From mboxrd@z Thu Jan 1 00:00:00 1970 From: NeilBrown Date: Mon, 22 Jul 2019 14:56:32 +1000 Subject: [lustre-devel] [PATCH 09/22] ext4: fix mballoc pa free mismatch In-Reply-To: <1563758631-29550-10-git-send-email-jsimmons@infradead.org> References: <1563758631-29550-1-git-send-email-jsimmons@infradead.org> <1563758631-29550-10-git-send-email-jsimmons@infradead.org> Message-ID: <87d0i2pszj.fsf@notabene.neil.brown.name> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: lustre-devel@lists.lustre.org On Sun, Jul 21 2019, James Simmons wrote: > Intoduce pa_error so we can find any mballoc pa cleanup problems. This patch seems to make sense, though > + BUG_ON(atomic_read(&sb->s_active) > 0 && pa->pa_free != free); should probably be a WARN_ON and > +#include seems un-called for. NeilBrown > > Signed-off-by: James Simmons > --- > fs/ext4/mballoc.c | 45 +++++++++++++++++++++++++++++++++++++++------ > fs/ext4/mballoc.h | 2 ++ > 2 files changed, 41 insertions(+), 6 deletions(-) > > diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c > index 483fc0f..463fba6 100644 > --- a/fs/ext4/mballoc.c > +++ b/fs/ext4/mballoc.c > @@ -3863,6 +3863,7 @@ static void ext4_mb_put_pa(struct ext4_allocation_context *ac, > INIT_LIST_HEAD(&pa->pa_group_list); > pa->pa_deleted = 0; > pa->pa_type = MB_INODE_PA; > + pa->pa_error = 0; > > mb_debug(1, "new inode pa %p: %llu/%u for %u\n", pa, > pa->pa_pstart, pa->pa_len, pa->pa_lstart); > @@ -3924,6 +3925,7 @@ static void ext4_mb_put_pa(struct ext4_allocation_context *ac, > INIT_LIST_HEAD(&pa->pa_group_list); > pa->pa_deleted = 0; > pa->pa_type = MB_GROUP_PA; > + pa->pa_error = 0; > > mb_debug(1, "new group pa %p: %llu/%u for %u\n", pa, > pa->pa_pstart, pa->pa_len, pa->pa_lstart); > @@ -3983,7 +3985,9 @@ static int ext4_mb_new_preallocation(struct ext4_allocation_context *ac) > unsigned long long grp_blk_start; > int free = 0; > > + assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group)); > BUG_ON(pa->pa_deleted == 0); > + BUG_ON(pa->pa_inode == NULL); > ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit); > grp_blk_start = pa->pa_pstart - EXT4_C2B(sbi, bit); > BUG_ON(group != e4b->bd_group && pa->pa_len != 0); > @@ -4006,12 +4010,19 @@ static int ext4_mb_new_preallocation(struct ext4_allocation_context *ac) > mb_free_blocks(pa->pa_inode, e4b, bit, next - bit); > bit = next + 1; > } > - if (free != pa->pa_free) { > - ext4_msg(e4b->bd_sb, KERN_CRIT, > - "pa %p: logic %lu, phys. %lu, len %lu", > - pa, (unsigned long) pa->pa_lstart, > - (unsigned long) pa->pa_pstart, > - (unsigned long) pa->pa_len); > + > + /* "free < pa->pa_free" means we maybe double alloc the same blocks, > + * otherwise maybe leave some free blocks unavailable, no need to BUG. > + */ > + if ((free > pa->pa_free && !pa->pa_error) || (free < pa->pa_free)) { > + ext4_error(sb, "pa free mismatch: [pa %p] " > + "[phy %lu] [logic %lu] [len %u] [free %u] " > + "[error %u] [inode %lu] [freed %u]", pa, > + (unsigned long)pa->pa_pstart, > + (unsigned long)pa->pa_lstart, > + (unsigned)pa->pa_len, (unsigned)pa->pa_free, > + (unsigned)pa->pa_error, pa->pa_inode->i_ino, > + free); > ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u", > free, pa->pa_free); > /* > @@ -4019,6 +4030,8 @@ static int ext4_mb_new_preallocation(struct ext4_allocation_context *ac) > * from the bitmap and continue. > */ > } > + /* do not verify if the file system is being umounted */ > + BUG_ON(atomic_read(&sb->s_active) > 0 && pa->pa_free != free); > atomic_add(free, &sbi->s_mb_discarded); > > return 0; > @@ -4764,6 +4777,26 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle, > ac->ac_b_ex.fe_len = 0; > ar->len = 0; > ext4_mb_show_ac(ac); > + if (ac->ac_pa) { > + struct ext4_prealloc_space *pa = ac->ac_pa; > + > + /* We can not make sure whether the bitmap has > + * been updated or not when fail case. So can > + * not revert pa_free back, just mark pa_error > + */ > + pa->pa_error++; > + ext4_error(sb, > + "Updating bitmap error: [err %d] " > + "[pa %p] [phy %lu] [logic %lu] " > + "[len %u] [free %u] [error %u] " > + "[inode %lu]", *errp, pa, > + (unsigned long)pa->pa_pstart, > + (unsigned long)pa->pa_lstart, > + (unsigned)pa->pa_len, > + (unsigned)pa->pa_free, > + (unsigned)pa->pa_error, > + pa->pa_inode ? pa->pa_inode->i_ino : 0); > + } > } > ext4_mb_release_context(ac); > out: > diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h > index 8325ad9..e00c3b7 100644 > --- a/fs/ext4/mballoc.h > +++ b/fs/ext4/mballoc.h > @@ -20,6 +20,7 @@ > #include > #include > #include > +#include > #include "ext4_jbd2.h" > #include "ext4.h" > > @@ -111,6 +112,7 @@ struct ext4_prealloc_space { > ext4_grpblk_t pa_len; /* len of preallocated chunk */ > ext4_grpblk_t pa_free; /* how many blocks are free */ > unsigned short pa_type; /* pa type. inode or group */ > + unsigned short pa_error; > spinlock_t *pa_obj_lock; > struct inode *pa_inode; /* hack, for history only */ > }; > -- > 1.8.3.1 -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 832 bytes Desc: not available URL: