Update bcachefs sources to 22fa8fc32e6a bcachefs: rcu_pending now wor…

…ks in userspace

.bcachefs_revision

@@ -1 +1 @@
-62439c6f1a6dba3fca1e57f352745d6e36dd1e31
+22fa8fc32e6aafb8bd76c6b746868dbdbc6a934d

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "bcachefs-tools"
-version = "1.11.1"
+version = "1.12.0"
 authors = ["Yuxuan Shui <[email protected]>", "Kayla Firestack <[email protected]>", "Kent Overstreet <[email protected]>" ]
 edition = "2021"
 rust-version = "1.70"

Makefile

@@ -1,4 +1,4 @@
-VERSION=1.11.1
+VERSION=1.12.0
 
 PREFIX?=/usr/local
 LIBEXECDIR?=$(PREFIX)/libexec

include/linux/generic-radix-tree.h

@@ -41,17 +41,71 @@
 #include <linux/limits.h>
 #include <linux/log2.h>
 #include <linux/math.h>
+#include <linux/slab.h>
 #include <linux/types.h>
 
 struct genradix_root;
 
 #define GENRADIX_NODE_SHIFT	9
 #define GENRADIX_NODE_SIZE	(1U << GENRADIX_NODE_SHIFT)
 
+#define GENRADIX_ARY		(GENRADIX_NODE_SIZE / sizeof(struct genradix_node *))
+#define GENRADIX_ARY_SHIFT	ilog2(GENRADIX_ARY)
+
+/* depth that's needed for a genradix that can address up to ULONG_MAX: */
+#define GENRADIX_MAX_DEPTH	\
+	DIV_ROUND_UP(BITS_PER_LONG - GENRADIX_NODE_SHIFT, GENRADIX_ARY_SHIFT)
+
+#define GENRADIX_DEPTH_MASK				\
+	((unsigned long) (roundup_pow_of_two(GENRADIX_MAX_DEPTH + 1) - 1))
+
+static inline int genradix_depth_shift(unsigned depth)
+{
+	return GENRADIX_NODE_SHIFT + GENRADIX_ARY_SHIFT * depth;
+}
+
+/*
+ * Returns size (of data, in bytes) that a tree of a given depth holds:
+ */
+static inline size_t genradix_depth_size(unsigned depth)
+{
+	return 1UL << genradix_depth_shift(depth);
+}
+
+static inline unsigned genradix_root_to_depth(struct genradix_root *r)
+{
+	return (unsigned long) r & GENRADIX_DEPTH_MASK;
+}
+
+static inline struct genradix_node *genradix_root_to_node(struct genradix_root *r)
+{
+	return (void *) ((unsigned long) r & ~GENRADIX_DEPTH_MASK);
+}
+
 struct __genradix {
 	struct genradix_root		*root;
 };
 
+struct genradix_node {
+	union {
+		/* Interior node: */
+		struct genradix_node	*children[GENRADIX_ARY];
+
+		/* Leaf: */
+		u8			data[GENRADIX_NODE_SIZE];
+	};
+};
+
+static inline struct genradix_node *genradix_alloc_node(gfp_t gfp_mask)
+{
+	return kzalloc(GENRADIX_NODE_SIZE, gfp_mask);
+}
+
+static inline void genradix_free_node(struct genradix_node *node)
+{
+	kfree(node);
+}
+
 /*
  * NOTE: currently, sizeof(_type) must not be larger than GENRADIX_NODE_SIZE:
  */
@@ -128,6 +182,30 @@ static inline size_t __idx_to_offset(size_t idx, size_t obj_size)
 #define __genradix_idx_to_offset(_radix, _idx)			\
 	__idx_to_offset(_idx, __genradix_obj_size(_radix))
 
+static inline void *__genradix_ptr_inlined(struct __genradix *radix, size_t offset)
+{
+	struct genradix_root *r = READ_ONCE(radix->root);
+	struct genradix_node *n = genradix_root_to_node(r);
+	unsigned level		= genradix_root_to_depth(r);
+	unsigned shift		= genradix_depth_shift(level);
+
+	if (unlikely(ilog2(offset) >= genradix_depth_shift(level)))
+		return NULL;
+
+	while (n && shift > GENRADIX_NODE_SHIFT) {
+		shift -= GENRADIX_ARY_SHIFT;
+		n = n->children[offset >> shift];
+		offset &= (1UL << shift) - 1;
+	}
+
+	return n ? &n->data[offset] : NULL;
+}
+
+#define genradix_ptr_inlined(_radix, _idx)			\
+	(__genradix_cast(_radix)				\
+	 __genradix_ptr_inlined(&(_radix)->tree,		\
+			__genradix_idx_to_offset(_radix, _idx)))
+
 void *__genradix_ptr(struct __genradix *, size_t);
 
 /**
@@ -142,7 +220,24 @@ void *__genradix_ptr(struct __genradix *, size_t);
 	 __genradix_ptr(&(_radix)->tree,			\
 			__genradix_idx_to_offset(_radix, _idx)))
 
-void *__genradix_ptr_alloc(struct __genradix *, size_t, gfp_t);
+void *__genradix_ptr_alloc(struct __genradix *, size_t,
+			   struct genradix_node **, gfp_t);
+
+#define genradix_ptr_alloc_inlined(_radix, _idx, _gfp)			\
+	(__genradix_cast(_radix)					\
+	 (__genradix_ptr_inlined(&(_radix)->tree,			\
+			__genradix_idx_to_offset(_radix, _idx)) ?:	\
+	  __genradix_ptr_alloc(&(_radix)->tree,				\
+			__genradix_idx_to_offset(_radix, _idx),		\
+			NULL, _gfp)))
+
+#define genradix_ptr_alloc_preallocated_inlined(_radix, _idx, _new_node, _gfp)\
+	(__genradix_cast(_radix)					\
+	 (__genradix_ptr_inlined(&(_radix)->tree,			\
+			__genradix_idx_to_offset(_radix, _idx)) ?:	\
+	  __genradix_ptr_alloc(&(_radix)->tree,				\
+			__genradix_idx_to_offset(_radix, _idx),		\
+			_new_node, _gfp)))
 
 /**
  * genradix_ptr_alloc - get a pointer to a genradix entry, allocating it
@@ -157,7 +252,13 @@ void *__genradix_ptr_alloc(struct __genradix *, size_t, gfp_t);
 	(__genradix_cast(_radix)				\
 	 __genradix_ptr_alloc(&(_radix)->tree,			\
 			__genradix_idx_to_offset(_radix, _idx),	\
-			_gfp))
+			NULL, _gfp))
+
+#define genradix_ptr_alloc_preallocated(_radix, _idx, _new_node, _gfp)\
+	(__genradix_cast(_radix)				\
+	 __genradix_ptr_alloc(&(_radix)->tree,			\
+			__genradix_idx_to_offset(_radix, _idx),	\
+			_new_node, _gfp))
 
 struct genradix_iter {
 	size_t			offset;

include/linux/mm.h

@@ -3,6 +3,7 @@
 #define _TOOLS_LINUX_MM_H
 
 #include <sys/syscall.h>
+#include <unistd.h>
 #include <linux/types.h>
 
 struct sysinfo {

include/linux/slab.h

@@ -18,6 +18,7 @@
 #define alloc_hooks(_do, ...)		_do
 
 #define ARCH_KMALLOC_MINALIGN		16
+#define ARCH_SLAB_MINALIGN		16
 #define KMALLOC_MAX_SIZE		SIZE_MAX
 
 #define MAX_PAGE_ORDER			10
@@ -102,6 +103,8 @@ static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
 	((size) != 0 && (n) > SIZE_MAX / (size)				\
 	 ? NULL : kmalloc((n) * (size), flags))
 
+#define kvmalloc_array_noprof(...)	kvmalloc_array(__VA_ARGS__)
+
 #define kcalloc(n, size, flags)		kmalloc_array(n, size, flags|__GFP_ZERO)
 
 #define kfree(p)			free((void *) p)

include/linux/srcu.h

@@ -1,6 +1,12 @@
 #ifndef __TOOLS_LINUX_SRCU_H
 #define __TOOLS_LINUX_SRCU_H
 
+#include <linux/rcupdate.h>
+
+#define NUM_ACTIVE_RCU_POLL_OLDSTATE	2
+
+typedef void (*rcu_callback_t)(struct rcu_head *head);
+
 struct srcu_struct {
 };
 
@@ -26,10 +32,35 @@ static inline unsigned long get_state_synchronize_srcu(struct srcu_struct *ssp)
 	return 0;
 }
 
+#undef poll_state_synchronize_rcu
+static inline bool poll_state_synchronize_rcu(unsigned long cookie)
+{
+	return false;
+}
+
+#undef start_poll_synchronize_rcu
+static inline unsigned long start_poll_synchronize_rcu()
+{
+	return 0;
+}
+
+static inline unsigned long get_state_synchronize_rcu()
+{
+	return 0;
+}
+
 static inline void synchronize_srcu_expedited(struct srcu_struct *ssp) {}
 
+static inline void srcu_barrier(struct srcu_struct *ssp) {}
+
 static inline void cleanup_srcu_struct(struct srcu_struct *ssp) {}
 
+static inline void call_srcu(struct srcu_struct *ssp, struct rcu_head *rhp,
+			     rcu_callback_t func)
+{
+	func(rhp);
+}
+
 static inline int init_srcu_struct(struct srcu_struct *ssp)
 {
 	return 0;

libbcachefs/alloc_background.c

@@ -30,6 +30,7 @@
 #include <linux/rcupdate.h>
 #include <linux/sched/task.h>
 #include <linux/sort.h>
+#include <linux/jiffies.h>
 
 static void bch2_discard_one_bucket_fast(struct bch_dev *, u64);
 
@@ -240,71 +241,73 @@ int bch2_alloc_v3_validate(struct bch_fs *c, struct bkey_s_c k,
 int bch2_alloc_v4_validate(struct bch_fs *c, struct bkey_s_c k,
 			   enum bch_validate_flags flags)
 {
-	struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
+	struct bch_alloc_v4 a;
 	int ret = 0;
 
-	bkey_fsck_err_on(alloc_v4_u64s_noerror(a.v) > bkey_val_u64s(k.k),
+	bkey_val_copy(&a, bkey_s_c_to_alloc_v4(k));
+
+	bkey_fsck_err_on(alloc_v4_u64s_noerror(&a) > bkey_val_u64s(k.k),
 			 c, alloc_v4_val_size_bad,
 			 "bad val size (%u > %zu)",
-			 alloc_v4_u64s_noerror(a.v), bkey_val_u64s(k.k));
+			 alloc_v4_u64s_noerror(&a), bkey_val_u64s(k.k));
 
-	bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
-			 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v),
+	bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(&a) &&
+			 BCH_ALLOC_V4_NR_BACKPOINTERS(&a),
 			 c, alloc_v4_backpointers_start_bad,
 			 "invalid backpointers_start");
 
-	bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type,
+	bkey_fsck_err_on(alloc_data_type(a, a.data_type) != a.data_type,
 			 c, alloc_key_data_type_bad,
 			 "invalid data type (got %u should be %u)",
-			 a.v->data_type, alloc_data_type(*a.v, a.v->data_type));
+			 a.data_type, alloc_data_type(a, a.data_type));
 
 	for (unsigned i = 0; i < 2; i++)
-		bkey_fsck_err_on(a.v->io_time[i] > LRU_TIME_MAX,
+		bkey_fsck_err_on(a.io_time[i] > LRU_TIME_MAX,
 				 c, alloc_key_io_time_bad,
 				 "invalid io_time[%s]: %llu, max %llu",
 				 i == READ ? "read" : "write",
-				 a.v->io_time[i], LRU_TIME_MAX);
+				 a.io_time[i], LRU_TIME_MAX);
 
-	unsigned stripe_sectors = BCH_ALLOC_V4_BACKPOINTERS_START(a.v) * sizeof(u64) >
+	unsigned stripe_sectors = BCH_ALLOC_V4_BACKPOINTERS_START(&a) * sizeof(u64) >
 		offsetof(struct bch_alloc_v4, stripe_sectors)
-		? a.v->stripe_sectors
+		? a.stripe_sectors
 		: 0;
 
-	switch (a.v->data_type) {
+	switch (a.data_type) {
 	case BCH_DATA_free:
 	case BCH_DATA_need_gc_gens:
 	case BCH_DATA_need_discard:
 		bkey_fsck_err_on(stripe_sectors ||
-				 a.v->dirty_sectors ||
-				 a.v->cached_sectors ||
-				 a.v->stripe,
+				 a.dirty_sectors ||
+				 a.cached_sectors ||
+				 a.stripe,
 				 c, alloc_key_empty_but_have_data,
 				 "empty data type free but have data %u.%u.%u %u",
 				 stripe_sectors,
-				 a.v->dirty_sectors,
-				 a.v->cached_sectors,
-				 a.v->stripe);
+				 a.dirty_sectors,
+				 a.cached_sectors,
+				 a.stripe);
 		break;
 	case BCH_DATA_sb:
 	case BCH_DATA_journal:
 	case BCH_DATA_btree:
 	case BCH_DATA_user:
 	case BCH_DATA_parity:
-		bkey_fsck_err_on(!a.v->dirty_sectors &&
+		bkey_fsck_err_on(!a.dirty_sectors &&
 				 !stripe_sectors,
 				 c, alloc_key_dirty_sectors_0,
 				 "data_type %s but dirty_sectors==0",
-				 bch2_data_type_str(a.v->data_type));
+				 bch2_data_type_str(a.data_type));
 		break;
 	case BCH_DATA_cached:
-		bkey_fsck_err_on(!a.v->cached_sectors ||
-				 a.v->dirty_sectors ||
+		bkey_fsck_err_on(!a.cached_sectors ||
+				 a.dirty_sectors ||
 				 stripe_sectors ||
-				 a.v->stripe,
+				 a.stripe,
 				 c, alloc_key_cached_inconsistency,
 				 "data type inconsistency");
 
-		bkey_fsck_err_on(!a.v->io_time[READ] &&
+		bkey_fsck_err_on(!a.io_time[READ] &&
 				 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs,
 				 c, alloc_key_cached_but_read_time_zero,
 				 "cached bucket with read_time == 0");
@@ -1872,26 +1875,26 @@ static void bch2_do_discards_work(struct work_struct *work)
 	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,
 			      bch2_err_str(ret));
 
-	bch2_write_ref_put(c, BCH_WRITE_REF_discard);
 	percpu_ref_put(&ca->io_ref);
+	bch2_write_ref_put(c, BCH_WRITE_REF_discard);
 }
 
 void bch2_dev_do_discards(struct bch_dev *ca)
 {
 	struct bch_fs *c = ca->fs;
 
-	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard))
 		return;
 
-	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard))
-		goto put_ioref;
+	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+		goto put_write_ref;
 
 	if (queue_work(c->write_ref_wq, &ca->discard_work))
 		return;
 
-	bch2_write_ref_put(c, BCH_WRITE_REF_discard);
-put_ioref:
 	percpu_ref_put(&ca->io_ref);
+put_write_ref:
+	bch2_write_ref_put(c, BCH_WRITE_REF_discard);
 }
 
 void bch2_do_discards(struct bch_fs *c)
@@ -2181,7 +2184,7 @@ int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
 	 * freespace/need_discard/need_gc_gens btrees as needed:
 	 */
 	while (1) {
-		if (last_updated + HZ * 10 < jiffies) {
+		if (time_after(jiffies, last_updated + HZ * 10)) {
 			bch_info(ca, "%s: currently at %llu/%llu",
 				 __func__, iter.pos.offset, ca->mi.nbuckets);
 			last_updated = jiffies;

libbcachefs/alloc_background_format.h

@@ -69,6 +69,7 @@ struct bch_alloc_v4 {
 	__u64			io_time[2];
 	__u32			stripe;
 	__u32			nr_external_backpointers;
+	/* end of fields in original version of alloc_v4 */
 	__u64			fragmentation_lru;
 	__u32			stripe_sectors;
 	__u32			pad;