New rescue mount options
cover New rescue mount options
Hello,
This is the next version of my rescue=all patches, this time broken up in to a few discrete patches, with some cleanups as well. I have a PR for the xfstest that exercises these options, it can be found here
https://github.com/btrfs/fstests/pull/35
This is the same idea as the previous versions, except I've made a mechanical change. Instead we have rescue=ignorebadroots, which ignores any global roots that we may not be able to read. We will still fail to mount if thinks like the chunk root or the tree root are corrupt, but this will allow us to mount read only if the extent root or csum root are completely hosed.
I've added a new patch this go around, rescue=ignoredatacsums. Somebody had indicated that they would prefer that the original rescue=all allowed us to continue to read csums if the root was still intact. In order to handle that usecase that's what you get with rescue=ignorebadroots, however if your csum tree is corrupt in the middle of the tree you could still end up with problems. Thus we have rescue=ignoredatacsums which will completely disable the csum tree as well.
And finally we have rescue=all, which simply enables ignoredatacsums, ignorebadroots, and notreelogreplay. We need an easy catch-all option for distros to fallback on to get users the highest probability of being able to recover their data, so we will use rescue=all to turn on all the fanciest rescue options that we have, and then use the discrete options for more fine grained recovery. Thanks,
Josef
Josef Bacik (5):
btrfs: unify the ro checking for mount options
btrfs: push the NODATASUM check into btrfs_lookup_bio_sums
btrfs: introduce rescue=ignorebadroots
btrfs: introduce rescue=ignoredatacsums
btrfs: introduce rescue=all
fs/btrfs/block-group.c | 48 +++++++++++++++++++++++++++ fs/btrfs/block-rsv.c | 8 +++++ fs/btrfs/compression.c | 17 ++++------ fs/btrfs/ctree.h | 2 ++ fs/btrfs/disk-io.c | 74 +++++++++++++++++++++++++++--------------- fs/btrfs/file-item.c | 4 +++ fs/btrfs/inode.c | 18 +++++++--- fs/btrfs/super.c | 49 ++++++++++++++++++++++++---- fs/btrfs/volumes.c | 7 ++++ 9 files changed, 179 insertions(+), 48 deletions(-)
Signature · 1 line
2.26.2
PATCH 1/5 btrfs: unify the ro checking for mount options
We're going to be adding more options that require RDONLY, so add a helper to do the check and error out if we don't have RDONLY set.
--- fs/btrfs/super.c | 23 ++++++++++++++++------- 1 file changed, 16 insertions(+), 7 deletions(-)
fs/btrfs/super.c +16 -7
@@ -458,6 +458,17 @@ static const match_table_t rescue_tokens = { {Opt_err, NULL}, }; +static bool check_ro_option (struct btrfs_fs_info *fs_info,unsigned long opt, +const char *opt_name) +{ +if (fs_info->mount_opt & opt) { +btrfs_err (fs_info,"%s must be used with ro mount option" , + opt_name); +return true; + } +return false; +} +static int parse_rescue_options (struct btrfs_fs_info *info,const char *options) {char *opts;
@@ -968,14 +979,12 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options, } } check: -/* -* Extra check for current option against current flag -*/ -if (btrfs_test_opt (info,NOLOGREPLAY ) && !(new_flags &SB_RDONLY )) { -btrfs_err (info, -"nologreplay must be used with ro mount option" ); +/* We're read-only, don't have to check. */ +if (new_flags &SB_RDONLY ) + goto out; + +if (check_ro_option (info,BTRFS_MOUNT_NOLOGREPLAY ,"nologreplay" )) ret = -EINVAL ; - } out: if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) && !btrfs_test_opt(info, FREE_SPACE_TREE) &&
Signature · 1 line
2.26.2
Re: [PATCH 1/5] btrfs: unify the ro checking for mount options
Quoted 4 lines · Josef Bacik
We're going to be adding more options that require RDONLY, so add a helper to do the check and error out if we don't have RDONLY set.
Signed-off-by Josef Bacik <[email protected]>
Thanks,
Qu
Quoted 47 lines
--- fs/btrfs/super.c | 23 ++++++++++++++++------- 1 file changed, 16 insertions(+), 7 deletions(-)fs/btrfs/super.c +16
-7@@ -458,6 +458,17 @@ static const match_table_t rescue_tokens = { {Opt_err, NULL}, }; +static bool check_ro_option (struct btrfs_fs_info *fs_info,unsigned long opt, +const char *opt_name) +{ +if (fs_info->mount_opt & opt) { +btrfs_err (fs_info,"%s must be used with ro mount option" , + opt_name); +return true; + } +return false; +} +static int parse_rescue_options (struct btrfs_fs_info *info,const char *options) {char *opts;@@ -968,14 +979,12 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options, } } check: -/* -* Extra check for current option against current flag -*/ -if (btrfs_test_opt (info,NOLOGREPLAY ) && !(new_flags &SB_RDONLY )) { -btrfs_err (info, -"nologreplay must be used with ro mount option" ); +/* We're read-only, don't have to check. */ +if (new_flags &SB_RDONLY ) + goto out; + +if (check_ro_option (info,BTRFS_MOUNT_NOLOGREPLAY ,"nologreplay" )) ret = -EINVAL ; - } out: if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) && !btrfs_test_opt(info, FREE_SPACE_TREE) &&
Re: [PATCH 1/5] btrfs: unify the ro checking for mount options
In reply to Josef Bacikfs/btrfs/super.c int btrfs_parse_options(struct btrfs_fs_info *info, char *options, · line 985
Quoted 7 lines · Josef Bacik
We're going to be adding more options that require RDONLY, so add a helper to do the check and error out if we don't have RDONLY set.
fs/btrfs/super.c +4
-0+/* We're read-only, don't have to check. */ +if (new_flags &SB_RDONLY ) + goto out; +
Why aren't you moving the SB_RDONLY check into the new check_ro_option() as well? This is what I would have thought this patch does after just reading the commit message.
Thanks,
Johannes
Re: [PATCH 1/5] btrfs: unify the ro checking for mount options
Quoted 4 lines · Johannes Thumshirn
Quoted 3 lines · Josef Bacik
We're going to be adding more options that require RDONLY, so add a helper to do the check and error out if we don't have RDONLY set.
Quoted 8 lines
fs/btrfs/super.c int btrfs_parse_options(struct btrfs_fs_info *info, char *options, · line 985
Quoted 4 lines
fs/btrfs/super.c +4
-0+/* We're read-only, don't have to check. */ +if (new_flags &SB_RDONLY ) + goto out; +Why aren't you moving the
SB_RDONLYcheck into the newcheck_ro_option()as well? This is what I would have thought this patch does after just reading the commit message.
To avoid the multiple calls if we're not read only, otherwise it'll be multiple function calls to check that that SB_RDONLY is set. The compiler will probably optimize that away, but I just went with this instead. I'm good either way if people have strong opinions one way or the other. Thanks,
Josef
Re: [PATCH 1/5] btrfs: unify the ro checking for mount options
Quoted 4 lines · Josef Bacik
To avoid the multiple calls if we're not read only, otherwise it'll be multiple function calls to check that that
SB_RDONLYis set. The compiler will probably optimize that away, but I just went with this instead. I'm good either way if people have strong opinions one way or the other.
The name check_ro_option() really suggests it checking for a read-only option and having multiple calls if RO isn't set really won't be a problem IMHO as we're not really in the fast-path here, but I don't have a strong opinion on it either.
PATCH 2/5 btrfs: push the NODATASUM check into btrfs_lookup_bio_sums
In reply to Josef BacikWhen we move to being able to handle NULL csum_roots it'll be cleaner to just check in btrfs_lookup_bio_sums instead of at all of the caller locations, so push the NODATASUM check into it as well so it's unified.
--- fs/btrfs/compression.c | 14 +++++--------- fs/btrfs/file-item.c | 3 +++ fs/btrfs/inode.c | 12 +++++++++--- 3 files changed, 17 insertions(+), 12 deletions(-)
fs/btrfs/compression.c +5 -9
@@ -722,11 +722,9 @@ blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,*/ refcount_inc (&cb->pending_bios); -if (!(BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM )) { - ret =btrfs_lookup_bio_sums (inode, comp_bio, - (u64)-1 , sums); -BUG_ON (ret);/* -ENOMEM */ - } + ret =btrfs_lookup_bio_sums (inode, comp_bio, (u64)-1 , + sums); +BUG_ON (ret);/* -ENOMEM */ nr_sectors =DIV_ROUND_UP (comp_bio->bi_iter.bi_size, fs_info->sectorsize);
@@ -751,10 +749,8 @@ blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, ret =btrfs_bio_wq_end_io (fs_info, comp_bio,BTRFS_WQ_ENDIO_DATA );BUG_ON (ret);/* -ENOMEM */ -if (!(BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM )) { - ret =btrfs_lookup_bio_sums (inode, comp_bio, (u64)-1 , sums); -BUG_ON (ret);/* -ENOMEM */ - } + ret =btrfs_lookup_bio_sums (inode, comp_bio, (u64)-1 , sums); +BUG_ON (ret);/* -ENOMEM */ ret =btrfs_map_bio (fs_info, comp_bio, mirror_num);if (ret) {
fs/btrfs/file-item.c +3 -0
@@ -272,6 +272,9 @@ blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,int count =0 ;u16 csum_size =btrfs_super_csum_size (fs_info->super_copy); +if (BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM ) +return BLK_STS_OK ; + path =btrfs_alloc_path ();if (!path)return BLK_STS_RESOURCE ;
fs/btrfs/inode.c +9 -3
@@ -2202,7 +2202,12 @@ blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio, mirror_num, bio_flags); goto out; - }else if (!skip_sum) { + }else { +/* +* Lookup bio sums does extra checks around whether we +* need to csum or not, which is why we ignore skip_sum +* here. +*/ ret =btrfs_lookup_bio_sums (inode, bio, (u64)-1 ,NULL );if (ret) goto out;
@@ -7781,7 +7786,6 @@ static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap, struct bio *dio_bio, loff_t file_offset) {const bool write = (bio_op (dio_bio) ==REQ_OP_WRITE ); -const bool csum = !(BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM );struct btrfs_fs_info *fs_info =btrfs_sb (inode->i_sb);const bool raid56 = (btrfs_data_alloc_profile (fs_info) &BTRFS_BLOCK_GROUP_RAID56_MASK );
@@ -7808,10 +7812,12 @@ static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,return BLK_QC_T_NONE ; } -if (!write && csum) { +if (!write) {/* * Load the csums up front to reduce csum tree searches and * contention when submitting bios. +* +* If we have csums disabled this will do nothing. */ status =btrfs_lookup_bio_sums (inode, dio_bio, file_offset, dip->csums);
Signature · 1 line
2.26.2
Re: [PATCH 2/5] btrfs: push the NODATASUM check into btrfs_lookup_bio_sums
Quoted 5 lines · Josef Bacik
When we move to being able to handle NULL
csum_rootsit'll be cleaner to just check inbtrfs_lookup_bio_sumsinstead of at all of the caller locations, so push the NODATASUM check into it as well so it's unified.Signed-off-by Josef Bacik <[email protected]>
But an off-topic question inlined below:
Quoted 11 lines
--- fs/btrfs/compression.c | 14 +++++--------- fs/btrfs/file-item.c | 3 +++ fs/btrfs/inode.c | 12 +++++++++--- 3 files changed, 17 insertions(+), 12 deletions(-)fs/btrfs/compression.c +0
-0
fs/btrfs/compression.c blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, · line 728
Quoted 7 lines
fs/btrfs/compression.c +0
-4@@ -722,11 +722,9 @@ blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,*/ refcount_inc (&cb->pending_bios); -if (!(BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM )) { - ret =btrfs_lookup_bio_sums (inode, comp_bio, - (u64)-1 , sums); -BUG_ON (ret);/* -ENOMEM */
Is it really possible to have compressed extent without data csum? Won't nodatacsum disable compression?
Or are we just here to handle some old compressed but not csumed data?
Thanks,
Qu
Quoted 75 lines
fs/btrfs/compression.c +5
-5- } + ret =btrfs_lookup_bio_sums (inode, comp_bio, (u64)-1 , + sums); +BUG_ON (ret);/* -ENOMEM */ nr_sectors =DIV_ROUND_UP (comp_bio->bi_iter.bi_size, fs_info->sectorsize);@@ -751,10 +749,8 @@ blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, ret =btrfs_bio_wq_end_io (fs_info, comp_bio,BTRFS_WQ_ENDIO_DATA );BUG_ON (ret);/* -ENOMEM */ -if (!(BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM )) { - ret =btrfs_lookup_bio_sums (inode, comp_bio, (u64)-1 , sums); -BUG_ON (ret);/* -ENOMEM */ - } + ret =btrfs_lookup_bio_sums (inode, comp_bio, (u64)-1 , sums); +BUG_ON (ret);/* -ENOMEM */ ret =btrfs_map_bio (fs_info, comp_bio, mirror_num);if (ret) {fs/btrfs/file-item.c +3
-0@@ -272,6 +272,9 @@ blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,int count =0 ;u16 csum_size =btrfs_super_csum_size (fs_info->super_copy); +if (BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM ) +return BLK_STS_OK ; + path =btrfs_alloc_path ();if (!path)return BLK_STS_RESOURCE ;fs/btrfs/inode.c +9
-3@@ -2202,7 +2202,12 @@ blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio, mirror_num, bio_flags); goto out; - }else if (!skip_sum) { + }else { +/* +* Lookup bio sums does extra checks around whether we +* need to csum or not, which is why we ignore skip_sum +* here. +*/ ret =btrfs_lookup_bio_sums (inode, bio, (u64)-1 ,NULL );if (ret) goto out;@@ -7781,7 +7786,6 @@ static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap, struct bio *dio_bio, loff_t file_offset) {const bool write = (bio_op (dio_bio) ==REQ_OP_WRITE ); -const bool csum = !(BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM );struct btrfs_fs_info *fs_info =btrfs_sb (inode->i_sb);const bool raid56 = (btrfs_data_alloc_profile (fs_info) &BTRFS_BLOCK_GROUP_RAID56_MASK );@@ -7808,10 +7812,12 @@ static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,return BLK_QC_T_NONE ; } -if (!write && csum) { +if (!write) {/* * Load the csums up front to reduce csum tree searches and * contention when submitting bios. +* +* If we have csums disabled this will do nothing. */ status =btrfs_lookup_bio_sums (inode, dio_bio, file_offset, dip->csums);
Re: [PATCH 2/5] btrfs: push the NODATASUM check into btrfs_lookup_bio_sums
Quoted 35 lines · Qu Wenruo
Quoted 5 lines · Josef Bacik
When we move to being able to handle NULL
csum_rootsit'll be cleaner to just check inbtrfs_lookup_bio_sumsinstead of at all of the caller locations, so push the NODATASUM check into it as well so it's unified.Signed-off-by Josef Bacik <[email protected]>Reviewed-by Qu Wenruo <[email protected]>But an off-topic question inlined below:
Quoted 11 lines
--- fs/btrfs/compression.c | 14 +++++--------- fs/btrfs/file-item.c | 3 +++ fs/btrfs/inode.c | 12 +++++++++--- 3 files changed, 17 insertions(+), 12 deletions(-)fs/btrfs/compression.c +0
-0fs/btrfs/compression.c blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, · line 728
Quoted 7 lines
fs/btrfs/compression.c +0
-4@@ -722,11 +722,9 @@ blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,*/ refcount_inc (&cb->pending_bios); -if (!(BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM )) { - ret =btrfs_lookup_bio_sums (inode, comp_bio, - (u64)-1 , sums); -BUG_ON (ret);/* -ENOMEM */ Is it really possible to have compressed extent without data csum? Won't nodatacsum disable compression?
Quoted 2 lines
Or are we just here to handle some old compressed but not csumed data?
We used to allow it, so I'm content to leave this here. Maybe at some point we'll allow it in the future, but IDK it doesn't hurt anything to handle it here. Thanks,
Josef
Re: [PATCH 2/5] btrfs: push the NODATASUM check into btrfs_lookup_bio_sums
In reply to Josef BacikLooks good,
PATCH 3/5 btrfs: introduce rescue=ignorebadroots
In reply to Josef BacikIn the face of extent root corruption, or any other core fs wide root corruption we will fail to mount the file system. This makes recovery kind of a pain, because you need to fall back to userspace tools to scrape off data. Instead provide a mechanism to gracefully handle bad roots, so we can at least mount read-only and possibly recover data from the file system.
--- fs/btrfs/block-group.c | 48 +++++++++++++++++++++++++++++++ fs/btrfs/block-rsv.c | 8 ++++++ fs/btrfs/compression.c | 3 +- fs/btrfs/ctree.h | 1 + fs/btrfs/disk-io.c | 65 ++++++++++++++++++++++++++---------------- fs/btrfs/file-item.c | 3 +- fs/btrfs/inode.c | 6 +++- fs/btrfs/super.c | 11 ++++++- fs/btrfs/volumes.c | 7 +++++ 9 files changed, 124 insertions(+), 28 deletions(-)
fs/btrfs/block-group.c +48 -0
@@ -1985,6 +1985,51 @@ static int read_one_block_group(struct btrfs_fs_info *info,return ret; } +static int fill_dummy_bgs (struct btrfs_fs_info *fs_info) +{ +struct extent_map_tree *em_tree = &fs_info->mapping_tree; +struct extent_map *em; +struct map_lookup *map; +struct btrfs_block_group *bg; +struct btrfs_space_info *space_info; +struct rb_node *node; +int ret =0 ; + +for (node =rb_first_cached (&em_tree->map); node; + node =rb_next (node)) { + em = rb_entry(node, struct extent_map, rb_node); + map = em->map_lookup; + bg =btrfs_create_block_group_cache (fs_info, em->start); +if (!bg) { + ret = -ENOMEM ; +break ; + } + +/* Fill dummy cache as FULL */ + bg->length = em->len; + bg->flags = map->type; + bg->last_byte_to_unpin = (u64)-1 ; + bg->cached =BTRFS_CACHE_FINISHED ; + bg->used = em->len; + bg->flags = map->type; + ret =btrfs_add_block_group_cache (fs_info, bg); +if (ret) { +btrfs_remove_free_space_cache (bg); +btrfs_put_block_group (bg); +break ; + } +btrfs_update_space_info (fs_info, bg->flags, em->len, em->len, +0 , &space_info); + bg->space_info = space_info; +link_block_group (bg); + +set_avail_alloc_bits (fs_info, bg->flags); + } +if (!ret) +btrfs_init_global_block_rsv (fs_info); +return ret; +} +int btrfs_read_block_groups (struct btrfs_fs_info *info) {struct btrfs_path *path;
@@ -1995,6 +2040,9 @@ int btrfs_read_block_groups(struct btrfs_fs_info *info)int need_clear =0 ;u64 cache_gen; +if (!info->extent_root) +return fill_dummy_bgs (info); + key.objectid =0 ; key.offset =0 ; key.type =BTRFS_BLOCK_GROUP_ITEM_KEY ;
fs/btrfs/block-rsv.c +8 -0
@@ -426,6 +426,14 @@ void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info) fs_info->delayed_block_rsv.space_info = space_info; fs_info->delayed_refs_rsv.space_info = space_info; +/* +* Our various recovery options can leave us with NULL roots, so check +* here and just bail before we go deref'ing NULLS everywhere. +*/ +if (!fs_info->extent_root || !fs_info->csum_root || + !fs_info->dev_root || !fs_info->chunk_root || !fs_info->tree_root) +return ; + fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv; fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv; fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
fs/btrfs/compression.c +2 -1
@@ -150,7 +150,8 @@ static int check_compressed_csum(struct btrfs_inode *inode, struct bio *bio,struct compressed_bio *cb = bio->bi_private;u8 *cb_sum = cb->sums; -if (inode->flags &BTRFS_INODE_NODATASUM ) +if (!fs_info->csum_root || + inode->flags &BTRFS_INODE_NODATASUM )return 0 ; shash->tfm = fs_info->csum_shash;
fs/btrfs/ctree.h +1 -0
@@ -1295,6 +1295,7 @@ static inline u32 BTRFS_MAX_XATTR_SIZE(const struct btrfs_fs_info *info)#define BTRFS_MOUNT_NOLOGREPLAY (1 << 27)#define BTRFS_MOUNT_REF_VERIFY (1 << 28)#define BTRFS_MOUNT_DISCARD_ASYNC (1 << 29) +#define BTRFS_MOUNT_IGNOREBADROOTS (1 << 30)#define BTRFS_DEFAULT_COMMIT_INTERVAL (30)#define BTRFS_DEFAULT_MAX_INLINE (2048)
fs/btrfs/disk-io.c +41 -24
102 lines not on this page · raw message
fs/btrfs/file-item.c +2 -1
@@ -272,7 +272,8 @@ blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,int count =0 ;u16 csum_size =btrfs_super_csum_size (fs_info->super_copy); -if (BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM ) +if (!fs_info->csum_root || +BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM )return BLK_STS_OK ; path =btrfs_alloc_path ();
fs/btrfs/inode.c +5 -1
@@ -2187,7 +2187,8 @@ blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,int skip_sum;int async = !atomic_read (&BTRFS_I (inode)->sync_writers); - skip_sum =BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM ; + skip_sum = (BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM ) || + !fs_info->csum_root;if (btrfs_is_free_space_inode (BTRFS_I (inode))) metadata =BTRFS_WQ_ENDIO_FREE_SPACE ;
@@ -2844,6 +2845,9 @@ int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u64 phy_offset,if (BTRFS_I (inode)->flags &BTRFS_INODE_NODATASUM )return 0 ; +if (!root->fs_info->csum_root) +return 0 ; +if (root->root_key.objectid ==BTRFS_DATA_RELOC_TREE_OBJECTID &&test_range_bit (io_tree, start, end,EXTENT_NODATASUM ,1 ,NULL )) {clear_extent_bits (io_tree, start, end,EXTENT_NODATASUM );
fs/btrfs/super.c +10 -1
@@ -360,6 +360,7 @@ enum { Opt_rescue, Opt_usebackuproot, Opt_nologreplay, + Opt_ignorebadroots,/* Deprecated options */ Opt_recovery,
@@ -455,6 +456,7 @@ static const match_table_t tokens = {static const match_table_t rescue_tokens = { {Opt_usebackuproot,"usebackuproot" }, {Opt_nologreplay,"nologreplay" }, + {Opt_ignorebadroots,"ignorebadroots" }, {Opt_err,NULL }, };
@@ -498,6 +500,10 @@ static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)btrfs_set_and_info (info,NOLOGREPLAY ,"disabling log replay at mount time" );break ; +case Opt_ignorebadroots: +btrfs_set_and_info (info,IGNOREBADROOTS , +"ignoring bad roots" ); +break ;case Opt_err:btrfs_info (info,"unrecognized rescue option '%s'" , p); ret = -EINVAL ;
@@ -983,7 +989,8 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options,if (new_flags &SB_RDONLY ) goto out; -if (check_ro_option (info,BTRFS_MOUNT_NOLOGREPLAY ,"nologreplay" )) +if (check_ro_option (info,BTRFS_MOUNT_NOLOGREPLAY ,"nologreplay" ) || +check_ro_option (info,BTRFS_MOUNT_IGNOREBADROOTS ,"ignorebadroots" )) ret = -EINVAL ; out: if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
@@ -1430,6 +1437,8 @@ static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)seq_puts (seq,",notreelog" );if (btrfs_test_opt (info,NOLOGREPLAY ))seq_puts (seq,",rescue=nologreplay" ); +if (btrfs_test_opt (info,IGNOREBADROOTS )) +seq_puts (seq,",rescue=ignorebadroots" );if (btrfs_test_opt (info,FLUSHONCOMMIT ))seq_puts (seq,",flushoncommit" );if (btrfs_test_opt (info,DISCARD_SYNC ))
fs/btrfs/volumes.c +7 -0
@@ -7656,6 +7656,13 @@ int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)u64 prev_dev_ext_end =0 ;int ret =0 ; +/* +* We don't have a dev_root because we mounted with ignorebadroots and +* failed to load the root, so skip the verification. +*/ +if (!root) +return 0 ; + key.objectid =1 ; key.type =BTRFS_DEV_EXTENT_KEY ; key.offset =0 ;
Signature · 1 line
2.26.2
Re: [PATCH 3/5] btrfs: introduce rescue=ignorebadroots
Quoted 8 lines · Josef Bacik
In the face of extent root corruption, or any other core fs wide root corruption we will fail to mount the file system. This makes recovery kind of a pain, because you need to fall back to userspace tools to scrape off data. Instead provide a mechanism to gracefully handle bad roots, so we can at least mount read-only and possibly recover data from the file system.
Signed-off-by Josef Bacik <[email protected]>
Mostly OK, but still a small problem inlined below. [...]
Quoted 4 lines
index
46f4efd58652..08b3ca60f3df100644
fs/btrfs/volumes.c int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) · line 7665
Quoted 10 lines
fs/btrfs/volumes.c +7
-0@@ -7656,6 +7656,13 @@ int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)u64 prev_dev_ext_end =0 ;int ret =0 ; +/* +* We don't have a dev_root because we mounted with ignorebadroots and +* failed to load the root, so skip the verification. +*/ +if (!root) +return 0 ; +
The check itself is mostly for write, to ensure we won't have missing/unnecessary dev extents to mess up chunk allocation.
For RO operations, the check makes little sense, and can be safely ignored for ignorebadroots.
Furthermore this only handles the case where the device tree root is corrupted.
But if only part of the device tree is corrupted, we still continue checking and fail to mount.
It's better to skip the whole check for dev extents if we're using ignorebadroots rescue option.
No matter if the root is corrupted or not.
Thanks,
Qu
Quoted 4 lines
fs/btrfs/volumes.c +0
-0key.objectid =1 ; key.type =BTRFS_DEV_EXTENT_KEY ; key.offset =0 ;
Re: [PATCH 3/5] btrfs: introduce rescue=ignorebadroots
Quoted 35 lines · Qu Wenruo
Quoted 8 lines · Josef Bacik
In the face of extent root corruption, or any other core fs wide root corruption we will fail to mount the file system. This makes recovery kind of a pain, because you need to fall back to userspace tools to scrape off data. Instead provide a mechanism to gracefully handle bad roots, so we can at least mount read-only and possibly recover data from the file system.
Signed-off-by Josef Bacik <[email protected]>Mostly OK, but still a small problem inlined below. [...]
Quoted 4 lines
index
46f4efd58652..08b3ca60f3df100644fs/btrfs/volumes.c int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) · line 7665
Quoted 10 lines
fs/btrfs/volumes.c +7
-0@@ -7656,6 +7656,13 @@ int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)u64 prev_dev_ext_end =0 ;int ret =0 ; +/* +* We don't have a dev_root because we mounted with ignorebadroots and +* failed to load the root, so skip the verification. +*/ +if (!root) +return 0 ; +The check itself is mostly for write, to ensure we won't have missing/unnecessary dev extents to mess up chunk allocation.
For RO operations, the check makes little sense, and can be safely ignored for ignorebadroots.
Quoted 9 lines
Furthermore this only handles the case where the device tree root is corrupted.
But if only part of the device tree is corrupted, we still continue checking and fail to mount.It's better to skip the whole check for dev extents if we're using ignorebadroots rescue option.
No matter if the root is corrupted or not.
Yeah good point, I'll fix this up. Thanks,
Josef
PATCH 4/5 btrfs: introduce rescue=ignoredatacsums
In reply to Josef BacikThere are cases where you can end up with bad data csums because of misbehaving applications. This happens when an application modifies a buffer in-flight when doing an O_DIRECT write. In order to recover the file we need a way to turn off data checksums so you can copy the file off, and then you can delete the file and restore it properly later.
--- fs/btrfs/ctree.h | 1 + fs/btrfs/disk-io.c | 21 ++++++++++++--------- fs/btrfs/super.c | 11 ++++++++++- 3 files changed, 23 insertions(+), 10 deletions(-)
fs/btrfs/ctree.h +1 -0
@@ -1296,6 +1296,7 @@ static inline u32 BTRFS_MAX_XATTR_SIZE(const struct btrfs_fs_info *info)#define BTRFS_MOUNT_REF_VERIFY (1 << 28)#define BTRFS_MOUNT_DISCARD_ASYNC (1 << 29)#define BTRFS_MOUNT_IGNOREBADROOTS (1 << 30) +#define BTRFS_MOUNT_IGNOREDATACSUMS (1 << 31)#define BTRFS_DEFAULT_COMMIT_INTERVAL (30)#define BTRFS_DEFAULT_MAX_INLINE (2048)
fs/btrfs/disk-io.c +12 -9
@@ -2269,16 +2269,19 @@ static int btrfs_read_roots(struct btrfs_fs_info *fs_info)btrfs_init_devices_late (fs_info); } - location.objectid =BTRFS_CSUM_TREE_OBJECTID ; - root =btrfs_read_tree_root (tree_root, &location); -if (IS_ERR (root)) { -if (!btrfs_test_opt (fs_info,IGNOREBADROOTS )) { - ret =PTR_ERR (root); - goto out; +/* If IGNOREDATASCUMS is set don't bother reading the csum root. */ +if (!btrfs_test_opt (fs_info,IGNOREDATACSUMS )) { + location.objectid =BTRFS_CSUM_TREE_OBJECTID ; + root =btrfs_read_tree_root (tree_root, &location); +if (IS_ERR (root)) { +if (!btrfs_test_opt (fs_info,IGNOREBADROOTS )) { + ret =PTR_ERR (root); + goto out; + } + }else { +set_bit (BTRFS_ROOT_TRACK_DIRTY , &root->state); + fs_info->csum_root = root; } - }else { -set_bit (BTRFS_ROOT_TRACK_DIRTY , &root->state); - fs_info->csum_root = root; }/*
fs/btrfs/super.c +10 -1
@@ -361,6 +361,7 @@ enum { Opt_usebackuproot, Opt_nologreplay, Opt_ignorebadroots, + Opt_ignoredatacsums,/* Deprecated options */ Opt_recovery,
@@ -457,6 +458,7 @@ static const match_table_t rescue_tokens = { {Opt_usebackuproot, "usebackuproot"}, {Opt_nologreplay, "nologreplay"}, {Opt_ignorebadroots, "ignorebadroots"}, + {Opt_ignoredatacsums, "ignoredatacsums"}, {Opt_err, NULL}, };
@@ -504,6 +506,10 @@ static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)btrfs_set_and_info (info,IGNOREBADROOTS ,"ignoring bad roots" );break ; +case Opt_ignoredatacsums: +btrfs_set_and_info (info,IGNOREDATACSUMS , +"ignoring data csums" ); +break ;case Opt_err:btrfs_info (info,"unrecognized rescue option '%s'" , p); ret = -EINVAL ;
@@ -990,7 +996,10 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options, goto out;if (check_ro_option (info,BTRFS_MOUNT_NOLOGREPLAY ,"nologreplay" ) || -check_ro_option (info,BTRFS_MOUNT_IGNOREBADROOTS ,"ignorebadroots" )) +check_ro_option (info,BTRFS_MOUNT_IGNOREBADROOTS , +"ignorebadroots" ) || +check_ro_option (info,BTRFS_MOUNT_IGNOREDATACSUMS , +"ignoredatacsums" )) ret = -EINVAL ; out: if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
Signature · 1 line
2.26.2
Re: [PATCH 4/5] btrfs: introduce rescue=ignoredatacsums
Quoted 31 lines · Josef Bacik
There are cases where you can end up with bad data csums because of misbehaving applications. This happens when an application modifies a buffer in-flight when doing an
O_DIRECTwrite. In order to recover the file we need a way to turn off data checksums so you can copy the file off, and then you can delete the file and restore it properly later.Signed-off-by Josef Bacik <[email protected]>--- fs/btrfs/ctree.h | 1 + fs/btrfs/disk-io.c | 21 ++++++++++++--------- fs/btrfs/super.c | 11 ++++++++++- 3 files changed, 23 insertions(+), 10 deletions(-)fs/btrfs/ctree.h +1
-0@@ -1296,6 +1296,7 @@ static inline u32 BTRFS_MAX_XATTR_SIZE(const struct btrfs_fs_info *info)#define BTRFS_MOUNT_REF_VERIFY (1 << 28)#define BTRFS_MOUNT_DISCARD_ASYNC (1 << 29)#define BTRFS_MOUNT_IGNOREBADROOTS (1 << 30) +#define BTRFS_MOUNT_IGNOREDATACSUMS (1 << 31)#define BTRFS_DEFAULT_COMMIT_INTERVAL (30)#define BTRFS_DEFAULT_MAX_INLINE (2048)fs/btrfs/disk-io.c +0
-0@@ -2269,16 +2269,19 @@ static int btrfs_read_roots(struct btrfs_fs_info *fs_info)btrfs_init_devices_late (fs_info);
fs/btrfs/disk-io.c static int btrfs_read_roots(struct btrfs_fs_info *fs_info) · line 2273
Quoted 10 lines
fs/btrfs/disk-io.c +2
-6} - location.objectid =BTRFS_CSUM_TREE_OBJECTID ; - root =btrfs_read_tree_root (tree_root, &location); -if (IS_ERR (root)) { -if (!btrfs_test_opt (fs_info,IGNOREBADROOTS )) { - ret =PTR_ERR (root); - goto out; +/* If IGNOREDATASCUMS is set don't bother reading the csum root. */ +if (!btrfs_test_opt (fs_info,IGNOREDATACSUMS )) {
This indeed matches the name, ignoredatacsums, no matter if the data csum matches or not.
I guess if the user is using this option, they really don't bother the datacsum and just want to grab whatever they can get.
Thanks,
Qu
Quoted 61 lines
fs/btrfs/disk-io.c +10
-3+ location.objectid =BTRFS_CSUM_TREE_OBJECTID ; + root =btrfs_read_tree_root (tree_root, &location); +if (IS_ERR (root)) { +if (!btrfs_test_opt (fs_info,IGNOREBADROOTS )) { + ret =PTR_ERR (root); + goto out; + } + }else { +set_bit (BTRFS_ROOT_TRACK_DIRTY , &root->state); + fs_info->csum_root = root; } - }else { -set_bit (BTRFS_ROOT_TRACK_DIRTY , &root->state); - fs_info->csum_root = root; }/* fs/btrfs/super.c +10
-1@@ -361,6 +361,7 @@ enum { Opt_usebackuproot, Opt_nologreplay, Opt_ignorebadroots, + Opt_ignoredatacsums,/* Deprecated options */ Opt_recovery,@@ -457,6 +458,7 @@ static const match_table_t rescue_tokens = { {Opt_usebackuproot, "usebackuproot"}, {Opt_nologreplay, "nologreplay"}, {Opt_ignorebadroots, "ignorebadroots"}, + {Opt_ignoredatacsums, "ignoredatacsums"}, {Opt_err, NULL}, };@@ -504,6 +506,10 @@ static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)btrfs_set_and_info (info,IGNOREBADROOTS ,"ignoring bad roots" );break ; +case Opt_ignoredatacsums: +btrfs_set_and_info (info,IGNOREDATACSUMS , +"ignoring data csums" ); +break ;case Opt_err:btrfs_info (info,"unrecognized rescue option '%s'" , p); ret = -EINVAL ;@@ -990,7 +996,10 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options, goto out;if (check_ro_option (info,BTRFS_MOUNT_NOLOGREPLAY ,"nologreplay" ) || -check_ro_option (info,BTRFS_MOUNT_IGNOREBADROOTS ,"ignorebadroots" )) +check_ro_option (info,BTRFS_MOUNT_IGNOREBADROOTS , +"ignorebadroots" ) || +check_ro_option (info,BTRFS_MOUNT_IGNOREDATACSUMS , +"ignoredatacsums" )) ret = -EINVAL ; out: if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
PATCH 5/5 btrfs: introduce rescue=all
In reply to Josef BacikNow that we have the building blocks for some better recovery options with corrupted file systems, add a rescue=all option to enable all of the relevant rescue options. This will allow distro's to simply default to rescue=all for the "oh dear lord the world's on fire" recovery without needing to know all the different options that we have and may add in the future.
--- fs/btrfs/super.c | 8 ++++++++ 1 file changed, 8 insertions(+)
fs/btrfs/super.c +8 -0
@@ -362,6 +362,7 @@ enum { Opt_nologreplay, Opt_ignorebadroots, Opt_ignoredatacsums, + Opt_all,/* Deprecated options */ Opt_recovery,
@@ -459,6 +460,7 @@ static const match_table_t rescue_tokens = { {Opt_nologreplay, "nologreplay"}, {Opt_ignorebadroots, "ignorebadroots"}, {Opt_ignoredatacsums, "ignoredatacsums"}, + {Opt_all, "all"}, {Opt_err, NULL}, };
@@ -510,6 +512,12 @@ static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)btrfs_set_and_info (info,IGNOREDATACSUMS ,"ignoring data csums" );break ; +case Opt_all: +btrfs_set_opt (info->mount_opt,IGNOREDATACSUMS ); +btrfs_set_opt (info->mount_opt,IGNOREBADROOTS ); +btrfs_set_opt (info->mount_opt,NOLOGREPLAY ); +btrfs_info (info,"enabling all of the rescue options" ); +break ;case Opt_err:btrfs_info (info,"unrecognized rescue option '%s'" , p); ret = -EINVAL ;
Signature · 1 line
2.26.2
Re: [PATCH 5/5] btrfs: introduce rescue=all
Quoted 8 lines · Josef Bacik
Now that we have the building blocks for some better recovery options with corrupted file systems, add a rescue=all option to enable all of the relevant rescue options. This will allow distro's to simply default to rescue=all for the "oh dear lord the world's on fire" recovery without needing to know all the different options that we have and may add in the future.
Signed-off-by Josef Bacik <[email protected]>
Thanks,
Qu
Quoted 38 lines
--- fs/btrfs/super.c | 8 ++++++++ 1 file changed, 8 insertions(+)fs/btrfs/super.c +8
-0@@ -362,6 +362,7 @@ enum { Opt_nologreplay, Opt_ignorebadroots, Opt_ignoredatacsums, + Opt_all,/* Deprecated options */ Opt_recovery,@@ -459,6 +460,7 @@ static const match_table_t rescue_tokens = { {Opt_nologreplay, "nologreplay"}, {Opt_ignorebadroots, "ignorebadroots"}, {Opt_ignoredatacsums, "ignoredatacsums"}, + {Opt_all, "all"}, {Opt_err, NULL}, };@@ -510,6 +512,12 @@ static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)btrfs_set_and_info (info,IGNOREDATACSUMS ,"ignoring data csums" );break ; +case Opt_all: +btrfs_set_opt (info->mount_opt,IGNOREDATACSUMS ); +btrfs_set_opt (info->mount_opt,IGNOREBADROOTS ); +btrfs_set_opt (info->mount_opt,NOLOGREPLAY ); +btrfs_info (info,"enabling all of the rescue options" ); +break ;case Opt_err:btrfs_info (info,"unrecognized rescue option '%s'" , p); ret = -EINVAL ;
Quoted 14 lines · Josef Bacik
Hello,
This is the next version of my rescue=all patches, this time broken up in to a few discrete patches, with some cleanups as well. I have a PR for the xfstest that exercises these options, it can be found here
https://github.com/btrfs/fstests/pull/35
This is the same idea as the previous versions, except I've made a mechanical change. Instead we have rescue=ignorebadroots, which ignores any global roots that we may not be able to read. We will still fail to mount if thinks like the chunk root or the tree root are corrupt, but this will allow us to mount read only if the extent root or csum root are completely hosed.
Quoted 7 lines
I've added a new patch this go around, rescue=ignoredatacsums. Somebody had indicated that they would prefer that the original rescue=all allowed us to continue to read csums if the root was still intact. In order to handle that usecase that's what you get with rescue=ignorebadroots, however if your csum tree is corrupt in the middle of the tree you could still end up with problems. Thus we have rescue=ignoredatacsums which will completely disable the csum tree as well.
The extra hard ignore for data csum is really good.
Quoted 7 lines
And finally we have rescue=all, which simply enables ignoredatacsums, ignorebadroots, and notreelogreplay. We need an easy catch-all option for distros to fallback on to get users the highest probability of being able to recover their data, so we will use rescue=all to turn on all the fanciest rescue options that we have, and then use the discrete options for more fine grained recovery. Thanks,
Now rescue=all makes more sense. It's just an alias for one to salvage as much data as possible.
Thanks,
Qu
Quoted 21 lines
Josef
Josef Bacik (5):
btrfs: unify the ro checking for mount options
btrfs: push the NODATASUM check intobtrfs_lookup_bio_sums
btrfs: introduce rescue=ignorebadroots
btrfs: introduce rescue=ignoredatacsums
btrfs: introduce rescue=allfs/btrfs/block-group.c | 48 +++++++++++++++++++++++++++ fs/btrfs/block-rsv.c | 8 +++++ fs/btrfs/compression.c | 17 ++++------ fs/btrfs/ctree.h | 2 ++ fs/btrfs/disk-io.c | 74 +++++++++++++++++++++++++++--------------- fs/btrfs/file-item.c | 4 +++ fs/btrfs/inode.c | 18 +++++++--- fs/btrfs/super.c | 49 ++++++++++++++++++++++++---- fs/btrfs/volumes.c | 7 ++++ 9 files changed, 179 insertions(+), 48 deletions(-)