ext4: make trim/discard optional (and off by default)

It is anticipated that when sb_issue_discard starts doing real work on trim-capable devices, we may see issues. Make this mount-time optional, and default it to off until we know that things are working out OK.

(Q: should we call this "discard" instead? What's the more common term users might expect ... ?)

---

index 6d94e06..87036af 100644
@@ -353,6 +353,12 @@ noauto_da_alloc		replacing existing files via patterns such as

system crashes before the delayed allocation blocks are forced to disk.

+trim Controls whether ext4 should issue TRIM/discard
+notrim(*) commands to the underlying block device when
+ blocks are freed. This is useful for SSD devices
+ and sparse/thinly-provisioned LUNs, but it is off
+ by default until sufficient testing has been done.
+
Data Mode
=========
There are 3 different data modes:

index 8825515..410adb6 100644
@@ -750,6 +750,7 @@ struct ext4_inode_info {

#define EXT4_MOUNT_DELALLOC 0x8000000 /* Delalloc support */
#define EXT4_MOUNT_DATA_ERR_ABORT 0x10000000 /* Abort on file data write */
#define EXT4_MOUNT_BLOCK_VALIDITY 0x20000000 /* Block validity checking */
+#define EXT4_MOUNT_TRIM 0x40000000 /* Issue TRIM requests */

#define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt
#define set_opt(o, opt) o |= EXT4_MOUNT_##opt

index bba1282..8a4f77b 100644
@@ -2529,7 +2529,6 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)

	struct ext4_group_info *db;
	int err, count = 0, count2 = 0;
	struct ext4_free_data *entry;
-	ext4_fsblk_t discard_block;
	struct list_head *l, *ltmp;

	list_for_each_safe(l, ltmp, &txn->t_private_list) {
@@ -2559,13 +2558,19 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
			page_cache_release(e4b.bd_bitmap_page);
		}
		ext4_unlock_group(sb, entry->group);

- discard_block = (ext4_fsblk_t) entry->group * EXT4_BLOCKS_PER_GROUP(sb)
- + entry->start_blk
- + le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
- trace_ext4_discard_blocks(sb, (unsigned long long)discard_block,
- entry->count);
- sb_issue_discard(sb, discard_block, entry->count);
-
+ if (test_opt(sb, TRIM)) {
+ ext4_fsblk_t discard_block;
+ struct ext4_super_block *es = EXT4_SB(sb)->s_es;
+
+ discard_block = (ext4_fsblk_t)entry->group *
+ EXT4_BLOCKS_PER_GROUP(sb)
+ + entry->start_blk
+ + le32_to_cpu(es->s_first_data_block);
+ trace_ext4_discard_blocks(sb,
+ (unsigned long long)discard_block,
+ entry->count);
+ sb_issue_discard(sb, discard_block, entry->count);
+ }

		kmem_cache_free(ext4_free_ext_cachep, entry);
		ext4_mb_release_desc(&e4b);
	}

index d4ca92a..fc4a8d8 100644
@@ -899,6 +899,9 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)

	if (test_opt(sb, NO_AUTO_DA_ALLOC))
		seq_puts(seq, ",noauto_da_alloc");

+ if (test_opt(sb, TRIM))
+ seq_puts(seq, ",trim");
+

	ext4_show_quota_options(seq, sb);

	return 0;

@@ -1079,7 +1082,8 @@ enum {
Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_block_validity, Opt_noblock_validity,
- Opt_inode_readahead_blks, Opt_journal_ioprio
+ Opt_inode_readahead_blks, Opt_journal_ioprio,
+ Opt_trim, Opt_notrim,
};

static const match_table_t tokens = {

@@ -1144,6 +1148,8 @@ static const match_table_t tokens = {
	{Opt_auto_da_alloc, "auto_da_alloc=%u"},
	{Opt_auto_da_alloc, "auto_da_alloc"},
	{Opt_noauto_da_alloc, "noauto_da_alloc"},
+	{Opt_trim, "trim"},
+	{Opt_notrim, "notrim"},
	{Opt_err, NULL},
};

@@ -1565,6 +1571,12 @@ set_qf_format:

			else
				set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
			break;

+ case Opt_trim:
+ set_opt(sbi->s_mount_opt, TRIM);
+ break;
+ case Opt_notrim:
+ clear_opt(sbi->s_mount_opt, TRIM);

+			break;
		default:
			ext4_msg(sb, KERN_ERR,
			       "Unrecognized mount option \"%s\" "

Users will be confused regardless of what we do here, but the actual discard only invokes ATA_TRIM commands on ATA devices. (SCSI uses its own command, either a WRITE_SAME with discard or UNMAP).

Not sure that any real user cares since they both end up doing roughly the same thing...

ric

It is anticipated that when sb_issue_discard starts doing real work on trim-capable devices, we may see issues. Make this mount-time optional, and default it to off until we know that things are working out OK.

V2: rename "trim" to "discard" to match btrfs & gfs2

---

index 6d94e06..26904ff 100644
@@ -353,6 +353,12 @@ noauto_da_alloc		replacing existing files via patterns such as

system crashes before the delayed allocation blocks are forced to disk.

+discard Controls whether ext4 should issue discard/TRIM
+nodiscard(*) commands to the underlying block device when
+ blocks are freed. This is useful for SSD devices
+ and sparse/thinly-provisioned LUNs, but it is off
+ by default until sufficient testing has been done.
+
Data Mode
=========
There are 3 different data modes:

index 8825515..05ce38b 100644
@@ -750,6 +750,7 @@ struct ext4_inode_info {

#define EXT4_MOUNT_DELALLOC 0x8000000 /* Delalloc support */
#define EXT4_MOUNT_DATA_ERR_ABORT 0x10000000 /* Abort on file data write */
#define EXT4_MOUNT_BLOCK_VALIDITY 0x20000000 /* Block validity checking */
+#define EXT4_MOUNT_DISCARD 0x40000000 /* Issue DISCARD requests */

#define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt
#define set_opt(o, opt) o |= EXT4_MOUNT_##opt

index bba1282..6e5a23a 100644
@@ -2529,7 +2529,6 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)

	struct ext4_group_info *db;
	int err, count = 0, count2 = 0;
	struct ext4_free_data *entry;
-	ext4_fsblk_t discard_block;
	struct list_head *l, *ltmp;

	list_for_each_safe(l, ltmp, &txn->t_private_list) {
@@ -2559,13 +2558,19 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
			page_cache_release(e4b.bd_bitmap_page);
		}
		ext4_unlock_group(sb, entry->group);

- discard_block = (ext4_fsblk_t) entry->group * EXT4_BLOCKS_PER_GROUP(sb)
- + entry->start_blk
- + le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
- trace_ext4_discard_blocks(sb, (unsigned long long)discard_block,
- entry->count);
- sb_issue_discard(sb, discard_block, entry->count);
-
+ if (test_opt(sb, DISCARD)) {
+ ext4_fsblk_t discard_block;
+ struct ext4_super_block *es = EXT4_SB(sb)->s_es;
+
+ discard_block = (ext4_fsblk_t)entry->group *
+ EXT4_BLOCKS_PER_GROUP(sb)
+ + entry->start_blk
+ + le32_to_cpu(es->s_first_data_block);
+ trace_ext4_discard_blocks(sb,
+ (unsigned long long)discard_block,
+ entry->count);
+ sb_issue_discard(sb, discard_block, entry->count);
+ }

		kmem_cache_free(ext4_free_ext_cachep, entry);
		ext4_mb_release_desc(&e4b);
	}

index d4ca92a..b9638d9 100644
@@ -899,6 +899,9 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)

	if (test_opt(sb, NO_AUTO_DA_ALLOC))
		seq_puts(seq, ",noauto_da_alloc");

+ if (test_opt(sb, DISCARD))
+ seq_puts(seq, ",discard");
+

	ext4_show_quota_options(seq, sb);

	return 0;

@@ -1079,7 +1082,8 @@ enum {
Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_block_validity, Opt_noblock_validity,
- Opt_inode_readahead_blks, Opt_journal_ioprio
+ Opt_inode_readahead_blks, Opt_journal_ioprio,
+ Opt_discard, Opt_nodiscard,
};

static const match_table_t tokens = {

@@ -1144,6 +1148,8 @@ static const match_table_t tokens = {
	{Opt_auto_da_alloc, "auto_da_alloc=%u"},
	{Opt_auto_da_alloc, "auto_da_alloc"},
	{Opt_noauto_da_alloc, "noauto_da_alloc"},
+	{Opt_discard, "discard"},
+	{Opt_nodiscard, "nodiscard"},
	{Opt_err, NULL},
};

@@ -1565,6 +1571,12 @@ set_qf_format:

			else
				set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
			break;

+ case Opt_discard:
+ set_opt(sbi->s_mount_opt, DISCARD);
+ break;
+ case Opt_nodiscard:
+ clear_opt(sbi->s_mount_opt, DISCARD);

+			break;
		default:
			ext4_msg(sb, KERN_ERR,
			       "Unrecognized mount option \"%s\" "

It is anticipated that when sb_issue_discard starts doing real work on trim-capable devices, we may see issues. Make this mount-time optional, and default it to off until we know that things are working out OK.

V2: rename "trim" to "discard" to match btrfs & gfs2
V3: fix mailer flowed text mangling, sorry.

---

index 6d94e06..26904ff 100644
@@ -353,6 +353,12 @@ noauto_da_alloc		replacing existing files via patterns such as
 			system crashes before the delayed allocation
 			blocks are forced to disk.
 
+discard		Controls whether ext4 should issue discard/TRIM
+nodiscard(*)		commands to the underlying block device when
+			blocks are freed.  This is useful for SSD devices
+			and sparse/thinly-provisioned LUNs, but it is off
+			by default until sufficient testing has been done.
+
 Data Mode
 =========
 There are 3 different data modes:

index 8825515..05ce38b 100644
@@ -750,6 +750,7 @@ struct ext4_inode_info {
 #define EXT4_MOUNT_DELALLOC		0x8000000 /* Delalloc support */
 #define EXT4_MOUNT_DATA_ERR_ABORT	0x10000000 /* Abort on file data write */
 #define EXT4_MOUNT_BLOCK_VALIDITY	0x20000000 /* Block validity checking */
+#define EXT4_MOUNT_DISCARD		0x40000000 /* Issue DISCARD requests */
 
 #define clear_opt(o, opt)		o &= ~EXT4_MOUNT_##opt
 #define set_opt(o, opt)			o |= EXT4_MOUNT_##opt

index bba1282..6e5a23a 100644
@@ -2529,7 +2529,6 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
 	struct ext4_group_info *db;
 	int err, count = 0, count2 = 0;
 	struct ext4_free_data *entry;
-	ext4_fsblk_t discard_block;
 	struct list_head *l, *ltmp;
 
 	list_for_each_safe(l, ltmp, &txn->t_private_list) {

@@ -2559,13 +2558,19 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
 			page_cache_release(e4b.bd_bitmap_page);
 		}
 		ext4_unlock_group(sb, entry->group);
-		discard_block = (ext4_fsblk_t) entry->group * EXT4_BLOCKS_PER_GROUP(sb)
-			+ entry->start_blk
-			+ le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
-		trace_ext4_discard_blocks(sb, (unsigned long long)discard_block,
-					  entry->count);
-		sb_issue_discard(sb, discard_block, entry->count);
-
+		if (test_opt(sb, DISCARD)) {
+			ext4_fsblk_t discard_block;
+			struct ext4_super_block *es = EXT4_SB(sb)->s_es;
+
+			discard_block = (ext4_fsblk_t)entry->group *
+						EXT4_BLOCKS_PER_GROUP(sb)
+					+ entry->start_blk
+					+ le32_to_cpu(es->s_first_data_block);
+			trace_ext4_discard_blocks(sb,
+					(unsigned long long)discard_block,
+					entry->count);
+			sb_issue_discard(sb, discard_block, entry->count);
+		}
 		kmem_cache_free(ext4_free_ext_cachep, entry);
 		ext4_mb_release_desc(&e4b);
 	}

index d4ca92a..b9638d9 100644
@@ -899,6 +899,9 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 	if (test_opt(sb, NO_AUTO_DA_ALLOC))
 		seq_puts(seq, ",noauto_da_alloc");
 
+	if (test_opt(sb, DISCARD))
+		seq_puts(seq, ",discard");
+
 	ext4_show_quota_options(seq, sb);
 
 	return 0;

@@ -1079,7 +1082,8 @@ enum {
 	Opt_usrquota, Opt_grpquota, Opt_i_version,
 	Opt_stripe, Opt_delalloc, Opt_nodelalloc,
 	Opt_block_validity, Opt_noblock_validity,
-	Opt_inode_readahead_blks, Opt_journal_ioprio
+	Opt_inode_readahead_blks, Opt_journal_ioprio,
+	Opt_discard, Opt_nodiscard,
 };
 
 static const match_table_t tokens = {

@@ -1144,6 +1148,8 @@ static const match_table_t tokens = {
 	{Opt_auto_da_alloc, "auto_da_alloc=%u"},
 	{Opt_auto_da_alloc, "auto_da_alloc"},
 	{Opt_noauto_da_alloc, "noauto_da_alloc"},
+	{Opt_discard, "discard"},
+	{Opt_nodiscard, "nodiscard"},
 	{Opt_err, NULL},
 };
 

@@ -1565,6 +1571,12 @@ set_qf_format:
 			else
 				set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
 			break;
+		case Opt_discard:
+			set_opt(sbi->s_mount_opt, DISCARD);
+			break;
+		case Opt_nodiscard:
+			clear_opt(sbi->s_mount_opt, DISCARD);
+			break;
 		default:
 			ext4_msg(sb, KERN_ERR,
 			       "Unrecognized mount option \"%s\" "

Queued, thanks.

- Ted