Merge tag 'xfs-zoned-allocator-2025-03-03' of git://git.infradead.org/users/hch/xfs into xfs-6.15-zoned_devices

xfs: add support for zoned devices

Add support for the new zoned space allocator and thus for zoned devices:

    https://zonedstorage.io/docs/introduction/zoned-storage

to XFS. This has been developed for and tested on both SMR hard drives,
which are the oldest and most common class of zoned devices:

   https://zonedstorage.io/docs/introduction/smr

and ZNS SSDs:

   https://zonedstorage.io/docs/introduction/zns

It has not been tested with zoned UFS devices, as their current capacity
points and performance characteristics aren't too interesting for XFS
use cases (but never say never).

Sequential write only zones are only supported for data using a new
allocator for the RT device, which maps each zone to a rtgroup which
is written sequentially.  All metadata and (for now) the log require
using randomly writable space. This means a realtime device is required
to support zoned storage, but for the common case of SMR hard drives
that contain random writable zones and sequential write required zones
on the same block device, the concept of an internal RT device is added
which means using XFS on a SMR HDD is as simple as:

$ mkfs.xfs /dev/sda
$ mount /dev/sda /mnt

When using NVMe ZNS SSDs that do not support conventional zones, the
traditional multi-device RT configuration is required.  E.g. for an
SSD with a conventional namespace 1 and a zoned namespace 2:

$ mkfs.xfs /dev/nvme0n1 -o rtdev=/dev/nvme0n2
$ mount -o rtdev=/dev/nvme0n2 /dev/nvme0n1 /mnt

The zoned allocator can also be used on conventional block devices, or
on conventional zones (e.g. when using an SMR HDD as the external RT
device).  For example using zoned XFS on normal SSDs shows very nice
performance advantages and write amplification reduction for intelligent
workloads like RocksDB.

Some work is still in progress or planned, but should not affect the
integration with the rest of XFS or the on-disk format:

 - support for quotas
 - support for reflinks

Note that the I/O path already supports reflink, but garbage collection
isn't refcount aware yet and would unshare shared blocks, thus rendering
the feature useless.

1 files changed, 120 insertions, 86 deletions

diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index b69356582b86..e65a659901d5 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -40,6 +40,7 @@
 #include "xfs_rtrmap_btree.h"
 #include "xfs_rtrefcount_btree.h"
 #include "scrub/stats.h"
+#include "xfs_zone_alloc.h"
 
 static DEFINE_MUTEX(xfs_uuid_table_mutex);
 static int xfs_uuid_table_size;
@@ -461,22 +462,38 @@ xfs_mount_reset_sbqflags(
 	return xfs_sync_sb(mp, false);
 }
 
+static const char *const xfs_free_pool_name[] = {
+	[XC_FREE_BLOCKS]	= "free blocks",
+	[XC_FREE_RTEXTENTS]	= "free rt extents",
+	[XC_FREE_RTAVAILABLE]	= "available rt extents",
+};
+
 uint64_t
-xfs_default_resblks(xfs_mount_t *mp)
+xfs_default_resblks(
+	struct xfs_mount	*mp,
+	enum xfs_free_counter	ctr)
 {
-	uint64_t resblks;
-
-	/*
-	 * We default to 5% or 8192 fsbs of space reserved, whichever is
-	 * smaller.  This is intended to cover concurrent allocation
-	 * transactions when we initially hit enospc. These each require a 4
-	 * block reservation. Hence by default we cover roughly 2000 concurrent
-	 * allocation reservations.
-	 */
-	resblks = mp->m_sb.sb_dblocks;
-	do_div(resblks, 20);
-	resblks = min_t(uint64_t, resblks, 8192);
-	return resblks;
+	switch (ctr) {
+	case XC_FREE_BLOCKS:
+		/*
+		 * Default to 5% or 8192 FSBs of space reserved, whichever is
+		 * smaller.
+		 *
+		 * This is intended to cover concurrent allocation transactions
+		 * when we initially hit ENOSPC.  These each require a 4 block
+		 * reservation. Hence by default we cover roughly 2000
+		 * concurrent allocation reservations.
+		 */
+		return min(div_u64(mp->m_sb.sb_dblocks, 20), 8192ULL);
+	case XC_FREE_RTEXTENTS:
+	case XC_FREE_RTAVAILABLE:
+		if (IS_ENABLED(CONFIG_XFS_RT) && xfs_has_zoned(mp))
+			return xfs_zoned_default_resblks(mp, ctr);
+		return 0;
+	default:
+		ASSERT(0);
+		return 0;
+	}
 }
 
 /* Ensure the summary counts are correct. */
@@ -543,7 +560,7 @@ xfs_check_summary_counts(
 	 * If we're mounting the rt volume after recovering the log, recompute
 	 * frextents from the rtbitmap file to fix the inconsistency.
 	 */
-	if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
+	if (xfs_has_realtime(mp) && !xfs_has_zoned(mp) && !xfs_is_clean(mp)) {
 		error = xfs_rtalloc_reinit_frextents(mp);
 		if (error)
 			return error;
@@ -678,6 +695,7 @@ xfs_mountfs(
 	uint			quotamount = 0;
 	uint			quotaflags = 0;
 	int			error = 0;
+	int			i;
 
 	xfs_sb_mount_common(mp, sbp);
 
@@ -747,27 +765,15 @@ xfs_mountfs(
 	/* enable fail_at_unmount as default */
 	mp->m_fail_unmount = true;
 
-	super_set_sysfs_name_id(mp->m_super);
-
-	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
-			       NULL, mp->m_super->s_id);
-	if (error)
-		goto out;
-
-	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
-			       &mp->m_kobj, "stats");
+	error = xfs_mount_sysfs_init(mp);
 	if (error)
-		goto out_remove_sysfs;
+		goto out_remove_scrub_stats;
 
 	xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
 
-	error = xfs_error_sysfs_init(mp);
-	if (error)
-		goto out_remove_scrub_stats;
-
 	error = xfs_errortag_init(mp);
 	if (error)
-		goto out_remove_error_sysfs;
+		goto out_remove_sysfs;
 
 	error = xfs_uuid_mount(mp);
 	if (error)
@@ -1031,6 +1037,12 @@ xfs_mountfs(
 	if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
 		xfs_log_clean(mp);
 
+	if (xfs_has_zoned(mp)) {
+		error = xfs_mount_zones(mp);
+		if (error)
+			goto out_rtunmount;
+	}
+
 	/*
 	 * Complete the quota initialisation, post-log-replay component.
 	 */
@@ -1046,22 +1058,28 @@ xfs_mountfs(
 	 * privileged transactions. This is needed so that transaction
 	 * space required for critical operations can dip into this pool
 	 * when at ENOSPC. This is needed for operations like create with
-	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
-	 * are not allowed to use this reserved space.
+	 * attr, unwritten extent conversion at ENOSPC, garbage collection
+	 * etc. Data allocations are not allowed to use this reserved space.
 	 *
 	 * This may drive us straight to ENOSPC on mount, but that implies
 	 * we were already there on the last unmount. Warn if this occurs.
 	 */
 	if (!xfs_is_readonly(mp)) {
-		error = xfs_reserve_blocks(mp, xfs_default_resblks(mp));
-		if (error)
-			xfs_warn(mp,
-	"Unable to allocate reserve blocks. Continuing without reserve pool.");
+		for (i = 0; i < XC_FREE_NR; i++) {
+			error = xfs_reserve_blocks(mp, i,
+					xfs_default_resblks(mp, i));
+			if (error)
+				xfs_warn(mp,
+"Unable to allocate reserve blocks. Continuing without reserve pool for %s.",
+					xfs_free_pool_name[i]);
+		}
 
 		/* Reserve AG blocks for future btree expansion. */
 		error = xfs_fs_reserve_ag_blocks(mp);
 		if (error && error != -ENOSPC)
 			goto out_agresv;
+
+		xfs_zone_gc_start(mp);
 	}
 
 	return 0;
@@ -1069,6 +1087,8 @@ xfs_mountfs(
  out_agresv:
 	xfs_fs_unreserve_ag_blocks(mp);
 	xfs_qm_unmount_quotas(mp);
+	if (xfs_has_zoned(mp))
+		xfs_unmount_zones(mp);
  out_rtunmount:
 	xfs_rtunmount_inodes(mp);
  out_rele_rip:
@@ -1116,13 +1136,10 @@ xfs_mountfs(
 	xfs_uuid_unmount(mp);
  out_remove_errortag:
 	xfs_errortag_del(mp);
- out_remove_error_sysfs:
-	xfs_error_sysfs_del(mp);
+ out_remove_sysfs:
+	xfs_mount_sysfs_del(mp);
  out_remove_scrub_stats:
 	xchk_stats_unregister(mp->m_scrub_stats);
-	xfs_sysfs_del(&mp->m_stats.xs_kobj);
- out_remove_sysfs:
-	xfs_sysfs_del(&mp->m_kobj);
  out:
 	return error;
 }
@@ -1148,8 +1165,12 @@ xfs_unmountfs(
 	xfs_inodegc_flush(mp);
 
 	xfs_blockgc_stop(mp);
+	if (!test_bit(XFS_OPSTATE_READONLY, &mp->m_opstate))
+		xfs_zone_gc_stop(mp);
 	xfs_fs_unreserve_ag_blocks(mp);
 	xfs_qm_unmount_quotas(mp);
+	if (xfs_has_zoned(mp))
+		xfs_unmount_zones(mp);
 	xfs_rtunmount_inodes(mp);
 	xfs_irele(mp->m_rootip);
 	if (mp->m_metadirip)
@@ -1173,7 +1194,7 @@ xfs_unmountfs(
 	 * we only every apply deltas to the superblock and hence the incore
 	 * value does not matter....
 	 */
-	error = xfs_reserve_blocks(mp, 0);
+	error = xfs_reserve_blocks(mp, XC_FREE_BLOCKS, 0);
 	if (error)
 		xfs_warn(mp, "Unable to free reserved block pool. "
 				"Freespace may not be correct on next mount.");
@@ -1195,10 +1216,8 @@ xfs_unmountfs(
 	xfs_free_rtgroups(mp, 0, mp->m_sb.sb_rgcount);
 	xfs_free_perag_range(mp, 0, mp->m_sb.sb_agcount);
 	xfs_errortag_del(mp);
-	xfs_error_sysfs_del(mp);
 	xchk_stats_unregister(mp->m_scrub_stats);
-	xfs_sysfs_del(&mp->m_stats.xs_kobj);
-	xfs_sysfs_del(&mp->m_kobj);
+	xfs_mount_sysfs_del(mp);
 }
 
 /*
@@ -1220,52 +1239,67 @@ xfs_fs_writable(
 	return true;
 }
 
+/*
+ * Estimate the amount of free space that is not available to userspace and is
+ * not explicitly reserved from the incore fdblocks.  This includes:
+ *
+ * - The minimum number of blocks needed to support splitting a bmap btree
+ * - The blocks currently in use by the freespace btrees because they record
+ *   the actual blocks that will fill per-AG metadata space reservations
+ */
+uint64_t
+xfs_freecounter_unavailable(
+	struct xfs_mount	*mp,
+	enum xfs_free_counter	ctr)
+{
+	if (ctr != XC_FREE_BLOCKS)
+		return 0;
+	return mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks);
+}
+
 void
 xfs_add_freecounter(
 	struct xfs_mount	*mp,
-	struct percpu_counter	*counter,
+	enum xfs_free_counter	ctr,
 	uint64_t		delta)
 {
-	bool			has_resv_pool = (counter == &mp->m_fdblocks);
+	struct xfs_freecounter	*counter = &mp->m_free[ctr];
 	uint64_t		res_used;
 
 	/*
 	 * If the reserve pool is depleted, put blocks back into it first.
 	 * Most of the time the pool is full.
 	 */
-	if (!has_resv_pool || mp->m_resblks == mp->m_resblks_avail) {
-		percpu_counter_add(counter, delta);
+	if (likely(counter->res_avail == counter->res_total)) {
+		percpu_counter_add(&counter->count, delta);
 		return;
 	}
 
 	spin_lock(&mp->m_sb_lock);
-	res_used = mp->m_resblks - mp->m_resblks_avail;
+	res_used = counter->res_total - counter->res_avail;
 	if (res_used > delta) {
-		mp->m_resblks_avail += delta;
+		counter->res_avail += delta;
 	} else {
 		delta -= res_used;
-		mp->m_resblks_avail = mp->m_resblks;
-		percpu_counter_add(counter, delta);
+		counter->res_avail = counter->res_total;
+		percpu_counter_add(&counter->count, delta);
 	}
 	spin_unlock(&mp->m_sb_lock);
 }
 
+
+/* Adjust in-core free blocks or RT extents. */
 int
 xfs_dec_freecounter(
 	struct xfs_mount	*mp,
-	struct percpu_counter	*counter,
+	enum xfs_free_counter	ctr,
 	uint64_t		delta,
 	bool			rsvd)
 {
-	int64_t			lcounter;
-	uint64_t		set_aside = 0;
+	struct xfs_freecounter	*counter = &mp->m_free[ctr];
 	s32			batch;
-	bool			has_resv_pool;
 
-	ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
-	has_resv_pool = (counter == &mp->m_fdblocks);
-	if (rsvd)
-		ASSERT(has_resv_pool);
+	ASSERT(ctr < XC_FREE_NR);
 
 	/*
 	 * Taking blocks away, need to be more accurate the closer we
@@ -1275,7 +1309,7 @@ xfs_dec_freecounter(
 	 * then make everything serialise as we are real close to
 	 * ENOSPC.
 	 */
-	if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
+	if (__percpu_counter_compare(&counter->count, 2 * XFS_FDBLOCKS_BATCH,
 				     XFS_FDBLOCKS_BATCH) < 0)
 		batch = 1;
 	else
@@ -1292,34 +1326,34 @@ xfs_dec_freecounter(
 	 * problems (i.e. transaction abort, pagecache discards, etc.) than
 	 * slightly premature -ENOSPC.
 	 */
-	if (has_resv_pool)
-		set_aside = xfs_fdblocks_unavailable(mp);
-	percpu_counter_add_batch(counter, -((int64_t)delta), batch);
-	if (__percpu_counter_compare(counter, set_aside,
-				     XFS_FDBLOCKS_BATCH) >= 0) {
-		/* we had space! */
-		return 0;
-	}
-
-	/*
-	 * lock up the sb for dipping into reserves before releasing the space
-	 * that took us to ENOSPC.
-	 */
-	spin_lock(&mp->m_sb_lock);
-	percpu_counter_add(counter, delta);
-	if (!has_resv_pool || !rsvd)
-		goto fdblocks_enospc;
-
-	lcounter = (long long)mp->m_resblks_avail - delta;
-	if (lcounter >= 0) {
-		mp->m_resblks_avail = lcounter;
+	percpu_counter_add_batch(&counter->count, -((int64_t)delta), batch);
+	if (__percpu_counter_compare(&counter->count,
+			xfs_freecounter_unavailable(mp, ctr),
+			XFS_FDBLOCKS_BATCH) < 0) {
+		/*
+		 * Lock up the sb for dipping into reserves before releasing the
+		 * space that took us to ENOSPC.
+		 */
+		spin_lock(&mp->m_sb_lock);
+		percpu_counter_add(&counter->count, delta);
+		if (!rsvd)
+			goto fdblocks_enospc;
+		if (delta > counter->res_avail) {
+			if (ctr == XC_FREE_BLOCKS)
+				xfs_warn_once(mp,
+"Reserve blocks depleted! Consider increasing reserve pool size.");
+			goto fdblocks_enospc;
+		}
+		counter->res_avail -= delta;
+		trace_xfs_freecounter_reserved(mp, ctr, delta, _RET_IP_);
 		spin_unlock(&mp->m_sb_lock);
-		return 0;
 	}
-	xfs_warn_once(mp,
-"Reserve blocks depleted! Consider increasing reserve pool size.");
+
+	/* we had space! */
+	return 0;
 
 fdblocks_enospc:
+	trace_xfs_freecounter_enospc(mp, ctr, delta, _RET_IP_);
 	spin_unlock(&mp->m_sb_lock);
 	return -ENOSPC;
 }