-From df0e45e0fe79ddc8c332cba8ab8678b37e55c160 Mon Sep 17 00:00:00 2001
+From a3426d68a9f018271cb42757f657b53701671a0b Mon Sep 17 00:00:00 2001
From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
-Date: Mon, 9 Nov 2026 08:41:09 -0500
+Date: Thu, 12 Nov 2026 11:46:48 -0500
Subject: [PATCH 00/12] btrfs: allocator hints (Kai Krakow's series)
To: linux-btrfs@vger.kernel.org
Follow-ups:
-Patch 4 adds a writable attribute under the devid kobject. Deleting a
-missing device frees the btrfs_device with its devid kobject still
-registered, so reading any devinfo attribute is a use-after-free; Cheng
-Yongkang's "[PATCH v2] btrfs: remove devid kobject when deleting a
-missing device" (2026-10-26, Reviewed-by: Qu Wenruo), posted separately,
-fixes that.
-
-Patch 9 moves the hint values out of the UAPI header into
-fs/btrfs/volumes.h: a CONFIG_ symbol in a UAPI header makes
-"make headers_install" (and the deb/rpm package targets) fail.
+Patch 9 drops the CONFIG_BTRFS_ALLOCATOR_HINTS guard around the hint
+values in include/uapi/linux/btrfs_tree.h: a CONFIG_ symbol in a UAPI
+header makes "make headers_install" (and the deb/rpm package targets)
+fail. The values are on-disk format (the meaning of dev_item.type),
+userspace needs them to read and set the hints, and defining them
+changes no userspace-visible structure, so they are always exported.
Patch 10 keeps the hint across device replace: the target took over the
source's dev item but not the in-memory type, so the next update of the
-item wrote type 0 over the hint.
+item wrote type 0 over the hint. The source's hint can still be changed
+during a replace and the target inherits it, so a replace can move to a
+device of a different type (e.g. SSD to HDD). A type write to the
+target, or to the source after the finish, returns -EBUSY instead of
+aborting the transaction.
Patch 11 stores a changed type without running a transaction from the
sysfs store: the device is marked dirty (BTRFS_DEV_STATE_TYPE_DIRTY) and
the item is written by the next commit. Running a transaction from the
store deadlocked against device remove. The store orders the new type
before the dirty bit, so a commit cannot write the old type and clear
-the bit.
+the bit. It takes chunk_mutex, which also closes the window at the end
+of a replace where a hint written to the source could be lost.
Patch 12 makes the available space estimates honor the hints. The
allocator keeps data off METADATA_ONLY and NONE_ONLY devices and
read-only, while statfs still offered 3 GiB. The patch shares the hint
rules between the allocator and the estimates, which are kept separately
for data and metadata. With it the same fill ends in a clean ENOSPC.
+In statfs only f_bavail (df Avail) changes; f_blocks and f_bfree (df
+Size and Used) still count every device.
Reproducer: alloc-hints-space-test.sh. Kai, this patch doubles as the
bug report for your series.
btrfs: honor allocator hints for fixed-width mirrors
Zygo Blaxell (4):
- btrfs: allocator hints: move the hint values out of the UAPI header
+ btrfs: allocator hints: define the hint values unconditionally in the
+ UAPI header
btrfs: allocator hints: keep the hint across device replace
btrfs: allocator hints: don't run a transaction from the type store
btrfs: allocator hints: make the available space estimates honor the
hints
- fs/btrfs/Kconfig | 54 ++++++++
- fs/btrfs/dev-replace.c | 9 ++
- fs/btrfs/space-info.c | 3 +-
- fs/btrfs/super.c | 4 +
- fs/btrfs/sysfs.c | 104 +++++++++++++++
- fs/btrfs/transaction.c | 3 +
- fs/btrfs/volumes.c | 282 ++++++++++++++++++++++++++++++++++++++---
- fs/btrfs/volumes.h | 58 ++++++++-
- 8 files changed, 496 insertions(+), 21 deletions(-)
+ fs/btrfs/Kconfig | 54 ++++++
+ fs/btrfs/dev-replace.c | 9 +
+ fs/btrfs/space-info.c | 3 +-
+ fs/btrfs/super.c | 4 +
+ fs/btrfs/sysfs.c | 104 ++++++++++++
+ fs/btrfs/transaction.c | 3 +
+ fs/btrfs/volumes.c | 282 ++++++++++++++++++++++++++++++--
+ fs/btrfs/volumes.h | 37 ++++-
+ include/uapi/linux/btrfs_tree.h | 18 ++
+ 9 files changed, 493 insertions(+), 21 deletions(-)
--
--- /dev/null
+From 6c5c219911f08069e629aaa12b0fbb22c6b3d064 Mon Sep 17 00:00:00 2001
+From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
+Date: Sun, 4 Oct 2026 00:44:02 -0400
+Subject: [PATCH 09/12] btrfs: allocator hints: define the hint values
+ unconditionally in the UAPI header
+To: linux-btrfs@vger.kernel.org
+
+The BTRFS_DEV_ALLOCATION_* values sit in include/uapi/linux/btrfs_tree.h
+inside #ifdef CONFIG_BTRFS_ALLOCATOR_HINTS, to keep them away from
+userspace until the interface is settled. A CONFIG_ symbol in a UAPI
+header is rejected by scripts/headers_install.sh:
+
+ error: include/uapi/linux/btrfs_tree.h: leak CONFIG_BTRFS_ALLOCATOR_HINTS to user-space
+
+so 'make headers_install' (and the deb/rpm package targets, which run
+it) fail with the patch set applied.
+
+The values are on-disk format: they give the meaning of the low bits of
+dev_item.type, and userspace (btrfs-progs) needs them to read and set a
+device's hint. Defining them changes no userspace-visible structure.
+So drop the guard and always export the values. The code that acts on
+the hints stays under CONFIG_BTRFS_ALLOCATOR_HINTS.
+
+---
+ include/uapi/linux/btrfs_tree.h | 7 -------
+ 1 file changed, 7 deletions(-)
+
+diff --git a/include/uapi/linux/btrfs_tree.h b/include/uapi/linux/btrfs_tree.h
+index 035b7385622bb..f1917364d127e 100644
+--- a/include/uapi/linux/btrfs_tree.h
++++ b/include/uapi/linux/btrfs_tree.h
+@@ -585,11 +585,6 @@ struct btrfs_node {
+ struct btrfs_key_ptr ptrs[];
+ } __attribute__ ((__packed__));
+
+-/*
+- * This defines the chunk allocation hints for a device. It uses an ifdef guard
+- * in UAPI to avoid exposing it to userspace.
+- */
+-#ifdef CONFIG_BTRFS_ALLOCATOR_HINTS
+ /* dev_item.type */
+
+ /* btrfs chunk allocation hints */
+@@ -608,8 +603,6 @@ struct btrfs_node {
+ #define BTRFS_DEV_ALLOCATION_NONE_ONLY (5ULL)
+ /* 6..7 are unused values */
+
+-#endif /* CONFIG_BTRFS_ALLOCATOR_HINTS */
+-
+ struct btrfs_dev_item {
+ /* the internal btrfs device id */
+ __le64 devid;
+--
+2.47.3
+
+++ /dev/null
-From 7fbece5a12d5b4477bd520ac74756a3b3b0aa12a Mon Sep 17 00:00:00 2001
-From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
-Date: Sun, 4 Oct 2026 00:44:02 -0400
-Subject: [PATCH 09/12] btrfs: allocator hints: move the hint values out of the
- UAPI header
-To: linux-btrfs@vger.kernel.org
-
-The BTRFS_DEV_ALLOCATION_* values sit in include/uapi/linux/btrfs_tree.h
-inside #ifdef CONFIG_BTRFS_ALLOCATOR_HINTS, to keep them away from
-userspace until the interface is settled. A CONFIG_ symbol in a UAPI
-header is rejected by scripts/headers_install.sh:
-
- error: include/uapi/linux/btrfs_tree.h: leak CONFIG_BTRFS_ALLOCATOR_HINTS to user-space
-
-so 'make headers_install' (and the deb/rpm package targets, which run
-it) fail with the patch set applied.
-
-Only fs/btrfs uses the values, so move them to fs/btrfs/volumes.h, still
-under the Kconfig guard. That keeps them out of userspace, which was the
-intent, without the leak.
-
----
- fs/btrfs/volumes.h | 21 +++++++++++++++++++++
- include/uapi/linux/btrfs_tree.h | 25 -------------------------
- 2 files changed, 21 insertions(+), 25 deletions(-)
-
-diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
-index da7c03871b838..91efbea27d04c 100644
---- a/fs/btrfs/volumes.h
-+++ b/fs/btrfs/volumes.h
-@@ -598,6 +598,27 @@ struct btrfs_io_context {
- struct btrfs_io_stripe stripes[];
- };
-
-+/*
-+ * Chunk allocation hints, stored in dev_item.type. Kept out of the UAPI
-+ * headers until the interface is settled.
-+ */
-+#ifdef CONFIG_BTRFS_ALLOCATOR_HINTS
-+#define BTRFS_DEV_ALLOCATION_MASK_BIT_COUNT 3
-+/* preferred data chunk, but metadata chunk allowed */
-+#define BTRFS_DEV_ALLOCATION_PREFERRED_DATA (0ULL)
-+/* preferred metadata chunk, but data chunk allowed */
-+#define BTRFS_DEV_ALLOCATION_PREFERRED_METADATA (1ULL)
-+/* only metadata chunk allowed */
-+#define BTRFS_DEV_ALLOCATION_METADATA_ONLY (2ULL)
-+/* only data chunk allowed */
-+#define BTRFS_DEV_ALLOCATION_DATA_ONLY (3ULL)
-+/* avoid chunk allocation if possible */
-+#define BTRFS_DEV_ALLOCATION_PREFERRED_NONE (4ULL)
-+/* deny chunk allocation */
-+#define BTRFS_DEV_ALLOCATION_NONE_ONLY (5ULL)
-+/* 6..7 are unused values */
-+#endif /* CONFIG_BTRFS_ALLOCATOR_HINTS */
-+
- struct btrfs_device_info {
- struct btrfs_device *dev;
- u64 dev_offset;
-diff --git a/include/uapi/linux/btrfs_tree.h b/include/uapi/linux/btrfs_tree.h
-index 035b7385622bb..47ee52859b458 100644
---- a/include/uapi/linux/btrfs_tree.h
-+++ b/include/uapi/linux/btrfs_tree.h
-@@ -585,31 +585,6 @@ struct btrfs_node {
- struct btrfs_key_ptr ptrs[];
- } __attribute__ ((__packed__));
-
--/*
-- * This defines the chunk allocation hints for a device. It uses an ifdef guard
-- * in UAPI to avoid exposing it to userspace.
-- */
--#ifdef CONFIG_BTRFS_ALLOCATOR_HINTS
--/* dev_item.type */
--
--/* btrfs chunk allocation hints */
--#define BTRFS_DEV_ALLOCATION_MASK_BIT_COUNT 3
--/* preferred data chunk, but metadata chunk allowed */
--#define BTRFS_DEV_ALLOCATION_PREFERRED_DATA (0ULL)
--/* preferred metadata chunk, but data chunk allowed */
--#define BTRFS_DEV_ALLOCATION_PREFERRED_METADATA (1ULL)
--/* only metadata chunk allowed */
--#define BTRFS_DEV_ALLOCATION_METADATA_ONLY (2ULL)
--/* only data chunk allowed */
--#define BTRFS_DEV_ALLOCATION_DATA_ONLY (3ULL)
--/* avoid chunk allocation if possible */
--#define BTRFS_DEV_ALLOCATION_PREFERRED_NONE (4ULL)
--/* deny chunk allocation */
--#define BTRFS_DEV_ALLOCATION_NONE_ONLY (5ULL)
--/* 6..7 are unused values */
--
--#endif /* CONFIG_BTRFS_ALLOCATOR_HINTS */
--
- struct btrfs_dev_item {
- /* the internal btrfs device id */
- __le64 devid;
---
-2.47.3
-
-From 1e8e85d87b5bea59f545e977e229199963980318 Mon Sep 17 00:00:00 2001
+From 389885a1676b3f3bccc445c43edc0911934f862c Mon Sep 17 00:00:00 2001
From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Date: Sun, 4 Oct 2026 00:44:58 -0400
Subject: [PATCH 10/12] btrfs: allocator hints: keep the hint across device
replace. Device remove has the same window between deleting the dev
item and removing the sysfs directory.
-Return -EBUSY for the replace target, for a device with the replace
-devid, and for the source of a running or suspended replace. An -ENOENT
-that still gets past these checks (remove, or a replace that finishes
-after them) ends the transaction without aborting it and returns -ENODEV.
+Return -EBUSY for the replace target and for a device with the replace
+devid. An -ENOENT that still gets past these checks (remove, or a
+replace that finishes after them) ends the transaction without aborting
+it and returns -ENODEV.
+
+The source's hint can still be changed during a replace, and the target
+inherits it at the finish, so a replace can move to a device of a
+different type, e.g. SSD to HDD. A write that races the finish can be
+lost: it misses the dev item and returns -ENODEV, or the finish has
+already copied the old type. That is tolerable, and the next patch
+closes the window.
---
- fs/btrfs/Kconfig | 9 +++++----
+ fs/btrfs/Kconfig | 8 ++++----
fs/btrfs/dev-replace.c | 6 ++++++
- fs/btrfs/sysfs.c | 28 ++++++++++++++++++++++++++++
- 3 files changed, 39 insertions(+), 4 deletions(-)
+ fs/btrfs/sysfs.c | 22 ++++++++++++++++++++++
+ 3 files changed, 32 insertions(+), 4 deletions(-)
diff --git a/fs/btrfs/Kconfig b/fs/btrfs/Kconfig
-index e14cddb83c8a3..038626bd24a68 100644
+index e14cddb83c8a3..72c1ced888a95 100644
--- a/fs/btrfs/Kconfig
+++ b/fs/btrfs/Kconfig
-@@ -102,10 +102,11 @@ config BTRFS_ALLOCATOR_HINTS
+@@ -102,10 +102,10 @@ config BTRFS_ALLOCATOR_HINTS
They cannot separate the two chunk classes and are outside the
tiering use case targeted by this feature.
- device. Reapply the hint after replacing a device. Hint updates are
- also not serialized with chunk allocation, so a concurrent new chunk
- may use either the old or the new setting.
-+ Device replace copies the allocation hint to the target device.
-+ The hint of the source and target devices cannot be changed while
-+ a replace is running or suspended. Hint updates are not serialized
-+ with chunk allocation, so a concurrent new chunk may use either the
-+ old or the new setting.
++ Device replace copies the allocation hint to the target device;
++ the target's hint cannot be changed until the replace finishes.
++ Hint updates are not serialized with chunk allocation, so a
++ concurrent new chunk may use either the old or the new setting.
You are advised to watch your free space closely with btrfs tools
instead of relying on df only.
memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
-index 16912c8f4f9e6..6457b74d67f93 100644
+index 16912c8f4f9e6..29cc1fba9e9c8 100644
--- a/fs/btrfs/sysfs.c
+++ b/fs/btrfs/sysfs.c
-@@ -19,6 +19,7 @@
- #include "transaction.h"
- #include "sysfs.h"
- #include "volumes.h"
-+#include "dev-replace.h"
- #include "space-info.h"
- #include "block-group.h"
- #include "qgroup.h"
-@@ -2021,6 +2022,7 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
- struct btrfs_device *device;
- int ret;
- struct btrfs_trans_handle *trans;
-+ bool replacing;
-
- u64 type, prev_type;
-
-@@ -2049,6 +2051,21 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
+@@ -2049,6 +2049,17 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
if (!btrfs_dev_allocation_hint_valid(type))
return -EINVAL;
+ /*
-+ * A replace target has no dev item of its own. When the replace
-+ * finishes, it takes over the source's dev item and type. Neither
-+ * device's type may change until then.
++ * A replace target has no dev item of its own until the replace
++ * finishes and it takes over the source's dev item and type. After
++ * the finish, the source has the replace devid and no dev item. The
++ * source's type may change during the replace: the target inherits
++ * it at the finish.
+ */
+ if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state) ||
+ device->devid == BTRFS_DEV_REPLACE_DEVID)
+ return -EBUSY;
-+ down_read(&fs_info->dev_replace.rwsem);
-+ replacing = btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
-+ fs_info->dev_replace.srcdev == device;
-+ up_read(&fs_info->dev_replace.rwsem);
-+ if (replacing)
-+ return -EBUSY;
+
trans = btrfs_start_transaction(root, 1);
if (IS_ERR(trans))
return PTR_ERR(trans);
-@@ -2058,6 +2075,17 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
+@@ -2058,6 +2069,17 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
ret = btrfs_update_device(trans, device);
-From 719bd5945ed892368c908d1fd3e5e5f04d8083c2 Mon Sep 17 00:00:00 2001
+From d49c657c24936286f68a2ab0d35d9b9d01498b3d Mon Sep 17 00:00:00 2001
From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Date: Sun, 4 Oct 2026 10:55:00 -0400
Subject: [PATCH 11/12] btrfs: allocator hints: don't run a transaction from
hint takes effect for new chunks at once and reaches the disk at the next
commit, sync or unmount.
-With the replace swap serialized by chunk_mutex, the source of a running
-replace no longer needs to refuse hint changes: the finish copies the
-hint and the dirty bit to the target. Only the replace target (devid 0,
-no dev item) still returns -EBUSY. Also check sb_rdonly() before
-BTRFS_FS_OPEN, so a read-only mount reports -EROFS rather than -EBUSY,
-and refuse seed devices, which are read-only and not written by the
-commit-time update.
+Taking chunk_mutex in the store also closes the race at the end of a
+device replace: the finish moves the hint and the dirty bit to the
+target under the same mutex, so a hint change on the source either
+reaches the target or, once the source has the replace devid, returns
+-EBUSY. The replace target (devid 0, no dev item) still returns -EBUSY.
+Also check sb_rdonly() before BTRFS_FS_OPEN, so a read-only mount
+reports -EROFS rather than -EBUSY, and refuse seed devices, which are
+read-only and not written by the commit-time update.
The store orders the new type before the dirty bit
(smp_mb__before_atomic(); the commit's test_and_clear_bit() is the other
machine.
---
- fs/btrfs/Kconfig | 9 ++---
+ fs/btrfs/Kconfig | 4 +-
fs/btrfs/dev-replace.c | 3 ++
- fs/btrfs/sysfs.c | 89 ++++++++++++++++--------------------------
+ fs/btrfs/sysfs.c | 83 +++++++++++++++++-------------------------
fs/btrfs/transaction.c | 3 ++
- fs/btrfs/volumes.c | 57 ++++++++++++++++++++++++++-
+ fs/btrfs/volumes.c | 57 ++++++++++++++++++++++++++++-
fs/btrfs/volumes.h | 7 ++++
- 6 files changed, 107 insertions(+), 61 deletions(-)
+ 6 files changed, 105 insertions(+), 52 deletions(-)
diff --git a/fs/btrfs/Kconfig b/fs/btrfs/Kconfig
-index 038626bd24a68..f097f9631a857 100644
+index 72c1ced888a95..f097f9631a857 100644
--- a/fs/btrfs/Kconfig
+++ b/fs/btrfs/Kconfig
-@@ -102,11 +102,10 @@ config BTRFS_ALLOCATOR_HINTS
+@@ -102,10 +102,10 @@ config BTRFS_ALLOCATOR_HINTS
They cannot separate the two chunk classes and are outside the
tiering use case targeted by this feature.
-- Device replace copies the allocation hint to the target device.
-- The hint of the source and target devices cannot be changed while
-- a replace is running or suspended. Hint updates are not serialized
-- with chunk allocation, so a concurrent new chunk may use either the
-- old or the new setting.
+ A hint change applies to new chunks at once and is written to the
+ device item at the next transaction commit (sync to force it).
-+ Device replace copies the allocation hint to the target device;
-+ the target's hint cannot be changed until the replace finishes.
+ Device replace copies the allocation hint to the target device;
+ the target's hint cannot be changed until the replace finishes.
+- Hint updates are not serialized with chunk allocation, so a
+- concurrent new chunk may use either the old or the new setting.
You are advised to watch your free space closely with btrfs tools
instead of relying on df only.
memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
-index 6457b74d67f93..6e1d0636f9af9 100644
+index 29cc1fba9e9c8..6e1d0636f9af9 100644
--- a/fs/btrfs/sysfs.c
+++ b/fs/btrfs/sysfs.c
-@@ -19,7 +19,6 @@
- #include "transaction.h"
- #include "sysfs.h"
- #include "volumes.h"
--#include "dev-replace.h"
- #include "space-info.h"
- #include "block-group.h"
- #include "qgroup.h"
-@@ -2018,31 +2017,22 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
+@@ -2017,30 +2017,22 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
const char *buf, size_t len)
{
struct btrfs_fs_info *fs_info;
struct btrfs_device *device;
int ret;
- struct btrfs_trans_handle *trans;
-- bool replacing;
-
- u64 type, prev_type;
+ u64 type;
ret = kstrtou64(buf, 0, &type);
if (ret < 0)
return -EINVAL;
-@@ -2051,55 +2041,44 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
+@@ -2049,51 +2041,44 @@ static ssize_t btrfs_devinfo_type_store(struct kobject *kobj,
if (!btrfs_dev_allocation_hint_valid(type))
return -EINVAL;
+ return -EROFS;
+
/*
-- * A replace target has no dev item of its own. When the replace
-- * finishes, it takes over the source's dev item and type. Neither
-- * device's type may change until then.
+- * A replace target has no dev item of its own until the replace
+- * finishes and it takes over the source's dev item and type. After
+- * the finish, the source has the replace devid and no dev item. The
+- * source's type may change during the replace: the target inherits
+- * it at the finish.
+ * No transaction here. Device remove drains this attribute while it
+ * holds a transaction handle and device_list_mutex, so a commit (which
+ * waits for that handle) or a join (which can wait for
+ mutex_lock(&fs_info->chunk_mutex);
if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state) ||
- device->devid == BTRFS_DEV_REPLACE_DEVID)
-- return -EBUSY;
-- down_read(&fs_info->dev_replace.rwsem);
-- replacing = btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
-- fs_info->dev_replace.srcdev == device;
-- up_read(&fs_info->dev_replace.rwsem);
-- if (replacing)
+ device->devid == BTRFS_DEV_REPLACE_DEVID) {
+ mutex_unlock(&fs_info->chunk_mutex);
return -EBUSY;
- ret = -ENODEV;
- goto abort;
- }
--
++ smp_mb__before_atomic();
++ set_bit(BTRFS_DEV_STATE_TYPE_DIRTY, &device->dev_state);
++ mutex_unlock(&fs_info->chunk_mutex);
+
- if (ret < 0) {
- btrfs_abort_transaction(trans, ret);
- btrfs_end_transaction(trans);
- goto abort;
- }
-+ smp_mb__before_atomic();
-+ set_bit(BTRFS_DEV_STATE_TYPE_DIRTY, &device->dev_state);
-+ mutex_unlock(&fs_info->chunk_mutex);
-
+-
- ret = btrfs_commit_transaction(trans);
- if (ret < 0)
- goto abort;
{
btrfs_dev_stat_inc(dev, index);
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
-index 91efbea27d04c..df5c6db129869 100644
+index da7c03871b838..181b8a4ea14b3 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -110,6 +110,8 @@ static_assert(BTRFS_RAID_RAID1C4 == 8);
/* Special value encoding failure to write primary super block. */
#define BTRFS_SUPER_PRIMARY_WRITE_ERROR (INT_MAX / 2)
-@@ -819,6 +821,11 @@ int btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
+@@ -798,6 +800,11 @@ int btrfs_init_devices_late(struct btrfs_fs_info *fs_info);
int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info);
int btrfs_init_writeback_bio_size(struct btrfs_fs_info *fs_info);
int btrfs_run_dev_stats(struct btrfs_trans_handle *trans);
-From df0e45e0fe79ddc8c332cba8ab8678b37e55c160 Mon Sep 17 00:00:00 2001
+From a3426d68a9f018271cb42757f657b53701671a0b Mon Sep 17 00:00:00 2001
From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Date: Sun, 18 Oct 2026 15:50:43 -0400
Subject: [PATCH 12/12] btrfs: allocator hints: make the available space
exclude for the type and picks hint tiers the way the allocator does.
btrfs_get_per_profile_avail() takes the block group type. Mixed block
groups ignore the hints, as in the allocator.
-- statfs leaves out devices that data chunks may not use.
+- statfs leaves out devices that data chunks may not use. Only the
+ available space changes: f_bavail (df Avail) leaves out the
+ unallocated space of those devices, while f_blocks (df Size) and
+ f_bfree (and so df Used) still count every device, including the ones
+ the hints keep data off. On a filesystem with a METADATA_ONLY device,
+ for example, Size minus Used is larger than Avail by about that
+ device's unallocated space.
- A type change recomputes the estimate.
PREFERRED_NONE devices still count: the allocator uses them as the last
BUG_ON(ndevs > ctl->ndevs);
ctl->ndevs = ndevs;
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
-index df5c6db129869..31538a7bf7fa6 100644
+index 181b8a4ea14b3..f8b15fc91f9ca 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -337,6 +337,13 @@ enum btrfs_read_policy {
spinlock_t per_profile_lock;
};
-@@ -947,18 +956,26 @@ const char *btrfs_bg_type_to_raid_name(u64 flags);
+@@ -926,18 +935,26 @@ const char *btrfs_bg_type_to_raid_name(u64 flags);
int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info);
bool btrfs_verify_dev_items(const struct btrfs_fs_info *fs_info);
void btrfs_update_per_profile_avail(struct btrfs_fs_info *fs_info);
nodiscard, replace -K. Limits: raid1 only; the remove race is timing
dependent (one run = tens of thousands of writes, not a proof).
-**Patch 9 (UAPI header).** Build-level only: `git grep CONFIG_
-misc-next/testing -- include/uapi/linux/btrfs.h include/uapi/linux/btrfs_tree.h`
-finds nothing, and the hint values live in `fs/btrfs/volumes.h` (checked
-2026-10-18). `make headers_install` itself was not run this round; the
-original failure (a CONFIG_ symbol in a UAPI header fails the CONFIG-leak
-check) was proven on 6.18 on 2026-10-04.
+**Patch 9 (UAPI header).** Superseded 2026-11-12: the values stay in
+the UAPI header without the guard, and `make headers_install` was run
+(see "Patch 9 rewritten" below). Before that, 2026-10-18: no CONFIG_ in
+the UAPI headers, values in `fs/btrfs/volumes.h`; the original failure
+was proven on 6.18 on 2026-10-04.
## The fix (patch 12; promoted 2026-10-18)
`forensics/ah1106/` (README, full-319.log, full-ctl.log, full-mn69.log).
This closes "e2 (patch 8) negative control: not built" below.
+## Patch 9 rewritten, patches 10-12 reworded (2026-11-12, uh1112)
+
+Zygo 11-12: patch 9 now drops the `CONFIG_BTRFS_ALLOCATOR_HINTS` guard
+and keeps the values in `include/uapi/linux/btrfs_tree.h` (on-disk
+dev_item.type, needed by btrfs-progs, no structure change) instead of
+moving them to `fs/btrfs/volumes.h`. Same day: patch 10 no longer
+refuses the replace source (option A: an SSD->HDD emergency replace or an
+HDD->SSD upgrade can re-hint the source; a write racing the finish may be
+lost on patch 10 alone, patch 11 closes it), patch 11's message says so,
+patch 12's message says only f_bavail changes in statfs. The series' end
+result is unchanged except for where the values are defined: the topic
+tip tree equals the UAPI-only candidate's, and the earlier test results
+above apply to it.
+
+Oracles (all three lanes; harness forensics `uh1112`):
+- topic old tip -> new tip = exactly the moved defines: the 7 values back
+ in btrfs_tree.h without the guard (no `CONFIG_` left in the file), the
+ block gone from volumes.h; testing and zb64 moved by the same delta,
+ nothing else (content lines compared).
+- range-diff: patches 1-8 (and Cheng's, and on 6.18 Qu's three
+ backports) `=`; 9 replaced; 10 changed (code + message); 11 changed
+ (message; its tree equals the UAPI-only candidate's); 12 message only.
+- fs/btrfs objects, lane config (ALLOCATOR_HINTS=y, BTRFS_ASSERT=y, no
+ KASAN), old zb64 vs new zb64: **not byte-identical**. Every
+ difference is a `__LINE__` constant 21 lower (ASSERT/WARN line
+ immediates in .text, line fields in `__bug_table`/.rodata): the 21
+ lines removed from volumes.h shift the inline helpers below them. The
+ probe "new zb64 + 21 blank lines in place of the block" is
+ byte-identical to the old zb64 in every object. So the code is the
+ same; no runtime gate.
+
+ | lane | objects | differ | line immediates -21 | data words -21 | other | padded probe |
+ |---|---|---|---|---|---|---|
+ | 6.18 | 64 | 15 | 26 | 53 | 0 | identical |
+ | misc-next | 75 | 19 | 35 | 86 | 0 | identical |
+ | 7.3 | 64 | 16 | 31 | 58 | 0 | identical |
+
+Build checks (bhive, vm-scripts `uh1112/run-uh1112.sh`):
+- per-commit `topic-buildcheck.sh` with `ENABLE_SYMS=BTRFS_ALLOCATOR_HINTS`
+ (=y from the Kconfig commit on) and again with the feature off: 6.18
+ 17/17, misc-next 14/14, 7.3 14/14 (lane base included), 0 failures, 0
+ new warnings, tip vmlinux links rc 0, both modes.
+- `make headers_install` at the topic tip, HINTS=y and =n: rc 0 on all
+ lanes, the installed `linux/btrfs_tree.h` has the 7 defines and no
+ `CONFIG_`. Control, the series before patch 9 (KeinNiemand's commit):
+ rc 2, `error: include/uapi/linux/btrfs_tree.h: leak
+ CONFIG_BTRFS_ALLOCATOR_HINTS to user-space`.
+- Not run: runtime tests (the code is unchanged, see the object
+ comparison). Patch 10 alone (source writable, transaction store) is
+ only build-checked; its finish race is closed by patch 11 and was not
+ exercised.
+
## Not covered / open
- **KERNEL BUG (found 2026-10-18): space estimates ignore the hints.** See
`alloc-hints-space-test.sh` reproduces what is very likely the same
defect.
- e2 (patch 8) negative control: done 2026-11-06 (ah1106, above).
-- `make headers_install` on the series (patch 9): not run this round.
+- `make headers_install` on the series (patch 9): done 2026-11-12, above.
- Striped profiles (raid0/10/5/6) and stripe_alloc data under hints: not
tested.