This add discard support to mkfs.nilfs2 which will be useful to solid
state devices or sparse/thin-provisioned storage.
The updated mkfs.nilfs2 will attempt to discard the device by default
unless -K option is specified.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@xxxxxxxxxxxxx>
---
man/mkfs.nilfs2.8 | 11 ++++++
sbin/mkfs/mkfs.c | 88 ++++++++++++++++++++++++++++++++++++++++++++++++-----
2 files changed, 91 insertions(+), 8 deletions(-)
diff --git a/man/mkfs.nilfs2.8 b/man/mkfs.nilfs2.8
index fb70c87..e9f7463 100644
--- a/man/mkfs.nilfs2.8
+++ b/man/mkfs.nilfs2.8
@@ -18,6 +18,9 @@ mkfs.nilfs2 \- create a NILFS2 filesystem
.B \-c
]
[
+.B \-K
+]
+[
.B \-L
.I volume-label
]
@@ -52,6 +55,9 @@ mkfs.nilfs2 \- create a NILFS2 filesystem
.B \-c
]
[
+.B \-K
+]
+[
.B \-L
.I volume-label
]
@@ -100,6 +106,11 @@ number of blocks per segment is 2048 (= 8MB with 4KB blocks).
.B \-c
Check the device for bad blocks before building the filesystem.
.TP
+.BI \-K
+Keep, do not attempt to discard blocks at mkfs time (discarding blocks
+initially is useful on solid state drives and sparse /
+thinly-provisioned storage).
+.TP
.BI \-L " new-volume-label"
Set the volume label for the filesystem to
.IR new-volume-label\fP.
diff --git a/sbin/mkfs/mkfs.c b/sbin/mkfs/mkfs.c
index 95f4408..092edfd 100644
--- a/sbin/mkfs/mkfs.c
+++ b/sbin/mkfs/mkfs.c
@@ -109,6 +109,7 @@ static int quiet = 0;
static int cflag = 0;
static int nflag = 0;
static int verbose = 0;
+static int discard = 1;
static unsigned long blocksize = NILFS_DEF_BLOCKSIZE;
static unsigned long blocks_per_segment = NILFS_DEF_BLKS_PER_SEG;
static unsigned long r_segments_percentage = NILFS_DEF_RESERVED_SEGMENTS;
@@ -288,6 +289,55 @@ static void cannot_allocate_memory(void);
static void too_small_segment(unsigned long, unsigned long);
/* I/O routines */
+#ifdef __linux__
+
+#ifndef BLKDISCARD
+#define BLKDISCARD _IO(0x12,119)
+#endif
+
+#ifndef BLKDISCARDZEROES
+#define BLKDISCARDZEROES _IO(0x12,124)
+#endif
+
+/**
+ * nilfs_mkfs_discard_range - issue discard command to the device
+ * @fd: file descriptor of the device
+ * @start: start offset of the region to discard (in bytes)
+ * @len: length of the region to discard (in bytes)
+ *
+ * Returns zero if the discard succeeds. Otherwise, -1 is returned.
+ */
+static int nilfs_mkfs_discard_range(int fd, __u64 start, __u64 len)
+{
+ __u64 range[2] = { start, len };
+ int ret;
+
+ ret = ioctl(fd, BLKDISCARD, &range);
+ if (verbose) {
+ pinfo("Discard device from %llu to %llu: %s.",
+ (unsigned long long)start,
+ (unsigned long long)start + len,
+ ret ? "failed" : "succeeded");
+ }
+ return ret;
+}
+
+/**
+ * nilfs_mkfs_discard_zeroes_data - get if discarded blocks are zeroed or not
+ * @fd: file descriptor of the device
+ */
+static int nilfs_mkfs_discard_zeroes_data(int fd)
+{
+ int discard_zeroes_data = 0;
+
+ ioctl(fd, BLKDISCARDZEROES, &discard_zeroes_data);
+ return discard_zeroes_data;
+}
+#else
+#define nilfs_mkfs_discard_range(fd, start, len) 1
+#define nilfs_mkfs_discard_zeroes_data(fd) 0
+#endif
+
static void disk_scan(const char *device);
static void check_mount(int fd, const char *device);
@@ -760,20 +810,39 @@ static int erase_disk_range(int fd, off_t offset, size_t count)
static int erase_disk(int fd, struct nilfs_disk_info *di)
{
+ const unsigned int sector_size = 512;
+ off_t start, end;
int ret;
- BUG_ON(di->dev_size < NILFS_DISK_ERASE_SIZE ||
- di->dev_size - NILFS_DISK_ERASE_SIZE < NILFS_SB_OFFSET_BYTES);
+ /*
+ * Define range of the partition that nilfs uses. This should
+ * not depend on the type of underlying device.
+ */
+ start = NILFS_SB_OFFSET_BYTES;
+ end = di->dev_size & ~((__u64)sector_size - 1);
+
+ BUG_ON(end < NILFS_DISK_ERASE_SIZE ||
+ end - NILFS_DISK_ERASE_SIZE < start);
+
+ if (discard) {
+ ret = nilfs_mkfs_discard_range(fd, start, end - start);
+ if (!ret && nilfs_mkfs_discard_zeroes_data(fd)) {
+ if (verbose)
+ pinfo("Discard succeeded and will return 0s "
+ " - skip wiping");
+ goto out;
+ }
+ }
/* Erase tail of partition */
- ret = erase_disk_range(fd, di->dev_size - NILFS_DISK_ERASE_SIZE,
+ ret = erase_disk_range(fd, end - NILFS_DISK_ERASE_SIZE,
NILFS_DISK_ERASE_SIZE);
if (ret == 0) {
/* Erase head of partition */
- ret = erase_disk_range(fd, NILFS_SB_OFFSET_BYTES,
- NILFS_DISK_ERASE_SIZE -
- NILFS_SB_OFFSET_BYTES);
+ ret = erase_disk_range(fd, start,
+ NILFS_DISK_ERASE_SIZE - start);
}
+out:
return ret;
}
@@ -877,7 +946,7 @@ static void parse_options(int argc, char *argv[])
{
int c, show_version_only = 0;
- while ((c = getopt(argc, argv, "b:B:cL:m:nqvVP:")) != EOF) {
+ while ((c = getopt(argc, argv, "b:B:cKL:m:nqvVP:")) != EOF) {
switch (c) {
case 'b':
blocksize = atol(optarg);
@@ -889,6 +958,9 @@ static void parse_options(int argc, char *argv[])
case 'c':
cflag++;
break;
+ case 'K':
+ discard = 0;
+ break;
case 'L':
strncpy(volume_label, optarg, sizeof(volume_label));
break;
@@ -945,7 +1017,7 @@ static void usage(void)
fprintf(stderr,
"Usage: %s [-b block-size] [-B blocks-per-segment] [-c] \n"
"[-L volume-label] [-m reserved-segments-percentage] \n"
- "[-nqvV] device\n",
+ "[-nqvKV] device\n",
progname);
exit(1);
}
--
1.7.3.2
--
To unsubscribe from this list: send the line "unsubscribe linux-nilfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
