/*
* __REC_SYNC -- sync the recno tree to disk.
*
* Parameters:
* dbp: pointer to access method
*
* Returns:
* RET_SUCCESS, RET_ERROR.
*/
int
__rec_sync(const DB *dbp, u_int flags)
{
struct iovec iov[2];
BTREE *t;
DBT data, key;
off_t off;
recno_t scursor, trec;
int status;
t = dbp->internal;
/* Toss any page pinned across calls. */
if (t->bt_pinned != NULL) {
mpool_put(t->bt_mp, t->bt_pinned, 0);
t->bt_pinned = NULL;
}
if (flags == R_RECNOSYNC)
return (__bt_sync(dbp, 0));
if (F_ISSET(t, R_RDONLY | R_INMEM) || !F_ISSET(t, R_MODIFIED))
return (RET_SUCCESS);
/* Read any remaining records into the tree. */
if (!F_ISSET(t, R_EOF) && t->bt_irec(t, MAX_REC_NUMBER) == RET_ERROR)
return (RET_ERROR);
/* Rewind the file descriptor. */
if (lseek(t->bt_rfd, (off_t)0, SEEK_SET) != 0)
return (RET_ERROR);
/* Save the cursor. */
scursor = t->bt_cursor.rcursor;
key.size = sizeof(recno_t);
key.data = &trec;
if (F_ISSET(t, R_FIXLEN)) {
/*
* We assume that fixed length records are all fixed length.
* Any that aren't are either EINVAL'd or corrected by the
* record put code.
*/
status = (dbp->seq)(dbp, &key, &data, R_FIRST);
while (status == RET_SUCCESS) {
if (write(t->bt_rfd, data.data, data.size) !=
(ssize_t) data.size)
return (RET_ERROR);
status = (dbp->seq)(dbp, &key, &data, R_NEXT);
}
} else {
iov[1].iov_base = &t->bt_bval;
iov[1].iov_len = 1;
status = (dbp->seq)(dbp, &key, &data, R_FIRST);
while (status == RET_SUCCESS) {
iov[0].iov_base = data.data;
iov[0].iov_len = data.size;
if (writev(t->bt_rfd, iov, 2) !=
(ssize_t) (data.size + 1))
return (RET_ERROR);
status = (dbp->seq)(dbp, &key, &data, R_NEXT);
}
}
/* Restore the cursor. */
t->bt_cursor.rcursor = scursor;
if (status == RET_ERROR)
return (RET_ERROR);
if ((off = lseek(t->bt_rfd, (off_t)0, SEEK_CUR)) == -1)
return (RET_ERROR);
if (ftruncate(t->bt_rfd, off))
return (RET_ERROR);
F_CLR(t, R_MODIFIED);
return (RET_SUCCESS);
}
DB *
__rec_open(const char *fname, int flags, mode_t mode, const RECNOINFO *openinfo,
int dflags)
{
BTREE *t;
BTREEINFO btopeninfo;
DB *dbp;
PAGE *h;
struct stat sb;
int rfd = -1; /* pacify gcc */
int sverrno;
dbp = NULL;
/* Open the user's file -- if this fails, we're done. */
if (fname != NULL) {
#ifndef O_CLOEXEC
#define O_CLOEXEC 0
#endif
if ((rfd = open(fname, flags | O_CLOEXEC, mode)) == -1)
return NULL;
#if O_CLOEXEC == 0
if (fcntl(rfd, F_SETFD, FD_CLOEXEC) == -1)
goto err;
#endif
}
/* Create a btree in memory (backed by disk). */
if (openinfo) {
if (openinfo->flags & ~(R_FIXEDLEN | R_NOKEY | R_SNAPSHOT))
goto einval;
btopeninfo.flags = 0;
btopeninfo.cachesize = openinfo->cachesize;
btopeninfo.maxkeypage = 0;
btopeninfo.minkeypage = 0;
btopeninfo.psize = openinfo->psize;
btopeninfo.compare = NULL;
btopeninfo.prefix = NULL;
btopeninfo.lorder = openinfo->lorder;
dbp = __bt_open(openinfo->bfname,
O_RDWR, S_IRUSR | S_IWUSR, &btopeninfo, dflags);
} else
dbp = __bt_open(NULL, O_RDWR, S_IRUSR | S_IWUSR, NULL, dflags);
if (dbp == NULL)
goto err;
/*
* Some fields in the tree structure are recno specific. Fill them
* in and make the btree structure look like a recno structure. We
* don't change the bt_ovflsize value, it's close enough and slightly
* bigger.
*/
t = dbp->internal;
if (openinfo) {
if (openinfo->flags & R_FIXEDLEN) {
F_SET(t, R_FIXLEN);
t->bt_reclen = openinfo->reclen;
if (t->bt_reclen == 0)
goto einval;
}
t->bt_bval = openinfo->bval;
} else
t->bt_bval = '\n';
F_SET(t, R_RECNO);
if (fname == NULL)
F_SET(t, R_EOF | R_INMEM);
else
t->bt_rfd = rfd;
if (fname != NULL) {
/*
* In 4.4BSD, stat(2) returns true for ISSOCK on pipes.
* Unfortunately, that's not portable, so we use lseek
* and check the errno values.
*/
errno = 0;
if (lseek(rfd, (off_t)0, SEEK_CUR) == -1 && errno == ESPIPE) {
switch (flags & O_ACCMODE) {
case O_RDONLY:
F_SET(t, R_RDONLY);
break;
default:
goto einval;
}
slow: if ((t->bt_rfp = fdopen(rfd, "r")) == NULL)
goto err;
F_SET(t, R_CLOSEFP);
t->bt_irec =
F_ISSET(t, R_FIXLEN) ? __rec_fpipe : __rec_vpipe;
} else {
switch (flags & O_ACCMODE) {
case O_RDONLY:
F_SET(t, R_RDONLY);
break;
case O_RDWR:
break;
default:
goto einval;
}
if (fstat(rfd, &sb))
goto err;
/*
* Kluge -- we'd like to test to see if the file is too
* big to mmap. Since, we don't know what size or type
* off_t's or size_t's are, what the largest unsigned
* integral type is, or what random insanity the local
* C compiler will perpetrate, doing the comparison in
* a portable way is flatly impossible. Hope that mmap
* fails if the file is too large.
*/
if (sb.st_size == 0)
F_SET(t, R_EOF);
else {
#ifdef MMAP_NOT_AVAILABLE
/*
* XXX
* Mmap doesn't work correctly on many current
* systems. In particular, it can fail subtly,
* with cache coherency problems. Don't use it
* for now.
*/
t->bt_msize = sb.st_size;
if ((t->bt_smap = mmap(NULL, t->bt_msize,
PROT_READ, MAP_FILE | MAP_PRIVATE, rfd,
(off_t)0)) == (caddr_t)-1)
goto slow;
t->bt_cmap = t->bt_smap;
t->bt_emap = t->bt_smap + sb.st_size;
t->bt_irec = F_ISSET(t, R_FIXLEN) ?
__rec_fmap : __rec_vmap;
F_SET(t, R_MEMMAPPED);
#else
goto slow;
#endif
}
}
}
/* Use the recno routines. */
dbp->close = __rec_close;
dbp->del = __rec_delete;
dbp->fd = __rec_fd;
dbp->get = __rec_get;
dbp->put = __rec_put;
dbp->seq = __rec_seq;
dbp->sync = __rec_sync;
/* If the root page was created, reset the flags. */
if ((h = mpool_get(t->bt_mp, P_ROOT, 0)) == NULL)
goto err;
if ((h->flags & P_TYPE) == P_BLEAF) {
F_CLR(h, P_TYPE);
F_SET(h, P_RLEAF);
mpool_put(t->bt_mp, h, MPOOL_DIRTY);
} else
mpool_put(t->bt_mp, h, 0);
if (openinfo && openinfo->flags & R_SNAPSHOT &&
!F_ISSET(t, R_EOF | R_INMEM) &&
t->bt_irec(t, MAX_REC_NUMBER) == RET_ERROR)
goto err;
return (dbp);
einval: errno = EINVAL;
err: sverrno = errno;
if (dbp != NULL)
(void)__bt_close(dbp);
if (fname != NULL)
(void)close(rfd);
errno = sverrno;
return (NULL);
}
/*
* __REC_PUT -- Add a recno item to the tree.
*
* Parameters:
* dbp: pointer to access method
* key: key
* data: data
* flag: R_CURSOR, R_IAFTER, R_IBEFORE, R_NOOVERWRITE
*
* Returns:
* RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is
* already in the tree and R_NOOVERWRITE specified.
*/
int
__rec_put(const DB *dbp, DBT *key, const DBT *data, u_int flags)
{
BTREE *t;
DBT fdata, tdata;
recno_t nrec;
int status;
t = dbp->internal;
/* Toss any page pinned across calls. */
if (t->bt_pinned != NULL) {
mpool_put(t->bt_mp, t->bt_pinned, 0);
t->bt_pinned = NULL;
}
/*
* If using fixed-length records, and the record is long, return
* EINVAL. If it's short, pad it out. Use the record data return
* memory, it's only short-term.
*/
if (F_ISSET(t, R_FIXLEN) && data->size != t->bt_reclen) {
if (data->size > t->bt_reclen)
goto einval;
if (t->bt_rdata.size < t->bt_reclen) {
t->bt_rdata.data =
reallocf(t->bt_rdata.data, t->bt_reclen);
if (t->bt_rdata.data == NULL)
return (RET_ERROR);
t->bt_rdata.size = t->bt_reclen;
}
memmove(t->bt_rdata.data, data->data, data->size);
memset((char *)t->bt_rdata.data + data->size,
t->bt_bval, t->bt_reclen - data->size);
fdata.data = t->bt_rdata.data;
fdata.size = t->bt_reclen;
} else {
fdata.data = data->data;
fdata.size = data->size;
}
switch (flags) {
case R_CURSOR:
if (!F_ISSET(&t->bt_cursor, CURS_INIT))
goto einval;
nrec = t->bt_cursor.rcursor;
break;
case R_SETCURSOR:
if ((nrec = *(recno_t *)key->data) == 0)
goto einval;
break;
case R_IAFTER:
if ((nrec = *(recno_t *)key->data) == 0) {
nrec = 1;
flags = R_IBEFORE;
}
break;
case 0:
case R_IBEFORE:
if ((nrec = *(recno_t *)key->data) == 0)
goto einval;
break;
case R_NOOVERWRITE:
if ((nrec = *(recno_t *)key->data) == 0)
goto einval;
if (nrec <= t->bt_nrecs)
return (RET_SPECIAL);
break;
default:
einval: errno = EINVAL;
return (RET_ERROR);
}
/*
* Make sure that records up to and including the put record are
* already in the database. If skipping records, create empty ones.
*/
if (nrec > t->bt_nrecs) {
if (!F_ISSET(t, R_EOF | R_INMEM) &&
t->bt_irec(t, nrec) == RET_ERROR)
return (RET_ERROR);
if (nrec > t->bt_nrecs + 1) {
if (F_ISSET(t, R_FIXLEN)) {
if ((tdata.data =
(void *)malloc(t->bt_reclen)) == NULL)
return (RET_ERROR);
tdata.size = t->bt_reclen;
memset(tdata.data, t->bt_bval, tdata.size);
} else {
tdata.data = NULL;
tdata.size = 0;
}
while (nrec > t->bt_nrecs + 1)
if (__rec_iput(t,
t->bt_nrecs, &tdata, 0) != RET_SUCCESS)
return (RET_ERROR);
if (F_ISSET(t, R_FIXLEN))
free(tdata.data);
}
}
if ((status = __rec_iput(t, nrec - 1, &fdata, flags)) != RET_SUCCESS)
return (status);
switch (flags) {
case R_IAFTER:
nrec++;
break;
case R_SETCURSOR:
t->bt_cursor.rcursor = nrec;
break;
}
F_SET(t, R_MODIFIED);
return (__rec_ret(t, NULL, nrec, key, NULL));
}
/*
* __REC_SEQ -- Recno sequential scan interface.
*
* Parameters:
* dbp: pointer to access method
* key: key for positioning and return value
* data: data return value
* flags: R_CURSOR, R_FIRST, R_LAST, R_NEXT, R_PREV.
*
* Returns:
* RET_ERROR, RET_SUCCESS or RET_SPECIAL if there's no next key.
*/
int
__rec_seq(const DB *dbp, DBT *key, DBT *data, u_int flags)
{
BTREE *t;
EPG *e;
recno_t nrec;
int status;
t = dbp->internal;
/* Toss any page pinned across calls. */
if (t->bt_pinned != NULL) {
mpool_put(t->bt_mp, t->bt_pinned, 0);
t->bt_pinned = NULL;
}
switch(flags) {
case R_CURSOR:
if ((nrec = *(recno_t *)key->data) == 0)
goto einval;
break;
case R_NEXT:
if (F_ISSET(&t->bt_cursor, CURS_INIT)) {
nrec = t->bt_cursor.rcursor + 1;
break;
}
/* FALLTHROUGH */
case R_FIRST:
nrec = 1;
break;
case R_PREV:
if (F_ISSET(&t->bt_cursor, CURS_INIT)) {
if ((nrec = t->bt_cursor.rcursor - 1) == 0)
return (RET_SPECIAL);
break;
}
/* FALLTHROUGH */
case R_LAST:
if (!F_ISSET(t, R_EOF | R_INMEM) &&
t->bt_irec(t, MAX_REC_NUMBER) == RET_ERROR)
return (RET_ERROR);
nrec = t->bt_nrecs;
break;
default:
einval: errno = EINVAL;
return (RET_ERROR);
}
if (t->bt_nrecs == 0 || nrec > t->bt_nrecs) {
if (!F_ISSET(t, R_EOF | R_INMEM) &&
(status = t->bt_irec(t, nrec)) != RET_SUCCESS)
return (status);
if (t->bt_nrecs == 0 || nrec > t->bt_nrecs)
return (RET_SPECIAL);
}
if ((e = __rec_search(t, nrec - 1, SEARCH)) == NULL)
return (RET_ERROR);
F_SET(&t->bt_cursor, CURS_INIT);
t->bt_cursor.rcursor = nrec;
status = __rec_ret(t, e, nrec, key, data);
if (F_ISSET(t, B_DB_LOCK))
mpool_put(t->bt_mp, e->page, 0);
else
t->bt_pinned = e->page;
return (status);
}
