/* ARGSUSED7 */
int /* ERRNO if error, 0 if successful. */
sam_create_ino(
sam_node_t *pip, /* pointer to parent directory inode. */
char *cp, /* pointer to the component name to create. */
vattr_t *vap, /* vattr ptr for type & mode information. */
vcexcl_t ex, /* exclusive create flag. */
int mode, /* file mode information. */
vnode_t **vpp, /* pointer pointer to returned vnode. */
cred_t *credp, /* credentials pointer. */
int filemode) /* open file mode */
{
int error = 0;
sam_node_t *ip;
struct sam_name name; /* If no entry, slot info is returned here */
int trans_size;
int issync;
int truncflag = 0;
int terr = 0;
#ifdef LQFS_TODO_LOCKFS
struct ulockfs *ulp;
#endif /* LQFS_TODO_LOCKFS */
/*
* Cannot set sticky bit unless superuser.
*/
if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(credp)) {
vap->va_mode &= ~VSVTX;
}
lookup_name:
#ifdef LQFS_TODO_LOCKFS
error = qfs_lockfs_begin(pip->mp, &ulp, ULOCKFS_CREATE_MASK);
if (error) {
return (error);
}
if (ulp) {
#endif /* LQFS_TODO_LOCKFS */
/* Start LQFS create transaction */
trans_size = (int)TOP_CREATE_SIZE(pip);
TRANS_BEGIN_CSYNC(pip->mp, issync, TOP_CREATE, trans_size);
#ifdef LQFS_TODO_LOCKFS
}
#endif /* LQFS_TODO_LOCKFS */
RW_LOCK_OS(&pip->data_rwl, RW_WRITER);
name.operation = SAM_CREATE;
if ((error = sam_lookup_name(pip, cp, &ip, &name, credp)) == ENOENT) {
if (((error = sam_create_name(pip, cp, &ip, &name,
vap, credp)) != 0) &&
IS_SAM_ENOSPC(error)) {
RW_UNLOCK_OS(&pip->data_rwl, RW_WRITER);
/*
* Temporarily end LQFS create transaction
*/
#ifdef LQFS_TODO_LOCKFS
if (ulp) {
#endif /* LQFS_TODO_LOCKFS */
TRANS_END_CSYNC(pip->mp, terr, issync,
TOP_CREATE, trans_size);
#ifdef LQFS_TODO_LOCKFS
}
#endif /* LQFS_TODO_LOCKFS */
error = sam_wait_space(pip, error);
if (error == 0) {
error = terr;
}
if (error) {
return (error);
}
goto lookup_name;
}
RW_UNLOCK_OS(&pip->data_rwl, RW_WRITER);
} else if (error == 0) { /* If entry already exists. */
RW_UNLOCK_OS(&pip->data_rwl, RW_WRITER);
error = EEXIST;
if (ex == NONEXCL) { /* If non-exclusive create */
if ((S_ISDIR(ip->di.mode) ||
S_ISATTRDIR(ip->di.mode)) &&
(mode & S_IWRITE)) {
/* Cannot create over an existing dir. */
error = EISDIR;
} else if (SAM_PRIVILEGE_INO(ip->di.version,
ip->di.id.ino)) {
/* Cannot create over privileged inodes */
error = EPERM;
} else if (mode) { /* Check mode if set */
error = sam_access_ino(ip, mode, FALSE, credp);
} else {
error = 0;
}
if ((error == 0) && S_ISREG(ip->di.mode) &&
(vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
/*
* If logging, do the truncate after the
* LQFS create transaction is logged.
*/
if (TRANS_ISTRANS(ip->mp)) {
truncflag++;
} else {
RW_LOCK_OS(&ip->inode_rwl, RW_WRITER);
error = sam_clear_file(ip, 0,
STALE_ARCHIVE, credp);
RW_UNLOCK_OS(&ip->inode_rwl, RW_WRITER);
}
if (error == 0) {
VNEVENT_CREATE_OS(SAM_ITOV(ip), NULL);
}
}
}
/*
* Cannot do the following as it caused a stale of
* offline copies.
*/
#if 0
if ((error == 0) && ((mode & O_CREAT) == 0)) {
TRANS_INODE(ip->mp, ip);
sam_mark_ino(ip, SAM_UPDATED|SAM_CHANGED);
}
#endif
if (error) {
VN_RELE(SAM_ITOV(ip)); /* Decrement v_count if error */
}
} else {
RW_UNLOCK_OS(&pip->data_rwl, RW_WRITER);
}
#ifdef LQFS_TODO_LOCKFS
if (ulp) {
#endif /* LQFS_TODO_LOCKFS */
TRANS_END_CSYNC(pip->mp, terr, issync, TOP_CREATE, trans_size);
/*
* If we haven't had a more interesting failure
* already, then anything that might've happened
* here should be reported.
*/
if (error == 0) {
error = terr;
}
#ifdef LQFS_TODO_LOCKFS
}
#endif /* LQFS_TODO_LOCKFS */
if (!error && truncflag) {
(void) TRANS_ITRUNC(ip, (u_offset_t)0, STALE_ARCHIVE, credp);
}
#ifdef LQFS_TODO_LOCKFS
if (ulp) {
qfs_lockfs_end(ulp);
}
#endif /* LQFS_TODO_LOCKFS */
if (error == 0) {
*vpp = SAM_ITOV(ip);
TRACE(T_SAM_CREATE_RET, SAM_ITOV(pip), (sam_tr_t)* vpp,
ip->di.id.ino, error);
}
return (error);
}
/*
* ufs_alloc_data - supports allocating space and reads or writes
* that involve changes to file length or space allocation.
*
* This function is more expensive, because of the UFS log transaction,
* so ufs_rdwr_data() should be used when space or file length changes
* will not occur.
*
* Inputs:
* fdb - A null pointer instructs this function to only allocate
* space for the specified offset and length.
* An actual fdbuffer instructs this function to perform
* the read or write operation.
* flags - defaults (zero value) to synchronous write
* B_READ - indicates read operation
* B_ASYNC - indicates perform operation asynchronously
*/
int
ufs_alloc_data(
vnode_t *vnodep,
u_offset_t offset,
size_t *len,
fdbuffer_t *fdbp,
int flags,
cred_t *credp)
{
struct inode *ip = VTOI(vnodep);
size_t done_len, io_len;
int contig;
u_offset_t uoff, io_off;
int error = 0; /* No error occured */
int offsetn; /* Start point this IO */
int nbytes; /* Number bytes in this IO */
daddr_t bn;
struct fs *fs;
struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
int i_size_changed = 0;
u_offset_t old_i_size;
struct ulockfs *ulp;
int trans_size;
int issync; /* UFS Log transaction */
/* synchronous when non-zero */
int io_started = 0; /* No IO started */
uint_t protp = PROT_ALL;
ASSERT((flags & B_WRITE) == 0);
/*
* Obey the lockfs protocol
*/
error = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, segkmap, 0, &protp);
if (error) {
if ((fdbp != NULL) && (flags & B_ASYNC)) {
fdb_ioerrdone(fdbp, error);
}
return (error);
}
if (ulp) {
/*
* Try to begin a UFS log transaction
*/
trans_size = TOP_GETPAGE_SIZE(ip);
TRANS_TRY_BEGIN_CSYNC(ufsvfsp, issync, TOP_GETPAGE,
trans_size, error);
if (error == EWOULDBLOCK) {
ufs_lockfs_end(ulp);
if ((fdbp != NULL) && (flags & B_ASYNC)) {
fdb_ioerrdone(fdbp, EDEADLK);
}
return (EDEADLK);
}
}
uoff = offset;
io_off = offset;
io_len = *len;
done_len = 0;
DEBUGF((CE_CONT, "?ufs_alloc: off %llx len %lx size %llx fdb: %p\n",
uoff, (io_len - done_len), ip->i_size, (void *)fdbp));
rw_enter(&ip->i_ufsvfs->vfs_dqrwlock, RW_READER);
rw_enter(&ip->i_contents, RW_WRITER);
ASSERT((ip->i_mode & IFMT) == IFREG);
fs = ip->i_fs;
while (error == 0 && done_len < io_len) {
uoff = (u_offset_t)(io_off + done_len);
offsetn = (int)blkoff(fs, uoff);
nbytes = (int)MIN(fs->fs_bsize - offsetn, io_len - done_len);
DEBUGF((CE_CONT, "?ufs_alloc_data: offset: %llx len %x\n",
uoff, nbytes));
if (uoff + nbytes > ip->i_size) {
/*
* We are extending the length of the file.
* bmap is used so that we are sure that
* if we need to allocate new blocks, that it
* is done here before we up the file size.
*/
DEBUGF((CE_CONT, "?ufs_alloc_data: grow %llx -> %llx\n",
ip->i_size, uoff + nbytes));
error = bmap_write(ip, uoff, (offsetn + nbytes),
BI_ALLOC_ONLY, NULL, credp);
if (ip->i_flag & (ICHG|IUPD))
ip->i_seq++;
if (error) {
DEBUGF((CE_CONT, "?ufs_alloc_data: grow "
"failed err: %d\n", error));
break;
}
if (fdbp != NULL) {
if (uoff >= ip->i_size) {
/*
* Desired offset is past end of bytes
* in file, so we have a hole.
*/
fdb_add_hole(fdbp, uoff - offset,
nbytes);
} else {
int contig;
buf_t *bp;
error = bmap_read(ip, uoff, &bn,
&contig);
if (error) {
break;
}
contig = ip->i_size - uoff;
contig = P2ROUNDUP(contig, DEV_BSIZE);
bp = fdb_iosetup(fdbp, uoff - offset,
contig, vnodep, flags);
bp->b_edev = ip->i_dev;
bp->b_dev = cmpdev(ip->i_dev);
bp->b_blkno = bn;
bp->b_file = ip->i_vnode;
bp->b_offset = (offset_t)uoff;
if (ufsvfsp->vfs_snapshot) {
fssnap_strategy(
&ufsvfsp->vfs_snapshot, bp);
} else {
(void) bdev_strategy(bp);
}
io_started = 1;
lwp_stat_update(LWP_STAT_OUBLK, 1);
if ((flags & B_ASYNC) == 0) {
error = biowait(bp);
fdb_iodone(bp);
if (error) {
break;
}
}
if (contig > (ip->i_size - uoff)) {
contig -= ip->i_size - uoff;
fdb_add_hole(fdbp,
ip->i_size - offset,
contig);
}
}
}
i_size_changed = 1;
old_i_size = ip->i_size;
UFS_SET_ISIZE(uoff + nbytes, ip);
TRANS_INODE(ip->i_ufsvfs, ip);
/*
* file has grown larger than 2GB. Set flag
* in superblock to indicate this, if it
* is not already set.
*/
if ((ip->i_size > MAXOFF32_T) &&
!(fs->fs_flags & FSLARGEFILES)) {
ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
mutex_enter(&ufsvfsp->vfs_lock);
fs->fs_flags |= FSLARGEFILES;
ufs_sbwrite(ufsvfsp);
mutex_exit(&ufsvfsp->vfs_lock);
}
} else {
/*
* The file length is not being extended.
*/
error = bmap_read(ip, uoff, &bn, &contig);
if (error) {
DEBUGF((CE_CONT, "?ufs_alloc_data: "
"bmap_read err: %d\n", error));
break;
}
if (bn != UFS_HOLE) {
/*
* Did not map a hole in the file
*/
int contig = P2ROUNDUP(nbytes, DEV_BSIZE);
buf_t *bp;
if (fdbp != NULL) {
bp = fdb_iosetup(fdbp, uoff - offset,
contig, vnodep, flags);
bp->b_edev = ip->i_dev;
bp->b_dev = cmpdev(ip->i_dev);
bp->b_blkno = bn;
bp->b_file = ip->i_vnode;
bp->b_offset = (offset_t)uoff;
if (ufsvfsp->vfs_snapshot) {
fssnap_strategy(
&ufsvfsp->vfs_snapshot, bp);
} else {
(void) bdev_strategy(bp);
}
io_started = 1;
lwp_stat_update(LWP_STAT_OUBLK, 1);
if ((flags & B_ASYNC) == 0) {
error = biowait(bp);
fdb_iodone(bp);
if (error) {
break;
}
}
}
} else {
/*
* We read a hole in the file.
* We have to allocate blocks for the hole.
*/
error = bmap_write(ip, uoff, (offsetn + nbytes),
BI_ALLOC_ONLY, NULL, credp);
if (ip->i_flag & (ICHG|IUPD))
ip->i_seq++;
if (error) {
DEBUGF((CE_CONT, "?ufs_alloc_data: fill"
" hole failed error: %d\n", error));
break;
}
if (fdbp != NULL) {
fdb_add_hole(fdbp, uoff - offset,
nbytes);
}
}
}
done_len += nbytes;
}
if (error) {
if (i_size_changed) {
/*
* Allocation of the blocks for the file failed.
* So truncate the file size back to its original size.
*/
(void) ufs_itrunc(ip, old_i_size, 0, credp);
}
}
DEBUGF((CE_CONT, "?ufs_alloc: uoff %llx len %lx\n",
uoff, (io_len - done_len)));
if ((offset + *len) < (NDADDR * fs->fs_bsize)) {
*len = (size_t)(roundup(offset + *len, fs->fs_fsize) - offset);
} else {
*len = (size_t)(roundup(offset + *len, fs->fs_bsize) - offset);
}
/*
* Flush cached pages.
*
* XXX - There should be no pages involved, since the I/O was performed
* through the device strategy routine and the page cache was bypassed.
* However, testing has demonstrated that this VOP_PUTPAGE is
* necessary. Without this, data might not always be read back as it
* was written.
*
*/
(void) VOP_PUTPAGE(vnodep, 0, 0, B_INVAL, credp);
rw_exit(&ip->i_contents);
rw_exit(&ip->i_ufsvfs->vfs_dqrwlock);
if ((fdbp != NULL) && (flags & B_ASYNC)) {
/*
* Show that no more asynchronous IO will be added
*/
fdb_ioerrdone(fdbp, error);
}
if (ulp) {
/*
* End the UFS Log transaction
*/
TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_GETPAGE,
trans_size);
ufs_lockfs_end(ulp);
}
if (io_started && (flags & B_ASYNC)) {
return (0);
} else {
return (error);
}
}
