Add Dhara FTL Support to LittleFS by singh-aditya-04 · Pull Request #7184 · Samsung/TizenRT

singh-aditya-04 · 2026-03-09T11:03:21Z

1) Integrates Dhara FTL with LittleFS filesystem
2) Fix autoformat and memory corruption
3) Fix nand erase to continue if bad block
4) Reduces internal fragmentation of Littlefs with dhara
5) Adds metadata cache for improved read speeds

Add support to littlefs VFS layer to choose the driver based on INODE. For MTD driver select the MTD defined functions otherwise use block device.

Register dhara as block driver on top of MTD device. This block device is used to mount littlfs on /mnt.

Dhara documentation shows that in case of successful read 0 should be returned. But current TizeRT code returns numnber of read.

These changes modifies the format_filesystem function to add support for formatting LittleFS over Dhara (MTD-based filesystem) in addition to the existing block device formatting capabilities.

autoformat not working in case mount fails. memory corruption due to wrong assignment for dhara. causing corruption of dhara structure.

in case nand erase encounter bad block, return eagain and continue erasing other blocks. Also fix return for nand_bread

Two changes are done for this to sync the behaviour between Dhara and littleFS 1) We change littlefs_blocksize from fs->geo.erasesize (128 KB) to fs->geo.blocksize(2 KB). 2) As we are reserving some sectors (pages) for dhara inside files os/fs/driver/mtd/dhara/journal.c and os/fs/driver/mtd/dhara/map.c so we are going to give total pages - reserved pages as available blocks to Littlefs. Signed-off-by: Aditya Singh <aditya.s4@samsung.com> Signed-off-by: Aashish Lakhwara <aashish.l@samsung.com>

We have varified dhara itself manages bad block checking so we do not need mtd to manage bad block checking as it increases flash read operations. Signed-off-by: Aditya Singh <aditya.s4@samsung.com> Signed-off-by: Aashish Lakhwara <aashish.l@samsung.com>

Taejun-Kwon · 2026-03-10T09:06:03Z

instead of extern, you should define constant of ioctl, and use driver->u.i_bops->ioctl()

Taejun-Kwon · 2026-03-10T09:24:59Z

This need to be checked.
fs->geo.erasesize * fs->geo.neraseblocks is same as size of entire of flash (allocated for littlefs) ?

Taejun-Kwon · 2026-03-10T09:26:08Z

prev is correct.

Taejun-Kwon · 2026-03-10T09:28:02Z

So erase size of MTD driver will be asigned.
But ftl will use blocksize of erase (Not erase internally). I mean lfs doesn't care flash is nor or nand. so if nand is attached, then blocksize will be used as a erasesize.
If I am right then geometry of ftl (include dhara) need to be updated.

Taejun-Kwon · 2026-03-10T09:32:06Z

if my comment is correct (https://github.com/Samsung/TizenRT/pull/7184/changes#r2910443698)
then this can be removed.

Taejun-Kwon · 2026-03-10T09:33:47Z

please remove this line

updated the PR please check

Taejun-Kwon · 2026-03-10T09:34:42Z

please put tab to each of code line.

updated Please check

Taejun-Kwon · 2026-03-10T09:37:53Z

what '64' means exactly? can we change constant value to some kind of formula?
(dev->geo->neraseblocks / dev->geo->blocksize I expect)

64 represents pages in one bock , sure i will change constant value to some kind of formula

Updated Please Check

Taejun-Kwon · 2026-03-10T09:39:03Z

if (page % 16 == 15)
But let's do not use '16' directly.

Updated please check

Taejun-Kwon · 2026-03-10T09:49:54Z

You reused this code from origin code of dhara. and I think this is implemented wrongly.
Because MTD_BREAD can be failed because of other issue, for example SPI.(of course w25n returns ERROR when status is ECC Fail. so I think we should put some TODO Comment here. I think kind of below logic is required

if (ret < 0) { dhara_insert_metadatacache(dev, mcache); if (check_ecc_check_logic) { dhara_set_error(err, DHARA_E_ECC); } return ret; }

Taejun-Kwon · 2026-03-12T08:35:31Z

coding rule,

if (((p + 1) % page_per_chunk) == 0) {

Taejun-Kwon · 2026-04-15T09:29:38Z

can you please add bracket ? according to our coding rule, bracket must be added for 1 line of code.

Taejun-Kwon · 2026-04-16T03:05:54Z

 CONFIG_MTD_NAND_MAXPAGESPARESIZE=128
 CONFIG_MTD_NAND_MAXSPAREECCBYTES=64
-CONFIG_MTD_NAND_BLOCKCHECK=y
+#CONFIG_MTD_NAND_BLOCKCHECK=y


enable CONFIG_RESOURCE_FS
2 disable CONFIG_AUTOMOUNT_ROMFS

Increase CONFIG_USERMAIN_STACKSIZE, but need to be tested.

Also some values are changed in flat_dev_ddr, can you please update entire of defconfig? we can change only fs, mtd, ftl and above configs I mentioned from flat_dev_ddr

Taejun-Kwon · 2026-04-16T04:46:13Z

prev code is proper.
Above layers expect number of written page

Taejun-Kwon · 2026-04-16T09:27:56Z

even if mount failed, bulkerase should not be happened. because of sudden power off, this can be reached.

A Metadata Cache based on Most Recently Used Metadata Page has been implemented inside file dhara.c that can be useful for faster page reads. Signed-off-by: Aditya Singh <aditya.s4@samsung.com> Signed-off-by: Aashish Lakhwara <aashish.l@samsung.com>w

Taejun-Kwon · 2026-04-17T08:37:24Z

 			if (fd < 0) {
-				continue;
+				/* Then Find Littlefs over Dhara */
+				snprintf(name, sizeof(name), "/dev/mtdblock%d", minor);


hmm I think this is very dangerous code, other partition which use bch can be erased.
I think we should allow access dhara_initialize_by_path from partitions.c and use '/dev/little%dp%d'

Taejun-Kwon · 2026-04-17T08:40:47Z

 		}
 		/* TODO Multi root of smartfs should be considered when it enabled */
-		ret = ioctl(fd, BIOC_BULKERASE, 0);
+		if (ext_dhara)


so lfs + little.c / lfs + dhara will work by same ioctl, BIOC_BULKERASE

Taejun-Kwon · 2026-04-17T08:45:36Z

I think someday we can use nand for internal flash so minor will be decided by defconfig.
hence below is enough for now.

#ifdef CONFIG_MTD_DHARA ... int ret = dhara_initialize(dhara_minor, mtd_part); ... #else little_initialize(minor, mtd_part, partref); #endif partinfo->littlefs_partno = g_partno;

Taejun-Kwon · 2026-04-17T08:45:55Z

please refer to my comment in fscmd.

Taejun-Kwon · 2026-04-17T08:47:21Z

instead of extern, you can add this to nand.h

Taejun-Kwon · 2026-04-17T10:03:33Z

so, if below code should use this function,
https://github.com/Samsung/TizenRT/pull/7184/changes#diff-2e8e8a310aaa1d9332c8ce08cf1a2b6af261d63277ce4d18579eff09097dda1bR981
then we can consider below.

#journal.c dhara_block_t dhara_journal_get_block_capacity(const struct dhara_journal *j) { const dhara_block_t max_bad = j->bb_last > j->bb_current ? j->bb_last : j->bb_current; const dhara_block_t good_blocks = j->nand->num_blocks - max_bad - 1; return good_blocks; } #map.c dhara_sector_t dhara_map_get_good_block(const struct dhara_map *m) { return dhara_journal_get_block_capacity(&m->journal); } #dhara.c static int dhara_ioctl(FAR struct inode *inode, int cmd, unsigned long arg) { FAR dhara_dev_t *dev; int ret; DEBUGASSERT(inode->i_private); dev = inode->i_private; /* No other block driver ioctl commands are not recognized by this * driver. Other possible MTD driver ioctl commands are passed through * to the MTD driver (unchanged). */ ret = MTD_IOCTL(dev->mtd, cmd, arg); if (ret < 0 && ret != -ENOTTY) { ferr("MTD ioctl(%04x) failed: %d\n", cmd, ret); return ret; } if (cmd == MTDIOC_GEOMETRY) { FAR struct mtd_geometry_s *geo = (FAR struct mtd_geometry_s *)arg; if (geo) { /* Populate the geometry structure with information needed to know * the capacity and how to access the device. */ geo->neraseblocks = dhara_map_get_good_block(dev->map); } } return ret; #lfs_Vfs.c static int littlefs_bind(FAR struct inode *driver, FAR const void *data, FAR void **handle) { ... fs->cfg.read_size = fs->geo.blocksize; fs->cfg.prog_size = fs->geo.blocksize; fs->cfg.block_size = fs->geo.blocksize; fs->cfg.block_count = fs->geo.neraseblocks * fs->geo.erasesize / fs->geo.blocksize; }

But is lfs should know number of capacity???

Taejun-Kwon · 2026-04-17T10:13:31Z

I think this function is not necessary.

Journal resume : It can call MTD_IOCTL directly, because struct dhara_journal include struct dhara_nand, and dhara_nand include dhara_dev_t.

dhara_ioctl : refer to comment in https://github.com/Samsung/TizenRT/pull/7184/changes#r3099185257

Can't understand why first format you considering. can you please explain it?

Taejun-Kwon · 2026-04-17T10:13:53Z

please check tab

Taejun-Kwon · 2026-04-17T10:15:49Z

I think lfs_format is enough. because we guarantee first mount during factory phase.

Taejun-Kwon · 2026-04-17T10:21:03Z

can you please check 'fs->cfg.sync = littlefs_sync_block;' can be used instead of this?
I think below can be used.

int lfs_check_format(lfs_t *lfs, const struct lfs_config *cfg) { .... cleanup: if (err == LFS_ERR_OK) { cfg->sync(cfg); } lfs_deinit(lfs); LFS_UNLOCK(cfg); return err; }

of course BIOC_FLUSH need to be added to ioctl.h

singh-aditya-04 added 8 commits January 29, 2026 14:25

fs-littlefs-apply-block-MTD-driver-logic-in-littlefs.patch

250e082

Add support to littlefs VFS layer to choose the driver based on INODE. For MTD driver select the MTD defined functions otherwise use block device.

board/common: add support for dhara device register

a513f2d

Register dhara as block driver on top of MTD device. This block device is used to mount littlfs on /mnt.

driver/mtd: fix return type of dhara_nand_read

08e4ee8

Dhara documentation shows that in case of successful read 0 should be returned. But current TizeRT code returns numnber of read.

system/utils: update format command to work with Dhara

9b8ec36

These changes modifies the format_filesystem function to add support for formatting LittleFS over Dhara (MTD-based filesystem) in addition to the existing block device formatting capabilities.

fs/littlefs: Fix autoformat and memory corruption

d66157d

autoformat not working in case mount fails. memory corruption due to wrong assignment for dhara. causing corruption of dhara structure.

driver/mtd: Fix nand erase to continue if bad block

361c20f

in case nand erase encounter bad block, return eagain and continue erasing other blocks. Also fix return for nand_bread

Taejun-Kwon reviewed Mar 10, 2026

View reviewed changes

singh-aditya-04 force-pushed the littlefs_with_dhara branch from 030fe79 to c817e72 Compare March 12, 2026 06:29

Taejun-Kwon reviewed Mar 12, 2026

View reviewed changes

Comment thread os/fs/driver/mtd/dhara.c Outdated

Copy link
Copy Markdown

Contributor

Taejun-Kwon Mar 12, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

coding rule,

if (((p + 1) % page_per_chunk) == 0) {

Taejun-Kwon reviewed Apr 16, 2026

View reviewed changes

singh-aditya-04 force-pushed the littlefs_with_dhara branch from c817e72 to 5c61406 Compare April 16, 2026 11:36

Taejun-Kwon reviewed Apr 17, 2026

View reviewed changes

Conversation

singh-aditya-04 commented Mar 9, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants