GFS2: stuck in inode wait, no glocks stuck
A message that is not in the archive (sent off-list or not yet received): 138792730.573041273614686623.JavaMail.root@zmail06.collab.prod.int.phx2.redhat.com
PATCH 1/1 GFS2: stuck in inode wait, no glocks stuck
Hi,
This patch changes the lock ordering when gfs2 reclaims unlinked dinodes, thereby avoiding a livelock.
Regards,
Bob Peterson
Red Hat GFS
--
GFS2: stuck in inode wait, no glocks stuck
This patch changes the lock ordering when gfs2 reclaims unlinked dinodes, thereby avoiding a livelock.
rhbz#583737
fs/gfs2/rgrp.c +30 -48
@@ -952,16 +952,14 @@ static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)* The inode, if one has been found, in inode. */ -static int try_rgrp_unlink (struct gfs2_rgrpd *rgd,u64 *last_unlinked, -u64 skip,struct inode **inode) +static u64 try_rgrp_unlink (struct gfs2_rgrpd *rgd,u64 *last_unlinked, +u64 skip) {u32 goal =0 , block;u64 no_addr;struct gfs2_sbd *sdp = rgd->rd_sbd;unsigned int n; -int error =0 ; - *inode =NULL ;for (;;) {if (goal >= rgd->rd_data)break ;
@@ -981,10 +979,7 @@ static int try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked,if (no_addr == skip)continue ; *last_unlinked = no_addr; - error =gfs2_unlinked_inode_lookup (rgd->rd_sbd->sd_vfs, - no_addr, inode); -if (*inode || error) -return error; +return no_addr; } rgd->rd_flags &= ~GFS2_RDF_CHECK ;
@@ -1069,11 +1064,12 @@ static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)* Try to acquire rgrp in way which avoids contending with others. * * Returns: errno +* unlinked: the block address of an unlinked block to be reclaimed */ -static struct inode *get_local_rgrp (struct gfs2_inode *ip,u64 *last_unlinked) +static int get_local_rgrp (struct gfs2_inode *ip,u64 *unlinked, +u64 *last_unlinked) { -struct inode *inode =NULL ;struct gfs2_sbd *sdp =GFS2_SB (&ip->i_inode);struct gfs2_rgrpd *rgd, *begin =NULL ;struct gfs2_alloc *al = ip->i_alloc;
@@ -1082,6 +1078,7 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)int loops =0 ;int error, rg_locked; + *unlinked =0 ; rgd =gfs2_blk2rgrpd (sdp, ip->i_goal);while (rgd) {
@@ -1103,29 +1100,19 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)because that would require an iput which can only happen after the rgrp is unlocked. */ if (!rg_locked && rgd->rd_flags &GFS2_RDF_CHECK ) - error =try_rgrp_unlink (rgd, last_unlinked, - ip->i_no_addr, &inode); + *unlinked =try_rgrp_unlink (rgd, last_unlinked, + ip->i_no_addr);if (!rg_locked)gfs2_glock_dq_uninit (&al->al_rgd_gh); -if (inode) { -if (error) { -if (inode->i_state &I_NEW ) -iget_failed (inode); -else -iput (inode); -return ERR_PTR (error); - } -return inode; - } -if (error) -return ERR_PTR (error); +if (*unlinked) +return -EAGAIN ;/* fall through */ case GLR_TRYFAILED : rgd =recent_rgrp_next (rgd);break ;default : -return ERR_PTR (error); +return error; } }
@@ -1148,22 +1135,12 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)if (try_rgrp_fit (rgd, al)) goto out;if (!rg_locked && rgd->rd_flags &GFS2_RDF_CHECK ) - error =try_rgrp_unlink (rgd, last_unlinked, - ip->i_no_addr, &inode); + *unlinked =try_rgrp_unlink (rgd, last_unlinked, + ip->i_no_addr);if (!rg_locked)gfs2_glock_dq_uninit (&al->al_rgd_gh); -if (inode) { -if (error) { -if (inode->i_state &I_NEW ) -iget_failed (inode); -else -iput (inode); -return ERR_PTR (error); - } -return inode; - } -if (error) -return ERR_PTR (error); +if (*unlinked) +return -EAGAIN ;break ;case GLR_TRYFAILED :
@@ -1171,7 +1148,7 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)break ;default : -return ERR_PTR (error); +return error; } rgd =gfs2_rgrpd_get_next (rgd);
@@ -1180,7 +1157,7 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)if (rgd == begin) {if (++loops >=3 ) -return ERR_PTR (-ENOSPC ); +return -ENOSPC ;if (!skipped) loops++; flags =0 ;
@@ -1200,7 +1177,7 @@ out:forward_rgrp_set (sdp, rgd); } -return NULL ; +return 0 ; }/**
@@ -1216,7 +1193,7 @@ int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line)struct gfs2_alloc *al = ip->i_alloc;struct inode *inode;int error =0 ; -u64 last_unlinked =NO_BLOCK ; +u64 last_unlinked =NO_BLOCK , unlinked;if (gfs2_assert_warn (sdp, al->al_requested))return -EINVAL ;
@@ -1232,14 +1209,19 @@ try_again:if (error)return error; - inode =get_local_rgrp (ip, &last_unlinked); -if (inode) { + error =get_local_rgrp (ip, &unlinked, &last_unlinked); +if (error) {if (ip !=GFS2_I (sdp->sd_rindex))gfs2_glock_dq_uninit (&al->al_ri_gh); -if (IS_ERR (inode)) -return PTR_ERR (inode); -iput (inode); +if (error != -EAGAIN ) +return error; + error =gfs2_unlinked_inode_lookup (ip->i_inode.i_sb, + unlinked, &inode); +if (inode) +iput (inode);gfs2_log_flush (sdp,NULL ); +if (error ==GLR_TRYFAILED ) + error =0 ; goto try_again; }
PATCH 1/1 GFS2: stuck in inode wait, no glocks stuck
Hi,
Now in the -nmw git tree. Thanks,
Steve.
Quoted 200 lines · Bob Peterson
Hi,
This patch changes the lock ordering when gfs2 reclaims unlinked dinodes, thereby avoiding a livelock.
Regards,
Bob Peterson
Red Hat GFSSigned-off-by Bob Peterson <[email protected]>--
GFS2: stuck in inode wait, no glocks stuckThis patch changes the lock ordering when gfs2 reclaims unlinked dinodes, thereby avoiding a livelock.
rhbz#583737
fs/gfs2/rgrp.c +30
-48@@ -952,16 +952,14 @@ static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al)* The inode, if one has been found, in inode. */ -static int try_rgrp_unlink (struct gfs2_rgrpd *rgd,u64 *last_unlinked, -u64 skip,struct inode **inode) +static u64 try_rgrp_unlink (struct gfs2_rgrpd *rgd,u64 *last_unlinked, +u64 skip) {u32 goal =0 , block;u64 no_addr;struct gfs2_sbd *sdp = rgd->rd_sbd;unsigned int n; -int error =0 ; - *inode =NULL ;for (;;) {if (goal >= rgd->rd_data)break ;@@ -981,10 +979,7 @@ static int try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked,if (no_addr == skip)continue ; *last_unlinked = no_addr; - error =gfs2_unlinked_inode_lookup (rgd->rd_sbd->sd_vfs, - no_addr, inode); -if (*inode || error) -return error; +return no_addr; } rgd->rd_flags &= ~GFS2_RDF_CHECK ;@@ -1069,11 +1064,12 @@ static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd)* Try to acquire rgrp in way which avoids contending with others. * * Returns: errno +* unlinked: the block address of an unlinked block to be reclaimed */ -static struct inode *get_local_rgrp (struct gfs2_inode *ip,u64 *last_unlinked) +static int get_local_rgrp (struct gfs2_inode *ip,u64 *unlinked, +u64 *last_unlinked) { -struct inode *inode =NULL ;struct gfs2_sbd *sdp =GFS2_SB (&ip->i_inode);struct gfs2_rgrpd *rgd, *begin =NULL ;struct gfs2_alloc *al = ip->i_alloc;@@ -1082,6 +1078,7 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)int loops =0 ;int error, rg_locked; + *unlinked =0 ; rgd =gfs2_blk2rgrpd (sdp, ip->i_goal);while (rgd) {@@ -1103,29 +1100,19 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)because that would require an iput which can only happen after the rgrp is unlocked. */ if (!rg_locked && rgd->rd_flags &GFS2_RDF_CHECK ) - error =try_rgrp_unlink (rgd, last_unlinked, - ip->i_no_addr, &inode); + *unlinked =try_rgrp_unlink (rgd, last_unlinked, + ip->i_no_addr);if (!rg_locked)gfs2_glock_dq_uninit (&al->al_rgd_gh); -if (inode) { -if (error) { -if (inode->i_state &I_NEW ) -iget_failed (inode); -else -iput (inode); -return ERR_PTR (error); - } -return inode; - } -if (error) -return ERR_PTR (error); +if (*unlinked) +return -EAGAIN ;/* fall through */ case GLR_TRYFAILED : rgd =recent_rgrp_next (rgd);break ;default : -return ERR_PTR (error); +return error; } }@@ -1148,22 +1135,12 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)if (try_rgrp_fit (rgd, al)) goto out;if (!rg_locked && rgd->rd_flags &GFS2_RDF_CHECK ) - error =try_rgrp_unlink (rgd, last_unlinked, - ip->i_no_addr, &inode); + *unlinked =try_rgrp_unlink (rgd, last_unlinked, + ip->i_no_addr);if (!rg_locked)gfs2_glock_dq_uninit (&al->al_rgd_gh); -if (inode) { -if (error) { -if (inode->i_state &I_NEW ) -iget_failed (inode); -else -iput (inode); -return ERR_PTR (error); - } -return inode; - } -if (error) -return ERR_PTR (error); +if (*unlinked) +return -EAGAIN ;break ;case GLR_TRYFAILED :@@ -1171,7 +1148,7 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)break ;default : -return ERR_PTR (error); +return error; } rgd =gfs2_rgrpd_get_next (rgd);@@ -1180,7 +1157,7 @@ static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked)if (rgd == begin) {if (++loops >=3 ) -return ERR_PTR (-ENOSPC ); +return -ENOSPC ;if (!skipped) loops++; flags =0 ;@@ -1200,7 +1177,7 @@ out:forward_rgrp_set (sdp, rgd); } -return NULL ; +return 0 ; }/** @@ -1216,7 +1193,7 @@ int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line)struct gfs2_alloc *al = ip->i_alloc;struct inode *inode;int error =0 ; -u64 last_unlinked =NO_BLOCK ; +u64 last_unlinked =NO_BLOCK , unlinked;if (gfs2_assert_warn (sdp, al->al_requested))return -EINVAL ;@@ -1232,14 +1209,19 @@ try_again:if (error)return error; - inode =get_local_rgrp (ip, &last_unlinked); -if (inode) { + error =get_local_rgrp (ip, &unlinked, &last_unlinked); +if (error) {if (ip !=GFS2_I (sdp->sd_rindex))gfs2_glock_dq_uninit (&al->al_ri_gh); -if (IS_ERR (inode)) -return PTR_ERR (inode); -iput (inode); +if (error != -EAGAIN ) +return error; + error =gfs2_unlinked_inode_lookup (ip->i_inode.i_sb, + unlinked, &inode); +if (inode) +iput (inode);gfs2_log_flush (sdp,NULL ); +if (error ==GLR_TRYFAILED ) + error =0 ; goto try_again; }