/*
* 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;
}
int jfs_get_block(struct inode *ip, sector_t lblock,
struct buffer_head *bh_result, int create)
{
s64 lblock64 = lblock;
int rc = 0;
xad_t xad;
s64 xaddr;
int xflag;
s32 xlen = bh_result->b_size >> ip->i_blkbits;
/*
* Take appropriate lock on inode
*/
if (create)
IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
else
IREAD_LOCK(ip, RDWRLOCK_NORMAL);
if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
(!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
xaddr) {
if (xflag & XAD_NOTRECORDED) {
if (!create)
/*
* Allocated but not recorded, read treats
* this as a hole
*/
goto unlock;
#ifdef _JFS_4K
XADoffset(&xad, lblock64);
XADlength(&xad, xlen);
XADaddress(&xad, xaddr);
#else /* _JFS_4K */
/*
* As long as block size = 4K, this isn't a problem.
* We should mark the whole page not ABNR, but how
* will we know to mark the other blocks BH_New?
*/
BUG();
#endif /* _JFS_4K */
rc = extRecord(ip, &xad);
if (rc)
goto unlock;
set_buffer_new(bh_result);
}
map_bh(bh_result, ip->i_sb, xaddr);
bh_result->b_size = xlen << ip->i_blkbits;
goto unlock;
}
if (!create)
goto unlock;
/*
* Allocate a new block
*/
#ifdef _JFS_4K
if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
goto unlock;
rc = extAlloc(ip, xlen, lblock64, &xad, false);
if (rc)
goto unlock;
set_buffer_new(bh_result);
map_bh(bh_result, ip->i_sb, addressXAD(&xad));
bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
#else /* _JFS_4K */
/*
* We need to do whatever it takes to keep all but the last buffers
* in 4K pages - see jfs_write.c
*/
BUG();
#endif /* _JFS_4K */
unlock:
/*
* Release lock on inode
*/
if (create)
IWRITE_UNLOCK(ip);
else
IREAD_UNLOCK(ip);
return rc;
}
/*
* NAME: jfs_statfs(vfsp, sfsp, crp)
*
* FUNCTION: get file system status <sfsp> from vfs <vfsp>
*
* PARAMETER: vfsp - virtual file system
* sfsp - file status information structure
* crp - credential
*
* RETURN: zero on success, non-zero on failure
*
* serialization: statfs() and extendfs() serializes by inode lock
* of the inode map for both inode map and block allocation map.
* All other access to fragment allocation map is serialized
* under VMM locking.
*
* note: percolation of file system information:
* support struct statfs (sys/statfs.h) for get file system status
* service call statfs().
* (XPG4.2 defines struct statvfs in sys/statvfs.h for statvfs()
* which requires statfs() and additional information)
*/
jfs_statfs(
register struct vfs *vfsp,
register struct statfs *statfsp,
struct ucred *crp)
{
register int32 rc;
register inode_t *ipmnt; /* mount inode */
inode_t *ipimap, *ipbmap;
cbuf_t *bpsuper;
struct superblock *sb = NULL;
int32 fsck_length, log_length;
NOISE(1,("jfs_statfs: vfs:0x%08x\n", vfsp));
/*
* get the file system stats from the superblock
*/
ipimap = (struct inode *)vfsp->vfs_data;
ipmnt = ipimap->i_ipmnt;
if (rc = readSuper(ipmnt, &bpsuper))
goto out;
sb = (struct superblock *)(bpsuper->cm_cdata);
/*
bcopy(sb->s_fname, statfsp->f_fname, sizeof(sb->s_fname));
*/
bcopy(sb->s_fpack, statfsp->f_fpack, sizeof(sb->s_fpack));
statfsp->f_bsize = PSIZE; /* preferred i/o block size */
statfsp->f_fsize = sb->s_bsize; /* fundamental block size */
fsck_length = lengthPXD(&(sb->s_fsckpxd));
log_length = lengthPXD(&(sb->s_logpxd));
rawRelease(bpsuper);
/* statfs()/extendfs() serialized by inode lock of the inode map
* for both inode map and block allocation map.
*/
IREAD_LOCK(ipimap);
/*
* get the block stats from the bmap
*/
ipbmap = ipmnt->i_ipbmap;
statfsp->f_blocks = (ipbmap->i_bmap->db_mapsize + fsck_length +
log_length) >> ipbmap->i_bmap->db_l2nbperpage;
statfsp->f_bfree = statfsp->f_bavail =
ipbmap->i_bmap->db_nfree >> ipbmap->i_bmap->db_l2nbperpage;
/*
* get the file stats from the ipimap
*/
statfsp->f_files = ipimap->i_imap->im_numinos;
statfsp->f_ffree = ipimap->i_imap->im_numfree;
/*
* fill in from vfs
*/
statfsp->f_fsid = vfsp->vfs_fsid;
statfsp->f_vfstype = MNT_XJFS;
statfsp->f_vfsnumber = vfsp->vfs_number;
statfsp->f_name_max = JFS_NAME_MAX;
/*
* fields in the statfs structure that we don't fill in ...
*
long f_version; version/type of statfs, 0 for now
long f_type; type of info, 0 for now
long f_vfsoff; reserved, for vfs specific data offset
long f_vfslen; reserved, for len of vfs specific data
long f_vfsvers; reserved, for vers of vfs specific data
*/
out:
IREAD_UNLOCK(ipimap);
return rc;
}
static int
jfs_get_blocks(struct inode *ip, sector_t lblock, unsigned long max_blocks,
struct buffer_head *bh_result, int create)
{
s64 lblock64 = lblock;
int no_size_check = 0;
int rc = 0;
int take_locks;
xad_t xad;
s64 xaddr;
int xflag;
s32 xlen;
/*
* If this is a special inode (imap, dmap) or directory,
* the lock should already be taken
*/
take_locks = ((JFS_IP(ip)->fileset != AGGREGATE_I) &&
!S_ISDIR(ip->i_mode));
/*
* Take appropriate lock on inode
*/
if (take_locks) {
if (create)
IWRITE_LOCK(ip);
else
IREAD_LOCK(ip);
}
/*
* A directory's "data" is the inode index table, but i_size is the
* size of the d-tree, so don't check the offset against i_size
*/
if (S_ISDIR(ip->i_mode))
no_size_check = 1;
if ((no_size_check ||
((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size)) &&
(xtLookup(ip, lblock64, max_blocks, &xflag, &xaddr, &xlen, no_size_check)
== 0) && xlen) {
if (xflag & XAD_NOTRECORDED) {
if (!create)
/*
* Allocated but not recorded, read treats
* this as a hole
*/
goto unlock;
#ifdef _JFS_4K
XADoffset(&xad, lblock64);
XADlength(&xad, xlen);
XADaddress(&xad, xaddr);
#else /* _JFS_4K */
/*
* As long as block size = 4K, this isn't a problem.
* We should mark the whole page not ABNR, but how
* will we know to mark the other blocks BH_New?
*/
BUG();
#endif /* _JFS_4K */
rc = extRecord(ip, &xad);
if (rc)
goto unlock;
set_buffer_new(bh_result);
}
map_bh(bh_result, ip->i_sb, xaddr);
bh_result->b_size = xlen << ip->i_blkbits;
goto unlock;
}
if (!create)
goto unlock;
/*
* Allocate a new block
*/
#ifdef _JFS_4K
if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
goto unlock;
rc = extAlloc(ip, max_blocks, lblock64, &xad, FALSE);
if (rc)
goto unlock;
set_buffer_new(bh_result);
map_bh(bh_result, ip->i_sb, addressXAD(&xad));
bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
#else /* _JFS_4K */
/*
* We need to do whatever it takes to keep all but the last buffers
* in 4K pages - see jfs_write.c
*/
BUG();
#endif /* _JFS_4K */
unlock:
/*
* Release lock on inode
*/
if (take_locks) {
if (create)
IWRITE_UNLOCK(ip);
else
IREAD_UNLOCK(ip);
}
return rc;
}
/*
* 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_readlink(vp, uiop, crp)
*
* FUNCTION: read a symbolic link <vp> into <uiop>
*
* PARAMETER: vp - pointer to the vnode that represents the
* symlink we want to read
* uiop - How much to read and where it goes
* crp - credential
*
* RETURN: EINVAL - if not a symbolic link
* errors from subroutines
*/
jfs_readlink(
struct vnode *vp, /* symlink vnode */
struct uio *uiop,
struct ucred *crp)
{
int32 rc = 0;
inode_t *ip = VP2IP(vp);
int32 cnt;
jbuf_t *bp;
NOISE(1,("jfs_readlink: ip:0x%08x\n", ip));
if (vp->v_vntype != VLNK)
return EINVAL;
IREAD_LOCK(ip);
/* validate the buffer size vs target path name size
* (AES requires ERANGE if the link name won't fit)
*/
if (ip->i_size > uiop->uio_resid)
{
rc = ERANGE;
goto out;
}
/*
* read the target path name
*/
if (ip->i_size <= IDATASIZE)
{
/*
* fast symbolic link
*
* read target path name inline from on-disk inode
*/
cnt = MIN(ip->i_size, uiop->uio_iov->iov_len);
rc = uiomove(ip->i_fastsymlink, cnt, UIO_READ, uiop);
}
else
{
/*
* read target path name from a single extent
*
* target path name <= PATH_MAX < buffer page size
*
* even though the data of symlink object (target
* path name) is treated as non-journaled user data,
* it is read/written thru buffer cache for performance.
*/
if (rc = bmRead(ip, 0, ip->i_size, bmREAD_PAGE, &bp))
goto out;
cnt = MIN(ip->i_size, uiop->uio_iov->iov_len);
rc = uiomove(bp->b_bdata, cnt, UIO_READ, uiop);
bmRelease(bp);
}
out:
IREAD_UNLOCK(ip);
NOISE(1,("jfs_readlink: rc:%d\n", rc));
return rc;
}
/*
* NAME: dbFree()
*
* FUNCTION: free the specified block range from the working block
* allocation map.
*
* the blocks will be free from the working map one dmap
* at a time.
*
* PARAMETERS:
* ip - pointer to in-core inode;
* blkno - starting block number to be freed.
* nblocks - number of blocks to be freed.
*
* RETURN VALUES:
* 0 - success
* -EIO - i/o error
*/
int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
{
struct metapage *mp;
struct dmap *dp;
int nb, rc;
s64 lblkno, rem;
struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
struct super_block *sb = ipbmap->i_sb;
IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
/* block to be freed better be within the mapsize. */
if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
IREAD_UNLOCK(ipbmap);
printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
(unsigned long long) blkno,
(unsigned long long) nblocks);
jfs_error(ip->i_sb,
"dbFree: block to be freed is outside the map");
return -EIO;
}
/**
* TRIM the blocks, when mounted with discard option
*/
if (JFS_SBI(sb)->flag & JFS_DISCARD)
if (JFS_SBI(sb)->minblks_trim <= nblocks)
jfs_issue_discard(ipbmap, blkno, nblocks);
/*
* free the blocks a dmap at a time.
*/
mp = NULL;
for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
/* release previous dmap if any */
if (mp) {
write_metapage(mp);
}
/* get the buffer for the current dmap. */
lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
if (mp == NULL) {
IREAD_UNLOCK(ipbmap);
return -EIO;
}
dp = (struct dmap *) mp->data;
/* determine the number of blocks to be freed from
* this dmap.
*/
nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
/* free the blocks. */
if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
jfs_error(ip->i_sb, "dbFree: error in block map\n");
release_metapage(mp);
IREAD_UNLOCK(ipbmap);
return (rc);
}
}
/* write the last buffer. */
write_metapage(mp);
IREAD_UNLOCK(ipbmap);
return (0);
}
