Skip to contentSkip to conversation

btrfs: simple quotas

PATCH Superseded Boris Burkov 3 participants 53 messages Jul 27, 2023 – Sep 13, 2023 Btrfs file system

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v5 cover btrfs: simple quotas

btrfs quota groups (qgroups) are a compelling feature of btrfs that allow flexible control for limiting subvolume data and metadata usage. However, due to btrfs's high level decision to tradeoff snapshot performance against ref-counting performance, qgroups suffer from non-trivial performance issues that make them unattractive in certain workloads. Particularly, frequent backref walking during writes and during commits can make operations increasingly expensive as the number of snapshots scales up. For that reason, we have never been able to commit to using qgroups in production at Meta, despite significant interest from people running container workloads, where we would benefit from protecting the rest of the host from a buggy application in a container running away with disk usage. This patch series introduces a simplified version of qgroups called
simple quotas (squotas) which never computes global reference counts for extents, and thus has similar performance characteristics to normal, quotas disabled, btrfs. The "trick" is that in simple quotas mode, we account all extents permanently to the subvolume in which they were originally created. That allows us to make all accounting 1:1 with extent item lifetime, removing the need to walk backrefs. However, this sacrifices the ability to compute shared vs. exclusive usage. It also results in counter-intuitive, though still predictable and simple accounting in the cases where an original extent is removed while a shared copy still exists. Qgroups is able to detect that case and count the remaining copy as an exclusive owner, while squotas is not. As a result, squotas works best when the original extent is immutable and outlives any clones.

==Format Change==
In order to track the original creating subvolume of a data extent in the face of reflinks, it is necessary to add additional accounting to the extent item. To save space, this is done with a new inline ref item. However, the downside of this approach is that it makes enabling squota an incompat change, denoted by the new incompat bit SIMPLE_QUOTA. When this bit is set and quotas are enabled, new extent items get the extra accounting, and freed extent items check for the accounting to find their creating subvolume. In addition, 1:1 with this incompat bit, the quota status item now tracks a "quota enablement generation" needed for properly handling deleting extents with predate enablement.

==API==
Squotas reuses the api of qgroups. The only difference is that when you enable quotas via btrfs quota enable, you pass the --simple flag. Squotas will always report exclusive == shared for each qgroup. Squotas deal with extent_item/metadata_item sizes and thus do not do anything special with compression. Squotas also introduce auto inheritance for nested subvols. The API is documented more fully in the documentation patches in btrfs-progs.

==Testing methodology==
Using updated btrfs-progs and fstests (relevant matching patch sets to be sent ASAP)
btrfs-progs: https://github.com/boryas/btrfs-progs/tree/squota-progs
fstests: https://github.com/boryas/fstests/tree/squota-test

I ran '-g auto' on fstests on the following configurations:
1a) baseline kernel/progs/fstests.
1b) squota kernel baseline progs/fstests.
2a) baseline kernel/progs/fstests. fstests configured to mkfs with quota 2b) squota kernel/progs/fstests. fstests configured to mkfs with squota

I compared 1a against 1b and 2a against 2b and detected no regressions. 2a/2b both exhibit regressions against 1a/1b which are largely issues with quota reservations in various complicated cases. I intend to run those down in the future, but they are not simple quota specific, as they are already broken with plain qgroups.

==Performance Testing==
I measured the performance of the change using fsperf. I ran with 3 configurations using the squota kernel:
- plain mkfs
- qgroup mkfs
- squota mkfs
And added a new performance test which creates 1000 files in a subvol, creates 100 snapshots of that subvol, then unshares extents in files in the snapshots. I measured write performance with fio and btrfs commit critical section performance side effects with bpftrace on 'wait_current_trans'.

The results for the test which measures unshare perf (unshare.py) with qgroup and squota compared to the baseline:

group test results
unshare results
metric baseline current stdev diff
========================================================================================
avg_commit_ms 162.13 285.75 3.14 76.24%
bg_count 16 16 0 0.00%

commits                           378.20           379          1.92      0.21%
elapsed                           201.40        270.40          1.34     34.26%

end_state_mount_ns 26036211.60 26004593.60 2281065.40 -0.12%
end_state_umount_ns 2.45e+09 2.55e+09 20740154.41 3.93%
max_commit_ms 425.80 594 53.34 39.50%
sys_cpu 0.10 0.06 0.06 -42.15%
wait_current_trans_calls 2945.60 3405.20 47.08 15.60%
wait_current_trans_ns_max 1.56e+08 3.43e+08 32659393.25 120.07%
wait_current_trans_ns_mean 1974875.35 28588482.55 1557588.84 1347.61%

wait_current_trans_ns_min            232           232         25.88      0.00%
wait_current_trans_ns_p50            718           740         22.80      3.06%
wait_current_trans_ns_p95     7711770.20      2.21e+08   17241032.09   2761.19%
wait_current_trans_ns_p99    67744932.29      2.68e+08   41275815.87    295.16%
write_bw_bytes                 653008.80     486344.40       4209.91    -25.52%
write_clat_ns_mean            6251404.78    8406837.89      39779.15     34.48%
write_clat_ns_p50             1656422.40    1643315.20      27415.68     -0.79%
write_clat_ns_p99               1.90e+08      3.20e+08       2097152     68.62%
write_io_kbytes                   128000        128000             0      0.00%
write_iops                        159.43        118.74          1.03    -25.52%
write_lat_ns_max                7.06e+08      9.80e+08   47324816.61     38.88%
write_lat_ns_mean             6251503.06    8406936.06      39780.83     34.48%
write_lat_ns_min                    3354          4648        616.06     38.58%

squota test results
unshare results
metric baseline current stdev diff
========================================================================================
avg_commit_ms 162.13 164.16 3.14 1.25%
bg_count 16 0 0 -100.00%

commits                           378.20        380.80          1.92      0.69%
elapsed                           201.40        208.20          1.34      3.38%
end_state_mount_ns           26036211.60   25840729.60    2281065.40     -0.75%
end_state_umount_ns             2.45e+09      3.01e+09   20740154.41     22.80%
max_commit_ms                     425.80        415.80         53.34     -2.35%

sys_cpu 0.10 0.08 0.06 -23.36%
wait_current_trans_calls 2945.60 2981.60 47.08 1.22%
wait_current_trans_ns_max 1.56e+08 1.12e+08 32659393.25 -27.86%
wait_current_trans_ns_mean 1974875.35 1064734.76 1557588.84 -46.09%

wait_current_trans_ns_min            232           238         25.88      2.59%
wait_current_trans_ns_p50            718           746         22.80      3.90%
wait_current_trans_ns_p95     7711770.20       1567.60   17241032.09    -99.98%
wait_current_trans_ns_p99    67744932.29   49880514.27   41275815.87    -26.37%
write_bw_bytes                 653008.80        631256       4209.91     -3.33%
write_clat_ns_mean            6251404.78    6476816.06      39779.15      3.61%
write_clat_ns_p50             1656422.40       1581056      27415.68     -4.55%
write_clat_ns_p99               1.90e+08      1.94e+08       2097152      2.21%
write_io_kbytes                   128000        128000             0      0.00%
write_iops                        159.43        154.12          1.03     -3.33%

write_lat_ns_max 7.06e+08 7.65e+08 47324816.61 8.38%
write_lat_ns_mean 6251503.06 6476912.76 39780.83 3.61%
write_lat_ns_min 3354 4062 616.06 21.11%

And the same, but only showing results where the deviation was outside of a 95% confidence interval for the mean (default significance highlighting in fsperf):
qgroup test results
unshare results
metric baseline current stdev diff
========================================================================================

avg_commit_ms                     162.13        285.75          3.14     76.24%
elapsed                           201.40        270.40          1.34     34.26%
end_state_umount_ns             2.45e+09      2.55e+09   20740154.41      3.93%
max_commit_ms                     425.80           594         53.34     39.50%
wait_current_trans_calls         2945.60       3405.20         47.08     15.60%
wait_current_trans_ns_max       1.56e+08      3.43e+08   32659393.25    120.07%
wait_current_trans_ns_mean    1974875.35   28588482.55    1557588.84   1347.61%
wait_current_trans_ns_p95     7711770.20      2.21e+08   17241032.09   2761.19%

wait_current_trans_ns_p99 67744932.29 2.68e+08 41275815.87 295.16%
write_bw_bytes 653008.80 486344.40 4209.91 -25.52%
write_clat_ns_mean 6251404.78 8406837.89 39779.15 34.48%
write_clat_ns_p99 1.90e+08 3.20e+08 2097152 68.62%
write_iops 159.43 118.74 1.03 -25.52%

write_lat_ns_max                7.06e+08      9.80e+08   47324816.61     38.88%
write_lat_ns_mean             6251503.06    8406936.06      39780.83     34.48%
write_lat_ns_min                    3354          4648        616.06     38.58%

squota test results
unshare results
metric baseline current stdev diff
========================================================================================
elapsed 201.40 208.20 1.34 3.38%
end_state_umount_ns 2.45e+09 3.01e+09 20740154.41 22.80%
write_bw_bytes 653008.80 631256 4209.91 -3.33%

write_clat_ns_mean            6251404.78    6476816.06      39779.15      3.61%
write_clat_ns_p50             1656422.40       1581056      27415.68     -4.55%

write_clat_ns_p99 1.90e+08 1.94e+08 2097152 2.21%
write_iops 159.43 154.12 1.03 -3.33%
write_lat_ns_mean 6251503.06 6476912.76 39780.83 3.61%

Particularly noteworthy are the massive regressions to wait_current_trans in qgroup mode as well as the solid regressions to bandwidth, iops and write latency. The regressions/improvements in squotas are modest in comparison in line with the expectation. I am still investigating the squota umount regression, particularly whether it is in the umount's final commit and represents a real performance problem with squotas.

---
Changelog:
v5:
* fix btrfs/187 failure in squota mode: relocation+dedupe led to drop refs with the wrong owning root coming first, followed by drop refs with the real owning root. The "bad" ones are never last, so fix it by letting the "good" ones set it on the head ref.
v4:
* drop unrelated patches folded into misc-next
* fix crash where check_committed_extent was reading the inline ref type on an extent item with no inline extents. (btrfs/192 *without* squotas enabled)
v3:
* u64 -> __le64 in new owner_ref_item (as caught by kernel test bot) v2:
* fix dumb formatting errors, unexpected/unrelated edits
* use command instead of status in ioctl
* fix the illegal GFP_KERNEL in delta fn (punted on pulling allocations out from the spin lock and using GFP_ATOMIC like other qgroups use cases for now. Plan to fix that in either v3 or a follow up series, as there are other places this is an issue for qgroups/squotas)
* improve boolean logic in head_ref init
* use list_count helper function instead of rolling my own bad one
* fixed the adjacent extents reloc cluster bug Josef noticed
* fixed a qgroups bug I introduced: it needs to be able to account extents while shutting down to not hit a warning in commit_transaction
* added a qgroup_status flag for simple quotas to not rely on the incompat bit directly. This allows disabling simple quotas and enabling qgroups.

Boris Burkov (18):
btrfs: introduce quota mode
btrfs: add new quota mode for simple quotas
btrfs: expose quota mode via sysfs
btrfs: add simple_quota incompat feature to sysfs
btrfs: flush reservations during quota disable
btrfs: create qgroup earlier in snapshot creation
btrfs: function for recording simple quota deltas
btrfs: rename tree_ref and data_ref owning_root
btrfs: track owning root in btrfs_ref
btrfs: track original extent owner in head_ref
btrfs: new inline ref storing owning subvol of data extents
btrfs: inline owner ref lookup helper
btrfs: record simple quota deltas
btrfs: simple quota auto hierarchy for nested subvols
btrfs: check generation when recording simple quota delta
btrfs: track metadata relocation cow with simple quota
btrfs: track data relocation with simple quota
btrfs: only set QUOTA_ENABLED when done reading qgroups

 fs/btrfs/accessors.h            |   6 +
 fs/btrfs/backref.c              |   3 +
 fs/btrfs/ctree.c                |  22 ++-
 fs/btrfs/ctree.h                |   1 +
 fs/btrfs/delayed-ref.c          |  35 ++--
 fs/btrfs/delayed-ref.h          |  32 +++-
 fs/btrfs/disk-io.c              |   5 +-
 fs/btrfs/extent-tree.c          | 242 ++++++++++++++++++++++-----
 fs/btrfs/extent-tree.h          |   6 +-
 fs/btrfs/file.c                 |  10 +-
 fs/btrfs/fs.h                   |   7 +-
 fs/btrfs/inode-item.c           |   2 +-
 fs/btrfs/ioctl.c                |   7 +-
 fs/btrfs/print-tree.c           |  12 ++
 fs/btrfs/qgroup.c               | 286 +++++++++++++++++++++++++++-----
 fs/btrfs/qgroup.h               |  28 +++-
 fs/btrfs/ref-verify.c           |   7 +-
 fs/btrfs/relocation.c           |  66 +++++++-
 fs/btrfs/root-tree.c            |   2 +-
 fs/btrfs/sysfs.c                |  28 ++++
 fs/btrfs/transaction.c          |  19 ++-
 fs/btrfs/tree-checker.c         |   3 +
 fs/btrfs/tree-log.c             |   3 +-
 include/uapi/linux/btrfs.h      |   2 +
 include/uapi/linux/btrfs_tree.h |  27 ++-
 25 files changed, 718 insertions(+), 143 deletions(-)
Signature · 1 line
2.41.0
#1 lore raw

PATCH v5 1/18 btrfs: introduce quota mode

In preparation for introducing simple quotas, change from a binary setting for quotas to an enum based mode. Initially, the possible modes are disabled/full. Full quotas is normal btrfs qgroups.

---
 fs/btrfs/qgroup.c | 7 +++++++
 fs/btrfs/qgroup.h | 6 ++++++
 2 files changed, 13 insertions(+)
fs/btrfs/qgroup.c +7 -0
index 2637d6b157ff..0a2085ae9bcd 100644
@@ -30,6 +30,13 @@
 #include "root-tree.h"
 #include "tree-checker.h"
 
+enum btrfs_qgroup_mode btrfs_qgroup_mode(struct btrfs_fs_info *fs_info)
+{
+	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+		return BTRFS_QGROUP_MODE_DISABLED;
+	return BTRFS_QGROUP_MODE_FULL;
+}
+
 /*
  * Helpers to access qgroup reservation
  *
fs/btrfs/qgroup.h +6 -0
index 7bffa10589d6..bb15e55f00b8 100644
@@ -250,6 +250,12 @@ enum {
 };
 
 int btrfs_quota_enable(struct btrfs_fs_info *fs_info);
+enum btrfs_qgroup_mode {
+	BTRFS_QGROUP_MODE_DISABLED,
+	BTRFS_QGROUP_MODE_FULL,
+};
+
+enum btrfs_qgroup_mode btrfs_qgroup_mode(struct btrfs_fs_info *fs_info);
 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
Signature · 1 line
2.41.0
#2 lore raw

PATCH v5 2/18 btrfs: add new quota mode for simple quotas

In reply to Boris Burkov

Add a new quota mode called "simple quotas". It can be enabled by the existing quota enable ioctl via a new command, and sets an incompat bit, as the implementation of simple quotas will make backwards incompatible changes to the disk format of the extent tree.

---
 fs/btrfs/delayed-ref.c          |  4 +-
 fs/btrfs/fs.h                   |  5 +-
 fs/btrfs/ioctl.c                |  3 +-
 fs/btrfs/qgroup.c               | 91 +++++++++++++++++++++++----------
 fs/btrfs/qgroup.h               |  4 +-
 fs/btrfs/root-tree.c            |  2 +-
 fs/btrfs/transaction.c          |  4 +-
 include/uapi/linux/btrfs.h      |  2 +
 include/uapi/linux/btrfs_tree.h | 14 ++++-
 9 files changed, 91 insertions(+), 38 deletions(-)
fs/btrfs/delayed-ref.c +2 -2
index 6a13cf00218b..a9b938d3a531 100644
@@ -898,7 +898,7 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 	}
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED &&
 	    !generic_ref->skip_qgroup) {
 		record = kzalloc(sizeof(*record), GFP_NOFS);
 		if (!record) {
@@ -1002,7 +1002,7 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 	}
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED &&
 	    !generic_ref->skip_qgroup) {
 		record = kzalloc(sizeof(*record), GFP_NOFS);
 		if (!record) {
fs/btrfs/fs.h +2 -3
index 203d2a267828..f76f450c2abf 100644
@@ -218,7 +218,8 @@ enum {
 	 BTRFS_FEATURE_INCOMPAT_NO_HOLES	|	\
 	 BTRFS_FEATURE_INCOMPAT_METADATA_UUID	|	\
 	 BTRFS_FEATURE_INCOMPAT_RAID1C34	|	\
-	 BTRFS_FEATURE_INCOMPAT_ZONED)
+	 BTRFS_FEATURE_INCOMPAT_ZONED		|	\
+	 BTRFS_FEATURE_INCOMPAT_SIMPLE_QUOTA)
 
 #ifdef CONFIG_BTRFS_DEBUG
 	/*
@@ -233,7 +234,6 @@ enum {
 
 #define BTRFS_FEATURE_INCOMPAT_SUPP		\
 	(BTRFS_FEATURE_INCOMPAT_SUPP_STABLE)
-
 #endif
 
 #define BTRFS_FEATURE_INCOMPAT_SAFE_SET			\
@@ -790,7 +790,6 @@ struct btrfs_fs_info {
 	struct lockdep_map btrfs_state_change_map[4];
 	struct lockdep_map btrfs_trans_pending_ordered_map;
 	struct lockdep_map btrfs_ordered_extent_map;
-
 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 	spinlock_t ref_verify_lock;
 	struct rb_root block_tree;
fs/btrfs/ioctl.c +2 -1
index a895d105464b..9b61bc62e439 100644
@@ -3691,7 +3691,8 @@ static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
 
 	switch (sa->cmd) {
 	case BTRFS_QUOTA_CTL_ENABLE:
-		ret = btrfs_quota_enable(fs_info);
+	case BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA:
+		ret = btrfs_quota_enable(fs_info, sa);
 		break;
 	case BTRFS_QUOTA_CTL_DISABLE:
 		ret = btrfs_quota_disable(fs_info);
fs/btrfs/qgroup.c +64 -27

303 lines not on this page · raw message

fs/btrfs/qgroup.h +3 -1
index bb15e55f00b8..d4c4d039585f 100644
@@ -249,13 +249,15 @@ enum {
 	ENUM_BIT(QGROUP_FREE),
 };
 
-int btrfs_quota_enable(struct btrfs_fs_info *fs_info);
 enum btrfs_qgroup_mode {
 	BTRFS_QGROUP_MODE_DISABLED,
 	BTRFS_QGROUP_MODE_FULL,
+	BTRFS_QGROUP_MODE_SIMPLE
 };
 
 enum btrfs_qgroup_mode btrfs_qgroup_mode(struct btrfs_fs_info *fs_info);
+int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
+		       struct btrfs_ioctl_quota_ctl_args *quota_ctl_args);
 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
fs/btrfs/root-tree.c +1 -1
index 859874579456..044a8c2710f8 100644
@@ -508,7 +508,7 @@ int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED) {
 		/* One for parent inode, two for dir entries */
 		qgroup_num_bytes = 3 * fs_info->nodesize;
 		ret = btrfs_qgroup_reserve_meta_prealloc(root,
fs/btrfs/transaction.c +2 -2
index 815f61d6b506..89ff15aa085f 100644
@@ -1529,11 +1529,11 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 	int ret;
 
 	/*
-	 * Save some performance in the case that qgroups are not
+	 * Save some performance in the case that full qgroups are not
 	 * enabled. If this check races with the ioctl, rescan will
 	 * kick in anyway.
 	 */
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 
 	/*
include/uapi/linux/btrfs.h +2 -0
index dbb8b96da50d..0e42f4a2121d 100644
@@ -333,6 +333,7 @@ struct btrfs_ioctl_fs_info_args {
 #define BTRFS_FEATURE_INCOMPAT_RAID1C34		(1ULL << 11)
 #define BTRFS_FEATURE_INCOMPAT_ZONED		(1ULL << 12)
 #define BTRFS_FEATURE_INCOMPAT_EXTENT_TREE_V2	(1ULL << 13)
+#define BTRFS_FEATURE_INCOMPAT_SIMPLE_QUOTA	(1ULL << 14)
 
 struct btrfs_ioctl_feature_flags {
 	__u64 compat_flags;
@@ -753,6 +754,7 @@ struct btrfs_ioctl_get_dev_stats {
 #define BTRFS_QUOTA_CTL_ENABLE	1
 #define BTRFS_QUOTA_CTL_DISABLE	2
 #define BTRFS_QUOTA_CTL_RESCAN__NOTUSED	3
+#define BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA 4
 struct btrfs_ioctl_quota_ctl_args {
 	__u64 cmd;
 	__u64 status;
include/uapi/linux/btrfs_tree.h +13 -1
index ab38d0f411fa..47aca414a41b 100644
@@ -1200,9 +1200,21 @@ static inline __u16 btrfs_qgroup_level(__u64 qgroupid)
  */
 #define BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT	(1ULL << 2)
 
+/*
+ * 3rd and 4th bits taken by non-persisted status flags in qgroup.h
+ */
+
+/*
+ * Whether or not this filesystem is using simple quotas.
+ * Not exactly the incompat bit, because we support using simple quotas,
+ * disabling it, then going back to full qgroup quotas.
+ */
+#define BTRFS_QGROUP_STATUS_FLAG_SIMPLE	(1ULL << 5)
+
 #define BTRFS_QGROUP_STATUS_FLAGS_MASK	(BTRFS_QGROUP_STATUS_FLAG_ON |		\
 					 BTRFS_QGROUP_STATUS_FLAG_RESCAN |	\
-					 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT)
+					 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |	\
+					 BTRFS_QGROUP_STATUS_FLAG_SIMPLE)
 
 #define BTRFS_QGROUP_STATUS_VERSION        1
 
Signature · 1 line
2.41.0
Reviewed-by on v5 2/18
#3 lore raw

Re: [PATCH v5 02/18] btrfs: add new quota mode for simple quotas

Quoted 55 lines · Boris Burkov

Add a new quota mode called "simple quotas". It can be enabled by the existing quota enable ioctl via a new command, and sets an incompat bit, as the implementation of simple quotas will make backwards incompatible changes to the disk format of the extent tree.

---
 fs/btrfs/delayed-ref.c          |  4 +-
 fs/btrfs/fs.h                   |  5 +-
 fs/btrfs/ioctl.c                |  3 +-
 fs/btrfs/qgroup.c               | 91 +++++++++++++++++++++++----------
 fs/btrfs/qgroup.h               |  4 +-
 fs/btrfs/root-tree.c            |  2 +-
 fs/btrfs/transaction.c          |  4 +-
 include/uapi/linux/btrfs.h      |  2 +
 include/uapi/linux/btrfs_tree.h | 14 ++++-
 9 files changed, 91 insertions(+), 38 deletions(-)
fs/btrfs/delayed-ref.c +2 -2
index 6a13cf00218b..a9b938d3a531 100644
@@ -898,7 +898,7 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 	}
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED &&
 	    !generic_ref->skip_qgroup) {
 		record = kzalloc(sizeof(*record), GFP_NOFS);
 		if (!record) {
@@ -1002,7 +1002,7 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 	}
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED &&
 	    !generic_ref->skip_qgroup) {
 		record = kzalloc(sizeof(*record), GFP_NOFS);
 		if (!record) {
fs/btrfs/fs.h +2 -1
index 203d2a267828..f76f450c2abf 100644
@@ -218,7 +218,8 @@ enum {
 	 BTRFS_FEATURE_INCOMPAT_NO_HOLES	|	\
 	 BTRFS_FEATURE_INCOMPAT_METADATA_UUID	|	\
 	 BTRFS_FEATURE_INCOMPAT_RAID1C34	|	\
-	 BTRFS_FEATURE_INCOMPAT_ZONED)
+	 BTRFS_FEATURE_INCOMPAT_ZONED		|	\
+	 BTRFS_FEATURE_INCOMPAT_SIMPLE_QUOTA)
 
 #ifdef CONFIG_BTRFS_DEBUG
 	/*

fs/btrfs/fs.h enum { · line 236

Quoted 4 lines
fs/btrfs/fs.h +0 -1
@@ -233,7 +234,6 @@ enum {
 
 #define BTRFS_FEATURE_INCOMPAT_SUPP		\
 	(BTRFS_FEATURE_INCOMPAT_SUPP_STABLE)
-

Extraneous newline change.

fs/btrfs/fs.h struct btrfs_fs_info { · line 793

Quoted 8 lines
fs/btrfs/fs.h +0 -1
 #endif
 
 #define BTRFS_FEATURE_INCOMPAT_SAFE_SET			\
@@ -790,7 +790,6 @@ struct btrfs_fs_info {
 	struct lockdep_map btrfs_state_change_map[4];
 	struct lockdep_map btrfs_trans_pending_ordered_map;
 	struct lockdep_map btrfs_ordered_extent_map;
-

Same here. Once you fix it up you can add

Thanks,

Josef

#4 lore raw

Re: [PATCH v5 02/18] btrfs: add new quota mode for simple quotas

In reply to Boris Burkov
Quoted 23 lines · Boris Burkov

Add a new quota mode called "simple quotas". It can be enabled by the existing quota enable ioctl via a new command, and sets an incompat bit, as the implementation of simple quotas will make backwards incompatible changes to the disk format of the extent tree.

---
 fs/btrfs/delayed-ref.c          |  4 +-
 fs/btrfs/fs.h                   |  5 +-
 fs/btrfs/ioctl.c                |  3 +-
 fs/btrfs/qgroup.c               | 91 +++++++++++++++++++++++----------
 fs/btrfs/qgroup.h               |  4 +-
 fs/btrfs/root-tree.c            |  2 +-
 fs/btrfs/transaction.c          |  4 +-
 include/uapi/linux/btrfs.h      |  2 +
 include/uapi/linux/btrfs_tree.h | 14 ++++-
 9 files changed, 91 insertions(+), 38 deletions(-)
fs/btrfs/delayed-ref.c +0 -0
index 6a13cf00218b..a9b938d3a531 100644

fs/btrfs/delayed-ref.c int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans, · line 901

Quoted 5 lines
fs/btrfs/delayed-ref.c +1 -1
@@ -898,7 +898,7 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 	}
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED &&

The expression

"btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED"

is repeated in many places, this should be a helper. Checking for the secific mode open coded is fine.

Quoted 27 lines
fs/btrfs/delayed-ref.c +1 -1
 	    !generic_ref->skip_qgroup) {
 		record = kzalloc(sizeof(*record), GFP_NOFS);
 		if (!record) {
@@ -1002,7 +1002,7 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 	}
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED &&
 	    !generic_ref->skip_qgroup) {
 		record = kzalloc(sizeof(*record), GFP_NOFS);
 		if (!record) {
fs/btrfs/fs.h +2 -1
index 203d2a267828..f76f450c2abf 100644
@@ -218,7 +218,8 @@ enum {
 	 BTRFS_FEATURE_INCOMPAT_NO_HOLES	|	\
 	 BTRFS_FEATURE_INCOMPAT_METADATA_UUID	|	\
 	 BTRFS_FEATURE_INCOMPAT_RAID1C34	|	\
-	 BTRFS_FEATURE_INCOMPAT_ZONED)
+	 BTRFS_FEATURE_INCOMPAT_ZONED		|	\
+	 BTRFS_FEATURE_INCOMPAT_SIMPLE_QUOTA)
 
 #ifdef CONFIG_BTRFS_DEBUG
 	/*

fs/btrfs/fs.h enum { · line 236

Quoted 4 lines
fs/btrfs/fs.h +0 -1
@@ -233,7 +234,6 @@ enum {
 
 #define BTRFS_FEATURE_INCOMPAT_SUPP		\
 	(BTRFS_FEATURE_INCOMPAT_SUPP_STABLE)
-

Keep the lines there please

fs/btrfs/fs.h struct btrfs_fs_info { · line 793

Quoted 8 lines
fs/btrfs/fs.h +0 -1
 #endif
 
 #define BTRFS_FEATURE_INCOMPAT_SAFE_SET			\
@@ -790,7 +790,6 @@ struct btrfs_fs_info {
 	struct lockdep_map btrfs_state_change_map[4];
 	struct lockdep_map btrfs_trans_pending_ordered_map;
 	struct lockdep_map btrfs_ordered_extent_map;
-

Same

Quoted 51 lines
fs/btrfs/fs.h +0 -0
 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 	spinlock_t ref_verify_lock;
 	struct rb_root block_tree;
fs/btrfs/ioctl.c +2 -1
index a895d105464b..9b61bc62e439 100644
@@ -3691,7 +3691,8 @@ static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
 
 	switch (sa->cmd) {
 	case BTRFS_QUOTA_CTL_ENABLE:
-		ret = btrfs_quota_enable(fs_info);
+	case BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA:
+		ret = btrfs_quota_enable(fs_info, sa);
 		break;
 	case BTRFS_QUOTA_CTL_DISABLE:
 		ret = btrfs_quota_disable(fs_info);
fs/btrfs/qgroup.c +6 -1
index 0a2085ae9bcd..558f66994667 100644
@@ -34,6 +34,8 @@ enum btrfs_qgroup_mode btrfs_qgroup_mode(struct btrfs_fs_info *fs_info)
 {
 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 		return BTRFS_QGROUP_MODE_DISABLED;
+	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE)
+		return BTRFS_QGROUP_MODE_SIMPLE;
 	return BTRFS_QGROUP_MODE_FULL;
 }
 
@@ -347,6 +349,8 @@ int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 
 static void qgroup_mark_inconsistent(struct btrfs_fs_info *fs_info)
 {
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
+		return;
 	fs_info->qgroup_flags |= (BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |
 				  BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
 				  BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
@@ -367,8 +371,9 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 	int ret = 0;
 	u64 flags = 0;
 	u64 rescan_progress = 0;
+	bool simple;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
 		return 0;
 
 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);

fs/btrfs/qgroup.c int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info) · line 427

Quoted 5 lines
fs/btrfs/qgroup.c +2 -0
@@ -418,14 +423,14 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 				 "old qgroup version, quota disabled");
 				goto out;
 			}
+			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, ptr);
+			simple = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE;

bool expressions assigned should be written like

			x = (a & b);

so it's clear that it's not a trivial statement. Similarly for '==' or ternary operator.

Quoted 32 lines
fs/btrfs/qgroup.c +4 -5
 			if (btrfs_qgroup_status_generation(l, ptr) !=
-			    fs_info->generation) {
+			    fs_info->generation && !simple) {
 				qgroup_mark_inconsistent(fs_info);
 				btrfs_err(fs_info,
 					"qgroup generation mismatch, marked as inconsistent");
 			}
-			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
-									  ptr);
 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 			goto next1;
 		}
@@ -557,7 +562,7 @@ bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
 	struct rb_node *node;
 	bool ret = false;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
 		return ret;
 	/*
 	 * Since we're unmounting, there is no race and no need to grab qgroup
@@ -956,7 +961,8 @@ static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
+int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
+		       struct btrfs_ioctl_quota_ctl_args *quota_ctl_args)
 {
 	struct btrfs_root *quota_root;
 	struct btrfs_root *tree_root = fs_info->tree_root;

fs/btrfs/qgroup.c int btrfs_quota_enable(struct btrfs_fs_info *fs_info) · line 977

Quoted 4 lines
fs/btrfs/qgroup.c +1 -0
@@ -968,6 +974,7 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 	struct btrfs_qgroup *qgroup = NULL;
 	struct btrfs_trans_handle *trans = NULL;
 	struct ulist *ulist = NULL;
+	bool simple = quota_ctl_args->cmd == BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA;
	const bool simple = ( ... == ...);
Quoted 60 lines
fs/btrfs/qgroup.c +17 -2
 	int ret = 0;
 	int slot;
 
@@ -1070,8 +1077,11 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 				 struct btrfs_qgroup_status_item);
 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
-	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
-				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
+	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON;
+	if (simple)
+		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
+	else
+		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
@@ -1187,8 +1197,14 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 	spin_lock(&fs_info->qgroup_lock);
 	fs_info->quota_root = quota_root;
 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
+	if (simple)
+		btrfs_set_fs_incompat(fs_info, SIMPLE_QUOTA);
 	spin_unlock(&fs_info->qgroup_lock);
 
+	/* Skip rescan for simple qgroups */
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
+		goto out_free_path;
+
 	ret = qgroup_rescan_init(fs_info, 0, 1);
 	if (!ret) {
 	        qgroup_rescan_zero_tracking(fs_info);
@@ -1302,6 +1318,7 @@ int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
 	quota_root = fs_info->quota_root;
 	fs_info->quota_root = NULL;
 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
+	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
 	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
 	spin_unlock(&fs_info->qgroup_lock);
 
@@ -1787,6 +1804,9 @@ int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
 	struct btrfs_qgroup_extent_record *entry;
 	u64 bytenr = record->bytenr;
 
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
+		return 0;
+
 	lockdep_assert_held(&delayed_refs->lock);
 	trace_btrfs_qgroup_trace_extent(fs_info, record);
 
@@ -1819,6 +1839,8 @@ int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
 	struct btrfs_backref_walk_ctx ctx = { 0 };
 	int ret;
 
+	if (btrfs_qgroup_mode(trans->fs_info) != BTRFS_QGROUP_MODE_FULL)
+		return 0;
 	/*
 	 * We are always called in a context where we are already holding a
 	 * transaction handle. Often we are called when adding a data delayed

fs/btrfs/qgroup.c int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr, · line 1899

Quoted 5 lines
fs/btrfs/qgroup.c +1 -1
@@ -1874,7 +1896,7 @@ int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
 	struct btrfs_delayed_ref_root *delayed_refs;
 	int ret;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL

Could this be written as an '==' condition?

fs/btrfs/qgroup.c int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr, · line 1900

Quoted 1 line
fs/btrfs/qgroup.c +0 -0
 	    || bytenr == 0 || num_bytes == 0)

With the rest it's not clear what exactly is it testing for.

fs/btrfs/qgroup.c int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans, · line 1932

Quoted 8 lines
fs/btrfs/qgroup.c +1 -1
 		return 0;
 	record = kzalloc(sizeof(*record), GFP_NOFS);
@@ -1907,7 +1929,7 @@ int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
 	u64 bytenr, num_bytes;
 
 	/* We can be called directly from walk_up_proc() */
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)

Reading more from the patch, the meaning of the "!= FULL" is either disabled or simple quotas, can this be inverted so we have something like "if no accounting"? This covers either quotas disabled or simple quotas.

Quoted 78 lines
fs/btrfs/qgroup.c +12 -6
 		return 0;
 
 	for (i = 0; i < nr; i++) {
@@ -2283,7 +2305,7 @@ static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
 	int level;
 	int ret;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 
 	/* Wrong parameter order */
@@ -2340,7 +2362,7 @@ int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
 	BUG_ON(root_eb == NULL);
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 
 	spin_lock(&fs_info->qgroup_lock);
@@ -2680,7 +2702,7 @@ int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
 	 * If quotas get disabled meanwhile, the resources need to be freed and
 	 * we can't just exit here.
 	 */
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL ||
 	    fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
 		goto out_free;
 
@@ -2768,6 +2790,9 @@ int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
 	u64 qgroup_to_skip;
 	int ret = 0;
 
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
+		return 0;
+
 	delayed_refs = &trans->transaction->delayed_refs;
 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
@@ -2883,7 +2908,7 @@ int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
 			qgroup_mark_inconsistent(fs_info);
 		spin_lock(&fs_info->qgroup_lock);
 	}
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED)
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
 	else
 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
@@ -2936,7 +2961,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 
 	if (!committing)
 		mutex_lock(&fs_info->qgroup_ioctl_lock);
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
 		goto out;
 
 	quota_root = fs_info->quota_root;
@@ -3010,7 +3035,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 		qgroup_dirty(fs_info, dstgroup);
 	}
 
-	if (srcid) {
+	if (srcid && btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL) {
 		srcgroup = find_qgroup_rb(fs_info, srcid);
 		if (!srcgroup)
 			goto unlock;
@@ -3302,6 +3327,9 @@ static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
 	int slot;
 	int ret;
 
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
+		return 1;
+
 	mutex_lock(&fs_info->qgroup_rescan_lock);
 	extent_root = btrfs_extent_root(fs_info,
 				fs_info->qgroup_rescan_progress.objectid);

fs/btrfs/qgroup.c static bool rescan_should_stop(struct btrfs_fs_info *fs_info) · line 3416

Quoted 7 lines
fs/btrfs/qgroup.c +2 -2
@@ -3384,8 +3412,8 @@ static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
 {
 	return btrfs_fs_closing(fs_info) ||
 		test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
-		!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
-			  fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
+		btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
+		fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;

The condition has become quite unreadable, please convert it to an 'if' or series of more 'ifs' grouping the conditions. Eg. the btrfs_fs_closing() can be a separate statement as it's not directly related to the quotas and rescan.

Quoted 14 lines
fs/btrfs/qgroup.c +3 -0
 }
 
 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
@@ -3399,6 +3427,9 @@ static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
 	bool stopped = false;
 	bool did_leaf_rescans = false;
 
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
+		return;
+
 	path = btrfs_alloc_path();
 	if (!path)
 		goto out;

fs/btrfs/qgroup.c qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid, · line 3537

Quoted 5 lines
fs/btrfs/qgroup.c +2 -0
@@ -3502,6 +3533,12 @@ qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 {
 	int ret = 0;
 
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) {
+		btrfs_warn(fs_info, "qgroup rescan init failed, running in simple mode. mode: %d\n",

No "\n" in the message helpers and I don't think we need the numeric value of the mode when it's stated in words

Quoted 165 lines
fs/btrfs/qgroup.c +13 -9
+			btrfs_qgroup_mode(fs_info));
+		return -EINVAL;
+	}
+
 	if (!init_flags) {
 		/* we're resuming qgroup rescan at mount time */
 		if (!(fs_info->qgroup_flags &
@@ -3532,7 +3569,7 @@ qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 			btrfs_warn(fs_info,
 			"qgroup rescan init failed, qgroup is not enabled");
 			ret = -EINVAL;
-		} else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
+		} else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED) {
 			/* Quota disable is in progress */
 			ret = -EBUSY;
 		}
@@ -3788,7 +3825,7 @@ static int qgroup_reserve_data(struct btrfs_inode *inode,
 	u64 to_reserve;
 	int ret;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
+	if (btrfs_qgroup_mode(root->fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
 	    !is_fstree(root->root_key.objectid) || len == 0)
 		return 0;
 
@@ -3920,7 +3957,7 @@ static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
 	int trace_op = QGROUP_RELEASE;
 	int ret;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
+	if (btrfs_qgroup_mode(inode->root->fs_info) == BTRFS_QGROUP_MODE_DISABLED)
 		return 0;
 
 	/* In release case, we shouldn't have @reserved */
@@ -4031,7 +4068,7 @@ int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	int ret;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
 		return 0;
 
@@ -4072,7 +4109,7 @@ void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
 	    !is_fstree(root->root_key.objectid))
 		return;
 
@@ -4088,7 +4125,7 @@ void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
 	    !is_fstree(root->root_key.objectid))
 		return;
 
@@ -4153,7 +4190,7 @@ void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
 	    !is_fstree(root->root_key.objectid))
 		return;
 	/* Same as btrfs_qgroup_free_meta_prealloc() */
@@ -4261,7 +4298,7 @@ int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
 	int level = btrfs_header_level(subvol_parent) - 1;
 	int ret = 0;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 
 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
@@ -4371,7 +4408,7 @@ int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
 	int ret = 0;
 	int i;
 
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
 		return 0;
fs/btrfs/qgroup.h +3 -1
index bb15e55f00b8..d4c4d039585f 100644
@@ -249,13 +249,15 @@ enum {
 	ENUM_BIT(QGROUP_FREE),
 };
 
-int btrfs_quota_enable(struct btrfs_fs_info *fs_info);
 enum btrfs_qgroup_mode {
 	BTRFS_QGROUP_MODE_DISABLED,
 	BTRFS_QGROUP_MODE_FULL,
+	BTRFS_QGROUP_MODE_SIMPLE
 };
 
 enum btrfs_qgroup_mode btrfs_qgroup_mode(struct btrfs_fs_info *fs_info);
+int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
+		       struct btrfs_ioctl_quota_ctl_args *quota_ctl_args);
 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
fs/btrfs/root-tree.c +1 -1
index 859874579456..044a8c2710f8 100644
@@ -508,7 +508,7 @@ int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
 
-	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED) {
 		/* One for parent inode, two for dir entries */
 		qgroup_num_bytes = 3 * fs_info->nodesize;
 		ret = btrfs_qgroup_reserve_meta_prealloc(root,
fs/btrfs/transaction.c +2 -2
index 815f61d6b506..89ff15aa085f 100644
@@ -1529,11 +1529,11 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 	int ret;
 
 	/*
-	 * Save some performance in the case that qgroups are not
+	 * Save some performance in the case that full qgroups are not
 	 * enabled. If this check races with the ioctl, rescan will
 	 * kick in anyway.
 	 */
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 
 	/*
include/uapi/linux/btrfs.h +2 -0
index dbb8b96da50d..0e42f4a2121d 100644
@@ -333,6 +333,7 @@ struct btrfs_ioctl_fs_info_args {
 #define BTRFS_FEATURE_INCOMPAT_RAID1C34		(1ULL << 11)
 #define BTRFS_FEATURE_INCOMPAT_ZONED		(1ULL << 12)
 #define BTRFS_FEATURE_INCOMPAT_EXTENT_TREE_V2	(1ULL << 13)
+#define BTRFS_FEATURE_INCOMPAT_SIMPLE_QUOTA	(1ULL << 14)
 
 struct btrfs_ioctl_feature_flags {
 	__u64 compat_flags;
@@ -753,6 +754,7 @@ struct btrfs_ioctl_get_dev_stats {
 #define BTRFS_QUOTA_CTL_ENABLE	1
 #define BTRFS_QUOTA_CTL_DISABLE	2
 #define BTRFS_QUOTA_CTL_RESCAN__NOTUSED	3
+#define BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA 4
 struct btrfs_ioctl_quota_ctl_args {
 	__u64 cmd;
 	__u64 status;
include/uapi/linux/btrfs_tree.h +0 -0
index ab38d0f411fa..47aca414a41b 100644

include/uapi/linux/btrfs_tree.h static inline __u16 btrfs_qgroup_level(__u64 qgroupid) · line 1205

Quoted 6 lines
include/uapi/linux/btrfs_tree.h +3 -0
@@ -1200,9 +1200,21 @@ static inline __u16 btrfs_qgroup_level(__u64 qgroupid)
  */
 #define BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT	(1ULL << 2)
 
+/*
+ * 3rd and 4th bits taken by non-persisted status flags in qgroup.h
+ */

If the bits are not persisted and used only internally or for in-memory tracking then they should be renumbered so you can use the value (1ULL << 3) in sequence.

include/uapi/linux/btrfs_tree.h static inline __u16 btrfs_qgroup_level(__u64 qgroupid) · line 1212

Quoted 7 lines
include/uapi/linux/btrfs_tree.h +7 -0
+
+/*
+ * Whether or not this filesystem is using simple quotas.
+ * Not exactly the incompat bit, because we support using simple quotas,
+ * disabling it, then going back to full qgroup quotas.
+ */
+#define BTRFS_QGROUP_STATUS_FLAG_SIMPLE	(1ULL << 5)

Here the context of 'SIMPLE' is not obvious, it's referring to accounting but adding that to the identifier would make it quite long so I'm not sure if we should do that. OTOH it's in the ioctl public API so clarity is important.

Quoted 11 lines
include/uapi/linux/btrfs_tree.h +3 -1
+
 #define BTRFS_QGROUP_STATUS_FLAGS_MASK	(BTRFS_QGROUP_STATUS_FLAG_ON |		\
 					 BTRFS_QGROUP_STATUS_FLAG_RESCAN |	\
-					 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT)
+					 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |	\
+					 BTRFS_QGROUP_STATUS_FLAG_SIMPLE)
 
 #define BTRFS_QGROUP_STATUS_VERSION        1
 
Signature · 1 line
2.41.0
#5 lore raw

PATCH v5 3/18 btrfs: expose quota mode via sysfs

In reply to Boris Burkov

Add a new sysfs file
/sys/fs/btrfs/<uuid>/qgroups/mode
which prints out the mode qgroups is running in. The possible modes are disabled, qgroup, and squota

---
 fs/btrfs/sysfs.c | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)
fs/btrfs/sysfs.c +26 -0
index b1d1ac25237b..e53614753391 100644
@@ -2086,6 +2086,31 @@ static ssize_t qgroup_enabled_show(struct kobject *qgroups_kobj,
 }
 BTRFS_ATTR(qgroups, enabled, qgroup_enabled_show);
 
+static ssize_t qgroup_mode_show(struct kobject *qgroups_kobj,
+				struct kobj_attribute *a,
+				char *buf)
+{
+	struct btrfs_fs_info *fs_info = to_fs_info(qgroups_kobj->parent);
+	char *mode = "";
+
+	spin_lock(&fs_info->qgroup_lock);
+	switch (btrfs_qgroup_mode(fs_info)) {
+	case BTRFS_QGROUP_MODE_DISABLED:
+		mode = "disabled";
+		break;
+	case BTRFS_QGROUP_MODE_FULL:
+		mode = "qgroup";
+		break;
+	case BTRFS_QGROUP_MODE_SIMPLE:
+		mode = "squota";
+		break;
+	}
+	spin_unlock(&fs_info->qgroup_lock);
+
+	return sysfs_emit(buf, "%s\n", mode);
+}
+BTRFS_ATTR(qgroups, mode, qgroup_mode_show);
+
 static ssize_t qgroup_inconsistent_show(struct kobject *qgroups_kobj,
 					struct kobj_attribute *a,
 					char *buf)
@@ -2148,6 +2173,7 @@ static struct attribute *qgroups_attrs[] = {
 	BTRFS_ATTR_PTR(qgroups, enabled),
 	BTRFS_ATTR_PTR(qgroups, inconsistent),
 	BTRFS_ATTR_PTR(qgroups, drop_subtree_threshold),
+	BTRFS_ATTR_PTR(qgroups, mode),
 	NULL
 };
 ATTRIBUTE_GROUPS(qgroups);
Signature · 1 line
2.41.0
#7 lore raw

Re: [PATCH v5 03/18] btrfs: expose quota mode via sysfs

In reply to Boris Burkov
Quoted 4 lines · Boris Burkov

Add a new sysfs file
/sys/fs/btrfs/<uuid>/qgroups/mode
which prints out the mode qgroups is running in. The possible modes are disabled, qgroup, and squota

Can you get the 'disabled' at all? Because when the quotas are disabled by ioctl the whole sysfs directory is gone, see btrfs_free_qgroup_config().

Quoted 20 lines
---
 fs/btrfs/sysfs.c | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)
fs/btrfs/sysfs.c +7 -0
index b1d1ac25237b..e53614753391 100644
@@ -2086,6 +2086,31 @@ static ssize_t qgroup_enabled_show(struct kobject *qgroups_kobj,
 }
 BTRFS_ATTR(qgroups, enabled, qgroup_enabled_show);
 
+static ssize_t qgroup_mode_show(struct kobject *qgroups_kobj,
+				struct kobj_attribute *a,
+				char *buf)
+{
+	struct btrfs_fs_info *fs_info = to_fs_info(qgroups_kobj->parent);
+	char *mode = "";
+

fs/btrfs/sysfs.c static ssize_t qgroup_enabled_show(struct kobject *qgroups_kobj, · line 2105

Quoted 10 lines
fs/btrfs/sysfs.c +10 -0
+	spin_lock(&fs_info->qgroup_lock);
+	switch (btrfs_qgroup_mode(fs_info)) {
+	case BTRFS_QGROUP_MODE_DISABLED:
+		mode = "disabled";
+		break;
+	case BTRFS_QGROUP_MODE_FULL:
+		mode = "qgroup";
+		break;
+	case BTRFS_QGROUP_MODE_SIMPLE:
+		mode = "squota";

You can do

	lock;
	switch (mode) {
	case FULL:   sysfs_emit(buf, "qgroup\n"); break;
	case SIMPLE: sysfs_emit(buf, "simple\n"); break;
	}
	unlock;
	return 7;

or track the return value from sysfs_emit so it's not so hacky.

Quoted 21 lines
fs/btrfs/sysfs.c +9 -0
+		break;
+	}
+	spin_unlock(&fs_info->qgroup_lock);
+
+	return sysfs_emit(buf, "%s\n", mode);
+}
+BTRFS_ATTR(qgroups, mode, qgroup_mode_show);
+
 static ssize_t qgroup_inconsistent_show(struct kobject *qgroups_kobj,
 					struct kobj_attribute *a,
 					char *buf)
@@ -2148,6 +2173,7 @@ static struct attribute *qgroups_attrs[] = {
 	BTRFS_ATTR_PTR(qgroups, enabled),
 	BTRFS_ATTR_PTR(qgroups, inconsistent),
 	BTRFS_ATTR_PTR(qgroups, drop_subtree_threshold),
+	BTRFS_ATTR_PTR(qgroups, mode),
 	NULL
 };
 ATTRIBUTE_GROUPS(qgroups);
Signature · 1 line
2.41.0
#8 lore raw

PATCH v5 4/18 btrfs: add simple_quota incompat feature to sysfs

In reply to Boris Burkov

Add an entry in the features directory for the new incompat flag

---
 fs/btrfs/sysfs.c | 2 ++
 1 file changed, 2 insertions(+)
fs/btrfs/sysfs.c +2 -0
index e53614753391..f62bba0068ca 100644
@@ -291,6 +291,7 @@ BTRFS_FEAT_ATTR_INCOMPAT(metadata_uuid, METADATA_UUID);
 BTRFS_FEAT_ATTR_COMPAT_RO(free_space_tree, FREE_SPACE_TREE);
 BTRFS_FEAT_ATTR_COMPAT_RO(block_group_tree, BLOCK_GROUP_TREE);
 BTRFS_FEAT_ATTR_INCOMPAT(raid1c34, RAID1C34);
+BTRFS_FEAT_ATTR_INCOMPAT(simple_quota, SIMPLE_QUOTA);
 #ifdef CONFIG_BLK_DEV_ZONED
 BTRFS_FEAT_ATTR_INCOMPAT(zoned, ZONED);
 #endif
@@ -322,6 +323,7 @@ static struct attribute *btrfs_supported_feature_attrs[] = {
 	BTRFS_FEAT_ATTR_PTR(free_space_tree),
 	BTRFS_FEAT_ATTR_PTR(raid1c34),
 	BTRFS_FEAT_ATTR_PTR(block_group_tree),
+	BTRFS_FEAT_ATTR_PTR(simple_quota),
 #ifdef CONFIG_BLK_DEV_ZONED
 	BTRFS_FEAT_ATTR_PTR(zoned),
 #endif
Signature · 1 line
2.41.0
#10 lore raw

Re: [PATCH v5 04/18] btrfs: add simple_quota incompat feature to sysfs

In reply to Boris Burkov
Quoted 12 lines · Boris Burkov

Add an entry in the features directory for the new incompat flag

---
 fs/btrfs/sysfs.c | 2 ++
 1 file changed, 2 insertions(+)
fs/btrfs/sysfs.c +0 -0
index e53614753391..f62bba0068ca 100644

fs/btrfs/sysfs.c BTRFS_FEAT_ATTR_INCOMPAT(metadata_uuid, METADATA_UUID); · line 294

Quoted 4 lines
fs/btrfs/sysfs.c +1 -0
@@ -291,6 +291,7 @@ BTRFS_FEAT_ATTR_INCOMPAT(metadata_uuid, METADATA_UUID);
 BTRFS_FEAT_ATTR_COMPAT_RO(free_space_tree, FREE_SPACE_TREE);
 BTRFS_FEAT_ATTR_COMPAT_RO(block_group_tree, BLOCK_GROUP_TREE);
 BTRFS_FEAT_ATTR_INCOMPAT(raid1c34, RAID1C34);
+BTRFS_FEAT_ATTR_INCOMPAT(simple_quota, SIMPLE_QUOTA);

I'm not sure if you mentioned in the cover letter or if we had discussed it before, but does this need to be a full incompat bit? I.e. no mount on older kernels, compared to a COMPAT_RO which would allow read-only mount.

Quoted 13 lines
fs/btrfs/sysfs.c +1 -0
 #ifdef CONFIG_BLK_DEV_ZONED
 BTRFS_FEAT_ATTR_INCOMPAT(zoned, ZONED);
 #endif
@@ -322,6 +323,7 @@ static struct attribute *btrfs_supported_feature_attrs[] = {
 	BTRFS_FEAT_ATTR_PTR(free_space_tree),
 	BTRFS_FEAT_ATTR_PTR(raid1c34),
 	BTRFS_FEAT_ATTR_PTR(block_group_tree),
+	BTRFS_FEAT_ATTR_PTR(simple_quota),
 #ifdef CONFIG_BLK_DEV_ZONED
 	BTRFS_FEAT_ATTR_PTR(zoned),
 #endif
Signature · 1 line
2.41.0
#11 lore raw

Re: [PATCH v5 04/18] btrfs: add simple_quota incompat feature to sysfs

Quoted 8 lines · David Sterba
Quoted 7 lines · Boris Burkov

Add an entry in the features directory for the new incompat flag

---
 fs/btrfs/sysfs.c | 2 ++
 1 file changed, 2 insertions(+)
Quoted 14 lines

fs/btrfs/sysfs.c BTRFS_FEAT_ATTR_INCOMPAT(metadata_uuid, METADATA_UUID); · line 294

Quoted 9 lines
fs/btrfs/sysfs.c +1 -0
index e53614753391..f62bba0068ca 100644
@@ -291,6 +291,7 @@ BTRFS_FEAT_ATTR_INCOMPAT(metadata_uuid, METADATA_UUID);
 BTRFS_FEAT_ATTR_COMPAT_RO(free_space_tree, FREE_SPACE_TREE);
 BTRFS_FEAT_ATTR_COMPAT_RO(block_group_tree, BLOCK_GROUP_TREE);
 BTRFS_FEAT_ATTR_INCOMPAT(raid1c34, RAID1C34);
+BTRFS_FEAT_ATTR_INCOMPAT(simple_quota, SIMPLE_QUOTA);

I'm not sure if you mentioned in the cover letter or if we had discussed it before, but does this need to be a full incompat bit? I.e. no mount on older kernels, compared to a COMPAT_RO which would allow read-only mount.

Unfortunately, as it is, simple quotas does need a full incompat bit. That is because of the details of how the kernel parses inline refs, but essentially that code relies on item size being fully exhausted by an iteration that steps forward by hard-code-computed inline ref size chunks. That parsing code blows up on the new structures in a way that can't be fixed, as far as I can tell.

To be COMPAT_RO, we would need to introduce an entirely new item. We discussed this in one of the early discussions and concluded that was not worth the space cost compared to an inline item.

Quoted 14 lines
Quoted 14 lines
fs/btrfs/sysfs.c +1 -0
 #ifdef CONFIG_BLK_DEV_ZONED
 BTRFS_FEAT_ATTR_INCOMPAT(zoned, ZONED);
 #endif
@@ -322,6 +323,7 @@ static struct attribute *btrfs_supported_feature_attrs[] = {
 	BTRFS_FEAT_ATTR_PTR(free_space_tree),
 	BTRFS_FEAT_ATTR_PTR(raid1c34),
 	BTRFS_FEAT_ATTR_PTR(block_group_tree),
+	BTRFS_FEAT_ATTR_PTR(simple_quota),
 #ifdef CONFIG_BLK_DEV_ZONED
 	BTRFS_FEAT_ATTR_PTR(zoned),
 #endif
Signature · 1 line
2.41.0
#12 lore raw

PATCH v5 5/18 btrfs: flush reservations during quota disable

In reply to Boris Burkov

The following sequence:
enable simple quotas
do some writes
reserve space
create ordered_extent
release rsv (store rsv_bytes in OE, mark QGROUP_RESERVED bits)
disable quotas
enable simple quotas
set qgroup rsv to 0 on all subvols
ordered_extent finishes
create delayed ref with rsv_bytes from before
run delayed ref

    record_simple_quota_delta
        free rsv_bytes (0 -> -rsv_delta)

results in us reliably underflowing the subvolume's qgroup rsv counter, because disabling/re-enabling quotas toggles reservation counters down to 0, but does not remove other file system state which represents successful acquisition of qgroup rsv space. Specifically metadata rsv counters on the root object and rsv_bytes on ordered_extent objects that have released their reservation as well as the corresponding QGROUP_RESERVED extent bits.

Normal qgroups gets away with this, I believe because it forces more work to happen on transaction commit, but I am not certain it is totally safe from the ordered_extent/leaked extent bit variant. Simple quotas hits this reliably.

The intent of the fix is to make disable take the time to clear that external to qgroups state as well: after flipping off the quota bit on fs_info, flush delalloc and ordered extents, clearing the extent bits along the way. This makes it so there are no ordered extents or meta prealloc hanging around from the first enablement period during the second.

---
 fs/btrfs/qgroup.c | 47 ++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 44 insertions(+), 3 deletions(-)
fs/btrfs/qgroup.c +44 -3
index 558f66994667..18f521716e8d 100644
@@ -1248,6 +1248,40 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 	return ret;
 }
 
+/*
+ * It is possible to have outstanding ordered extents
+ * which reserved bytes before we disabled. We need to fully flush
+ * delalloc, ordered extents, and a commit to ensure that
+ * we don't leak such reservations, only to have them come back
+ * if we re-enable.
+ *
+ * i.e.:
+ * enable simple quotas
+ * reserve space
+ * release it, store rsv_bytes in OE
+ * disable quotas
+ * enable simple quotas (qgroup rsv are all 0)
+ * OE finishes
+ * run delayed refs
+ * free rsv_bytes, resulting in miscounting or even underflow
+ */
+static int flush_reservations(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_trans_handle *trans;
+	int ret;
+
+	ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
+	if (ret)
+		return ret;
+	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
+	trans = btrfs_join_transaction(fs_info->tree_root);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+	btrfs_commit_transaction(trans);
+
+	return ret;
+}
+
 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
 {
 	struct btrfs_root *quota_root;
@@ -1292,6 +1326,10 @@ int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 	btrfs_qgroup_wait_for_completion(fs_info, false);
 
+	ret = flush_reservations(fs_info);
+	if (ret)
+		goto out;
+
 	/*
 	 * 1 For the root item
 	 *
@@ -1353,7 +1391,7 @@ int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
 	if (ret && trans)
 		btrfs_end_transaction(trans);
 	else if (trans)
-		ret = btrfs_end_transaction(trans);
+		ret = btrfs_commit_transaction(trans);
 	mutex_unlock(&fs_info->cleaner_mutex);
 
 	return ret;
@@ -3957,8 +3995,11 @@ static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
 	int trace_op = QGROUP_RELEASE;
 	int ret;
 
-	if (btrfs_qgroup_mode(inode->root->fs_info) == BTRFS_QGROUP_MODE_DISABLED)
-		return 0;
+	if (btrfs_qgroup_mode(inode->root->fs_info) == BTRFS_QGROUP_MODE_DISABLED) {
+		extent_changeset_init(&changeset);
+		return clear_record_extent_bits(&inode->io_tree, start, start + len - 1,
+				       EXTENT_QGROUP_RESERVED, &changeset);
+	}
 
 	/* In release case, we shouldn't have @reserved */
 	WARN_ON(!free && reserved);
Signature · 1 line
2.41.0
#13 lore raw

PATCH v5 6/18 btrfs: create qgroup earlier in snapshot creation

In reply to Boris Burkov

Pull creating the qgroup earlier in the snapshot. This allows simple quotas qgroups to see all the metadata writes related to the snapshot being created and to be born with the root node accounted.

---
 fs/btrfs/qgroup.c      | 3 +++
 fs/btrfs/transaction.c | 6 ++++++
 2 files changed, 9 insertions(+)
fs/btrfs/qgroup.c +3 -0
index 18f521716e8d..8e3a4ced3077 100644
@@ -1672,6 +1672,9 @@ int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
 	struct btrfs_qgroup *qgroup;
 	int ret = 0;
 
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
+		return 0;
+
 	mutex_lock(&fs_info->qgroup_ioctl_lock);
 	if (!fs_info->quota_root) {
 		ret = -ENOTCONN;
fs/btrfs/transaction.c +6 -0
index 89ff15aa085f..25217888e897 100644
@@ -1722,6 +1722,12 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	}
 	btrfs_release_path(path);
 
+	ret = btrfs_create_qgroup(trans, objectid);
+	if (ret) {
+		btrfs_abort_transaction(trans, ret);
+		goto fail;
+	}
+
 	/*
 	 * pull in the delayed directory update
 	 * and the delayed inode item
Signature · 1 line
2.41.0
Reviewed-by on v5 6/18
#14 lore raw

Re: [PATCH v5 06/18] btrfs: create qgroup earlier in snapshot creation

Quoted 5 lines · Boris Burkov

Pull creating the qgroup earlier in the snapshot. This allows simple quotas qgroups to see all the metadata writes related to the snapshot being created and to be born with the root node accounted.

Thanks,

Josef

#15 lore raw

Re: [PATCH v5 06/18] btrfs: create qgroup earlier in snapshot creation

In reply to Boris Burkov
Quoted 29 lines · Boris Burkov

Pull creating the qgroup earlier in the snapshot. This allows simple quotas qgroups to see all the metadata writes related to the snapshot being created and to be born with the root node accounted.

---
 fs/btrfs/qgroup.c      | 3 +++
 fs/btrfs/transaction.c | 6 ++++++
 2 files changed, 9 insertions(+)
fs/btrfs/qgroup.c +3 -0
index 18f521716e8d..8e3a4ced3077 100644
@@ -1672,6 +1672,9 @@ int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
 	struct btrfs_qgroup *qgroup;
 	int ret = 0;
 
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
+		return 0;
+
 	mutex_lock(&fs_info->qgroup_ioctl_lock);
 	if (!fs_info->quota_root) {
 		ret = -ENOTCONN;
fs/btrfs/transaction.c +0 -0
index 89ff15aa085f..25217888e897 100644

fs/btrfs/transaction.c static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, · line 1727

Quoted 6 lines
fs/btrfs/transaction.c +3 -0
@@ -1722,6 +1722,12 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	}
 	btrfs_release_path(path);
 
+	ret = btrfs_create_qgroup(trans, objectid);
+	if (ret) {
+		btrfs_abort_transaction(trans, ret);

This adds and error case to the middle of a transaction commit. Snapshots are created in two parts, first is the ioctl adding the structure and then commit actually creates that. So the first phase preallocates what's needed (the root_item and path) and should do the same with the qgroups as much as possible.

Also check all the things that btrfs_create_qgroup() does, searches the qgroup tree, adds the new item, takes the qgroup_ioctl_lock mutex, and adds the sysfs entry (that does allocations under GFP_KERNEL). If you really need to create the qgroup like that then it needs much more care.

#16 lore raw

Re: [PATCH v5 06/18] btrfs: create qgroup earlier in snapshot creation

Quoted 43 lines · David Sterba
Quoted 29 lines · Boris Burkov

Pull creating the qgroup earlier in the snapshot. This allows simple quotas qgroups to see all the metadata writes related to the snapshot being created and to be born with the root node accounted.

---
 fs/btrfs/qgroup.c      | 3 +++
 fs/btrfs/transaction.c | 6 ++++++
 2 files changed, 9 insertions(+)
fs/btrfs/qgroup.c +3 -0
index 18f521716e8d..8e3a4ced3077 100644
@@ -1672,6 +1672,9 @@ int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
 	struct btrfs_qgroup *qgroup;
 	int ret = 0;
 
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
+		return 0;
+
 	mutex_lock(&fs_info->qgroup_ioctl_lock);
 	if (!fs_info->quota_root) {
 		ret = -ENOTCONN;
fs/btrfs/transaction.c +0 -0
index 89ff15aa085f..25217888e897 100644

fs/btrfs/transaction.c static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, · line 1727

Quoted 6 lines
fs/btrfs/transaction.c +3 -0
@@ -1722,6 +1722,12 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	}
 	btrfs_release_path(path);
 
+	ret = btrfs_create_qgroup(trans, objectid);
+	if (ret) {
+		btrfs_abort_transaction(trans, ret);

This adds and error case to the middle of a transaction commit. Snapshots are created in two parts, first is the ioctl adding the structure and then commit actually creates that. So the first phase preallocates what's needed (the root_item and path) and should do the same with the qgroups as much as possible.

Quoted 3 lines

Also check all the things that btrfs_create_qgroup() does, searches the qgroup tree, adds the new item, takes the qgroup_ioctl_lock mutex, and adds the sysfs entry (that does allocations under GFP_KERNEL).

I believe it does it with GFP_NOFS via allocating "prealloc". I might be missing another allocation under the covers. That's covered below, though.

Quoted 2 lines

If you really need to create the qgroup like that then it needs much more care.

As I understand it, the way that the qgroup gets created currently is by qgroup_account_snapshot which calls btrfs_qgroup_inherit in this same function.

btrfs_create_qgroup consists of:
- lock qgroup_ioctl_lock
- do an rbtree lookup for the qgid
- do a NOFS "prealloc" allocation for the qgroup struct
- add the qgroup item
- add it to the rbtree
- add it to sysfs (using the above nofs prealloc)

With the exception of the qgroup_ioctl_lock, all those are in btrfs_qgroup_inherit (and much more).

So that is all happening within create_pending_snapshot and thus the commit critical section. It also does other work like backref walks, and committing the roots.

Am I missing something important about the relative parts of create_pending_snapshots where this work is happening? My intent was to pull it up to before the run_delayed_refs in create_pending_snapshots so that the new dir metadata item gets counted correctly. I think I may have gotten delayed_refs and delayed_items confused and pulled it up *too* far, and can probably stuff it earlier in that account function or something.

Apologies if I am fundamentally misunderstanding something here.

#17 lore raw

PATCH v5 7/18 btrfs: function for recording simple quota deltas

In reply to Boris Burkov

Rather than re-computing shared/exclusive ownership based on backrefs and walking roots for implicit backrefs, simple quotas does an increment when creating an extent and a decrement when deleting it. Add the API for the extent item code to use to track those events.

Also add a helper function to make collecting parent qgroups in a ulist easier for functions like this.

---
 fs/btrfs/qgroup.c | 73 +++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/qgroup.h | 11 ++++++-
 2 files changed, 83 insertions(+), 1 deletion(-)
fs/btrfs/qgroup.c +73 -0
index 8e3a4ced3077..dedc532669f4 100644
@@ -332,6 +332,35 @@ static int del_relation_rb(struct btrfs_fs_info *fs_info,
 	return -ENOENT;
 }
 
+static int qgroup_collect_parents(struct btrfs_qgroup *qgroup,
+				  struct ulist *ul)
+{
+	struct ulist_iterator uiter;
+	struct ulist_node *unode;
+	struct btrfs_qgroup_list *glist;
+	struct btrfs_qgroup *qg;
+	int ret = 0;
+
+	ulist_reinit(ul);
+	ret = ulist_add(ul, qgroup->qgroupid,
+			qgroup_to_aux(qgroup), GFP_ATOMIC);
+	if (ret < 0)
+		goto out;
+	ULIST_ITER_INIT(&uiter);
+	while ((unode = ulist_next(ul, &uiter))) {
+		qg = unode_aux_to_qgroup(unode);
+		list_for_each_entry(glist, &qg->groups, next_group) {
+			ret = ulist_add(ul, glist->group->qgroupid,
+					qgroup_to_aux(glist->group), GFP_ATOMIC);
+			if (ret < 0)
+				goto out;
+		}
+	}
+	ret = 0;
+out:
+	return ret;
+}
+
 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 			       u64 rfer, u64 excl)
@@ -4535,3 +4564,47 @@ void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
 	}
 	*root = RB_ROOT;
 }
+
+int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
+				    struct btrfs_simple_quota_delta *delta)
+{
+	int ret;
+	struct ulist *ul = fs_info->qgroup_ulist;
+	struct btrfs_qgroup *qgroup;
+	struct ulist_iterator uiter;
+	struct ulist_node *unode;
+	struct btrfs_qgroup *qg;
+	u64 root = delta->root;
+	u64 num_bytes = delta->num_bytes;
+	int sign = delta->is_inc ? 1 : -1;
+
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE)
+		return 0;
+
+	if (!is_fstree(root))
+		return 0;
+
+	spin_lock(&fs_info->qgroup_lock);
+	qgroup = find_qgroup_rb(fs_info, root);
+	if (!qgroup) {
+		ret = -ENOENT;
+		goto out;
+	}
+	ret = qgroup_collect_parents(qgroup, ul);
+	if (ret)
+		goto out;
+
+	ULIST_ITER_INIT(&uiter);
+	while ((unode = ulist_next(ul, &uiter))) {
+		qg = unode_aux_to_qgroup(unode);
+		qg->excl += num_bytes * sign;
+		qg->rfer += num_bytes * sign;
+		qgroup_dirty(fs_info, qg);
+	}
+
+out:
+	spin_unlock(&fs_info->qgroup_lock);
+	if (!ret && delta->rsv_bytes)
+		btrfs_qgroup_free_refroot(fs_info, root, delta->rsv_bytes, BTRFS_QGROUP_RSV_DATA);
+	return ret;
+}
fs/btrfs/qgroup.h +10 -1
index d4c4d039585f..94d85b4fbebd 100644
@@ -235,6 +235,14 @@ struct btrfs_qgroup {
 	struct kobject kobj;
 };
 
+struct btrfs_simple_quota_delta {
+	u64 root; /* The fstree root this delta counts against */
+	u64 num_bytes; /* The number of bytes in the extent being counted */
+	u64 rsv_bytes; /* The number of bytes reserved for this extent */
+	bool is_inc; /* Whether we are using or freeing the extent */
+	bool is_data; /* Whether the extent is data or metadata */
+};
+
 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
 {
 	return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
@@ -447,5 +455,6 @@ int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
 		struct btrfs_root *root, struct extent_buffer *eb);
 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info);
-
+int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
+				    struct btrfs_simple_quota_delta *delta);
 #endif
Signature · 1 line
2.41.0
Reviewed-by on v5 7/18
#18 lore raw

Re: [PATCH v5 07/18] btrfs: function for recording simple quota deltas

Quoted 9 lines · Boris Burkov

Rather than re-computing shared/exclusive ownership based on backrefs and walking roots for implicit backrefs, simple quotas does an increment when creating an extent and a decrement when deleting it. Add the API for the extent item code to use to track those events.

Also add a helper function to make collecting parent qgroups in a ulist easier for functions like this.

Thanks,

Josef

#19 lore raw

Re: [PATCH v5 07/18] btrfs: function for recording simple quota deltas

In reply to Boris Burkov
Quoted 24 lines · Boris Burkov

Rather than re-computing shared/exclusive ownership based on backrefs and walking roots for implicit backrefs, simple quotas does an increment when creating an extent and a decrement when deleting it. Add the API for the extent item code to use to track those events.

Also add a helper function to make collecting parent qgroups in a ulist easier for functions like this.

---
 fs/btrfs/qgroup.c | 73 +++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/qgroup.h | 11 ++++++-
 2 files changed, 83 insertions(+), 1 deletion(-)
fs/btrfs/qgroup.c +2 -0
index 8e3a4ced3077..dedc532669f4 100644
@@ -332,6 +332,35 @@ static int del_relation_rb(struct btrfs_fs_info *fs_info,
 	return -ENOENT;
 }
 
+static int qgroup_collect_parents(struct btrfs_qgroup *qgroup,
+				  struct ulist *ul)

fs/btrfs/qgroup.c static int del_relation_rb(struct btrfs_fs_info *fs_info, · line 346

Quoted 10 lines
fs/btrfs/qgroup.c +10 -0
+{
+	struct ulist_iterator uiter;
+	struct ulist_node *unode;
+	struct btrfs_qgroup_list *glist;
+	struct btrfs_qgroup *qg;
+	int ret = 0;
+
+	ulist_reinit(ul);
+	ret = ulist_add(ul, qgroup->qgroupid,
+			qgroup_to_aux(qgroup), GFP_ATOMIC);

Qu has sent a series to get rid of the GFP_ATOMIC allocations when processing qgruops, so this would be good to port to the qgroup iterators as well but it's a recent change and can be done later as an optimization.

Quoted 21 lines
fs/btrfs/qgroup.c +17 -0
+	if (ret < 0)
+		goto out;
+	ULIST_ITER_INIT(&uiter);
+	while ((unode = ulist_next(ul, &uiter))) {
+		qg = unode_aux_to_qgroup(unode);
+		list_for_each_entry(glist, &qg->groups, next_group) {
+			ret = ulist_add(ul, glist->group->qgroupid,
+					qgroup_to_aux(glist->group), GFP_ATOMIC);
+			if (ret < 0)
+				goto out;
+		}
+	}
+	ret = 0;
+out:
+	return ret;
+}
+
 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 			       u64 rfer, u64 excl)

fs/btrfs/qgroup.c void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans) · line 4568

Quoted 5 lines
fs/btrfs/qgroup.c +2 -0
@@ -4535,3 +4564,47 @@ void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
 	}
 	*root = RB_ROOT;
 }
+
+int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,

You can abbreviate all the 'simple_quota' in identifiers as 'squota'.

fs/btrfs/qgroup.c void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans) · line 4579

Quoted 11 lines
fs/btrfs/qgroup.c +11 -0
+				    struct btrfs_simple_quota_delta *delta)
+{
+	int ret;
+	struct ulist *ul = fs_info->qgroup_ulist;
+	struct btrfs_qgroup *qgroup;
+	struct ulist_iterator uiter;
+	struct ulist_node *unode;
+	struct btrfs_qgroup *qg;
+	u64 root = delta->root;
+	u64 num_bytes = delta->num_bytes;
+	int sign = delta->is_inc ? 1 : -1;
	const int sign = (delta->is_inc ? 1 : -1);
Quoted 36 lines
fs/btrfs/qgroup.c +31 -0
+
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE)
+		return 0;
+
+	if (!is_fstree(root))
+		return 0;
+
+	spin_lock(&fs_info->qgroup_lock);
+	qgroup = find_qgroup_rb(fs_info, root);
+	if (!qgroup) {
+		ret = -ENOENT;
+		goto out;
+	}
+	ret = qgroup_collect_parents(qgroup, ul);
+	if (ret)
+		goto out;
+
+	ULIST_ITER_INIT(&uiter);
+	while ((unode = ulist_next(ul, &uiter))) {
+		qg = unode_aux_to_qgroup(unode);
+		qg->excl += num_bytes * sign;
+		qg->rfer += num_bytes * sign;
+		qgroup_dirty(fs_info, qg);
+	}
+
+out:
+	spin_unlock(&fs_info->qgroup_lock);
+	if (!ret && delta->rsv_bytes)
+		btrfs_qgroup_free_refroot(fs_info, root, delta->rsv_bytes, BTRFS_QGROUP_RSV_DATA);
+	return ret;
+}
fs/btrfs/qgroup.h +0 -0
index d4c4d039585f..94d85b4fbebd 100644

fs/btrfs/qgroup.h struct btrfs_qgroup { · line 238

Quoted 4 lines
fs/btrfs/qgroup.h +1 -0
@@ -235,6 +235,14 @@ struct btrfs_qgroup {
 	struct kobject kobj;
 };
 
+struct btrfs_simple_quota_delta {

struct btrfs_squota_delta

fs/btrfs/qgroup.h struct btrfs_qgroup { · line 243

Quoted 5 lines
fs/btrfs/qgroup.h +5 -0
+	u64 root; /* The fstree root this delta counts against */
+	u64 num_bytes; /* The number of bytes in the extent being counted */
+	u64 rsv_bytes; /* The number of bytes reserved for this extent */
+	bool is_inc; /* Whether we are using or freeing the extent */
+	bool is_data; /* Whether the extent is data or metadata */

Please put the comments on separate lines before the struct member definitions.

fs/btrfs/qgroup.h int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans, · line 459

Quoted 12 lines
fs/btrfs/qgroup.h +4 -1
+};
+
 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
 {
 	return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
@@ -447,5 +455,6 @@ int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
 		struct btrfs_root *root, struct extent_buffer *eb);
 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info);
-
+int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
+				    struct btrfs_simple_quota_delta *delta);

Please keep the newline before the last #endif

Quoted 3 lines
fs/btrfs/qgroup.h +0 -0
 #endif
Signature · 1 line
2.41.0
#20 lore raw

PATCH v5 8/18 btrfs: rename tree_ref and data_ref owning_root

In reply to Boris Burkov

commit 113479d5b8eb ("btrfs: rename root fields in delayed refs structs") changed these from ref_root to owning_root. However, there are many circumstances where that name is not really accurate and the root on the ref struct _is_ the referring root. In general, these are not the owning root, though it does happen in some ref merging cases involving overwrites during snapshots and similar.

Simple quotas cares quite a bit about tracking the original owner of an extent through delayed refs, so rename these back to free up the name for the real owning root (which will live on the generic btrfs_ref and the head ref)

---
 fs/btrfs/delayed-ref.c | 10 +++++-----
 fs/btrfs/delayed-ref.h | 12 ++++++------
 fs/btrfs/extent-tree.c | 10 +++++-----
 fs/btrfs/ref-verify.c  |  4 ++--
 4 files changed, 18 insertions(+), 18 deletions(-)
fs/btrfs/delayed-ref.c +5 -5
index a9b938d3a531..f0bae1e1c455 100644
@@ -885,7 +885,7 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 	u64 parent = generic_ref->parent;
 	u8 ref_type;
 
-	is_system = (generic_ref->tree_ref.owning_root == BTRFS_CHUNK_TREE_OBJECTID);
+	is_system = (generic_ref->tree_ref.ref_root == BTRFS_CHUNK_TREE_OBJECTID);
 
 	ASSERT(generic_ref->type == BTRFS_REF_METADATA && generic_ref->action);
 	ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS);
@@ -914,14 +914,14 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 		ref_type = BTRFS_TREE_BLOCK_REF_KEY;
 
 	init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes,
-				generic_ref->tree_ref.owning_root, action,
+				generic_ref->tree_ref.ref_root, action,
 				ref_type);
-	ref->root = generic_ref->tree_ref.owning_root;
+	ref->root = generic_ref->tree_ref.ref_root;
 	ref->parent = parent;
 	ref->level = level;
 
 	init_delayed_ref_head(head_ref, record, bytenr, num_bytes,
-			      generic_ref->tree_ref.owning_root, 0, action,
+			      generic_ref->tree_ref.ref_root, 0, action,
 			      false, is_system);
 	head_ref->extent_op = extent_op;
 
@@ -974,7 +974,7 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 	u64 bytenr = generic_ref->bytenr;
 	u64 num_bytes = generic_ref->len;
 	u64 parent = generic_ref->parent;
-	u64 ref_root = generic_ref->data_ref.owning_root;
+	u64 ref_root = generic_ref->data_ref.ref_root;
 	u64 owner = generic_ref->data_ref.ino;
 	u64 offset = generic_ref->data_ref.offset;
 	u8 ref_type;
fs/btrfs/delayed-ref.h +6 -6
index b8e14b0ba5f1..a71eff78469c 100644
@@ -188,8 +188,8 @@ enum btrfs_ref_type {
 struct btrfs_data_ref {
 	/* For EXTENT_DATA_REF */
 
-	/* Original root this data extent belongs to */
-	u64 owning_root;
+	/* Root which owns this data reference */
+	u64 ref_root;
 
 	/* Inode which refers to this data extent */
 	u64 ino;
@@ -212,11 +212,11 @@ struct btrfs_tree_ref {
 	int level;
 
 	/*
-	 * Root which owns this tree block.
+	 * Root which owns this tree block reference.
 	 *
 	 * For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
 	 */
-	u64 owning_root;
+	u64 ref_root;
 
 	/* For non-skinny metadata, no special member needed */
 };
@@ -294,7 +294,7 @@ static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref,
 	generic_ref->real_root = mod_root ?: root;
 #endif
 	generic_ref->tree_ref.level = level;
-	generic_ref->tree_ref.owning_root = root;
+	generic_ref->tree_ref.ref_root = root;
 	generic_ref->type = BTRFS_REF_METADATA;
 	if (skip_qgroup || !(is_fstree(root) &&
 			     (!mod_root || is_fstree(mod_root))))
@@ -312,7 +312,7 @@ static inline void btrfs_init_data_ref(struct btrfs_ref *generic_ref,
 	/* If @real_root not set, use @root as fallback */
 	generic_ref->real_root = mod_root ?: ref_root;
 #endif
-	generic_ref->data_ref.owning_root = ref_root;
+	generic_ref->data_ref.ref_root = ref_root;
 	generic_ref->data_ref.ino = ino;
 	generic_ref->data_ref.offset = offset;
 	generic_ref->type = BTRFS_REF_DATA;
fs/btrfs/extent-tree.c +5 -5
index 04ceb9d25d3e..018e288ccf7d 100644
@@ -1365,7 +1365,7 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 	ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
 	       generic_ref->action);
 	BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
-	       generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
+	       generic_ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID);
 
 	if (generic_ref->type == BTRFS_REF_METADATA)
 		ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
@@ -3307,9 +3307,9 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
 	 * tree, just update pinning info and exit early.
 	 */
 	if ((ref->type == BTRFS_REF_METADATA &&
-	     ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
+	     ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) ||
 	    (ref->type == BTRFS_REF_DATA &&
-	     ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
+	     ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
 		/* unlocks the pinned mutex */
 		btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
 		ret = 0;
@@ -3320,9 +3320,9 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
 	}
 
 	if (!((ref->type == BTRFS_REF_METADATA &&
-	       ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
+	       ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) ||
 	      (ref->type == BTRFS_REF_DATA &&
-	       ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
+	       ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
 		btrfs_ref_tree_mod(fs_info, ref);
 
 	return ret;
fs/btrfs/ref-verify.c +2 -2
index 95d28497de7c..b7b3bd86f5e2 100644
@@ -681,10 +681,10 @@ int btrfs_ref_tree_mod(struct btrfs_fs_info *fs_info,
 
 	if (generic_ref->type == BTRFS_REF_METADATA) {
 		if (!parent)
-			ref_root = generic_ref->tree_ref.owning_root;
+			ref_root = generic_ref->tree_ref.ref_root;
 		owner = generic_ref->tree_ref.level;
 	} else if (!parent) {
-		ref_root = generic_ref->data_ref.owning_root;
+		ref_root = generic_ref->data_ref.ref_root;
 		owner = generic_ref->data_ref.ino;
 		offset = generic_ref->data_ref.offset;
 	}
Signature · 1 line
2.41.0
#21 lore raw

PATCH v5 9/18 btrfs: track owning root in btrfs_ref

In reply to Boris Burkov

While data extents require us to store additional inline refs to track the original owner on free, this information is available implicitly for metadata. It is found in the owner field of the header of the tree block. Even if other trees refer to this block and the original ref goes away, we will not rewrite that header field, so it will reliably give the original owner.

In addition, there is a relocation case where a new data extent needs to have an owning root separate from the referring root wired through delayed refs.

To use it for recording simple quota deltas, we need to wire this root id through from when we create the delayed ref until we fully process it. Store it in the generic btrfs_ref struct of the delayed ref.

---
 fs/btrfs/delayed-ref.h |  7 +++++--
 fs/btrfs/extent-tree.c | 19 +++++++++++--------
 fs/btrfs/file.c        | 10 +++++-----
 fs/btrfs/inode-item.c  |  2 +-
 fs/btrfs/relocation.c  | 17 ++++++++++-------
 fs/btrfs/tree-log.c    |  3 ++-
 6 files changed, 34 insertions(+), 24 deletions(-)
fs/btrfs/delayed-ref.h +5 -2
index a71eff78469c..0729850a9193 100644
@@ -239,6 +239,7 @@ struct btrfs_ref {
 #endif
 	u64 bytenr;
 	u64 len;
+	u64 owning_root;
 
 	/* Bytenr of the parent tree block */
 	u64 parent;
@@ -278,16 +279,18 @@ static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_in
 }
 
 static inline void btrfs_init_generic_ref(struct btrfs_ref *generic_ref,
-				int action, u64 bytenr, u64 len, u64 parent)
+				int action, u64 bytenr, u64 len, u64 parent, u64 owning_root)
 {
 	generic_ref->action = action;
 	generic_ref->bytenr = bytenr;
 	generic_ref->len = len;
 	generic_ref->parent = parent;
+	generic_ref->owning_root = owning_root;
 }
 
 static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref,
-				int level, u64 root, u64 mod_root, bool skip_qgroup)
+				int level, u64 root, u64 mod_root,
+				bool skip_qgroup)
 {
 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 	/* If @real_root not set, use @root as fallback */
fs/btrfs/extent-tree.c +11 -8
index 018e288ccf7d..4f0115553cd3 100644
@@ -2389,7 +2389,7 @@ static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
 			num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
 			key.offset -= btrfs_file_extent_offset(buf, fi);
 			btrfs_init_generic_ref(&generic_ref, action, bytenr,
-					       num_bytes, parent);
+					       num_bytes, parent, ref_root);
 			btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
 					    key.offset, root->root_key.objectid,
 					    for_reloc);
@@ -2402,8 +2402,9 @@ static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
 		} else {
 			bytenr = btrfs_node_blockptr(buf, i);
 			num_bytes = fs_info->nodesize;
+			/* We don't know the owning_root, use 0 */
 			btrfs_init_generic_ref(&generic_ref, action, bytenr,
-					       num_bytes, parent);
+					       num_bytes, parent, 0);
 			btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
 					    root->root_key.objectid, for_reloc);
 			if (inc)
@@ -3220,7 +3221,7 @@ void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 	int ret;
 
 	btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
-			       buf->start, buf->len, parent);
+			       buf->start, buf->len, parent, btrfs_header_owner(buf));
 	btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
 			    root_id, 0, false);
 
@@ -4677,12 +4678,14 @@ int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 				     struct btrfs_key *ins)
 {
 	struct btrfs_ref generic_ref = { 0 };
+	u64 root_objectid = root->root_key.objectid;
+	u64 owning_root = root_objectid;
 
-	BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
+	BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
 
 	btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
-			       ins->objectid, ins->offset, 0);
-	btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner,
+			       ins->objectid, ins->offset, 0, owning_root);
+	btrfs_init_data_ref(&generic_ref, root_objectid, owner,
 			    offset, 0, false);
 	btrfs_ref_tree_mod(root->fs_info, &generic_ref);
 
@@ -4894,7 +4897,7 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 		extent_op->level = level;
 
 		btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
-				       ins.objectid, ins.offset, parent);
+				       ins.objectid, ins.offset, parent, btrfs_header_owner(buf));
 		btrfs_init_tree_ref(&generic_ref, level, root_objectid,
 				    root->root_key.objectid, false);
 		btrfs_ref_tree_mod(fs_info, &generic_ref);
@@ -5315,7 +5318,7 @@ static noinline int do_walk_down(struct btrfs_trans_handle *trans,
 		find_next_key(path, level, &wc->drop_progress);
 
 		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
-				       fs_info->nodesize, parent);
+				       fs_info->nodesize, parent, btrfs_header_owner(next));
 		btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
 				    0, false);
 		ret = btrfs_free_extent(trans, &ref);
fs/btrfs/file.c +5 -5
index fd03e689a6be..83b651d823bb 100644
@@ -373,7 +373,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			if (update_refs && disk_bytenr > 0) {
 				btrfs_init_generic_ref(&ref,
 						BTRFS_ADD_DELAYED_REF,
-						disk_bytenr, num_bytes, 0);
+						disk_bytenr, num_bytes, 0, root->root_key.objectid);
 				btrfs_init_data_ref(&ref,
 						root->root_key.objectid,
 						new_key.objectid,
@@ -463,7 +463,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			} else if (update_refs && disk_bytenr > 0) {
 				btrfs_init_generic_ref(&ref,
 						BTRFS_DROP_DELAYED_REF,
-						disk_bytenr, num_bytes, 0);
+						disk_bytenr, num_bytes, 0, root->root_key.objectid);
 				btrfs_init_data_ref(&ref,
 						root->root_key.objectid,
 						key.objectid,
@@ -745,7 +745,7 @@ int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 		btrfs_mark_buffer_dirty(leaf);
 
 		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
-				       num_bytes, 0);
+				       num_bytes, 0, root->root_key.objectid);
 		btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
 				    orig_offset, 0, false);
 		ret = btrfs_inc_extent_ref(trans, &ref);
@@ -771,7 +771,7 @@ int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 	other_start = end;
 	other_end = 0;
 	btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
-			       num_bytes, 0);
+			       num_bytes, 0, root->root_key.objectid);
 	btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
 			    0, false);
 	if (extent_mergeable(leaf, path->slots[0] + 1,
@@ -2290,7 +2290,7 @@ static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
 
 		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
 				       extent_info->disk_offset,
-				       extent_info->disk_len, 0);
+				       extent_info->disk_len, 0, root->root_key.objectid);
 		ref_offset = extent_info->file_offset - extent_info->data_offset;
 		btrfs_init_data_ref(&ref, root->root_key.objectid,
 				    btrfs_ino(inode), ref_offset, 0, false);
fs/btrfs/inode-item.c +1 -1
index 4c322b720a80..4a56bf679de6 100644
@@ -676,7 +676,7 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 			bytes_deleted += extent_num_bytes;
 
 			btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
-					extent_start, extent_num_bytes, 0);
+					extent_start, extent_num_bytes, 0, root->root_key.objectid);
 			btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
 					control->ino, extent_offset,
 					root->root_key.objectid, false);
fs/btrfs/relocation.c +10 -7
index 9db2e6fa2cb2..3161a48d5970 100644
@@ -1158,7 +1158,7 @@ int replace_file_extents(struct btrfs_trans_handle *trans,
 
 		key.offset -= btrfs_file_extent_offset(leaf, fi);
 		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
-				       num_bytes, parent);
+				       num_bytes, parent, root->root_key.objectid);
 		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
 				    key.objectid, key.offset,
 				    root->root_key.objectid, false);
@@ -1169,7 +1169,7 @@ int replace_file_extents(struct btrfs_trans_handle *trans,
 		}
 
 		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
-				       num_bytes, parent);
+				       num_bytes, parent, root->root_key.objectid);
 		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
 				    key.objectid, key.offset,
 				    root->root_key.objectid, false);
@@ -1382,7 +1382,8 @@ int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
 		btrfs_mark_buffer_dirty(path->nodes[level]);
 
 		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
-				       blocksize, path->nodes[level]->start);
+				       blocksize, path->nodes[level]->start,
+				       src->root_key.objectid);
 		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
 				    0, true);
 		ret = btrfs_inc_extent_ref(trans, &ref);
@@ -1391,7 +1392,7 @@ int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
 			break;
 		}
 		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
-				       blocksize, 0);
+				       blocksize, 0, dest->root_key.objectid);
 		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0,
 				    true);
 		ret = btrfs_inc_extent_ref(trans, &ref);
@@ -1400,8 +1401,9 @@ int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
 			break;
 		}
 
+		/* We don't know the real owning_root, use 0 */
 		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
-				       blocksize, path->nodes[level]->start);
+				       blocksize, path->nodes[level]->start, 0);
 		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
 				    0, true);
 		ret = btrfs_free_extent(trans, &ref);
@@ -1410,8 +1412,9 @@ int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
 			break;
 		}
 
+		/* We don't know the real owning_root, use 0 */
 		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
-				       blocksize, 0);
+				       blocksize, 0, 0);
 		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid,
 				    0, true);
 		ret = btrfs_free_extent(trans, &ref);
@@ -2491,7 +2494,7 @@ static int do_relocation(struct btrfs_trans_handle *trans,
 
 			btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
 					       node->eb->start, blocksize,
-					       upper->eb->start);
+					       upper->eb->start, btrfs_header_owner(upper->eb));
 			btrfs_init_tree_ref(&ref, node->level,
 					    btrfs_header_owner(upper->eb),
 					    root->root_key.objectid, false);
fs/btrfs/tree-log.c +2 -1
index 8ad7e7e38d18..51aaaefaf39d 100644
@@ -767,7 +767,8 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 			} else if (ret == 0) {
 				btrfs_init_generic_ref(&ref,
 						BTRFS_ADD_DELAYED_REF,
-						ins.objectid, ins.offset, 0);
+						ins.objectid, ins.offset, 0,
+						root->root_key.objectid);
 				btrfs_init_data_ref(&ref,
 						root->root_key.objectid,
 						key->objectid, offset, 0, false);
Signature · 1 line
2.41.0
Reviewed-by on v5 9/18
#22 lore raw

Re: [PATCH v5 09/18] btrfs: track owning root in btrfs_ref

Quoted 11 lines · Boris Burkov

While data extents require us to store additional inline refs to track the original owner on free, this information is available implicitly for metadata. It is found in the owner field of the header of the tree block. Even if other trees refer to this block and the original ref goes away, we will not rewrite that header field, so it will reliably give the original owner.

In addition, there is a relocation case where a new data extent needs to have an owning root separate from the referring root wired through delayed refs.

Quoted 5 lines

To use it for recording simple quota deltas, we need to wire this root id through from when we create the delayed ref until we fully process it. Store it in the generic btrfs_ref struct of the delayed ref.

Thanks,

Josef

#23 lore raw

PATCH v5 10/18 btrfs: track original extent owner in head_ref

In reply to Boris Burkov

Simple quotas requires tracking the original creating root of any given extent. This gets complicated when multiple subvolumes create overlapping/contradictory refs in the same transaction. For example, due to modifying or deleting an extent while also snapshotting it.

To resolve this in a general way, take advantage of the fact that we are essentially already tracking this for handling releasing reservations. The head ref coalesces the various refs and uses must_insert_reserved to check if it needs to create an extent/free reservation. Store the ref that set must_insert_reserved as the owning ref on the head ref.

Note that this can result in writing an extent for the very first time with an owner different from its only ref, but it will look the same as if you first created it with the original owning ref, then added the other ref, then removed the owning ref.

---
 fs/btrfs/delayed-ref.c | 20 ++++++++++++++++----
 fs/btrfs/delayed-ref.h |  7 +++++++
 2 files changed, 23 insertions(+), 4 deletions(-)
fs/btrfs/delayed-ref.c +16 -4
index f0bae1e1c455..28ba7a9eb3c3 100644
@@ -623,6 +623,16 @@ static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans,
 	BUG_ON(existing->is_data != update->is_data);
 
 	spin_lock(&existing->lock);
+
+	/*
+	 * When freeing an extent, we may not know the owning root
+	 * when we first create the head_ref. However, some deref before the
+	 * last deref will know it, so we just need to update the head_ref
+	 * accordingly
+	 */
+	if (!existing->owning_root)
+		existing->owning_root = update->owning_root;
+
 	if (update->must_insert_reserved) {
 		/* if the extent was freed and then
 		 * reallocated before the delayed ref
@@ -632,6 +642,7 @@ static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans,
 		 * Set it again here
 		 */
 		existing->must_insert_reserved = update->must_insert_reserved;
+		existing->owning_root = update->owning_root;
 
 		/*
 		 * update the num_bytes so we make sure the accounting
@@ -694,7 +705,7 @@ static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref,
 				  struct btrfs_qgroup_extent_record *qrecord,
 				  u64 bytenr, u64 num_bytes, u64 ref_root,
 				  u64 reserved, int action, bool is_data,
-				  bool is_system)
+				  bool is_system, u64 owning_root)
 {
 	int count_mod = 1;
 	bool must_insert_reserved = false;
@@ -735,6 +746,7 @@ static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref,
 	head_ref->num_bytes = num_bytes;
 	head_ref->ref_mod = count_mod;
 	head_ref->must_insert_reserved = must_insert_reserved;
+	head_ref->owning_root = owning_root;
 	head_ref->is_data = is_data;
 	head_ref->is_system = is_system;
 	head_ref->ref_tree = RB_ROOT_CACHED;
@@ -922,7 +934,7 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 
 	init_delayed_ref_head(head_ref, record, bytenr, num_bytes,
 			      generic_ref->tree_ref.ref_root, 0, action,
-			      false, is_system);
+			      false, is_system, generic_ref->owning_root);
 	head_ref->extent_op = extent_op;
 
 	delayed_refs = &trans->transaction->delayed_refs;
@@ -1014,7 +1026,7 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 	}
 
 	init_delayed_ref_head(head_ref, record, bytenr, num_bytes, ref_root,
-			      reserved, action, true, false);
+			      reserved, action, true, false, generic_ref->owning_root);
 	head_ref->extent_op = NULL;
 
 	delayed_refs = &trans->transaction->delayed_refs;
@@ -1060,7 +1072,7 @@ int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 
 	init_delayed_ref_head(head_ref, NULL, bytenr, num_bytes, 0, 0,
-			      BTRFS_UPDATE_DELAYED_HEAD, false, false);
+			      BTRFS_UPDATE_DELAYED_HEAD, false, false, 0);
 	head_ref->extent_op = extent_op;
 
 	delayed_refs = &trans->transaction->delayed_refs;
fs/btrfs/delayed-ref.h +7 -0
index 0729850a9193..71f0a6e5d583 100644
@@ -117,6 +117,13 @@ struct btrfs_delayed_ref_head {
 	 * the free has happened.
 	 */
 	bool must_insert_reserved;
+
+	/*
+	 * The root which triggered the allocation when
+	 * must_insert_reserved is true
+	 */
+	u64 owning_root;
+
 	bool is_data;
 	bool is_system;
 	bool processing;
Signature · 1 line
2.41.0
Reviewed-by on v5 10/18
#24 lore raw

Re: [PATCH v5 10/18] btrfs: track original extent owner in head_ref

Quoted 11 lines · Boris Burkov

Simple quotas requires tracking the original creating root of any given extent. This gets complicated when multiple subvolumes create overlapping/contradictory refs in the same transaction. For example, due to modifying or deleting an extent while also snapshotting it.

To resolve this in a general way, take advantage of the fact that we are essentially already tracking this for handling releasing reservations. The head ref coalesces the various refs and uses must_insert_reserved to check if it needs to create an extent/free reservation. Store the ref that set must_insert_reserved as the owning ref on the head ref.

Quoted 6 lines

Note that this can result in writing an extent for the very first time with an owner different from its only ref, but it will look the same as if you first created it with the original owning ref, then added the other ref, then removed the owning ref.

Thanks,

Josef

#25 lore raw

Re: [PATCH v5 10/18] btrfs: track original extent owner in head_ref

In reply to Boris Burkov
Quoted 27 lines · Boris Burkov

Simple quotas requires tracking the original creating root of any given extent. This gets complicated when multiple subvolumes create overlapping/contradictory refs in the same transaction. For example, due to modifying or deleting an extent while also snapshotting it.

To resolve this in a general way, take advantage of the fact that we are essentially already tracking this for handling releasing reservations. The head ref coalesces the various refs and uses must_insert_reserved to check if it needs to create an extent/free reservation. Store the ref that set must_insert_reserved as the owning ref on the head ref.

Note that this can result in writing an extent for the very first time with an owner different from its only ref, but it will look the same as if you first created it with the original owning ref, then added the other ref, then removed the owning ref.

---
 fs/btrfs/delayed-ref.c | 20 ++++++++++++++++----
 fs/btrfs/delayed-ref.h |  7 +++++++
 2 files changed, 23 insertions(+), 4 deletions(-)
fs/btrfs/delayed-ref.c +0 -0
index f0bae1e1c455..28ba7a9eb3c3 100644

fs/btrfs/delayed-ref.c static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans, · line 631

Quoted 9 lines
fs/btrfs/delayed-ref.c +6 -0
@@ -623,6 +623,16 @@ static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans,
 	BUG_ON(existing->is_data != update->is_data);
 
 	spin_lock(&existing->lock);
+
+	/*
+	 * When freeing an extent, we may not know the owning root
+	 * when we first create the head_ref. However, some deref before the
+	 * last deref will know it, so we just need to update the head_ref
+	 * accordingly

We now write comments as full sentences so please use a "." at the end.

Quoted 64 lines
fs/btrfs/delayed-ref.c +10 -4
+	 */
+	if (!existing->owning_root)
+		existing->owning_root = update->owning_root;
+
 	if (update->must_insert_reserved) {
 		/* if the extent was freed and then
 		 * reallocated before the delayed ref
@@ -632,6 +642,7 @@ static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans,
 		 * Set it again here
 		 */
 		existing->must_insert_reserved = update->must_insert_reserved;
+		existing->owning_root = update->owning_root;
 
 		/*
 		 * update the num_bytes so we make sure the accounting
@@ -694,7 +705,7 @@ static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref,
 				  struct btrfs_qgroup_extent_record *qrecord,
 				  u64 bytenr, u64 num_bytes, u64 ref_root,
 				  u64 reserved, int action, bool is_data,
-				  bool is_system)
+				  bool is_system, u64 owning_root)
 {
 	int count_mod = 1;
 	bool must_insert_reserved = false;
@@ -735,6 +746,7 @@ static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref,
 	head_ref->num_bytes = num_bytes;
 	head_ref->ref_mod = count_mod;
 	head_ref->must_insert_reserved = must_insert_reserved;
+	head_ref->owning_root = owning_root;
 	head_ref->is_data = is_data;
 	head_ref->is_system = is_system;
 	head_ref->ref_tree = RB_ROOT_CACHED;
@@ -922,7 +934,7 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 
 	init_delayed_ref_head(head_ref, record, bytenr, num_bytes,
 			      generic_ref->tree_ref.ref_root, 0, action,
-			      false, is_system);
+			      false, is_system, generic_ref->owning_root);
 	head_ref->extent_op = extent_op;
 
 	delayed_refs = &trans->transaction->delayed_refs;
@@ -1014,7 +1026,7 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 	}
 
 	init_delayed_ref_head(head_ref, record, bytenr, num_bytes, ref_root,
-			      reserved, action, true, false);
+			      reserved, action, true, false, generic_ref->owning_root);
 	head_ref->extent_op = NULL;
 
 	delayed_refs = &trans->transaction->delayed_refs;
@@ -1060,7 +1072,7 @@ int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 
 	init_delayed_ref_head(head_ref, NULL, bytenr, num_bytes, 0, 0,
-			      BTRFS_UPDATE_DELAYED_HEAD, false, false);
+			      BTRFS_UPDATE_DELAYED_HEAD, false, false, 0);
 	head_ref->extent_op = extent_op;
 
 	delayed_refs = &trans->transaction->delayed_refs;
fs/btrfs/delayed-ref.h +0 -0
index 0729850a9193..71f0a6e5d583 100644

fs/btrfs/delayed-ref.h struct btrfs_delayed_ref_head { · line 125

Quoted 9 lines
fs/btrfs/delayed-ref.h +6 -0
@@ -117,6 +117,13 @@ struct btrfs_delayed_ref_head {
 	 * the free has happened.
 	 */
 	bool must_insert_reserved;
+
+	/*
+	 * The root which triggered the allocation when
+	 * must_insert_reserved is true
+	 */
+	u64 owning_root;

Please reorder this so it's not among bools, this would create a gap of 7 bytes before that.

Quoted 6 lines
fs/btrfs/delayed-ref.h +1 -0
+
 	bool is_data;
 	bool is_system;
 	bool processing;
Signature · 1 line
2.41.0
#26 lore raw

PATCH v5 11/18 btrfs: new inline ref storing owning subvol of data extents

In reply to Boris Burkov

In order to implement simple quota groups, we need to be able to associate a data extent with the subvolume that created it. Once you account for reflink, this information cannot be recovered without explicitly storing it. Options for storing it are:
- a new key/item
- a new extent inline ref item

The former is backwards compatible, but wastes space, the latter is incompat, but is efficient in space and reuses the existing inline ref machinery, while only abusing it a tiny amount -- specifically, the new item is not a ref, per-se.

---
 fs/btrfs/accessors.h            |  4 +++
 fs/btrfs/backref.c              |  3 ++
 fs/btrfs/extent-tree.c          | 56 ++++++++++++++++++++++++++-------
 fs/btrfs/print-tree.c           | 12 +++++++
 fs/btrfs/ref-verify.c           |  3 ++
 fs/btrfs/tree-checker.c         |  3 ++
 include/uapi/linux/btrfs_tree.h |  6 ++++
 7 files changed, 76 insertions(+), 11 deletions(-)
fs/btrfs/accessors.h +4 -0
index 8cfc8214109c..a23045c05937 100644
@@ -349,6 +349,8 @@ BTRFS_SETGET_FUNCS(extent_data_ref_count, struct btrfs_extent_data_ref, count, 3
 
 BTRFS_SETGET_FUNCS(shared_data_ref_count, struct btrfs_shared_data_ref, count, 32);
 
+BTRFS_SETGET_FUNCS(extent_owner_ref_root_id, struct btrfs_extent_owner_ref, root_id, 64);
+
 BTRFS_SETGET_FUNCS(extent_inline_ref_type, struct btrfs_extent_inline_ref,
 		   type, 8);
 BTRFS_SETGET_FUNCS(extent_inline_ref_offset, struct btrfs_extent_inline_ref,
@@ -365,6 +367,8 @@ static inline u32 btrfs_extent_inline_ref_size(int type)
 	if (type == BTRFS_EXTENT_DATA_REF_KEY)
 		return sizeof(struct btrfs_extent_data_ref) +
 		       offsetof(struct btrfs_extent_inline_ref, offset);
+	if (type == BTRFS_EXTENT_OWNER_REF_KEY)
+		return sizeof(struct btrfs_extent_inline_ref);
 	return 0;
 }
 
fs/btrfs/backref.c +3 -0
index 79336fa853db..d5bb6a880713 100644
@@ -1129,6 +1129,9 @@ static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx,
 						       count, sc, GFP_NOFS);
 			break;
 		}
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			WARN_ON(!btrfs_fs_incompat(ctx->fs_info, SIMPLE_QUOTA));
+			break;
 		default:
 			WARN_ON(1);
 		}
fs/btrfs/extent-tree.c +45 -11

149 lines not on this page · raw message

fs/btrfs/print-tree.c +12 -0
index aa06d9ca911d..3fac15ce0db0 100644
@@ -80,12 +80,20 @@ static void print_extent_data_ref(const struct extent_buffer *eb,
 	       btrfs_extent_data_ref_count(eb, ref));
 }
 
+static void print_extent_owner_ref(const struct extent_buffer *eb,
+				   struct btrfs_extent_owner_ref *ref)
+{
+	WARN_ON(!btrfs_fs_incompat(eb->fs_info, SIMPLE_QUOTA));
+	pr_cont("extent data owner root %llu\n", btrfs_extent_owner_ref_root_id(eb, ref));
+}
+
 static void print_extent_item(const struct extent_buffer *eb, int slot, int type)
 {
 	struct btrfs_extent_item *ei;
 	struct btrfs_extent_inline_ref *iref;
 	struct btrfs_extent_data_ref *dref;
 	struct btrfs_shared_data_ref *sref;
+	struct btrfs_extent_owner_ref *oref;
 	struct btrfs_disk_key key;
 	unsigned long end;
 	unsigned long ptr;
@@ -159,6 +167,10 @@ static void print_extent_item(const struct extent_buffer *eb, int slot, int type
 			"\t\t\t(parent %llu not aligned to sectorsize %u)\n",
 				     offset, eb->fs_info->sectorsize);
 			break;
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
+			print_extent_owner_ref(eb, oref);
+			break;
 		default:
 			pr_cont("(extent %llu has INVALID ref type %d)\n",
 				  eb->start, type);
fs/btrfs/ref-verify.c +3 -0
index b7b3bd86f5e2..c0660233feb4 100644
@@ -485,6 +485,9 @@ static int process_extent_item(struct btrfs_fs_info *fs_info,
 			ret = add_shared_data_ref(fs_info, offset, count,
 						  key->objectid, key->offset);
 			break;
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
+			break;
 		default:
 			btrfs_err(fs_info, "invalid key type in iref");
 			ret = -EINVAL;
fs/btrfs/tree-checker.c +3 -0
index 038dfa8f1788..72d29ab74a01 100644
@@ -1451,6 +1451,9 @@ static int check_extent_item(struct extent_buffer *leaf,
 			}
 			inline_refs += btrfs_shared_data_ref_count(leaf, sref);
 			break;
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
+			break;
 		default:
 			extent_err(leaf, slot, "unknown inline ref type: %u",
 				   inline_type);
include/uapi/linux/btrfs_tree.h +6 -0
index 47aca414a41b..eacb26caf3c6 100644
@@ -226,6 +226,8 @@
 
 #define BTRFS_SHARED_DATA_REF_KEY	184
 
+#define BTRFS_EXTENT_OWNER_REF_KEY	190
+
 /*
  * block groups give us hints into the extent allocation trees.  Which
  * blocks are free etc etc
@@ -783,6 +785,10 @@ struct btrfs_shared_data_ref {
 	__le32 count;
 } __attribute__ ((__packed__));
 
+struct btrfs_extent_owner_ref {
+	__le64 root_id;
+} __attribute__ ((__packed__));
+
 struct btrfs_extent_inline_ref {
 	__u8 type;
 	__le64 offset;
Signature · 1 line
2.41.0
Reviewed-by on v5 11/18
#27 lore raw

Re: [PATCH v5 11/18] btrfs: new inline ref storing owning subvol of data extents

Quoted 7 lines · Boris Burkov

In order to implement simple quota groups, we need to be able to associate a data extent with the subvolume that created it. Once you account for reflink, this information cannot be recovered without explicitly storing it. Options for storing it are:
- a new key/item
- a new extent inline ref item

Quoted 6 lines

The former is backwards compatible, but wastes space, the latter is incompat, but is efficient in space and reuses the existing inline ref machinery, while only abusing it a tiny amount -- specifically, the new item is not a ref, per-se.

Thanks,

Josef

#28 lore raw

Re: [PATCH v5 11/18] btrfs: new inline ref storing owning subvol of data extents

In reply to Boris Burkov
Quoted 50 lines · Boris Burkov

In order to implement simple quota groups, we need to be able to associate a data extent with the subvolume that created it. Once you account for reflink, this information cannot be recovered without explicitly storing it. Options for storing it are:
- a new key/item
- a new extent inline ref item

The former is backwards compatible, but wastes space, the latter is incompat, but is efficient in space and reuses the existing inline ref machinery, while only abusing it a tiny amount -- specifically, the new item is not a ref, per-se.

---
 fs/btrfs/accessors.h            |  4 +++
 fs/btrfs/backref.c              |  3 ++
 fs/btrfs/extent-tree.c          | 56 ++++++++++++++++++++++++++-------
 fs/btrfs/print-tree.c           | 12 +++++++
 fs/btrfs/ref-verify.c           |  3 ++
 fs/btrfs/tree-checker.c         |  3 ++
 include/uapi/linux/btrfs_tree.h |  6 ++++
 7 files changed, 76 insertions(+), 11 deletions(-)
fs/btrfs/accessors.h +4 -0
index 8cfc8214109c..a23045c05937 100644
@@ -349,6 +349,8 @@ BTRFS_SETGET_FUNCS(extent_data_ref_count, struct btrfs_extent_data_ref, count, 3
 
 BTRFS_SETGET_FUNCS(shared_data_ref_count, struct btrfs_shared_data_ref, count, 32);
 
+BTRFS_SETGET_FUNCS(extent_owner_ref_root_id, struct btrfs_extent_owner_ref, root_id, 64);
+
 BTRFS_SETGET_FUNCS(extent_inline_ref_type, struct btrfs_extent_inline_ref,
 		   type, 8);
 BTRFS_SETGET_FUNCS(extent_inline_ref_offset, struct btrfs_extent_inline_ref,
@@ -365,6 +367,8 @@ static inline u32 btrfs_extent_inline_ref_size(int type)
 	if (type == BTRFS_EXTENT_DATA_REF_KEY)
 		return sizeof(struct btrfs_extent_data_ref) +
 		       offsetof(struct btrfs_extent_inline_ref, offset);
+	if (type == BTRFS_EXTENT_OWNER_REF_KEY)
+		return sizeof(struct btrfs_extent_inline_ref);
 	return 0;
 }
 
fs/btrfs/backref.c +0 -0
index 79336fa853db..d5bb6a880713 100644

fs/btrfs/backref.c static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx, · line 1133

Quoted 5 lines
fs/btrfs/backref.c +2 -0
@@ -1129,6 +1129,9 @@ static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx,
 						       count, sc, GFP_NOFS);
 			break;
 		}
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			WARN_ON(!btrfs_fs_incompat(ctx->fs_info, SIMPLE_QUOTA));

Please turn this to an ASSERT, we must catch this during development but hte warning makes no sense for end users.

Quoted 9 lines
fs/btrfs/backref.c +1 -0
+			break;
 		default:
 			WARN_ON(1);
 		}
fs/btrfs/extent-tree.c +0 -0
index 4f0115553cd3..c6d537bf5ad4 100644

fs/btrfs/extent-tree.c int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, · line 349

Quoted 8 lines
fs/btrfs/extent-tree.c +2 -0
@@ -342,9 +342,13 @@ int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
 				     struct btrfs_extent_inline_ref *iref,
 				     enum btrfs_inline_ref_type is_data)
 {
+	struct btrfs_fs_info *fs_info = eb->fs_info;
 	int type = btrfs_extent_inline_ref_type(eb, iref);
 	u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
 
+	if (type == BTRFS_EXTENT_OWNER_REF_KEY && btrfs_fs_incompat(fs_info, SIMPLE_QUOTA))

I think the conditions should be swapped, first you check that squotas are enabled then that the type is the one you look for.

fs/btrfs/extent-tree.c int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, · line 360

Quoted 11 lines
fs/btrfs/extent-tree.c +3 -1
+		return type;
+
 	if (type == BTRFS_TREE_BLOCK_REF_KEY ||
 	    type == BTRFS_SHARED_BLOCK_REF_KEY ||
 	    type == BTRFS_SHARED_DATA_REF_KEY ||
@@ -353,26 +357,25 @@ int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
 			if (type == BTRFS_TREE_BLOCK_REF_KEY)
 				return type;
 			if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
-				ASSERT(eb->fs_info);
+				ASSERT(fs_info);

I'm not sure what's the point of this assertion, each eb is created with a valid fs_info. It's not in your new code so you can keep it but would be good to remove it evenntually.

Quoted 5 lines
fs/btrfs/extent-tree.c +0 -1
 				/*
 				 * Every shared one has parent tree block,
 				 * which must be aligned to sector size.
 				 */
-				if (offset &&

fs/btrfs/extent-tree.c int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, · line 372

Quoted 10 lines
fs/btrfs/extent-tree.c +2 -2
-				    IS_ALIGNED(offset, eb->fs_info->sectorsize))
+				if (offset && IS_ALIGNED(offset, fs_info->sectorsize))
 					return type;
 			}
 		} else if (is_data == BTRFS_REF_TYPE_DATA) {
 			if (type == BTRFS_EXTENT_DATA_REF_KEY)
 				return type;
 			if (type == BTRFS_SHARED_DATA_REF_KEY) {
-				ASSERT(eb->fs_info);
+				ASSERT(fs_info);

Same.

Quoted 20 lines
fs/btrfs/extent-tree.c +2 -2
 				/*
 				 * Every shared one has parent tree block,
 				 * which must be aligned to sector size.
 				 */
 				if (offset &&
-				    IS_ALIGNED(offset, eb->fs_info->sectorsize))
+				    IS_ALIGNED(offset, fs_info->sectorsize))
 					return type;
 			}
 		} else {
@@ -382,7 +385,7 @@ int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
 	}
 
 	btrfs_print_leaf(eb);
-	btrfs_err(eb->fs_info,
+	btrfs_err(fs_info,
 		  "eb %llu iref 0x%lx invalid extent inline ref type %d",
 		  eb->start, (unsigned long)iref, type);
 	WARN_ON(1);

fs/btrfs/extent-tree.c int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, · line 898

Quoted 5 lines
fs/btrfs/extent-tree.c +2 -0
@@ -891,6 +894,11 @@ int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
 		}
 		iref = (struct btrfs_extent_inline_ref *)ptr;
 		type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
+		if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
+			WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));

ASSERT()

Quoted 58 lines
fs/btrfs/extent-tree.c +20 -4
+			ptr += btrfs_extent_inline_ref_size(type);
+			continue;
+		}
 		if (type == BTRFS_REF_TYPE_INVALID) {
 			err = -EUCLEAN;
 			goto out;
@@ -1684,6 +1692,8 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 		 node->type == BTRFS_SHARED_DATA_REF_KEY)
 		ret = run_delayed_data_ref(trans, node, extent_op,
 					   insert_reserved);
+	else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY)
+		ret = 0;
 	else
 		BUG();
 	if (ret && insert_reserved)
@@ -2250,6 +2260,7 @@ static noinline int check_committed_ref(struct btrfs_root *root,
 	struct btrfs_extent_item *ei;
 	struct btrfs_key key;
 	u32 item_size;
+	u32 expected_size;
 	int type;
 	int ret;
 
@@ -2276,10 +2287,22 @@ static noinline int check_committed_ref(struct btrfs_root *root,
 	ret = 1;
 	item_size = btrfs_item_size(leaf, path->slots[0]);
 	ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
+	expected_size = sizeof(*ei) + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY);
+
+	/* No inline refs; we need to bail before checking for owner ref */
+	if (item_size == sizeof(*ei))
+		goto out;
+
+	/* Check for an owner ref; skip over it to the real inline refs */
+	iref = (struct btrfs_extent_inline_ref *)(ei + 1);
+	type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
+	if (btrfs_fs_incompat(fs_info, SIMPLE_QUOTA) && type == BTRFS_EXTENT_OWNER_REF_KEY) {
+		expected_size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
+		iref = (struct btrfs_extent_inline_ref *)(iref + 1);
+	}
 
 	/* If extent item has more than 1 inline ref then it's shared */
-	if (item_size != sizeof(*ei) +
-	    btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
+	if (item_size != expected_size)
 		goto out;
 
 	/*
@@ -2291,8 +2314,6 @@ static noinline int check_committed_ref(struct btrfs_root *root,
 	     btrfs_root_last_snapshot(&root->root_item)))
 		goto out;
 
-	iref = (struct btrfs_extent_inline_ref *)(ei + 1);
-
 	/* If this extent has SHARED_DATA_REF then it's shared */
 	type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
 	if (type != BTRFS_EXTENT_DATA_REF_KEY)

fs/btrfs/extent-tree.c static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, · line 4573

Quoted 10 lines
fs/btrfs/extent-tree.c +2 -0
@@ -4543,18 +4564,23 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_root *extent_root;
 	int ret;
 	struct btrfs_extent_item *extent_item;
+	struct btrfs_extent_owner_ref *oref;
 	struct btrfs_extent_inline_ref *iref;
 	struct btrfs_path *path;
 	struct extent_buffer *leaf;
 	int type;
 	u32 size;
+	bool simple_quota = btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE;
	const bool simple_quota = (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE);
Quoted 15 lines
fs/btrfs/extent-tree.c +4 -1
 	if (parent > 0)
 		type = BTRFS_SHARED_DATA_REF_KEY;
 	else
 		type = BTRFS_EXTENT_DATA_REF_KEY;
 
-	size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
+	size = sizeof(*extent_item);
+	if (simple_quota)
+		size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
+	size += btrfs_extent_inline_ref_size(type);
 
 	path = btrfs_alloc_path();
 	if (!path)

fs/btrfs/extent-tree.c static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, · line 4604

Quoted 4 lines
fs/btrfs/extent-tree.c +1 -0
@@ -4575,8 +4601,16 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 	btrfs_set_extent_flags(leaf, extent_item,
 			       flags | BTRFS_EXTENT_FLAG_DATA);
 
+

Stray newline

Quoted 17 lines
fs/btrfs/extent-tree.c +7 -0
 	iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
+	if (simple_quota) {
+		btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY);
+		oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
+		btrfs_set_extent_owner_ref_root_id(leaf, oref, root_objectid);
+		iref = (struct btrfs_extent_inline_ref *)(oref + 1);
+	}
 	btrfs_set_extent_inline_ref_type(leaf, iref, type);
+
 	if (parent > 0) {
 		struct btrfs_shared_data_ref *ref;
 		ref = (struct btrfs_shared_data_ref *)(iref + 1);
fs/btrfs/print-tree.c +0 -0
index aa06d9ca911d..3fac15ce0db0 100644

fs/btrfs/print-tree.c static void print_extent_data_ref(const struct extent_buffer *eb, · line 84

Quoted 5 lines
fs/btrfs/print-tree.c +2 -0
@@ -80,12 +80,20 @@ static void print_extent_data_ref(const struct extent_buffer *eb,
 	       btrfs_extent_data_ref_count(eb, ref));
 }
 
+static void print_extent_owner_ref(const struct extent_buffer *eb,
+				   struct btrfs_extent_owner_ref *ref)

ref type can be also made const

fs/btrfs/print-tree.c static void print_extent_data_ref(const struct extent_buffer *eb, · line 86

Quoted 2 lines
fs/btrfs/print-tree.c +2 -0
+{
+	WARN_ON(!btrfs_fs_incompat(eb->fs_info, SIMPLE_QUOTA));

ASSERT()

Quoted 57 lines
fs/btrfs/print-tree.c +8 -0
+	pr_cont("extent data owner root %llu\n", btrfs_extent_owner_ref_root_id(eb, ref));
+}
+
 static void print_extent_item(const struct extent_buffer *eb, int slot, int type)
 {
 	struct btrfs_extent_item *ei;
 	struct btrfs_extent_inline_ref *iref;
 	struct btrfs_extent_data_ref *dref;
 	struct btrfs_shared_data_ref *sref;
+	struct btrfs_extent_owner_ref *oref;
 	struct btrfs_disk_key key;
 	unsigned long end;
 	unsigned long ptr;
@@ -159,6 +167,10 @@ static void print_extent_item(const struct extent_buffer *eb, int slot, int type
 			"\t\t\t(parent %llu not aligned to sectorsize %u)\n",
 				     offset, eb->fs_info->sectorsize);
 			break;
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
+			print_extent_owner_ref(eb, oref);
+			break;
 		default:
 			pr_cont("(extent %llu has INVALID ref type %d)\n",
 				  eb->start, type);
fs/btrfs/ref-verify.c +3 -0
index b7b3bd86f5e2..c0660233feb4 100644
@@ -485,6 +485,9 @@ static int process_extent_item(struct btrfs_fs_info *fs_info,
 			ret = add_shared_data_ref(fs_info, offset, count,
 						  key->objectid, key->offset);
 			break;
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
+			break;
 		default:
 			btrfs_err(fs_info, "invalid key type in iref");
 			ret = -EINVAL;
fs/btrfs/tree-checker.c +3 -0
index 038dfa8f1788..72d29ab74a01 100644
@@ -1451,6 +1451,9 @@ static int check_extent_item(struct extent_buffer *leaf,
 			}
 			inline_refs += btrfs_shared_data_ref_count(leaf, sref);
 			break;
+		case BTRFS_EXTENT_OWNER_REF_KEY:
+			WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
+			break;
 		default:
 			extent_err(leaf, slot, "unknown inline ref type: %u",
 				   inline_type);
include/uapi/linux/btrfs_tree.h +0 -0
index 47aca414a41b..eacb26caf3c6 100644

include/uapi/linux/btrfs_tree.h line 229

Quoted 4 lines
include/uapi/linux/btrfs_tree.h +1 -0
@@ -226,6 +226,8 @@
 
 #define BTRFS_SHARED_DATA_REF_KEY	184
 
+#define BTRFS_EXTENT_OWNER_REF_KEY	190

Please add some description of the new key and use number 188 so it leaves roughly equal space around it.

Quoted 17 lines
include/uapi/linux/btrfs_tree.h +5 -0
+
 /*
  * block groups give us hints into the extent allocation trees.  Which
  * blocks are free etc etc
@@ -783,6 +785,10 @@ struct btrfs_shared_data_ref {
 	__le32 count;
 } __attribute__ ((__packed__));
 
+struct btrfs_extent_owner_ref {
+	__le64 root_id;
+} __attribute__ ((__packed__));
+
 struct btrfs_extent_inline_ref {
 	__u8 type;
 	__le64 offset;
Signature · 1 line
2.41.0
#29 lore raw

PATCH v5 12/18 btrfs: inline owner ref lookup helper

In reply to Boris Burkov

Inline ref parsing is a bit tricky and relies on a decent amount of implicit information, so I think it is beneficial to have a helper function for reading the owner ref, if only to "document" the format, along with the write path.

The main subtlety of note which I was missing by open-coding this was that it is important to check whether or not inline refs are present *at all*. i.e., if we are writing out a new extent under squotas, we will always use a big enough item for the inline ref and have it. However, it is possible that some random item predating squotas will not have any inline refs. In that case, trying to read the "type" field of the first inline ref will just be reading garbage in the form of whatever is in the next item.

This will be used by the extent free-ing path, which looks up data extent owners as well as a relocation path which needs to grab the owner before relocating an extent.

---
 fs/btrfs/extent-tree.c | 51 ++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/extent-tree.h |  3 +++
 2 files changed, 54 insertions(+)
fs/btrfs/extent-tree.c +51 -0
index c6d537bf5ad4..09fb321fa560 100644
@@ -2805,6 +2805,57 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
 	return 0;
 }
 
+/*
+ * Helper to parse an extent item's inline extents looking for a simple
+ * quotas owner ref.
+ *
+ * @fs_info  - the btrfs_fs_info for this mount
+ * @leaf     - a leaf in the extent tree containing the extent item
+ * @slot     - the slot in the leaf where the extent item is found
+ *
+ * Returns the objectid of the root that originally allocated the extent item
+ * if the inline owner ref is expected and present, otherwise 0.
+ *
+ * If an extent item has an owner ref item, it will be the first
+ * inline ref item. Therefore the logic is to check whether there are
+ * any inline ref items, then check the type of the first one.
+ *
+ */
+u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
+				struct extent_buffer *leaf,
+				int slot)
+{
+	struct btrfs_extent_item *ei;
+	struct btrfs_extent_inline_ref *iref;
+	struct btrfs_extent_owner_ref *oref;
+	unsigned long ptr;
+	unsigned long end;
+	int type;
+
+	if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA))
+		return 0;
+
+	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
+	ptr = (unsigned long)(ei + 1);
+	end = (unsigned long)ei + btrfs_item_size(leaf, slot);
+
+	/* No inline ref items of any kind, can't check type */
+	if (ptr == end)
+		return 0;
+
+	iref = (struct btrfs_extent_inline_ref *)ptr;
+	type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
+
+	/* We found an owner ref, get the root out of it */
+	if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
+		oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
+		return btrfs_extent_owner_ref_root_id(leaf, oref);
+	}
+
+	/* We have inline refs, but not an owner ref */
+	return 0;
+}
+
 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
 				     u64 bytenr, u64 num_bytes, bool is_data)
 {
fs/btrfs/extent-tree.h +3 -0
index b9e148adcd28..7c27652880a2 100644
@@ -141,6 +141,9 @@ int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
 				struct extent_buffer *eb, u64 flags);
 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref);
 
+u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
+				struct extent_buffer *leaf,
+				int slot);
 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
 			       u64 start, u64 len, int delalloc);
 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, u64 len);
Signature · 1 line
2.41.0
#30 lore raw

Re: [PATCH v5 12/18] btrfs: inline owner ref lookup helper

Quoted 30 lines · Boris Burkov

Inline ref parsing is a bit tricky and relies on a decent amount of implicit information, so I think it is beneficial to have a helper function for reading the owner ref, if only to "document" the format, along with the write path.

The main subtlety of note which I was missing by open-coding this was that it is important to check whether or not inline refs are present *at all*. i.e., if we are writing out a new extent under squotas, we will always use a big enough item for the inline ref and have it. However, it is possible that some random item predating squotas will not have any inline refs. In that case, trying to read the "type" field of the first inline ref will just be reading garbage in the form of whatever is in the next item.

This will be used by the extent free-ing path, which looks up data extent owners as well as a relocation path which needs to grab the owner before relocating an extent.

---
 fs/btrfs/extent-tree.c | 51 ++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/extent-tree.h |  3 +++
 2 files changed, 54 insertions(+)
fs/btrfs/extent-tree.c +0 -0
index c6d537bf5ad4..09fb321fa560 100644

fs/btrfs/extent-tree.c int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) · line 2810

Quoted 6 lines
fs/btrfs/extent-tree.c +3 -0
@@ -2805,6 +2805,57 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
 	return 0;
 }
 
+/*
+ * Helper to parse an extent item's inline extents looking for a simple
+ * quotas owner ref.

"Parse an extent item's ..."

fs/btrfs/extent-tree.c int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) · line 2812

Quoted 2 lines
fs/btrfs/extent-tree.c +2 -0
+ *
+ * @fs_info  - the btrfs_fs_info for this mount

The argument description format is " @fs_info:"

fs/btrfs/extent-tree.c int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) · line 2821

Quoted 9 lines
fs/btrfs/extent-tree.c +9 -0
+ * @leaf     - a leaf in the extent tree containing the extent item
+ * @slot     - the slot in the leaf where the extent item is found
+ *
+ * Returns the objectid of the root that originally allocated the extent item
+ * if the inline owner ref is expected and present, otherwise 0.
+ *
+ * If an extent item has an owner ref item, it will be the first
+ * inline ref item. Therefore the logic is to check whether there are
+ * any inline ref items, then check the type of the first one.

Please format comments so they use the 80 columns with acceptable overflows 3-4 chars if it makes it look nicer.

Quoted 56 lines
fs/btrfs/extent-tree.c +37 -0
+ *
+ */
+u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
+				struct extent_buffer *leaf,
+				int slot)
+{
+	struct btrfs_extent_item *ei;
+	struct btrfs_extent_inline_ref *iref;
+	struct btrfs_extent_owner_ref *oref;
+	unsigned long ptr;
+	unsigned long end;
+	int type;
+
+	if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA))
+		return 0;
+
+	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
+	ptr = (unsigned long)(ei + 1);
+	end = (unsigned long)ei + btrfs_item_size(leaf, slot);
+
+	/* No inline ref items of any kind, can't check type */
+	if (ptr == end)
+		return 0;
+
+	iref = (struct btrfs_extent_inline_ref *)ptr;
+	type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
+
+	/* We found an owner ref, get the root out of it */
+	if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
+		oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
+		return btrfs_extent_owner_ref_root_id(leaf, oref);
+	}
+
+	/* We have inline refs, but not an owner ref */
+	return 0;
+}
+
 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
 				     u64 bytenr, u64 num_bytes, bool is_data)
 {
fs/btrfs/extent-tree.h +3 -0
index b9e148adcd28..7c27652880a2 100644
@@ -141,6 +141,9 @@ int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
 				struct extent_buffer *eb, u64 flags);
 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref);
 
+u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
+				struct extent_buffer *leaf,
+				int slot);
 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
 			       u64 start, u64 len, int delalloc);
 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, u64 len);
Signature · 1 line
2.41.0
#31 lore raw

PATCH v5 13/18 btrfs: record simple quota deltas

In reply to Boris Burkov

At the moment that we run delayed refs, we make the final ref-count based decision on creating/removing extent (and metadata) items. Therefore, it is exactly the spot to hook up simple quotas.

There are a few important subtleties to the fields we must collect to accurately track simple quotas, particularly when removing an extent. When removing a data extent, the ref could be in any tree (due to reflink, for example) and so we need to recover the owning root id from the owner ref item. When removing a metadata extent, we know the owning root from the owner field in the header when we create the delayed ref, so we can recover it from there.

We must also be careful to handle reservations properly to not leaked reserved space. The happy path is freeing the reservation when the simple quota delta runs on a data extent. If that doesn't happen, due to refs canceling out or some error, the ref head already has the must_insert_reserved machinery to handle this, so we piggy back on that and use it to clean up the reserved data.

---
 fs/btrfs/delayed-ref.c |  1 +
 fs/btrfs/delayed-ref.h |  6 +++
 fs/btrfs/extent-tree.c | 85 +++++++++++++++++++++++++++++++++++++-----
 3 files changed, 82 insertions(+), 10 deletions(-)
fs/btrfs/delayed-ref.c +1 -0
index 28ba7a9eb3c3..874c1853d9b1 100644
@@ -745,6 +745,7 @@ static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref,
 	head_ref->bytenr = bytenr;
 	head_ref->num_bytes = num_bytes;
 	head_ref->ref_mod = count_mod;
+	head_ref->reserved_bytes = reserved;
 	head_ref->must_insert_reserved = must_insert_reserved;
 	head_ref->owning_root = owning_root;
 	head_ref->is_data = is_data;
fs/btrfs/delayed-ref.h +6 -0
index 71f0a6e5d583..221d400dd88f 100644
@@ -104,6 +104,12 @@ struct btrfs_delayed_ref_head {
 	 */
 	int ref_mod;
 
+	/*
+	 * Track reserved bytes when setting must_insert_reserved.
+	 * On success or cleanup, we will need to free the reservation.
+	 */
+	u64 reserved_bytes;
+
 	/*
 	 * when a new extent is allocated, it is just reserved in memory
 	 * The actual extent isn't inserted into the extent allocation tree
fs/btrfs/extent-tree.c +75 -10

221 lines not on this page · raw message

Signature · 1 line
2.41.0
Reviewed-by on v5 13/18
#32 lore raw

Re: [PATCH v5 13/18] btrfs: record simple quota deltas

Quoted 12 lines · Boris Burkov

At the moment that we run delayed refs, we make the final ref-count based decision on creating/removing extent (and metadata) items. Therefore, it is exactly the spot to hook up simple quotas.

There are a few important subtleties to the fields we must collect to accurately track simple quotas, particularly when removing an extent. When removing a data extent, the ref could be in any tree (due to reflink, for example) and so we need to recover the owning root id from the owner ref item. When removing a metadata extent, we know the owning root from the owner field in the header when we create the delayed ref, so we can recover it from there.

Quoted 8 lines

We must also be careful to handle reservations properly to not leaked reserved space. The happy path is freeing the reservation when the simple quota delta runs on a data extent. If that doesn't happen, due to refs canceling out or some error, the ref head already has the must_insert_reserved machinery to handle this, so we piggy back on that and use it to clean up the reserved data.

Thanks,

Josef

#33 lore raw

Re: [PATCH v5 13/18] btrfs: record simple quota deltas

In reply to Boris Burkov
Quoted 169 lines · Boris Burkov

At the moment that we run delayed refs, we make the final ref-count based decision on creating/removing extent (and metadata) items. Therefore, it is exactly the spot to hook up simple quotas.

There are a few important subtleties to the fields we must collect to accurately track simple quotas, particularly when removing an extent. When removing a data extent, the ref could be in any tree (due to reflink, for example) and so we need to recover the owning root id from the owner ref item. When removing a metadata extent, we know the owning root from the owner field in the header when we create the delayed ref, so we can recover it from there.

We must also be careful to handle reservations properly to not leaked reserved space. The happy path is freeing the reservation when the simple quota delta runs on a data extent. If that doesn't happen, due to refs canceling out or some error, the ref head already has the must_insert_reserved machinery to handle this, so we piggy back on that and use it to clean up the reserved data.

---
 fs/btrfs/delayed-ref.c |  1 +
 fs/btrfs/delayed-ref.h |  6 +++
 fs/btrfs/extent-tree.c | 85 +++++++++++++++++++++++++++++++++++++-----
 3 files changed, 82 insertions(+), 10 deletions(-)
fs/btrfs/delayed-ref.c +1 -0
index 28ba7a9eb3c3..874c1853d9b1 100644
@@ -745,6 +745,7 @@ static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref,
 	head_ref->bytenr = bytenr;
 	head_ref->num_bytes = num_bytes;
 	head_ref->ref_mod = count_mod;
+	head_ref->reserved_bytes = reserved;
 	head_ref->must_insert_reserved = must_insert_reserved;
 	head_ref->owning_root = owning_root;
 	head_ref->is_data = is_data;
fs/btrfs/delayed-ref.h +6 -0
index 71f0a6e5d583..221d400dd88f 100644
@@ -104,6 +104,12 @@ struct btrfs_delayed_ref_head {
 	 */
 	int ref_mod;
 
+	/*
+	 * Track reserved bytes when setting must_insert_reserved.
+	 * On success or cleanup, we will need to free the reservation.
+	 */
+	u64 reserved_bytes;
+
 	/*
 	 * when a new extent is allocated, it is just reserved in memory
 	 * The actual extent isn't inserted into the extent allocation tree
fs/btrfs/extent-tree.c +30 -5
index 09fb321fa560..1b5efd03ef83 100644
@@ -47,6 +47,7 @@
 
 
 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
+			       struct btrfs_delayed_ref_head *href,
 			       struct btrfs_delayed_ref_node *node, u64 parent,
 			       u64 root_objectid, u64 owner_objectid,
 			       u64 owner_offset, int refs_to_drop,
@@ -1482,6 +1483,7 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 }
 
 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
+				struct btrfs_delayed_ref_head *href,
 				struct btrfs_delayed_ref_node *node,
 				struct btrfs_delayed_extent_op *extent_op,
 				bool insert_reserved)
@@ -1505,18 +1507,28 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 	ref_root = ref->root;
 
 	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
+		struct btrfs_simple_quota_delta delta = {
+			.root = href->owning_root,
+			.num_bytes = node->num_bytes,
+			.rsv_bytes = href->reserved_bytes,
+			.is_data = true,
+			.is_inc	= true,
+		};
+
 		if (extent_op)
 			flags |= extent_op->flags_to_set;
 		ret = alloc_reserved_file_extent(trans, parent, ref_root,
 						 flags, ref->objectid,
 						 ref->offset, &ins,
 						 node->ref_mod);
+		if (!ret)
+			ret = btrfs_record_simple_quota_delta(trans->fs_info, &delta);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
 		ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
 					     ref->objectid, ref->offset,
 					     node->ref_mod, extent_op);
 	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
-		ret = __btrfs_free_extent(trans, node, parent,
+		ret = __btrfs_free_extent(trans, href, node, parent,
 					  ref_root, ref->objectid,
 					  ref->offset, node->ref_mod,
 					  extent_op);
@@ -1632,11 +1644,13 @@ static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
 }
 
 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
+				struct btrfs_delayed_ref_head *href,
 				struct btrfs_delayed_ref_node *node,
 				struct btrfs_delayed_extent_op *extent_op,
 				bool insert_reserved)
 {
 	int ret = 0;
+	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_delayed_tree_ref *ref;
 	u64 parent = 0;
 	u64 ref_root = 0;
@@ -1656,13 +1670,23 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 		return -EIO;
 	}
 	if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
+		struct btrfs_simple_quota_delta delta = {
+			.root = href->owning_root,
+			.num_bytes = fs_info->nodesize,
+			.rsv_bytes = 0,
+			.is_data = false,
+			.is_inc = true,
+		};
+
 		BUG_ON(!extent_op || !extent_op->update_flags);
 		ret = alloc_reserved_tree_block(trans, node, extent_op);
+		if (!ret)
+			btrfs_record_simple_quota_delta(fs_info, &delta);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
 		ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
 					     ref->level, 0, 1, extent_op);
 	} else if (node->action == BTRFS_DROP_DELAYED_REF) {
-		ret = __btrfs_free_extent(trans, node, parent, ref_root,
+		ret = __btrfs_free_extent(trans, href, node, parent, ref_root,
 					  ref->level, 0, 1, extent_op);
 	} else {
 		BUG();
@@ -1672,6 +1696,7 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 
 /* helper function to actually process a single delayed ref entry */
 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
+			       struct btrfs_delayed_ref_head *href,
 			       struct btrfs_delayed_ref_node *node,
 			       struct btrfs_delayed_extent_op *extent_op,
 			       bool insert_reserved)
@@ -1686,12 +1711,12 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 
 	if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
 	    node->type == BTRFS_SHARED_BLOCK_REF_KEY)
-		ret = run_delayed_tree_ref(trans, node, extent_op,
+		ret = run_delayed_tree_ref(trans, href, node, extent_op,
 					   insert_reserved);
 	else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
 		 node->type == BTRFS_SHARED_DATA_REF_KEY)
-		ret = run_delayed_data_ref(trans, node, extent_op,
-					   insert_reserved);
+		ret = run_delayed_data_ref(trans, href, node,
+					   extent_op, insert_reserved);
 	else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY)
 		ret = 0;
 	else

fs/btrfs/extent-tree.c void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, · line 1817

Quoted 5 lines
fs/btrfs/extent-tree.c +2 -0
@@ -1788,6 +1813,11 @@ void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
 		spin_unlock(&delayed_refs->lock);
 		nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
 	}
+	if (head->must_insert_reserved && head->is_data &&
+	    btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)

The check for qgroup mode should be the first condition.

Quoted 126 lines
fs/btrfs/extent-tree.c +43 -5
+		btrfs_qgroup_free_refroot(fs_info, head->owning_root,
+					  head->reserved_bytes,
+					  BTRFS_QGROUP_RSV_DATA);
 
 	btrfs_delayed_refs_rsv_release(fs_info, nr_items);
 }
@@ -1934,8 +1964,8 @@ static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
 		locked_ref->extent_op = NULL;
 		spin_unlock(&locked_ref->lock);
 
-		ret = run_one_delayed_ref(trans, ref, extent_op,
-					  must_insert_reserved);
+		ret = run_one_delayed_ref(trans, locked_ref, ref,
+					  extent_op, must_insert_reserved);
 
 		btrfs_free_delayed_extent_op(extent_op);
 		if (ret) {
@@ -2857,11 +2887,12 @@ u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
 }
 
 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
-				     u64 bytenr, u64 num_bytes, bool is_data)
+				     u64 bytenr, struct btrfs_simple_quota_delta *delta)
 {
 	int ret;
+	u64 num_bytes = delta->num_bytes;
 
-	if (is_data) {
+	if (delta->is_data) {
 		struct btrfs_root *csum_root;
 
 		csum_root = btrfs_csum_root(trans->fs_info, bytenr);
@@ -2872,6 +2903,12 @@ static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
 		}
 	}
 
+	ret = btrfs_record_simple_quota_delta(trans->fs_info, delta);
+	if (ret) {
+		btrfs_abort_transaction(trans, ret);
+		return ret;
+	}
+
 	ret = add_to_free_space_tree(trans, bytenr, num_bytes);
 	if (ret) {
 		btrfs_abort_transaction(trans, ret);
@@ -2952,6 +2989,7 @@ static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
  * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
  */
 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
+			       struct btrfs_delayed_ref_head *href,
 			       struct btrfs_delayed_ref_node *node, u64 parent,
 			       u64 root_objectid, u64 owner_objectid,
 			       u64 owner_offset, int refs_to_drop,
@@ -2974,6 +3012,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 	u64 bytenr = node->bytenr;
 	u64 num_bytes = node->num_bytes;
 	bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
+	u64 delayed_ref_root = href->owning_root;
 
 	extent_root = btrfs_extent_root(info, bytenr);
 	ASSERT(extent_root);
@@ -3172,6 +3211,14 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			}
 		}
 	} else {
+		struct btrfs_simple_quota_delta delta = {
+			.root = delayed_ref_root,
+			.num_bytes = num_bytes,
+			.rsv_bytes = 0,
+			.is_data = is_data,
+			.is_inc = false,
+		};
+
 		/* In this branch refs == 1 */
 		if (found_extent) {
 			if (is_data && refs_to_drop !=
@@ -3210,6 +3257,16 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 				num_to_del = 2;
 			}
 		}
+		/*
+		 * We can't infer the data owner from the delayed ref, so we
+		 * need to try to get it from the owning ref item.
+		 *
+		 * If it is not present, then that extent was not written under
+		 * simple quotas mode, so we don't need to account for its
+		 * deletion.
+		 */
+		if (is_data)
+			delta.root = btrfs_get_extent_owner_root(trans->fs_info, leaf, extent_slot);
 
 		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
 				      num_to_del);
@@ -3219,7 +3276,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 		}
 		btrfs_release_path(path);
 
-		ret = do_free_extent_accounting(trans, bytenr, num_bytes, is_data);
+		ret = do_free_extent_accounting(trans, bytenr, &delta);
 	}
 	btrfs_release_path(path);
 
@@ -4790,6 +4847,13 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	int ret;
 	struct btrfs_block_group *block_group;
 	struct btrfs_space_info *space_info;
+	struct btrfs_simple_quota_delta delta = {
+		.root = root_objectid,
+		.num_bytes = ins->offset,
+		.rsv_bytes = 0,
+		.is_data = true,
+		.is_inc = true,
+	};
 
 	/*
 	 * Mixed block groups will exclude before processing the log so we only
@@ -4818,6 +4882,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 					 offset, ins, 1);
 	if (ret)
 		btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
+	ret = btrfs_record_simple_quota_delta(fs_info, &delta);
 	btrfs_put_block_group(block_group);
 	return ret;
 }
Signature · 1 line
2.41.0
#34 lore raw

PATCH v5 14/18 btrfs: simple quota auto hierarchy for nested subvols

In reply to Boris Burkov

Consider the following sequence:
- enable quotas
- create subvol S id 256 at dir outer/
- create a qgroup 1/100
- add 0/256 (S's auto qgroup) to 1/100
- create subvol T id 257 at dir outer/inner/

With full qgroups, there is no relationship between 0/257 and either of 0/256 or 1/100. There is an inherit feature that the creator of inner/ can use to specify it ought to be in 1/100.

Simple quotas are targeted at container isolation, where such automatic inheritance for not necessarily trusted/controlled nested subvol creation would be quite helpful. Therefore, add a new default behavior for simple quotas: when you create a nested subvol, automatically inherit as parents any parents of the qgroup of the subvol the new inode is going in.

In our example, 257/0 would also be under 1/100, allowing easy control of a total quota over an arbitrary hierarchy of subvolumes.

I think this _might_ be a generally useful behavior, so it could be interesting to put it behind a new inheritance flag that simple quotas always use while traditional quotas let the user specify, but this is a minimally intrusive change to start.

---
 fs/btrfs/ioctl.c       |  2 +-
 fs/btrfs/qgroup.c      | 44 +++++++++++++++++++++++++++++++++++++++---
 fs/btrfs/qgroup.h      |  6 +++---
 fs/btrfs/transaction.c | 13 +++++++++----
 4 files changed, 54 insertions(+), 11 deletions(-)
fs/btrfs/ioctl.c +1 -1
index 9b61bc62e439..c9b069077fd0 100644
@@ -652,7 +652,7 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 	/* Tree log can't currently deal with an inode which is a new root. */
 	btrfs_set_log_full_commit(trans);
 
-	ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
+	ret = btrfs_qgroup_inherit(trans, 0, objectid, root->root_key.objectid, inherit);
 	if (ret)
 		goto out;
 
fs/btrfs/qgroup.c +41 -3

74 lines not on this page · raw message

fs/btrfs/qgroup.h +3 -3
index 94d85b4fbebd..ce6fa8694ca7 100644
@@ -271,8 +271,7 @@ int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
 				     bool interruptible);
-int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
-			      u64 dst);
+int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst);
 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
 			      u64 dst);
 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
@@ -366,7 +365,8 @@ int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
 int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
-			 u64 objectid, struct btrfs_qgroup_inherit *inherit);
+			 u64 objectid, u64 inode_rootid,
+			 struct btrfs_qgroup_inherit *inherit);
 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
 			       u64 ref_root, u64 num_bytes,
 			       enum btrfs_qgroup_rsv_type type);
fs/btrfs/transaction.c +9 -4
index 25217888e897..fb857147df57 100644
@@ -1529,13 +1529,14 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 	int ret;
 
 	/*
-	 * Save some performance in the case that full qgroups are not
+	 * Save some performance in the case that qgroups are not
 	 * enabled. If this check races with the ioctl, rescan will
 	 * kick in anyway.
 	 */
 	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 
+
 	/*
 	 * Ensure dirty @src will be committed.  Or, after coming
 	 * commit_fs_roots() and switch_commit_roots(), any dirty but not
@@ -1572,7 +1573,7 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 
 	/* Now qgroup are all updated, we can inherit it to new qgroups */
 	ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
-				   inherit);
+				   parent->root_key.objectid, inherit);
 	if (ret < 0)
 		goto out;
 
@@ -1839,8 +1840,12 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	 * To co-operate with that hack, we do hack again.
 	 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
 	 */
-	ret = qgroup_account_snapshot(trans, root, parent_root,
-				      pending->inherit, objectid);
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL)
+		ret = qgroup_account_snapshot(trans, root, parent_root,
+					      pending->inherit, objectid);
+	else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
+		ret = btrfs_qgroup_inherit(trans, root->root_key.objectid, objectid,
+					   parent_root->root_key.objectid, pending->inherit);
 	if (ret < 0)
 		goto fail;
 
Signature · 1 line
2.41.0
Reviewed-by on v5 14/18
#35 lore raw

Re: [PATCH v5 14/18] btrfs: simple quota auto hierarchy for nested subvols

Quoted 123 lines · Boris Burkov

Consider the following sequence:
- enable quotas
- create subvol S id 256 at dir outer/
- create a qgroup 1/100
- add 0/256 (S's auto qgroup) to 1/100
- create subvol T id 257 at dir outer/inner/

With full qgroups, there is no relationship between 0/257 and either of 0/256 or 1/100. There is an inherit feature that the creator of inner/ can use to specify it ought to be in 1/100.

Simple quotas are targeted at container isolation, where such automatic inheritance for not necessarily trusted/controlled nested subvol creation would be quite helpful. Therefore, add a new default behavior for simple quotas: when you create a nested subvol, automatically inherit as parents any parents of the qgroup of the subvol the new inode is going in.

In our example, 257/0 would also be under 1/100, allowing easy control of a total quota over an arbitrary hierarchy of subvolumes.

I think this _might_ be a generally useful behavior, so it could be interesting to put it behind a new inheritance flag that simple quotas always use while traditional quotas let the user specify, but this is a minimally intrusive change to start.

---
 fs/btrfs/ioctl.c       |  2 +-
 fs/btrfs/qgroup.c      | 44 +++++++++++++++++++++++++++++++++++++++---
 fs/btrfs/qgroup.h      |  6 +++---
 fs/btrfs/transaction.c | 13 +++++++++----
 4 files changed, 54 insertions(+), 11 deletions(-)
fs/btrfs/ioctl.c +1 -1
index 9b61bc62e439..c9b069077fd0 100644
@@ -652,7 +652,7 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 	/* Tree log can't currently deal with an inode which is a new root. */
 	btrfs_set_log_full_commit(trans);
 
-	ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
+	ret = btrfs_qgroup_inherit(trans, 0, objectid, root->root_key.objectid, inherit);
 	if (ret)
 		goto out;
 
fs/btrfs/qgroup.c +40 -3
index dedc532669f4..58e9ed0deedd 100644
@@ -1550,8 +1550,7 @@ static int quick_update_accounting(struct btrfs_fs_info *fs_info,
 	return ret;
 }
 
-int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
-			      u64 dst)
+int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_qgroup *parent;
@@ -2991,6 +2990,40 @@ int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
 	return ret;
 }
 
+static int qgroup_auto_inherit(struct btrfs_fs_info *fs_info,
+			       u64 inode_rootid,
+			       struct btrfs_qgroup_inherit **inherit)
+{
+	int i = 0;
+	u64 num_qgroups = 0;
+	struct btrfs_qgroup *inode_qg;
+	struct btrfs_qgroup_list *qg_list;
+
+	if (*inherit)
+		return -EEXIST;
+
+	inode_qg = find_qgroup_rb(fs_info, inode_rootid);
+	if (!inode_qg)
+		return -ENOENT;
+
+	num_qgroups = list_count_nodes(&inode_qg->groups);
+
+	if (!num_qgroups)
+		return 0;
+
+	*inherit = kzalloc(sizeof(**inherit) + num_qgroups * sizeof(u64), GFP_NOFS);
+	if (!*inherit)
+		return -ENOMEM;
+	(*inherit)->num_qgroups = num_qgroups;
+
+	list_for_each_entry(qg_list, &inode_qg->groups, next_group) {
+		u64 qg_id = qg_list->group->qgroupid;
+		*((u64 *)((*inherit)+1) + i) = qg_id;
+	}
+
+	return 0;
+}
+
 /*
  * Copy the accounting information between qgroups. This is necessary
  * when a snapshot or a subvolume is created. Throwing an error will
@@ -2998,7 +3031,8 @@ int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
  * when a readonly fs is a reasonable outcome.
  */
 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
-			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
+			 u64 objectid, u64 inode_rootid,
+			 struct btrfs_qgroup_inherit *inherit)
 {
 	int ret = 0;
 	int i;
@@ -3040,6 +3074,9 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 		goto out;
 	}
 
+	if (!inherit && btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
+		qgroup_auto_inherit(fs_info, inode_rootid, &inherit);
+
 	if (inherit) {
 		i_qgroups = (u64 *)(inherit + 1);
 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +

fs/btrfs/qgroup.c int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, · line 3106

Quoted 4 lines
fs/btrfs/qgroup.c +1 -0
@@ -3066,6 +3103,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 	if (ret)
 		goto out;
 
+

Extraneous whitespace change. Once fixed you can add

Thanks,

Josef

#36 lore raw

Re: [PATCH v5 14/18] btrfs: simple quota auto hierarchy for nested subvols

In reply to Boris Burkov
Quoted 77 lines · Boris Burkov

Consider the following sequence:
- enable quotas
- create subvol S id 256 at dir outer/
- create a qgroup 1/100
- add 0/256 (S's auto qgroup) to 1/100
- create subvol T id 257 at dir outer/inner/

With full qgroups, there is no relationship between 0/257 and either of 0/256 or 1/100. There is an inherit feature that the creator of inner/ can use to specify it ought to be in 1/100.

Simple quotas are targeted at container isolation, where such automatic inheritance for not necessarily trusted/controlled nested subvol creation would be quite helpful. Therefore, add a new default behavior for simple quotas: when you create a nested subvol, automatically inherit as parents any parents of the qgroup of the subvol the new inode is going in.

In our example, 257/0 would also be under 1/100, allowing easy control of a total quota over an arbitrary hierarchy of subvolumes.

I think this _might_ be a generally useful behavior, so it could be interesting to put it behind a new inheritance flag that simple quotas always use while traditional quotas let the user specify, but this is a minimally intrusive change to start.

---
 fs/btrfs/ioctl.c       |  2 +-
 fs/btrfs/qgroup.c      | 44 +++++++++++++++++++++++++++++++++++++++---
 fs/btrfs/qgroup.h      |  6 +++---
 fs/btrfs/transaction.c | 13 +++++++++----
 4 files changed, 54 insertions(+), 11 deletions(-)
fs/btrfs/ioctl.c +1 -1
index 9b61bc62e439..c9b069077fd0 100644
@@ -652,7 +652,7 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 	/* Tree log can't currently deal with an inode which is a new root. */
 	btrfs_set_log_full_commit(trans);
 
-	ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
+	ret = btrfs_qgroup_inherit(trans, 0, objectid, root->root_key.objectid, inherit);
 	if (ret)
 		goto out;
 
fs/btrfs/qgroup.c +13 -2
index dedc532669f4..58e9ed0deedd 100644
@@ -1550,8 +1550,7 @@ static int quick_update_accounting(struct btrfs_fs_info *fs_info,
 	return ret;
 }
 
-int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
-			      u64 dst)
+int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_qgroup *parent;
@@ -2991,6 +2990,40 @@ int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
 	return ret;
 }
 
+static int qgroup_auto_inherit(struct btrfs_fs_info *fs_info,
+			       u64 inode_rootid,
+			       struct btrfs_qgroup_inherit **inherit)
+{
+	int i = 0;
+	u64 num_qgroups = 0;
+	struct btrfs_qgroup *inode_qg;
+	struct btrfs_qgroup_list *qg_list;
+
+	if (*inherit)
+		return -EEXIST;
+

fs/btrfs/qgroup.c int btrfs_run_qgroups(struct btrfs_trans_handle *trans) · line 3014

Quoted 10 lines
fs/btrfs/qgroup.c +10 -0
+	inode_qg = find_qgroup_rb(fs_info, inode_rootid);
+	if (!inode_qg)
+		return -ENOENT;
+
+	num_qgroups = list_count_nodes(&inode_qg->groups);
+
+	if (!num_qgroups)
+		return 0;
+
+	*inherit = kzalloc(sizeof(**inherit) + num_qgroups * sizeof(u64), GFP_NOFS);

There's kcalloc that verifies the potential multiplication overflow.

fs/btrfs/qgroup.c int btrfs_run_qgroups(struct btrfs_trans_handle *trans) · line 3021

Quoted 7 lines
fs/btrfs/qgroup.c +7 -0
+	if (!*inherit)
+		return -ENOMEM;
+	(*inherit)->num_qgroups = num_qgroups;
+
+	list_for_each_entry(qg_list, &inode_qg->groups, next_group) {
+		u64 qg_id = qg_list->group->qgroupid;
+		*((u64 *)((*inherit)+1) + i) = qg_id;

What does this do?

fs/btrfs/qgroup.c int btrfs_run_qgroups(struct btrfs_trans_handle *trans) · line 3022

Quoted 1 line
fs/btrfs/qgroup.c +1 -0
+	}

Instead of reusing the *inherit for operations, please add a local variable and at the end assign it to *inherit.

Quoted 18 lines
fs/btrfs/qgroup.c +6 -1
+
+	return 0;
+}
+
 /*
  * Copy the accounting information between qgroups. This is necessary
  * when a snapshot or a subvolume is created. Throwing an error will
@@ -2998,7 +3031,8 @@ int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
  * when a readonly fs is a reasonable outcome.
  */
 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
-			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
+			 u64 objectid, u64 inode_rootid,
+			 struct btrfs_qgroup_inherit *inherit)
 {
 	int ret = 0;
 	int i;

fs/btrfs/qgroup.c int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, · line 3077

Quoted 4 lines
fs/btrfs/qgroup.c +1 -0
@@ -3040,6 +3074,9 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 		goto out;
 	}
 
+	if (!inherit && btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)

Swap the conditions please

fs/btrfs/qgroup.c int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, · line 3106

Quoted 10 lines
fs/btrfs/qgroup.c +3 -0
+		qgroup_auto_inherit(fs_info, inode_rootid, &inherit);
+
 	if (inherit) {
 		i_qgroups = (u64 *)(inherit + 1);
 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
@@ -3066,6 +3103,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
 	if (ret)
 		goto out;
 
+

Stray newline

Quoted 34 lines
fs/btrfs/qgroup.c +0 -0
 	/*
 	 * add qgroup to all inherited groups
 	 */
fs/btrfs/qgroup.h +3 -3
index 94d85b4fbebd..ce6fa8694ca7 100644
@@ -271,8 +271,7 @@ int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
 				     bool interruptible);
-int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
-			      u64 dst);
+int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst);
 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
 			      u64 dst);
 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
@@ -366,7 +365,8 @@ int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
 int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
-			 u64 objectid, struct btrfs_qgroup_inherit *inherit);
+			 u64 objectid, u64 inode_rootid,
+			 struct btrfs_qgroup_inherit *inherit);
 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
 			       u64 ref_root, u64 num_bytes,
 			       enum btrfs_qgroup_rsv_type type);
fs/btrfs/transaction.c +0 -0
index 25217888e897..fb857147df57 100644
@@ -1529,13 +1529,14 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 	int ret;
 

fs/btrfs/transaction.c static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, · line 1539

Quoted 10 lines
fs/btrfs/transaction.c +2 -1
 	/*
-	 * Save some performance in the case that full qgroups are not
+	 * Save some performance in the case that qgroups are not
 	 * enabled. If this check races with the ioctl, rescan will
 	 * kick in anyway.
 	 */
 	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_FULL)
 		return 0;
 
+

Again

Quoted 29 lines
fs/btrfs/transaction.c +7 -3
 	/*
 	 * Ensure dirty @src will be committed.  Or, after coming
 	 * commit_fs_roots() and switch_commit_roots(), any dirty but not
@@ -1572,7 +1573,7 @@ static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
 
 	/* Now qgroup are all updated, we can inherit it to new qgroups */
 	ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
-				   inherit);
+				   parent->root_key.objectid, inherit);
 	if (ret < 0)
 		goto out;
 
@@ -1839,8 +1840,12 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	 * To co-operate with that hack, we do hack again.
 	 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
 	 */
-	ret = qgroup_account_snapshot(trans, root, parent_root,
-				      pending->inherit, objectid);
+	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL)
+		ret = qgroup_account_snapshot(trans, root, parent_root,
+					      pending->inherit, objectid);
+	else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
+		ret = btrfs_qgroup_inherit(trans, root->root_key.objectid, objectid,
+					   parent_root->root_key.objectid, pending->inherit);
 	if (ret < 0)
 		goto fail;
 
Signature · 1 line
2.41.0
#37 lore raw

PATCH v5 15/18 btrfs: check generation when recording simple quota delta

In reply to Boris Burkov

Simple quotas count extents only from the moment the feature is enabled. Therefore, if we do something like:
1. create subvol S
2. write F in S
3. enable quotas
4. remove F
5. write G in S

then after 3. and 4. we would expect the simple quota usage of S to be 0 (putting aside some metadata extents that might be written) and after 5., it should be the size of G plus metadata. Therefore, we need to be able to determine whether a particular quota delta we are processing predates simple quota enablement.

To do this, store the transaction id when quotas were enabled. In fs_info for immediate use and in the quota status item to make it recoverable on mount. When we see a delta, check if the generation of the extent item is less than that of quota enablement. If so, we should ignore the delta from this extent.

---
 fs/btrfs/accessors.h            |  2 ++
 fs/btrfs/extent-tree.c          |  4 ++++
 fs/btrfs/fs.h                   |  2 ++
 fs/btrfs/qgroup.c               | 14 ++++++++++++--
 fs/btrfs/qgroup.h               |  1 +
 include/uapi/linux/btrfs_tree.h |  7 +++++++
 6 files changed, 28 insertions(+), 2 deletions(-)
fs/btrfs/accessors.h +2 -0
index a23045c05937..513f8edbd98e 100644
@@ -970,6 +970,8 @@ BTRFS_SETGET_FUNCS(qgroup_status_flags, struct btrfs_qgroup_status_item,
 		   flags, 64);
 BTRFS_SETGET_FUNCS(qgroup_status_rescan, struct btrfs_qgroup_status_item,
 		   rescan, 64);
+BTRFS_SETGET_FUNCS(qgroup_status_enable_gen, struct btrfs_qgroup_status_item,
+		   enable_gen, 64);
 
 /* btrfs_qgroup_info_item */
 BTRFS_SETGET_FUNCS(qgroup_info_generation, struct btrfs_qgroup_info_item,
fs/btrfs/extent-tree.c +4 -0
index 1b5efd03ef83..395ab46e520b 100644
@@ -1513,6 +1513,7 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = href->reserved_bytes,
 			.is_data = true,
 			.is_inc	= true,
+			.generation = trans->transid,
 		};
 
 		if (extent_op)
@@ -1676,6 +1677,7 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = false,
 			.is_inc = true,
+			.generation = trans->transid,
 		};
 
 		BUG_ON(!extent_op || !extent_op->update_flags);
@@ -3217,6 +3219,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = is_data,
 			.is_inc = false,
+			.generation = btrfs_extent_generation(leaf, ei),
 		};
 
 		/* In this branch refs == 1 */
@@ -4850,6 +4853,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_simple_quota_delta delta = {
 		.root = root_objectid,
 		.num_bytes = ins->offset,
+		.generation = trans->transid,
 		.rsv_bytes = 0,
 		.is_data = true,
 		.is_inc = true,
fs/btrfs/fs.h +2 -0
index f76f450c2abf..da7b623ff15f 100644
@@ -802,6 +802,8 @@ struct btrfs_fs_info {
 	spinlock_t eb_leak_lock;
 	struct list_head allocated_ebs;
 #endif
+
+	u64 quota_enable_gen;
 };
 
 static inline void btrfs_set_last_root_drop_gen(struct btrfs_fs_info *fs_info,
fs/btrfs/qgroup.c +12 -2
index 58e9ed0deedd..a8a603242431 100644
@@ -454,6 +454,8 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 			}
 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, ptr);
 			simple = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
+			if (simple)
+				fs_info->quota_enable_gen = btrfs_qgroup_status_enable_gen(l, ptr);
 			if (btrfs_qgroup_status_generation(l, ptr) !=
 			    fs_info->generation && !simple) {
 				qgroup_mark_inconsistent(fs_info);
@@ -1107,10 +1109,12 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON;
-	if (simple)
+	if (simple) {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
-	else
+		btrfs_set_qgroup_status_enable_gen(leaf, ptr, trans->transid);
+	} else {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
+	}
 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
@@ -1202,6 +1206,8 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 		goto out_free_path;
 	}
 
+	fs_info->quota_enable_gen = trans->transid;
+
 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
 	/*
 	 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
@@ -4622,6 +4628,10 @@ int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
 	if (!is_fstree(root))
 		return 0;
 
+	/* If the extent predates enabling quotas, don't count it. */
+	if (delta->generation < fs_info->quota_enable_gen)
+		return 0;
+
 	spin_lock(&fs_info->qgroup_lock);
 	qgroup = find_qgroup_rb(fs_info, root);
 	if (!qgroup) {
fs/btrfs/qgroup.h +1 -0
index ce6fa8694ca7..ae1ce14b365c 100644
@@ -241,6 +241,7 @@ struct btrfs_simple_quota_delta {
 	u64 rsv_bytes; /* The number of bytes reserved for this extent */
 	bool is_inc; /* Whether we are using or freeing the extent */
 	bool is_data; /* Whether the extent is data or metadata */
+	u64 generation; /* The generation the extent was created in */
 };
 
 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
include/uapi/linux/btrfs_tree.h +7 -0
index eacb26caf3c6..1120ce3dae42 100644
@@ -1242,6 +1242,13 @@ struct btrfs_qgroup_status_item {
 	 * of the scan. It contains a logical address
 	 */
 	__le64 rescan;
+
+	/*
+	 * the generation when quotas are enabled. Used by simple quotas to
+	 * avoid decrementing when freeing an extent that was written before
+	 * enable.
+	 */
+	__le64 enable_gen;
 } __attribute__ ((__packed__));
 
 struct btrfs_qgroup_info_item {
Signature · 1 line
2.41.0
Reviewed-by on v5 15/18
#38 lore raw

Re: [PATCH v5 15/18] btrfs: check generation when recording simple quota delta

Quoted 14 lines · Boris Burkov

Simple quotas count extents only from the moment the feature is enabled. Therefore, if we do something like:
1. create subvol S
2. write F in S
3. enable quotas
4. remove F
5. write G in S

then after 3. and 4. we would expect the simple quota usage of S to be 0 (putting aside some metadata extents that might be written) and after 5., it should be the size of G plus metadata. Therefore, we need to be able to determine whether a particular quota delta we are processing predates simple quota enablement.

Quoted 7 lines

To do this, store the transaction id when quotas were enabled. In fs_info for immediate use and in the quota status item to make it recoverable on mount. When we see a delta, check if the generation of the extent item is less than that of quota enablement. If so, we should ignore the delta from this extent.

Thanks,

Josef

#39 lore raw

Re: [PATCH v5 15/18] btrfs: check generation when recording simple quota delta

In reply to Boris Burkov
Quoted 84 lines · Boris Burkov

Simple quotas count extents only from the moment the feature is enabled. Therefore, if we do something like:
1. create subvol S
2. write F in S
3. enable quotas
4. remove F
5. write G in S

then after 3. and 4. we would expect the simple quota usage of S to be 0 (putting aside some metadata extents that might be written) and after 5., it should be the size of G plus metadata. Therefore, we need to be able to determine whether a particular quota delta we are processing predates simple quota enablement.

To do this, store the transaction id when quotas were enabled. In fs_info for immediate use and in the quota status item to make it recoverable on mount. When we see a delta, check if the generation of the extent item is less than that of quota enablement. If so, we should ignore the delta from this extent.

---
 fs/btrfs/accessors.h            |  2 ++
 fs/btrfs/extent-tree.c          |  4 ++++
 fs/btrfs/fs.h                   |  2 ++
 fs/btrfs/qgroup.c               | 14 ++++++++++++--
 fs/btrfs/qgroup.h               |  1 +
 include/uapi/linux/btrfs_tree.h |  7 +++++++
 6 files changed, 28 insertions(+), 2 deletions(-)
fs/btrfs/accessors.h +2 -0
index a23045c05937..513f8edbd98e 100644
@@ -970,6 +970,8 @@ BTRFS_SETGET_FUNCS(qgroup_status_flags, struct btrfs_qgroup_status_item,
 		   flags, 64);
 BTRFS_SETGET_FUNCS(qgroup_status_rescan, struct btrfs_qgroup_status_item,
 		   rescan, 64);
+BTRFS_SETGET_FUNCS(qgroup_status_enable_gen, struct btrfs_qgroup_status_item,
+		   enable_gen, 64);
 
 /* btrfs_qgroup_info_item */
 BTRFS_SETGET_FUNCS(qgroup_info_generation, struct btrfs_qgroup_info_item,
fs/btrfs/extent-tree.c +4 -0
index 1b5efd03ef83..395ab46e520b 100644
@@ -1513,6 +1513,7 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = href->reserved_bytes,
 			.is_data = true,
 			.is_inc	= true,
+			.generation = trans->transid,
 		};
 
 		if (extent_op)
@@ -1676,6 +1677,7 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = false,
 			.is_inc = true,
+			.generation = trans->transid,
 		};
 
 		BUG_ON(!extent_op || !extent_op->update_flags);
@@ -3217,6 +3219,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = is_data,
 			.is_inc = false,
+			.generation = btrfs_extent_generation(leaf, ei),
 		};
 
 		/* In this branch refs == 1 */
@@ -4850,6 +4853,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_simple_quota_delta delta = {
 		.root = root_objectid,
 		.num_bytes = ins->offset,
+		.generation = trans->transid,
 		.rsv_bytes = 0,
 		.is_data = true,
 		.is_inc = true,
fs/btrfs/fs.h +0 -0
index f76f450c2abf..da7b623ff15f 100644

fs/btrfs/fs.h struct btrfs_fs_info { · line 806

Quoted 5 lines
fs/btrfs/fs.h +2 -0
@@ -802,6 +802,8 @@ struct btrfs_fs_info {
 	spinlock_t eb_leak_lock;
 	struct list_head allocated_ebs;
 #endif
+
+	u64 quota_enable_gen;

Please move it to the other quota/qgroup related members, at the end of fs_info there's only debugging stuff.

Quoted 56 lines
fs/btrfs/fs.h +0 -0
 };
 
 static inline void btrfs_set_last_root_drop_gen(struct btrfs_fs_info *fs_info,
fs/btrfs/qgroup.c +12 -2
index 58e9ed0deedd..a8a603242431 100644
@@ -454,6 +454,8 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 			}
 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, ptr);
 			simple = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
+			if (simple)
+				fs_info->quota_enable_gen = btrfs_qgroup_status_enable_gen(l, ptr);
 			if (btrfs_qgroup_status_generation(l, ptr) !=
 			    fs_info->generation && !simple) {
 				qgroup_mark_inconsistent(fs_info);
@@ -1107,10 +1109,12 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON;
-	if (simple)
+	if (simple) {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
-	else
+		btrfs_set_qgroup_status_enable_gen(leaf, ptr, trans->transid);
+	} else {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
+	}
 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
@@ -1202,6 +1206,8 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 		goto out_free_path;
 	}
 
+	fs_info->quota_enable_gen = trans->transid;
+
 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
 	/*
 	 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
@@ -4622,6 +4628,10 @@ int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
 	if (!is_fstree(root))
 		return 0;
 
+	/* If the extent predates enabling quotas, don't count it. */
+	if (delta->generation < fs_info->quota_enable_gen)
+		return 0;
+
 	spin_lock(&fs_info->qgroup_lock);
 	qgroup = find_qgroup_rb(fs_info, root);
 	if (!qgroup) {
fs/btrfs/qgroup.h +0 -0
index ce6fa8694ca7..ae1ce14b365c 100644

fs/btrfs/qgroup.h struct btrfs_simple_quota_delta { · line 244

Quoted 4 lines
fs/btrfs/qgroup.h +1 -0
@@ -241,6 +241,7 @@ struct btrfs_simple_quota_delta {
 	u64 rsv_bytes; /* The number of bytes reserved for this extent */
 	bool is_inc; /* Whether we are using or freeing the extent */
 	bool is_data; /* Whether the extent is data or metadata */
+	u64 generation; /* The generation the extent was created in */

Please reorder it so it does not leave gaps between struct members.

Quoted 8 lines
fs/btrfs/qgroup.h +0 -0
 };
 
 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
include/uapi/linux/btrfs_tree.h +0 -0
index eacb26caf3c6..1120ce3dae42 100644

include/uapi/linux/btrfs_tree.h struct btrfs_qgroup_status_item { · line 1251

Quoted 10 lines
include/uapi/linux/btrfs_tree.h +7 -0
@@ -1242,6 +1242,13 @@ struct btrfs_qgroup_status_item {
 	 * of the scan. It contains a logical address
 	 */
 	__le64 rescan;
+
+	/*
+	 * the generation when quotas are enabled. Used by simple quotas to
+	 * avoid decrementing when freeing an extent that was written before
+	 * enable.
+	 */
+	__le64 enable_gen;

This is public interface and btrfs_qgroup_status_item is used in many places in user space at least in btrfs-progs. This needs a lot of sanity checks.

Quoted 5 lines
include/uapi/linux/btrfs_tree.h +0 -0
 } __attribute__ ((__packed__));
 
 struct btrfs_qgroup_info_item {
Signature · 1 line
2.41.0
#40 lore raw

Re: [PATCH v5 15/18] btrfs: check generation when recording simple quota delta

Quoted 171 lines · David Sterba
Quoted 84 lines · Boris Burkov

Simple quotas count extents only from the moment the feature is enabled. Therefore, if we do something like:
1. create subvol S
2. write F in S
3. enable quotas
4. remove F
5. write G in S

then after 3. and 4. we would expect the simple quota usage of S to be 0 (putting aside some metadata extents that might be written) and after 5., it should be the size of G plus metadata. Therefore, we need to be able to determine whether a particular quota delta we are processing predates simple quota enablement.

To do this, store the transaction id when quotas were enabled. In fs_info for immediate use and in the quota status item to make it recoverable on mount. When we see a delta, check if the generation of the extent item is less than that of quota enablement. If so, we should ignore the delta from this extent.

---
 fs/btrfs/accessors.h            |  2 ++
 fs/btrfs/extent-tree.c          |  4 ++++
 fs/btrfs/fs.h                   |  2 ++
 fs/btrfs/qgroup.c               | 14 ++++++++++++--
 fs/btrfs/qgroup.h               |  1 +
 include/uapi/linux/btrfs_tree.h |  7 +++++++
 6 files changed, 28 insertions(+), 2 deletions(-)
fs/btrfs/accessors.h +2 -0
index a23045c05937..513f8edbd98e 100644
@@ -970,6 +970,8 @@ BTRFS_SETGET_FUNCS(qgroup_status_flags, struct btrfs_qgroup_status_item,
 		   flags, 64);
 BTRFS_SETGET_FUNCS(qgroup_status_rescan, struct btrfs_qgroup_status_item,
 		   rescan, 64);
+BTRFS_SETGET_FUNCS(qgroup_status_enable_gen, struct btrfs_qgroup_status_item,
+		   enable_gen, 64);
 
 /* btrfs_qgroup_info_item */
 BTRFS_SETGET_FUNCS(qgroup_info_generation, struct btrfs_qgroup_info_item,
fs/btrfs/extent-tree.c +4 -0
index 1b5efd03ef83..395ab46e520b 100644
@@ -1513,6 +1513,7 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = href->reserved_bytes,
 			.is_data = true,
 			.is_inc	= true,
+			.generation = trans->transid,
 		};
 
 		if (extent_op)
@@ -1676,6 +1677,7 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = false,
 			.is_inc = true,
+			.generation = trans->transid,
 		};
 
 		BUG_ON(!extent_op || !extent_op->update_flags);
@@ -3217,6 +3219,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = is_data,
 			.is_inc = false,
+			.generation = btrfs_extent_generation(leaf, ei),
 		};
 
 		/* In this branch refs == 1 */
@@ -4850,6 +4853,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_simple_quota_delta delta = {
 		.root = root_objectid,
 		.num_bytes = ins->offset,
+		.generation = trans->transid,
 		.rsv_bytes = 0,
 		.is_data = true,
 		.is_inc = true,
fs/btrfs/fs.h +0 -0
index f76f450c2abf..da7b623ff15f 100644

fs/btrfs/fs.h struct btrfs_fs_info { · line 806

Quoted 5 lines
fs/btrfs/fs.h +2 -0
@@ -802,6 +802,8 @@ struct btrfs_fs_info {
 	spinlock_t eb_leak_lock;
 	struct list_head allocated_ebs;
 #endif
+
+	u64 quota_enable_gen;

Please move it to the other quota/qgroup related members, at the end of fs_info there's only debugging stuff.

Quoted 56 lines
fs/btrfs/fs.h +0 -0
 };
 
 static inline void btrfs_set_last_root_drop_gen(struct btrfs_fs_info *fs_info,
fs/btrfs/qgroup.c +12 -2
index 58e9ed0deedd..a8a603242431 100644
@@ -454,6 +454,8 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 			}
 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, ptr);
 			simple = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
+			if (simple)
+				fs_info->quota_enable_gen = btrfs_qgroup_status_enable_gen(l, ptr);
 			if (btrfs_qgroup_status_generation(l, ptr) !=
 			    fs_info->generation && !simple) {
 				qgroup_mark_inconsistent(fs_info);
@@ -1107,10 +1109,12 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON;
-	if (simple)
+	if (simple) {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
-	else
+		btrfs_set_qgroup_status_enable_gen(leaf, ptr, trans->transid);
+	} else {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
+	}
 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
@@ -1202,6 +1206,8 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 		goto out_free_path;
 	}
 
+	fs_info->quota_enable_gen = trans->transid;
+
 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
 	/*
 	 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
@@ -4622,6 +4628,10 @@ int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
 	if (!is_fstree(root))
 		return 0;
 
+	/* If the extent predates enabling quotas, don't count it. */
+	if (delta->generation < fs_info->quota_enable_gen)
+		return 0;
+
 	spin_lock(&fs_info->qgroup_lock);
 	qgroup = find_qgroup_rb(fs_info, root);
 	if (!qgroup) {
fs/btrfs/qgroup.h +0 -0
index ce6fa8694ca7..ae1ce14b365c 100644

fs/btrfs/qgroup.h struct btrfs_simple_quota_delta { · line 244

Quoted 4 lines
fs/btrfs/qgroup.h +1 -0
@@ -241,6 +241,7 @@ struct btrfs_simple_quota_delta {
 	u64 rsv_bytes; /* The number of bytes reserved for this extent */
 	bool is_inc; /* Whether we are using or freeing the extent */
 	bool is_data; /* Whether the extent is data or metadata */
+	u64 generation; /* The generation the extent was created in */

Please reorder it so it does not leave gaps between struct members.

Quoted 14 lines
fs/btrfs/qgroup.h +0 -0
 };
 
 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
include/uapi/linux/btrfs_tree.h +3 -0
index eacb26caf3c6..1120ce3dae42 100644
@@ -1242,6 +1242,13 @@ struct btrfs_qgroup_status_item {
 	 * of the scan. It contains a logical address
 	 */
 	__le64 rescan;
+
+	/*
+	 * the generation when quotas are enabled. Used by simple quotas to
Quoted 8 lines

include/uapi/linux/btrfs_tree.h struct btrfs_qgroup_status_item { · line 1251

Quoted 4 lines
include/uapi/linux/btrfs_tree.h +4 -0
+	 * avoid decrementing when freeing an extent that was written before
+	 * enable.
+	 */
+	__le64 enable_gen;

This is public interface and btrfs_qgroup_status_item is used in many places in user space at least in btrfs-progs. This needs a lot of sanity checks.

Totally agreed in principle, but not exactly sure how to proceed in practice. I would definitely appreciate some tips/help!

How we interact with the new field:
- When enabling squota, set it, the incompat bit, and the status flag
- When reading in the qgroup status_item, if the status flag is set, then read the enable_gen.

I believe this prevents us from ever reading garbage while trying to read an old fs (status flag won't be set) and it prevents any btrfs-progs from getting confused by a wrong-sized status item, since it would choke on the incompat bit first.

Am I missing some other case? I can try to make it more explicitly zeroed when we enable qgroups but not squotas? I can add an ASSERT that the incompat bit is set as expected when we read the status item with the flag on (that seems good no matter what)?

I can also write a wrapper for getting it which does the incompat/status flag checking to make it more clear that it isn't safe to read in general. Or a comment on the struct saying it depends on the incompat bit?

Thanks for all the review, by the way.

Quoted 6 lines
Quoted 6 lines
include/uapi/linux/btrfs_tree.h +0 -0
 } __attribute__ ((__packed__));
 
 struct btrfs_qgroup_info_item {
Signature · 1 line
2.41.0
#41 lore raw

Re: [PATCH v5 15/18] btrfs: check generation when recording simple quota delta

Quoted 199 lines · Boris Burkov
Quoted 171 lines · David Sterba
Quoted 84 lines · Boris Burkov

Simple quotas count extents only from the moment the feature is enabled. Therefore, if we do something like:
1. create subvol S
2. write F in S
3. enable quotas
4. remove F
5. write G in S

then after 3. and 4. we would expect the simple quota usage of S to be 0 (putting aside some metadata extents that might be written) and after 5., it should be the size of G plus metadata. Therefore, we need to be able to determine whether a particular quota delta we are processing predates simple quota enablement.

To do this, store the transaction id when quotas were enabled. In fs_info for immediate use and in the quota status item to make it recoverable on mount. When we see a delta, check if the generation of the extent item is less than that of quota enablement. If so, we should ignore the delta from this extent.

---
 fs/btrfs/accessors.h            |  2 ++
 fs/btrfs/extent-tree.c          |  4 ++++
 fs/btrfs/fs.h                   |  2 ++
 fs/btrfs/qgroup.c               | 14 ++++++++++++--
 fs/btrfs/qgroup.h               |  1 +
 include/uapi/linux/btrfs_tree.h |  7 +++++++
 6 files changed, 28 insertions(+), 2 deletions(-)
fs/btrfs/accessors.h +2 -0
index a23045c05937..513f8edbd98e 100644
@@ -970,6 +970,8 @@ BTRFS_SETGET_FUNCS(qgroup_status_flags, struct btrfs_qgroup_status_item,
 		   flags, 64);
 BTRFS_SETGET_FUNCS(qgroup_status_rescan, struct btrfs_qgroup_status_item,
 		   rescan, 64);
+BTRFS_SETGET_FUNCS(qgroup_status_enable_gen, struct btrfs_qgroup_status_item,
+		   enable_gen, 64);
 
 /* btrfs_qgroup_info_item */
 BTRFS_SETGET_FUNCS(qgroup_info_generation, struct btrfs_qgroup_info_item,
fs/btrfs/extent-tree.c +4 -0
index 1b5efd03ef83..395ab46e520b 100644
@@ -1513,6 +1513,7 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = href->reserved_bytes,
 			.is_data = true,
 			.is_inc	= true,
+			.generation = trans->transid,
 		};
 
 		if (extent_op)
@@ -1676,6 +1677,7 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = false,
 			.is_inc = true,
+			.generation = trans->transid,
 		};
 
 		BUG_ON(!extent_op || !extent_op->update_flags);
@@ -3217,6 +3219,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = is_data,
 			.is_inc = false,
+			.generation = btrfs_extent_generation(leaf, ei),
 		};
 
 		/* In this branch refs == 1 */
@@ -4850,6 +4853,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_simple_quota_delta delta = {
 		.root = root_objectid,
 		.num_bytes = ins->offset,
+		.generation = trans->transid,
 		.rsv_bytes = 0,
 		.is_data = true,
 		.is_inc = true,
fs/btrfs/fs.h +0 -0
index f76f450c2abf..da7b623ff15f 100644

fs/btrfs/fs.h struct btrfs_fs_info { · line 806

Quoted 5 lines
fs/btrfs/fs.h +2 -0
@@ -802,6 +802,8 @@ struct btrfs_fs_info {
 	spinlock_t eb_leak_lock;
 	struct list_head allocated_ebs;
 #endif
+
+	u64 quota_enable_gen;

Please move it to the other quota/qgroup related members, at the end of fs_info there's only debugging stuff.

Quoted 56 lines
fs/btrfs/fs.h +0 -0
 };
 
 static inline void btrfs_set_last_root_drop_gen(struct btrfs_fs_info *fs_info,
fs/btrfs/qgroup.c +12 -2
index 58e9ed0deedd..a8a603242431 100644
@@ -454,6 +454,8 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 			}
 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, ptr);
 			simple = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
+			if (simple)
+				fs_info->quota_enable_gen = btrfs_qgroup_status_enable_gen(l, ptr);
 			if (btrfs_qgroup_status_generation(l, ptr) !=
 			    fs_info->generation && !simple) {
 				qgroup_mark_inconsistent(fs_info);
@@ -1107,10 +1109,12 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON;
-	if (simple)
+	if (simple) {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
-	else
+		btrfs_set_qgroup_status_enable_gen(leaf, ptr, trans->transid);
+	} else {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
+	}
 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
@@ -1202,6 +1206,8 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 		goto out_free_path;
 	}
 
+	fs_info->quota_enable_gen = trans->transid;
+
 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
 	/*
 	 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
@@ -4622,6 +4628,10 @@ int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
 	if (!is_fstree(root))
 		return 0;
 
+	/* If the extent predates enabling quotas, don't count it. */
+	if (delta->generation < fs_info->quota_enable_gen)
+		return 0;
+
 	spin_lock(&fs_info->qgroup_lock);
 	qgroup = find_qgroup_rb(fs_info, root);
 	if (!qgroup) {
fs/btrfs/qgroup.h +0 -0
index ce6fa8694ca7..ae1ce14b365c 100644

fs/btrfs/qgroup.h struct btrfs_simple_quota_delta { · line 244

Quoted 4 lines
fs/btrfs/qgroup.h +1 -0
@@ -241,6 +241,7 @@ struct btrfs_simple_quota_delta {
 	u64 rsv_bytes; /* The number of bytes reserved for this extent */
 	bool is_inc; /* Whether we are using or freeing the extent */
 	bool is_data; /* Whether the extent is data or metadata */
+	u64 generation; /* The generation the extent was created in */

Please reorder it so it does not leave gaps between struct members.

Quoted 14 lines
fs/btrfs/qgroup.h +0 -0
 };
 
 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
include/uapi/linux/btrfs_tree.h +3 -0
index eacb26caf3c6..1120ce3dae42 100644
@@ -1242,6 +1242,13 @@ struct btrfs_qgroup_status_item {
 	 * of the scan. It contains a logical address
 	 */
 	__le64 rescan;
+
+	/*
+	 * the generation when quotas are enabled. Used by simple quotas to
Quoted 8 lines

include/uapi/linux/btrfs_tree.h struct btrfs_qgroup_status_item { · line 1251

Quoted 4 lines
include/uapi/linux/btrfs_tree.h +4 -0
+	 * avoid decrementing when freeing an extent that was written before
+	 * enable.
+	 */
+	__le64 enable_gen;

This is public interface and btrfs_qgroup_status_item is used in many places in user space at least in btrfs-progs. This needs a lot of sanity checks.

Totally agreed in principle, but not exactly sure how to proceed in practice. I would definitely appreciate some tips/help!

How we interact with the new field:
- When enabling squota, set it, the incompat bit, and the status flag
- When reading in the qgroup status_item, if the status flag is set, then read the enable_gen.

I believe this prevents us from ever reading garbage while trying to read an old fs (status flag won't be set) and it prevents any btrfs-progs from getting confused by a wrong-sized status item, since it would choke on the incompat bit first.

Am I missing some other case? I can try to make it more explicitly zeroed when we enable qgroups but not squotas? I can add an ASSERT that the incompat bit is set as expected when we read the status item with the flag on (that seems good no matter what)?

Quoted 4 lines

I can also write a wrapper for getting it which does the incompat/status flag checking to make it more clear that it isn't safe to read in general. Or a comment on the struct saying it depends on the incompat bit?

All of the above makes sense and I had something like that in mind when writing the comment. The wrappers can make sure the bit is set when reading the item. I think there's an example in existing code that versions an item based on size, I can't find it now (probably something from the send/receive time where several new struct members were added).

I just noticed we have versioning for the qgoup status item, BTRFS_QGROUP_STATUS_VERSION is now 1 and has backward compatibility handling. We can probably use version 2 for squotas, in addition to the helpers with sanity checks.

#42 lore raw

Re: [PATCH v5 15/18] btrfs: check generation when recording simple quota delta

Quoted 211 lines · David Sterba
Quoted 199 lines · Boris Burkov
Quoted 171 lines · David Sterba
Quoted 84 lines · Boris Burkov

Simple quotas count extents only from the moment the feature is enabled. Therefore, if we do something like:
1. create subvol S
2. write F in S
3. enable quotas
4. remove F
5. write G in S

then after 3. and 4. we would expect the simple quota usage of S to be 0 (putting aside some metadata extents that might be written) and after 5., it should be the size of G plus metadata. Therefore, we need to be able to determine whether a particular quota delta we are processing predates simple quota enablement.

To do this, store the transaction id when quotas were enabled. In fs_info for immediate use and in the quota status item to make it recoverable on mount. When we see a delta, check if the generation of the extent item is less than that of quota enablement. If so, we should ignore the delta from this extent.

---
 fs/btrfs/accessors.h            |  2 ++
 fs/btrfs/extent-tree.c          |  4 ++++
 fs/btrfs/fs.h                   |  2 ++
 fs/btrfs/qgroup.c               | 14 ++++++++++++--
 fs/btrfs/qgroup.h               |  1 +
 include/uapi/linux/btrfs_tree.h |  7 +++++++
 6 files changed, 28 insertions(+), 2 deletions(-)
fs/btrfs/accessors.h +2 -0
index a23045c05937..513f8edbd98e 100644
@@ -970,6 +970,8 @@ BTRFS_SETGET_FUNCS(qgroup_status_flags, struct btrfs_qgroup_status_item,
 		   flags, 64);
 BTRFS_SETGET_FUNCS(qgroup_status_rescan, struct btrfs_qgroup_status_item,
 		   rescan, 64);
+BTRFS_SETGET_FUNCS(qgroup_status_enable_gen, struct btrfs_qgroup_status_item,
+		   enable_gen, 64);
 
 /* btrfs_qgroup_info_item */
 BTRFS_SETGET_FUNCS(qgroup_info_generation, struct btrfs_qgroup_info_item,
fs/btrfs/extent-tree.c +4 -0
index 1b5efd03ef83..395ab46e520b 100644
@@ -1513,6 +1513,7 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = href->reserved_bytes,
 			.is_data = true,
 			.is_inc	= true,
+			.generation = trans->transid,
 		};
 
 		if (extent_op)
@@ -1676,6 +1677,7 @@ static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = false,
 			.is_inc = true,
+			.generation = trans->transid,
 		};
 
 		BUG_ON(!extent_op || !extent_op->update_flags);
@@ -3217,6 +3219,7 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 			.rsv_bytes = 0,
 			.is_data = is_data,
 			.is_inc = false,
+			.generation = btrfs_extent_generation(leaf, ei),
 		};
 
 		/* In this branch refs == 1 */
@@ -4850,6 +4853,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_simple_quota_delta delta = {
 		.root = root_objectid,
 		.num_bytes = ins->offset,
+		.generation = trans->transid,
 		.rsv_bytes = 0,
 		.is_data = true,
 		.is_inc = true,
fs/btrfs/fs.h +0 -0
index f76f450c2abf..da7b623ff15f 100644

fs/btrfs/fs.h struct btrfs_fs_info { · line 806

Quoted 5 lines
fs/btrfs/fs.h +2 -0
@@ -802,6 +802,8 @@ struct btrfs_fs_info {
 	spinlock_t eb_leak_lock;
 	struct list_head allocated_ebs;
 #endif
+
+	u64 quota_enable_gen;

Please move it to the other quota/qgroup related members, at the end of fs_info there's only debugging stuff.

Quoted 56 lines
fs/btrfs/fs.h +0 -0
 };
 
 static inline void btrfs_set_last_root_drop_gen(struct btrfs_fs_info *fs_info,
fs/btrfs/qgroup.c +12 -2
index 58e9ed0deedd..a8a603242431 100644
@@ -454,6 +454,8 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 			}
 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, ptr);
 			simple = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
+			if (simple)
+				fs_info->quota_enable_gen = btrfs_qgroup_status_enable_gen(l, ptr);
 			if (btrfs_qgroup_status_generation(l, ptr) !=
 			    fs_info->generation && !simple) {
 				qgroup_mark_inconsistent(fs_info);
@@ -1107,10 +1109,12 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON;
-	if (simple)
+	if (simple) {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_SIMPLE;
-	else
+		btrfs_set_qgroup_status_enable_gen(leaf, ptr, trans->transid);
+	} else {
 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
+	}
 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
@@ -1202,6 +1206,8 @@ int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
 		goto out_free_path;
 	}
 
+	fs_info->quota_enable_gen = trans->transid;
+
 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
 	/*
 	 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
@@ -4622,6 +4628,10 @@ int btrfs_record_simple_quota_delta(struct btrfs_fs_info *fs_info,
 	if (!is_fstree(root))
 		return 0;
 
+	/* If the extent predates enabling quotas, don't count it. */
+	if (delta->generation < fs_info->quota_enable_gen)
+		return 0;
+
 	spin_lock(&fs_info->qgroup_lock);
 	qgroup = find_qgroup_rb(fs_info, root);
 	if (!qgroup) {
fs/btrfs/qgroup.h +0 -0
index ce6fa8694ca7..ae1ce14b365c 100644

fs/btrfs/qgroup.h struct btrfs_simple_quota_delta { · line 244

Quoted 4 lines
fs/btrfs/qgroup.h +1 -0
@@ -241,6 +241,7 @@ struct btrfs_simple_quota_delta {
 	u64 rsv_bytes; /* The number of bytes reserved for this extent */
 	bool is_inc; /* Whether we are using or freeing the extent */
 	bool is_data; /* Whether the extent is data or metadata */
+	u64 generation; /* The generation the extent was created in */

Please reorder it so it does not leave gaps between struct members.

Quoted 14 lines
fs/btrfs/qgroup.h +0 -0
 };
 
 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
include/uapi/linux/btrfs_tree.h +3 -0
index eacb26caf3c6..1120ce3dae42 100644
@@ -1242,6 +1242,13 @@ struct btrfs_qgroup_status_item {
 	 * of the scan. It contains a logical address
 	 */
 	__le64 rescan;
+
+	/*
+	 * the generation when quotas are enabled. Used by simple quotas to
Quoted 8 lines

include/uapi/linux/btrfs_tree.h struct btrfs_qgroup_status_item { · line 1251

Quoted 4 lines
include/uapi/linux/btrfs_tree.h +4 -0
+	 * avoid decrementing when freeing an extent that was written before
+	 * enable.
+	 */
+	__le64 enable_gen;

This is public interface and btrfs_qgroup_status_item is used in many places in user space at least in btrfs-progs. This needs a lot of sanity checks.

Totally agreed in principle, but not exactly sure how to proceed in practice. I would definitely appreciate some tips/help!

How we interact with the new field:
- When enabling squota, set it, the incompat bit, and the status flag
- When reading in the qgroup status_item, if the status flag is set, then read the enable_gen.

I believe this prevents us from ever reading garbage while trying to read an old fs (status flag won't be set) and it prevents any btrfs-progs from getting confused by a wrong-sized status item, since it would choke on the incompat bit first.

Am I missing some other case? I can try to make it more explicitly zeroed when we enable qgroups but not squotas? I can add an ASSERT that the incompat bit is set as expected when we read the status item with the flag on (that seems good no matter what)?

Quoted 4 lines

I can also write a wrapper for getting it which does the incompat/status flag checking to make it more clear that it isn't safe to read in general. Or a comment on the struct saying it depends on the incompat bit?

All of the above makes sense and I had something like that in mind when writing the comment. The wrappers can make sure the bit is set when reading the item. I think there's an example in existing code that versions an item based on size, I can't find it now (probably something from the send/receive time where several new struct members were added).

Quoted 4 lines

I just noticed we have versioning for the qgoup status item, BTRFS_QGROUP_STATUS_VERSION is now 1 and has backward compatibility handling. We can probably use version 2 for squotas, in addition to the helpers with sanity checks.

I made the helper/validation changes in V6, but forgot to address the BTRFS_QGROUP_STATUS_VERSION idea. Right now, that field is for preventing forward compatibility as well as backward (for lack of a better term?) The check on it is a !=, so if you bump the version, you can no longer honor old fs-es qgroups, which is not the case with this change. Since the version has never been bumped, I believe we can safely change it to a backwards compatibility check and use it for squotas.

#43 lore raw

PATCH v5 16/18 btrfs: track metadata relocation cow with simple quota

In reply to Boris Burkov

Relocation cows metadata blocks in two cases for the reloc root:
- copying the subvol root item when creating the reloc root
- copying a btree node when there is a cow during relocation

In both cases, the resulting btree node hits an abnormal code path with respect to the owner field in its btrfs_header. It first creates the root item for the new objectid, which populates the reloc root id, and it at this point that delayed refs are created.

Later, it fully copies the old node into the new node (including the original owner field) which overwrites it. This results in a simple quotas mismatch where we run the delayed ref for the reloc root which has no simple quota effect (reloc root is not an fstree) but when we ultimately delete the node, the owner is the real original fstree and we do free the space.

To work around this without tampering with the behavior of relocation, add a parameter to btrfs_add_tree_block that lets the relocation code path specify a different owning root than the "operating" root (in this case, owning root is the real root and the operating root is the reloc root). These can naturally be plumbed into delayed refs that have the same concept.

Note that this is a double count in some sense, but a relatively natural one, as there are really two extents, and the old one will be deleted soon. This is consistent with how data relocation extents are accounted by simple quotas.

---
 fs/btrfs/ctree.c       | 22 ++++++++++++++--------
 fs/btrfs/disk-io.c     |  4 ++--
 fs/btrfs/extent-tree.c |  8 ++++++--
 fs/btrfs/extent-tree.h |  3 ++-
 fs/btrfs/ioctl.c       |  2 +-
 5 files changed, 25 insertions(+), 14 deletions(-)
fs/btrfs/ctree.c +14 -8
index a4cb4b642987..cb0d4535de37 100644
@@ -316,6 +316,7 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 	int ret = 0;
 	int level;
 	struct btrfs_disk_key disk_key;
+	u64 reloc_src_root = 0;
 
 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 		trans->transid != fs_info->running_transaction->transid);
@@ -328,9 +329,11 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 	else
 		btrfs_node_key(buf, &disk_key, 0);
 
+	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
+		reloc_src_root = btrfs_header_owner(buf);
 	cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
 				     &disk_key, level, buf->start, 0,
-				     BTRFS_NESTING_NEW_ROOT);
+				     BTRFS_NESTING_NEW_ROOT, reloc_src_root);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -522,6 +525,7 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	int last_ref = 0;
 	int unlock_orig = 0;
 	u64 parent_start = 0;
+	u64 reloc_src_root = 0;
 
 	if (*cow_ret == buf)
 		unlock_orig = 1;
@@ -540,12 +544,14 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	else
 		btrfs_node_key(buf, &disk_key, 0);
 
-	if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
-		parent_start = parent->start;
-
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
+		if (parent)
+			parent_start = parent->start;
+		reloc_src_root = btrfs_header_owner(buf);
+	}
 	cow = btrfs_alloc_tree_block(trans, root, parent_start,
 				     root->root_key.objectid, &disk_key, level,
-				     search_start, empty_size, nest);
+				     search_start, empty_size, nest, reloc_src_root);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -2956,7 +2962,7 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 
 	c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
 				   &lower_key, level, root->node->start, 0,
-				   BTRFS_NESTING_NEW_ROOT);
+				   BTRFS_NESTING_NEW_ROOT, 0);
 	if (IS_ERR(c))
 		return PTR_ERR(c);
 
@@ -3100,7 +3106,7 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 
 	split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
 				       &disk_key, level, c->start, 0,
-				       BTRFS_NESTING_SPLIT);
+				       BTRFS_NESTING_SPLIT, 0);
 	if (IS_ERR(split))
 		return PTR_ERR(split);
 
@@ -3853,7 +3859,7 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
 	right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
 				       &disk_key, 0, l->start, 0,
 				       num_doubles ? BTRFS_NESTING_NEW_ROOT :
-				       BTRFS_NESTING_SPLIT);
+				       BTRFS_NESTING_SPLIT, 0);
 	if (IS_ERR(right))
 		return PTR_ERR(right);
 
fs/btrfs/disk-io.c +2 -2
index b4495d4c1533..e2b0e11800fc 100644
@@ -862,7 +862,7 @@ struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 	root->root_key.offset = 0;
 
 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
-				      BTRFS_NESTING_NORMAL);
+				      BTRFS_NESTING_NORMAL, 0);
 	if (IS_ERR(leaf)) {
 		ret = PTR_ERR(leaf);
 		leaf = NULL;
@@ -939,7 +939,7 @@ int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
 	 */
 
 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
-			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
+			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL, 0);
 	if (IS_ERR(leaf))
 		return PTR_ERR(leaf);
 
fs/btrfs/extent-tree.c +6 -2
index 395ab46e520b..50db75529a83 100644
@@ -4989,7 +4989,8 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 					     const struct btrfs_disk_key *key,
 					     int level, u64 hint,
 					     u64 empty_size,
-					     enum btrfs_lock_nesting nest)
+					     enum btrfs_lock_nesting nest,
+					     u64 reloc_src_root)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_key ins;
@@ -5001,6 +5002,7 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 	int ret;
 	u32 blocksize = fs_info->nodesize;
 	bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
+	u64 owning_root;
 
 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 	if (btrfs_is_testing(fs_info)) {
@@ -5027,11 +5029,13 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 		ret = PTR_ERR(buf);
 		goto out_free_reserved;
 	}
+	owning_root = btrfs_header_owner(buf);
 
 	if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
 		if (parent == 0)
 			parent = ins.objectid;
 		flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
+		owning_root = reloc_src_root;
 	} else
 		BUG_ON(parent > 0);
 
@@ -5051,7 +5055,7 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 		extent_op->level = level;
 
 		btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
-				       ins.objectid, ins.offset, parent, btrfs_header_owner(buf));
+				       ins.objectid, ins.offset, parent, owning_root);
 		btrfs_init_tree_ref(&generic_ref, level, root_objectid,
 				    root->root_key.objectid, false);
 		btrfs_ref_tree_mod(fs_info, &generic_ref);
fs/btrfs/extent-tree.h +2 -1
index 7c27652880a2..99b11e278ae4 100644
@@ -118,7 +118,8 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 					     const struct btrfs_disk_key *key,
 					     int level, u64 hint,
 					     u64 empty_size,
-					     enum btrfs_lock_nesting nest);
+					     enum btrfs_lock_nesting nest,
+					     u64 reloc_src_root);
 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 			   u64 root_id,
 			   struct extent_buffer *buf,
fs/btrfs/ioctl.c +1 -1
index c9b069077fd0..f3807def6596 100644
@@ -657,7 +657,7 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 		goto out;
 
 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
-				      BTRFS_NESTING_NORMAL);
+				      BTRFS_NESTING_NORMAL, 0);
 	if (IS_ERR(leaf)) {
 		ret = PTR_ERR(leaf);
 		goto out;
Signature · 1 line
2.41.0
#44 lore raw

Re: [PATCH v5 16/18] btrfs: track metadata relocation cow with simple quota

Quoted 144 lines · Boris Burkov

Relocation cows metadata blocks in two cases for the reloc root:
- copying the subvol root item when creating the reloc root
- copying a btree node when there is a cow during relocation

In both cases, the resulting btree node hits an abnormal code path with respect to the owner field in its btrfs_header. It first creates the root item for the new objectid, which populates the reloc root id, and it at this point that delayed refs are created.

Later, it fully copies the old node into the new node (including the original owner field) which overwrites it. This results in a simple quotas mismatch where we run the delayed ref for the reloc root which has no simple quota effect (reloc root is not an fstree) but when we ultimately delete the node, the owner is the real original fstree and we do free the space.

To work around this without tampering with the behavior of relocation, add a parameter to btrfs_add_tree_block that lets the relocation code path specify a different owning root than the "operating" root (in this case, owning root is the real root and the operating root is the reloc root). These can naturally be plumbed into delayed refs that have the same concept.

Note that this is a double count in some sense, but a relatively natural one, as there are really two extents, and the old one will be deleted soon. This is consistent with how data relocation extents are accounted by simple quotas.

---
 fs/btrfs/ctree.c       | 22 ++++++++++++++--------
 fs/btrfs/disk-io.c     |  4 ++--
 fs/btrfs/extent-tree.c |  8 ++++++--
 fs/btrfs/extent-tree.h |  3 ++-
 fs/btrfs/ioctl.c       |  2 +-
 5 files changed, 25 insertions(+), 14 deletions(-)
fs/btrfs/ctree.c +14 -8
index a4cb4b642987..cb0d4535de37 100644
@@ -316,6 +316,7 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 	int ret = 0;
 	int level;
 	struct btrfs_disk_key disk_key;
+	u64 reloc_src_root = 0;
 
 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 		trans->transid != fs_info->running_transaction->transid);
@@ -328,9 +329,11 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 	else
 		btrfs_node_key(buf, &disk_key, 0);
 
+	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
+		reloc_src_root = btrfs_header_owner(buf);
 	cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
 				     &disk_key, level, buf->start, 0,
-				     BTRFS_NESTING_NEW_ROOT);
+				     BTRFS_NESTING_NEW_ROOT, reloc_src_root);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -522,6 +525,7 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	int last_ref = 0;
 	int unlock_orig = 0;
 	u64 parent_start = 0;
+	u64 reloc_src_root = 0;
 
 	if (*cow_ret == buf)
 		unlock_orig = 1;
@@ -540,12 +544,14 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 	else
 		btrfs_node_key(buf, &disk_key, 0);
 
-	if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
-		parent_start = parent->start;
-
+	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
+		if (parent)
+			parent_start = parent->start;
+		reloc_src_root = btrfs_header_owner(buf);
+	}
 	cow = btrfs_alloc_tree_block(trans, root, parent_start,
 				     root->root_key.objectid, &disk_key, level,
-				     search_start, empty_size, nest);
+				     search_start, empty_size, nest, reloc_src_root);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -2956,7 +2962,7 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 
 	c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
 				   &lower_key, level, root->node->start, 0,
-				   BTRFS_NESTING_NEW_ROOT);
+				   BTRFS_NESTING_NEW_ROOT, 0);
 	if (IS_ERR(c))
 		return PTR_ERR(c);
 
@@ -3100,7 +3106,7 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 
 	split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
 				       &disk_key, level, c->start, 0,
-				       BTRFS_NESTING_SPLIT);
+				       BTRFS_NESTING_SPLIT, 0);
 	if (IS_ERR(split))
 		return PTR_ERR(split);
 
@@ -3853,7 +3859,7 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
 	right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
 				       &disk_key, 0, l->start, 0,
 				       num_doubles ? BTRFS_NESTING_NEW_ROOT :
-				       BTRFS_NESTING_SPLIT);
+				       BTRFS_NESTING_SPLIT, 0);
 	if (IS_ERR(right))
 		return PTR_ERR(right);
 
fs/btrfs/disk-io.c +2 -2
index b4495d4c1533..e2b0e11800fc 100644
@@ -862,7 +862,7 @@ struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 	root->root_key.offset = 0;
 
 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
-				      BTRFS_NESTING_NORMAL);
+				      BTRFS_NESTING_NORMAL, 0);
 	if (IS_ERR(leaf)) {
 		ret = PTR_ERR(leaf);
 		leaf = NULL;
@@ -939,7 +939,7 @@ int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
 	 */
 
 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
-			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
+			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL, 0);
 	if (IS_ERR(leaf))
 		return PTR_ERR(leaf);
 
fs/btrfs/extent-tree.c +0 -0
index 395ab46e520b..50db75529a83 100644

fs/btrfs/extent-tree.c struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, · line 4993

Quoted 6 lines
fs/btrfs/extent-tree.c +2 -1
@@ -4989,7 +4989,8 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 					     const struct btrfs_disk_key *key,
 					     int level, u64 hint,
 					     u64 empty_size,
-					     enum btrfs_lock_nesting nest)
+					     enum btrfs_lock_nesting nest,
+					     u64 reloc_src_root)

Please move the new parameter before 'nest'.

Quoted 63 lines
fs/btrfs/extent-tree.c +4 -1
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_key ins;
@@ -5001,6 +5002,7 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 	int ret;
 	u32 blocksize = fs_info->nodesize;
 	bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
+	u64 owning_root;
 
 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 	if (btrfs_is_testing(fs_info)) {
@@ -5027,11 +5029,13 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 		ret = PTR_ERR(buf);
 		goto out_free_reserved;
 	}
+	owning_root = btrfs_header_owner(buf);
 
 	if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
 		if (parent == 0)
 			parent = ins.objectid;
 		flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
+		owning_root = reloc_src_root;
 	} else
 		BUG_ON(parent > 0);
 
@@ -5051,7 +5055,7 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 		extent_op->level = level;
 
 		btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
-				       ins.objectid, ins.offset, parent, btrfs_header_owner(buf));
+				       ins.objectid, ins.offset, parent, owning_root);
 		btrfs_init_tree_ref(&generic_ref, level, root_objectid,
 				    root->root_key.objectid, false);
 		btrfs_ref_tree_mod(fs_info, &generic_ref);
fs/btrfs/extent-tree.h +2 -1
index 7c27652880a2..99b11e278ae4 100644
@@ -118,7 +118,8 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 					     const struct btrfs_disk_key *key,
 					     int level, u64 hint,
 					     u64 empty_size,
-					     enum btrfs_lock_nesting nest);
+					     enum btrfs_lock_nesting nest,
+					     u64 reloc_src_root);
 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 			   u64 root_id,
 			   struct extent_buffer *buf,
fs/btrfs/ioctl.c +1 -1
index c9b069077fd0..f3807def6596 100644
@@ -657,7 +657,7 @@ static noinline int create_subvol(struct mnt_idmap *idmap,
 		goto out;
 
 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
-				      BTRFS_NESTING_NORMAL);
+				      BTRFS_NESTING_NORMAL, 0);
 	if (IS_ERR(leaf)) {
 		ret = PTR_ERR(leaf);
 		goto out;
Signature · 1 line
2.41.0
#45 lore raw

PATCH v5 17/18 btrfs: track data relocation with simple quota

In reply to Boris Burkov

Relocation data allocations are quite tricky for simple quotas. The basic data relocation sequence is (ignoring details that aren't relevant to this fix):
- create a fake relocation data fs root
- create a fake relocation inode in that root
- foreach data extent:
- preallocate a data extent on behalf of the fake inode
- copy over the data
- foreach extent
- swap the refs so that the original file extent now refers to the new extent item
- drop the fake root, dropping its refs on the old extents, which lets us delete them.

Done naively, this results in storing an extent item in the extent tree whose owner_ref points at the relocation data root and a no-op squota recording, since the reloc root is not a legit fstree. So far, that's OK. The problem comes when you do the swap, and leave an extent item owned by this bogus root as the real permanent extents of the file. If the file then drops that ref, we free it and no-op account that against the fake relocation root. Essentially, this means that relocation is simple quota "extent laundering", since we re-own the extents into a fake root.

Simple quotas very intentionally doesn't have a mechanism for transferring ownership of extents, as that is exactly the complicated thing we are trying to avoid with the new design. Further, it cannot be correctly done in this case, since at the time you create the new "real" refs, there is no way to know which was the original owner before relocation unless we track it.

Therefore, it makes more sense to trick the preallocation to handle relocation as a special case and note the proper owner ref from the beginning. That way, we never write out an extent item without the correct owner ref that it will eventually have.

This could be done by wiring a special root parameter all the way through the allocation code path, but to avoid that special case touching all the code, take advantage of the serial nature of relocation to store the src root on the relocation root object. Then when we finish the prealloc, if it happens to be this case, prepare the delayed ref appropriately.

We must also add logic to handle relocating adjacent extents with different owning roots. Those cannot be preallocated together in a cluster as it would lose the separate ownership information.

This is obviously a smelly bit of code, but I think it is the best solution to the problem, given the relocation implementation.

---
 fs/btrfs/ctree.h       |  1 +
 fs/btrfs/extent-tree.c | 13 ++++++-----
 fs/btrfs/relocation.c  | 49 +++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 57 insertions(+), 6 deletions(-)
fs/btrfs/ctree.h +1 -0
index f2d2b313bde5..577186994188 100644
@@ -333,6 +333,7 @@ struct btrfs_root {
 #ifdef CONFIG_BTRFS_DEBUG
 	struct list_head leak_list;
 #endif
+	u64 relocation_src_root;
 };
 
 static inline bool btrfs_root_readonly(const struct btrfs_root *root)
fs/btrfs/extent-tree.c +8 -5
index 50db75529a83..eb132200833d 100644
@@ -58,7 +58,7 @@ static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 				      u64 parent, u64 root_objectid,
 				      u64 flags, u64 owner, u64 offset,
-				      struct btrfs_key *ins, int ref_mod);
+				      struct btrfs_key *ins, int ref_mod, u64 oref_root);
 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 				     struct btrfs_delayed_ref_node *node,
 				     struct btrfs_delayed_extent_op *extent_op);
@@ -1521,7 +1521,7 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 		ret = alloc_reserved_file_extent(trans, parent, ref_root,
 						 flags, ref->objectid,
 						 ref->offset, &ins,
-						 node->ref_mod);
+						 node->ref_mod, href->owning_root);
 		if (!ret)
 			ret = btrfs_record_simple_quota_delta(trans->fs_info, &delta);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
@@ -4669,7 +4669,7 @@ static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 				      u64 parent, u64 root_objectid,
 				      u64 flags, u64 owner, u64 offset,
-				      struct btrfs_key *ins, int ref_mod)
+				      struct btrfs_key *ins, int ref_mod, u64 oref_root)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_root *extent_root;
@@ -4717,7 +4717,7 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 	if (simple_quota) {
 		btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY);
 		oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
-		btrfs_set_extent_owner_ref_root_id(leaf, oref, root_objectid);
+		btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root);
 		iref = (struct btrfs_extent_inline_ref *)(oref + 1);
 	}
 	btrfs_set_extent_inline_ref_type(leaf, iref, type);
@@ -4828,6 +4828,9 @@ int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 
 	BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
 
+	if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root))
+		owning_root = root->relocation_src_root;
+
 	btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
 			       ins->objectid, ins->offset, 0, owning_root);
 	btrfs_init_data_ref(&generic_ref, root_objectid, owner,
@@ -4883,7 +4886,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	spin_unlock(&space_info->lock);
 
 	ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
-					 offset, ins, 1);
+					 offset, ins, 1, root_objectid);
 	if (ret)
 		btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
 	ret = btrfs_record_simple_quota_delta(fs_info, &delta);
fs/btrfs/relocation.c +48 -1

74 lines not on this page · raw message

Signature · 1 line
2.41.0
Reviewed-by on v5 17/18
#46 lore raw

Re: [PATCH v5 17/18] btrfs: track data relocation with simple quota

Quoted 43 lines · Boris Burkov

Relocation data allocations are quite tricky for simple quotas. The basic data relocation sequence is (ignoring details that aren't relevant to this fix):
- create a fake relocation data fs root
- create a fake relocation inode in that root
- foreach data extent:
- preallocate a data extent on behalf of the fake inode
- copy over the data
- foreach extent
- swap the refs so that the original file extent now refers to the new extent item
- drop the fake root, dropping its refs on the old extents, which lets us delete them.

Done naively, this results in storing an extent item in the extent tree whose owner_ref points at the relocation data root and a no-op squota recording, since the reloc root is not a legit fstree. So far, that's OK. The problem comes when you do the swap, and leave an extent item owned by this bogus root as the real permanent extents of the file. If the file then drops that ref, we free it and no-op account that against the fake relocation root. Essentially, this means that relocation is simple quota "extent laundering", since we re-own the extents into a fake root.

Simple quotas very intentionally doesn't have a mechanism for transferring ownership of extents, as that is exactly the complicated thing we are trying to avoid with the new design. Further, it cannot be correctly done in this case, since at the time you create the new "real" refs, there is no way to know which was the original owner before relocation unless we track it.

Therefore, it makes more sense to trick the preallocation to handle relocation as a special case and note the proper owner ref from the beginning. That way, we never write out an extent item without the correct owner ref that it will eventually have.

This could be done by wiring a special root parameter all the way through the allocation code path, but to avoid that special case touching all the code, take advantage of the serial nature of relocation to store the src root on the relocation root object. Then when we finish the prealloc, if it happens to be this case, prepare the delayed ref appropriately.

Quoted 8 lines

We must also add logic to handle relocating adjacent extents with different owning roots. Those cannot be preallocated together in a cluster as it would lose the separate ownership information.

This is obviously a smelly bit of code, but I think it is the best solution to the problem, given the relocation implementation.

Thanks,

Josef

#47 lore raw

PATCH v5 18/18 btrfs: only set QUOTA_ENABLED when done reading qgroups

In reply to Boris Burkov

In open_ctree, we set BTRFS_FS_QUOTA_ENABLED as soon as we see a quota_root, as opposed to after we are done setting up the qgroup structures. In the quota_enable path, we wait until after the structures are set up. Likewise, in disable, we clear the bit before tearing down the structures. I feel that this organization is less surprising for the open_ctree path.

I don't believe this fixes any actual bug, but avoids potential confusion when using btrfs_qgroup_mode in an intermediate state where we are enabled but haven't yet setup the qgroup status flags. It also avoids any risk of calling a qgroup function and attempting to use the qgroup rbtrees before they exist/are setup.

This all occurs before we do rw setup, so I believe it should be mostly a no-op.

---
 fs/btrfs/disk-io.c |  1 -
 fs/btrfs/qgroup.c  | 15 +++++++--------
 2 files changed, 7 insertions(+), 9 deletions(-)
fs/btrfs/disk-io.c +0 -1
index e2b0e11800fc..874685c84df2 100644
@@ -2254,7 +2254,6 @@ static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
 	root = btrfs_read_tree_root(tree_root, &location);
 	if (!IS_ERR(root)) {
 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
-		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 		fs_info->quota_root = root;
 	}
 
fs/btrfs/qgroup.c +7 -8
index a8a603242431..1f915d70b99d 100644
@@ -402,7 +402,7 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 	u64 rescan_progress = 0;
 	bool simple;
 
-	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
+	if (!fs_info->quota_root)
 		return 0;
 
 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
@@ -565,13 +565,12 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 out:
 	btrfs_free_path(path);
 	fs_info->qgroup_flags |= flags;
-	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
-		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
-	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
-		 ret >= 0)
-		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
-
-	if (ret < 0) {
+	if (ret >= 0) {
+		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)
+			set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
+		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
+			ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
+	} else {
 		ulist_free(fs_info->qgroup_ulist);
 		fs_info->qgroup_ulist = NULL;
 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
Signature · 1 line
2.41.0
Reviewed-by on v5 18/18
#48 lore raw

Re: [PATCH v5 18/18] btrfs: only set QUOTA_ENABLED when done reading qgroups

Quoted 13 lines · Boris Burkov

In open_ctree, we set BTRFS_FS_QUOTA_ENABLED as soon as we see a quota_root, as opposed to after we are done setting up the qgroup structures. In the quota_enable path, we wait until after the structures are set up. Likewise, in disable, we clear the bit before tearing down the structures. I feel that this organization is less surprising for the open_ctree path.

I don't believe this fixes any actual bug, but avoids potential confusion when using btrfs_qgroup_mode in an intermediate state where we are enabled but haven't yet setup the qgroup status flags. It also avoids any risk of calling a qgroup function and attempting to use the qgroup rbtrees before they exist/are setup.

Quoted 4 lines

This all occurs before we do rw setup, so I believe it should be mostly a no-op.

Thanks,

Josef

#49 lore raw
In reply to Boris Burkov
Quoted 39 lines · Boris Burkov

btrfs quota groups (qgroups) are a compelling feature of btrfs that allow flexible control for limiting subvolume data and metadata usage. However, due to btrfs's high level decision to tradeoff snapshot performance against ref-counting performance, qgroups suffer from non-trivial performance issues that make them unattractive in certain workloads. Particularly, frequent backref walking during writes and during commits can make operations increasingly expensive as the number of snapshots scales up. For that reason, we have never been able to commit to using qgroups in production at Meta, despite significant interest from people running container workloads, where we would benefit from protecting the rest of the host from a buggy application in a container running away with disk usage. This patch series introduces a simplified version of qgroups called
simple quotas (squotas) which never computes global reference counts for extents, and thus has similar performance characteristics to normal, quotas disabled, btrfs. The "trick" is that in simple quotas mode, we account all extents permanently to the subvolume in which they were originally created. That allows us to make all accounting 1:1 with extent item lifetime, removing the need to walk backrefs. However, this sacrifices the ability to compute shared vs. exclusive usage. It also results in counter-intuitive, though still predictable and simple accounting in the cases where an original extent is removed while a shared copy still exists. Qgroups is able to detect that case and count the remaining copy as an exclusive owner, while squotas is not. As a result, squotas works best when the original extent is immutable and outlives any clones.

==Format Change==
In order to track the original creating subvolume of a data extent in the face of reflinks, it is necessary to add additional accounting to the extent item. To save space, this is done with a new inline ref item. However, the downside of this approach is that it makes enabling squota an incompat change, denoted by the new incompat bit SIMPLE_QUOTA. When this bit is set and quotas are enabled, new extent items get the extra accounting, and freed extent items check for the accounting to find their creating subvolume. In addition, 1:1 with this incompat bit, the quota status item now tracks a "quota enablement generation" needed for properly handling deleting extents with predate enablement.

Quoted 2 lines

==API==
Squotas reuses the api of qgroups.

So apart from the accounting, the hierarchy of qgroups can be still built as before, right? In the example you create a group 1/100 so I assume that it's still qgroups from the outside, and that the limits can be set.

Because if not, then squotas would make more sense as a separate infrastructure, under quotas. Like that quotas are the abstraction while qgroups or squota would be the implementation.

Quoted 7 lines

The only difference is that when you
enable quotas via btrfs quota enable, you pass the --simple flag. Squotas will always report exclusive == shared for each qgroup. Squotas deal with extent_item/metadata_item sizes and thus do not do anything special with compression. Squotas also introduce auto inheritance for nested subvols. The API is documented more fully in the documentation patches in btrfs-progs.

The lack of exclusive size sharing will be confusing I guess, so we need to make it clear in the documentation and in the UI that it's either full or simple mode.

I've added the patchset to for-next, we may need an iteration or two to fix some issues I've seen so far but on the fundamental level I think it's ok.

#50 lore raw
Quoted 40 lines · David Sterba
Quoted 39 lines · Boris Burkov

btrfs quota groups (qgroups) are a compelling feature of btrfs that allow flexible control for limiting subvolume data and metadata usage. However, due to btrfs's high level decision to tradeoff snapshot performance against ref-counting performance, qgroups suffer from non-trivial performance issues that make them unattractive in certain workloads. Particularly, frequent backref walking during writes and during commits can make operations increasingly expensive as the number of snapshots scales up. For that reason, we have never been able to commit to using qgroups in production at Meta, despite significant interest from people running container workloads, where we would benefit from protecting the rest of the host from a buggy application in a container running away with disk usage. This patch series introduces a simplified version of qgroups called
simple quotas (squotas) which never computes global reference counts for extents, and thus has similar performance characteristics to normal, quotas disabled, btrfs. The "trick" is that in simple quotas mode, we account all extents permanently to the subvolume in which they were originally created. That allows us to make all accounting 1:1 with extent item lifetime, removing the need to walk backrefs. However, this sacrifices the ability to compute shared vs. exclusive usage. It also results in counter-intuitive, though still predictable and simple accounting in the cases where an original extent is removed while a shared copy still exists. Qgroups is able to detect that case and count the remaining copy as an exclusive owner, while squotas is not. As a result, squotas works best when the original extent is immutable and outlives any clones.

==Format Change==
In order to track the original creating subvolume of a data extent in the face of reflinks, it is necessary to add additional accounting to the extent item. To save space, this is done with a new inline ref item. However, the downside of this approach is that it makes enabling squota an incompat change, denoted by the new incompat bit SIMPLE_QUOTA. When this bit is set and quotas are enabled, new extent items get the extra accounting, and freed extent items check for the accounting to find their creating subvolume. In addition, 1:1 with this incompat bit, the quota status item now tracks a "quota enablement generation" needed for properly handling deleting extents with predate enablement.

Quoted 7 lines
Quoted 2 lines

==API==
Squotas reuses the api of qgroups.

So apart from the accounting, the hierarchy of qgroups can be still built as before, right? In the example you create a group 1/100 so I assume that it's still qgroups from the outside, and that the limits can be set.

Yes, you can create quota group hierarchies with the same nesting behavior. I am only changing the accounting methodology (and added auto hierarchy)

Quoted 13 lines

Because if not, then squotas would make more sense as a separate infrastructure, under quotas. Like that quotas are the abstraction while qgroups or squota would be the implementation.

Quoted 7 lines

The only difference is that when you
enable quotas via btrfs quota enable, you pass the --simple flag. Squotas will always report exclusive == shared for each qgroup. Squotas deal with extent_item/metadata_item sizes and thus do not do anything special with compression. Squotas also introduce auto inheritance for nested subvols. The API is documented more fully in the documentation patches in btrfs-progs.

Quoted 3 lines

The lack of exclusive size sharing will be confusing I guess, so we need to make it clear in the documentation and in the UI that it's either full or simple mode.

I am happy to iterate on that. I think always reporting as shared=0, since the *ownership* is exclusive. I opted for making them equal since it sort of both shared usage (we don't know if it's shared nor when it will be freed) and exclusive usage (belongs to this subvol by owner ref)

Quoted 4 lines

I've added the patchset to for-next, we may need an iteration or two to fix some issues I've seen so far but on the fundamental level I think it's ok.

#51 lore raw
Quoted 44 lines · Boris Burkov
Quoted 43 lines · David Sterba
Quoted 41 lines · Boris Burkov

btrfs quota groups (qgroups) are a compelling feature of btrfs that allow flexible control for limiting subvolume data and metadata usage. However, due to btrfs's high level decision to tradeoff snapshot performance against ref-counting performance, qgroups suffer from non-trivial performance issues that make them unattractive in certain workloads. Particularly, frequent backref walking during writes and during commits can make operations increasingly expensive as the number of snapshots scales up. For that reason, we have never been able to commit to using qgroups in production at Meta, despite significant interest from people running container workloads, where we would benefit from protecting the rest of the host from a buggy application in a container running away with disk usage. This patch series introduces a simplified version of qgroups called
simple quotas (squotas) which never computes global reference counts for extents, and thus has similar performance characteristics to normal, quotas disabled, btrfs. The "trick" is that in simple quotas mode, we account all extents permanently to the subvolume in which they were originally created. That allows us to make all accounting 1:1 with extent item lifetime, removing the need to walk backrefs. However, this sacrifices the ability to compute shared vs. exclusive usage. It also results in counter-intuitive, though still predictable and simple accounting in the cases where an original extent is removed while a shared copy still exists. Qgroups is able to detect that case and count the remaining copy as an exclusive owner, while squotas is not. As a result, squotas works best when the original extent is immutable and outlives any clones.

==Format Change==
In order to track the original creating subvolume of a data extent in the face of reflinks, it is necessary to add additional accounting to the extent item. To save space, this is done with a new inline ref item. However, the downside of this approach is that it makes enabling squota an incompat change, denoted by the new incompat bit SIMPLE_QUOTA. When this bit is set and quotas are enabled, new extent items get the extra accounting, and freed extent items check for the accounting to find their creating subvolume. In addition, 1:1 with this incompat bit, the quota status item now tracks a "quota enablement generation" needed for properly handling deleting extents with predate enablement.

==API==
Squotas reuses the api of qgroups.

Quoted 8 lines
Quoted 4 lines

So apart from the accounting, the hierarchy of qgroups can be still built as before, right? In the example you create a group 1/100 so I assume that it's still qgroups from the outside, and that the limits can be set.

Yes, you can create quota group hierarchies with the same nesting behavior. I am only changing the accounting methodology (and added auto hierarchy)

OK, makes sense. The hierarchy does not need to be used and is probably less practical for the simple accounting. What I had in mind was some kind of flat hierarchy, now that the simple accounting is there. People were asking about that in the past, wit the drawback of lack of shared/exclusive accounting. Adding a separate subcommands and tooling around flat quotas could be done but with squotas as well, just "don't use the hierarchy".

Quoted 12 lines
Quoted 12 lines

Because if not, then squotas would make more sense as a separate infrastructure, under quotas. Like that quotas are the abstraction while qgroups or squota would be the implementation.

Quoted 7 lines

The only difference is that when you
enable quotas via btrfs quota enable, you pass the --simple flag. Squotas will always report exclusive == shared for each qgroup. Squotas deal with extent_item/metadata_item sizes and thus do not do anything special with compression. Squotas also introduce auto inheritance for nested subvols. The API is documented more fully in the documentation patches in btrfs-progs.

Quoted 8 lines
Quoted 3 lines

The lack of exclusive size sharing will be confusing I guess, so we need to make it clear in the documentation and in the UI that it's either full or simple mode.

I am happy to iterate on that. I think always reporting as shared=0, since the *ownership* is exclusive. I opted for making them equal since it sort of both shared usage (we don't know if it's shared nor when it will be freed) and exclusive usage (belongs to this subvol by owner ref)

I agree with that reasoning.

#52 lore raw
In reply to David Sterba
Quoted 2 lines · David Sterba

On Thu, Jul 27, 2023 at 03:12:47PM -0700, Boris Burkov wrote:
I've added the patchset to for-next,

There's a merge conflict due to Filipe's delayed refs changes, "btrfs: record simple quota deltas" new parameter to __btrfs_free_extent, run_delayed_data_ref and maybe others. I may resolve that for for-next but this could duplicate work if you that too so I can wait for a resend with other things updated.