/*
* Copyright 2000, International Business Machines Corporation and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
#include <afsconfig.h>
#include <afs/param.h>
-RCSID
- ("$Header$");
+#include <roken.h>
+#include <afs/opr.h>
-#include <sys/types.h>
-#ifdef AFS_NT40_ENV
-#include <winsock2.h>
+#ifdef AFS_PTHREAD_ENV
+# include <opr/lock.h>
#else
-#include <sys/file.h>
-#include <netinet/in.h>
+# include <opr/lockstub.h>
#endif
-#include <errno.h>
-#include <string.h>
-#include <lock.h>
-#include <rx/xdr.h>
-
-
+#include <afs/afsutil.h>
#define UBIK_INTERNALS
#include "ubik.h"
#define PHSIZE 128
static struct buffer {
- struct ubik_dbase *dbase; /* dbase within which the buffer resides */
- afs_int32 file; /* Unique cache key */
- afs_int32 page; /* page number */
+ struct ubik_dbase *dbase; /*!< dbase within which the buffer resides */
+ afs_int32 file; /*!< Unique cache key */
+ afs_int32 page; /*!< page number */
struct buffer *lru_next;
struct buffer *lru_prev;
- struct buffer *hashNext; /* next dude in hash table */
- char *data; /* ptr to the data */
- char lockers; /* usage ref count */
- char dirty; /* is buffer modified */
- char hashIndex; /* back ptr to hash table */
+ struct buffer *hashNext; /*!< next dude in hash table */
+ char *data; /*!< ptr to the data */
+ char lockers; /*!< usage ref count */
+ char dirty; /*!< is buffer modified */
+ char hashIndex; /*!< back ptr to hash table */
} *Buffers;
#define pHash(page) ((page) & (PHSIZE-1))
afs_int32 ubik_nBuffers = NBUFFERS;
-static struct buffer *phTable[PHSIZE]; /* page hash table */
+static struct buffer *phTable[PHSIZE]; /*!< page hash table */
static struct buffer *LruBuffer;
static int nbuffers;
static int calls = 0, ios = 0, lastb = 0;
static char *BufferData;
static struct buffer *newslot(struct ubik_dbase *adbase, afs_int32 afid,
afs_int32 apage);
-static initd = 0;
#define BADFID 0xffffffff
-static DTrunc(struct ubik_dbase *dbase, afs_int32 fid, afs_int32 length);
+static int DTrunc(struct ubik_trans *atrans, afs_int32 fid, afs_int32 length);
static struct ubik_trunc *freeTruncList = 0;
-/* remove a transaction from the database's active transaction list. Don't free it */
+/*!
+ * \brief Remove a transaction from the database's active transaction list. Don't free it.
+ */
static int
unthread(struct ubik_trans *atrans)
{
return 2; /* no entry */
}
-/* some debugging assistance */
-int
+/*!
+ * \brief some debugging assistance
+ */
+void
udisk_Debug(struct ubik_debug *aparm)
{
struct buffer *tb;
aparm->writeLockedPages++;
}
}
- return 0;
}
-/* log format is defined here, and implicitly in recovery.c
+/*!
+ * \brief Write an opcode to the log.
+ *
+ * log format is defined here, and implicitly in recovery.c
*
* 4 byte opcode, followed by parameters, each 4 bytes long. All integers
* are in logged in network standard byte order, in case we want to move logs
* from machine-to-machine someday.
*
- * Begin transaction: opcode
- * Commit transaction: opcode, version (8 bytes)
- * Truncate file: opcode, file number, length
- * Abort transaction: opcode
- * Write data: opcode, file, position, length, <length> data bytes
+ * Begin transaction: opcode \n
+ * Commit transaction: opcode, version (8 bytes) \n
+ * Truncate file: opcode, file number, length \n
+ * Abort transaction: opcode \n
+ * Write data: opcode, file, position, length, <length> data bytes \n
*/
-
-/* write an opcode to the log */
-int
+static int
udisk_LogOpcode(struct ubik_dbase *adbase, afs_int32 aopcode, int async)
{
- struct ubik_stat ustat;
afs_int32 code;
- /* figure out where to write */
- code = (*adbase->stat) (adbase, LOGFILE, &ustat);
- if (code < 0)
- return code;
-
/* setup data and do write */
aopcode = htonl(aopcode);
- code =
- (*adbase->write) (adbase, LOGFILE, (char *)&aopcode, ustat.size,
- sizeof(afs_int32));
+ code = (*adbase->buffered_append)(adbase, LOGFILE, &aopcode, sizeof(afs_int32));
if (code != sizeof(afs_int32))
return UIOERROR;
return code;
}
-/* log a commit, never syncing */
-int
+/*!
+ * \brief Log a commit, never syncing.
+ */
+static int
udisk_LogEnd(struct ubik_dbase *adbase, struct ubik_version *aversion)
{
afs_int32 code;
afs_int32 data[3];
- struct ubik_stat ustat;
-
- /* figure out where to write */
- code = (*adbase->stat) (adbase, LOGFILE, &ustat);
- if (code)
- return code;
/* setup data */
data[0] = htonl(LOGEND);
/* do write */
code =
- (*adbase->write) (adbase, LOGFILE, (char *)data, ustat.size,
- 3 * sizeof(afs_int32));
+ (*adbase->buffered_append)(adbase, LOGFILE, data, 3 * sizeof(afs_int32));
if (code != 3 * sizeof(afs_int32))
return UIOERROR;
return code;
}
-/* log a truncate operation, never syncing */
-int
+/*!
+ * \brief Log a truncate operation, never syncing.
+ */
+static int
udisk_LogTruncate(struct ubik_dbase *adbase, afs_int32 afile,
afs_int32 alength)
{
afs_int32 code;
afs_int32 data[3];
- struct ubik_stat ustat;
-
- /* figure out where to write */
- code = (*adbase->stat) (adbase, LOGFILE, &ustat);
- if (code < 0)
- return code;
/* setup data */
data[0] = htonl(LOGTRUNCATE);
/* do write */
code =
- (*adbase->write) (adbase, LOGFILE, (char *)data, ustat.size,
- 3 * sizeof(afs_int32));
+ (*adbase->buffered_append)(adbase, LOGFILE, data, 3 * sizeof(afs_int32));
if (code != 3 * sizeof(afs_int32))
return UIOERROR;
return 0;
}
-/* write some data to the log, never syncing */
-int
-udisk_LogWriteData(struct ubik_dbase *adbase, afs_int32 afile, char *abuffer,
+/*!
+ * \brief Write some data to the log, never syncing.
+ */
+static int
+udisk_LogWriteData(struct ubik_dbase *adbase, afs_int32 afile, void *abuffer,
afs_int32 apos, afs_int32 alen)
{
- struct ubik_stat ustat;
afs_int32 code;
afs_int32 data[4];
- afs_int32 lpos;
-
- /* find end of log */
- code = (*adbase->stat) (adbase, LOGFILE, &ustat);
- lpos = ustat.size;
- if (code < 0)
- return code;
/* setup header */
data[0] = htonl(LOGDATA);
/* write header */
code =
- (*adbase->write) (adbase, LOGFILE, (char *)data, lpos, 4 * sizeof(afs_int32));
+ (*adbase->buffered_append)(adbase, LOGFILE, data, 4 * sizeof(afs_int32));
if (code != 4 * sizeof(afs_int32))
return UIOERROR;
- lpos += 4 * sizeof(afs_int32);
/* write data */
- code = (*adbase->write) (adbase, LOGFILE, abuffer, lpos, alen);
+ code = (*adbase->buffered_append)(adbase, LOGFILE, abuffer, alen);
if (code != alen)
return UIOERROR;
return 0;
}
-static int
-DInit(int abuffers)
+int
+udisk_Init(int abuffers)
{
/* Initialize the venus buffer system. */
int i;
struct buffer *tb;
- Buffers = (struct buffer *)malloc(abuffers * sizeof(struct buffer));
- memset(Buffers, 0, abuffers * sizeof(struct buffer));
- BufferData = (char *)malloc(abuffers * UBIK_PAGESIZE);
+ Buffers = calloc(abuffers, sizeof(struct buffer));
+ BufferData = malloc(abuffers * UBIK_PAGESIZE);
nbuffers = abuffers;
for (i = 0; i < PHSIZE; i++)
phTable[i] = 0;
return 0;
}
-/* Take a buffer and mark it as the least recently used buffer */
+/*!
+ * \brief Take a buffer and mark it as the least recently used buffer.
+ */
static void
Dlru(struct buffer *abuf)
{
LruBuffer = abuf;
}
-/* Take a buffer and mark it as the most recently used buffer */
+/*!
+ * \brief Take a buffer and mark it as the most recently used buffer.
+ */
static void
Dmru(struct buffer *abuf)
{
LruBuffer->lru_prev = abuf;
}
-/* get a pointer to a particular buffer */
+static_inline int
+MatchBuffer(struct buffer *buf, int page, afs_int32 fid,
+ struct ubik_trans *atrans)
+{
+ if (buf->page != page) {
+ return 0;
+ }
+ if (buf->file != fid) {
+ return 0;
+ }
+ if (atrans->type == UBIK_READTRANS && buf->dirty) {
+ /* if 'buf' is dirty, it has uncommitted changes; we do not want to
+ * see uncommitted changes if we are a read transaction, so skip over
+ * it. */
+ return 0;
+ }
+ if (buf->dbase != atrans->dbase) {
+ return 0;
+ }
+ return 1;
+}
+
+/*!
+ * \brief Get a pointer to a particular buffer.
+ */
static char *
-DRead(struct ubik_dbase *dbase, afs_int32 fid, int page)
+DRead(struct ubik_trans *atrans, afs_int32 fid, int page)
{
/* Read a page from the disk. */
- struct buffer *tb, *lastbuffer;
+ struct buffer *tb, *lastbuffer, *found_tb = NULL;
afs_int32 code;
+ struct ubik_dbase *dbase = atrans->dbase;
calls++;
lastbuffer = LruBuffer->lru_prev;
- if ((lastbuffer->page == page) && (lastbuffer->file == fid)
- && (lastbuffer->dbase == dbase)) {
+ /* Skip for write transactions for a clean page - this may not be the right page to use */
+ if (MatchBuffer(lastbuffer, page, fid, atrans)
+ && (atrans->type == UBIK_READTRANS || lastbuffer->dirty)) {
tb = lastbuffer;
tb->lockers++;
lastb++;
return tb->data;
}
for (tb = phTable[pHash(page)]; tb; tb = tb->hashNext) {
- if (tb->page == page && tb->file == fid && tb->dbase == dbase) {
- Dmru(tb);
- tb->lockers++;
- return tb->data;
+ if (MatchBuffer(tb, page, fid, atrans)) {
+ if (tb->dirty || atrans->type == UBIK_READTRANS) {
+ found_tb = tb;
+ break;
+ }
+ /* Remember this clean page - we might use it */
+ found_tb = tb;
}
}
+ /* For a write transaction, use a matching clean page if no dirty one was found */
+ if (found_tb) {
+ Dmru(found_tb);
+ found_tb->lockers++;
+ return found_tb->data;
+ }
+
/* can't find it */
tb = newslot(dbase, fid, page);
if (!tb)
tb->file = BADFID;
Dlru(tb);
tb->lockers--;
- ubik_print("Ubik: Error reading database file: errno=%d\n", errno);
+ ViceLog(0, ("Ubik: Error reading database file: errno=%d\n", errno));
return 0;
}
ios++;
return tb->data;
}
-/* zap truncated pages */
+/*!
+ * \brief Zap truncated pages.
+ */
static int
-DTrunc(struct ubik_dbase *dbase, afs_int32 fid, afs_int32 length)
+DTrunc(struct ubik_trans *atrans, afs_int32 fid, afs_int32 length)
{
afs_int32 maxPage;
struct buffer *tb;
int i;
+ struct ubik_dbase *dbase = atrans->dbase;
maxPage = (length + UBIK_PAGESIZE - 1) >> UBIK_LOGPAGESIZE; /* first invalid page now in file */
for (i = 0, tb = Buffers; i < nbuffers; i++, tb++) {
return 0;
}
-/* allocate a truncation entry. We allocate special entries representing truncations, rather than
- performing them immediately, so that we can abort a transaction easily by simply purging
- the in-core memory buffers and discarding these truncation entries.
-*/
+/*!
+ * \brief Allocate a truncation entry.
+ *
+ * We allocate special entries representing truncations, rather than
+ * performing them immediately, so that we can abort a transaction easily by simply purging
+ * the in-core memory buffers and discarding these truncation entries.
+ */
static struct ubik_trunc *
GetTrunc(void)
{
struct ubik_trunc *tt;
if (!freeTruncList) {
- freeTruncList =
- (struct ubik_trunc *)malloc(sizeof(struct ubik_trunc));
+ freeTruncList = malloc(sizeof(struct ubik_trunc));
freeTruncList->next = (struct ubik_trunc *)0;
}
tt = freeTruncList;
return tt;
}
-/* free a truncation entry */
+/*!
+ * \brief Free a truncation entry.
+ */
static int
PutTrunc(struct ubik_trunc *at)
{
return 0;
}
-/* find a truncation entry for a file, if any */
+/*!
+ * \brief Find a truncation entry for a file, if any.
+ */
static struct ubik_trunc *
FindTrunc(struct ubik_trans *atrans, afs_int32 afile)
{
return (struct ubik_trunc *)0;
}
-/* do truncates associated with trans, and free them */
+/*!
+ * \brief Do truncates associated with \p atrans, and free them.
+ */
static int
DoTruncs(struct ubik_trans *atrans)
{
struct ubik_trunc *tt, *nt;
- int (*tproc) ();
+ int (*tproc) (struct ubik_dbase *, afs_int32, afs_int32);
afs_int32 rcode = 0, code;
tproc = atrans->dbase->truncate;
for (tt = atrans->activeTruncs; tt; tt = nt) {
nt = tt->next;
- DTrunc(atrans->dbase, tt->file, tt->length); /* zap pages from buffer cache */
+ DTrunc(atrans, tt->file, tt->length); /* zap pages from buffer cache */
code = (*tproc) (atrans->dbase, tt->file, tt->length);
if (code)
rcode = code;
return (rcode);
}
-/* mark a fid as invalid */
+/*!
+ * \brief Mark an \p fid as invalid.
+ */
int
udisk_Invalidate(struct ubik_dbase *adbase, afs_int32 afid)
{
return 0;
}
-/* move this page into the correct hash bucket */
+/*!
+ * \brief Move this page into the correct hash bucket.
+ */
static int
FixupBucket(struct buffer *ap)
{
return 0;
}
-/* create a new slot for a particular dbase page */
+/*!
+ * \brief Create a new slot for a particular dbase page.
+ */
static struct buffer *
newslot(struct ubik_dbase *adbase, afs_int32 afid, afs_int32 apage)
{
if (pp == 0) {
/* There are no unlocked buffers that don't need to be written to the disk. */
- ubik_print
- ("Ubik: Internal Error: Unable to find free buffer in ubik cache\n");
+ ViceLog(0, ("Ubik: Internal Error: Unable to find free buffer in ubik cache\n"));
return NULL;
}
return pp;
}
-/* Release a buffer, specifying whether or not the buffer has been modified by the locker. */
+/*!
+ * \brief Release a buffer, specifying whether or not the buffer has been modified by the locker.
+ */
static void
DRelease(char *ap, int flag)
{
return;
}
-/* flush all modified buffers, leaves dirty bits set (they're cleared
- * by DSync). Note interaction with DSync: you call this thing first,
- * writing the buffers to the disk. Then you call DSync to sync all the
+/*!
+ * \brief Flush all modified buffers, leaves dirty bits set (they're cleared
+ * by DSync()).
+ *
+ * \note Note interaction with DSync(): you call this thing first,
+ * writing the buffers to the disk. Then you call DSync() to sync all the
* files that were written, and to clear the dirty bits. You should
* always call DFlush/DSync as a pair.
*/
static int
-DFlush(struct ubik_dbase *adbase)
+DFlush(struct ubik_trans *atrans)
{
int i;
afs_int32 code;
struct buffer *tb;
+ struct ubik_dbase *adbase = atrans->dbase;
tb = Buffers;
for (i = 0; i < nbuffers; i++, tb++) {
return 0;
}
-/* flush all modified buffers */
+/*!
+ * \brief Flush all modified buffers.
+ */
static int
-DAbort(struct ubik_dbase *adbase)
+DAbort(struct ubik_trans *atrans)
{
int i;
struct buffer *tb;
return 0;
}
-/* must only be called after DFlush, due to its interpretation of dirty flag */
+/**
+ * Invalidate any buffers that are duplicates of abuf. Duplicate buffers
+ * can appear if a read transaction reads a page that is dirty, then that
+ * dirty page is synced. The read transaction will skip over the dirty page,
+ * and create a new buffer, and when the dirty page is synced, it will be
+ * identical (except for contents) to the read-transaction buffer.
+ */
+static void
+DedupBuffer(struct buffer *abuf)
+{
+ struct buffer *tb;
+ for (tb = phTable[pHash(abuf->page)]; tb; tb = tb->hashNext) {
+ if (tb->page == abuf->page && tb != abuf && tb->file == abuf->file
+ && tb->dbase == abuf->dbase) {
+
+ tb->file = BADFID;
+ Dlru(tb);
+ }
+ }
+}
+
+/*!
+ * \attention DSync() must only be called after DFlush(), due to its interpretation of dirty flag.
+ */
static int
-DSync(struct ubik_dbase *adbase)
+DSync(struct ubik_trans *atrans)
{
int i;
afs_int32 code;
struct buffer *tb;
afs_int32 file;
afs_int32 rCode;
+ struct ubik_dbase *adbase = atrans->dbase;
rCode = 0;
while (1) {
if (tb->dirty == 1) {
if (file == BADFID)
file = tb->file;
- if (file != BADFID && tb->file == file)
+ if (file != BADFID && tb->file == file) {
tb->dirty = 0;
+ DedupBuffer(tb);
+ }
}
}
if (file == BADFID)
return rCode;
}
-/* Same as read, only do not even try to read the page */
+/*!
+ * \brief Same as DRead(), only do not even try to read the page.
+ */
static char *
-DNew(struct ubik_dbase *dbase, afs_int32 fid, int page)
+DNew(struct ubik_trans *atrans, afs_int32 fid, int page)
{
struct buffer *tb;
+ struct ubik_dbase *dbase = atrans->dbase;
if ((tb = newslot(dbase, fid, page)) == 0)
return NULL;
return tb->data;
}
-/* read data from database */
+/*!
+ * \brief Read data from database.
+ */
int
-udisk_read(struct ubik_trans *atrans, afs_int32 afile, char *abuffer,
+udisk_read(struct ubik_trans *atrans, afs_int32 afile, void *abuffer,
afs_int32 apos, afs_int32 alen)
{
char *bp;
afs_int32 offset, len, totalLen;
- struct ubik_dbase *dbase;
if (atrans->flags & TRDONE)
return UDONE;
totalLen = 0;
- dbase = atrans->dbase;
while (alen > 0) {
- bp = DRead(dbase, afile, apos >> UBIK_LOGPAGESIZE);
+ bp = DRead(atrans, afile, apos >> UBIK_LOGPAGESIZE);
if (!bp)
return UEOF;
/* otherwise, min of remaining bytes and end of buffer to user mode */
if (len > alen)
len = alen;
memcpy(abuffer, bp + offset, len);
- abuffer += len;
+ abuffer = (char *)abuffer + len;
apos += len;
alen -= len;
totalLen += len;
return 0;
}
-/* truncate file */
+/*!
+ * \brief Truncate file.
+ */
int
udisk_truncate(struct ubik_trans *atrans, afs_int32 afile, afs_int32 alength)
{
return code;
}
-/* write data to database, using logs */
+/*!
+ * \brief Write data to database, using logs.
+ */
int
-udisk_write(struct ubik_trans *atrans, afs_int32 afile, char *abuffer,
+udisk_write(struct ubik_trans *atrans, afs_int32 afile, void *abuffer,
afs_int32 apos, afs_int32 alen)
{
char *bp;
afs_int32 offset, len, totalLen;
- struct ubik_dbase *dbase;
struct ubik_trunc *tt;
afs_int32 code;
if (atrans->type != UBIK_WRITETRANS)
return UBADTYPE;
- dbase = atrans->dbase;
/* first write the data to the log */
- code = udisk_LogWriteData(dbase, afile, abuffer, apos, alen);
+ code = udisk_LogWriteData(atrans->dbase, afile, abuffer, apos, alen);
if (code)
return code;
/* now update vm */
totalLen = 0;
while (alen > 0) {
- bp = DRead(dbase, afile, apos >> UBIK_LOGPAGESIZE);
+ bp = DRead(atrans, afile, apos >> UBIK_LOGPAGESIZE);
if (!bp) {
- bp = DNew(dbase, afile, apos >> UBIK_LOGPAGESIZE);
+ bp = DNew(atrans, afile, apos >> UBIK_LOGPAGESIZE);
if (!bp)
return UIOERROR;
- memset(bp, 0, UBIK_PAGESIZE);
}
/* otherwise, min of remaining bytes and end of buffer to user mode */
offset = apos & (UBIK_PAGESIZE - 1);
if (len > alen)
len = alen;
memcpy(bp + offset, abuffer, len);
- abuffer += len;
+ abuffer = (char *)abuffer + len;
apos += len;
alen -= len;
totalLen += len;
return 0;
}
-/* begin a new local transaction */
+/*!
+ * \brief Begin a new local transaction.
+ */
int
udisk_begin(struct ubik_dbase *adbase, int atype, struct ubik_trans **atrans)
{
afs_int32 code;
struct ubik_trans *tt;
- *atrans = (struct ubik_trans *)NULL;
- /* Make sure system is initialized before doing anything */
- if (!initd) {
- initd = 1;
- DInit(ubik_nBuffers);
- }
+ *atrans = NULL;
if (atype == UBIK_WRITETRANS) {
- if (adbase->flags & DBWRITING)
+ if (adbase->dbFlags & DBWRITING)
return USYNC;
code = udisk_LogOpcode(adbase, LOGNEW, 0);
if (code)
return code;
}
- tt = (struct ubik_trans *)malloc(sizeof(struct ubik_trans));
- memset(tt, 0, sizeof(struct ubik_trans));
+ tt = calloc(1, sizeof(struct ubik_trans));
tt->dbase = adbase;
tt->next = adbase->activeTrans;
adbase->activeTrans = tt;
tt->type = atype;
if (atype == UBIK_READTRANS)
adbase->readers++;
- else if (atype == UBIK_WRITETRANS)
- adbase->flags |= DBWRITING;
+ else if (atype == UBIK_WRITETRANS) {
+ UBIK_VERSION_LOCK;
+ adbase->dbFlags |= DBWRITING;
+ UBIK_VERSION_UNLOCK;
+ }
*atrans = tt;
return 0;
}
-/* commit transaction */
+/*!
+ * \brief Commit transaction.
+ */
int
udisk_commit(struct ubik_trans *atrans)
{
struct ubik_dbase *dbase;
afs_int32 code = 0;
struct ubik_version oldversion, newversion;
+ afs_int32 now = FT_ApproxTime();
if (atrans->flags & TRDONE)
return (UTWOENDS);
/* On the first write to the database. We update the versions */
if (ubeacon_AmSyncSite() && !(urecovery_state & UBIK_RECLABELDB)) {
+ UBIK_VERSION_LOCK;
+ if (version_globals.ubik_epochTime < UBIK_MILESTONE
+ || version_globals.ubik_epochTime > now) {
+ ViceLog(0,
+ ("Ubik: New database label %d is out of the valid range (%d - %d)\n",
+ version_globals.ubik_epochTime, UBIK_MILESTONE, now));
+ panic("Writing Ubik DB label\n");
+ }
oldversion = dbase->version;
- newversion.epoch = FT_ApproxTime();;
+ newversion.epoch = version_globals.ubik_epochTime;
newversion.counter = 1;
code = (*dbase->setlabel) (dbase, 0, &newversion);
- if (code)
- return (code);
- ubik_epochTime = newversion.epoch;
+ if (code) {
+ UBIK_VERSION_UNLOCK;
+ return code;
+ }
+
dbase->version = newversion;
+ UBIK_VERSION_UNLOCK;
+
+ urecovery_state |= UBIK_RECLABELDB;
/* Ignore the error here. If the call fails, the site is
- * marked down and when we detect it is up again, we will
+ * marked down and when we detect it is up again, we will
* send the entire database to it.
*/
- ContactQuorum(DISK_SetVersion, atrans, 1 /*CStampVersion */ ,
- &oldversion, &newversion);
- urecovery_state |= UBIK_RECLABELDB;
+ ContactQuorum_DISK_SetVersion( atrans, 1 /*CStampVersion */ ,
+ &oldversion, &newversion);
}
+ UBIK_VERSION_LOCK;
dbase->version.counter++; /* bump commit count */
-#ifdef AFS_PTHREAD_ENV
- assert(pthread_cond_broadcast(&dbase->version_cond) == 0);
-#else
- LWP_NoYieldSignal(&dbase->version);
-#endif
code = udisk_LogEnd(dbase, &dbase->version);
if (code) {
dbase->version.counter--;
- return (code);
+ UBIK_VERSION_UNLOCK;
+ return code;
}
+ UBIK_VERSION_UNLOCK;
/* If we fail anytime after this, then panic and let the
- * recovery replay the log.
+ * recovery replay the log.
*/
- code = DFlush(dbase); /* write dirty pages to respective files */
+ code = DFlush(atrans); /* write dirty pages to respective files */
if (code)
panic("Writing Ubik DB modifications\n");
- code = DSync(dbase); /* sync the files and mark pages not dirty */
+ code = DSync(atrans); /* sync the files and mark pages not dirty */
if (code)
panic("Synchronizing Ubik DB modifications\n");
return code;
}
-/* abort transaction */
+/*!
+ * \brief Abort transaction.
+ */
int
udisk_abort(struct ubik_trans *atrans)
{
return UTWOENDS;
/* Check if we are the write trans before logging abort, lest we
- * abort a good write trans in progress.
- * We don't really care if the LOGABORT gets to the log because we
- * truncate the log next. If the truncate fails, we panic; for
+ * abort a good write trans in progress.
+ * We don't really care if the LOGABORT gets to the log because we
+ * truncate the log next. If the truncate fails, we panic; for
* otherwise, the log entries remain. On restart, replay of the log
* will do nothing because the abort is there or no LogEnd opcode.
*/
dbase = atrans->dbase;
- if (atrans->type == UBIK_WRITETRANS && dbase->flags & DBWRITING) {
+ if (atrans->type == UBIK_WRITETRANS && dbase->dbFlags & DBWRITING) {
udisk_LogOpcode(dbase, LOGABORT, 1);
code = (*dbase->truncate) (dbase, LOGFILE, 0);
if (code)
panic("Truncating Ubik logfile during an abort\n");
- DAbort(dbase); /* remove all dirty pages */
+ DAbort(atrans); /* remove all dirty pages */
}
/* When the transaction is marked done, it also means the logfile
return 0;
}
-/* destroy a transaction after it has been committed or aborted. if
- * it hasn't committed before you call this routine, we'll abort the
+/*!
+ * \brief Destroy a transaction after it has been committed or aborted.
+ *
+ * If it hasn't committed before you call this routine, we'll abort the
* transaction for you.
*/
int
{
struct ubik_dbase *dbase;
-#if defined(UBIK_PAUSE)
- /* Another thread is trying to lock this transaction.
- * That can only be an RPC doing SDISK_Lock.
- * Unlock the transaction, 'cause otherwise the other
- * thread will never wake up. Don't free it because
- * the caller will do that already.
- */
- if (atrans->flags & TRSETLOCK) {
- atrans->flags |= TRSTALE;
- ulock_relLock(atrans);
- return;
- }
-#endif /* UBIK_PAUSE */
if (!(atrans->flags & TRDONE))
udisk_abort(atrans);
dbase = atrans->dbase;
/* check if we are the write trans before unsetting the DBWRITING bit, else
* we could be unsetting someone else's bit.
*/
- if (atrans->type == UBIK_WRITETRANS && dbase->flags & DBWRITING) {
- dbase->flags &= ~DBWRITING;
+ if (atrans->type == UBIK_WRITETRANS && dbase->dbFlags & DBWRITING) {
+ UBIK_VERSION_LOCK;
+ dbase->dbFlags &= ~DBWRITING;
+ UBIK_VERSION_UNLOCK;
} else {
dbase->readers--;
}
/* Wakeup any writers waiting in BeginTrans() */
#ifdef AFS_PTHREAD_ENV
- assert(pthread_cond_broadcast(&dbase->flags_cond) == 0);
+ opr_cv_broadcast(&dbase->flags_cond);
#else
- LWP_NoYieldSignal(&dbase->flags);
+ LWP_NoYieldSignal(&dbase->dbFlags);
#endif
return 0;
}