigned32(pPageHeader + kiPageRightChildOffset);
695 /* Each cell is a 4-byte integer page number and a varint rowid
696 * which is greater than the rowid of items in that sub-tree (this
697 * module ignores ordering). The offset is from the beginning of the
698 * page, not from the page header.
700 pCellOffsets = pPageHeader + kiPageInteriorHeaderBytes;
701 iCellOffset = decodeUnsigned16(pCellOffsets + pCursor->iChild*2);
702 if( iCellOffset<=pCursor->nPageSize-4 ){
703 return decodeUnsigned32(PageData(pCursor->pPage, iCellOffset));
706 /* TODO(shess): Check for cell overlaps? Cells require 4 bytes plus
707 * a varint. Check could be identical to leaf check (or even a
708 * shared helper testing for "Cells starting in this range"?).
711 /* If the offset is broken, return an invalid page number. */
715 static int interiorCursorEOF(RecoverInteriorCursor *pCursor){
716 /* Find a parent with remaining children. EOF if none found. */
717 while( pCursor && pCursor->iChild>=pCursor->nChildren ){
718 pCursor = pCursor->pParent;
720 return pCursor==NULL;
723 /* Internal helper. Used to detect if iPage would cause a loop. */
724 static int interiorCursorPageInUse(RecoverInteriorCursor *pCursor,
726 /* Find any parent using the indicated page. */
727 while( pCursor && pCursor->pPage->pgno!=iPage ){
728 pCursor = pCursor->pParent;
730 return pCursor!=NULL;
733 /* Get the next page from the interior cursor at *ppCursor. Returns
734 * SQLITE_ROW with the page in *ppPage, or SQLITE_DONE if out of
735 * pages, or the error SQLite returned.
737 * If the tree is uneven, then when the cursor attempts to get a new
738 * interior page from the parent cursor, it may get a non-interior
739 * page. In that case, the new page is returned, and *ppCursor is
740 * updated to point to the parent cursor (this cursor is freed).
742 /* TODO(shess): I've tried to avoid recursion in most of this code,
743 * but this case is more challenging because the recursive call is in
744 * the middle of operation. One option for converting it without
745 * adding memory management would be to retain the head pointer and
746 * use a helper to "back up" as needed. Another option would be to
747 * reverse the list during traversal.
749 static int interiorCursorNextPage(RecoverInteriorCursor **ppCursor,
751 RecoverInteriorCursor *pCursor = *ppCursor;
754 const unsigned char *pPageHeader; /* Header of found page. */
756 /* Find a valid child page which isn't on the stack. */
757 while( pCursor->iChild<pCursor->nChildren ){
758 const unsigned iPage = interiorCursorChildPage(pCursor);
760 if( interiorCursorPageInUse(pCursor, iPage) ){
761 fprintf(stderr, "Loop detected at %d\n", iPage);
763 int rc = sqlite3PagerAcquire(pCursor->pPage->pPager, iPage, ppPage, 0);
770 /* This page has no more children. Get next page from parent. */
771 if( !pCursor->pParent ){
774 rc = interiorCursorNextPage(&pCursor->pParent, ppPage);
775 if( rc!=SQLITE_ROW ){
779 /* If a non-interior page is received, that either means that the
780 * tree is uneven, or that a child was re-used (say as an overflow
781 * page). Remove this cursor and let the caller handle the page.
783 pPageHeader = PageHeader(*ppPage);
784 if( pPageHeader[kiPageTypeOffset]!=kTableInteriorPage ){
785 *ppCursor = pCursor->pParent;
786 pCursor->pParent = NULL;
787 interiorCursorDestroy(pCursor);
791 /* Iterate the new page. */
792 interiorCursorSetPage(pCursor, *ppPage);
796 assert(NULL); /* NOTREACHED() */
797 return SQLITE_CORRUPT;
800 /* Large rows are spilled to overflow pages. The row's main page
801 * stores the overflow page number after the local payload, with a
802 * linked list forward from there as necessary. overflowMaybeCreate()
803 * and overflowGetSegment() provide an abstraction for accessing such
804 * data while centralizing the code.
806 * overflowDestroy - releases all resources associated with the structure.
807 * overflowMaybeCreate - create the overflow structure if it is needed
808 * to represent the given record. See function comment.
809 * overflowGetSegment - fetch a segment from the record, accounting
810 * for overflow pages. Segments which are not
811 * entirely contained with a page are constructed
812 * into a buffer which is returned. See function comment.
814 typedef struct RecoverOverflow RecoverOverflow;
815 struct RecoverOverflow {
816 RecoverOverflow *pNextOverflow;
821 static void overflowDestroy(RecoverOverflow *pOverflow){
823 RecoverOverflow *p = pOverflow;
824 pOverflow = p->pNextOverflow;
827 sqlite3PagerUnref(p->pPage);
831 memset(p, 0xA5, sizeof(*p));
836 /* Internal helper. Used to detect if iPage would cause a loop. */
837 static int overflowPageInUse(RecoverOverflow *pOverflow, unsigned iPage){
838 while( pOverflow && pOverflow->pPage->pgno!=iPage ){
839 pOverflow = pOverflow->pNextOverflow;
841 return pOverflow!=NULL;
844 /* Setup to access an nRecordBytes record beginning at iRecordOffset
845 * in pPage. If nRecordBytes can be satisfied entirely from pPage,
846 * then no overflow pages are needed an *pnLocalRecordBytes is set to
847 * nRecordBytes. Otherwise, *ppOverflow is set to the head of a list
848 * of overflow pages, and *pnLocalRecordBytes is set to the number of
849 * bytes local to pPage.
851 * overflowGetSegment() will do the right thing regardless of whether
852 * those values are set to be in-page or not.
854 static int overflowMaybeCreate(DbPage *pPage, unsigned nPageSize,
855 unsigned iRecordOffset, unsigned nRecordBytes,
856 unsigned *pnLocalRecordBytes,
857 RecoverOverflow **ppOverflow){
858 unsigned nLocalRecordBytes; /* Record bytes in the leaf page. */
859 unsigned iNextPage; /* Next page number for record data. */
860 unsigned nBytes; /* Maximum record bytes as of current page. */
862 RecoverOverflow *pFirstOverflow; /* First in linked list of pages. */
863 RecoverOverflow *pLastOverflow; /* End of linked list. */
865 /* Calculations from the "Table B-Tree Leaf Cell" part of section
866 * 1.5 of http://www.sqlite.org/fileformat2.html . maxLocal and
867 * minLocal to match naming in btree.c.
869 const unsigned maxLocal = nPageSize - 35;
870 const unsigned minLocal = ((nPageSize-12)*32/255)-23; /* m */
872 /* Always fit anything smaller than maxLocal. */
873 if( nRecordBytes<=maxLocal ){
874 *pnLocalRecordBytes = nRecordBytes;
879 /* Calculate the remainder after accounting for minLocal on the leaf
880 * page and what packs evenly into overflow pages. If the remainder
881 * does not fit into maxLocal, then a partially-full overflow page
882 * will be required in any case, so store as little as possible locally.
884 nLocalRecordBytes = minLocal+((nRecordBytes-minLocal)%(nPageSize-4));
885 if( maxLocal<nLocalRecordBytes ){
886 nLocalRecordBytes = minLocal;
889 /* Don't read off the end of the page. */
890 if( iRecordOffset+nLocalRecordBytes+4>nPageSize ){
891 return SQLITE_CORRUPT;
894 /* First overflow page number is after the local bytes. */
896 decodeUnsigned32(PageData(pPage, iRecordOffset + nLocalRecordBytes));
897 nBytes = nLocalRecordBytes;
899 /* While there are more pages to read, and more bytes are needed,
902 pFirstOverflow = pLastOverflow = NULL;
904 while( iNextPage && nBytes<nRecordBytes ){
905 RecoverOverflow *pOverflow; /* New overflow page for the list. */
907 rc = sqlite3PagerAcquire(pPage->pPager, iNextPage, &pPage, 0);
912 pOverflow = sqlite3_malloc(sizeof(RecoverOverflow));
914 sqlite3PagerUnref(pPage);
918 memset(pOverflow, 0, sizeof(*pOverflow));
919 pOverflow->pPage = pPage;
920 pOverflow->nPageSize = nPageSize;
922 if( !pFirstOverflow ){
923 pFirstOverflow = pOverflow;
925 pLastOverflow->pNextOverflow = pOverflow;
927 pLastOverflow = pOverflow;
929 iNextPage = decodeUnsigned32(pPage->pData);
930 nBytes += nPageSize-4;
933 if( overflowPageInUse(pFirstOverflow, iNextPage) ){
934 fprintf(stderr, "Overflow loop detected at %d\n", iNextPage);
940 /* If there were not enough pages, or too many, things are corrupt.
941 * Not having enough pages is an obvious problem, all the data
942 * cannot be read. Too many pages means that the contents of the
943 * row between the main page and the overflow page(s) is
944 * inconsistent (most likely one or more of the overflow pages does
945 * not really belong to this row).
947 if( rc==SQLITE_OK && (nBytes<nRecordBytes || iNextPage) ){
952 *ppOverflow = pFirstOverflow;
953 *pnLocalRecordBytes = nLocalRecordBytes;
954 }else if( pFirstOverflow ){
955 overflowDestroy(pFirstOverflow);
960 /* Use in concert with overflowMaybeCreate() to efficiently read parts
961 * of a potentially-overflowing record. pPage and iRecordOffset are
962 * the values passed into overflowMaybeCreate(), nLocalRecordBytes and
963 * pOverflow are the values returned by that call.
965 * On SQLITE_OK, *ppBase points to nRequestBytes of data at
966 * iRequestOffset within the record. If the data exists contiguously
967 * in a page, a direct pointer is returned, otherwise a buffer from
968 * sqlite3_malloc() is returned with the data. *pbFree is set true if
969 * sqlite3_free() should be called on *ppBase.
971 /* Operation of this function is subtle. At any time, pPage is the
972 * current page, with iRecordOffset and nLocalRecordBytes being record
973 * data within pPage, and pOverflow being the overflow page after
974 * pPage. This allows the code to handle both the initial leaf page
975 * and overflow pages consistently by adjusting the values
978 static int overflowGetSegment(DbPage *pPage, unsigned iRecordOffset,
979 unsigned nLocalRecordBytes,
980 RecoverOverflow *pOverflow,
981 unsigned iRequestOffset, unsigned nRequestBytes,
982 unsigned char **ppBase, int *pbFree){
983 unsigned nBase; /* Amount of data currently collected. */
984 unsigned char *pBase; /* Buffer to collect record data into. */
986 /* Skip to the page containing the start of the data. */
987 while( iRequestOffset>=nLocalRecordBytes && pOverflow ){
988 /* Factor out current page's contribution. */
989 iRequestOffset -= nLocalRecordBytes;
991 /* Move forward to the next page in the list. */
992 pPage = pOverflow->pPage;
994 nLocalRecordBytes = pOverflow->nPageSize - iRecordOffset;
995 pOverflow = pOverflow->pNextOverflow;
998 /* If the requested data is entirely within this page, return a
999 * pointer into the page.
1001 if( iRequestOffset+nRequestBytes<=nLocalRecordBytes ){
1002 /* TODO(shess): "assignment discards qualifiers from pointer target type"
1003 * Having ppBase be const makes sense, but sqlite3_free() takes non-const.
1005 *ppBase = (unsigned char *)PageData(pPage, iRecordOffset + iRequestOffset);
1010 /* The data range would require additional pages. */
1012 /* Should never happen, the range is outside the nRecordBytes
1013 * passed to overflowMaybeCreate().
1015 assert(NULL); /* NOTREACHED */
1016 return SQLITE_ERROR;
1019 /* Get a buffer to construct into. */
1021 pBase = sqlite3_malloc(nRequestBytes);
1023 return SQLITE_NOMEM;
1025 while( nBase<nRequestBytes ){
1026 /* Copy over data present on this page. */
1027 unsigned nCopyBytes = nRequestBytes - nBase;
1028 if( nLocalRecordBytes-iRequestOffset<nCopyBytes ){
1029 nCopyBytes = nLocalRecordBytes - iRequestOffset;
1031 memcpy(pBase + nBase, PageData(pPage, iRecordOffset + iRequestOffset),
1033 nBase += nCopyBytes;
1036 /* Copy from start of record data in future pages. */
1039 /* Move forward to the next page in the list. Should match
1040 * first while() loop.
1042 pPage = pOverflow->pPage;
1044 nLocalRecordBytes = pOverflow->nPageSize - iRecordOffset;
1045 pOverflow = pOverflow->pNextOverflow;
1046 }else if( nBase<nRequestBytes ){
1047 /* Ran out of overflow pages with data left to deliver. Not
1048 * possible if the requested range fits within nRecordBytes
1049 * passed to overflowMaybeCreate() when creating pOverflow.
1051 assert(NULL); /* NOTREACHED */
1052 sqlite3_free(pBase);
1053 return SQLITE_ERROR;
1056 assert( nBase==nRequestBytes );
1062 /* Primary structure for iterating the contents of a table.
1064 * leafCursorDestroy - release all resources associated with the cursor.
1065 * leafCursorCreate - create a cursor to iterate items from tree at
1066 * the provided root page.
1067 * leafCursorNextValidCell - get the cursor ready to access data from
1068 * the next valid cell in the table.
1069 * leafCursorCellRowid - get the current cell's rowid.
1070 * leafCursorCellColumns - get current cell's column count.
1071 * leafCursorCellColInfo - get type and data for a column in current cell.
1073 * leafCursorNextValidCell skips cells which fail simple integrity
1074 * checks, such as overlapping other cells, or being located at
1075 * impossible offsets, or where header data doesn't correctly describe
1076 * payload data. Returns SQLITE_ROW if a valid cell is found,
1077 * SQLITE_DONE if all pages in the tree were exhausted.
1079 * leafCursorCellColInfo() accounts for overflow pages in the style of
1080 * overflowGetSegment().
1082 typedef struct RecoverLeafCursor RecoverLeafCursor;
1083 struct RecoverLeafCursor {
1084 RecoverInteriorCursor *pParent; /* Parent node to this node. */
1085 DbPage *pPage; /* Reference to leaf page. */
1086 unsigned nPageSize; /* Size of pPage. */
1087 unsigned nCells; /* Number of cells in pPage. */
1088 unsigned iCell; /* Current cell. */
1090 /* Info parsed from data in iCell. */
1091 i64 iRowid; /* rowid parsed. */
1092 unsigned nRecordCols; /* how many items in the record. */
1093 u64 iRecordOffset; /* offset to record data. */
1094 /* TODO(shess): nRecordBytes and nRecordHeaderBytes are used in
1095 * leafCursorCellColInfo() to prevent buffer overruns.
1096 * leafCursorCellDecode() already verified that the cell is valid, so
1097 * those checks should be redundant.
1099 u64 nRecordBytes; /* Size of record data. */
1100 unsigned nLocalRecordBytes; /* Amount of record data in-page. */
1101 unsigned nRecordHeaderBytes; /* Size of record header data. */
1102 unsigned char *pRecordHeader; /* Pointer to record header data. */
1103 int bFreeRecordHeader; /* True if record header requires free. */
1104 RecoverOverflow *pOverflow; /* Cell overflow info, if needed. */
1107 /* Internal helper shared between next-page and create-cursor. If
1108 * pPage is a leaf page, it will be stored in the cursor and state
1109 * initialized for reading cells.
1111 * If pPage is an interior page, a new parent cursor is created and
1112 * injected on the stack. This is necessary to handle trees with
1113 * uneven depth, but also is used during initial setup.
1115 * If pPage is not a table page at all, it is discarded.
1117 * If SQLITE_OK is returned, the caller no longer owns pPage,
1118 * otherwise the caller is responsible for discarding it.
1120 static int leafCursorLoadPage(RecoverLeafCursor *pCursor, DbPage *pPage){
1121 const unsigned char *pPageHeader; /* Header of *pPage */
1123 /* Release the current page. */
1124 if( pCursor->pPage ){
1125 sqlite3PagerUnref(pCursor->pPage);
1126 pCursor->pPage = NULL;
1127 pCursor->iCell = pCursor->nCells = 0;
1130 /* If the page is an unexpected interior node, inject a new stack
1131 * layer and try again from there.
1133 pPageHeader = PageHeader(pPage);
1134 if( pPageHeader[kiPageTypeOffset]==kTableInteriorPage ){
1135 RecoverInteriorCursor *pParent;
1136 int rc = interiorCursorCreate(pCursor->pParent, pPage, pCursor->nPageSize,
1138 if( rc!=SQLITE_OK ){
1141 pCursor->pParent = pParent;
1145 /* Not a leaf page, skip it. */
1146 if( pPageHeader[kiPageTypeOffset]!=kTableLeafPage ){
1147 sqlite3PagerUnref(pPage);
1151 /* Take ownership of the page and start decoding. */
1152 pCursor->pPage = pPage;
1154 pCursor->nCells = decodeUnsigned16(pPageHeader + kiPageCellCountOffset);
1158 /* Get the next leaf-level page in the tree. Returns SQLITE_ROW when
1159 * a leaf page is found, SQLITE_DONE when no more leaves exist, or any
1160 * error which occurred.
1162 static int leafCursorNextPage(RecoverLeafCursor *pCursor){
1163 if( !pCursor->pParent ){
1167 /* Repeatedly load the parent's next child page until a leaf is found. */
1170 int rc = interiorCursorNextPage(&pCursor->pParent, &pNextPage);
1171 if( rc!=SQLITE_ROW ){
1172 assert( rc==SQLITE_DONE );
1176 rc = leafCursorLoadPage(pCursor, pNextPage);
1177 if( rc!=SQLITE_OK ){
1178 sqlite3PagerUnref(pNextPage);
1181 } while( !pCursor->pPage );
1186 static void leafCursorDestroyCellData(RecoverLeafCursor *pCursor){
1187 if( pCursor->bFreeRecordHeader ){
1188 sqlite3_free(pCursor->pRecordHeader);
1190 pCursor->bFreeRecordHeader = 0;
1191 pCursor->pRecordHeader = NULL;
1193 if( pCursor->pOverflow ){
1194 overflowDestroy(pCursor->pOverflow);
1195 pCursor->pOverflow = NULL;
1199 static void leafCursorDestroy(RecoverLeafCursor *pCursor){
1200 leafCursorDestroyCellData(pCursor);
1202 if( pCursor->pParent ){
1203 interiorCursorDestroy(pCursor->pParent);
1204 pCursor->pParent = NULL;
1207 if( pCursor->pPage ){
1208 sqlite3PagerUnref(pCursor->pPage);
1209 pCursor->pPage = NULL;
1212 memset(pCursor, 0xA5, sizeof(*pCursor));
1213 sqlite3_free(pCursor);
1216 /* Create a cursor to iterate the rows from the leaf pages of a table
1217 * rooted at iRootPage.
1219 /* TODO(shess): recoverOpen() calls this to setup the cursor, and I
1220 * think that recoverFilter() may make a hard assumption that the
1221 * cursor returned will turn up at least one valid cell.
1223 * The cases I can think of which break this assumption are:
1224 * - pPage is a valid leaf page with no valid cells.
1225 * - pPage is a valid interior page with no valid leaves.
1226 * - pPage is a valid interior page who's leaves contain no valid cells.
1227 * - pPage is not a valid leaf or interior page.
1229 static int leafCursorCreate(Pager *pPager, unsigned nPageSize,
1230 u32 iRootPage, RecoverLeafCursor **ppCursor){
1231 DbPage *pPage; /* Reference to page at iRootPage. */
1232 RecoverLeafCursor *pCursor; /* Leaf cursor being constructed. */
1235 /* Start out with the root page. */
1236 rc = sqlite3PagerAcquire(pPager, iRootPage, &pPage, 0);
1237 if( rc!=SQLITE_OK ){
1241 pCursor = sqlite3_malloc(sizeof(RecoverLeafCursor));
1243 sqlite3PagerUnref(pPage);
1244 return SQLITE_NOMEM;
1246 memset(pCursor, 0, sizeof(*pCursor));
1248 pCursor->nPageSize = nPageSize;
1250 rc = leafCursorLoadPage(pCursor, pPage);
1251 if( rc!=SQLITE_OK ){
1252 sqlite3PagerUnref(pPage);
1253 leafCursorDestroy(pCursor);
1257 /* pPage wasn't a leaf page, find the next leaf page. */
1258 if( !pCursor->pPage ){
1259 rc = leafCursorNextPage(pCursor);
1260 if( rc!=SQLITE_DONE && rc!=SQLITE_ROW ){
1261 leafCursorDestroy(pCursor);
1266 *ppCursor = pCursor;
1270 /* Useful for setting breakpoints. */
1271 static int ValidateError(){
1272 return SQLITE_ERROR;
1275 /* Setup the cursor for reading the information from cell iCell. */
1276 static int leafCursorCellDecode(RecoverLeafCursor *pCursor){
1277 const unsigned char *pPageHeader; /* Header of current page. */
1278 const unsigned char *pCellOffsets; /* Pointer to page's cell offsets. */
1279 unsigned iCellOffset; /* Offset of current cell (iCell). */
1280 const unsigned char *pCell; /* Pointer to data at iCellOffset. */
1281 unsigned nCellMaxBytes; /* Maximum local size of iCell. */
1282 unsigned iEndOffset; /* End of iCell's in-page data. */
1283 u64 nRecordBytes; /* Expected size of cell, w/overflow. */
1284 u64 iRowid; /* iCell's rowid (in table). */
1285 unsigned nRead; /* Amount of cell read. */
1286 unsigned nRecordHeaderRead; /* Header data read. */
1287 u64 nRecordHeaderBytes; /* Header size expected. */
1288 unsigned nRecordCols; /* Columns read from header. */
1289 u64 nRecordColBytes; /* Bytes in payload for those columns. */
1293 assert( pCursor->iCell<pCursor->nCells );
1295 leafCursorDestroyCellData(pCursor);
1297 /* Find the offset to the row. */
1298 pPageHeader = PageHeader(pCursor->pPage);
1299 pCellOffsets = pPageHeader + knPageLeafHeaderBytes;
1300 iCellOffset = decodeUnsigned16(pCellOffsets + pCursor->iCell*2);
1301 if( iCellOffset>=pCursor->nPageSize ){
1302 return ValidateError();
1305 pCell = PageData(pCursor->pPage, iCellOffset);
1306 nCellMaxBytes = pCursor->nPageSize - iCellOffset;
1308 /* B-tree leaf cells lead with varint record size, varint rowid and
1309 * varint header size.
1311 /* TODO(shess): The smallest page size is 512 bytes, which has an m
1312 * of 39. Three varints need at most 27 bytes to encode. I think.
1314 if( !checkVarints(pCell, nCellMaxBytes, 3) ){
1315 return ValidateError();
1318 nRead = getVarint(pCell, &nRecordBytes);
1319 assert( iCellOffset+nRead<=pCursor->nPageSize );
1320 pCursor->nRecordBytes = nRecordBytes;
1322 nRead += getVarint(pCell + nRead, &iRowid);
1323 assert( iCellOffset+nRead<=pCursor->nPageSize );
1324 pCursor->iRowid = (i64)iRowid;
1326 pCursor->iRecordOffset = iCellOffset + nRead;
1328 /* Start overflow setup here because nLocalRecordBytes is needed to
1329 * check cell overlap.
1331 rc = overflowMaybeCreate(pCursor->pPage, pCursor->nPageSize,
1332 pCursor->iRecordOffset, pCursor->nRecordBytes,
1333 &pCursor->nLocalRecordBytes,
1334 &pCursor->pOverflow);
1335 if( rc!=SQLITE_OK ){
1336 return ValidateError();
1339 /* Check that no other cell starts within this cell. */
1340 iEndOffset = pCursor->iRecordOffset + pCursor->nLocalRecordBytes;
1341 for( i=0; i<pCursor->nCells; ++i ){
1342 const unsigned iOtherOffset = decodeUnsigned16(pCellOffsets + i*2);
1343 if( iOtherOffset>iCellOffset && iOtherOffset<iEndOffset ){
1344 return ValidateError();
1348 nRecordHeaderRead = getVarint(pCell + nRead, &nRecordHeaderBytes);
1349 assert( nRecordHeaderBytes<=nRecordBytes );
1350 pCursor->nRecordHeaderBytes = nRecordHeaderBytes;
1352 /* Large headers could overflow if pages are small. */
1353 rc = overflowGetSegment(pCursor->pPage,
1354 pCursor->iRecordOffset, pCursor->nLocalRecordBytes,
1355 pCursor->pOverflow, 0, nRecordHeaderBytes,
1356 &pCursor->pRecordHeader, &pCursor->bFreeRecordHeader);
1357 if( rc!=SQLITE_OK ){
1358 return ValidateError();
1361 /* Tally up the column count and size of data. */
1363 nRecordColBytes = 0;
1364 while( nRecordHeaderRead<nRecordHeaderBytes ){
1365 u64 iSerialType; /* Type descriptor for current column. */
1366 if( !checkVarint(pCursor->pRecordHeader + nRecordHeaderRead,
1367 nRecordHeaderBytes - nRecordHeaderRead) ){
1368 return ValidateError();
1370 nRecordHeaderRead += getVarint(pCursor->pRecordHeader + nRecordHeaderRead,
1372 if( iSerialType==10 || iSerialType==11 ){
1373 return ValidateError();
1375 nRecordColBytes += SerialTypeLength(iSerialType);
1378 pCursor->nRecordCols = nRecordCols;
1380 /* Parsing the header used as many bytes as expected. */
1381 if( nRecordHeaderRead!=nRecordHeaderBytes ){
1382 return ValidateError();
1385 /* Calculated record is size of expected record. */
1386 if( nRecordHeaderBytes+nRecordColBytes!=nRecordBytes ){
1387 return ValidateError();
1393 static i64 leafCursorCellRowid(RecoverLeafCursor *pCursor){
1394 return pCursor->iRowid;
1397 static unsigned leafCursorCellColumns(RecoverLeafCursor *pCursor){
1398 return pCursor->nRecordCols;
1401 /* Get the column info for the cell. Pass NULL for ppBase to prevent
1402 * retrieving the data segment. If *pbFree is true, *ppBase must be
1403 * freed by the caller using sqlite3_free().
1405 static int leafCursorCellColInfo(RecoverLeafCursor *pCursor,
1406 unsigned iCol, u64 *piColType,
1407 unsigned char **ppBase, int *pbFree){
1408 const unsigned char *pRecordHeader; /* Current cell's header. */
1409 u64 nRecordHeaderBytes; /* Bytes in pRecordHeader. */
1410 unsigned nRead; /* Bytes read from header. */
1411 u64 iColEndOffset; /* Offset to end of column in cell. */
1412 unsigned nColsSkipped; /* Count columns as procesed. */
1413 u64 iSerialType; /* Type descriptor for current column. */
1415 /* Implicit NULL for columns past the end. This case happens when
1416 * rows have not been updated since an ALTER TABLE added columns.
1417 * It is more convenient to address here than in callers.
1419 if( iCol>=pCursor->nRecordCols ){
1428 /* Must be able to decode header size. */
1429 pRecordHeader = pCursor->pRecordHeader;
1430 if( !checkVarint(pRecordHeader, pCursor->nRecordHeaderBytes) ){
1431 return SQLITE_CORRUPT;
1434 /* Rather than caching the header size and how many bytes it took,
1435 * decode it every time.
1437 nRead = getVarint(pRecordHeader, &nRecordHeaderBytes);
1438 assert( nRecordHeaderBytes==pCursor->nRecordHeaderBytes );
1440 /* Scan forward to the indicated column. Scans to _after_ column
1441 * for later range checking.
1443 /* TODO(shess): This could get expensive for very wide tables. An
1444 * array of iSerialType could be built in leafCursorCellDecode(), but
1445 * the number of columns is dynamic per row, so it would add memory
1446 * management complexity. Enough info to efficiently forward
1447 * iterate could be kept, if all clients forward iterate
1448 * (recoverColumn() may not).
1452 while( nColsSkipped<=iCol && nRead<nRecordHeaderBytes ){
1453 if( !checkVarint(pRecordHeader + nRead, nRecordHeaderBytes - nRead) ){
1454 return SQLITE_CORRUPT;
1456 nRead += getVarint(pRecordHeader + nRead, &iSerialType);
1457 iColEndOffset += SerialTypeLength(iSerialType);
1461 /* Column's data extends past record's end. */
1462 if( nRecordHeaderBytes+iColEndOffset>pCursor->nRecordBytes ){
1463 return SQLITE_CORRUPT;
1466 *piColType = iSerialType;
1468 const u32 nColBytes = SerialTypeLength(iSerialType);
1470 /* Offset from start of record to beginning of column. */
1471 const unsigned iColOffset = nRecordHeaderBytes+iColEndOffset-nColBytes;
1473 return overflowGetSegment(pCursor->pPage, pCursor->iRecordOffset,
1474 pCursor->nLocalRecordBytes, pCursor->pOverflow,
1475 iColOffset, nColBytes, ppBase, pbFree);
1480 static int leafCursorNextValidCell(RecoverLeafCursor *pCursor){
1484 /* Move to the next cell. */
1487 /* No more cells, get the next leaf. */
1488 if( pCursor->iCell>=pCursor->nCells ){
1489 rc = leafCursorNextPage(pCursor);
1490 if( rc!=SQLITE_ROW ){
1493 assert( pCursor->iCell==0 );
1496 /* If the cell is valid, indicate that a row is available. */
1497 rc = leafCursorCellDecode(pCursor);
1498 if( rc==SQLITE_OK ){
1502 /* Iterate until done or a valid row is found. */
1503 /* TODO(shess): Remove debugging output. */
1504 fprintf(stderr, "Skipping invalid cell\n");
1506 return SQLITE_ERROR;
1509 typedef struct Recover Recover;
1512 sqlite3 *db; /* Host database connection */
1513 char *zDb; /* Database containing target table */
1514 char *zTable; /* Target table */
1515 unsigned nCols; /* Number of columns in target table */
1516 unsigned char *pTypes; /* Types of columns in target table */
1519 /* Internal helper for deleting the module. */
1520 static void recoverRelease(Recover *pRecover){
1521 sqlite3_free(pRecover->zDb);
1522 sqlite3_free(pRecover->zTable);
1523 sqlite3_free(pRecover->pTypes);
1524 memset(pRecover, 0xA5, sizeof(*pRecover));
1525 sqlite3_free(pRecover);
1528 /* Helper function for initializing the module. Forward-declared so
1529 * recoverCreate() and recoverConnect() can see it.
1531 static int recoverInit(
1532 sqlite3 *, void *, int, const char *const*, sqlite3_vtab **, char **
1535 static int recoverCreate(
1538 int argc, const char *const*argv,
1539 sqlite3_vtab **ppVtab,
1543 return recoverInit(db, pAux, argc, argv, ppVtab, pzErr);
1546 /* This should never be called. */
1547 static int recoverConnect(
1550 int argc, const char *const*argv,
1551 sqlite3_vtab **ppVtab,
1555 return recoverInit(db, pAux, argc, argv, ppVtab, pzErr);
1558 /* No indices supported. */
1559 static int recoverBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){
1564 /* Logically, this should never be called. */
1565 static int recoverDisconnect(sqlite3_vtab *pVtab){
1567 recoverRelease((Recover*)pVtab);
1571 static int recoverDestroy(sqlite3_vtab *pVtab){
1573 recoverRelease((Recover*)pVtab);
1577 typedef struct RecoverCursor RecoverCursor;
1578 struct RecoverCursor {
1579 sqlite3_vtab_cursor base;
1580 RecoverLeafCursor *pLeafCursor;
1585 static int recoverOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
1586 Recover *pRecover = (Recover*)pVTab;
1587 u32 iRootPage; /* Root page of the backing table. */
1588 int iEncoding; /* UTF encoding for backing database. */
1589 unsigned nPageSize; /* Size of pages in backing database. */
1590 Pager *pPager; /* Backing database pager. */
1591 RecoverLeafCursor *pLeafCursor; /* Cursor to read table's leaf pages. */
1592 RecoverCursor *pCursor; /* Cursor to read rows from leaves. */
1598 rc = getRootPage(pRecover->db, pRecover->zDb, pRecover->zTable,
1600 if( rc!=SQLITE_OK ){
1605 rc = getEncoding(pRecover->db, pRecover->zDb, &iEncoding);
1606 if( rc!=SQLITE_OK ){
1610 rc = GetPager(pRecover->db, pRecover->zDb, &pPager, &nPageSize);
1611 if( rc!=SQLITE_OK ){
1615 rc = leafCursorCreate(pPager, nPageSize, iRootPage, &pLeafCursor);
1616 if( rc!=SQLITE_OK ){
1620 pCursor = sqlite3_malloc(sizeof(RecoverCursor));
1622 leafCursorDestroy(pLeafCursor);
1623 return SQLITE_NOMEM;
1625 memset(pCursor, 0, sizeof(*pCursor));
1626 pCursor->base.pVtab = pVTab;
1627 pCursor->pLeafCursor = pLeafCursor;
1628 pCursor->iEncoding = iEncoding;
1630 *ppCursor = (sqlite3_vtab_cursor*)pCursor;
1634 static int recoverClose(sqlite3_vtab_cursor *cur){
1635 RecoverCursor *pCursor = (RecoverCursor*)cur;
1637 if( pCursor->pLeafCursor ){
1638 leafCursorDestroy(pCursor->pLeafCursor);
1639 pCursor->pLeafCursor = NULL;
1641 memset(pCursor, 0xA5, sizeof(*pCursor));
1646 /* Helpful place to set a breakpoint. */
1647 static int RecoverInvalidCell(){
1648 return SQLITE_ERROR;
1651 /* Returns SQLITE_OK if the cell has an appropriate number of columns
1652 * with the appropriate types of data.
1654 static int recoverValidateLeafCell(Recover *pRecover, RecoverCursor *pCursor){
1657 /* If the row's storage has too many columns, skip it. */
1658 if( leafCursorCellColumns(pCursor->pLeafCursor)>pRecover->nCols ){
1659 return RecoverInvalidCell();
1662 /* Skip rows with unexpected types. */
1663 for( i=0; i<pRecover->nCols; ++i ){
1664 u64 iType; /* Storage type of column i. */
1668 if( (pRecover->pTypes[i]&MASK_ROWID) ){
1672 rc = leafCursorCellColInfo(pCursor->pLeafCursor, i, &iType, NULL, NULL);
1673 assert( rc==SQLITE_OK );
1674 if( rc!=SQLITE_OK || !SerialTypeIsCompatible(iType, pRecover->pTypes[i]) ){
1675 return RecoverInvalidCell();
1682 static int recoverNext(sqlite3_vtab_cursor *pVtabCursor){
1683 RecoverCursor *pCursor = (RecoverCursor*)pVtabCursor;
1684 Recover *pRecover = (Recover*)pCursor->base.pVtab;
1689 /* Scan forward to the next cell with valid storage, then check that
1690 * the stored data matches the schema.
1692 while( (rc = leafCursorNextValidCell(pCursor->pLeafCursor))==SQLITE_ROW ){
1693 if( recoverValidateLeafCell(pRecover, pCursor)==SQLITE_OK ){
1698 if( rc==SQLITE_DONE ){
1703 assert( rc!=SQLITE_OK );
1707 static int recoverFilter(
1708 sqlite3_vtab_cursor *pVtabCursor,
1709 int idxNum, const char *idxStr,
1710 int argc, sqlite3_value **argv
1712 RecoverCursor *pCursor = (RecoverCursor*)pVtabCursor;
1713 Recover *pRecover = (Recover*)pCursor->base.pVtab;
1718 /* Load the first cell, and iterate forward if it's not valid. */
1719 /* TODO(shess): What happens if no cells at all are valid? */
1720 rc = leafCursorCellDecode(pCursor->pLeafCursor);
1721 if( rc!=SQLITE_OK || recoverValidateLeafCell(pRecover, pCursor)!=SQLITE_OK ){
1722 return recoverNext(pVtabCursor);
1728 static int recoverEof(sqlite3_vtab_cursor *pVtabCursor){
1729 RecoverCursor *pCursor = (RecoverCursor*)pVtabCursor;
1731 return pCursor->bEOF;
1734 static int recoverColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){
1735 RecoverCursor *pCursor = (RecoverCursor*)cur;
1736 Recover *pRecover = (Recover*)pCursor->base.pVtab;
1737 u64 iColType; /* Storage type of column i. */
1738 unsigned char *pColData; /* Column i's data. */
1739 int shouldFree; /* Non-zero if pColData should be freed. */
1744 if( i>=pRecover->nCols ){
1745 return SQLITE_ERROR;
1749 if( (pRecover->pTypes[i]&MASK_ROWID) ){
1750 sqlite3_result_int64(ctx, leafCursorCellRowid(pCursor->pLeafCursor));
1756 rc = leafCursorCellColInfo(pCursor->pLeafCursor, i, &iColType,
1757 &pColData, &shouldFree);
1758 if( rc!=SQLITE_OK ){
1761 /* recoverValidateLeafCell() should guarantee that this will never
1764 if( !SerialTypeIsCompatible(iColType, pRecover->pTypes[i]) ){
1766 sqlite3_free(pColData);
1768 return SQLITE_ERROR;
1772 case 0 : sqlite3_result_null(ctx); break;
1773 case 1 : sqlite3_result_int64(ctx, decodeSigned(pColData, 1)); break;
1774 case 2 : sqlite3_result_int64(ctx, decodeSigned(pColData, 2)); break;
1775 case 3 : sqlite3_result_int64(ctx, decodeSigned(pColData, 3)); break;
1776 case 4 : sqlite3_result_int64(ctx, decodeSigned(pColData, 4)); break;
1777 case 5 : sqlite3_result_int64(ctx, decodeSigned(pColData, 6)); break;
1778 case 6 : sqlite3_result_int64(ctx, decodeSigned(pColData, 8)); break;
1779 case 7 : sqlite3_result_double(ctx, decodeFloat64(pColData)); break;
1780 case 8 : sqlite3_result_int(ctx, 0); break;
1781 case 9 : sqlite3_result_int(ctx, 1); break;
1782 case 10 : assert( iColType!=10 ); break;
1783 case 11 : assert( iColType!=11 ); break;
1786 u32 l = SerialTypeLength(iColType);
1788 /* If pColData was already allocated, arrange to pass ownership. */
1789 sqlite3_destructor_type pFn = SQLITE_TRANSIENT;
1795 if( SerialTypeIsBlob(iColType) ){
1796 sqlite3_result_blob(ctx, pColData, l, pFn);
1798 if( pCursor->iEncoding==SQLITE_UTF16LE ){
1799 sqlite3_result_text16le(ctx, (const void*)pColData, l, pFn);
1800 }else if( pCursor->iEncoding==SQLITE_UTF16BE ){
1801 sqlite3_result_text16be(ctx, (const void*)pColData, l, pFn);
1803 sqlite3_result_text(ctx, (const char*)pColData, l, pFn);
1809 sqlite3_free(pColData);
1814 static int recoverRowid(sqlite3_vtab_cursor *pVtabCursor, sqlite_int64 *pRowid){
1815 RecoverCursor *pCursor = (RecoverCursor*)pVtabCursor;
1817 *pRowid = leafCursorCellRowid(pCursor->pLeafCursor);
1821 static sqlite3_module recoverModule = {
1823 recoverCreate, /* xCreate - create a table */
1824 recoverConnect, /* xConnect - connect to an existing table */
1825 recoverBestIndex, /* xBestIndex - Determine search strategy */
1826 recoverDisconnect, /* xDisconnect - Disconnect from a table */
1827 recoverDestroy, /* xDestroy - Drop a table */
1828 recoverOpen, /* xOpen - open a cursor */
1829 recoverClose, /* xClose - close a cursor */
1830 recoverFilter, /* xFilter - configure scan constraints */
1831 recoverNext, /* xNext - advance a cursor */
1832 recoverEof, /* xEof */
1833 recoverColumn, /* xColumn - read data */
1834 recoverRowid, /* xRowid - read data */
1835 0, /* xUpdate - write data */
1836 0, /* xBegin - begin transaction */
1837 0, /* xSync - sync transaction */
1838 0, /* xCommit - commit transaction */
1839 0, /* xRollback - rollback transaction */
1840 0, /* xFindFunction - function overloading */
1841 0, /* xRename - rename the table */
1844 int recoverVtableInit(sqlite3 *db){
1845 return sqlite3_create_module_v2(db, "recover", &recoverModule, NULL, 0);
1848 /* This section of code is for parsing the create input and
1849 * initializing the module.
1852 /* Find the next word in zText and place the endpoints in pzWord*.
1853 * Returns true if the word is non-empty. "Word" is defined as
1854 * ASCII alphanumeric plus '_' at this time.
1856 static int findWord(const char *zText,
1857 const char **pzWordStart, const char **pzWordEnd){
1859 while( ascii_isspace(*zText) ){
1862 *pzWordStart = zText;
1863 while( ascii_isalnum(*zText) || *zText=='_' ){
1866 r = zText>*pzWordStart; /* In case pzWordStart==pzWordEnd */
1871 /* Return true if the next word in zText is zWord, also setting
1872 * *pzContinue to the character after the word.
1874 static int expectWord(const char *zText, const char *zWord,
1875 const char **pzContinue){
1876 const char *zWordStart, *zWordEnd;
1877 if( findWord(zText, &zWordStart, &zWordEnd) &&
1878 ascii_strncasecmp(zWord, zWordStart, zWordEnd - zWordStart)==0 ){
1879 *pzContinue = zWordEnd;
1885 /* Parse the name and type information out of parameter. In case of
1886 * success, *pzNameStart/End contain the name of the column,
1887 * *pzTypeStart/End contain the top-level type, and *pTypeMask has the
1888 * type mask to use for the column.
1890 static int findNameAndType(const char *parameter,
1891 const char **pzNameStart, const char **pzNameEnd,
1892 const char **pzTypeStart, const char **pzTypeEnd,
1893 unsigned char *pTypeMask){
1894 unsigned nNameLen; /* Length of found name. */
1895 const char *zEnd; /* Current end of parsed column information. */
1896 int bNotNull; /* Non-zero if NULL is not allowed for name. */
1897 int bStrict; /* Non-zero if column requires exact type match. */
1898 const char *zDummy; /* Dummy parameter, result unused. */
1901 /* strictMask is used for STRICT, strictMask|otherMask if STRICT is
1902 * not supplied. zReplace provides an alternate type to expose to
1907 unsigned char strictMask;
1908 unsigned char otherMask;
1909 const char *zReplace;
1912 MASK_INTEGER | MASK_FLOAT | MASK_BLOB | MASK_TEXT | MASK_NULL,
1915 { "ROWID", MASK_INTEGER | MASK_ROWID, 0, "INTEGER", },
1916 { "INTEGER", MASK_INTEGER | MASK_NULL, 0, NULL, },
1917 { "FLOAT", MASK_FLOAT | MASK_NULL, MASK_INTEGER, NULL, },
1918 { "NUMERIC", MASK_INTEGER | MASK_FLOAT | MASK_NULL, MASK_TEXT, NULL, },
1919 { "TEXT", MASK_TEXT | MASK_NULL, MASK_BLOB, NULL, },
1920 { "BLOB", MASK_BLOB | MASK_NULL, 0, NULL, },
1923 if( !findWord(parameter, pzNameStart, pzNameEnd) ){
1924 return SQLITE_MISUSE;
1927 /* Manifest typing, accept any storage type. */
1928 if( !findWord(*pzNameEnd, pzTypeStart, pzTypeEnd) ){
1929 *pzTypeEnd = *pzTypeStart = "";
1930 *pTypeMask = MASK_INTEGER | MASK_FLOAT | MASK_BLOB | MASK_TEXT | MASK_NULL;
1934 nNameLen = *pzTypeEnd - *pzTypeStart;
1935 for( i=0; i<ArraySize(kTypeInfo); ++i ){
1936 if( ascii_strncasecmp(kTypeInfo[i].zName, *pzTypeStart, nNameLen)==0 ){
1940 if( i==ArraySize(kTypeInfo) ){
1941 return SQLITE_MISUSE;
1946 if( expectWord(zEnd, "STRICT", &zEnd) ){
1947 /* TODO(shess): Ick. But I don't want another single-purpose
1950 if( kTypeInfo[i].zReplace && !kTypeInfo[i].zReplace[0] ){
1951 return SQLITE_MISUSE;
1957 if( expectWord(zEnd, "NOT", &zEnd) ){
1958 if( expectWord(zEnd, "NULL", &zEnd) ){
1961 /* Anything other than NULL after NOT is an error. */
1962 return SQLITE_MISUSE;
1966 /* Anything else is an error. */
1967 if( findWord(zEnd, &zDummy, &zDummy) ){
1968 return SQLITE_MISUSE;
1971 *pTypeMask = kTypeInfo[i].strictMask;
1973 *pTypeMask |= kTypeInfo[i].otherMask;
1976 *pTypeMask &= ~MASK_NULL;
1978 if( kTypeInfo[i].zReplace ){
1979 *pzTypeStart = kTypeInfo[i].zReplace;
1980 *pzTypeEnd = *pzTypeStart + strlen(*pzTypeStart);
1985 /* Parse the arguments, placing type masks in *pTypes and the exposed
1986 * schema in *pzCreateSql (for sqlite3_declare_vtab).
1988 static int ParseColumnsAndGenerateCreate(unsigned nCols,
1989 const char *const *pCols,
1991 unsigned char *pTypes,
1994 char *zCreateSql = sqlite3_mprintf("CREATE TABLE x(");
1996 return SQLITE_NOMEM;
1999 for( i=0; i<nCols; i++ ){
2000 const char *zSep = (i < nCols - 1 ? ", " : ")");
2001 const char *zNotNull = "";
2002 const char *zNameStart, *zNameEnd;
2003 const char *zTypeStart, *zTypeEnd;
2004 int rc = findNameAndType(pCols[i],
2005 &zNameStart, &zNameEnd,
2006 &zTypeStart, &zTypeEnd,
2008 if( rc!=SQLITE_OK ){
2009 *pzErr = sqlite3_mprintf("unable to parse column %d", i);
2010 sqlite3_free(zCreateSql);
2014 if( !(pTypes[i]&MASK_NULL) ){
2015 zNotNull = " NOT NULL";
2018 /* Add name and type to the create statement. */
2019 zCreateSql = sqlite3_mprintf("%z%.*s %.*s%s%s",
2021 zNameEnd - zNameStart, zNameStart,
2022 zTypeEnd - zTypeStart, zTypeStart,
2025 return SQLITE_NOMEM;
2029 *pzCreateSql = zCreateSql;
2033 /* Helper function for initializing the module. */
2034 /* argv[0] module name
2035 * argv[1] db name for virtual table
2036 * argv[2] virtual table name
2037 * argv[3] backing table name
2040 /* TODO(shess): Since connect isn't supported, could inline into
2043 /* TODO(shess): Explore cases where it would make sense to set *pzErr. */
2044 static int recoverInit(
2045 sqlite3 *db, /* Database connection */
2046 void *pAux, /* unused */
2047 int argc, const char *const*argv, /* Parameters to CREATE TABLE statement */
2048 sqlite3_vtab **ppVtab, /* OUT: New virtual table */
2049 char **pzErr /* OUT: Error message, if any */
2051 const unsigned kTypeCol = 4; /* First argument with column type info. */
2052 Recover *pRecover; /* Virtual table structure being created. */
2053 char *zDot; /* Any dot found in "db.table" backing. */
2054 u32 iRootPage; /* Root page of backing table. */
2055 char *zCreateSql; /* Schema of created virtual table. */
2058 /* Require to be in the temp database. */
2059 if( ascii_strcasecmp(argv[1], "temp")!=0 ){
2060 *pzErr = sqlite3_mprintf("recover table must be in temp database");
2061 return SQLITE_MISUSE;
2064 /* Need the backing table and at least one column. */
2065 if( argc<=kTypeCol ){
2066 *pzErr = sqlite3_mprintf("no columns specified");
2067 return SQLITE_MISUSE;
2070 pRecover = sqlite3_malloc(sizeof(Recover));
2072 return SQLITE_NOMEM;
2074 memset(pRecover, 0, sizeof(*pRecover));
2075 pRecover->base.pModule = &recoverModule;
2078 /* Parse out db.table, assuming main if no dot. */
2079 zDot = strchr(argv[3], '.');
2081 pRecover->zDb = sqlite3_strdup(db->aDb[0].zName);
2082 pRecover->zTable = sqlite3_strdup(argv[3]);
2083 }else if( zDot>argv[3] && zDot[1]!='\0' ){
2084 pRecover->zDb = sqlite3_strndup(argv[3], zDot - argv[3]);
2085 pRecover->zTable = sqlite3_strdup(zDot + 1);
2087 /* ".table" or "db." not allowed. */
2088 *pzErr = sqlite3_mprintf("ill-formed table specifier");
2089 recoverRelease(pRecover);
2090 return SQLITE_ERROR;
2093 pRecover->nCols = argc - kTypeCol;
2094 pRecover->pTypes = sqlite3_malloc(pRecover->nCols);
2095 if( !pRecover->zDb || !pRecover->zTable || !pRecover->pTypes ){
2096 recoverRelease(pRecover);
2097 return SQLITE_NOMEM;
2100 /* Require the backing table to exist. */
2101 /* TODO(shess): Be more pedantic about the form of the descriptor
2102 * string. This already fails for poorly-formed strings, simply
2103 * because there won't be a root page, but it would make more sense
2106 rc = getRootPage(pRecover->db, pRecover->zDb, pRecover->zTable, &iRootPage);
2107 if( rc!=SQLITE_OK ){
2108 *pzErr = sqlite3_mprintf("unable to find backing table");
2109 recoverRelease(pRecover);
2113 /* Parse the column definitions. */
2114 rc = ParseColumnsAndGenerateCreate(pRecover->nCols, argv + kTypeCol,
2115 &zCreateSql, pRecover->pTypes, pzErr);
2116 if( rc!=SQLITE_OK ){
2117 recoverRelease(pRecover);
2121 rc = sqlite3_declare_vtab(db, zCreateSql);
2122 sqlite3_free(zCreateSql);
2123 if( rc!=SQLITE_OK ){
2124 recoverRelease(pRecover);
2128 *ppVtab = (sqlite3_vtab *)pRecover;