/*
* Remove unneeded leaves from the old tree.
* Remove directories from the lookup chains.
*/
void
removeoldleaves(void)
{
struct entry *ep;
ino_t i;
Vprintf(stdout, "Mark entries to be removed.\n");
for (i = ROOTINO + 1; i < maxino; i++) {
ep = lookupino(i);
if (ep == NULL)
continue;
if (TSTINO(i, usedinomap))
continue;
for ( ; ep != NULL; ep = ep->e_links) {
Dprintf(stdout, "%s: REMOVE\n", myname(ep));
if (ep->e_type == LEAF) {
removeleaf(ep);
freeentry(ep);
} else {
mktempname(ep);
deleteino(ep->e_ino);
ep->e_next = removelist;
removelist = ep;
}
}
}
}
void freenode( struct list *l, struct node *n, void (*freeentry)(void *) ) {
unlistnode(l, n);
if (freeentry != NULL) {
freeentry(n->d.v);
}
FREE(n);
}
/*
* Remove invalid whiteouts from the old tree.
* Remove unneeded leaves from the old tree.
* Remove directories from the lookup chains.
*/
void
removeoldleaves(void)
{
struct entry *ep, *nextep;
ino_t i, mydirino;
vprintf(stdout, "Mark entries to be removed.\n");
if ((ep = lookupino(WINO))) {
vprintf(stdout, "Delete whiteouts\n");
for ( ; ep != NULL; ep = nextep) {
nextep = ep->e_links;
mydirino = ep->e_parent->e_ino;
/*
* We remove all whiteouts that are in directories
* that have been removed or that have been dumped.
*/
if (TSTINO(mydirino, usedinomap) &&
!TSTINO(mydirino, dumpmap))
continue;
delwhiteout(ep);
freeentry(ep);
}
}
for (i = ROOTINO + 1; i < maxino; i++) {
ep = lookupino(i);
if (ep == NULL)
continue;
if (TSTINO(i, usedinomap))
continue;
for ( ; ep != NULL; ep = ep->e_links) {
dprintf(stdout, "%s: REMOVE\n", myname(ep));
if (ep->e_type == LEAF) {
removeleaf(ep);
freeentry(ep);
} else {
mktempname(ep);
deleteino(ep->e_ino);
ep->e_next = removelist;
removelist = ep;
}
}
}
}
static void
_insertdb(Db *db, Entry *e)
{
Entry *o, *ne;
ne = allocentry();
*ne = *e;
o = nil;
insertavl(db->avl, (Avl*)ne, (Avl**)&o);
if(o)
freeentry(o);
}
static void
_removedb(Db *db, char *name)
{
Entry *e, k;
memset(&k, 0, sizeof k);
k.name = name;
e = nil;
deleteavl(db->avl, (Avl*)&k, (Avl**)&e);
if(e)
freeentry(e);
}
/*
* Find unreferenced link names.
*/
void
findunreflinks(void)
{
struct entry *ep, *np;
ino_t i;
vprintf(stdout, "Find unreferenced names.\n");
for (i = ROOTINO; i < maxino; i++) {
ep = lookupino(i);
if (ep == NULL || ep->e_type == LEAF || TSTINO(i, dumpmap) == 0)
continue;
for (np = ep->e_entries; np != NULL; np = np->e_sibling) {
if (np->e_flags == 0) {
dprintf(stdout,
"%s: remove unreferenced name\n",
myname(np));
removeleaf(np);
freeentry(np);
}
}
}
/*
* Any leaves remaining in removed directories is unreferenced.
*/
for (ep = removelist; ep != NULL; ep = ep->e_next) {
for (np = ep->e_entries; np != NULL; np = np->e_sibling) {
if (np->e_type == LEAF) {
if (np->e_flags != 0)
badentry(np, "unreferenced with flags");
dprintf(stdout,
"%s: remove unreferenced name\n",
myname(np));
removeleaf(np);
freeentry(np);
}
}
}
}
/* ARGSUSED */
long
deletefile(char *name, ino_t ino, int type)
{
long descend = hflag ? GOOD : FAIL;
struct entry *ep;
if (TSTINO(ino, dumpmap) == 0)
return (descend);
ep = lookupname(name);
if (ep != NULL) {
ep->e_flags &= ~NEW;
ep->e_flags |= REMOVED;
if (ep->e_type != NODE)
freeentry(ep);
}
return (descend);
}
/*
* Remove old nodes (directories).
* Note that this routine runs in O(N*D) where:
* N is the number of directory entries to be removed.
* D is the maximum depth of the tree.
* If N == D this can be quite slow. If the list were
* topologically sorted, the deletion could be done in
* time O(N).
*/
void
removeoldnodes(void)
{
struct entry *ep, **prev;
long change;
vprintf(stdout, "Remove old nodes (directories).\n");
do {
change = 0;
prev = &removelist;
for (ep = removelist; ep != NULL; ep = *prev) {
if (ep->e_entries != NULL) {
prev = &ep->e_next;
continue;
}
*prev = ep->e_next;
removenode(ep);
freeentry(ep);
change++;
}
} while (change);
for (ep = removelist; ep != NULL; ep = ep->e_next)
badentry(ep, "cannot remove, non-empty");
}
/*
* For each directory entry on the incremental tape, determine which
* category it falls into as follows:
* KEEP - entries that are to be left alone.
* NEW - new entries to be added.
* EXTRACT - files that must be updated with new contents.
* LINK - new links to be added.
* Renames are done at the same time.
*/
long
nodeupdates(char *name, ino_t ino, int type)
{
struct entry *ep, *np, *ip;
long descend = GOOD;
int lookuptype = 0;
int key = 0;
/* key values */
# define ONTAPE 0x1 /* inode is on the tape */
# define INOFND 0x2 /* inode already exists */
# define NAMEFND 0x4 /* name already exists */
# define MODECHG 0x8 /* mode of inode changed */
/*
* This routine is called once for each element in the
* directory hierarchy, with a full path name.
* The "type" value is incorrectly specified as LEAF for
* directories that are not on the dump tape.
*
* Check to see if the file is on the tape.
*/
if (TSTINO(ino, dumpmap))
key |= ONTAPE;
/*
* Check to see if the name exists, and if the name is a link.
*/
np = lookupname(name);
if (np != NULL) {
key |= NAMEFND;
ip = lookupino(np->e_ino);
if (ip == NULL)
panic("corrupted symbol table\n");
if (ip != np)
lookuptype = LINK;
}
/*
* Check to see if the inode exists, and if one of its links
* corresponds to the name (if one was found).
*/
ip = lookupino(ino);
if (ip != NULL) {
key |= INOFND;
for (ep = ip->e_links; ep != NULL; ep = ep->e_links) {
if (ep == np) {
ip = ep;
break;
}
}
}
/*
* If both a name and an inode are found, but they do not
* correspond to the same file, then both the inode that has
* been found and the inode corresponding to the name that
* has been found need to be renamed. The current pathname
* is the new name for the inode that has been found. Since
* all files to be deleted have already been removed, the
* named file is either a now unneeded link, or it must live
* under a new name in this dump level. If it is a link, it
* can be removed. If it is not a link, it is given a
* temporary name in anticipation that it will be renamed
* when it is later found by inode number.
*/
if (((key & (INOFND|NAMEFND)) == (INOFND|NAMEFND)) && ip != np) {
if (lookuptype == LINK) {
removeleaf(np);
freeentry(np);
} else {
dprintf(stdout, "name/inode conflict, mktempname %s\n",
myname(np));
mktempname(np);
}
np = NULL;
key &= ~NAMEFND;
}
if ((key & ONTAPE) &&
(((key & INOFND) && ip->e_type != type) ||
((key & NAMEFND) && np->e_type != type)))
key |= MODECHG;
/*
* Decide on the disposition of the file based on its flags.
* Note that we have already handled the case in which
* a name and inode are found that correspond to different files.
* Thus if both NAMEFND and INOFND are set then ip == np.
*/
switch (key) {
/*
* A previously existing file has been found.
* Mark it as KEEP so that other links to the inode can be
* detected, and so that it will not be reclaimed by the search
* for unreferenced names.
*/
case INOFND|NAMEFND:
ip->e_flags |= KEEP;
dprintf(stdout, "[%s] %s: %s\n", keyval(key), name,
flagvalues(ip));
break;
/*
* A file on the tape has a name which is the same as a name
* corresponding to a different file in the previous dump.
* Since all files to be deleted have already been removed,
* this file is either a now unneeded link, or it must live
* under a new name in this dump level. If it is a link, it
* can simply be removed. If it is not a link, it is given a
* temporary name in anticipation that it will be renamed
* when it is later found by inode number (see INOFND case
* below). The entry is then treated as a new file.
*/
case ONTAPE|NAMEFND:
case ONTAPE|NAMEFND|MODECHG:
if (lookuptype == LINK) {
removeleaf(np);
freeentry(np);
} else {
mktempname(np);
}
/* FALLTHROUGH */
/*
* A previously non-existent file.
* Add it to the file system, and request its extraction.
* If it is a directory, create it immediately.
* (Since the name is unused there can be no conflict)
*/
case ONTAPE:
ep = addentry(name, ino, type);
if (type == NODE)
newnode(ep);
ep->e_flags |= NEW|KEEP;
dprintf(stdout, "[%s] %s: %s\n", keyval(key), name,
flagvalues(ep));
break;
/*
* A file with the same inode number, but a different
* name has been found. If the other name has not already
* been found (indicated by the KEEP flag, see above) then
* this must be a new name for the file, and it is renamed.
* If the other name has been found then this must be a
* link to the file. Hard links to directories are not
* permitted, and are either deleted or converted to
* symbolic links. Finally, if the file is on the tape,
* a request is made to extract it.
*/
case ONTAPE|INOFND:
if (type == LEAF && (ip->e_flags & KEEP) == 0)
ip->e_flags |= EXTRACT;
/* FALLTHROUGH */
case INOFND:
if ((ip->e_flags & KEEP) == 0) {
renameit(myname(ip), name);
moveentry(ip, name);
ip->e_flags |= KEEP;
dprintf(stdout, "[%s] %s: %s\n", keyval(key), name,
flagvalues(ip));
break;
}
if (ip->e_type == NODE) {
descend = FAIL;
fprintf(stderr,
"deleted hard link %s to directory %s\n",
name, myname(ip));
break;
}
ep = addentry(name, ino, type|LINK);
ep->e_flags |= NEW;
dprintf(stdout, "[%s] %s: %s|LINK\n", keyval(key), name,
flagvalues(ep));
break;
/*
* A previously known file which is to be updated. If it is a link,
* then all names referring to the previous file must be removed
* so that the subset of them that remain can be recreated.
*/
case ONTAPE|INOFND|NAMEFND:
if (lookuptype == LINK) {
removeleaf(np);
freeentry(np);
ep = addentry(name, ino, type|LINK);
if (type == NODE)
newnode(ep);
ep->e_flags |= NEW|KEEP;
dprintf(stdout, "[%s] %s: %s|LINK\n", keyval(key), name,
flagvalues(ep));
break;
}
if (type == LEAF && lookuptype != LINK)
np->e_flags |= EXTRACT;
np->e_flags |= KEEP;
dprintf(stdout, "[%s] %s: %s\n", keyval(key), name,
flagvalues(np));
break;
/*
* An inode is being reused in a completely different way.
* Normally an extract can simply do an "unlink" followed
* by a "creat". Here we must do effectively the same
* thing. The complications arise because we cannot really
* delete a directory since it may still contain files
* that we need to rename, so we delete it from the symbol
* table, and put it on the list to be deleted eventually.
* Conversely if a directory is to be created, it must be
* done immediately, rather than waiting until the
* extraction phase.
*/
case ONTAPE|INOFND|MODECHG:
case ONTAPE|INOFND|NAMEFND|MODECHG:
if (ip->e_flags & KEEP) {
badentry(ip, "cannot KEEP and change modes");
break;
}
if (ip->e_type == LEAF) {
/* changing from leaf to node */
for (ip = lookupino(ino); ip != NULL; ip = ip->e_links) {
if (ip->e_type != LEAF)
badentry(ip, "NODE and LEAF links to same inode");
removeleaf(ip);
freeentry(ip);
}
ip = addentry(name, ino, type);
newnode(ip);
} else {
/* changing from node to leaf */
if ((ip->e_flags & TMPNAME) == 0)
mktempname(ip);
deleteino(ip->e_ino);
ip->e_next = removelist;
removelist = ip;
ip = addentry(name, ino, type);
}
ip->e_flags |= NEW|KEEP;
dprintf(stdout, "[%s] %s: %s\n", keyval(key), name,
flagvalues(ip));
break;
/*
* A hard link to a directory that has been removed.
* Ignore it.
*/
case NAMEFND:
dprintf(stdout, "[%s] %s: Extraneous name\n", keyval(key),
name);
descend = FAIL;
break;
/*
* If we find a directory entry for a file that is not on
* the tape, then we must have found a file that was created
* while the dump was in progress. Since we have no contents
* for it, we discard the name knowing that it will be on the
* next incremental tape.
*/
case 0:
fprintf(stderr, "%s: (inode %ju) not found on tape\n",
name, (uintmax_t)ino);
break;
/*
* If any of these arise, something is grievously wrong with
* the current state of the symbol table.
*/
case INOFND|NAMEFND|MODECHG:
case NAMEFND|MODECHG:
case INOFND|MODECHG:
fprintf(stderr, "[%s] %s: inconsistent state\n", keyval(key),
name);
break;
/*
* These states "cannot" arise for any state of the symbol table.
*/
case ONTAPE|MODECHG:
case MODECHG:
default:
panic("[%s] %s: impossible state\n", keyval(key), name);
break;
}
return (descend);
}
static int
parse_log(void)
{
pkgentry_t *ent, *look;
avl_index_t where;
int num = 0;
int logfd;
struct stat stb;
int mlen = strlen(marker);
off_t realend;
ptrdiff_t off;
char *p, *q, *map;
logfd = open(PKGLOG, O_RDONLY);
if (logfd < 0) {
if (errno == ENOENT)
return (0);
progerr(gettext("cannot read "PKGLOG": %s"), strerror(errno));
exit(2);
}
if (fstat(logfd, &stb) != 0) {
progerr(gettext("cannot stat "PKGLOG": %s"), strerror(errno));
exit(2);
}
if (stb.st_size == 0) {
(void) close(logfd);
/* Force pkgdump && remove of the logfile. */
return (1);
}
map = mmap(0, stb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
logfd, 0);
(void) close(logfd);
if (map == (char *)-1) {
progerr(gettext("Cannot mmap the "PKGLOG": %s"),
strerror(errno));
exit(2);
}
cind = 0;
realend = stb.st_size;
if (memcmp(map + realend - mlen, marker, mlen) != 0) {
progerr(gettext(PKGLOG" is not complete"));
map[stb.st_size - 1] = '\0'; /* for strstr() */
realend = 0;
for (p = map; q = strstr(p, marker); ) {
if (q == map || q[-1] == '\n')
realend = q - map + mlen;
p = q + mlen;
}
progerr(gettext("Ignoring %ld bytes from log"),
(long)(stb.st_size - realend));
}
for (off = 0; off < realend; off += q - p) {
p = map + off;
q = memchr(p, '\n', realend - off);
if (q == NULL)
break;
q++;
num++;
if (p[0] == '#' || p[0] == '\n') {
if (memcmp(marker, p, mlen) == 0)
cind = 0;
else
handle_comments(p, q - p);
continue;
}
ent = parse_line(p + 1, q - (p + 1) - 1, p[0] != '-');
if (ent == NULL)
continue;
look = avl_find(list, ent, &where);
/*
* The log can be replayed; so any value of "look" is
* not unexpected.
*/
switch (p[0]) {
case '+':
case '=':
if (look != NULL)
swapentry(look, ent);
else
avl_insert(list, ent, where);
break;
case '-':
if (look != NULL) {
avl_remove(list, look);
freeentry(look);
}
freeentry(ent);
break;
default:
freeentry(ent);
progerr(gettext("log %d: bad line"), num);
break;
}
}
(void) munmap(map, stb.st_size);
/* Force pkgdump && remove of the logfile if there are no valid mods. */
return (num == 0 ? 1 : num);
}
static void
parse_contents(void)
{
int cnt;
pkgentry_t *ent, *e2;
avl_index_t where;
int num = 0;
struct stat stb;
ptrdiff_t off;
char *p, *q, *map;
pkgentry_t *lastentry = NULL;
int d;
int cntserrs = 0;
cnt = open(CONTENTS, O_RDONLY);
cind = 0;
if (cnt == -1) {
if (errno == ENOENT)
return;
exit(99);
}
if (fstat(cnt, &stb) != 0) {
(void) close(cnt);
exit(99);
}
if (stb.st_size == 0) {
(void) close(cnt);
return;
}
map = mmap(0, stb.st_size, PROT_READ, MAP_PRIVATE, cnt, 0);
(void) close(cnt);
if (map == (char *)-1)
return;
(void) madvise(map, stb.st_size, MADV_WILLNEED);
for (off = 0; off < stb.st_size; off += q - p) {
p = map + off;
q = memchr(p, '\n', stb.st_size - off);
if (q == NULL)
break;
q++;
num++;
if (p[0] == '#' || p[0] == '\n') {
handle_comments(p, q - p);
continue;
}
ent = parse_line(p, q - p - 1, B_TRUE);
if (ent == NULL) {
cntserrs++;
continue;
}
/*
* We save time by assuming the database is sorted; by
* using avl_insert_here(), building the tree is nearly free.
* lastentry always contains the last entry in the AVL tree.
*/
if (lastentry == NULL) {
avl_add(list, ent);
lastentry = ent;
} else if ((d = avlcmp(ent, lastentry)) == 1) {
avl_insert_here(list, ent, lastentry, AVL_AFTER);
lastentry = ent;
} else if (d == 0 ||
(e2 = avl_find(list, ent, &where)) != NULL) {
/*
* This can only happen if the contents file is bad;
* this can, e.g., happen with the old SQL contents DB,
* it didn't sort properly. Assume the first one
* is the correct one, but who knows?
*/
if (d == 0)
e2 = lastentry;
if (strcmp(ent->line, e2->line) != 0) {
progerr(gettext("two entries for %.*s"),
ent->pathlen, ent->line);
cntserrs++;
}
freeentry(ent);
} else {
/* Out of order: not an error for us, really. */
progerr(gettext("bad read of contents file"));
logerr(gettext("pathname: Unknown"));
logerr(gettext(
"problem: unable to read pathname field"));
if (one_shot)
exit(2);
avl_insert(list, ent, where);
}
}
cind = 0;
(void) munmap(map, stb.st_size);
/* By default, we ignore bad lines, keep them in a copy. */
if (cntserrs > 0 && stb.st_nlink == 1) {
char bcf[sizeof (BADCONTENTS)];
(void) strcpy(bcf, BADCONTENTS);
if (mktemp(bcf) != NULL) {
(void) link(CONTENTS, bcf);
syslog(LOG_WARNING, "A bad contents file was saved: %s",
bcf);
}
}
}
/*
* Entry: a single line without the NEWLINE
* Return: the package entry with the path determined.
*/
static pkgentry_t *
parse_line(char *buf, int blen, boolean_t full)
{
char *t;
pkgentry_t *p;
int nfields;
p = umem_cache_alloc(ecache, UMEM_NOFAIL);
buf = p->line = mystrcpy(buf, blen + 1);
p->len = blen + 1;
t = buf;
while (!IS_ST0Q[(unsigned char)*t++])
;
p->pathlen = t - buf - 1;
if (p->pathlen == 0 || p->pathlen >= PATH_MAX) {
progerr("bad entry read in contents file");
logerr("pathname: Unknown");
logerr("problem: unable to read pathname field");
if (one_shot)
exit(2);
}
if (t[-1] == '=')
while (!IS_ST0[(unsigned char)*t++])
;
/* Partial as found in the "-" entries for log */
if (t[-1] == '\0') {
if (full)
goto badline;
p->pkgoff = -1;
return (p);
}
switch (*t) {
case '?':
nfields = 0;
break;
case 's':
case 'l':
/* Fields: class */
nfields = 1;
break;
case 'p':
case 'x':
case 'd':
/* class mode owner group */
nfields = 4;
break;
case 'f':
case 'e':
case 'v':
/* class mode owner group size csum time */
nfields = 7;
break;
case 'c':
case 'b':
/* class major minor mode owner group */
nfields = 6;
break;
default:
progerr("bad entry read in contents file");
logerr("pathname: %.*s", p->pathlen, p->line);
logerr("problem: unknown ftype");
freeentry(p);
if (one_shot)
exit(2);
return (NULL);
}
p->pkgoff = t + fieldoff(t, nfields + 1) - buf;
if (p->line[p->pkgoff] != '\0' || p->pkgoff == p->len - 1)
return (p);
badline:
progerr(gettext("bad entry read in contents file"));
logerr(gettext("pathname: Unknown"));
logerr(gettext("problem: unknown ftype"));
freeentry(p);
if (one_shot)
exit(2);
return (NULL);
}
