/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred,
struct thread *td)
{
struct vnode *ovp = vp;
int32_t lastblock;
struct inode *oip;
int32_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
uint32_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct m_ext2fs *fs;
struct buf *bp;
int offset, size, level;
e4fs_daddr_t count, nblocks, blocksreleased = 0;
int error, i, allerror;
off_t osize;
#ifdef INVARIANTS
struct bufobj *bo;
#endif
oip = VTOI(ovp);
#ifdef INVARIANTS
bo = &ovp->v_bufobj;
#endif
ASSERT_VOP_LOCKED(vp, "ext2_truncate");
if (length < 0)
return (EINVAL);
if (ovp->v_type == VLNK &&
oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
#ifdef INVARIANTS
if (length != 0)
panic("ext2_truncate: partial truncate of symlink");
#endif
bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 0));
}
fs = oip->i_e2fs;
osize = oip->i_size;
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
if (length > oip->i_e2fs->e2fs_maxfilesize)
return (EFBIG);
vnode_pager_setsize(ovp, length);
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, flags);
if (error) {
vnode_pager_setsize(vp, osize);
return (error);
}
oip->i_size = length;
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, !DOINGASYNC(ovp)));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessible again because
* of subsequent file growth.
*/
/* I don't understand the comment above */
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset, cred, &bp, flags);
if (error)
return (error);
oip->i_size = length;
size = blksize(fs, oip, lbn);
bzero((char *)bp->b_data + offset, (u_int)(size - offset));
allocbuf(bp, size);
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->e2fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->e2fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[NDADDR + level] = oip->i_ib[level];
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
}
for (i = 0; i < NDADDR; i++) {
oldblks[i] = oip->i_db[i];
if (i > lastblock)
oip->i_db[i] = 0;
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = ext2_update(ovp, !DOINGASYNC(ovp));
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < NDADDR; i++) {
newblks[i] = oip->i_db[i];
oip->i_db[i] = oldblks[i];
}
for (i = 0; i < NIADDR; i++) {
newblks[NDADDR + i] = oip->i_ib[i];
oip->i_ib[i] = oldblks[NDADDR + i];
}
oip->i_size = osize;
error = vtruncbuf(ovp, cred, length, (int)fs->e2fs_bsize);
if (error && (allerror == 0))
allerror = error;
vnode_pager_setsize(ovp, length);
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ext2_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ext2_blkfree(oip, bn, fs->e2fs_fsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ext2_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ext2_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ext2_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef INVARIANTS
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("itrunc1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("itrunc2");
BO_LOCK(bo);
if (length == 0 && (bo->bo_dirty.bv_cnt != 0 ||
bo->bo_clean.bv_cnt != 0))
panic("itrunc3");
BO_UNLOCK(bo);
#endif /* INVARIANTS */
/*
* Put back the real size.
*/
oip->i_size = length;
if (oip->i_blocks >= blocksreleased)
oip->i_blocks -= blocksreleased;
else /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
vnode_pager_setsize(ovp, length);
return (allerror);
}
void
source(int argc, char *argv[])
{
struct stat stb;
static BUF buffer;
BUF *bp;
off_t i;
int amt, fd, haderr, indx, result;
char *last, *name, buf[BUFSIZ];
for (indx = 0; indx < argc; ++indx) {
name = argv[indx];
if ((fd = open(name, O_RDONLY, 0)) < 0)
goto syserr;
if (fstat(fd, &stb)) {
syserr: run_err("%s: %s", name, strerror(errno));
goto next;
}
switch (stb.st_mode & S_IFMT) {
case S_IFREG:
break;
case S_IFDIR:
if (iamrecursive) {
rsource(name, &stb);
goto next;
}
/* FALLTHROUGH */
default:
run_err("%s: not a regular file", name);
goto next;
}
if ((last = strrchr(name, '/')) == NULL)
last = name;
else
++last;
if (pflag) {
/*
* Make it compatible with possible future
* versions expecting microseconds.
*/
(void)snprintf(buf, sizeof(buf), "T%ld 0 %ld 0\n",
(long)stb.st_mtim.tv_sec,
(long)stb.st_atim.tv_sec);
(void)write(rem, buf, strlen(buf));
if (response() < 0)
goto next;
}
#define MODEMASK (S_ISUID|S_ISGID|S_ISTXT|S_IRWXU|S_IRWXG|S_IRWXO)
(void)snprintf(buf, sizeof(buf), "C%04o %jd %s\n",
stb.st_mode & MODEMASK, (intmax_t)stb.st_size, last);
(void)write(rem, buf, strlen(buf));
if (response() < 0)
goto next;
if ((bp = allocbuf(&buffer, fd, BUFSIZ)) == NULL) {
next: if (fd >= 0)
(void)close(fd);
continue;
}
/* Keep writing after an error so that we stay sync'd up. */
for (haderr = i = 0; i < stb.st_size; i += bp->cnt) {
amt = bp->cnt;
if (i + amt > stb.st_size)
amt = stb.st_size - i;
if (!haderr) {
result = read(fd, bp->buf, amt);
if (result != amt)
haderr = result >= 0 ? EIO : errno;
}
if (haderr)
(void)write(rem, bp->buf, amt);
else {
result = write(rem, bp->buf, amt);
if (result != amt)
haderr = result >= 0 ? EIO : errno;
}
}
if (close(fd) && !haderr)
haderr = errno;
if (!haderr)
(void)write(rem, "", 1);
else
run_err("%s: %s", name, strerror(haderr));
(void)response();
}
}
void
source(int argc, char **argv)
{
struct stat stb;
static BUF buffer;
BUF *bp;
off_t i, statbytes;
size_t amt;
int fd = -1, haderr, indx;
char *last, *name, buf[16384], encname[MAXPATHLEN];
int len;
for (indx = 0; indx < argc; ++indx) {
fd = -1;
name = argv[indx];
statbytes = 0;
len = strlen(name);
while (len > 1 && name[len-1] == '/')
name[--len] = '\0';
if ((fd = open(name, O_RDONLY|O_NONBLOCK, 0)) < 0)
goto syserr;
if (strchr(name, '\n') != NULL) {
strvisx(encname, name, len, VIS_NL);
name = encname;
}
if (fstat(fd, &stb) < 0) {
syserr: run_err("%s: %s", name, strerror(errno));
goto next;
}
if (stb.st_size < 0) {
run_err("%s: %s", name, "Negative file size");
goto next;
}
unset_nonblock(fd);
switch (stb.st_mode & S_IFMT) {
case S_IFREG:
break;
case S_IFDIR:
if (iamrecursive) {
rsource(name, &stb);
goto next;
}
/* FALLTHROUGH */
default:
run_err("%s: not a regular file", name);
goto next;
}
if ((last = strrchr(name, '/')) == NULL)
last = name;
else
++last;
curfile = last;
if (pflag) {
/*
* Make it compatible with possible future
* versions expecting microseconds.
*/
(void) snprintf(buf, sizeof buf, "T%lu 0 %lu 0\n",
(u_long) (stb.st_mtime < 0 ? 0 : stb.st_mtime),
(u_long) (stb.st_atime < 0 ? 0 : stb.st_atime));
if (verbose_mode) {
fprintf(stderr, "File mtime %ld atime %ld\n",
(long)stb.st_mtime, (long)stb.st_atime);
fprintf(stderr, "Sending file timestamps: %s",
buf);
}
(void) atomicio(vwrite, remout, buf, strlen(buf));
if (response() < 0)
goto next;
}
#define FILEMODEMASK (S_ISUID|S_ISGID|S_IRWXU|S_IRWXG|S_IRWXO)
snprintf(buf, sizeof buf, "C%04o %lld %s\n",
(u_int) (stb.st_mode & FILEMODEMASK),
(long long)stb.st_size, last);
if (verbose_mode) {
fprintf(stderr, "Sending file modes: %s", buf);
}
(void) atomicio(vwrite, remout, buf, strlen(buf));
if (response() < 0)
goto next;
if ((bp = allocbuf(&buffer, fd, COPY_BUFLEN)) == NULL) {
next: if (fd != -1) {
(void) close(fd);
fd = -1;
}
continue;
}
if (showprogress)
start_progress_meter(curfile, stb.st_size, &statbytes);
set_nonblock(remout);
for (haderr = i = 0; i < stb.st_size; i += bp->cnt) {
amt = bp->cnt;
if (i + (off_t)amt > stb.st_size)
amt = stb.st_size - i;
if (!haderr) {
if (atomicio(read, fd, bp->buf, amt) != amt)
haderr = errno;
}
/* Keep writing after error to retain sync */
if (haderr) {
(void)atomicio(vwrite, remout, bp->buf, amt);
continue;
}
if (atomicio6(vwrite, remout, bp->buf, amt, scpio,
&statbytes) != amt)
haderr = errno;
}
unset_nonblock(remout);
if (showprogress)
stop_progress_meter();
if (fd != -1) {
if (close(fd) < 0 && !haderr)
haderr = errno;
fd = -1;
}
if (!haderr)
(void) atomicio(vwrite, remout, empty, 1);
else
run_err("%s: %s", name, strerror(haderr));
(void) response();
}
}
void
sink(int argc, char **argv)
{
static BUF buffer;
struct stat stb;
enum {
YES, NO, DISPLAYED
} wrerr;
BUF *bp;
off_t i;
size_t j, count;
int amt, exists, first, ofd;
mode_t mode, omode, mask;
off_t size, statbytes;
int setimes, targisdir, wrerrno = 0;
char ch, *cp, *np, *targ, *vect[1], buf[16384];
const char *why;
struct timeval tv[2];
#define atime tv[0]
#define mtime tv[1]
#define SCREWUP(str) { why = str; goto screwup; }
setimes = targisdir = 0;
mask = umask(0);
if (!pflag)
(void) umask(mask);
if (argc != 1) {
run_err("ambiguous target");
exit(1);
}
targ = *argv;
if (targetshouldbedirectory)
verifydir(targ);
(void) atomicio(vwrite, remout, empty, 1);
if (stat(targ, &stb) == 0 && S_ISDIR(stb.st_mode))
targisdir = 1;
for (first = 1;; first = 0) {
cp = buf;
if (atomicio(read, remin, cp, 1) != 1)
return;
if (*cp++ == '\n')
SCREWUP("unexpected <newline>");
do {
if (atomicio(read, remin, &ch, sizeof(ch)) != sizeof(ch))
SCREWUP("lost connection");
*cp++ = ch;
} while (cp < &buf[sizeof(buf) - 1] && ch != '\n');
*cp = 0;
if (verbose_mode)
fprintf(stderr, "Sink: %s", buf);
if (buf[0] == '\01' || buf[0] == '\02') {
if (iamremote == 0)
(void) atomicio(vwrite, STDERR_FILENO,
buf + 1, strlen(buf + 1));
if (buf[0] == '\02')
exit(1);
++errs;
continue;
}
if (buf[0] == 'E') {
(void) atomicio(vwrite, remout, empty, 1);
return;
}
if (ch == '\n')
*--cp = 0;
cp = buf;
if (*cp == 'T') {
setimes++;
cp++;
mtime.tv_sec = strtol(cp, &cp, 10);
if (!cp || *cp++ != ' ')
SCREWUP("mtime.sec not delimited");
mtime.tv_usec = strtol(cp, &cp, 10);
if (!cp || *cp++ != ' ')
SCREWUP("mtime.usec not delimited");
atime.tv_sec = strtol(cp, &cp, 10);
if (!cp || *cp++ != ' ')
SCREWUP("atime.sec not delimited");
atime.tv_usec = strtol(cp, &cp, 10);
if (!cp || *cp++ != '\0')
SCREWUP("atime.usec not delimited");
(void) atomicio(vwrite, remout, empty, 1);
continue;
}
if (*cp != 'C' && *cp != 'D') {
/*
* Check for the case "rcp remote:foo\* local:bar".
* In this case, the line "No match." can be returned
* by the shell before the rcp command on the remote is
* executed so the ^Aerror_message convention isn't
* followed.
*/
if (first) {
run_err("%s", cp);
exit(1);
}
SCREWUP("expected control record");
}
mode = 0;
for (++cp; cp < buf + 5; cp++) {
if (*cp < '0' || *cp > '7')
SCREWUP("bad mode");
mode = (mode << 3) | (*cp - '0');
}
if (*cp++ != ' ')
SCREWUP("mode not delimited");
for (size = 0; isdigit((unsigned char)*cp);)
size = size * 10 + (*cp++ - '0');
if (*cp++ != ' ')
SCREWUP("size not delimited");
if ((strchr(cp, '/') != NULL) || (strcmp(cp, "..") == 0)) {
run_err("error: unexpected filename: %s", cp);
exit(1);
}
if (targisdir) {
static char *namebuf;
static size_t cursize;
size_t need;
need = strlen(targ) + strlen(cp) + 250;
if (need > cursize) {
if (namebuf)
xfree(namebuf);
namebuf = xmalloc(need);
cursize = need;
}
(void) snprintf(namebuf, need, "%s%s%s", targ,
strcmp(targ, "/") ? "/" : "", cp);
np = namebuf;
} else
np = targ;
curfile = cp;
exists = stat(np, &stb) == 0;
if (buf[0] == 'D') {
int mod_flag = pflag;
if (!iamrecursive)
SCREWUP("received directory without -r");
if (exists) {
if (!S_ISDIR(stb.st_mode)) {
errno = ENOTDIR;
goto bad;
}
if (pflag)
(void) chmod(np, mode);
} else {
/* Handle copying from a read-only
directory */
mod_flag = 1;
if (mkdir(np, mode | S_IRWXU) < 0)
goto bad;
}
vect[0] = xstrdup(np);
sink(1, vect);
if (setimes) {
setimes = 0;
if (utimes(vect[0], tv) < 0)
run_err("%s: set times: %s",
vect[0], strerror(errno));
}
if (mod_flag)
(void) chmod(vect[0], mode);
if (vect[0])
xfree(vect[0]);
continue;
}
omode = mode;
mode |= S_IWRITE;
if ((ofd = open(np, O_WRONLY|O_CREAT, mode)) < 0) {
bad: run_err("%s: %s", np, strerror(errno));
continue;
}
(void) atomicio(vwrite, remout, empty, 1);
if ((bp = allocbuf(&buffer, ofd, COPY_BUFLEN)) == NULL) {
(void) close(ofd);
continue;
}
cp = bp->buf;
wrerr = NO;
statbytes = 0;
if (showprogress)
start_progress_meter(curfile, size, &statbytes);
set_nonblock(remin);
for (count = i = 0; i < size; i += bp->cnt) {
amt = bp->cnt;
if (i + amt > size)
amt = size - i;
count += amt;
do {
j = atomicio6(read, remin, cp, amt,
scpio, &statbytes);
if (j == 0) {
run_err("%s", j != EPIPE ?
strerror(errno) :
"dropped connection");
exit(1);
}
amt -= j;
cp += j;
} while (amt > 0);
if (count == bp->cnt) {
/* Keep reading so we stay sync'd up. */
if (wrerr == NO) {
if (atomicio(vwrite, ofd, bp->buf,
count) != count) {
wrerr = YES;
wrerrno = errno;
}
}
count = 0;
cp = bp->buf;
}
}
unset_nonblock(remin);
if (showprogress)
stop_progress_meter();
if (count != 0 && wrerr == NO &&
atomicio(vwrite, ofd, bp->buf, count) != count) {
wrerr = YES;
wrerrno = errno;
}
if (wrerr == NO && (!exists || S_ISREG(stb.st_mode)) &&
ftruncate(ofd, size) != 0) {
run_err("%s: truncate: %s", np, strerror(errno));
wrerr = DISPLAYED;
}
if (pflag) {
if (exists || omode != mode)
if (fchmod(ofd, omode)) {
run_err("%s: set mode: %s",
np, strerror(errno));
wrerr = DISPLAYED;
}
} else {
if (!exists && omode != mode)
if (fchmod(ofd, omode & ~mask)) {
run_err("%s: set mode: %s",
np, strerror(errno));
wrerr = DISPLAYED;
}
}
if (close(ofd) == -1) {
wrerr = YES;
wrerrno = errno;
}
(void) response();
if (setimes && wrerr == NO) {
setimes = 0;
if (utimes(np, tv) < 0) {
run_err("%s: set times: %s",
np, strerror(errno));
wrerr = DISPLAYED;
}
}
switch (wrerr) {
case YES:
run_err("%s: %s", np, strerror(wrerrno));
break;
case NO:
(void) atomicio(vwrite, remout, empty, 1);
break;
case DISPLAYED:
break;
}
}
screwup:
run_err("protocol error: %s", why);
exit(1);
}
void
source(int argc, char **argv)
{
struct stat stb;
static BUF buffer;
BUF *bp;
off_t i, amt, statbytes;
size_t result;
int fd = -1, haderr, indx;
char *last, *name, buf[2048];
int len;
for (indx = 0; indx < argc; ++indx) {
name = argv[indx];
statbytes = 0;
len = strlen(name);
while (len > 1 && name[len-1] == '/')
name[--len] = '\0';
if (strchr(name, '\n') != NULL) {
run_err("%s: skipping, filename contains a newline",
name);
goto next;
}
if ((fd = open(name, O_RDONLY, 0)) < 0)
goto syserr;
if (fstat(fd, &stb) < 0) {
syserr: run_err("%s: %s", name, strerror(errno));
goto next;
}
switch (stb.st_mode & S_IFMT) {
case S_IFREG:
break;
case S_IFDIR:
if (iamrecursive) {
rsource(name, &stb);
goto next;
}
/* FALLTHROUGH */
default:
run_err("%s: not a regular file", name);
goto next;
}
if ((last = strrchr(name, '/')) == NULL)
last = name;
else
++last;
curfile = last;
if (pflag) {
/*
* Make it compatible with possible future
* versions expecting microseconds.
*/
(void) snprintf(buf, sizeof buf, "T%lu 0 %lu 0\n",
(u_long) stb.st_mtime,
(u_long) stb.st_atime);
(void) atomicio(vwrite, remout, buf, strlen(buf));
if (response() < 0)
goto next;
}
#define FILEMODEMASK (S_ISUID|S_ISGID|S_IRWXU|S_IRWXG|S_IRWXO)
snprintf(buf, sizeof buf, "C%04o %lld %s\n",
(u_int) (stb.st_mode & FILEMODEMASK),
(long long)stb.st_size, last);
if (verbose_mode) {
fprintf(stderr, "Sending file modes: %s", buf);
}
(void) atomicio(vwrite, remout, buf, strlen(buf));
if (response() < 0)
goto next;
if ((bp = allocbuf(&buffer, fd, 2048)) == NULL) {
next: if (fd != -1) {
(void) close(fd);
fd = -1;
}
continue;
}
#if PROGRESS_METER
if (showprogress)
start_progress_meter(curfile, stb.st_size, &statbytes);
#endif
/* Keep writing after an error so that we stay sync'd up. */
for (haderr = i = 0; i < stb.st_size; i += bp->cnt) {
amt = bp->cnt;
if (i + amt > stb.st_size)
amt = stb.st_size - i;
if (!haderr) {
result = atomicio(read, fd, bp->buf, amt);
if (result != amt)
haderr = errno;
}
if (haderr)
(void) atomicio(vwrite, remout, bp->buf, amt);
else {
result = atomicio(vwrite, remout, bp->buf, amt);
if (result != amt)
haderr = errno;
statbytes += result;
}
}
#ifdef PROGRESS_METER
if (showprogress)
stop_progress_meter();
#endif
if (fd != -1) {
if (close(fd) < 0 && !haderr)
haderr = errno;
fd = -1;
}
if (!haderr)
(void) atomicio(vwrite, remout, "", 1);
else
run_err("%s: %s", name, strerror(haderr));
(void) response();
}
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp = vp;
ufs_daddr_t lastblock;
struct inode *oip;
ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, blocksreleased = 0;
int i;
int aflags, error, allerror;
off_t osize;
oip = VTOI(ovp);
fs = oip->i_fs;
if (length < 0)
return (EINVAL);
if (length > fs->fs_maxfilesize)
return (EFBIG);
if (ovp->v_type == VLNK &&
(oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif /* DIAGNOSTIC */
bzero((char *)&oip->i_shortlink, (uint)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 0));
}
if (fs->fs_ronly)
panic("ffs_truncate: read-only filesystem");
#ifdef QUOTA
error = ufs_getinoquota(oip);
if (error)
return (error);
#endif
ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
if (DOINGSOFTDEP(ovp)) {
if (length > 0 || softdep_slowdown(ovp)) {
/*
* If a file is only partially truncated, then
* we have to clean up the data structures
* describing the allocation past the truncation
* point. Finding and deallocating those structures
* is a lot of work. Since partial truncation occurs
* rarely, we solve the problem by syncing the file
* so that it will have no data structures left.
*/
if ((error = VOP_FSYNC(ovp, MNT_WAIT, 0)) != 0)
return (error);
} else {
#ifdef QUOTA
(void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0);
#endif
softdep_setup_freeblocks(oip, length);
vinvalbuf(ovp, 0, 0, 0);
nvnode_pager_setsize(ovp, 0, fs->fs_bsize, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 0));
}
}
osize = oip->i_size;
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*
* nvextendbuf() only breads the old buffer. The blocksize
* of the new buffer must be specified so it knows how large
* to make the VM object.
*/
if (osize < length) {
nvextendbuf(vp, osize, length,
blkoffsize(fs, oip, osize), /* oblksize */
blkoffresize(fs, length), /* nblksize */
blkoff(fs, osize),
blkoff(fs, length),
0);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
/* BALLOC will reallocate the fragment at the old EOF */
error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
if (error)
return (error);
oip->i_size = length;
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 1));
}
/*
* Shorten the size of the file.
*
* NOTE: The block size specified in nvtruncbuf() is the blocksize
* of the buffer containing length prior to any reallocation
* of the block.
*/
allerror = nvtruncbuf(ovp, length, blkoffsize(fs, oip, length),
blkoff(fs, length), 0);
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
if (error)
return (error);
/*
* When we are doing soft updates and the UFS_BALLOC
* above fills in a direct block hole with a full sized
* block that will be truncated down to a fragment below,
* we must flush out the block dependency with an FSYNC
* so that we do not get a soft updates inconsistency
* when we create the fragment below.
*
* nvtruncbuf() may have re-dirtied the underlying block
* as part of its truncation zeroing code. To avoid a
* 'locking against myself' panic in the second fsync we
* can simply undirty the bp since the redirtying was
* related to areas of the buffer that we are going to
* throw away anyway, and we will b*write() the remainder
* anyway down below.
*/
if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize) {
bundirty(bp);
error = VOP_FSYNC(ovp, MNT_WAIT, 0);
if (error) {
bdwrite(bp);
return (error);
}
}
oip->i_size = length;
size = blksize(fs, oip, lbn);
#if 0
/* remove - nvtruncbuf deals with this */
if (ovp->v_type != VDIR)
bzero((char *)bp->b_data + offset,
(uint)(size - offset));
#endif
/* Kirk's code has reallocbuf(bp, size, 1) here */
allocbuf(bp, size);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ffs_indirtrunc below.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
for (i = NDADDR - 1; i > lastblock; i--)
oip->i_db[i] = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
error = ffs_update(ovp, 1);
if (error && allerror == 0)
allerror = error;
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
oip->i_size = osize;
if (error && allerror == 0)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ffs_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ffs_blkfree(oip, bn, fs->fs_bsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ffs_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ffs_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ffs_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("ffs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("ffs_truncate2");
if (length == 0 && !RB_EMPTY(&ovp->v_rbdirty_tree))
panic("ffs_truncate3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = length;
oip->i_blocks -= blocksreleased;
if (oip->i_blocks < 0) /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
#ifdef QUOTA
(void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
}
/*
* Truncate the inode ip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(vnode *vp, off_t length, int flags, Ucred *cred)
{
print("HARVEY TODO: %s\n", __func__);
#if 0
struct inode *ip;
ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR];
ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR];
ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR];
ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno;
struct bufobj *bo;
struct fs *fs;
struct buf *bp;
struct ufsmount *ump;
int softdeptrunc, journaltrunc;
int needextclean, extblocks;
int offset, size, level, nblocks;
int i, error, allerror, indiroff, waitforupdate;
off_t osize;
ip = VTOI(vp);
ump = VFSTOUFS(vp->v_mount);
fs = ump->um_fs;
bo = &vp->v_bufobj;
ASSERT_VOP_LOCKED(vp, "ffs_truncate");
if (length < 0)
return (EINVAL);
if (length > fs->fs_maxfilesize)
return (EFBIG);
#ifdef QUOTA
error = getinoquota(ip);
if (error)
return (error);
#endif
/*
* Historically clients did not have to specify which data
* they were truncating. So, if not specified, we assume
* traditional behavior, e.g., just the normal data.
*/
if ((flags & (IO_EXT | IO_NORMAL)) == 0)
flags |= IO_NORMAL;
if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
flags |= IO_SYNC;
waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp);
/*
* If we are truncating the extended-attributes, and cannot
* do it with soft updates, then do it slowly here. If we are
* truncating both the extended attributes and the file contents
* (e.g., the file is being unlinked), then pick it off with
* soft updates below.
*/
allerror = 0;
needextclean = 0;
softdeptrunc = 0;
journaltrunc = DOINGSUJ(vp);
if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
softdeptrunc = !softdep_slowdown(vp);
extblocks = 0;
datablocks = DIP(ip, i_blocks);
if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
datablocks -= extblocks;
}
if ((flags & IO_EXT) && extblocks > 0) {
if (length != 0)
panic("ffs_truncate: partial trunc of extdata");
if (softdeptrunc || journaltrunc) {
if ((flags & IO_NORMAL) == 0)
goto extclean;
needextclean = 1;
} else {
if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
#ifdef QUOTA
(void) chkdq(ip, -extblocks, NOCRED, 0);
#endif
vinvalbuf(vp, V_ALT, 0, 0);
vn_pages_remove(vp,
OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
osize = ip->i_din2->di_extsize;
ip->i_din2->di_blocks -= extblocks;
ip->i_din2->di_extsize = 0;
for (i = 0; i < UFS_NXADDR; i++) {
oldblks[i] = ip->i_din2->di_extb[i];
ip->i_din2->di_extb[i] = 0;
}
ip->i_flag |= IN_CHANGE;
if ((error = ffs_update(vp, waitforupdate)))
return (error);
for (i = 0; i < UFS_NXADDR; i++) {
if (oldblks[i] == 0)
continue;
ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i],
sblksize(fs, osize, i), ip->i_number,
vp->v_type, nil);
}
}
}
if ((flags & IO_NORMAL) == 0)
return (0);
if (vp->v_type == VLNK &&
(ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
datablocks == 0)) {
#ifdef INVARIANTS
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif
bzero(SHORTLINK(ip), (uint)ip->i_size);
ip->i_size = 0;
DIP_SET(ip, i_size, 0);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (needextclean)
goto extclean;
return (ffs_update(vp, waitforupdate));
}
if (ip->i_size == length) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (needextclean)
goto extclean;
return (ffs_update(vp, 0));
}
if (fs->fs_ronly)
panic("ffs_truncate: read-only filesystem");
if (IS_SNAPSHOT(ip))
ffs_snapremove(vp);
vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
osize = ip->i_size;
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
vnode_pager_setsize(vp, length);
flags |= BA_CLRBUF;
error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
if (error) {
vnode_pager_setsize(vp, osize);
return (error);
}
ip->i_size = length;
DIP_SET(ip, i_size, length);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(vp))
bdwrite(bp);
else
bawrite(bp);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(vp, waitforupdate));
}
/*
* Lookup block number for a given offset. Zero length files
* have no blocks, so return a blkno of -1.
*/
lbn = lblkno(fs, length - 1);
if (length == 0) {
blkno = -1;
} else if (lbn < UFS_NDADDR) {
blkno = DIP(ip, i_db[lbn]);
} else {
error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize,
cred, BA_METAONLY, &bp);
if (error)
return (error);
indiroff = (lbn - UFS_NDADDR) % NINDIR(fs);
if (I_IS_UFS1(ip))
blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
else
blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
/*
* If the block number is non-zero, then the indirect block
* must have been previously allocated and need not be written.
* If the block number is zero, then we may have allocated
* the indirect block and hence need to write it out.
*/
if (blkno != 0)
brelse(bp);
else if (flags & IO_SYNC)
bwrite(bp);
else
bdwrite(bp);
}
/*
* If the block number at the new end of the file is zero,
* then we must allocate it to ensure that the last block of
* the file is allocated. Soft updates does not handle this
* case, so here we have to clean up the soft updates data
* structures describing the allocation past the truncation
* point. Finding and deallocating those structures is a lot of
* work. Since partial truncation with a hole at the end occurs
* rarely, we solve the problem by syncing the file so that it
* will have no soft updates data structures left.
*/
if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
if (blkno != 0 && DOINGSOFTDEP(vp)) {
if (softdeptrunc == 0 && journaltrunc == 0) {
/*
* If soft updates cannot handle this truncation,
* clean up soft dependency data structures and
* fall through to the synchronous truncation.
*/
if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
} else {
flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
if (journaltrunc)
softdep_journal_freeblocks(ip, cred, length,
flags);
else
softdep_setup_freeblocks(ip, length, flags);
ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
if (journaltrunc == 0) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
error = ffs_update(vp, 0);
}
return (error);
}
}
/*
* Shorten the size of the file. If the last block of the
* shortened file is unallocated, we must allocate it.
* Additionally, if the file is not being truncated to a
* block boundary, the contents of the partial block
* following the end of the file must be zero'ed in
* case it ever becomes accessible again because of
* subsequent file growth. Directories however are not
* zero'ed as they should grow back initialized to empty.
*/
offset = blkoff(fs, length);
if (blkno != 0 && offset == 0) {
ip->i_size = length;
DIP_SET(ip, i_size, length);
} else {
lbn = lblkno(fs, length);
flags |= BA_CLRBUF;
error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
if (error)
return (error);
/*
* When we are doing soft updates and the UFS_BALLOC
* above fills in a direct block hole with a full sized
* block that will be truncated down to a fragment below,
* we must flush out the block dependency with an FSYNC
* so that we do not get a soft updates inconsistency
* when we create the fragment below.
*/
if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
(error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
ip->i_size = length;
DIP_SET(ip, i_size, length);
size = blksize(fs, ip, lbn);
if (vp->v_type != VDIR && offset != 0)
bzero((char *)bp->b_data + offset,
(uint)(size - offset));
/* Kirk's code has reallocbuf(bp, size, 1) here */
allocbuf(bp, size);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(vp))
bdwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - UFS_NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ffs_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]);
if (lastiblock[level] < 0) {
DIP_SET(ip, i_ib[level], 0);
lastiblock[level] = -1;
}
}
for (i = 0; i < UFS_NDADDR; i++) {
oldblks[i] = DIP(ip, i_db[i]);
if (i > lastblock)
DIP_SET(ip, i_db[i], 0);
}
ip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = ffs_update(vp, waitforupdate);
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < UFS_NDADDR; i++) {
newblks[i] = DIP(ip, i_db[i]);
DIP_SET(ip, i_db[i], oldblks[i]);
}
for (i = 0; i < UFS_NIADDR; i++) {
newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]);
DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]);
}
ip->i_size = osize;
DIP_SET(ip, i_size, osize);
error = vtruncbuf(vp, cred, length, fs->fs_bsize);
if (error && (allerror == 0))
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -UFS_NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = DIP(ip, i_ib[level]);
if (bn != 0) {
error = ffs_indirtrunc(ip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
DIP_SET(ip, i_ib[level], 0);
ffs_blkfree(ump, fs, ump->um_devvp, bn,
fs->fs_bsize, ip->i_number,
vp->v_type, nil);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = UFS_NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = DIP(ip, i_db[i]);
if (bn == 0)
continue;
DIP_SET(ip, i_db[i], 0);
bsize = blksize(fs, ip, i);
ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number,
vp->v_type, nil);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = DIP(ip, i_db[lastblock]);
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, ip, lastblock);
ip->i_size = length;
DIP_SET(ip, i_size, length);
newspace = blksize(fs, ip, lastblock);
if (newspace == 0)
panic("ffs_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ffs_blkfree(ump, fs, ump->um_devvp, bn,
oldspace - newspace, ip->i_number, vp->v_type, nil);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef INVARIANTS
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level]))
panic("ffs_truncate1");
for (i = 0; i < UFS_NDADDR; i++)
if (newblks[i] != DIP(ip, i_db[i]))
panic("ffs_truncate2");
BO_LOCK(bo);
if (length == 0 &&
(fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
(bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
panic("ffs_truncate3");
BO_UNLOCK(bo);
#endif /* INVARIANTS */
/*
* Put back the real size.
*/
ip->i_size = length;
DIP_SET(ip, i_size, length);
if (DIP(ip, i_blocks) >= blocksreleased)
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
else /* sanity */
DIP_SET(ip, i_blocks, 0);
ip->i_flag |= IN_CHANGE;
#ifdef QUOTA
(void) chkdq(ip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
extclean:
if (journaltrunc)
softdep_journal_freeblocks(ip, cred, length, IO_EXT);
else
softdep_setup_freeblocks(ip, length, IO_EXT);
return (ffs_update(vp, waitforupdate));
#endif // 0
return 0;
}
/*
* Get a block of requested size that is associated with
* a given vnode and block offset. If it is found in the
* block cache, mark it as having been found, make it busy
* and return it. Otherwise, return an empty block of the
* correct size. It is up to the caller to insure that the
* cached blocks be of the correct size.
*/
struct buf *
getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo)
{
struct bufhashhdr *bh;
struct buf *bp, *nbp = NULL;
int s, err;
/*
* XXX
* The following is an inlined version of 'incore()', but with
* the 'invalid' test moved to after the 'busy' test. It's
* necessary because there are some cases in which the NFS
* code sets B_INVAL prior to writing data to the server, but
* in which the buffers actually contain valid data. In this
* case, we can't allow the system to allocate a new buffer for
* the block until the write is finished.
*/
bh = BUFHASH(vp, blkno);
start:
LIST_FOREACH(bp, BUFHASH(vp, blkno), b_hash) {
if (bp->b_lblkno != blkno || bp->b_vp != vp)
continue;
s = splbio();
if (ISSET(bp->b_flags, B_BUSY)) {
SET(bp->b_flags, B_WANTED);
err = tsleep(bp, slpflag | (PRIBIO + 1), "getblk",
slptimeo);
splx(s);
if (err)
return (NULL);
goto start;
}
if (!ISSET(bp->b_flags, B_INVAL)) {
SET(bp->b_flags, (B_BUSY | B_CACHE));
bremfree(bp);
splx(s);
break;
}
splx(s);
}
if (bp == NULL) {
if (nbp == NULL && getnewbuf(slpflag, slptimeo, &nbp) != 0) {
goto start;
}
bp = nbp;
binshash(bp, bh);
bp->b_blkno = bp->b_lblkno = blkno;
s = splbio();
bgetvp(vp, bp);
splx(s);
} else if (nbp != NULL) {
/*
* Set B_AGE so that buffer appear at BQ_CLEAN head
* and gets reused ASAP.
*/
SET(nbp->b_flags, B_AGE);
brelse(nbp);
}
allocbuf(bp, size);
return (bp);
}
static int
fuse_write_biobackend(struct vnode *vp, struct uio *uio,
struct ucred *cred, struct fuse_filehandle *fufh, int ioflag)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct buf *bp;
daddr_t lbn;
int bcount;
int n, on, err = 0;
const int biosize = fuse_iosize(vp);
KASSERT(uio->uio_rw == UIO_WRITE, ("ncl_write mode"));
FS_DEBUG("resid=%zx offset=%jx fsize=%jx\n",
uio->uio_resid, uio->uio_offset, fvdat->filesize);
if (vp->v_type != VREG)
return (EIO);
if (uio->uio_offset < 0)
return (EINVAL);
if (uio->uio_resid == 0)
return (0);
if (ioflag & IO_APPEND)
uio_setoffset(uio, fvdat->filesize);
/*
* Find all of this file's B_NEEDCOMMIT buffers. If our writes
* would exceed the local maximum per-file write commit size when
* combined with those, we must decide whether to flush,
* go synchronous, or return err. We don't bother checking
* IO_UNIT -- we just make all writes atomic anyway, as there's
* no point optimizing for something that really won't ever happen.
*/
do {
if (fuse_isdeadfs(vp)) {
err = ENXIO;
break;
}
lbn = uio->uio_offset / biosize;
on = uio->uio_offset & (biosize - 1);
n = MIN((unsigned)(biosize - on), uio->uio_resid);
FS_DEBUG2G("lbn %ju, on %d, n %d, uio offset %ju, uio resid %zd\n",
(uintmax_t)lbn, on, n,
(uintmax_t)uio->uio_offset, uio->uio_resid);
again:
/*
* Handle direct append and file extension cases, calculate
* unaligned buffer size.
*/
if (uio->uio_offset == fvdat->filesize && n) {
/*
* Get the buffer (in its pre-append state to maintain
* B_CACHE if it was previously set). Resize the
* nfsnode after we have locked the buffer to prevent
* readers from reading garbage.
*/
bcount = on;
FS_DEBUG("getting block from OS, bcount %d\n", bcount);
bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
if (bp != NULL) {
long save;
err = fuse_vnode_setsize(vp, cred,
uio->uio_offset + n);
if (err) {
brelse(bp);
break;
}
save = bp->b_flags & B_CACHE;
bcount += n;
allocbuf(bp, bcount);
bp->b_flags |= save;
}
} else {
/*
* Obtain the locked cache block first, and then
* adjust the file's size as appropriate.
*/
bcount = on + n;
if ((off_t)lbn * biosize + bcount < fvdat->filesize) {
if ((off_t)(lbn + 1) * biosize < fvdat->filesize)
bcount = biosize;
else
bcount = fvdat->filesize -
(off_t)lbn *biosize;
}
FS_DEBUG("getting block from OS, bcount %d\n", bcount);
bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
if (bp && uio->uio_offset + n > fvdat->filesize) {
err = fuse_vnode_setsize(vp, cred,
uio->uio_offset + n);
if (err) {
brelse(bp);
break;
}
}
}
if (!bp) {
err = EINTR;
break;
}
/*
* Issue a READ if B_CACHE is not set. In special-append
* mode, B_CACHE is based on the buffer prior to the write
* op and is typically set, avoiding the read. If a read
* is required in special append mode, the server will
* probably send us a short-read since we extended the file
* on our end, resulting in b_resid == 0 and, thusly,
* B_CACHE getting set.
*
* We can also avoid issuing the read if the write covers
* the entire buffer. We have to make sure the buffer state
* is reasonable in this case since we will not be initiating
* I/O. See the comments in kern/vfs_bio.c's getblk() for
* more information.
*
* B_CACHE may also be set due to the buffer being cached
* normally.
*/
if (on == 0 && n == bcount) {
bp->b_flags |= B_CACHE;
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
fuse_io_strategy(vp, bp);
if ((err = bp->b_error)) {
brelse(bp);
break;
}
}
if (bp->b_wcred == NOCRED)
bp->b_wcred = crhold(cred);
/*
* If dirtyend exceeds file size, chop it down. This should
* not normally occur but there is an append race where it
* might occur XXX, so we log it.
*
* If the chopping creates a reverse-indexed or degenerate
* situation with dirtyoff/end, we 0 both of them.
*/
if (bp->b_dirtyend > bcount) {
FS_DEBUG("FUSE append race @%lx:%d\n",
(long)bp->b_blkno * biosize,
bp->b_dirtyend - bcount);
bp->b_dirtyend = bcount;
}
if (bp->b_dirtyoff >= bp->b_dirtyend)
bp->b_dirtyoff = bp->b_dirtyend = 0;
/*
* If the new write will leave a contiguous dirty
* area, just update the b_dirtyoff and b_dirtyend,
* otherwise force a write rpc of the old dirty area.
*
* While it is possible to merge discontiguous writes due to
* our having a B_CACHE buffer ( and thus valid read data
* for the hole), we don't because it could lead to
* significant cache coherency problems with multiple clients,
* especially if locking is implemented later on.
*
* as an optimization we could theoretically maintain
* a linked list of discontinuous areas, but we would still
* have to commit them separately so there isn't much
* advantage to it except perhaps a bit of asynchronization.
*/
if (bp->b_dirtyend > 0 &&
(on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
/*
* Yes, we mean it. Write out everything to "storage"
* immediatly, without hesitation. (Apart from other
* reasons: the only way to know if a write is valid
* if its actually written out.)
*/
bwrite(bp);
if (bp->b_error == EINTR) {
err = EINTR;
break;
}
goto again;
}
err = uiomove((char *)bp->b_data + on, n, uio);
/*
* Since this block is being modified, it must be written
* again and not just committed. Since write clustering does
* not work for the stage 1 data write, only the stage 2
* commit rpc, we have to clear B_CLUSTEROK as well.
*/
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
if (err) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = err;
brelse(bp);
break;
}
/*
* Only update dirtyoff/dirtyend if not a degenerate
* condition.
*/
if (n) {
if (bp->b_dirtyend > 0) {
bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
bp->b_dirtyend = MAX((on + n), bp->b_dirtyend);
} else {
bp->b_dirtyoff = on;
bp->b_dirtyend = on + n;
}
vfs_bio_set_valid(bp, on, n);
}
err = bwrite(bp);
if (err)
break;
} while (uio->uio_resid > 0 && n > 0);
if (fuse_sync_resize && (fvdat->flag & FN_SIZECHANGE) != 0)
fuse_vnode_savesize(vp, cred);
return (err);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp = vp;
daddr_t lastblock;
struct inode *oip;
daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct ext2_sb_info *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, blocksreleased = 0;
int i;
int aflags, error, allerror;
off_t osize;
/*
kprintf("ext2_truncate called %d to %d\n", VTOI(ovp)->i_number, length);
*/ /*
* negative file sizes will totally break the code below and
* are not meaningful anyways.
*/
if (length < 0)
return EFBIG;
oip = VTOI(ovp);
if (ovp->v_type == VLNK &&
oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
#if DIAGNOSTIC
if (length != 0)
panic("ext2_truncate: partial truncate of symlink");
#endif
bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (EXT2_UPDATE(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (EXT2_UPDATE(ovp, 0));
}
#if QUOTA
if ((error = ext2_getinoquota(oip)) != 0)
return (error);
#endif
fs = oip->i_e2fs;
osize = oip->i_size;
ext2_discard_prealloc(oip);
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
vnode_pager_setsize(ovp, length);
error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, aflags);
if (error) {
vnode_pager_setsize(ovp, osize);
return (error);
}
oip->i_size = length;
if (aflags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (EXT2_UPDATE(ovp, 1));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessable again because
* of subsequent file growth.
*/
/* I don't understand the comment above */
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2_balloc(oip, lbn, offset, cred, &bp, aflags);
if (error)
return (error);
oip->i_size = length;
size = blksize(fs, oip, lbn);
bzero((char *)bp->b_data + offset, (u_int)(size - offset));
allocbuf(bp, size);
if (aflags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->s_blocksize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->s_blocksize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2_indirtrunc below.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
for (i = NDADDR - 1; i > lastblock; i--)
oip->i_db[i] = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = EXT2_UPDATE(ovp, 1);
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
oip->i_size = osize;
error = vtruncbuf(ovp, length, (int)fs->s_blocksize);
if (error && (allerror == 0))
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ext2_indirtrunc(oip, indir_lbn[level],
fsbtodoff(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ext2_blkfree(oip, bn, fs->s_frag_size);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ext2_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ext2_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#if DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("itrunc1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("itrunc2");
if (length == 0 && (!RB_EMPTY(&ovp->v_rbdirty_tree) ||
!RB_EMPTY(&ovp->v_rbclean_tree)))
panic("itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = length;
oip->i_blocks -= blocksreleased;
if (oip->i_blocks < 0) /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
vnode_pager_setsize(ovp, length);
#if QUOTA
ext2_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
}
/*
* Find a buffer which is available for use.
* If free memory for buffer space and an empty header from the empty list,
* use that. Otherwise, select something from a free list.
* Preference is to AGE list, then LRU list.
*/
static struct buf *
getnewbuf(int sz)
{
struct buf *bp;
int x;
x = splbio();
start:
/* can we constitute a new buffer? */
if (freebufspace > sz
&& bfreelist[BQ_EMPTY].av_forw != (struct buf *)bfreelist+BQ_EMPTY) {
caddr_t addr;
/*#define notyet*/
#ifndef notyet
if ((addr = malloc (sz, M_TEMP, M_WAITOK)) == 0) goto tryfree;
#else /* notyet */
/* get new memory buffer */
if (round_page(sz) == sz)
addr = (caddr_t) kmem_alloc_wired_wait(buffer_map, sz);
else
addr = (caddr_t) malloc (sz, M_TEMP, M_WAITOK);
/*if ((addr = malloc (sz, M_TEMP, M_NOWAIT)) == 0) goto tryfree;*/
bzero(addr, sz);
#endif /* notyet */
freebufspace -= sz;
allocbufspace += sz;
bp = bfreelist[BQ_EMPTY].av_forw;
bp->b_flags = B_BUSY | B_INVAL;
bremfree(bp);
bp->b_un.b_addr = addr;
bp->b_bufsize = sz; /* 20 Aug 92*/
goto fillin;
}
tryfree:
if (bfreelist[BQ_AGE].av_forw != (struct buf *)bfreelist+BQ_AGE) {
bp = bfreelist[BQ_AGE].av_forw;
bremfree(bp);
} else if (bfreelist[BQ_LRU].av_forw != (struct buf *)bfreelist+BQ_LRU) {
bp = bfreelist[BQ_LRU].av_forw;
bremfree(bp);
} else {
/* wait for a free buffer of any kind */
(bfreelist + BQ_AGE)->b_flags |= B_WANTED;
sleep(bfreelist, PRIBIO);
splx(x);
return (0);
}
/* if we are a delayed write, convert to an async write! */
if (bp->b_flags & B_DELWRI) {
bp->b_flags |= B_BUSY;
bawrite (bp);
goto start;
}
if(bp->b_vp)
brelvp(bp);
/* we are not free, nor do we contain interesting data */
if (bp->b_rcred != NOCRED) crfree(bp->b_rcred); /* 25 Apr 92*/
if (bp->b_wcred != NOCRED) crfree(bp->b_wcred);
bp->b_flags = B_BUSY;
fillin:
bremhash(bp);
splx(x);
bp->b_dev = NODEV;
bp->b_vp = NULL;
bp->b_blkno = bp->b_lblkno = 0;
bp->b_iodone = 0;
bp->b_error = 0;
bp->b_wcred = bp->b_rcred = NOCRED;
if (bp->b_bufsize != sz)
allocbuf(bp, sz);
bp->b_bcount = bp->b_bufsize = sz;
bp->b_dirtyoff = bp->b_dirtyend = 0;
return (bp);
}
/* VOP_BWRITE 1 time */
int
lfs_fragextend(struct vnode *vp, int osize, int nsize, daddr_t lbn, struct buf **bpp,
kauth_cred_t cred)
{
struct inode *ip;
struct lfs *fs;
long frags;
int error;
extern long locked_queue_bytes;
size_t obufsize;
ip = VTOI(vp);
fs = ip->i_lfs;
frags = (long)lfs_numfrags(fs, nsize - osize);
error = 0;
ASSERT_NO_SEGLOCK(fs);
/*
* Get the seglock so we don't enlarge blocks while a segment
* is being written. If we're called with bpp==NULL, though,
* we are only pretending to change a buffer, so we don't have to
* lock.
*/
top:
if (bpp) {
rw_enter(&fs->lfs_fraglock, RW_READER);
LFS_DEBUG_COUNTLOCKED("frag");
}
if (!ISSPACE(fs, frags, cred)) {
error = ENOSPC;
goto out;
}
/*
* If we are not asked to actually return the block, all we need
* to do is allocate space for it. UBC will handle dirtying the
* appropriate things and making sure it all goes to disk.
* Don't bother to read in that case.
*/
if (bpp && (error = bread(vp, lbn, osize, 0, bpp))) {
goto out;
}
#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
if ((error = lfs_chkdq(ip, frags, cred, 0))) {
if (bpp)
brelse(*bpp, 0);
goto out;
}
#endif
/*
* Adjust accounting for lfs_avail. If there's not enough room,
* we will have to wait for the cleaner, which we can't do while
* holding a block busy or while holding the seglock. In that case,
* release both and start over after waiting.
*/
if (bpp && ((*bpp)->b_oflags & BO_DELWRI)) {
if (!lfs_fits(fs, frags)) {
if (bpp)
brelse(*bpp, 0);
#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
lfs_chkdq(ip, -frags, cred, 0);
#endif
rw_exit(&fs->lfs_fraglock);
lfs_availwait(fs, frags);
goto top;
}
lfs_sb_subavail(fs, frags);
}
mutex_enter(&lfs_lock);
lfs_sb_subbfree(fs, frags);
mutex_exit(&lfs_lock);
ip->i_lfs_effnblks += frags;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp) {
obufsize = (*bpp)->b_bufsize;
allocbuf(*bpp, nsize, 1);
/* Adjust locked-list accounting */
if (((*bpp)->b_flags & B_LOCKED) != 0 &&
(*bpp)->b_iodone == NULL) {
mutex_enter(&lfs_lock);
locked_queue_bytes += (*bpp)->b_bufsize - obufsize;
mutex_exit(&lfs_lock);
}
memset((char *)((*bpp)->b_data) + osize, 0, (u_int)(nsize - osize));
}
out:
if (bpp) {
rw_exit(&fs->lfs_fraglock);
}
return (error);
}
void
sink(int argc, char *argv[])
{
static BUF buffer;
static struct stat stb;
struct timeval tv[2];
enum { YES, NO, DISPLAYED } wrerr;
BUF *bp;
off_t i, j;
int exists, first, mask, mode, ofd, omode;
int setimes, targisdir, wrerrno = 0;
size_t amt, count, size;
char ch, *cp, *np, *targ, *why, *vect[1];
static char buf[BUFSIZ];
#define atime tv[0]
#define mtime tv[1]
#define SCREWUP(str) { why = str; goto screwup; }
setimes = targisdir = 0;
#ifndef __GNO__
mask = umask(0);
if (!pflag)
(void)umask(mask);
#endif
if (argc != 1) {
run_err("ambiguous target");
exit(1);
}
targ = *argv;
if (targetshouldbedirectory)
verifydir(targ);
(void)write(rem, "", 1);
if (stat(targ, &stb) == 0 && S_ISDIR(stb.st_mode))
targisdir = 1;
for (first = 1;; first = 0) {
cp = buf;
if (read(rem, cp, 1) <= 0)
return;
if (*cp++ == '\n')
SCREWUP("unexpected <newline>");
do {
if (read(rem, &ch, sizeof(ch)) != sizeof(ch))
SCREWUP("lost connection");
*cp++ = ch;
} while (cp < &buf[BUFSIZ - 1] && ch != '\n');
*cp = 0;
if (buf[0] == '\01' || buf[0] == '\02') {
if (iamremote == 0)
(void)write(STDERR_FILENO,
buf + 1, strlen(buf + 1));
if (buf[0] == '\02')
exit(1);
++errs;
continue;
}
if (buf[0] == 'E') {
(void)write(rem, "", 1);
return;
}
if (ch == '\n')
*--cp = 0;
#define getnum(t) (t) = 0; while (isdigit(*cp)) (t) = (t) * 10 + (*cp++ - '0');
cp = buf;
if (*cp == 'T') {
setimes++;
cp++;
getnum(mtime.tv_sec);
if (*cp++ != ' ')
SCREWUP("mtime.sec not delimited");
getnum(mtime.tv_usec);
if (*cp++ != ' ')
SCREWUP("mtime.usec not delimited");
getnum(atime.tv_sec);
if (*cp++ != ' ')
SCREWUP("atime.sec not delimited");
getnum(atime.tv_usec);
if (*cp++ != '\0')
SCREWUP("atime.usec not delimited");
(void)write(rem, "", 1);
continue;
}
if (*cp != 'C' && *cp != 'D') {
/*
* Check for the case "rcp remote:foo\* local:bar".
* In this case, the line "No match." can be returned
* by the shell before the rcp command on the remote is
* executed so the ^Aerror_message convention isn't
* followed.
*/
if (first) {
run_err("%s", cp);
exit(1);
}
SCREWUP("expected control record");
}
mode = 0;
for (++cp; cp < buf + 5; cp++) {
if (*cp < '0' || *cp > '7')
SCREWUP("bad mode");
mode = (mode << 3) | (*cp - '0');
}
if (*cp++ != ' ')
SCREWUP("mode not delimited");
for (size = 0; isdigit(*cp);)
size = size * 10 + (*cp++ - '0');
if (*cp++ != ' ')
SCREWUP("size not delimited");
if (targisdir) {
static char *namebuf;
static int cursize;
size_t need;
need = strlen(targ) + strlen(cp) + 250;
if (need > cursize) {
if (!(namebuf = malloc(need)))
run_err("%s", strerror(errno));
}
(void)snprintf(namebuf, need, "%s%s%s", targ,
*targ ? "/" : "", cp);
np = namebuf;
} else
np = targ;
exists = stat(np, &stb) == 0;
if (buf[0] == 'D') {
int mod_flag = pflag;
if (exists) {
if (!S_ISDIR(stb.st_mode)) {
errno = ENOTDIR;
goto bad;
}
if (pflag)
(void)chmod(np, mode);
} else {
/* Handle copying from a read-only directory */
mod_flag = 1;
#ifdef __GNO__
if (mkdir(np) < 0)
#else
if (mkdir(np, mode | S_IRWXU) < 0)
#endif
goto bad;
}
vect[0] = np;
sink(1, vect);
if (setimes) {
setimes = 0;
#ifndef __GNO__
if (utimes(np, tv) < 0)
run_err("%s: set times: %s",
np, strerror(errno));
#endif
}
if (mod_flag)
(void)chmod(np, mode);
continue;
}
omode = mode;
mode |= S_IWRITE;
if ((ofd = open(np, O_WRONLY|O_CREAT, mode)) < 0) {
bad: run_err("%s: %s", np, strerror(errno));
continue;
}
(void)write(rem, "", 1);
if ((bp = allocbuf(&buffer, ofd, BUFSIZ)) == NULL) {
(void)close(ofd);
continue;
}
cp = bp->buf;
wrerr = NO;
for (count = i = 0; i < size; i += BUFSIZ) {
amt = BUFSIZ;
if (i + amt > size)
amt = size - i;
count += amt;
do {
j = read(rem, cp, amt);
if (j <= 0) {
run_err("%s", j ? strerror(errno) :
"dropped connection");
exit(1);
}
amt -= j;
cp += j;
} while (amt > 0);
if (count == bp->cnt) {
/* Keep reading so we stay sync'd up. */
if (wrerr == NO) {
j = write(ofd, bp->buf, count);
if (j != count) {
wrerr = YES;
wrerrno = j >= 0 ? EIO : errno;
}
}
count = 0;
cp = bp->buf;
}
}
if (count != 0 && wrerr == NO &&
(j = write(ofd, bp->buf, count)) != count) {
wrerr = YES;
wrerrno = j >= 0 ? EIO : errno;
}
if (ftruncate(ofd, size)) {
run_err("%s: truncate: %s", np, strerror(errno));
wrerr = DISPLAYED;
}
#ifndef __GNO__
if (pflag) {
if (exists || omode != mode)
if (fchmod(ofd, omode))
run_err("%s: set mode: %s",
np, strerror(errno));
} else {
if (!exists && omode != mode)
if (fchmod(ofd, omode & ~mask))
run_err("%s: set mode: %s",
np, strerror(errno));
}
#endif
(void)close(ofd);
(void)response();
if (setimes && wrerr == NO) {
setimes = 0;
#ifndef __GNO__
if (utimes(np, tv) < 0) {
run_err("%s: set times: %s",
np, strerror(errno));
wrerr = DISPLAYED;
}
#endif
}
switch(wrerr) {
case YES:
run_err("%s: %s", np, strerror(wrerrno));
break;
case NO:
(void)write(rem, "", 1);
break;
case DISPLAYED:
break;
}
}
screwup:
run_err("protocol error: %s", why);
exit(1);
}
void
source(int argc, char **argv)
{
struct stat stb;
static BUF buffer;
BUF *bp;
off_t i;
off_t amt;
int fd, haderr, indx, result;
char *last, *name, buf[BUFSIZ];
for (indx = 0; indx < argc; ++indx) {
name = argv[indx];
if ((fd = open(name, O_RDONLY, 0)) < 0)
goto syserr;
if (fstat(fd, &stb)) {
syserr: run_err("%s: %s", name, strerror(errno));
goto next;
}
if (S_ISDIR(stb.st_mode) && iamrecursive) {
rsource(name, &stb);
goto next;
} else if (!S_ISREG(stb.st_mode)) {
run_err("%s: not a regular file", name);
goto next;
}
if ((last = strrchr(name, '/')) == NULL)
last = name;
else
++last;
if (pflag) {
/*
* Make it compatible with possible future
* versions expecting microseconds.
*/
snprintf(buf, sizeof(buf), "T%ld 0 %ld 0\n",
(long)stb.st_mtime,
(long)stb.st_atime);
write(remout, buf, strlen(buf));
if (response() < 0)
goto next;
}
#undef MODEMASK
#define MODEMASK (S_ISUID|S_ISGID|S_ISVTX|S_IRWXU|S_IRWXG|S_IRWXO)
snprintf(buf, sizeof(buf), "C%04o %lu %s\n",
(unsigned int)(stb.st_mode & MODEMASK),
(unsigned long)stb.st_size,
last);
write(remout, buf, strlen(buf));
if (response() < 0)
goto next;
if ((bp = allocbuf(&buffer, fd, BUFSIZ)) == NULL) {
next: close(fd);
continue;
}
/* Keep writing after an error so that we stay sync'd up. */
for (haderr = i = 0; i < stb.st_size; i += bp->cnt) {
amt = bp->cnt;
if (i + amt > stb.st_size)
amt = stb.st_size - i;
if (!haderr) {
result = read(fd, bp->buf, (size_t)amt);
if (result != amt)
haderr = result >= 0 ? EIO : errno;
}
if (haderr)
write(remout, bp->buf, amt);
else {
result = write(remout, bp->buf, (size_t)amt);
if (result != amt)
haderr = result >= 0 ? EIO : errno;
}
}
if (close(fd) && !haderr)
haderr = errno;
if (!haderr)
write(remout, "", 1);
else
run_err("%s: %s", name, strerror(haderr));
response();
}
}
int
lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
{
daddr_t lastblock;
struct inode *oip = VTOI(ovp);
daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR];
/* XXX ondisk32 */
int32_t newblks[ULFS_NDADDR + ULFS_NIADDR];
struct lfs *fs;
struct buf *bp;
int offset, size, level;
daddr_t count, rcount;
daddr_t blocksreleased = 0, real_released = 0;
int i, nblocks;
int aflags, error, allerror = 0;
off_t osize;
long lastseg;
size_t bc;
int obufsize, odb;
int usepc;
if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
KASSERT(oip->i_size == 0);
return 0;
}
if (length < 0)
return (EINVAL);
/*
* Just return and not update modification times.
*/
if (oip->i_size == length) {
/* still do a uvm_vnp_setsize() as writesize may be larger */
uvm_vnp_setsize(ovp, length);
return (0);
}
fs = oip->i_lfs;
if (ovp->v_type == VLNK &&
(oip->i_size < fs->um_maxsymlinklen ||
(fs->um_maxsymlinklen == 0 &&
oip->i_ffs1_blocks == 0))) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("lfs_truncate: partial truncate of symlink");
#endif
memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
oip->i_size = oip->i_ffs1_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
lfs_imtime(fs);
osize = oip->i_size;
usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
ASSERT_NO_SEGLOCK(fs);
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
if (length > fs->um_maxfilesize)
return (EFBIG);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
if (usepc) {
if (lfs_lblkno(fs, osize) < ULFS_NDADDR &&
lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) &&
lfs_blkroundup(fs, osize) != osize) {
off_t eob;
eob = lfs_blkroundup(fs, osize);
uvm_vnp_setwritesize(ovp, eob);
error = ulfs_balloc_range(ovp, osize,
eob - osize, cred, aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
if (ioflag & IO_SYNC) {
mutex_enter(ovp->v_interlock);
VOP_PUTPAGES(ovp,
trunc_page(osize & lfs_sb_getbmask(fs)),
round_page(eob),
PGO_CLEANIT | PGO_SYNCIO);
}
}
uvm_vnp_setwritesize(ovp, length);
error = ulfs_balloc_range(ovp, length - 1, 1, cred,
aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
uvm_vnp_setsize(ovp, length);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
KASSERT(ovp->v_size == oip->i_size);
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
} else {
error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
error = lfs_balloc(ovp, length - 1, 1, cred,
aflags, &bp);
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
oip->i_ffs1_size = oip->i_size = length;
uvm_vnp_setsize(ovp, length);
(void) VOP_BWRITE(bp->b_vp, bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
}
}
if ((error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 0)
return (error);
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundary, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever becomes accessible again because
* of subsequent file growth. Directories however are not
* zero'ed as they should grow back initialized to empty.
*/
offset = lfs_blkoff(fs, length);
lastseg = -1;
bc = 0;
if (ovp != fs->lfs_ivnode)
lfs_seglock(fs, SEGM_PROT);
if (offset == 0) {
oip->i_size = oip->i_ffs1_size = length;
} else if (!usepc) {
lbn = lfs_lblkno(fs, length);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
obufsize = bp->b_bufsize;
odb = lfs_btofsb(fs, bp->b_bcount);
oip->i_size = oip->i_ffs1_size = length;
size = lfs_blksize(fs, oip, lbn);
if (ovp->v_type != VDIR)
memset((char *)bp->b_data + offset, 0,
(u_int)(size - offset));
allocbuf(bp, size, 1);
if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
mutex_enter(&lfs_lock);
locked_queue_bytes -= obufsize - bp->b_bufsize;
mutex_exit(&lfs_lock);
}
if (bp->b_oflags & BO_DELWRI) {
lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size));
/* XXX shouldn't this wake up on lfs_availsleep? */
}
(void) VOP_BWRITE(bp->b_vp, bp);
} else { /* vp->v_type == VREG && length < osize && offset != 0 */
/*
* When truncating a regular file down to a non-block-aligned
* size, we must zero the part of last block which is past
* the new EOF. We must synchronously flush the zeroed pages
* to disk since the new pages will be invalidated as soon
* as we inform the VM system of the new, smaller size.
* We must do this before acquiring the GLOCK, since fetching
* the pages will acquire the GLOCK internally.
* So there is a window where another thread could see a whole
* zeroed page past EOF, but that's life.
*/
daddr_t xlbn;
voff_t eoz;
aflags = ioflag & IO_SYNC ? B_SYNC : 0;
error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
xlbn = lfs_lblkno(fs, length);
size = lfs_blksize(fs, oip, xlbn);
eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize);
ubc_zerorange(&ovp->v_uobj, length, eoz - length,
UBC_UNMAP_FLAG(ovp));
if (round_page(eoz) > round_page(length)) {
mutex_enter(ovp->v_interlock);
error = VOP_PUTPAGES(ovp, round_page(length),
round_page(eoz),
PGO_CLEANIT | PGO_DEACTIVATE |
((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
}
}
genfs_node_wrlock(ovp);
oip->i_size = oip->i_ffs1_size = length;
uvm_vnp_setsize(ovp, length);
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
if (length > QUAD_MAX - lfs_sb_getbsize(fs))
lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs));
else
lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1;
lastiblock[SINGLE] = lastblock - ULFS_NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs);
nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
/*
* Record changed file and block pointers before we start
* freeing blocks. lastiblock values are also normalized to -1
* for calls to lfs_indirtrunc below.
*/
memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
newblks[ULFS_NDADDR+level] = 0;
lastiblock[level] = -1;
}
for (i = ULFS_NDADDR - 1; i > lastblock; i--)
newblks[i] = 0;
oip->i_size = oip->i_ffs1_size = osize;
error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
if (error && !allerror)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -ULFS_NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ffs1_ib[level];
if (bn != 0) {
error = lfs_indirtrunc(oip, indir_lbn[level],
bn, lastiblock[level],
level, &count, &rcount,
&lastseg, &bc);
if (error)
allerror = error;
real_released += rcount;
blocksreleased += count;
if (lastiblock[level] < 0) {
if (oip->i_ffs1_ib[level] > 0)
real_released += nblocks;
blocksreleased += nblocks;
oip->i_ffs1_ib[level] = 0;
lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs),
&lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = ULFS_NDADDR - 1; i > lastblock; i--) {
long bsize, obsize;
bn = oip->i_ffs1_db[i];
if (bn == 0)
continue;
bsize = lfs_blksize(fs, oip, i);
if (oip->i_ffs1_db[i] > 0) {
/* Check for fragment size changes */
obsize = oip->i_lfs_fragsize[i];
real_released += lfs_btofsb(fs, obsize);
oip->i_lfs_fragsize[i] = 0;
} else
obsize = 0;
blocksreleased += lfs_btofsb(fs, bsize);
oip->i_ffs1_db[i] = 0;
lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_ffs1_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
#if 0
long olddspace;
#endif
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = lfs_blksize(fs, oip, lastblock);
#if 0
olddspace = oip->i_lfs_fragsize[lastblock];
#endif
oip->i_size = oip->i_ffs1_size = length;
newspace = lfs_blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: newspace");
if (oldspace - newspace > 0) {
blocksreleased += lfs_btofsb(fs, oldspace - newspace);
}
#if 0
if (bn > 0 && olddspace - newspace > 0) {
/* No segment accounting here, just vnode */
real_released += lfs_btofsb(fs, olddspace - newspace);
}
#endif
}
done:
/* Finish segment accounting corrections */
lfs_update_seguse(fs, oip, lastseg, bc);
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if ((newblks[ULFS_NDADDR + level] == 0) !=
((oip->i_ffs1_ib[level]) == 0)) {
panic("lfs itrunc1");
}
for (i = 0; i < ULFS_NDADDR; i++)
if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
panic("lfs itrunc2");
}
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("lfs itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = oip->i_ffs1_size = length;
oip->i_lfs_effnblks -= blocksreleased;
oip->i_ffs1_blocks -= real_released;
mutex_enter(&lfs_lock);
lfs_sb_addbfree(fs, blocksreleased);
mutex_exit(&lfs_lock);
#ifdef DIAGNOSTIC
if (oip->i_size == 0 &&
(oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
printf("lfs_truncate: truncate to 0 but %d blks/%jd effblks\n",
oip->i_ffs1_blocks, (intmax_t)oip->i_lfs_effnblks);
panic("lfs_truncate: persistent blocks");
}
#endif
/*
* If we truncated to zero, take us off the paging queue.
*/
mutex_enter(&lfs_lock);
if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
oip->i_flags &= ~IN_PAGING;
TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
}
mutex_exit(&lfs_lock);
oip->i_flag |= IN_CHANGE;
#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
(void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
genfs_node_unlock(ovp);
errout:
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
if (ovp != fs->lfs_ivnode)
lfs_segunlock(fs);
return (allerror ? allerror : error);
}