/*
* NAME: jfs_mknod(dvp, name, mode, dev, crp)
*
* FUNCTION: make a new object in directory <dvp> with mode = <mode>,
* name = <pname>, and rdev = <dev>.
*
* PARAMETER: dvp - directory vnode
* name - name of new object
* mode - create mode (rwxrwxrwx).
* dev - new device number if block/character-special file
* crp - credential
*
* RETURN: Errors from subroutines
*
* note: JFS allows mknod() to create a special file.
* XPG4.2: the only portable use of mknod() is to create a FIFO-special file
* with mode = S_IFIFO and dev = 0.
*/
jfs_mknod(
struct vnode *dvp,
caddr_t name,
mode_t mode,
dev_t dev,
struct ucred *crp)
{
int32 rc;
int32 tid;
inode_t *dip = VP2IP(dvp);
inode_t *ip;
ino_t ino;
component_t dname;
struct vfs *vfsp = dvp->v_vfsp;
btstack_t btstack;
inode_t *iplist[2];
/* JFS does NOT support mknod() of directory */
if ((mode & IFMT) == IFDIR)
return EISDIR;
if ((mode & IFMT) != IFIFO && (mode & IFMT) != IFSOCK)
if (rc = privcheck_cr(DEV_CONFIG, crp))
return rc;
IWRITE_LOCK(dip);
txBegin(dip->i_ipmnt, &tid, 0);
/* validate search+write permission on parent directory */
if (rc = iAccess(dip, IEXEC|IWRITE, crp))
goto out1;
/*
* scan parent directory for entry/freespace
* (dtSearch() returns parent directory page pinned)
*/
dname.name = name;
dname.namlen = strlen(name);
if (rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))
goto out1;
/*
* allocate in-memory+on-disk inode:
* (iAlloc() returns new, locked inode)
*/
if (rc = iAlloc(vfsp, dip, mode, &ip, crp))
{
/* release parent directory page */
BT_PUTSEARCH(&btstack);
goto out1;
}
/*
* create entry in parent directory
* (dtInsert() releases parent directory page)
*/
ino = ip->i_number;
if (rc = dtInsert(tid, dip, &dname, &ino, &btstack))
{
/* discard the new inode */
ip->i_nlink = 0;
goto out2;
}
/*
* fo a block or character special file, the passed-in device number
* needs to be set into the inode's device field and the gnode's
* "real device" field.
*/
if ((ip->i_mode & IFMT) == IFBLK || (ip->i_mode & IFMT) == IFCHR)
{
ip->i_rdev = dev;
IP2GP(ip)->gn_rdev = dev;
}
imark(ip, IACC|ICHG|IUPD|IFSYNC);
/* update parent directory inode */
imark(dip, ICHG|IUPD|IFSYNC);
/*
* insert entry for the new file to dnlc
*/
ncEnter(dip->i_ipimap, dip->i_number, &dname, ino, NULL);
iplist[0] = dip;
iplist[1] = ip;
rc = txCommit(tid, 2, &iplist[0], 0);
out2:
IWRITE_UNLOCK(ip);
ICACHE_LOCK();
iput(ip, vfsp);
ICACHE_UNLOCK();
out1:
IWRITE_UNLOCK(dip);
txEnd(tid);
return rc;
}
/*
* NAME: pathlookup_pc
*
* FUNCTION: Lookup path, preserving case.
* Lookup path, copying each component into output string
* preserving its case.
*
* PARAMETERS: vfsp - pointer to VFS
* path - full path name
* path_pc - output - full path name w/preserved case
*
* RETURN: errors from subroutines.
*
* NOTES: We assume path is in canonical form: "X:\<path>" where there
* are no extraneous backslashes and . and .. have been removed.
*
* dtFind is not very efficient for finding a directory entry
* without wildcards, but infolevel 7 does not seem to actually
* be used anywhere, so it must not be too important.
*/
int32 pathlookup_pc(
struct vfs *vfsp,
UniChar *path, /* input - Path */
UniChar *path_pc) /* output - path w/preserved case */
{
int32 comp_len;
uint32 count;
UniChar *component;
struct dirent *dbuf;
inode_t *ip;
int32 offset;
UniChar *outptr = path_pc;
component_t pattern;
int rc;
UniChar *slash;
int32 tbytes;
struct vnode *vp;
if (vfsp == NULL)
{
jERROR(2,("pathlookup_pc: invalid VPB!\n"));
return ENOTDIR;
}
/* Copy "X:\" to output string */
UniStrncpy(outptr, path, 3);
outptr += 3;
path += 3;
if (*path == 0)
{
/* Trivial case "X:\" */
*outptr = 0;
return NO_ERROR;
}
component = (UniChar *)allocpool(unipool, 0);
if (component == 0)
return ENOSPC;
pattern.name = component;
dbuf = (struct dirent *)allocpool(dirent_pool, 0);
if (dbuf == 0)
{
freepool(unipool, (caddr_t *)component);
return ENOSPC;
}
vp = vfsp->vfs_mntd; /* vnode of root directory */
jfs_hold(vp);
while (path)
{
slash = UniStrchr(path, '\\');
if (slash)
{
comp_len = slash - path;
slash++;
}
else
comp_len = UniStrlen(path);
UniStrncpy(component, path, comp_len);
component[comp_len] = 0;
UniStrupr(component); /* Convert to upper case */
pattern.namlen = comp_len;
path = slash;
offset = 0;
count = 1;
tbytes = 0;
rc = dtFind(VP2IP(vp), &pattern, 0, &offset, &count, PSIZE,
&tbytes, dbuf);
jfs_rele(vp);
if (rc || (count == 0))
{
freepool(dirent_pool, (caddr_t *)dbuf);
freepool(unipool, (caddr_t *)component);
return ENOENT;
}
UniStrncpy(outptr, dbuf->d_name, dbuf->d_namlen);
outptr += dbuf->d_namlen;
if (path)
{
ICACHE_LOCK();
rc = iget(vfsp, dbuf->d_ino, &ip, 0);
ICACHE_UNLOCK();
if (rc)
{
freepool(dirent_pool, (caddr_t *)dbuf);
freepool(unipool, (caddr_t *)component);
return rc;
}
vp = IP2VP(ip);
*(outptr++) = '\\';
}
else
*outptr = 0;
}
freepool(dirent_pool, (caddr_t *)dbuf);
freepool(unipool, (caddr_t *)component);
return NO_ERROR;
}
/*
* NAME: jfs_lookup(dvp, vpp, pname, flag, vattrp, crp)
*
* FUNCTION: resolve <pname> in directory <dvp> to vnode <vpp>
* with a reference acquired and attribute <vattrp>.
*
* PARAMETERS: dvp _ directory vnode
* vpp - object vnode (out)
* pname - object name
* flag -
* vattrp - object attribute (out)
* crp - credential
*
* RETURN: errors from subroutines.
*/
jfs_lookup(
struct vnode *dvp,
struct vnode **vpp,
UniChar *pname, /* NULL terminated */
int32 flag,
struct vattr *vattrp,
struct ucred *crp)
{
int32 rc = 0;
struct vfs *vfsp = dvp->v_vfsp;
inode_t *dip = VP2IP(dvp); /* parent directory inode */
ino_t ino; /* object i_number */
inode_t *ip; /* object inode */
component_t dname; /* object name */
ncookie_t ncookie;
btstack_t btstack;
NOISE(1,("jfs_lookup: dip:0x%08x name:%s\n", dip, pname));
*vpp = NULL;
/* <dvp> must be a directory */
if ((dip->i_mode & IFMT) != IFDIR)
return ENOTDIR;
IREAD_LOCK(dip);
if (dip->i_nlink == 0)
{
rc = ENOENT;
goto out;
}
/*
* resolve name to i_number via dnlc/directory lookup
*/
getInumber:
/*
* for "." or "..", lookup directory inode
*/
if (pname[0] == '.')
{
/* looking up ".." */
if (pname[1] == '.' && pname[2] == '\0')
{
ino = dip->i_parent;
goto getInode;
}
/* looking up "." */
else if (pname[1] == '\0')
{
ip = dip;
jfs_hold(dvp);
*vpp = dvp;
goto getAttribute;
}
}
/*
* search dnlc/directory
*/
dname.name = pname;
dname.namlen = UniStrlen(pname);
if ((ino = ncLookup(dip->i_ipimap, dip->i_number, &dname, &ncookie)) == 0)
{
/*
* dnlc miss: search directory
*/
if (rc = dtSearch(dip, &dname, &ino, &btstack, JFS_LOOKUP))
goto out;
/* insert name entry to dnlc */
ncEnter(dip->i_ipimap, dip->i_number, &dname, ino, &ncookie);
}
/*
* resolve i_number to inode/vnode with a reference
*/
getInode:
ICACHE_LOCK();
rc = iget(vfsp, ino, &ip, 0);
ICACHE_UNLOCK();
if (rc)
goto out;
*vpp = IP2VP(ip);
/*
* get attribute
*/
getAttribute:
if (vattrp != NULL)
get_vattr(ip, vattrp);
out:
IREAD_UNLOCK(dip);
NOISE(1,("jfs_lookup: rc:%d\n", rc));
return rc;
}
/*
* NAME: jfs_checkpath
*
* FUNCTION: Search the directory tree rooted with topar_ip to see if
* it includes from_ip
*
* PARAMETERS:
* from_ip - pointer to original directory inode
* frompar_ip - pointer to parent of from_ip
* topar_ip - pointer to destination parent inode
* vfst -
*
* NOTES: RENAME_LOCK() of mount inode is used to serialize renaming
* within an aggregate. This prevents someone from modifying
* behind us as we walk up the tree. Since we hold the rename
* lock, we don't need to lock each parent along the directory
* tree. We already have a lock on a directory child, so none of
* the directories we visit will be empty. Therefore, they cannot
* be removed until our rename is complete. This will work as long
* as we continue to not support unlink() of a directory.
*
* RETURN: 0 - from_ip is not a parent of topar_ip.
* != 0 - from_ip is a parent or some other error.
*/
static int32
jfs_checkpath( inode_t *from_ip,
inode_t *frompar_ip,
inode_t *topar_ip,
struct vfs *vfst)
{
ino_t par_ino;
inode_t *parent_ip;
int32 rc = 0;
if (topar_ip->i_number == ROOT_I)
{
/*
* We have reached the root and not found the source so we know
* we don't have an orphan
*/
return 0;
}
/*
* Walk up the destination directory tree looking for orphans
*/
par_ino = ((dtroot_t *)&(topar_ip->i_btroot))->header.idotdot;
while(1)
{
if (par_ino <= 0)
{
/* We have hit a bad dotdot entry */
assert(0);
}
if (par_ino == ROOT_I ||
par_ino == frompar_ip->i_number)
{
rc = 0;
break;
}
if (par_ino == from_ip->i_number)
{
/*
* Uh-oh: we have an orphan situation
*/
rc = EINVAL;
break;
}
ICACHE_LOCK();
rc = iget(vfst, par_ino, &parent_ip, 0);
ICACHE_UNLOCK();
if (rc != 0)
{
break;
}
par_ino = ((dtroot_t *)&(parent_ip->i_btroot))->header.idotdot;
ICACHE_LOCK();
iput(parent_ip, vfst);
ICACHE_UNLOCK();
}
return rc;
}
/*
* NAME: jfs_rename
*
* FUNCTION: rename a file or directory
*
* PARAMETER: from_vp - pointer to source vnode
* frompar_vp - pointer to source parent vnode
* from_name - original name
* to_vp - pointer to target vnode (if it exists)
* topar_vp - pointer to target parent vnode
* to_name - new name
* crp - pointer to caller's credentials
*
* RETURN: errors from subroutines
*
* JFS supports sticky bit permission.
*
* normally, to rename an object, user needs execute and write
* permission of the directory containing the object as well as
* on the target directory.
* EACCESS: if S_ISVTX (aka sticky) bit is set for a directory,
* a file in the directory can be renamed only if
* the user has write permission for the directory, and
* either owns the file, owns the directory, or
* is the superuser (or have appropriate privileges).
* (e.g., /tmp in which any user can create a file but
* should not be able to delete or rename files owned
* by others) [XPG4.2]
*
* Basic algorithm is:
*
* 1) Check validity of parameters and then obtain locks on the source parent
* directory, destination parent directory, source, and the destination
* inode. This ensures neither the source or destination will be deleted out
* from underneath us. It also ensures the source or destination won't be
* modified before we finish.
* 2) Link source to destination. If destination already exists,
* delete it first.
* 3) Unlink source reference to inode if still around. If a
* directory was moved and the parent of the destination
* is different from the source, patch the ".." entry in the
* directory.
*
* The transaction log needs to have the changes journalled in such a
* way that a crash meets the requirement that the destination always
* exists (if it existed before). For directory objects it is critical
* that the changes in i-numbers occur in such a way that loops and
* orphans aren't created by a crash.
*/
int32
jfs_rename(
struct vnode *from_vp, /* source vnode */
struct vnode *frompar_vp, /* source parent vnode */
UniChar *from_name, /* source name */
struct vnode *to_vp, /* destination vnode */
struct vnode *topar_vp, /* destination parent vnode */
UniChar *to_name, /* destination name */
struct dasd_usage *from_dlim, /* dasd usage structure for source F226941 */
struct dasd_usage *to_dlim, /* dasd usage structure for dest F226941 */
uint32 flags) /* INEWNAME if new pathname not 8.3 */
{
inode_t *ipmnt;
int32 tid = -1; /* Transaction ID */
int32 txCount = 0; /* Count of inodes in txList */
inode_t *txList[4]; /* List of modified inodes */
inode_t *from_ip = VP2IP (from_vp); /* Source inode */
inode_t *frompar_ip = VP2IP (frompar_vp); /* Source parent inode */
inode_t *to_ip = to_vp ? VP2IP (to_vp) : 0;
/* Target inode (if it exists) */
inode_t *topar_ip = VP2IP (topar_vp); /* Target parent inode */
struct vfs *vfsf = frompar_vp->v_vfsp;
struct vfs *vfst = topar_vp->v_vfsp;
int32 doingdirectory = 0, /* Source inode is a directory */
newparent = 0; /* New i-number of parent */
int32 got_from = 0, got_to = 0;
btstack_t btstack; /* Temporary stack for B-tree struct */
component_t dname; /* Directory name structure */
ino_t ino; /* I-number for directory searches */
int32 error = 0, rc;
tlock_t *tlck; /* Transaction lock */
dtlock_t *dtlck; /* dtree line lock */
lv_t *lv; /* line lock vector */
// BEGIN F226941
#ifdef _JFS_OS2
int64 blocks_moving; /* Number of blocks being moved */
int64 orig_fromblocks; /* original dasd usage of from dir */
int64 orig_toblocks; /* original dasd usage of to dir */
uint32 start_here; /* index in limits array */
int32 i;
inode_t *lock_arr[64]; /* Array of inodes to lock */
inode_t **lock_list; /* List of inodes to lock */
int32 num_locks; /* Number of inodes to lock */
int32 upper_limit; /* Upper limit to number of inodes
* we need to lock */
#ifdef _JFS_FASTDASD
int32 first_locked; /* index in lock_list array of first
* inode still locked. D233382
*/
#endif /* _JFS_FASTDASD */
#endif /* _JFS_OS2 */
// BEGIN F226941
assert( to_name != NULL && from_name != NULL);
ipmnt = frompar_ip->i_ipmnt;
RENAME_LOCK(ipmnt);
/*
* Make sure that we are not renaming the "." or ".." entries in
* a directory.
*/
if (from_name[0] == '.' &&
(!from_name[1] || (from_name[1] == '.' && !from_name[2])))
{
error = EINVAL;
goto abortit;
}
/*
* Now we want to verify the proper realationship between the inodes.
* If everything checks out we need to obtain locks on the source parent
* directory, destination parent directory, source, and destination
* inodes. This allows us to proceed knowing that noone can delete
* either the source or destination from underneath us. It also allows
* us to know noone will be modifying the destination so we can delete
* it safely. We need the lock on the source since we might need to
* modify its .. entry.
*/
/*
* In OS/2, the destination cannot exist unless it is the same as
* the source. This is to allow DosMove to change the case of a
* filename.
*/
if (to_ip && (from_ip != to_ip))
{
error = EINVAL;
goto abortit;
}
/*
* Check for cross-device rename.
*/
if (frompar_ip->i_dev != topar_ip->i_dev)
{
error = EXDEV;
goto abortit;
}
/*
* Check if source is the destination parent. Can't do this
* since this would orphan everything under the source.
*/
if (from_ip == topar_ip)
{
error = EINVAL;
goto abortit;
}
/*
* Now we know both the parents are okay. We will lock both the
* parent inodes so we can guarantee the children will not be
* deleted underneath us. We will also lock the source.
* When locking the inodes we will use the protocol of locking
* regular files first, then directories. When locking more than
* one of each type, we do it in descending inode order. We also need
* to make sure we don't attempt to lock one inode twice if the
* two parents are the same.
*/
if ((from_ip->i_mode & IFMT) == IFDIR)
doingdirectory = 1;
// BEGIN F226941
if (from_dlim->num_limits)
{
/* We need to lock all directories from the root to each of
* the from and to parent directories. The DASD usage lists
* will contain at least one common entry (the root). We need
* to compile a list of all unique inodes to lock. The upper
* limit will be the sum of both num_limits fields.
*/
if (frompar_ip == topar_ip)
upper_limit = from_dlim->num_limits + 1;
else
upper_limit = from_dlim->num_limits +
to_dlim->num_limits;
if (upper_limit > 64)
lock_list = (inode_t **)
xmalloc(upper_limit* sizeof(inode_t *),
0, kernel_heap);
else
lock_list = lock_arr;
/* Put from parents dasd usage list into lock_list */
for (num_locks = 0; num_locks < from_dlim->num_limits;
num_locks++)
lock_list[num_locks] = from_dlim->pLimits[num_locks];
/*
* We need to lock regular files before directories. If
* we are moving a regular file, lock it now, otherwise add
* it to the lock list.
*/
if (doingdirectory)
lock_list[num_locks++] = from_ip;
else
IWRITE_LOCK(from_ip);
if (frompar_ip != topar_ip)
{
/* Add unique members of to parent's dasd usage list */
for (i = 0; i < to_dlim->num_limits; i++)
if (to_dlim->pLimits[i] != lock_list[i])
break;
start_here = i; /* Needed by over_limit() */
while (i < to_dlim->num_limits)
lock_list[num_locks++] = to_dlim->pLimits[i++];
}
/* Lock them! */
sort_and_lock(0, num_locks, lock_list);
/*
* If this was a regular file, add the inode to the lock list
* now, so we include it in the txCommit, and unlock it
*/
if (!doingdirectory)
lock_list[num_locks++] = from_ip;
from_dlim->flag |= DLIM_DIRS_LOCKED;
frompar_ip->i_dasdlim = from_dlim;
if (frompar_ip != topar_ip) // D230860
{
to_dlim->flag |= DLIM_DIRS_LOCKED;
topar_ip->i_dasdlim = to_dlim;
}
}
else
// END F226941
{
if (doingdirectory)
{
if (frompar_ip == topar_ip)
iwritelocklist(2, frompar_ip, from_ip);
else
iwritelocklist(3, frompar_ip, topar_ip,
from_ip);
}
else
{
if (frompar_ip == topar_ip)
{
IWRITE_LOCK(from_ip);
IWRITE_LOCK(frompar_ip);
}
else
{
IWRITE_LOCK(from_ip);
iwritelocklist(2, frompar_ip, topar_ip);
}
}
}
/*
* Now that we have the source parent locked we can lookup the
* source inode to make sure it hasn't changed since we did our
* initial verification. If it has we will need to release our
* locks and repeat our verification with the new version.
* Otherwise we know the source is valid and we can continue.
*/
dname.name = from_name;
dname.namlen = UniStrlen (from_name);
rc = dtSearch(frompar_ip, &dname, &ino, &btstack, JFS_LOOKUP);
switch (rc)
{
case 0:
/*
* An entry was found, need to see if it is the
* same inode as we had before.
*/
if (ino != from_ip->i_number)
{
rc = ENOENT;
goto cleanup;
}
break;
default:
/*
* Either the source cannot be found or
* something went wrong in the search
*/
error = rc;
goto cleanup;
}
/*
* If changing case, no need to search for destination inode
*/
if (!to_ip)
{
/*
* Now we can lookup the destination inode to make sure it
* hasn't been created since we did our verifications. If it
* has we abort.
*/
dname.name = to_name;
dname.namlen = UniStrlen (to_name);
if (dname.namlen > JFS_NAME_MAX-1)
{
error = ERROR_FILENAME_EXCED_RANGE;
goto cleanup;
}
rc = dtSearch(topar_ip, &dname, &ino, &btstack, JFS_LOOKUP);
switch (rc)
{
case 0:
/*
* We found an entry; fail with EEXIST
*/
rc = EEXIST;
goto cleanup;
case ENOENT:
break;
default:
/*
* Some error from dtSearch(). cleanup and
* return
*/
error = rc;
goto cleanup;
}
}
/*
* Now we need to validate parameters and check for proper inodes
*/
/*
* Make sure the source inode has a positive link count
*/
if (from_ip->i_nlink <= 0)
{
error = EINVAL;
goto cleanup;
}
if ((topar_ip->i_mode & IFMT) != IFDIR)
{
error = EINVAL;
goto cleanup;
}
/* Check for the appropriate permissions on the source */
if (doingdirectory && (frompar_ip->i_number != topar_ip->i_number))
{
/*
* Account for ".." in new directory. When source and
* destination have the same parent we don't fool with the link
* count.
*/
if ((nlink_t)topar_ip->i_nlink >= LINK_MAX)
{
error = EMLINK;
goto cleanup;
}
/*
* If ".." must be changed (i.e. the directory gets a new
* parent) then the source directory must not be in the
* directory hierarchy above the target, as this would orphan
* everything below the source directory.
*/
newparent = topar_ip->i_number;
/* RENAME_LOCK(topar_ip->i_ipmnt); */
if (error = jfs_checkpath(from_ip, frompar_ip, topar_ip, vfst))
goto cleanup;
}
/*
* Check for write-protected media
*/
if (isReadOnly(topar_ip))
{
error = EROFS;
goto cleanup;
}
// BEGIN F226941
/*
* Check to see if there is room in the destination directory
*/
blocks_moving = doingdirectory ? DASDUSED(&from_ip->i_DASD) :
from_ip->i_nblocks;
if ((topar_ip != frompar_ip) && (topar_ip->i_dasdlim) &&
over_limit(topar_ip->i_dasdlim, blocks_moving, start_here))
{
error = ERROR_DISK_FULL;
goto cleanup;
}
orig_fromblocks = frompar_ip->i_nblocks;
if (topar_ip != frompar_ip)
orig_toblocks = topar_ip->i_nblocks;
// END F226941
/*
* Perform rename
*/
txBegin(topar_ip->i_ipmnt, &tid, 0);
dname.name = to_name;
dname.namlen = UniStrlen (to_name);
if (to_ip != NULL)
{
/*
* We change the case of the name in place.
*/
ino = to_ip->i_number;
if (error = dtChangeCase(tid, frompar_ip, &dname, &ino,
JFS_RENAME))
goto cleanup;
imark(topar_ip, ICHG|IUPD|IFSYNC);
}
else
{
/*
* We already know the destination does not exist, so link
* source inode as destination.
*/
if (error = dtSearch(topar_ip, &dname, &ino, &btstack,
JFS_CREATE))
{
/*
* Problem, can't find where to put this guy, or it
* already exists!!
*/
goto cleanup;
}
ino = from_ip->i_number;
if (error = dtInsert(tid, topar_ip, &dname, &ino, &btstack))
{
/*
* Failed adding source to destination parent
*/
goto cleanup;
}
imark(topar_ip, ICHG|IUPD|IFSYNC);
/* Insert entry for the new file to name cache */
ncEnter(topar_ip->i_ipimap, topar_ip->i_number, &dname,
from_ip->i_number, NULL);
/* Remove source name from source parent and name cache */
dname.name = from_name;
dname.namlen = UniStrlen (from_name);
ino = from_ip->i_number;
ncDelete(frompar_ip->i_ipimap, frompar_ip->i_number, &dname);
if (error = dtDelete(tid, frompar_ip, &dname, &ino, JFS_REMOVE))
{
/*
* Another unexpected error -- the original file does not
* appear to exist ...
*/
assert(0);
}
imark(frompar_ip, ICHG|IUPD|IFSYNC);
/*
* If this is a directory we need to update the ".." i-number
* in the inode. Also, there is now one fewer ".." referencing
* the source parent directory. Decrement the link count to
* the source parent directory for this.
*/
if (doingdirectory && newparent)
{
/* linelock header of dtree root (containing idotdot)
*/
tlck = txLock(tid, from_ip,
(jbuf_t *)&from_ip->i_bxflag,
tlckDTREE|tlckBTROOT);
dtlck = (dtlock_t *)&tlck->lock;
ASSERT(dtlck->index == 0);
lv = (lv_t *)&dtlck->lv[0];
lv->offset = 0;
lv->length = 1;
dtlck->index++;
((dtroot_t *) &from_ip->i_btroot)->header.idotdot =
newparent;
frompar_ip->i_nlink--;
topar_ip->i_nlink++;
/* RENAME_UNLOCK(topar_ip->i_ipmnt); */
}
/* Set or reset INEWNAME flag as appropriate */
from_ip->i_mode = (from_ip->i_mode & ~INEWNAME) |
(flags & INEWNAME);
imark(from_ip, ICHG|IFSYNC); // D231252
txList[txCount++] = from_ip; // D231252
}
// BEGIN F226941
/*
* Modify dasd usage for source and destination directories.
* Number of blocks for object being moved didn't change.
*/
if (frompar_ip == topar_ip)
{
DLIM_UPDATE(tid, frompar_ip, frompar_ip->i_nblocks
- orig_fromblocks);
}
else
{
DLIM_UPDATE(tid, topar_ip, topar_ip->i_nblocks +
blocks_moving - orig_toblocks);
DLIM_UPDATE(tid, frompar_ip, frompar_ip->i_nblocks -
(orig_fromblocks + blocks_moving));
}
// BEGIN D233382
#ifdef _JFS_FASTDASD
if (from_dlim->num_limits)
{
/*
* Let's remove any ancestor directories at the beginning of
* the lock list that aren't directly involved in the rename.
* i_dasdlim will be non-zero for frompar_ip & topar_ip
*/
for(i = 0; i < num_locks; i++)
{
if (lock_list[i]->i_dasdlim)
{
first_locked = i;
break;
}
IWRITE_UNLOCK(lock_list[i]);
}
ASSERT(i < num_locks); // We shouldn't have unlocked them all.
}
#endif /* _JFS_FASTDASD */
// END D233382
#ifndef _JFS_FASTDASD // D233382
if ((from_dlim->flag & DLIM_LOGGED) || (to_dlim->flag & DLIM_LOGGED))
{
error = txCommit(tid, num_locks, lock_list, 0);
from_dlim->flag &= ~DLIM_LOGGED;
to_dlim->flag &= ~DLIM_LOGGED;
}
else
#endif /* _JFS_FASTDASD */ // D233382
// END F226941
{
if (frompar_ip != topar_ip)
{
/*
* Different parent, so we need to add the other parent
* to our transaction list
*/
txList[txCount++] = topar_ip;
}
/* Add source parent directory to transaction list */
txList[txCount++] = frompar_ip;
assert(txCount <= 4);
error = txCommit(tid, txCount, txList, 0);
}
cleanup:
/*
* Come to this label when all locks have been taken and ready to leave.
* error should be set to return value.
*/
if (tid != -1) txEnd(tid);
// BEGIN F226941
if (from_dlim->num_limits)
{
frompar_ip->i_dasdlim->flag &= ~DLIM_DIRS_LOCKED;
frompar_ip->i_dasdlim = 0;
if (frompar_ip != topar_ip)
{
topar_ip->i_dasdlim->flag &= ~DLIM_DIRS_LOCKED;
topar_ip->i_dasdlim = 0;
}
#ifdef _JFS_FASTDASD
for(i = first_locked; i < num_locks; i++) // D233382
#else
for(i = 0; i < num_locks; i++)
#endif
IWRITE_UNLOCK(lock_list[i]);
if (lock_list != lock_arr)
xmfree((void *)lock_list, kernel_heap);
}
else
// END F226941
{
IWRITE_UNLOCK(from_ip);
IWRITE_UNLOCK(frompar_ip);
if (frompar_ip != topar_ip)
{
IWRITE_UNLOCK(topar_ip);
}
}
abortit:
/*
* Come to this label when none of the locks have been taken and ready
* to leave. error should be set to return value.
*/
if (to_ip && got_to)
{
ICACHE_LOCK();
iput(to_ip, vfst);
ICACHE_UNLOCK();
}
if (got_from)
{
ICACHE_LOCK();
iput(from_ip, vfsf);
ICACHE_UNLOCK();
}
RENAME_UNLOCK(ipmnt);
return error;
}
/*
* NAME: readdir
*
* FUNCTION: read directory according to specifications
* in directory search structure
*
* PARAMETER:
*
* RETURN: EINVAL - if not a directory
* errors from subroutines
*
* note:
* N.B. directory file offset encodes (directory page number,
* entry index number), and shold NOT be interpreted/modified
* by caller (lseek()) except that intial offset set to 0.
*
* no guarantees can be made that the exact offset
* requested can be found if directory has been updated
* by other threads between consecutive readdir()s.
* transfer length of zero signals start offset beyond eof.
*
* unused space in the directory are not returned to the user,
* i.e., more than requested size may have to be read
* from directory to fill the user's buffer.
*/
readdir(
struct vnode *dvp, /* inode of directory being read */
struct fsfd *fsfp, /* directory search information */
char *ubuf, /* user's data area */
uint32 ubytes, /* size of user's data area */
uint32 *matchcnt, /* count of entries returned */
uint32 level, /* level of output struct */
uint32 flags, /* offsets needed in output? */
EAOP *eaopp, /* pointer to EAOP */
struct ucred *crp)
{
int32 rc = 0;
int32 ReturnCode = NO_ERROR;
inode_t *dip; /* directory inode */
inode_t *ip; /* object inode */
uint32 matches; /* output matches found */
uint32 dtmatches; /* matches found per dtFind call */
uint32 position; /* offsets in output */
uint32 count; /* output buffer count */
int32 tbytes; /* byte count in dirent buffer */
struct dirent *dbuf; /* dirent buffer */
struct dirent *dbufp; /* dirent buffer */
uint32 ffhdsize; /* size of ffbuf header */
component_t lastmatch; /* pointer to last matching entry */
char *ffbuf; /* output buffer pointer */
char *nxbuf; /* output buffer pointer */
char *bufp; /* output buffer pointer */
MMPHPrereaddir(); /* MMPH Performance Hook */
/* set state from search structure
*/
dip = VP2IP(dvp);
position = flags & FF_GETPOS;
/* validate request */
if (ubytes == 0)
{
rc = EINVAL;
goto readdir_Exit;
}
/* continuous read of empty directory ? */
if (fsfp->fsd_offset == -1)
{
rc = ERROR_NO_MORE_FILES;
goto readdir_Exit;
}
dbuf = (struct dirent *)allocpool(dirent_pool, 0); // D228565
if (dbuf == 0) // D228565
{
rc = ENOMEM;
goto readdir_Exit;
}
/* set up variable to manipulate output buffer pointers
* based on level.
*/
if (level == 1)
ffhdsize = FFBUFHD;
else if (level == 11)
ffhdsize = FFBUFHD3L;
else if (level < 11)
ffhdsize = FFBUFHD2;
else
ffhdsize = FFBUFHD4L;
if (position)
ffhdsize += sizeof(uint32);
ffbuf = ubuf;
count = 0;
matches = *matchcnt;
*matchcnt = 0;
while ((*matchcnt < matches) && (rc == 0))
{
IREAD_LOCK(dip);
/* directory became void when last link was removed */
if ((dip->i_nlink == 0) || ((dip->i_mode & IFMT) != IFDIR))
{
IREAD_UNLOCK(dip);
freepool(dirent_pool, (caddr_t *)dbuf);
rc = ENOTDIR;
goto readdir_Exit;
}
/* fill a directory buffer.
* read on-disk structure (struct ldtentry_t) and
* translate into readdir() structure (struct dirent).
*/
tbytes = 0;
dtmatches = matches - *matchcnt;
dbufp = dbuf; // D228565
rc = dtFind(dip, &fsfp->fsd_pattern, fsfp->fsd_lastmatch,
&fsfp->fsd_offset, &dtmatches, PSIZE, &tbytes, dbufp);
IREAD_UNLOCK(dip);
if (rc)
{
freepool(dirent_pool, (caddr_t *)dbuf);
goto readdir_Exit;
}
/* copy translate buffer to user FileFindBuf buffer */
while ((*matchcnt < matches) && (ReturnCode == NO_ERROR))
{
uint32 namlen;
/* translation buffer empty? */
if (tbytes == 0)
break;
/* get size of next name */
namlen = dbufp->d_namlen;
/* user buffer full?
* the +1 here is to allow for the null character
* terminating the name string.
*/
if ((count + ffhdsize + namlen + 1) > ubytes)
{
rc = ERROR_BUFFER_OVERFLOW;
break;
}
/* get the inode for the file
*/
ICACHE_LOCK();
rc = iget(dvp->v_vfsp, dbufp->d_ino, &ip, 0);
ICACHE_UNLOCK();
if (rc)
goto try_next;
nxbuf = ffbuf;
/* fill in file search info for files that have
* the proper attributes; ignore others.
*/
rc = get_fileinfo(ip, &nxbuf, ubytes, dbufp->d_name,
namlen, fsfp->fsd_havattr, level,
eaopp, flags);
if ((rc == ERROR_BUFFER_OVERFLOW) && (*matchcnt == 0) &&
((level == FIL_QUERYEASFROMLIST) ||
(level == FIL_QUERYEASFROMLISTL)))
{
/* Can't fit EA in buffer, try without
* getting EA
*/
if (level == FIL_QUERYEASFROMLIST)
level = FIL_QUERYEASIZE;
else
level = FIL_QUERYEASIZEL;
ReturnCode = ERROR_EAS_DIDNT_FIT;
rc = get_fileinfo(ip, &nxbuf, ubytes,
dbufp->d_name, namlen,
fsfp->fsd_havattr, level,
eaopp, flags);
}
/* release the inode */
jfs_rele(IP2VP(ip));
if (rc == 0)
{
/* set offset if requested */
if (position)
{
rc = KernCopyOut(ffbuf,
&dbufp->d_offset,
sizeof(int32));
if (rc)
{
/* This is very unlikely to
* happen! */
ASSERT(0);
break;
}
}
/* update output buffer count */
count += nxbuf - ffbuf;
/* move to next entry in output buffer */
ffbuf = nxbuf;
/* update match count */
*matchcnt += 1;
}
else if (rc != -1)
break;
try_next:
/* rc == -1 indicates no attribute match,
* just keep going.
*/
rc = 0;
/* save name for next call setup */
lastmatch.name = dbufp->d_name;
lastmatch.namlen = namlen;
/* update dirent buffer count */
tbytes -= dbufp->d_reclen;
/* move to next entry in dirent buffer */
dbufp = (struct dirent *)
((caddr_t)dbufp + dbufp->d_reclen);
}
/* We don't want to continue if ReturnCode = ERROR_EAS_DIDNT_FIT
*/
if (rc == 0)
rc = ReturnCode;
/* set return code for end of directory with no matches */
if (fsfp->fsd_offset == -1)
rc = ERROR_NO_MORE_FILES;
else if ((rc == 0) || (rc == ERROR_EAS_DIDNT_FIT))
{
/* save last matching name for next call */
UniStrncpy(fsfp->fsd_lastmatch,lastmatch.name,
lastmatch.namlen);
fsfp->fsd_lastmatch[lastmatch.namlen] = '\0';
}
}
/* claim success if we return any entries */
if (*matchcnt != 0)
rc = ReturnCode;
freepool(dirent_pool, (caddr_t *)dbuf);
readdir_Exit:
MMPHPostreaddir(); /* MMPH Performance Hook */
return rc;
}
/*
* NAME: jfs_mkdir(dvp, name, mode, crp)
*
* FUNCTION: create a child directory in the parent directory <dvp>
* with name = <name> and mode = <mode>
*
* PARAMETER: dvp - parent directory vnode
* name - name of child directory
* mode - create mode (rwxrwxrwx).
* crp - credential
*
* RETURN: Errors from subroutines
*
* note:
* EACCESS: user needs search+write permission on the parent directory
*/
jfs_mkdir(
struct vnode *dvp,
UniChar *name,
mode_t mode,
#ifdef _JFS_OS2
EAOP *pEABuf,
#endif /* _JFS_OS2 */
struct ucred *crp)
{
int32 rc = 0, rc1 = 0;
int32 tid; /* transaction id */
struct vfs *vfsp = dvp->v_vfsp;
inode_t *dip = VP2IP(dvp); /* parent directory inode */
struct dasd_usage *du; // F226941
inode_t *ip = NULL; /* child directory inode */
ino_t ino;
component_t dname; /* child directory name */
btstack_t btstack;
inode_t *iplist[2];
int64 orig_nblocks; // F226941
#ifdef _JFS_LAZYCOMMIT
tblock_t *tblk; // D230860
#endif
#ifdef _JFS_OS2
FEALIST *pfealist = NULL;
#endif /* _JFS_OS2 */
NOISE(1,("jfs_mkdir: dip:0x%08x name:%s\n", dip, name));
/*
* the named file exists for "." or ".."
*/
if (name[0] == '.')
{
if ((name[1] == '.' && name[2] == '\0') ||
name[1] == '\0')
return EEXIST;
}
#ifdef _JFS_OS2
/* validate the EAOP buffer, FEALIST size storage location and the
* entire FEALIST storage area. Once all the storage has been
* validated, the entire FEALIST is validated for format and the size is
* computed and compared against the limit.
*/
if (pEABuf)
{
if (rc = jfs_ValidateUserFEAList(pEABuf, &pfealist,
&pEABuf->oError))
{
/* something failed -- bail out */
return rc;
}
}
#endif /* _JFS_OS2 */
if (dip->i_nlink == 0)
{
rc = ENOENT;
goto out1;
}
/* link count overflow on parent directory ? */
if (dip->i_nlink >= LINK_MAX)
{
rc = EMLINK;
goto out1;
}
/*
* search parent directory for entry/freespace
* (dtSearch() returns parent directory page pinned)
*/
dname.name = name;
dname.namlen = UniStrlen(name);
if (dname.namlen > JFS_NAME_MAX-1)
{
rc = ERROR_FILENAME_EXCED_RANGE;
goto out1;
}
// BEGIN D233382
#ifdef _JFS_LAZYCOMMIT
if (isReadOnly(dip))
{
rc = EROFS;
goto out1;
}
/*
* Either iAlloc() or txBegin() may block. Deadlock can occur if we
* block there while holding dtree page, so we allocate the inode &
* begin the transaction before we search the directory.
*/
if (rc = iAlloc(vfsp, dip, IFDIR|mode|IDIRECTORY, &ip, crp))
goto out1;
txBegin(dip->i_ipmnt, &tid, 0);
if (rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))
{
ip->i_nlink = 0;
ICACHE_LOCK();
iput(ip, vfsp);
ICACHE_UNLOCK();
txEnd(tid);
goto out1;
}
tblk = &TxBlock[tid]; // D230860
tblk->xflag |= COMMIT_CREATE; // D230860
tblk->ip = ip; // D230860
#else /* ! _JFS_LAZYCOMMIT */
// END D233382
if (rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))
goto out1;
if (isReadOnly(dip))
{
rc = EROFS;
/* release parent directory page */
BT_PUTSEARCH(&btstack);
goto out1;
}
/*
* allocate on-disk/in-memory inode for child directory:
* (iAlloc() returns new, locked inode)
*/
if (rc = iAlloc(vfsp, dip, IFDIR|mode|IDIRECTORY, &ip, crp))
{
/* release parent directory page */
BT_PUTSEARCH(&btstack);
jEVENT(0,("jfs_mkdir: iAlloc error(%d)\n", rc));
goto out1;
}
txBegin(dip->i_ipmnt, &tid, 0);
#endif /* _JFS_LAZYCOMMIT */
du = ip->i_dasdlim = dip->i_dasdlim; // F226941
orig_nblocks = dip->i_nblocks; // F226941
iplist[0] = dip;
iplist[1] = ip;
/*
* initialize the child directory in-line in inode
*/
dtInitRoot(tid, ip, dip->i_number);
/*
* create entry in parent directory for child directory
* (dtInsert() releases parent directory page)
*/
ino = ip->i_number;
if (rc = dtInsert(tid, dip, &dname, &ino, &btstack))
{
/* discard new directory inode */
ip->i_nlink = 0;
if (rc == EIO)
txAbort(tid, 1); /* Marks Filesystem dirty */
else
txAbort(tid, 0); /* Filesystem full */
txEnd(tid);
goto out3;
}
/* update child directory inode */
// BEGIN F226941
setDASDLIMIT(&ip->i_DASD, 0);
setDASDUSED(&ip->i_DASD, 0);
// END F226941
ip->i_nlink++; /* for '.' */
imark(ip, IACC|ICHG|IUPD|IFSYNC);
/* update parent directory inode */
dip->i_nlink++; /* for '..' from child directory */
imark(dip, ICHG|IUPD|IFSYNC);
/*
* insert entry for the new file to dnlc
*/
ncEnter(dip->i_ipimap, dip->i_number, &dname, ino, NULL);
#ifdef _JFS_OS2
if ((rc1 == 0) && pfealist) // F226941
rc1 = jfs_InitializeFEAList(ip, tid, pfealist);
if ((rc1 == 0) && (dip->i_acl.flag) && (dip->i_acl.size))
rc1 = jfs_CopyACL(ip, dip, tid);
#endif /* _JFS_OS2 */
// BEGIN FF226941
DLIM_UPDATE(tid, dip, dip->i_nblocks + ip->i_nblocks - orig_nblocks);
setDASDUSED(&ip->i_DASD, ip->i_nblocks); // D233382
#ifndef _JFS_FASTDASD // D233382
/*
* If the transaction modified the ancestors of the inode, the
* parent will be in the dasd usage list. Otherwise, transaction
* will only change new directory and parent.
*/
if (du && (du->flag & DLIM_LOGGED))
rc = dasd_commit(tid, ip, 0);
else
#endif /* _JFS_FASTDASD */ // D233382
// END FF226941
rc = txCommit(tid, 2, &iplist[0], 0);
txEnd(tid);
#ifdef _JFS_OS2
/* If we successfully added the directory, but failed adding the EA or
* ACL, we must cleanup the created directory entry and return the
* error.
*/
if (rc1)
{
// BEGIN D230860
if ((du) && (du->first_locked))
{
/*
* txCommit unlocked one or more inodes.
* We need to unlock all the directories &
* relock them.
*/
dip->i_dasdlim = 0;
DLIM_WRITE_UNLOCK_DETACH(ip, du);
DLIM_WRITE_LOCK_ATTACH(ip, du);
dip->i_dasdlim = du;
}
// END D230860
jfs_xrmdir(dip, ip, name);
rc = rc1;
}
#endif /* _JFS_OS2 */
out3:
ip->i_dasdlim = 0; // F226941
IWRITE_UNLOCK(ip);
ICACHE_LOCK();
iput(ip, vfsp);
ICACHE_UNLOCK();
out1:
#ifdef _JFS_OS2
/*
* this buffer was allocated by
* jfs_ValidateUserFEAList() at twice the
* size of the given list to provide buffer
* space for eliminating duplicate names
*/
if (pfealist)
jfs_EABufFree((char *)pfealist, (pfealist->cbList << 1));
#endif /* _JFS_OS2 */
NOISE(1,("jfs_mkdir: rc:%d\n", rc));
return rc;
}
/*
* NAME: jfs_defragfs()
*
* FUNCTION: relocate specified extent for defragfs()
* optionally commiting the operation.
*/
int32 jfs_defragfs(
char *pData, /* pointer to buffer containing plist */
uint32 lenData, /* length of buffer */
uint16 *pbufsize) /* pointer of buffer length */
{
int32 rc = 0;
defragfs_t pList, *p = &pList;
uint32 xtype;
int64 offset, xoff, oxaddr, nxaddr;
int32 xlen;
inode_t *ipmnt, *ipimap, *ipbmap;
inode_t *ip = NULL;
xad_t oxad;
pxd_t opxd;
mode_t imode;
int32 tid;
inode_t *iplist[1];
struct vfs *vfsp;
if (rc = KernCopyIn(&pList, pData, lenData))
return rc;
/* get the 'mount' inode */
for (vfsp = vfs_anchor; vfsp != NULL; vfsp = vfsp->vfs_next)
if (vfsp->vfs_vpfsi->vpi_drive == pList.dev)
break;
if (vfsp == NULL)
return EINVAL;
xtype = pList.flag;
/* sync at start of defragfs ? */
if (xtype & DEFRAGFS_SYNC)
{
jEVENT(0,("jfs_defragfs: DEFRAGFS_SYNC\n"));
if ((vfsp->vfs_flag & VFS_READONLY) ||
(vfsp->vfs_flag & VFS_ACCEPT))
return 0;
ipimap = (inode_t *)vfsp->vfs_data;
ipmnt = ipimap->i_ipmnt;
ipbmap = ipmnt->i_ipbmap;
/* sync the file system */
iSyncFS(vfsp);
/* write dirty pages of imap */
diSync(ipimap);
/* write dirty pages of bmap */
dbSync(ipbmap);
return 0;
}
else if (!(xtype & DEFRAGFS_RELOCATE))
return EINVAL;
if (vfsp->vfs_flag & VFS_READONLY)
return EROFS;
if (vfsp->vfs_flag & VFS_ACCEPT)
return EINVAL;
/* get the relocation parameter */
xoff = pList.xoff;
oxaddr = pList.old_xaddr;
nxaddr = pList.new_xaddr;
xlen = pList.xlen;
jEVENT(0,("jfs_defragfs: i:%d xtype:0x%08x xoff:%lld xlen:%d xaddr:%lld:%lld\n",
pList.ino, xtype, xoff, xlen, oxaddr, nxaddr));
/* get the object inode if it exist */
ICACHE_LOCK();
rc = iget(vfsp, pList.ino, &ip, 0);
ICACHE_UNLOCK();
if(rc)
{
jEVENT(0,("jfs_defragfs: stale target object.\n"));
rc = ESTALE; /* stale object ENOENT */
goto out1;
}
IWRITE_LOCK(ip);
/* validate inode */
if (ip->i_nlink == 0
|| ip->i_gen != pList.gen
|| ip->i_fileset != pList.fileset
|| ip->i_inostamp != pList.inostamp)
{
jEVENT(0,("jfs_defragfs: stale target object.\n"));
rc = ESTALE; /* stale object ENOENT */
goto out1;
}
/* validate object type: regular file or directory */
imode = ip->i_mode & IFMT;
switch(imode)
{
case IFREG:
case IFDIR:
break;
default:
rc = ESTALE; /* invalid object type ENOENT */
goto out1;
}
/*
* try to allocate new destination extent
*/
if (rc = dbAllocExact(ip, nxaddr, xlen))
{
jEVENT(0,("jfs_defragfs: stale destination extent.\n"));
rc = ENOSPC;
goto out1;
}
iBindCache(ip);
/*
* regular file:
*/
if (imode == IFREG)
{
/*
* automatic commit before and after each relocation
* may be skipped after more experience;
*/
/*
* commit any update before relocation
*/
if (ip->i_flag & IUPD)
{
ip->i_flag |= IFSYNC;
txBegin(ip->i_ipmnt, &tid, 0);
iplist[0] = ip;
rc = txCommit(tid, 1, &iplist[0], 0);
if (rc)
goto out2;
txEnd(tid);
}
/*
* relocate either xtpage or data extent
*/
txBegin(ip->i_ipmnt, &tid, 0);
/* source extent xad */
XADoffset(&oxad, xoff);
XADaddress(&oxad, oxaddr);
XADlength(&oxad, xlen);
rc = xtRelocate(tid, ip, &oxad, nxaddr, xtype);
if (rc)
goto out2;
iplist[0] = ip;
rc = txCommit(tid, 1, &iplist[0], 0);
if (rc)
goto out2;
txEnd(tid);
goto out1;
}
/*
* directory:
*/
else /* IFDIR */
{
/*
* relocate dtpage
*/
txBegin(ip->i_ipmnt, &tid, 0);
/* source extent pxd */
PXDaddress(&opxd, oxaddr);
PXDlength(&opxd, xlen);
rc = dtRelocate(tid, ip, xoff, &opxd, nxaddr);
if (rc)
goto out2;
iplist[0] = ip;
rc = txCommit(tid, 1, &iplist[0], 0);
if (rc)
goto out2;
txEnd(tid);
goto out1;
}
out2:
dbFree(ip, nxaddr, xlen) ;
out1:
if (ip)
{
IWRITE_UNLOCK(ip);
ICACHE_LOCK();
iput(ip, NULL);
ICACHE_UNLOCK();
}
jEVENT(0,("jfs_defragfs: rc=%d\n", rc));
return (rc);
}
