xfs_dquot.c

linux-2.6.15.6

	id = be32_to_cpu(dqp->q_core.d_id);
	nmaps = 1;
	newdquot = B_FALSE;

	/*
	 * If we don't know where the dquot lives, find out.
	 */
	if (dqp->q_blkno == (xfs_daddr_t) 0) {
		/* We use the id as an index */
		dqp->q_fileoffset = (xfs_fileoff_t)id / XFS_QM_DQPERBLK(mp);
		nmaps = 1;
		quotip = XFS_DQ_TO_QIP(dqp);
		xfs_ilock(quotip, XFS_ILOCK_SHARED);
		/*
		 * Return if this type of quotas is turned off while we didn't
		 * have an inode lock
		 */
		if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
			xfs_iunlock(quotip, XFS_ILOCK_SHARED);
			return (ESRCH);
		}
		/*
		 * Find the block map; no allocations yet
		 */
		error = xfs_bmapi(NULL, quotip, dqp->q_fileoffset,
				  XFS_DQUOT_CLUSTER_SIZE_FSB,
				  XFS_BMAPI_METADATA,
				  NULL, 0, &map, &nmaps, NULL);

		xfs_iunlock(quotip, XFS_ILOCK_SHARED);
		if (error)
			return (error);
		ASSERT(nmaps == 1);
		ASSERT(map.br_blockcount == 1);

		/*
		 * offset of dquot in the (fixed sized) dquot chunk.
		 */
		dqp->q_bufoffset = (id % XFS_QM_DQPERBLK(mp)) *
			sizeof(xfs_dqblk_t);
		if (map.br_startblock == HOLESTARTBLOCK) {
			/*
			 * We don't allocate unless we're asked to
			 */
			if (!(flags & XFS_QMOPT_DQALLOC))
				return (ENOENT);

			ASSERT(tp);
			if ((error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
						dqp->q_fileoffset, &bp)))
				return (error);
			tp = *tpp;
			newdquot = B_TRUE;
		} else {
			/*
			 * store the blkno etc so that we don't have to do the
			 * mapping all the time
			 */
			dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
		}
	}
	ASSERT(dqp->q_blkno != DELAYSTARTBLOCK);
	ASSERT(dqp->q_blkno != HOLESTARTBLOCK);

	/*
	 * Read in the buffer, unless we've just done the allocation
	 * (in which case we already have the buf).
	 */
	if (! newdquot) {
		xfs_dqtrace_entry(dqp, "DQTOBP READBUF");
		if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
					       dqp->q_blkno,
					       XFS_QI_DQCHUNKLEN(mp),
					       0, &bp))) {
			return (error);
		}
		if (error || !bp)
			return XFS_ERROR(error);
	}
	ASSERT(XFS_BUF_ISBUSY(bp));
	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);

	/*
	 * calculate the location of the dquot inside the buffer.
	 */
	ddq = (xfs_disk_dquot_t *)((char *)XFS_BUF_PTR(bp) + dqp->q_bufoffset);

	/*
	 * A simple sanity check in case we got a corrupted dquot...
	 */
	if (xfs_qm_dqcheck(ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
			   flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
			   "dqtobp")) {
		if (!(flags & XFS_QMOPT_DQREPAIR)) {
			xfs_trans_brelse(tp, bp);
			return XFS_ERROR(EIO);
		}
		XFS_BUF_BUSY(bp); /* We dirtied this */
	}

	*O_bpp = bp;
	*O_ddpp = ddq;

	return (0);
}


/*
 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
 * and release the buffer immediately.
 *
 */
/* ARGSUSED */
STATIC int
xfs_qm_dqread(
	xfs_trans_t	**tpp,
	xfs_dqid_t	id,
	xfs_dquot_t	*dqp,	/* dquot to get filled in */
	uint		flags)
{
	xfs_disk_dquot_t *ddqp;
	xfs_buf_t	 *bp;
	int		 error;
	xfs_trans_t	 *tp;

	ASSERT(tpp);

	/*
	 * get a pointer to the on-disk dquot and the buffer containing it
	 * dqp already knows its own type (GROUP/USER).
	 */
	xfs_dqtrace_entry(dqp, "DQREAD");
	if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) {
		return (error);
	}
	tp = *tpp;

	/* copy everything from disk dquot to the incore dquot */
	memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
	ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
	xfs_qm_dquot_logitem_init(dqp);

	/*
	 * Reservation counters are defined as reservation plus current usage
	 * to avoid having to add everytime.
	 */
	dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
	dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
	dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);

	/* Mark the buf so that this will stay incore a little longer */
	XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);

	/*
	 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
	 * So we need to release with xfs_trans_brelse().
	 * The strategy here is identical to that of inodes; we lock
	 * the dquot in xfs_qm_dqget() before making it accessible to
	 * others. This is because dquots, like inodes, need a good level of
	 * concurrency, and we don't want to take locks on the entire buffers
	 * for dquot accesses.
	 * Note also that the dquot buffer may even be dirty at this point, if
	 * this particular dquot was repaired. We still aren't afraid to
	 * brelse it because we have the changes incore.
	 */
	ASSERT(XFS_BUF_ISBUSY(bp));
	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
	xfs_trans_brelse(tp, bp);

	return (error);
}


/*
 * allocate an incore dquot from the kernel heap,
 * and fill its core with quota information kept on disk.
 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
 * if it wasn't already allocated.
 */
STATIC int
xfs_qm_idtodq(
	xfs_mount_t	*mp,
	xfs_dqid_t	id,	 /* gid or uid, depending on type */
	uint		type,	 /* UDQUOT or GDQUOT */
	uint		flags,	 /* DQALLOC, DQREPAIR */
	xfs_dquot_t	**O_dqpp)/* OUT : incore dquot, not locked */
{
	xfs_dquot_t	*dqp;
	int		error;
	xfs_trans_t	*tp;
	int		cancelflags=0;

	dqp = xfs_qm_dqinit(mp, id, type);
	tp = NULL;
	if (flags & XFS_QMOPT_DQALLOC) {
		tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
		if ((error = xfs_trans_reserve(tp,
				       XFS_QM_DQALLOC_SPACE_RES(mp),
				       XFS_WRITE_LOG_RES(mp) +
					      BBTOB(XFS_QI_DQCHUNKLEN(mp)) - 1 +
					      128,
				       0,
				       XFS_TRANS_PERM_LOG_RES,
				       XFS_WRITE_LOG_COUNT))) {
			cancelflags = 0;
			goto error0;
		}
		cancelflags = XFS_TRANS_RELEASE_LOG_RES;
	}

	/*
	 * Read it from disk; xfs_dqread() takes care of
	 * all the necessary initialization of dquot's fields (locks, etc)
	 */
	if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) {
		/*
		 * This can happen if quotas got turned off (ESRCH),
		 * or if the dquot didn't exist on disk and we ask to
		 * allocate (ENOENT).
		 */
		xfs_dqtrace_entry(dqp, "DQREAD FAIL");
		cancelflags |= XFS_TRANS_ABORT;
		goto error0;
	}
	if (tp) {
		if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
					     NULL)))
			goto error1;
	}

	*O_dqpp = dqp;
	return (0);

 error0:
	ASSERT(error);
	if (tp)
		xfs_trans_cancel(tp, cancelflags);
 error1:
	xfs_qm_dqdestroy(dqp);
	*O_dqpp = NULL;
	return (error);
}

/*
 * Lookup a dquot in the incore dquot hashtable. We keep two separate
 * hashtables for user and group dquots; and, these are global tables
 * inside the XQM, not per-filesystem tables.
 * The hash chain must be locked by caller, and it is left locked
 * on return. Returning dquot is locked.
 */
STATIC int
xfs_qm_dqlookup(
	xfs_mount_t		*mp,
	xfs_dqid_t		id,
	xfs_dqhash_t		*qh,
	xfs_dquot_t		**O_dqpp)
{
	xfs_dquot_t		*dqp;
	uint			flist_locked;
	xfs_dquot_t		*d;

	ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));

	flist_locked = B_FALSE;

	/*
	 * Traverse the hashchain looking for a match
	 */
	for (dqp = qh->qh_next; dqp != NULL; dqp = dqp->HL_NEXT) {
		/*
		 * We already have the hashlock. We don't need the
		 * dqlock to look at the id field of the dquot, since the
		 * id can't be modified without the hashlock anyway.
		 */
		if (be32_to_cpu(dqp->q_core.d_id) == id && dqp->q_mount == mp) {
			xfs_dqtrace_entry(dqp, "DQFOUND BY LOOKUP");
			/*
			 * All in core dquots must be on the dqlist of mp
			 */
			ASSERT(dqp->MPL_PREVP != NULL);

			xfs_dqlock(dqp);
			if (dqp->q_nrefs == 0) {
				ASSERT (XFS_DQ_IS_ON_FREELIST(dqp));
				if (! xfs_qm_freelist_lock_nowait(xfs_Gqm)) {
					xfs_dqtrace_entry(dqp, "DQLOOKUP: WANT");

					/*
					 * We may have raced with dqreclaim_one()
					 * (and lost). So, flag that we don't
					 * want the dquot to be reclaimed.
					 */
					dqp->dq_flags |= XFS_DQ_WANT;
					xfs_dqunlock(dqp);
					xfs_qm_freelist_lock(xfs_Gqm);
					xfs_dqlock(dqp);
					dqp->dq_flags &= ~(XFS_DQ_WANT);
				}
				flist_locked = B_TRUE;
			}

			/*
			 * id couldn't have changed; we had the hashlock all
			 * along
			 */
			ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);

			if (flist_locked) {
				if (dqp->q_nrefs != 0) {
					xfs_qm_freelist_unlock(xfs_Gqm);
					flist_locked = B_FALSE;
				} else {
					/*
					 * take it off the freelist
					 */
					xfs_dqtrace_entry(dqp,
							"DQLOOKUP: TAKEOFF FL");
					XQM_FREELIST_REMOVE(dqp);
					/* xfs_qm_freelist_print(&(xfs_Gqm->
							qm_dqfreelist),
							"after removal"); */
				}
			}

			/*
			 * grab a reference
			 */
			XFS_DQHOLD(dqp);

			if (flist_locked)
				xfs_qm_freelist_unlock(xfs_Gqm);
			/*
			 * move the dquot to the front of the hashchain
			 */
			ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
			if (dqp->HL_PREVP != &qh->qh_next) {
				xfs_dqtrace_entry(dqp,
						  "DQLOOKUP: HASH MOVETOFRONT");
				if ((d = dqp->HL_NEXT))
					d->HL_PREVP = dqp->HL_PREVP;
				*(dqp->HL_PREVP) = d;
				d = qh->qh_next;
				d->HL_PREVP = &dqp->HL_NEXT;
				dqp->HL_NEXT = d;
				dqp->HL_PREVP = &qh->qh_next;
				qh->qh_next = dqp;
			}
			xfs_dqtrace_entry(dqp, "LOOKUP END");
			*O_dqpp = dqp;
			ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
			return (0);
		}
	}

	*O_dqpp = NULL;
	ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
	return (1);
}

/*
 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
 * a locked dquot, doing an allocation (if requested) as needed.
 * When both an inode and an id are given, the inode's id takes precedence.
 * That is, if the id changes while we don't hold the ilock inside this
 * function, the new dquot is returned, not necessarily the one requested
 * in the id argument.
 */
int
xfs_qm_dqget(
	xfs_mount_t	*mp,
	xfs_inode_t	*ip,	  /* locked inode (optional) */
	xfs_dqid_t	id,	  /* uid/projid/gid depending on type */
	uint		type,	  /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
	uint		flags,	  /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
	xfs_dquot_t	**O_dqpp) /* OUT : locked incore dquot */
{
	xfs_dquot_t	*dqp;
	xfs_dqhash_t	*h;
	uint		version;
	int		error;

	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
	if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
	    (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
	    (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
		return (ESRCH);
	}
	h = XFS_DQ_HASH(mp, id, type);

#ifdef DEBUG
	if (xfs_do_dqerror) {
		if ((xfs_dqerror_target == mp->m_ddev_targp) &&
		    (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
			cmn_err(CE_DEBUG, "Returning error in dqget");
			return (EIO);
		}
	}
#endif

 again:

#ifdef DEBUG
	ASSERT(type == XFS_DQ_USER ||
	       type == XFS_DQ_PROJ ||
	       type == XFS_DQ_GROUP);
	if (ip) {
		ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
		if (type == XFS_DQ_USER)
			ASSERT(ip->i_udquot == NULL);
		else
			ASSERT(ip->i_gdquot == NULL);
	}
#endif
	XFS_DQ_HASH_LOCK(h);

	/*
	 * Look in the cache (hashtable).
	 * The chain is kept locked during lookup.
	 */
	if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
		XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
		/*
		 * The dquot was found, moved to the front of the chain,
		 * taken off the freelist if it was on it, and locked
		 * at this point. Just unlock the hashchain and return.
		 */
		ASSERT(*O_dqpp);
		ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
		XFS_DQ_HASH_UNLOCK(h);
		xfs_dqtrace_entry(*O_dqpp, "DQGET DONE (FROM CACHE)");
		return (0);	/* success */
	}
	XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);

	/*
	 * Dquot cache miss. We don't want to keep the inode lock across
	 * a (potential) disk read. Also we don't want to deal with the lock
	 * ordering between quotainode and this inode. OTOH, dropping the inode
	 * lock here means dealing with a chown that can happen before
	 * we re-acquire the lock.
	 */
	if (ip)
		xfs_iunlock(ip, XFS_ILOCK_EXCL);
	/*
	 * Save the hashchain version stamp, and unlock the chain, so that
	 * we don't keep the lock across a disk read
	 */
	version = h->qh_version;
	XFS_DQ_HASH_UNLOCK(h);

	/*
	 * Allocate the dquot on the kernel heap, and read the ondisk
	 * portion off the disk. Also, do all the necessary initialization
	 * This can return ENOENT if dquot didn't exist on disk and we didn't
	 * ask it to allocate; ESRCH if quotas got turned off suddenly.
	 */
	if ((error = xfs_qm_idtodq(mp, id, type,
				  flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
					   XFS_QMOPT_DOWARN),
				  &dqp))) {
		if (ip)
			xfs_ilock(ip, XFS_ILOCK_EXCL);
		return (error);
	}

	/*
	 * See if this is mount code calling to look at the overall quota limits
	 * which are stored in the id == 0 user or group's dquot.
	 * Since we may not have done a quotacheck by this point, just return
	 * the dquot without attaching it to any hashtables, lists, etc, or even
	 * taking a reference.
	 * The caller must dqdestroy this once done.
	 */
	if (flags & XFS_QMOPT_DQSUSER) {
		ASSERT(id == 0);
		ASSERT(! ip);
		goto dqret;
	}

	/*
	 * Dquot lock comes after hashlock in the lock ordering
	 */
	if (ip) {
		xfs_ilock(ip, XFS_ILOCK_EXCL);
		if (! XFS_IS_DQTYPE_ON(mp, type)) {
			/* inode stays locked on return */
			xfs_qm_dqdestroy(dqp);
			return XFS_ERROR(ESRCH);
		}
		/*
		 * A dquot could be attached to this inode by now, since
		 * we had dropped the ilock.
		 */
		if (type == XFS_DQ_USER) {
			if (ip->i_udquot) {
				xfs_qm_dqdestroy(dqp);
				dqp = ip->i_udquot;
				xfs_dqlock(dqp);
				goto dqret;
			}
		} else {
			if (ip->i_gdquot) {
				xfs_qm_dqdestroy(dqp);
				dqp = ip->i_gdquot;
				xfs_dqlock(dqp);
				goto dqret;
			}
		}
	}

	/*
	 * Hashlock comes after ilock in lock order
	 */
	XFS_DQ_HASH_LOCK(h);
	if (version != h->qh_version) {
		xfs_dquot_t *tmpdqp;
		/*
		 * Now, see if somebody else put the dquot in the
		 * hashtable before us. This can happen because we didn't
		 * keep the hashchain lock. We don't have to worry about
		 * lock order between the two dquots here since dqp isn't
		 * on any findable lists yet.
		 */
		if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
			/*
			 * Duplicate found. Just throw away the new dquot
			 * and start over.
			 */
			xfs_qm_dqput(tmpdqp);
			XFS_DQ_HASH_UNLOCK(h);
			xfs_qm_dqdestroy(dqp);
			XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
			goto again;
		}