Merge tag 'vectorized-scrub-6.10_2024-04-23' of...

Merge tag 'vectorized-scrub-6.10_2024-04-23' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.10-mergeC

xfs: vectorize scrub kernel calls

Create a vectorized version of the metadata scrub and repair ioctl, and
adapt xfs_scrub to use that.  This mitigates the impact of system call
overhead on xfs_scrub runtime.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>

* tag 'vectorized-scrub-6.10_2024-04-23' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: introduce vectored scrub mode
  xfs: move xfs_ioc_scrub_metadata to scrub.c
  xfs: reduce the rate of cond_resched calls inside scrub

fs/xfs/libxfs/xfs_fs.h

@@ -725,6 +725,15 @@ struct xfs_scrub_metadata {
 /* Number of scrub subcommands. */
 #define XFS_SCRUB_TYPE_NR	29
 
+/*
+ * This special type code only applies to the vectored scrub implementation.
+ *
+ * If any of the previous scrub vectors recorded runtime errors or have
+ * sv_flags bits set that match the OFLAG bits in the barrier vector's
+ * sv_flags, set the barrier's sv_ret to -ECANCELED and return to userspace.
+ */
+#define XFS_SCRUB_TYPE_BARRIER	(0xFFFFFFFF)
+
 /* i: Repair this metadata. */
 #define XFS_SCRUB_IFLAG_REPAIR		(1u << 0)
 
@@ -769,6 +778,29 @@ struct xfs_scrub_metadata {
 				 XFS_SCRUB_OFLAG_NO_REPAIR_NEEDED)
 #define XFS_SCRUB_FLAGS_ALL	(XFS_SCRUB_FLAGS_IN | XFS_SCRUB_FLAGS_OUT)
 
+/* Vectored scrub calls to reduce the number of kernel transitions. */
+
+struct xfs_scrub_vec {
+	__u32 sv_type;		/* XFS_SCRUB_TYPE_* */
+	__u32 sv_flags;		/* XFS_SCRUB_FLAGS_* */
+	__s32 sv_ret;		/* 0 or a negative error code */
+	__u32 sv_reserved;	/* must be zero */
+};
+
+/* Vectored metadata scrub control structure. */
+struct xfs_scrub_vec_head {
+	__u64 svh_ino;		/* inode number. */
+	__u32 svh_gen;		/* inode generation. */
+	__u32 svh_agno;		/* ag number. */
+	__u32 svh_flags;	/* XFS_SCRUB_VEC_FLAGS_* */
+	__u16 svh_rest_us;	/* wait this much time between vector items */
+	__u16 svh_nr;		/* number of svh_vectors */
+	__u64 svh_reserved;	/* must be zero */
+	__u64 svh_vectors;	/* pointer to buffer of xfs_scrub_vec */
+};
+
+#define XFS_SCRUB_VEC_FLAGS_ALL		(0)
+
 /*
  * ioctl limits
  */
@@ -928,6 +960,7 @@ struct xfs_getparents_by_handle {
 #define XFS_IOC_AG_GEOMETRY	_IOWR('X', 61, struct xfs_ag_geometry)
 #define XFS_IOC_GETPARENTS	_IOWR('X', 62, struct xfs_getparents)
 #define XFS_IOC_GETPARENTS_BY_HANDLE _IOWR('X', 63, struct xfs_getparents_by_handle)
+#define XFS_IOC_SCRUBV_METADATA	_IOWR('X', 64, struct xfs_scrub_vec_head)
 
 /*
  * ioctl commands that replace IRIX syssgi()'s

fs/xfs/scrub/common.h

@@ -6,31 +6,6 @@
 #ifndef __XFS_SCRUB_COMMON_H__
 #define __XFS_SCRUB_COMMON_H__
 
-/*
- * We /could/ terminate a scrub/repair operation early.  If we're not
- * in a good place to continue (fatal signal, etc.) then bail out.
- * Note that we're careful not to make any judgements about *error.
- */
-static inline bool
-xchk_should_terminate(
-	struct xfs_scrub	*sc,
-	int			*error)
-{
-	/*
-	 * If preemption is disabled, we need to yield to the scheduler every
-	 * few seconds so that we don't run afoul of the soft lockup watchdog
-	 * or RCU stall detector.
-	 */
-	cond_resched();
-
-	if (fatal_signal_pending(current)) {
-		if (*error == 0)
-			*error = -EINTR;
-		return true;
-	}
-	return false;
-}
-
 int xchk_trans_alloc(struct xfs_scrub *sc, uint resblks);
 int xchk_trans_alloc_empty(struct xfs_scrub *sc);
 void xchk_trans_cancel(struct xfs_scrub *sc);

fs/xfs/scrub/scrub.c

@@ -21,6 +21,7 @@
 #include "xfs_exchmaps.h"
 #include "xfs_dir2.h"
 #include "xfs_parent.h"
+#include "xfs_icache.h"
 #include "scrub/scrub.h"
 #include "scrub/common.h"
 #include "scrub/trace.h"
@@ -578,7 +579,7 @@ xchk_scrub_create_subord(
 }
 
 /* Dispatch metadata scrubbing. */
-int
+STATIC int
 xfs_scrub_metadata(
 	struct file			*file,
 	struct xfs_scrub_metadata	*sm)
@@ -620,6 +621,7 @@ xfs_scrub_metadata(
 	sc->sm = sm;
 	sc->ops = &meta_scrub_ops[sm->sm_type];
 	sc->sick_mask = xchk_health_mask_for_scrub_type(sm->sm_type);
+	sc->relax = INIT_XCHK_RELAX;
 retry_op:
 	/*
 	 * When repairs are allowed, prevent freezing or readonly remount while
@@ -723,3 +725,176 @@ xfs_scrub_metadata(
 	run.retries++;
 	goto retry_op;
 }
+
+/* Scrub one aspect of one piece of metadata. */
+int
+xfs_ioc_scrub_metadata(
+	struct file			*file,
+	void				__user *arg)
+{
+	struct xfs_scrub_metadata	scrub;
+	int				error;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (copy_from_user(&scrub, arg, sizeof(scrub)))
+		return -EFAULT;
+
+	error = xfs_scrub_metadata(file, &scrub);
+	if (error)
+		return error;
+
+	if (copy_to_user(arg, &scrub, sizeof(scrub)))
+		return -EFAULT;
+
+	return 0;
+}
+
+/* Decide if there have been any scrub failures up to this point. */
+static inline int
+xfs_scrubv_check_barrier(
+	struct xfs_mount		*mp,
+	const struct xfs_scrub_vec	*vectors,
+	const struct xfs_scrub_vec	*stop_vec)
+{
+	const struct xfs_scrub_vec	*v;
+	__u32				failmask;
+
+	failmask = stop_vec->sv_flags & XFS_SCRUB_FLAGS_OUT;
+
+	for (v = vectors; v < stop_vec; v++) {
+		if (v->sv_type == XFS_SCRUB_TYPE_BARRIER)
+			continue;
+
+		/*
+		 * Runtime errors count as a previous failure, except the ones
+		 * used to ask userspace to retry.
+		 */
+		switch (v->sv_ret) {
+		case -EBUSY:
+		case -ENOENT:
+		case -EUSERS:
+		case 0:
+			break;
+		default:
+			return -ECANCELED;
+		}
+
+		/*
+		 * If any of the out-flags on the scrub vector match the mask
+		 * that was set on the barrier vector, that's a previous fail.
+		 */
+		if (v->sv_flags & failmask)
+			return -ECANCELED;
+	}
+
+	return 0;
+}
+
+/* Vectored scrub implementation to reduce ioctl calls. */
+int
+xfs_ioc_scrubv_metadata(
+	struct file			*file,
+	void				__user *arg)
+{
+	struct xfs_scrub_vec_head	head;
+	struct xfs_scrub_vec_head	__user *uhead = arg;
+	struct xfs_scrub_vec		*vectors;
+	struct xfs_scrub_vec		__user *uvectors;
+	struct xfs_inode		*ip_in = XFS_I(file_inode(file));
+	struct xfs_mount		*mp = ip_in->i_mount;
+	struct xfs_scrub_vec		*v;
+	size_t				vec_bytes;
+	unsigned int			i;
+	int				error = 0;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (copy_from_user(&head, uhead, sizeof(head)))
+		return -EFAULT;
+
+	if (head.svh_reserved)
+		return -EINVAL;
+	if (head.svh_flags & ~XFS_SCRUB_VEC_FLAGS_ALL)
+		return -EINVAL;
+	if (head.svh_nr == 0)
+		return 0;
+
+	vec_bytes = array_size(head.svh_nr, sizeof(struct xfs_scrub_vec));
+	if (vec_bytes > PAGE_SIZE)
+		return -ENOMEM;
+
+	uvectors = (void __user *)(uintptr_t)head.svh_vectors;
+	vectors = memdup_user(uvectors, vec_bytes);
+	if (IS_ERR(vectors))
+		return PTR_ERR(vectors);
+
+	trace_xchk_scrubv_start(ip_in, &head);
+
+	for (i = 0, v = vectors; i < head.svh_nr; i++, v++) {
+		if (v->sv_reserved) {
+			error = -EINVAL;
+			goto out_free;
+		}
+
+		if (v->sv_type == XFS_SCRUB_TYPE_BARRIER &&
+		    (v->sv_flags & ~XFS_SCRUB_FLAGS_OUT)) {
+			error = -EINVAL;
+			goto out_free;
+		}
+
+		trace_xchk_scrubv_item(mp, &head, i, v);
+	}
+
+	/* Run all the scrubbers. */
+	for (i = 0, v = vectors; i < head.svh_nr; i++, v++) {
+		struct xfs_scrub_metadata	sm = {
+			.sm_type		= v->sv_type,
+			.sm_flags		= v->sv_flags,
+			.sm_ino			= head.svh_ino,
+			.sm_gen			= head.svh_gen,
+			.sm_agno		= head.svh_agno,
+		};
+
+		if (v->sv_type == XFS_SCRUB_TYPE_BARRIER) {
+			v->sv_ret = xfs_scrubv_check_barrier(mp, vectors, v);
+			if (v->sv_ret) {
+				trace_xchk_scrubv_barrier_fail(mp, &head, i, v);
+				break;
+			}
+
+			continue;
+		}
+
+		v->sv_ret = xfs_scrub_metadata(file, &sm);
+		v->sv_flags = sm.sm_flags;
+
+		trace_xchk_scrubv_outcome(mp, &head, i, v);
+
+		if (head.svh_rest_us) {
+			ktime_t		expires;
+
+			expires = ktime_add_ns(ktime_get(),
+					head.svh_rest_us * 1000);
+			set_current_state(TASK_KILLABLE);
+			schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
+		}
+
+		if (fatal_signal_pending(current)) {
+			error = -EINTR;
+			goto out_free;
+		}
+	}
+
+	if (copy_to_user(uvectors, vectors, vec_bytes) ||
+	    copy_to_user(uhead, &head, sizeof(head))) {
+		error = -EFAULT;
+		goto out_free;
+	}
+
+out_free:
+	kfree(vectors);
+	return error;
+}

fs/xfs/scrub/scrub.h

@@ -8,6 +8,49 @@
 
 struct xfs_scrub;
 
+struct xchk_relax {
+	unsigned long	next_resched;
+	unsigned int	resched_nr;
+	bool		interruptible;
+};
+
+/* Yield to the scheduler at most 10x per second. */
+#define XCHK_RELAX_NEXT		(jiffies + (HZ / 10))
+
+#define INIT_XCHK_RELAX	\
+	(struct xchk_relax){ \
+		.next_resched	= XCHK_RELAX_NEXT, \
+		.resched_nr	= 0, \
+		.interruptible	= true, \
+	}
+
+/*
+ * Relax during a scrub operation and exit if there's a fatal signal pending.
+ *
+ * If preemption is disabled, we need to yield to the scheduler every now and
+ * then so that we don't run afoul of the soft lockup watchdog or RCU stall
+ * detector.  cond_resched calls are somewhat expensive (~5ns) so we want to
+ * ratelimit this to 10x per second.  Amortize the cost of the other checks by
+ * only doing it once every 100 calls.
+ */
+static inline int xchk_maybe_relax(struct xchk_relax *widget)
+{
+	/* Amortize the cost of scheduling and checking signals. */
+	if (likely(++widget->resched_nr < 100))
+		return 0;
+	widget->resched_nr = 0;
+
+	if (unlikely(widget->next_resched <= jiffies)) {
+		cond_resched();
+		widget->next_resched = XCHK_RELAX_NEXT;
+	}
+
+	if (widget->interruptible && fatal_signal_pending(current))
+		return -EINTR;
+
+	return 0;
+}
+
 /*
  * Standard flags for allocating memory within scrub.  NOFS context is
  * configured by the process allocation scope.  Scrub and repair must be able
@@ -123,6 +166,9 @@ struct xfs_scrub {
 	 */
 	unsigned int			sick_mask;
 
+	/* next time we want to cond_resched() */
+	struct xchk_relax		relax;
+
 	/* State tracking for single-AG operations. */
 	struct xchk_ag			sa;
 };
@@ -167,6 +213,24 @@ struct xfs_scrub_subord *xchk_scrub_create_subord(struct xfs_scrub *sc,
 		unsigned int subtype);
 void xchk_scrub_free_subord(struct xfs_scrub_subord *sub);
 
+/*
+ * We /could/ terminate a scrub/repair operation early.  If we're not
+ * in a good place to continue (fatal signal, etc.) then bail out.
+ * Note that we're careful not to make any judgements about *error.
+ */
+static inline bool
+xchk_should_terminate(
+	struct xfs_scrub	*sc,
+	int			*error)
+{
+	if (xchk_maybe_relax(&sc->relax)) {
+		if (*error == 0)
+			*error = -EINTR;
+		return true;
+	}
+	return false;
+}
+
 /* Metadata scrubbers */
 int xchk_tester(struct xfs_scrub *sc);
 int xchk_superblock(struct xfs_scrub *sc);

fs/xfs/scrub/trace.h

@@ -69,6 +69,7 @@ TRACE_DEFINE_ENUM(XFS_SCRUB_TYPE_QUOTACHECK);
 TRACE_DEFINE_ENUM(XFS_SCRUB_TYPE_NLINKS);
 TRACE_DEFINE_ENUM(XFS_SCRUB_TYPE_HEALTHY);
 TRACE_DEFINE_ENUM(XFS_SCRUB_TYPE_DIRTREE);
+TRACE_DEFINE_ENUM(XFS_SCRUB_TYPE_BARRIER);
 
 #define XFS_SCRUB_TYPE_STRINGS \
 	{ XFS_SCRUB_TYPE_PROBE,		"probe" }, \
@@ -99,7 +100,8 @@ TRACE_DEFINE_ENUM(XFS_SCRUB_TYPE_DIRTREE);
 	{ XFS_SCRUB_TYPE_QUOTACHECK,	"quotacheck" }, \
 	{ XFS_SCRUB_TYPE_NLINKS,	"nlinks" }, \
 	{ XFS_SCRUB_TYPE_HEALTHY,	"healthy" }, \
-	{ XFS_SCRUB_TYPE_DIRTREE,	"dirtree" }
+	{ XFS_SCRUB_TYPE_DIRTREE,	"dirtree" }, \
+	{ XFS_SCRUB_TYPE_BARRIER,	"barrier" }
 
 #define XFS_SCRUB_FLAG_STRINGS \
 	{ XFS_SCRUB_IFLAG_REPAIR,		"repair" }, \
@@ -208,6 +210,81 @@ DEFINE_EVENT(xchk_fsgate_class, name, \
 DEFINE_SCRUB_FSHOOK_EVENT(xchk_fsgates_enable);
 DEFINE_SCRUB_FSHOOK_EVENT(xchk_fsgates_disable);
 
+DECLARE_EVENT_CLASS(xchk_vector_head_class,
+	TP_PROTO(struct xfs_inode *ip, struct xfs_scrub_vec_head *vhead),
+	TP_ARGS(ip, vhead),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(xfs_ino_t, ino)
+		__field(xfs_agnumber_t, agno)
+		__field(xfs_ino_t, inum)
+		__field(unsigned int, gen)
+		__field(unsigned int, flags)
+		__field(unsigned short, rest_us)
+		__field(unsigned short, nr_vecs)
+	),
+	TP_fast_assign(
+		__entry->dev = ip->i_mount->m_super->s_dev;
+		__entry->ino = ip->i_ino;
+		__entry->agno = vhead->svh_agno;
+		__entry->inum = vhead->svh_ino;
+		__entry->gen = vhead->svh_gen;
+		__entry->flags = vhead->svh_flags;
+		__entry->rest_us = vhead->svh_rest_us;
+		__entry->nr_vecs = vhead->svh_nr;
+	),
+	TP_printk("dev %d:%d ino 0x%llx agno 0x%x inum 0x%llx gen 0x%x flags 0x%x rest_us %u nr_vecs %u",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ino,
+		  __entry->agno,
+		  __entry->inum,
+		  __entry->gen,
+		  __entry->flags,
+		  __entry->rest_us,
+		  __entry->nr_vecs)
+)
+#define DEFINE_SCRUBV_HEAD_EVENT(name) \
+DEFINE_EVENT(xchk_vector_head_class, name, \
+	TP_PROTO(struct xfs_inode *ip, struct xfs_scrub_vec_head *vhead), \
+	TP_ARGS(ip, vhead))
+
+DEFINE_SCRUBV_HEAD_EVENT(xchk_scrubv_start);
+
+DECLARE_EVENT_CLASS(xchk_vector_class,
+	TP_PROTO(struct xfs_mount *mp, struct xfs_scrub_vec_head *vhead,
+		 unsigned int vec_nr, struct xfs_scrub_vec *v),
+	TP_ARGS(mp, vhead, vec_nr, v),
+	TP_STRUCT__entry(
+		__field(dev_t, dev)
+		__field(unsigned int, vec_nr)
+		__field(unsigned int, vec_type)
+		__field(unsigned int, vec_flags)
+		__field(int, vec_ret)
+	),
+	TP_fast_assign(
+		__entry->dev = mp->m_super->s_dev;
+		__entry->vec_nr = vec_nr;
+		__entry->vec_type = v->sv_type;
+		__entry->vec_flags = v->sv_flags;
+		__entry->vec_ret = v->sv_ret;
+	),
+	TP_printk("dev %d:%d vec[%u] type %s flags %s ret %d",
+		  MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->vec_nr,
+		  __print_symbolic(__entry->vec_type, XFS_SCRUB_TYPE_STRINGS),
+		  __print_flags(__entry->vec_flags, "|", XFS_SCRUB_FLAG_STRINGS),
+		  __entry->vec_ret)
+)
+#define DEFINE_SCRUBV_EVENT(name) \
+DEFINE_EVENT(xchk_vector_class, name, \
+	TP_PROTO(struct xfs_mount *mp, struct xfs_scrub_vec_head *vhead, \
+		 unsigned int vec_nr, struct xfs_scrub_vec *v), \
+	TP_ARGS(mp, vhead, vec_nr, v))
+
+DEFINE_SCRUBV_EVENT(xchk_scrubv_barrier_fail);
+DEFINE_SCRUBV_EVENT(xchk_scrubv_item);
+DEFINE_SCRUBV_EVENT(xchk_scrubv_outcome);
+
 TRACE_EVENT(xchk_op_error,
 	TP_PROTO(struct xfs_scrub *sc, xfs_agnumber_t agno,
 		 xfs_agblock_t bno, int error, void *ret_ip),

fs/xfs/scrub/xfarray.c

@@ -7,9 +7,9 @@
 #include "xfs_fs.h"
 #include "xfs_shared.h"
 #include "xfs_format.h"
+#include "scrub/scrub.h"
 #include "scrub/xfile.h"
 #include "scrub/xfarray.h"
-#include "scrub/scrub.h"
 #include "scrub/trace.h"
 
 /*
@@ -486,6 +486,9 @@ xfarray_sortinfo_alloc(
 
 	xfarray_sortinfo_lo(si)[0] = 0;
 	xfarray_sortinfo_hi(si)[0] = array->nr - 1;
+	si->relax = INIT_XCHK_RELAX;
+	if (flags & XFARRAY_SORT_KILLABLE)
+		si->relax.interruptible = false;
 
 	trace_xfarray_sort(si, nr_bytes);
 	*infop = si;
@@ -503,10 +506,7 @@ xfarray_sort_terminated(
 	 * few seconds so that we don't run afoul of the soft lockup watchdog
 	 * or RCU stall detector.
 	 */
-	cond_resched();
-
-	if ((si->flags & XFARRAY_SORT_KILLABLE) &&
-	    fatal_signal_pending(current)) {
+	if (xchk_maybe_relax(&si->relax)) {
 		if (*error == 0)
 			*error = -EINTR;
 		return true;

fs/xfs/scrub/xfarray.h

@@ -127,6 +127,9 @@ struct xfarray_sortinfo {
 	/* XFARRAY_SORT_* flags; see below. */
 	unsigned int		flags;
 
+	/* next time we want to cond_resched() */
+	struct xchk_relax	relax;
+
 	/* Cache a folio here for faster scanning for pivots */
 	struct folio		*folio;
 

fs/xfs/scrub/xfile.c

@@ -10,9 +10,9 @@
 #include "xfs_log_format.h"
 #include "xfs_trans_resv.h"
 #include "xfs_mount.h"
+#include "scrub/scrub.h"
 #include "scrub/xfile.h"
 #include "scrub/xfarray.h"
-#include "scrub/scrub.h"
 #include "scrub/trace.h"
 #include <linux/shmem_fs.h>
 

fs/xfs/scrub/xfs_scrub.h

@@ -7,9 +7,11 @@
 #define __XFS_SCRUB_H__
 
 #ifndef CONFIG_XFS_ONLINE_SCRUB
-# define xfs_scrub_metadata(file, sm)	(-ENOTTY)
+# define xfs_ioc_scrub_metadata(f, a)	(-ENOTTY)
+# define xfs_ioc_scrubv_metadata(f, a)	(-ENOTTY)
 #else
-int xfs_scrub_metadata(struct file *file, struct xfs_scrub_metadata *sm);
+int xfs_ioc_scrub_metadata(struct file *file, void __user *arg);
+int xfs_ioc_scrubv_metadata(struct file *file, void __user *arg);
 #endif /* CONFIG_XFS_ONLINE_SCRUB */
 
 #endif	/* __XFS_SCRUB_H__ */

fs/xfs/xfs_ioctl.c

@@ -1055,30 +1055,6 @@ xfs_ioc_getfsmap(
 	return error;
 }
 
-STATIC int
-xfs_ioc_scrub_metadata(
-	struct file			*file,
-	void				__user *arg)
-{
-	struct xfs_scrub_metadata	scrub;
-	int				error;
-
-	if (!capable(CAP_SYS_ADMIN))
-		return -EPERM;
-
-	if (copy_from_user(&scrub, arg, sizeof(scrub)))
-		return -EFAULT;
-
-	error = xfs_scrub_metadata(file, &scrub);
-	if (error)
-		return error;
-
-	if (copy_to_user(arg, &scrub, sizeof(scrub)))
-		return -EFAULT;
-
-	return 0;
-}
-
 int
 xfs_ioc_swapext(
 	xfs_swapext_t	*sxp)
@@ -1437,6 +1413,8 @@ xfs_file_ioctl(
 	case FS_IOC_GETFSMAP:
 		return xfs_ioc_getfsmap(ip, arg);
 
+	case XFS_IOC_SCRUBV_METADATA:
+		return xfs_ioc_scrubv_metadata(filp, arg);
 	case XFS_IOC_SCRUB_METADATA:
 		return xfs_ioc_scrub_metadata(filp, arg);