slru.c

postgresql8.3.4源码,开源数据库

/*-------------------------------------------------------------------------
 *
 * slru.c
 *		Simple LRU buffering for transaction status logfiles
 *
 * We use a simple least-recently-used scheme to manage a pool of page
 * buffers.  Under ordinary circumstances we expect that write
 * traffic will occur mostly to the latest page (and to the just-prior
 * page, soon after a page transition).  Read traffic will probably touch
 * a larger span of pages, but in any case a fairly small number of page
 * buffers should be sufficient.  So, we just search the buffers using plain
 * linear search; there's no need for a hashtable or anything fancy.
 * The management algorithm is straight LRU except that we will never swap
 * out the latest page (since we know it's going to be hit again eventually).
 *
 * We use a control LWLock to protect the shared data structures, plus
 * per-buffer LWLocks that synchronize I/O for each buffer.  The control lock
 * must be held to examine or modify any shared state.	A process that is
 * reading in or writing out a page buffer does not hold the control lock,
 * only the per-buffer lock for the buffer it is working on.
 *
 * "Holding the control lock" means exclusive lock in all cases except for
 * SimpleLruReadPage_ReadOnly(); see comments for SlruRecentlyUsed() for
 * the implications of that.
 *
 * When initiating I/O on a buffer, we acquire the per-buffer lock exclusively
 * before releasing the control lock.  The per-buffer lock is released after
 * completing the I/O, re-acquiring the control lock, and updating the shared
 * state.  (Deadlock is not possible here, because we never try to initiate
 * I/O when someone else is already doing I/O on the same buffer.)
 * To wait for I/O to complete, release the control lock, acquire the
 * per-buffer lock in shared mode, immediately release the per-buffer lock,
 * reacquire the control lock, and then recheck state (since arbitrary things
 * could have happened while we didn't have the lock).
 *
 * As with the regular buffer manager, it is possible for another process
 * to re-dirty a page that is currently being written out.	This is handled
 * by re-setting the page's page_dirty flag.
 *
 *
 * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 * $PostgreSQL: pgsql/src/backend/access/transam/slru.c,v 1.44 2008/01/01 19:45:48 momjian Exp $
 *
 *-------------------------------------------------------------------------
 */
#include "postgres.h"

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

#include "access/slru.h"
#include "access/transam.h"
#include "access/xlog.h"
#include "storage/fd.h"
#include "storage/shmem.h"
#include "miscadmin.h"


/*
 * Define segment size.  A page is the same BLCKSZ as is used everywhere
 * else in Postgres.  The segment size can be chosen somewhat arbitrarily;
 * we make it 32 pages by default, or 256Kb, i.e. 1M transactions for CLOG
 * or 64K transactions for SUBTRANS.
 *
 * Note: because TransactionIds are 32 bits and wrap around at 0xFFFFFFFF,
 * page numbering also wraps around at 0xFFFFFFFF/xxxx_XACTS_PER_PAGE (where
 * xxxx is CLOG or SUBTRANS, respectively), and segment numbering at
 * 0xFFFFFFFF/xxxx_XACTS_PER_PAGE/SLRU_PAGES_PER_SEGMENT.  We need
 * take no explicit notice of that fact in this module, except when comparing
 * segment and page numbers in SimpleLruTruncate (see PagePrecedes()).
 *
 * Note: this file currently assumes that segment file names will be four
 * hex digits.	This sets a lower bound on the segment size (64K transactions
 * for 32-bit TransactionIds).
 */
#define SLRU_PAGES_PER_SEGMENT	32

#define SlruFileName(ctl, path, seg) \
	snprintf(path, MAXPGPATH, "%s/%04X", (ctl)->Dir, seg)

/*
 * During SimpleLruFlush(), we will usually not need to write/fsync more
 * than one or two physical files, but we may need to write several pages
 * per file.  We can consolidate the I/O requests by leaving files open
 * until control returns to SimpleLruFlush().  This data structure remembers
 * which files are open.
 */
#define MAX_FLUSH_BUFFERS	16

typedef struct SlruFlushData
{
	int			num_files;		/* # files actually open */
	int			fd[MAX_FLUSH_BUFFERS];	/* their FD's */
	int			segno[MAX_FLUSH_BUFFERS];		/* their log seg#s */
} SlruFlushData;

/*
 * Macro to mark a buffer slot "most recently used".  Note multiple evaluation
 * of arguments!
 *
 * The reason for the if-test is that there are often many consecutive
 * accesses to the same page (particularly the latest page).  By suppressing
 * useless increments of cur_lru_count, we reduce the probability that old
 * pages' counts will "wrap around" and make them appear recently used.
 *
 * We allow this code to be executed concurrently by multiple processes within
 * SimpleLruReadPage_ReadOnly().  As long as int reads and writes are atomic,
 * this should not cause any completely-bogus values to enter the computation.
 * However, it is possible for either cur_lru_count or individual
 * page_lru_count entries to be "reset" to lower values than they should have,
 * in case a process is delayed while it executes this macro.  With care in
 * SlruSelectLRUPage(), this does little harm, and in any case the absolute
 * worst possible consequence is a nonoptimal choice of page to evict.	The
 * gain from allowing concurrent reads of SLRU pages seems worth it.
 */
#define SlruRecentlyUsed(shared, slotno)	\
	do { \
		int		new_lru_count = (shared)->cur_lru_count; \
		if (new_lru_count != (shared)->page_lru_count[slotno]) { \
			(shared)->cur_lru_count = ++new_lru_count; \
			(shared)->page_lru_count[slotno] = new_lru_count; \
		} \
	} while (0)

/* Saved info for SlruReportIOError */
typedef enum
{
	SLRU_OPEN_FAILED,
	SLRU_SEEK_FAILED,
	SLRU_READ_FAILED,
	SLRU_WRITE_FAILED,
	SLRU_FSYNC_FAILED,
	SLRU_CLOSE_FAILED
} SlruErrorCause;

static SlruErrorCause slru_errcause;
static int	slru_errno;


static void SimpleLruZeroLSNs(SlruCtl ctl, int slotno);
static void SimpleLruWaitIO(SlruCtl ctl, int slotno);
static bool SlruPhysicalReadPage(SlruCtl ctl, int pageno, int slotno);
static bool SlruPhysicalWritePage(SlruCtl ctl, int pageno, int slotno,
					  SlruFlush fdata);
static void SlruReportIOError(SlruCtl ctl, int pageno, TransactionId xid);
static int	SlruSelectLRUPage(SlruCtl ctl, int pageno);


/*
 * Initialization of shared memory
 */

Size
SimpleLruShmemSize(int nslots, int nlsns)
{
	Size		sz;

	/* we assume nslots isn't so large as to risk overflow */
	sz = MAXALIGN(sizeof(SlruSharedData));
	sz += MAXALIGN(nslots * sizeof(char *));	/* page_buffer[] */
	sz += MAXALIGN(nslots * sizeof(SlruPageStatus));	/* page_status[] */
	sz += MAXALIGN(nslots * sizeof(bool));		/* page_dirty[] */
	sz += MAXALIGN(nslots * sizeof(int));		/* page_number[] */
	sz += MAXALIGN(nslots * sizeof(int));		/* page_lru_count[] */
	sz += MAXALIGN(nslots * sizeof(LWLockId));	/* buffer_locks[] */

	if (nlsns > 0)
		sz += MAXALIGN(nslots * nlsns * sizeof(XLogRecPtr));	/* group_lsn[] */

	return BUFFERALIGN(sz) + BLCKSZ * nslots;
}

void
SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
			  LWLockId ctllock, const char *subdir)
{
	SlruShared	shared;
	bool		found;

	shared = (SlruShared) ShmemInitStruct(name,
										  SimpleLruShmemSize(nslots, nlsns),
										  &found);

	if (!IsUnderPostmaster)
	{
		/* Initialize locks and shared memory area */
		char	   *ptr;
		Size		offset;
		int			slotno;

		Assert(!found);

		memset(shared, 0, sizeof(SlruSharedData));

		shared->ControlLock = ctllock;

		shared->num_slots = nslots;
		shared->lsn_groups_per_page = nlsns;

		shared->cur_lru_count = 0;

		/* shared->latest_page_number will be set later */

		ptr = (char *) shared;
		offset = MAXALIGN(sizeof(SlruSharedData));
		shared->page_buffer = (char **) (ptr + offset);
		offset += MAXALIGN(nslots * sizeof(char *));
		shared->page_status = (SlruPageStatus *) (ptr + offset);
		offset += MAXALIGN(nslots * sizeof(SlruPageStatus));
		shared->page_dirty = (bool *) (ptr + offset);
		offset += MAXALIGN(nslots * sizeof(bool));
		shared->page_number = (int *) (ptr + offset);
		offset += MAXALIGN(nslots * sizeof(int));
		shared->page_lru_count = (int *) (ptr + offset);
		offset += MAXALIGN(nslots * sizeof(int));
		shared->buffer_locks = (LWLockId *) (ptr + offset);
		offset += MAXALIGN(nslots * sizeof(LWLockId));

		if (nlsns > 0)
		{
			shared->group_lsn = (XLogRecPtr *) (ptr + offset);
			offset += MAXALIGN(nslots * nlsns * sizeof(XLogRecPtr));
		}

		ptr += BUFFERALIGN(offset);
		for (slotno = 0; slotno < nslots; slotno++)
		{
			shared->page_buffer[slotno] = ptr;
			shared->page_status[slotno] = SLRU_PAGE_EMPTY;
			shared->page_dirty[slotno] = false;
			shared->page_lru_count[slotno] = 0;
			shared->buffer_locks[slotno] = LWLockAssign();
			ptr += BLCKSZ;
		}
	}
	else
		Assert(found);

	/*
	 * Initialize the unshared control struct, including directory path. We
	 * assume caller set PagePrecedes.
	 */
	ctl->shared = shared;
	ctl->do_fsync = true;		/* default behavior */
	StrNCpy(ctl->Dir, subdir, sizeof(ctl->Dir));
}

/*
 * Initialize (or reinitialize) a page to zeroes.
 *
 * The page is not actually written, just set up in shared memory.
 * The slot number of the new page is returned.
 *
 * Control lock must be held at entry, and will be held at exit.
 */
int
SimpleLruZeroPage(SlruCtl ctl, int pageno)
{
	SlruShared	shared = ctl->shared;
	int			slotno;

	/* Find a suitable buffer slot for the page */
	slotno = SlruSelectLRUPage(ctl, pageno);
	Assert(shared->page_status[slotno] == SLRU_PAGE_EMPTY ||
		   (shared->page_status[slotno] == SLRU_PAGE_VALID &&
			!shared->page_dirty[slotno]) ||
		   shared->page_number[slotno] == pageno);

	/* Mark the slot as containing this page */
	shared->page_number[slotno] = pageno;
	shared->page_status[slotno] = SLRU_PAGE_VALID;
	shared->page_dirty[slotno] = true;
	SlruRecentlyUsed(shared, slotno);

	/* Set the buffer to zeroes */
	MemSet(shared->page_buffer[slotno], 0, BLCKSZ);

	/* Set the LSNs for this new page to zero */
	SimpleLruZeroLSNs(ctl, slotno);

	/* Assume this page is now the latest active page */
	shared->latest_page_number = pageno;

	return slotno;
}

/*
 * Zero all the LSNs we store for this slru page.
 *
 * This should be called each time we create a new page, and each time we read
 * in a page from disk into an existing buffer.  (Such an old page cannot
 * have any interesting LSNs, since we'd have flushed them before writing
 * the page in the first place.)
 */
static void
SimpleLruZeroLSNs(SlruCtl ctl, int slotno)
{
	SlruShared	shared = ctl->shared;

	if (shared->lsn_groups_per_page > 0)
		MemSet(&shared->group_lsn[slotno * shared->lsn_groups_per_page], 0,
			   shared->lsn_groups_per_page * sizeof(XLogRecPtr));
}

/*
 * Wait for any active I/O on a page slot to finish.  (This does not
 * guarantee that new I/O hasn't been started before we return, though.
 * In fact the slot might not even contain the same page anymore.)
 *
 * Control lock must be held at entry, and will be held at exit.
 */
static void
SimpleLruWaitIO(SlruCtl ctl, int slotno)
{
	SlruShared	shared = ctl->shared;

	/* See notes at top of file */
	LWLockRelease(shared->ControlLock);
	LWLockAcquire(shared->buffer_locks[slotno], LW_SHARED);
	LWLockRelease(shared->buffer_locks[slotno]);
	LWLockAcquire(shared->ControlLock, LW_EXCLUSIVE);

	/*
	 * If the slot is still in an io-in-progress state, then either someone
	 * already started a new I/O on the slot, or a previous I/O failed and
	 * neglected to reset the page state.  That shouldn't happen, really, but
	 * it seems worth a few extra cycles to check and recover from it. We can
	 * cheaply test for failure by seeing if the buffer lock is still held (we
	 * assume that transaction abort would release the lock).
	 */
	if (shared->page_status[slotno] == SLRU_PAGE_READ_IN_PROGRESS ||
		shared->page_status[slotno] == SLRU_PAGE_WRITE_IN_PROGRESS)
	{
		if (LWLockConditionalAcquire(shared->buffer_locks[slotno], LW_SHARED))
		{
			/* indeed, the I/O must have failed */
			if (shared->page_status[slotno] == SLRU_PAGE_READ_IN_PROGRESS)
				shared->page_status[slotno] = SLRU_PAGE_EMPTY;
			else	/* write_in_progress */
			{
				shared->page_status[slotno] = SLRU_PAGE_VALID;
				shared->page_dirty[slotno] = true;
			}
			LWLockRelease(shared->buffer_locks[slotno]);
		}
	}
}

/*
 * Find a page in a shared buffer, reading it in if necessary.
 * The page number must correspond to an already-initialized page.
 *
 * If write_ok is true then it is OK to return a page that is in
 * WRITE_IN_PROGRESS state; it is the caller's responsibility to be sure
 * that modification of the page is safe.  If write_ok is false then we
 * will not return the page until it is not undergoing active I/O.
 *
 * The passed-in xid is used only for error reporting, and may be
 * InvalidTransactionId if no specific xid is associated with the action.
 *
 * Return value is the shared-buffer slot number now holding the page.
 * The buffer's LRU access info is updated.
 *
 * Control lock must be held at entry, and will be held at exit.
 */
int
SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok,
				  TransactionId xid)
{
	SlruShared	shared = ctl->shared;

	/* Outer loop handles restart if we must wait for someone else's I/O */
	for (;;)
	{
		int			slotno;
		bool		ok;

		/* See if page already is in memory; if not, pick victim slot */
		slotno = SlruSelectLRUPage(ctl, pageno);

		/* Did we find the page in memory? */
		if (shared->page_number[slotno] == pageno &&
			shared->page_status[slotno] != SLRU_PAGE_EMPTY)
		{
			/*
			 * If page is still being read in, we must wait for I/O.  Likewise
			 * if the page is being written and the caller said that's not OK.
			 */
			if (shared->page_status[slotno] == SLRU_PAGE_READ_IN_PROGRESS ||
				(shared->page_status[slotno] == SLRU_PAGE_WRITE_IN_PROGRESS &&
				 !write_ok))
			{
				SimpleLruWaitIO(ctl, slotno);
				/* Now we must recheck state from the top */
				continue;
			}