md.c

postgresql8.3.4源码,开源数据库

/*-------------------------------------------------------------------------
 *
 * md.c
 *	  This code manages relations that reside on magnetic disk.
 *
 * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 *
 * IDENTIFICATION
 *	  $PostgreSQL: pgsql/src/backend/storage/smgr/md.c,v 1.135.2.1 2008/04/18 06:48:50 heikki Exp $
 *
 *-------------------------------------------------------------------------
 */
#include "postgres.h"

#include <unistd.h>
#include <fcntl.h>
#include <sys/file.h>

#include "catalog/catalog.h"
#include "miscadmin.h"
#include "postmaster/bgwriter.h"
#include "storage/fd.h"
#include "storage/bufmgr.h"
#include "storage/smgr.h"
#include "utils/hsearch.h"
#include "utils/memutils.h"


/* interval for calling AbsorbFsyncRequests in mdsync */
#define FSYNCS_PER_ABSORB		10

/* special values for the segno arg to RememberFsyncRequest */
#define FORGET_RELATION_FSYNC	(InvalidBlockNumber)
#define FORGET_DATABASE_FSYNC	(InvalidBlockNumber-1)
#define UNLINK_RELATION_REQUEST (InvalidBlockNumber-2)

/*
 * On Windows, we have to interpret EACCES as possibly meaning the same as
 * ENOENT, because if a file is unlinked-but-not-yet-gone on that platform,
 * that's what you get.  Ugh.  This code is designed so that we don't
 * actually believe these cases are okay without further evidence (namely,
 * a pending fsync request getting revoked ... see mdsync).
 */
#ifndef WIN32
#define FILE_POSSIBLY_DELETED(err)	((err) == ENOENT)
#else
#define FILE_POSSIBLY_DELETED(err)	((err) == ENOENT || (err) == EACCES)
#endif

/*
 *	The magnetic disk storage manager keeps track of open file
 *	descriptors in its own descriptor pool.  This is done to make it
 *	easier to support relations that are larger than the operating
 *	system's file size limit (often 2GBytes).  In order to do that,
 *	we break relations up into "segment" files that are each shorter than
 *	the OS file size limit.  The segment size is set by the RELSEG_SIZE
 *	configuration constant in pg_config_manual.h.
 *
 *	On disk, a relation must consist of consecutively numbered segment
 *	files in the pattern
 *		-- Zero or more full segments of exactly RELSEG_SIZE blocks each
 *		-- Exactly one partial segment of size 0 <= size < RELSEG_SIZE blocks
 *		-- Optionally, any number of inactive segments of size 0 blocks.
 *	The full and partial segments are collectively the "active" segments.
 *	Inactive segments are those that once contained data but are currently
 *	not needed because of an mdtruncate() operation.  The reason for leaving
 *	them present at size zero, rather than unlinking them, is that other
 *	backends and/or the bgwriter might be holding open file references to
 *	such segments.	If the relation expands again after mdtruncate(), such
 *	that a deactivated segment becomes active again, it is important that
 *	such file references still be valid --- else data might get written
 *	out to an unlinked old copy of a segment file that will eventually
 *	disappear.
 *
 *	The file descriptor pointer (md_fd field) stored in the SMgrRelation
 *	cache is, therefore, just the head of a list of MdfdVec objects, one
 *	per segment.  But note the md_fd pointer can be NULL, indicating
 *	relation not open.
 *
 *	Also note that mdfd_chain == NULL does not necessarily mean the relation
 *	doesn't have another segment after this one; we may just not have
 *	opened the next segment yet.  (We could not have "all segments are
 *	in the chain" as an invariant anyway, since another backend could
 *	extend the relation when we weren't looking.)  We do not make chain
 *	entries for inactive segments, however; as soon as we find a partial
 *	segment, we assume that any subsequent segments are inactive.
 *
 *	All MdfdVec objects are palloc'd in the MdCxt memory context.
 *
 *	Defining LET_OS_MANAGE_FILESIZE disables the segmentation logic,
 *	for use on machines that support large files.  Beware that that
 *	code has not been tested in a long time and is probably bit-rotted.
 */

typedef struct _MdfdVec
{
	File		mdfd_vfd;		/* fd number in fd.c's pool */
	BlockNumber mdfd_segno;		/* segment number, from 0 */
#ifndef LET_OS_MANAGE_FILESIZE	/* for large relations */
	struct _MdfdVec *mdfd_chain;	/* next segment, or NULL */
#endif
} MdfdVec;

static MemoryContext MdCxt;		/* context for all md.c allocations */


/*
 * In some contexts (currently, standalone backends and the bgwriter process)
 * we keep track of pending fsync operations: we need to remember all relation
 * segments that have been written since the last checkpoint, so that we can
 * fsync them down to disk before completing the next checkpoint.  This hash
 * table remembers the pending operations.	We use a hash table mostly as
 * a convenient way of eliminating duplicate requests.
 *
 * We use a similar mechanism to remember no-longer-needed files that can
 * be deleted after the next checkpoint, but we use a linked list instead of
 * a hash table, because we don't expect there to be any duplicate requests.
 *
 * (Regular backends do not track pending operations locally, but forward
 * them to the bgwriter.)
 */
typedef struct
{
	RelFileNode rnode;			/* the targeted relation */
	BlockNumber segno;			/* which segment */
} PendingOperationTag;

typedef uint16 CycleCtr;		/* can be any convenient integer size */

typedef struct
{
	PendingOperationTag tag;	/* hash table key (must be first!) */
	bool		canceled;		/* T => request canceled, not yet removed */
	CycleCtr	cycle_ctr;		/* mdsync_cycle_ctr when request was made */
} PendingOperationEntry;

typedef struct
{
	RelFileNode rnode;			/* the dead relation to delete */
	CycleCtr	cycle_ctr;		/* mdckpt_cycle_ctr when request was made */
} PendingUnlinkEntry;

static HTAB *pendingOpsTable = NULL;
static List *pendingUnlinks = NIL;

static CycleCtr mdsync_cycle_ctr = 0;
static CycleCtr mdckpt_cycle_ctr = 0;


typedef enum					/* behavior for mdopen & _mdfd_getseg */
{
	EXTENSION_FAIL,				/* ereport if segment not present */
	EXTENSION_RETURN_NULL,		/* return NULL if not present */
	EXTENSION_CREATE			/* create new segments as needed */
} ExtensionBehavior;

/* local routines */
static MdfdVec *mdopen(SMgrRelation reln, ExtensionBehavior behavior);
static void register_dirty_segment(SMgrRelation reln, MdfdVec *seg);
static void register_unlink(RelFileNode rnode);
static MdfdVec *_fdvec_alloc(void);

#ifndef LET_OS_MANAGE_FILESIZE
static MdfdVec *_mdfd_openseg(SMgrRelation reln, BlockNumber segno,
			  int oflags);
#endif
static MdfdVec *_mdfd_getseg(SMgrRelation reln, BlockNumber blkno,
			 bool isTemp, ExtensionBehavior behavior);
static BlockNumber _mdnblocks(SMgrRelation reln, MdfdVec *seg);


/*
 *	mdinit() -- Initialize private state for magnetic disk storage manager.
 */
void
mdinit(void)
{
	MdCxt = AllocSetContextCreate(TopMemoryContext,
								  "MdSmgr",
								  ALLOCSET_DEFAULT_MINSIZE,
								  ALLOCSET_DEFAULT_INITSIZE,
								  ALLOCSET_DEFAULT_MAXSIZE);

	/*
	 * Create pending-operations hashtable if we need it.  Currently, we need
	 * it if we are standalone (not under a postmaster) OR if we are a
	 * bootstrap-mode subprocess of a postmaster (that is, a startup or
	 * bgwriter process).
	 */
	if (!IsUnderPostmaster || IsBootstrapProcessingMode())
	{
		HASHCTL		hash_ctl;

		MemSet(&hash_ctl, 0, sizeof(hash_ctl));
		hash_ctl.keysize = sizeof(PendingOperationTag);
		hash_ctl.entrysize = sizeof(PendingOperationEntry);
		hash_ctl.hash = tag_hash;
		hash_ctl.hcxt = MdCxt;
		pendingOpsTable = hash_create("Pending Ops Table",
									  100L,
									  &hash_ctl,
								   HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);
		pendingUnlinks = NIL;
	}
}

/*
 *	mdcreate() -- Create a new relation on magnetic disk.
 *
 * If isRedo is true, it's okay for the relation to exist already.
 */
void
mdcreate(SMgrRelation reln, bool isRedo)
{
	char	   *path;
	File		fd;

	if (isRedo && reln->md_fd != NULL)
		return;					/* created and opened already... */

	Assert(reln->md_fd == NULL);

	path = relpath(reln->smgr_rnode);

	fd = PathNameOpenFile(path, O_RDWR | O_CREAT | O_EXCL | PG_BINARY, 0600);

	if (fd < 0)
	{
		int			save_errno = errno;

		/*
		 * During bootstrap, there are cases where a system relation will be
		 * accessed (by internal backend processes) before the bootstrap
		 * script nominally creates it.  Therefore, allow the file to exist
		 * already, even if isRedo is not set.	(See also mdopen)
		 */
		if (isRedo || IsBootstrapProcessingMode())
			fd = PathNameOpenFile(path, O_RDWR | PG_BINARY, 0600);
		if (fd < 0)
		{
			pfree(path);
			/* be sure to report the error reported by create, not open */
			errno = save_errno;
			ereport(ERROR,
					(errcode_for_file_access(),
					 errmsg("could not create relation %u/%u/%u: %m",
							reln->smgr_rnode.spcNode,
							reln->smgr_rnode.dbNode,
							reln->smgr_rnode.relNode)));
		}
	}

	pfree(path);

	reln->md_fd = _fdvec_alloc();

	reln->md_fd->mdfd_vfd = fd;
	reln->md_fd->mdfd_segno = 0;
#ifndef LET_OS_MANAGE_FILESIZE
	reln->md_fd->mdfd_chain = NULL;
#endif
}

/*
 *	mdunlink() -- Unlink a relation.
 *
 * Note that we're passed a RelFileNode --- by the time this is called,
 * there won't be an SMgrRelation hashtable entry anymore.
 *
 * Actually, we don't unlink the first segment file of the relation, but
 * just truncate it to zero length, and record a request to unlink it after
 * the next checkpoint.  Additional segments can be unlinked immediately,
 * however.  Leaving the empty file in place prevents that relfilenode
 * number from being reused.  The scenario this protects us from is:
 * 1. We delete a relation (and commit, and actually remove its file).
 * 2. We create a new relation, which by chance gets the same relfilenode as
 *	  the just-deleted one (OIDs must've wrapped around for that to happen).
 * 3. We crash before another checkpoint occurs.
 * During replay, we would delete the file and then recreate it, which is fine
 * if the contents of the file were repopulated by subsequent WAL entries.
 * But if we didn't WAL-log insertions, but instead relied on fsyncing the
 * file after populating it (as for instance CLUSTER and CREATE INDEX do),
 * the contents of the file would be lost forever.	By leaving the empty file
 * until after the next checkpoint, we prevent reassignment of the relfilenode
 * number until it's safe, because relfilenode assignment skips over any
 * existing file.
 *
 * If isRedo is true, it's okay for the relation to be already gone.
 * Also, we should remove the file immediately instead of queuing a request
 * for later, since during redo there's no possibility of creating a
 * conflicting relation.
 *
 * Note: any failure should be reported as WARNING not ERROR, because
 * we are usually not in a transaction anymore when this is called.
 */
void
mdunlink(RelFileNode rnode, bool isRedo)
{
	char	   *path;
	int			ret;

	/*
	 * We have to clean out any pending fsync requests for the doomed
	 * relation, else the next mdsync() will fail.
	 */
	ForgetRelationFsyncRequests(rnode);

	path = relpath(rnode);

	/*
	 * Delete or truncate the first segment, or only segment if not doing
	 * segmenting
	 */
	if (isRedo)
		ret = unlink(path);
	else
	{
		/* truncate(2) would be easier here, but Windows hasn't got it */
		int			fd;

		fd = BasicOpenFile(path, O_RDWR | PG_BINARY, 0);
		if (fd >= 0)
		{
			int			save_errno;

			ret = ftruncate(fd, 0);
			save_errno = errno;
			close(fd);
			errno = save_errno;
		}
		else
			ret = -1;
	}
	if (ret < 0)
	{
		if (!isRedo || errno != ENOENT)
			ereport(WARNING,
					(errcode_for_file_access(),
					 errmsg("could not remove relation %u/%u/%u: %m",
							rnode.spcNode,
							rnode.dbNode,
							rnode.relNode)));
	}

#ifndef LET_OS_MANAGE_FILESIZE
	/* Delete the additional segments, if any */
	else
	{
		char	   *segpath = (char *) palloc(strlen(path) + 12);
		BlockNumber segno;

		/*
		 * Note that because we loop until getting ENOENT, we will correctly
		 * remove all inactive segments as well as active ones.
		 */
		for (segno = 1;; segno++)
		{
			sprintf(segpath, "%s.%u", path, segno);
			if (unlink(segpath) < 0)
			{
				/* ENOENT is expected after the last segment... */
				if (errno != ENOENT)
					ereport(WARNING,
							(errcode_for_file_access(),
							 errmsg("could not remove segment %u of relation %u/%u/%u: %m",
									segno,
									rnode.spcNode,
									rnode.dbNode,
									rnode.relNode)));
				break;
			}
		}
		pfree(segpath);
	}
#endif

	pfree(path);

	/* Register request to unlink first segment later */
	if (!isRedo)
		register_unlink(rnode);
}

/*
 *	mdextend() -- Add a block to the specified relation.
 *
 *		The semantics are nearly the same as mdwrite(): write at the
 *		specified position.  However, this is to be used for the case of
 *		extending a relation (i.e., blocknum is at or beyond the current
 *		EOF).  Note that we assume writing a block beyond current EOF
 *		causes intervening file space to become filled with zeroes.
 */
void
mdextend(SMgrRelation reln, BlockNumber blocknum, char *buffer, bool isTemp)
{
	long		seekpos;
	int			nbytes;
	MdfdVec    *v;

	/* This assert is too expensive to have on normally ... */
#ifdef CHECK_WRITE_VS_EXTEND
	Assert(blocknum >= mdnblocks(reln));
#endif

	/*