static void
swabvid(struct cpu_disklabel *pcpul)
{
M_32_SWAP(pcpul->vid_oss);
M_16_SWAP(pcpul->vid_osl);
/*
M_16_SWAP(pcpul->vid_osa_u);
M_16_SWAP(pcpul->vid_osa_l);
*/
M_32_SWAP(pcpul->vid_cas);
}
/**
* Flush out header onto disk
*/
static int
_ffdb_hput_header (ffdb_htab_t* hashp)
{
ffdb_hashhdr_t *whdrp;
ffdb_hashhdr_t whdr;
unsigned int num_copied = 0;
unsigned int chksum = 0;
whdrp = &hashp->hdr;
if (hashp->mborder == LITTLE_ENDIAN) {
whdrp = &whdr;
_ffdb_swap_header_copy(&hashp->hdr, whdrp);
}
/* calculate checksum value */
chksum = __ffdb_crc32_checksum (chksum, (const unsigned char *)whdrp,
sizeof(ffdb_hashhdr_t) - sizeof(unsigned int));
/* hashp->hdr.chksum = chksum; */
if (hashp->mborder == LITTLE_ENDIAN)
M_32_SWAP(chksum);
whdrp->chksum = chksum;
/* write the header */
lseek(hashp->fp, 0, SEEK_SET);
num_copied = write(hashp->fp, whdrp, sizeof(ffdb_hashhdr_t));
if (num_copied != sizeof(ffdb_hashhdr_t)) {
fprintf(stderr, "hash: could not write hash header");
return -1;
}
return 0;
}
static void
swap_header(HTAB *hashp)
{
HASHHDR *hdrp;
int i;
hdrp = &hashp->hdr;
M_32_SWAP(hdrp->magic);
M_32_SWAP(hdrp->version);
M_32_SWAP(hdrp->lorder);
M_32_SWAP(hdrp->bsize);
M_32_SWAP(hdrp->bshift);
M_32_SWAP(hdrp->dsize);
M_32_SWAP(hdrp->ssize);
M_32_SWAP(hdrp->sshift);
M_32_SWAP(hdrp->ovfl_point);
M_32_SWAP(hdrp->last_freed);
M_32_SWAP(hdrp->max_bucket);
M_32_SWAP(hdrp->high_mask);
M_32_SWAP(hdrp->low_mask);
M_32_SWAP(hdrp->ffactor);
M_32_SWAP(hdrp->nkeys);
M_32_SWAP(hdrp->hdrpages);
M_32_SWAP(hdrp->h_charkey);
for (i = 0; i < NCACHED; i++) {
M_32_SWAP(hdrp->spares[i]);
M_16_SWAP(hdrp->bitmaps[i]);
}
}
/*
* __BT_OPEN -- Open a btree.
*
* Creates and fills a DB struct, and calls the routine that actually
* opens the btree.
*
* Parameters:
* fname: filename (NULL for in-memory trees)
* flags: open flag bits
* mode: open permission bits
* b: BTREEINFO pointer
*
* Returns:
* NULL on failure, pointer to DB on success.
*
*/
DB *
__bt_open(const char *fname, int flags, mode_t mode, const BTREEINFO *openinfo,
int dflags)
{
struct stat sb;
BTMETA m;
BTREE *t;
BTREEINFO b;
DB *dbp;
pgno_t ncache;
ssize_t nr;
size_t temp;
int machine_lorder;
t = NULL;
/*
* Intention is to make sure all of the user's selections are okay
* here and then use them without checking. Can't be complete, since
* we don't know the right page size, lorder or flags until the backing
* file is opened. Also, the file's page size can cause the cachesize
* to change.
*/
machine_lorder = byteorder();
if (openinfo) {
b = *openinfo;
/* Flags: R_DUP. */
if (b.flags & ~(R_DUP))
goto einval;
/*
* Page size must be indx_t aligned and >= MINPSIZE. Default
* page size is set farther on, based on the underlying file
* transfer size.
*/
if (b.psize &&
(b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
b.psize & (sizeof(indx_t) - 1)))
goto einval;
/* Minimum number of keys per page; absolute minimum is 2. */
if (b.minkeypage) {
if (b.minkeypage < 2)
goto einval;
} else
b.minkeypage = DEFMINKEYPAGE;
/* If no comparison, use default comparison and prefix. */
if (b.compare == NULL) {
b.compare = __bt_defcmp;
if (b.prefix == NULL)
b.prefix = __bt_defpfx;
}
if (b.lorder == 0)
b.lorder = machine_lorder;
} else {
b.compare = __bt_defcmp;
b.cachesize = 0;
b.flags = 0;
b.lorder = machine_lorder;
b.minkeypage = DEFMINKEYPAGE;
b.prefix = __bt_defpfx;
b.psize = 0;
}
/* Check for the ubiquitous PDP-11. */
if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
goto einval;
/* Allocate and initialize DB and BTREE structures. */
if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
goto err;
memset(t, 0, sizeof(BTREE));
t->bt_fd = -1; /* Don't close unopened fd on error. */
t->bt_lorder = b.lorder;
t->bt_order = NOT;
t->bt_cmp = b.compare;
t->bt_pfx = b.prefix;
t->bt_rfd = -1;
if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
goto err;
memset(t->bt_dbp, 0, sizeof(DB));
if (t->bt_lorder != machine_lorder)
F_SET(t, B_NEEDSWAP);
dbp->type = DB_BTREE;
dbp->internal = t;
dbp->close = __bt_close;
dbp->del = __bt_delete;
dbp->fd = __bt_fd;
dbp->get = __bt_get;
dbp->put = __bt_put;
dbp->seq = __bt_seq;
dbp->sync = __bt_sync;
/*
* If no file name was supplied, this is an in-memory btree and we
* open a backing temporary file. Otherwise, it's a disk-based tree.
*/
if (fname) {
switch (flags & O_ACCMODE) {
case O_RDONLY:
F_SET(t, B_RDONLY);
break;
case O_RDWR:
break;
case O_WRONLY:
default:
goto einval;
}
if ((t->bt_fd = open(fname, flags, mode)) == -1)
goto err;
if (fcntl(t->bt_fd, F_SETFD, FD_CLOEXEC) == -1)
goto err;
} else {
if ((flags & O_ACCMODE) != O_RDWR)
goto einval;
if ((t->bt_fd = tmp()) == -1)
goto err;
F_SET(t, B_INMEM);
}
if (fcntl(t->bt_fd, F_SETFD, FD_CLOEXEC) == -1)
goto err;
if (fstat(t->bt_fd, &sb))
goto err;
if (sb.st_size) {
if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
goto err;
if (nr != sizeof(BTMETA))
goto eftype;
/*
* Read in the meta-data. This can change the notion of what
* the lorder, page size and flags are, and, when the page size
* changes, the cachesize value can change too. If the user
* specified the wrong byte order for an existing database, we
* don't bother to return an error, we just clear the NEEDSWAP
* bit.
*/
if (m.magic == BTREEMAGIC)
F_CLR(t, B_NEEDSWAP);
else {
F_SET(t, B_NEEDSWAP);
M_32_SWAP(m.magic);
M_32_SWAP(m.version);
M_32_SWAP(m.psize);
M_32_SWAP(m.free);
M_32_SWAP(m.nrecs);
M_32_SWAP(m.flags);
}
if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
goto eftype;
if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
m.psize & (sizeof(indx_t) - 1))
goto eftype;
if (m.flags & ~SAVEMETA)
goto eftype;
b.psize = m.psize;
F_SET(t, m.flags);
t->bt_free = m.free;
t->bt_nrecs = m.nrecs;
} else {
/*
* Set the page size to the best value for I/O to this file.
* Don't overflow the page offset type.
*/
if (b.psize == 0) {
b.psize = sb.st_blksize;
if (b.psize < MINPSIZE)
b.psize = MINPSIZE;
if (b.psize > MAX_PAGE_OFFSET + 1)
b.psize = MAX_PAGE_OFFSET + 1;
}
/* Set flag if duplicates permitted. */
if (!(b.flags & R_DUP))
F_SET(t, B_NODUPS);
t->bt_free = P_INVALID;
t->bt_nrecs = 0;
F_SET(t, B_METADIRTY);
}
t->bt_psize = b.psize;
/* Set the cache size; must be a multiple of the page size. */
if (b.cachesize && b.cachesize & (b.psize - 1))
b.cachesize += (~b.cachesize & (b.psize - 1)) + 1;
if (b.cachesize < b.psize * MINCACHE)
b.cachesize = b.psize * MINCACHE;
/* Calculate number of pages to cache. */
ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
/*
* The btree data structure requires that at least two keys can fit on
* a page, but other than that there's no fixed requirement. The user
* specified a minimum number per page, and we translated that into the
* number of bytes a key/data pair can use before being placed on an
* overflow page. This calculation includes the page header, the size
* of the index referencing the leaf item and the size of the leaf item
* structure. Also, don't let the user specify a minkeypage such that
* a key/data pair won't fit even if both key and data are on overflow
* pages.
*/
temp = (t->bt_psize - BTDATAOFF) / b.minkeypage -
(sizeof(indx_t) + NBLEAFDBT(0, 0));
_DBFIT(temp, indx_t);
t->bt_ovflsize = (indx_t)temp;
if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
t->bt_ovflsize =
NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
/* Initialize the buffer pool. */
if ((t->bt_mp =
mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
goto err;
if (!F_ISSET(t, B_INMEM))
mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
/* Create a root page if new tree. */
if (nroot(t) == RET_ERROR)
goto err;
/* Global flags. */
if (dflags & DB_LOCK)
F_SET(t, B_DB_LOCK);
if (dflags & DB_SHMEM)
F_SET(t, B_DB_SHMEM);
if (dflags & DB_TXN)
F_SET(t, B_DB_TXN);
return (dbp);
einval: errno = EINVAL;
goto err;
eftype: errno = EFTYPE;
goto err;
err: if (t) {
if (t->bt_dbp)
free(t->bt_dbp);
if (t->bt_fd != -1)
(void)close(t->bt_fd);
free(t);
}
return (NULL);
}
/*
* __BT_BPGIN, __BT_BPGOUT --
* Convert host-specific number layout to/from the host-independent
* format stored on disk.
*
* Parameters:
* t: tree
* pg: page number
* h: page to convert
*/
void
__bt_pgin(void *t, pgno_t pg, void *pp)
{
PAGE *h;
indx_t i, top;
u_char flags;
char *p;
if (!F_ISSET(((BTREE *)t), B_NEEDSWAP))
return;
if (pg == P_META) {
mswap(pp);
return;
}
h = pp;
M_32_SWAP(h->pgno);
M_32_SWAP(h->prevpg);
M_32_SWAP(h->nextpg);
M_32_SWAP(h->flags);
M_16_SWAP(h->lower);
M_16_SWAP(h->upper);
top = NEXTINDEX(h);
if ((h->flags & P_TYPE) == P_BINTERNAL)
for (i = 0; i < top; i++) {
M_16_SWAP(h->linp[i]);
p = (char *)GETBINTERNAL(h, i);
P_32_SWAP(p);
p += sizeof(u_int32_t);
P_32_SWAP(p);
p += sizeof(pgno_t);
if (*(u_char *)p & P_BIGKEY) {
p += sizeof(u_char);
P_32_SWAP(p);
p += sizeof(pgno_t);
P_32_SWAP(p);
}
}
else if ((h->flags & P_TYPE) == P_BLEAF)
for (i = 0; i < top; i++) {
M_16_SWAP(h->linp[i]);
p = (char *)GETBLEAF(h, i);
P_32_SWAP(p);
p += sizeof(u_int32_t);
P_32_SWAP(p);
p += sizeof(u_int32_t);
flags = *(u_char *)p;
if (flags & (P_BIGKEY | P_BIGDATA)) {
p += sizeof(u_char);
if (flags & P_BIGKEY) {
P_32_SWAP(p);
p += sizeof(pgno_t);
P_32_SWAP(p);
}
if (flags & P_BIGDATA) {
p += sizeof(u_int32_t);
P_32_SWAP(p);
p += sizeof(pgno_t);
P_32_SWAP(p);
}
}
}
}
/**
* Swap headers only if the byte order is little endian
*/
static void
_ffdb_swap_header (ffdb_htab_t* hashp)
{
ffdb_hashhdr_t *hdrp;
int i;
hdrp = &hashp->hdr;
M_32_SWAP(hdrp->magic);
M_32_SWAP(hdrp->version);
M_32_SWAP(hdrp->lorder);
M_32_SWAP(hdrp->bsize);
M_32_SWAP(hdrp->bshift);
M_32_SWAP(hdrp->ovfl_point);
M_32_SWAP(hdrp->max_bucket);
M_32_SWAP(hdrp->high_mask);
M_32_SWAP(hdrp->low_mask);
M_32_SWAP(hdrp->ffactor);
M_32_SWAP(hdrp->nkeys);
M_32_SWAP(hdrp->hdrpages);
M_32_SWAP(hdrp->uinfolen);
M_16_SWAP(hdrp->uinfo_page);
M_16_SWAP(hdrp->uinfo_npages);
M_16_SWAP(hdrp->cfig_page);
M_16_SWAP(hdrp->cfig_npages);
M_32_SWAP(hdrp->num_cfigs);
M_32_SWAP(hdrp->h_charkey);
M_32_SWAP(hdrp->num_moved_pages);
for (i = 0; i < NCACHED; i++)
M_32_SWAP(hdrp->spares[i]);
for (i = 0; i < NCACHED; i++)
M_32_SWAP(hdrp->free_pages[i]);
M_32_SWAP(hdrp->chksum);
}