bcachefs: Log truncate operations

Previously, we guaranteed atomicity of truncate after unclean shutdown
with the BCH_INODE_I_SIZE_DIRTY flag - which required a full scan of the
inodes btree.

Recently the deleted inodes btree was added so that we no longer have to
scan for deleted inodes, but truncate was unfinished and that change
left it broken.

This patch uses the new logged operations btree to fix truncate
atomicity; we now log an operation that can be replayed at the start of
a truncate.

Signed-off-by: Kent Overstreet <[e-mail]>

fs/bcachefs/bcachefs_format.h

···
 	x(backpointer,		28)			\
 	x(inode_v3,		29)			\
 	x(bucket_gens,		30)			\
-	x(snapshot_tree,	31)
+	x(snapshot_tree,	31)			\
+	x(logged_op_truncate,	32)
 
 enum bch_bkey_type {
 #define x(name, nr) KEY_TYPE_##name	= nr,
···
 	__BCH_INODE_NODUMP		= 3,
 	__BCH_INODE_NOATIME		= 4,
 
-	__BCH_INODE_I_SIZE_DIRTY	= 5,
-	__BCH_INODE_I_SECTORS_DIRTY	= 6,
+	__BCH_INODE_I_SIZE_DIRTY	= 5, /* obsolete */
+	__BCH_INODE_I_SECTORS_DIRTY	= 6, /* obsolete */
 	__BCH_INODE_UNLINKED		= 7,
 	__BCH_INODE_BACKPTR_UNTRUSTED	= 8,
 
···
 } __packed __aligned(8);
 
 #define LRU_ID_STRIPES		(1U << 16)
+
+/* Logged operations btree: */
+
+struct bch_logged_op_truncate {
+	struct bch_val		v;
+	__le32			subvol;
+	__le32			pad;
+	__le64			inum;
+	__le64			new_i_size;
+};
 
 /* Optional/variable size superblock sections: */
 
···
 	x(deleted_inodes,	16,	BTREE_ID_SNAPSHOTS,			\
 	  BIT_ULL(KEY_TYPE_set))						\
 	x(logged_ops,		17,	0,					\
-	  0)
+	  BIT_ULL(KEY_TYPE_logged_op_truncate))
 
 enum btree_id {
 #define x(name, nr, ...) BTREE_ID_##name = nr,

fs/bcachefs/bkey_methods.c

···
 #include "error.h"
 #include "extents.h"
 #include "inode.h"
+#include "io_misc.h"
 #include "lru.h"
 #include "quota.h"
 #include "reflink.h"

fs/bcachefs/io_misc.c

···
 #include "inode.h"
 #include "io_misc.h"
 #include "io_write.h"
+#include "logged_ops.h"
 #include "subvolume.h"
 
 /* Overwrites whatever was present with zeroes: */
···
 	return ret;
 }
 
+/* truncate: */
+
+void bch2_logged_op_truncate_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
+{
+	struct bkey_s_c_logged_op_truncate op = bkey_s_c_to_logged_op_truncate(k);
+
+	prt_printf(out, "subvol=%u", le32_to_cpu(op.v->subvol));
+	prt_printf(out, " inum=%llu", le64_to_cpu(op.v->inum));
+	prt_printf(out, " new_i_size=%llu", le64_to_cpu(op.v->new_i_size));
+}
+
 static int truncate_set_isize(struct btree_trans *trans,
 			      subvol_inum inum,
 			      u64 new_i_size)
···
 	return ret;
 }
 
-int bch2_truncate(struct bch_fs *c, subvol_inum inum, u64 new_i_size, u64 *i_sectors_delta)
+static int __bch2_resume_logged_op_truncate(struct btree_trans *trans,
+					    struct bkey_i *op_k,
+					    u64 *i_sectors_delta)
 {
-	struct btree_trans trans;
+	struct bch_fs *c = trans->c;
 	struct btree_iter fpunch_iter;
+	struct bkey_i_logged_op_truncate *op = bkey_i_to_logged_op_truncate(op_k);
+	subvol_inum inum = { le32_to_cpu(op->v.subvol), le64_to_cpu(op->v.inum) };
+	u64 new_i_size = le64_to_cpu(op->v.new_i_size);
 	int ret;
 
-	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
-	bch2_trans_iter_init(&trans, &fpunch_iter, BTREE_ID_extents,
+	ret = commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
+			truncate_set_isize(trans, inum, new_i_size));
+	if (ret)
+		goto err;
+
+	bch2_trans_iter_init(trans, &fpunch_iter, BTREE_ID_extents,
 			     POS(inum.inum, round_up(new_i_size, block_bytes(c)) >> 9),
 			     BTREE_ITER_INTENT);
+	ret = bch2_fpunch_at(trans, &fpunch_iter, inum, U64_MAX, i_sectors_delta);
+	bch2_trans_iter_exit(trans, &fpunch_iter);
 
-	ret = commit_do(&trans, NULL, NULL, BTREE_INSERT_NOFAIL,
-			truncate_set_isize(&trans, inum, new_i_size));
-	if (ret)
-		goto err;
-
-	ret = bch2_fpunch_at(&trans, &fpunch_iter, inum, U64_MAX, i_sectors_delta);
 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 		ret = 0;
-	if (ret)
-		goto err;
 err:
-	bch2_trans_iter_exit(&trans, &fpunch_iter);
-	bch2_trans_exit(&trans);
-
-	bch2_fs_fatal_err_on(ret, c, "%s: error truncating %u:%llu: %s",
-			    __func__, inum.subvol, inum.inum, bch2_err_str(ret));
+	bch2_logged_op_finish(trans, op_k);
 	return ret;
+}
+
+int bch2_resume_logged_op_truncate(struct btree_trans *trans, struct bkey_i *op_k)
+{
+	return __bch2_resume_logged_op_truncate(trans, op_k, NULL);
+}
+
+int bch2_truncate(struct bch_fs *c, subvol_inum inum, u64 new_i_size, u64 *i_sectors_delta)
+{
+	struct bkey_i_logged_op_truncate op;
+
+	bkey_logged_op_truncate_init(&op.k_i);
+	op.v.subvol	= cpu_to_le32(inum.subvol);
+	op.v.inum	= cpu_to_le64(inum.inum);
+	op.v.new_i_size	= cpu_to_le64(new_i_size);
+
+	return bch2_trans_run(c,
+		bch2_logged_op_start(&trans, &op.k_i) ?:
+		__bch2_resume_logged_op_truncate(&trans, &op.k_i, i_sectors_delta));
 }
 
 static int adjust_i_size(struct btree_trans *trans, subvol_inum inum, u64 offset, s64 len)

fs/bcachefs/io_misc.h

···
 		   subvol_inum, u64, s64 *);
 int bch2_fpunch(struct bch_fs *c, subvol_inum, u64, u64, s64 *);
 
+void bch2_logged_op_truncate_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+
+#define bch2_bkey_ops_logged_op_truncate ((struct bkey_ops) {	\
+	.val_to_text	= bch2_logged_op_truncate_to_text,	\
+	.min_val_size	= 24,					\
+})
+
+int bch2_resume_logged_op_truncate(struct btree_trans *, struct bkey_i *);
+
 int bch2_truncate(struct bch_fs *, subvol_inum, u64, u64 *);
 int bch2_fcollapse_finsert(struct bch_fs *, subvol_inum, u64, u64, bool, s64 *);
 

fs/bcachefs/logged_ops.c

···
 #include "bkey_buf.h"
 #include "btree_update.h"
 #include "error.h"
+#include "io_misc.h"
 #include "logged_ops.h"
 #include "super.h"
 

fs/bcachefs/logged_ops.h

···
 
 #include "bkey.h"
 
-#define BCH_LOGGED_OPS()
+#define BCH_LOGGED_OPS()			\
+	x(truncate)
 
 static inline int bch2_logged_op_update(struct btree_trans *trans, struct bkey_i *op)
 {