diff options
Diffstat (limited to 'db/pdfile.h')
| -rw-r--r-- | db/pdfile.h | 236 |
1 files changed, 113 insertions, 123 deletions
diff --git a/db/pdfile.h b/db/pdfile.h index d268aac..91f4877 100644 --- a/db/pdfile.h +++ b/db/pdfile.h @@ -29,8 +29,9 @@ #include "../util/mmap.h" #include "diskloc.h" #include "jsobjmanipulator.h" -#include "namespace.h" +#include "namespace-inl.h" #include "client.h" +#include "mongommf.h" namespace mongo { @@ -45,53 +46,60 @@ namespace mongo { /* low level - only drops this ns */ void dropNS(const string& dropNs); - + /* deletes this ns, indexes and cursors */ - void dropCollection( const string &name, string &errmsg, BSONObjBuilder &result ); + void dropCollection( const string &name, string &errmsg, BSONObjBuilder &result ); bool userCreateNS(const char *ns, BSONObj j, string& err, bool logForReplication, bool *deferIdIndex = 0); shared_ptr<Cursor> findTableScan(const char *ns, const BSONObj& order, const DiskLoc &startLoc=DiskLoc()); -// -1 if library unavailable. + // -1 if library unavailable. boost::intmax_t freeSpace( const string &path = dbpath ); + bool isValidNS( const StringData& ns ); + /*---------------------------------------------------------------------*/ class MongoDataFile { friend class DataFileMgr; friend class BasicCursor; public: - MongoDataFile(int fn) : fileNo(fn) { } + MongoDataFile(int fn) : _mb(0), fileNo(fn) { } void open(const char *filename, int requestedDataSize = 0, bool preallocateOnly = false); - /* allocate a new extent from this datafile. + /* allocate a new extent from this datafile. @param capped - true if capped collection @param loops is our recursion check variable - you want to pass in zero */ Extent* createExtent(const char *ns, int approxSize, bool capped = false, int loops = 0); - DataFileHeader *getHeader() { - return header; - } + DataFileHeader *getHeader() { return header(); } + + unsigned long long length() const { return mmf.length(); } /* return max size an extent may be */ static int maxSize(); - + + /** fsync */ void flush( bool sync ); - + + /** only use fore debugging */ + Extent* debug_getExtent(DiskLoc loc) { return _getExtent( loc ); } private: void badOfs(int) const; - + void badOfs2(int) const; int defaultSize( const char *filename ) const; - Extent* getExtent(DiskLoc loc); - Extent* _getExtent(DiskLoc loc); + Extent* getExtent(DiskLoc loc) const; + Extent* _getExtent(DiskLoc loc) const; Record* recordAt(DiskLoc dl); Record* makeRecord(DiskLoc dl, int size); - void grow(DiskLoc dl, int size); + void grow(DiskLoc dl, int size); - MMF mmf; - MMF::Pointer _p; - DataFileHeader *header; + char* p() const { return (char *) _mb; } + DataFileHeader* header() { return (DataFileHeader*) _mb; } + + MongoMMF mmf; + void *_mb; // the memory mapped view int fileNo; }; @@ -110,9 +118,9 @@ namespace mongo { NamespaceDetails *d, NamespaceDetailsTransient *nsdt, Record *toupdate, const DiskLoc& dl, - const char *buf, int len, OpDebug& debug, bool &changedId, bool god=false); + const char *buf, int len, OpDebug& debug, bool god=false); - // The object o may be updated if modified on insert. + // The object o may be updated if modified on insert. void insertAndLog( const char *ns, const BSONObj &o, bool god = false ); /** @param obj both and in and out param -- insert can sometimes modify an object (such as add _id). */ @@ -122,7 +130,6 @@ namespace mongo { void insertNoReturnVal(const char *ns, BSONObj o, bool god = false); DiskLoc insert(const char *ns, const void *buf, int len, bool god = false, const BSONElement &writeId = BSONElement(), bool mayAddIndex = true); - void deleteRecord(const char *ns, Record *todelete, const DiskLoc& dl, bool cappedOK = false, bool noWarn = false); static shared_ptr<Cursor> findAll(const char *ns, const DiskLoc &startLoc = DiskLoc()); /* special version of insert for transaction logging -- streamlined a bit. @@ -134,9 +141,10 @@ namespace mongo { static Extent* getExtent(const DiskLoc& dl); static Record* getRecord(const DiskLoc& dl); static DeletedRecord* makeDeletedRecord(const DiskLoc& dl, int len); - static void grow(const DiskLoc& dl, int len); - /* does not clean up indexes, etc. : just deletes the record in the pdfile. */ + void deleteRecord(const char *ns, Record *todelete, const DiskLoc& dl, bool cappedOK = false, bool noWarn = false); + + /* does not clean up indexes, etc. : just deletes the record in the pdfile. use deleteRecord() to unindex */ void _deleteRecord(NamespaceDetails *d, const char *ns, Record *todelete, const DiskLoc& dl); private: @@ -175,7 +183,10 @@ namespace mongo { int extentOfs; int nextOfs; int prevOfs; + + /** be careful when referencing this that your write intent was correct */ char data[4]; + int netLength() { return lengthWithHeaders - HeaderSize; } @@ -192,6 +203,12 @@ namespace mongo { /* get the next record in the namespace, traversing extents as necessary */ DiskLoc getNext(const DiskLoc& myLoc); DiskLoc getPrev(const DiskLoc& myLoc); + + struct NP { + int nextOfs; + int prevOfs; + }; + NP* np() { return (NP*) &nextOfs; } }; /* extents are datafile regions where all the records within the region @@ -206,13 +223,14 @@ namespace mongo { DiskLoc myLoc; DiskLoc xnext, xprev; /* next/prev extent for this namespace */ - /* which namespace this extent is for. this is just for troubleshooting really + /* which namespace this extent is for. this is just for troubleshooting really and won't even be correct if the collection were renamed! */ - Namespace nsDiagnostic; + Namespace nsDiagnostic; int length; /* size of the extent, including these fields */ - DiskLoc firstRecord, lastRecord; + DiskLoc firstRecord; + DiskLoc lastRecord; char _extentData[4]; static int HeaderSize() { return sizeof(Extent)-4; } @@ -224,7 +242,7 @@ namespace mongo { void dump(iostream& s) { s << " loc:" << myLoc.toString() << " xnext:" << xnext.toString() << " xprev:" << xprev.toString() << '\n'; - s << " nsdiag:" << nsDiagnostic.buf << '\n'; + s << " nsdiag:" << nsDiagnostic.toString() << '\n'; s << " size:" << length << " firstRecord:" << firstRecord.toString() << " lastRecord:" << lastRecord.toString() << '\n'; } @@ -237,9 +255,8 @@ namespace mongo { /* like init(), but for a reuse case */ DiskLoc reuse(const char *nsname); - void assertOk() { - assert(magic == 0x41424344); - } + bool isOk() const { return magic == 0x41424344; } + void assertOk() const { assert(isOk()); } Record* newRecord(int len); @@ -251,19 +268,38 @@ namespace mongo { return (Record *) (((char *) this) + x); } - Extent* getNextExtent() { - return xnext.isNull() ? 0 : DataFileMgr::getExtent(xnext); - } - Extent* getPrevExtent() { - return xprev.isNull() ? 0 : DataFileMgr::getExtent(xprev); - } - + Extent* getNextExtent() { return xnext.isNull() ? 0 : DataFileMgr::getExtent(xnext); } + Extent* getPrevExtent() { return xprev.isNull() ? 0 : DataFileMgr::getExtent(xprev); } + static int maxSize(); + static int minSize() { return 0x100; } + /** + * @param len lengt of record we need + * @param lastRecord size of last extent which is a factor in next extent size + */ + static int followupSize(int len, int lastExtentLen); + + /** + * @param len lengt of record we need + */ + static int initialSize(int len); + + struct FL { + DiskLoc firstRecord; + DiskLoc lastRecord; + }; + /** often we want to update just the firstRecord and lastRecord fields. + this helper is for that -- for use with getDur().writing() method + */ + FL* fl() { return (FL*) &firstRecord; } + private: + DiskLoc _reuse(const char *nsname); }; - /* + /* a datafile - i.e. the "dbname.<#>" files : + ---------------------- - Header + DataFileHeader ---------------------- Extent (for a particular namespace) Record @@ -273,7 +309,6 @@ namespace mongo { more Extents... ---------------------- */ - class DataFileHeader { public: int version; @@ -287,35 +322,27 @@ namespace mongo { enum { HeaderSize = 8192 }; - bool currentVersion() const { - return ( version == VERSION ) && ( versionMinor == VERSION_MINOR ); - } - - bool uninitialized() const { - if ( version == 0 ) return true; - return false; - } + bool isCurrentVersion() const { return ( version == VERSION ) && ( versionMinor == VERSION_MINOR ); } - /*Record* __getRecord(DiskLoc dl) { - int ofs = dl.getOfs(); - assert( ofs >= HeaderSize ); - return (Record*) (((char *) this) + ofs); - }*/ + bool uninitialized() const { return version == 0; } - void init(int fileno, int filelength) { + void init(int fileno, int filelength, const char* filename) { if ( uninitialized() ) { - assert(filelength > 32768 ); + if( !(filelength > 32768 ) ) { + massert(13640, str::stream() << "DataFileHeader looks corrupt at file open filelength:" << filelength << " fileno:" << fileno, false); + } + getDur().createdFile(filename, filelength); assert( HeaderSize == 8192 ); - fileLength = filelength; - version = VERSION; - versionMinor = VERSION_MINOR; - unused.setOfs( fileno, HeaderSize ); + DataFileHeader *h = getDur().writing(this); + h->fileLength = filelength; + h->version = VERSION; + h->versionMinor = VERSION_MINOR; + h->unused.set( fileno, HeaderSize ); assert( (data-(char*)this) == HeaderSize ); - unusedLength = fileLength - HeaderSize - 16; - //memcpy(data+unusedLength, " \nthe end\n", 16); + h->unusedLength = fileLength - HeaderSize - 16; } } - + bool isEmpty() const { return uninitialized() || ( unusedLength == fileLength - HeaderSize - 16 ); } @@ -323,13 +350,13 @@ namespace mongo { #pragma pack() - inline Extent* MongoDataFile::_getExtent(DiskLoc loc) { + inline Extent* MongoDataFile::_getExtent(DiskLoc loc) const { loc.assertOk(); - Extent *e = (Extent *) _p.at(loc.getOfs(), Extent::HeaderSize()); + Extent *e = (Extent *) (p()+loc.getOfs()); return e; } - inline Extent* MongoDataFile::getExtent(DiskLoc loc) { + inline Extent* MongoDataFile::getExtent(DiskLoc loc) const { Extent *e = _getExtent(loc); e->assertOk(); return e; @@ -344,18 +371,13 @@ namespace mongo { inline Record* MongoDataFile::recordAt(DiskLoc dl) { int ofs = dl.getOfs(); if( ofs < DataFileHeader::HeaderSize ) badOfs(ofs); // will uassert - external call to keep out of the normal code path - return (Record*) _p.at(ofs, -1); + return (Record*) (p()+ofs); } - inline void MongoDataFile::grow(DiskLoc dl, int size) { - int ofs = dl.getOfs(); - _p.grow(ofs, size); - } - - inline Record* MongoDataFile::makeRecord(DiskLoc dl, int size) { + inline Record* MongoDataFile::makeRecord(DiskLoc dl, int size) { int ofs = dl.getOfs(); - assert( ofs >= DataFileHeader::HeaderSize ); - return (Record*) _p.at(ofs, size); + if( ofs < DataFileHeader::HeaderSize ) badOfs(ofs); // will uassert - external call to keep out of the normal code path + return (Record*) (p()+ofs); } inline DiskLoc Record::getNext(const DiskLoc& myLoc) { @@ -395,50 +417,23 @@ namespace mongo { return BSONObj(rec()); } inline DeletedRecord* DiskLoc::drec() const { - assert( fileNo != -1 ); + assert( _a != -1 ); return (DeletedRecord*) rec(); } inline Extent* DiskLoc::ext() const { return DataFileMgr::getExtent(*this); } - - /*---------------------------------------------------------------------*/ + inline const BtreeBucket* DiskLoc::btree() const { + assert( _a != -1 ); + return (const BtreeBucket *) rec()->data; + } } // namespace mongo -#include "rec.h" #include "database.h" namespace mongo { - // Heritable class to implement an operation that may be applied to all - // files in a database using _applyOpToDataFiles() - class FileOp { - public: - virtual ~FileOp() {} - // Return true if file exists and operation successful - virtual bool apply( const boost::filesystem::path &p ) = 0; - virtual const char * op() const = 0; - }; - - void _applyOpToDataFiles( const char *database, FileOp &fo, bool afterAllocator = false, const string& path = dbpath ); - - inline void _deleteDataFiles(const char *database) { - if ( directoryperdb ) { - BOOST_CHECK_EXCEPTION( boost::filesystem::remove_all( boost::filesystem::path( dbpath ) / database ) ); - return; - } - class : public FileOp { - virtual bool apply( const boost::filesystem::path &p ) { - return boost::filesystem::remove( p ); - } - virtual const char * op() const { - return "remove"; - } - } deleter; - _applyOpToDataFiles( database, deleter, true ); - } - boost::intmax_t dbSize( const char *database ); inline NamespaceIndex* nsindex(const char *ns) { @@ -462,11 +457,6 @@ namespace mongo { return nsindex(ns)->details(ns); } - inline MongoDataFile& DiskLoc::pdf() const { - assert( fileNo != -1 ); - return *cc().database()->getFile(fileNo); - } - inline Extent* DataFileMgr::getExtent(const DiskLoc& dl) { assert( dl.a() != -1 ); return cc().database()->getFile(dl.a())->getExtent(dl); @@ -477,30 +467,30 @@ namespace mongo { return cc().database()->getFile(dl.a())->recordAt(dl); } - BOOST_STATIC_ASSERT( 16 == sizeof(DeletedRecord) ); - - inline void DataFileMgr::grow(const DiskLoc& dl, int len) { - assert( dl.a() != -1 ); - cc().database()->getFile(dl.a())->grow(dl, len); - } + BOOST_STATIC_ASSERT( 16 == sizeof(DeletedRecord) ); - inline DeletedRecord* DataFileMgr::makeDeletedRecord(const DiskLoc& dl, int len) { + inline DeletedRecord* DataFileMgr::makeDeletedRecord(const DiskLoc& dl, int len) { assert( dl.a() != -1 ); return (DeletedRecord*) cc().database()->getFile(dl.a())->makeRecord(dl, sizeof(DeletedRecord)); } - + void ensureHaveIdIndex(const char *ns); - + bool dropIndexes( NamespaceDetails *d, const char *ns, const char *name, string &errmsg, BSONObjBuilder &anObjBuilder, bool maydeleteIdIndex ); /** - * @return true if ns is ok + * @return true if ns is 'normal'. $ used for collections holding index data, which do not contain BSON objects in their records. + * special case for the local.oplog.$main ns -- naming it as such was a mistake. */ - inline bool nsDollarCheck( const char* ns ){ + inline bool isANormalNSName( const char* ns ) { if ( strchr( ns , '$' ) == 0 ) return true; - return strcmp( ns, "local.oplog.$main" ) == 0; } + + inline BSONObj::BSONObj(const Record *r) { + init(r->data, false); + } + } // namespace mongo |