1 files changed, 117 insertions, 43 deletions

diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 2456c41b215e..a233f45875e9 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -7,6 +7,7 @@
 #include "btree_update.h"
 #include "buckets.h"
 #include "clock.h"
+#include "enumerated_ref.h"
 #include "error.h"
 #include "extents.h"
 #include "extent_update.h"
@@ -48,7 +49,8 @@ static void nocow_flush_endio(struct bio *_bio)
 	struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio);
 
 	closure_put(bio->cl);
-	percpu_ref_put(&bio->ca->io_ref);
+	enumerated_ref_put(&bio->ca->io_ref[WRITE],
+			   BCH_DEV_WRITE_REF_nocow_flush);
 	bio_put(&bio->bio);
 }
 
@@ -69,11 +71,12 @@ void bch2_inode_flush_nocow_writes_async(struct bch_fs *c,
 	memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
 
 	for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) {
-		rcu_read_lock();
-		ca = rcu_dereference(c->devs[dev]);
-		if (ca && !percpu_ref_tryget(&ca->io_ref))
-			ca = NULL;
-		rcu_read_unlock();
+		scoped_guard(rcu) {
+			ca = rcu_dereference(c->devs[dev]);
+			if (ca && !enumerated_ref_tryget(&ca->io_ref[WRITE],
+							 BCH_DEV_WRITE_REF_nocow_flush))
+				ca = NULL;
+		}
 
 		if (!ca)
 			continue;
@@ -144,10 +147,24 @@ int __must_check bch2_write_inode_size(struct bch_fs *c,
 void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
 			   struct quota_res *quota_res, s64 sectors)
 {
-	bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c,
-				"inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)",
-				inode->v.i_ino, (u64) inode->v.i_blocks, sectors,
-				inode->ei_inode.bi_sectors);
+	if (unlikely((s64) inode->v.i_blocks + sectors < 0)) {
+		struct printbuf buf = PRINTBUF;
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf, "inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)",
+			   inode->v.i_ino, (u64) inode->v.i_blocks, sectors,
+			   inode->ei_inode.bi_sectors);
+
+		bool print = bch2_count_fsck_err(c, vfs_inode_i_blocks_underflow, &buf);
+		if (print)
+			bch2_print_str(c, KERN_ERR, buf.buf);
+		printbuf_exit(&buf);
+
+		if (sectors < 0)
+			sectors = -inode->v.i_blocks;
+		else
+			sectors = 0;
+	}
+
 	inode->v.i_blocks += sectors;
 
 #ifdef CONFIG_BCACHEFS_QUOTA
@@ -167,6 +184,34 @@ void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
 
 /* fsync: */
 
+static int bch2_get_inode_journal_seq_trans(struct btree_trans *trans, subvol_inum inum,
+					    u64 *seq)
+{
+	struct printbuf buf = PRINTBUF;
+	struct bch_inode_unpacked u;
+	struct btree_iter iter;
+	int ret = bch2_inode_peek(trans, &iter, &u, inum, 0);
+	if (ret)
+		return ret;
+
+	u64 cur_seq = journal_cur_seq(&trans->c->journal);
+	*seq = min(cur_seq, u.bi_journal_seq);
+
+	if (fsck_err_on(u.bi_journal_seq > cur_seq,
+			trans, inode_journal_seq_in_future,
+			"inode journal seq in future (currently at %llu)\n%s",
+			cur_seq,
+			(bch2_inode_unpacked_to_text(&buf, &u),
+			buf.buf))) {
+		u.bi_journal_seq = cur_seq;
+		ret = bch2_inode_write(trans, &iter, &u);
+	}
+fsck_err:
+	bch2_trans_iter_exit(trans, &iter);
+	printbuf_exit(&buf);
+	return ret;
+}
+
 /*
  * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an
  * insert trigger: look up the btree inode instead
@@ -177,14 +222,15 @@ static int bch2_flush_inode(struct bch_fs *c,
 	if (c->opts.journal_flush_disabled)
 		return 0;
 
-	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fsync))
+	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_fsync))
 		return -EROFS;
 
-	struct bch_inode_unpacked u;
-	int ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u) ?:
-		  bch2_journal_flush_seq(&c->journal, u.bi_journal_seq, TASK_INTERRUPTIBLE) ?:
+	u64 seq;
+	int ret = bch2_trans_commit_do(c, NULL, NULL, 0,
+			bch2_get_inode_journal_seq_trans(trans, inode_inum(inode), &seq)) ?:
+		  bch2_journal_flush_seq(&c->journal, seq, TASK_INTERRUPTIBLE) ?:
 		  bch2_inode_flush_nocow_writes(c, inode);
-	bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
+	enumerated_ref_put(&c->writes, BCH_WRITE_REF_fsync);
 	return ret;
 }
 
@@ -222,7 +268,7 @@ static inline int range_has_data(struct bch_fs *c, u32 subvol,
 				 struct bpos end)
 {
 	return bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents, start, end,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents, start, end,
 						    subvol, 0, k, ({
 			bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k);
 		})));
@@ -256,7 +302,7 @@ static int __bch2_truncate_folio(struct bch_inode_info *inode,
 
 		folio = __filemap_get_folio(mapping, index,
 					    FGP_LOCK|FGP_CREAT, GFP_KERNEL);
-		if (IS_ERR_OR_NULL(folio)) {
+		if (IS_ERR(folio)) {
 			ret = -ENOMEM;
 			goto out;
 		}
@@ -437,6 +483,7 @@ int bchfs_truncate(struct mnt_idmap *idmap,
 	ret = bch2_truncate_folio(inode, iattr->ia_size);
 	if (unlikely(ret < 0))
 		goto err;
+	ret = 0;
 
 	truncate_setsize(&inode->v, iattr->ia_size);
 
@@ -472,11 +519,20 @@ int bchfs_truncate(struct mnt_idmap *idmap,
 		goto err;
 	}
 
-	bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks &&
-				!bch2_journal_error(&c->journal), c,
-				"inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
-				inode->v.i_ino, (u64) inode->v.i_blocks,
-				inode->ei_inode.bi_sectors);
+	if (unlikely(!inode->v.i_size && inode->v.i_blocks &&
+		     !bch2_journal_error(&c->journal))) {
+		struct printbuf buf = PRINTBUF;
+		bch2_log_msg_start(c, &buf);
+		prt_printf(&buf,
+			   "inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
+			   inode->v.i_ino, (u64) inode->v.i_blocks,
+			   inode->ei_inode.bi_sectors);
+
+		bool print = bch2_count_fsck_err(c, vfs_inode_i_blocks_not_zero_at_truncate, &buf);
+		if (print)
+			bch2_print_str(c, KERN_ERR, buf.buf);
+		printbuf_exit(&buf);
+	}
 
 	ret = bch2_setattr_nonsize(idmap, inode, iattr);
 err:
@@ -606,9 +662,9 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 		if (ret)
 			goto bkey_err;
 
-		bch2_btree_iter_set_snapshot(&iter, snapshot);
+		bch2_btree_iter_set_snapshot(trans, &iter, snapshot);
 
-		k = bch2_btree_iter_peek_slot(&iter);
+		k = bch2_btree_iter_peek_slot(trans, &iter);
 		if ((ret = bkey_err(k)))
 			goto bkey_err;
 
@@ -619,13 +675,13 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 		/* already reserved */
 		if (bkey_extent_is_reservation(k) &&
 		    bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) {
-			bch2_btree_iter_advance(&iter);
+			bch2_btree_iter_advance(trans, &iter);
 			continue;
 		}
 
 		if (bkey_extent_is_data(k.k) &&
 		    !(mode & FALLOC_FL_ZERO_RANGE)) {
-			bch2_btree_iter_advance(&iter);
+			bch2_btree_iter_advance(trans, &iter);
 			continue;
 		}
 
@@ -646,7 +702,7 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 				if (ret)
 					goto bkey_err;
 			}
-			bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
+			bch2_btree_iter_set_pos(trans, &iter, POS(iter.pos.inode, hole_start));
 
 			if (ret)
 				goto bkey_err;
@@ -765,7 +821,7 @@ long bch2_fallocate_dispatch(struct file *file, int mode,
 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
 	long ret;
 
-	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate))
+	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_fallocate))
 		return -EROFS;
 
 	inode_lock(&inode->v);
@@ -789,7 +845,7 @@ long bch2_fallocate_dispatch(struct file *file, int mode,
 err:
 	bch2_pagecache_block_put(inode);
 	inode_unlock(&inode->v);
-	bch2_write_ref_put(c, BCH_WRITE_REF_fallocate);
+	enumerated_ref_put(&c->writes, BCH_WRITE_REF_fallocate);
 
 	return bch2_err_class(ret);
 }
@@ -806,7 +862,7 @@ static int quota_reserve_range(struct bch_inode_info *inode,
 	u64 sectors = end - start;
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter,
+		for_each_btree_key_in_subvolume_max(trans, iter,
 				BTREE_ID_extents,
 				POS(inode->v.i_ino, start),
 				POS(inode->v.i_ino, end - 1),
@@ -877,11 +933,18 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
 	bch2_mark_pagecache_unallocated(src, pos_src >> 9,
 				   (pos_src + aligned_len) >> 9);
 
+	/*
+	 * XXX: we'd like to be telling bch2_remap_range() if we have
+	 * permission to write to the source file, and thus if io path option
+	 * changes should be propagated through the copy, but we need mnt_idmap
+	 * from the pathwalk, awkward
+	 */
 	ret = bch2_remap_range(c,
 			       inode_inum(dst), pos_dst >> 9,
 			       inode_inum(src), pos_src >> 9,
 			       aligned_len >> 9,
-			       pos_dst + len, &i_sectors_delta);
+			       pos_dst + len, &i_sectors_delta,
+			       false);
 	if (ret < 0)
 		goto err;
 
@@ -922,7 +985,7 @@ static loff_t bch2_seek_data(struct file *file, u64 offset)
 		return -ENXIO;
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents,
 				   POS(inode->v.i_ino, offset >> 9),
 				   POS(inode->v.i_ino, U64_MAX),
 				   inum.subvol, 0, k, ({
@@ -958,21 +1021,32 @@ static loff_t bch2_seek_hole(struct file *file, u64 offset)
 		return -ENXIO;
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents,
 				   POS(inode->v.i_ino, offset >> 9),
 				   POS(inode->v.i_ino, U64_MAX),
 				   inum.subvol, BTREE_ITER_slots, k, ({
-			if (k.k->p.inode != inode->v.i_ino) {
-				next_hole = bch2_seek_pagecache_hole(&inode->v,
-						offset, MAX_LFS_FILESIZE, 0, false);
-				break;
-			} else if (!bkey_extent_is_data(k.k)) {
-				next_hole = bch2_seek_pagecache_hole(&inode->v,
-						max(offset, bkey_start_offset(k.k) << 9),
-						k.k->p.offset << 9, 0, false);
-
-				if (next_hole < k.k->p.offset << 9)
+			if (k.k->p.inode != inode->v.i_ino ||
+			    !bkey_extent_is_data(k.k)) {
+				loff_t start_offset = k.k->p.inode == inode->v.i_ino
+					? max(offset, bkey_start_offset(k.k) << 9)
+					: offset;
+				loff_t end_offset = k.k->p.inode == inode->v.i_ino
+					? MAX_LFS_FILESIZE
+					: k.k->p.offset << 9;
+
+				/*
+				 * Found a hole in the btree, now make sure it's
+				 * a hole in the pagecache. We might have to
+				 * keep searching if this hole is entirely dirty
+				 * in the page cache:
+				 */
+				bch2_trans_unlock(trans);
+				loff_t pagecache_hole = bch2_seek_pagecache_hole(&inode->v,
+								start_offset, end_offset, 0, false);
+				if (pagecache_hole < end_offset) {
+					next_hole = pagecache_hole;
 					break;
+				}
 			} else {
 				offset = max(offset, bkey_start_offset(k.k) << 9);
 			}