github-mirrors/thirdparty-littlefs
1
0
Fork 0
mirror of https://github.com/eledio-devices/thirdparty-littlefs.git synced 2025-10-31 08:42:40 +01:00
Code Issues Packages Projects Releases Wiki Activity

Added building blocks for dynamic wear-leveling

Browse Source 
Initially, littlefs relied entirely on bad-block detection for
wear-leveling. Conceptually, at the end of a devices lifespan, all
blocks would be worn evenly, even if they weren't worn out at the same
time. However, this doesn't work for all devices, rather than causing
corruption during writes, wear reduces a devices "sticking power",
causing bits to flip over time. This means for many devices, true
wear-leveling (dynamic or static) is required.

Fortunately, way back at the beginning, littlefs was designed to do full
dynamic wear-leveling, only dropping it when making the retrospectively
short-sighted realization that bad-block detection is theoretically
sufficient. We can enable dynamic wear-leveling with only a few tweaks
to littlefs. These can be implemented without breaking backwards
compatibility.

1. Evict metadata-pairs after a certain number of writes. Eviction in
   this case is identical to a relocation to recover from a bad block.
   We move our data and stick the old block back into our pool of
   blocks.

   For knowing when to evict, we already have a revision count for each
   metadata-pair which gives us enough information. We add the
   configuration option block_cycles and evict when our revision count
   is a multiple of this value.

2. Now all blocks participate in COW behaviour. However we don't store
   the state of our allocator, so every boot cycle we reuse the first
   blocks on storage. This is very bad on a microcontroller, where we
   may reboot often. We need a way to spread our usage across the disk.

   To pull this off, we can simply randomize which block we start our
   allocator at. But we need a random number generator that is different
   on each boot. Fortunately we have a great source of entropy, our
   filesystem. So we seed our block allocator with a simple hash of the
   CRCs on our metadata-pairs. This can be done for free since we
   already need to scan the metadata-pairs during mount.

What we end up with is a uniform distribution of wear on storage. The
wear is not perfect, if a block is used for metadata it gets more wear,
and the randomization may not be exact. But we can never actually get
perfect wear-leveling, since we're already resigned to dynamic
wear-leveling at the file level.

With the addition of metadata logging, we end up with a really
interesting two-stage wear-leveling algorithm. At the low-level,
metadata is statically wear-leveled. At the high-level, blocks are
dynamically wear-leveled.

---

This specific commit implements the first step, eviction of metadata
pairs. Entertwining this into the already complicated compact logic was
a bit annoying, however we can combine the logic for superblock
expansion with the logic for metadata-pair eviction.
This commit is contained in:
Christopher Haster
2018-08-08 16:34:56 -05:00
parent 20b669a23d
commit e4a0d586d5
4 changed files with 356 additions and 306 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.travis.yml
View File

@@ -19,7 +19,7 @@ script:
# run tests with a few different configurations
- make test QUIET=1 CFLAGS+="-DLFS_READ_SIZE=1 -DLFS_CACHE_SIZE=4"
- make test QUIET=1 CFLAGS+="-DLFS_READ_SIZE=512 -DLFS_CACHE_SIZE=512"
- make test QUIET=1 CFLAGS+="-DLFS_READ_SIZE=512 -DLFS_CACHE_SIZE=512 -DLFS_BLOCK_CYCLES=16"
- make test QUIET=1 CFLAGS+="-DLFS_BLOCK_COUNT=1023 -DLFS_LOOKAHEAD=2048"
- make clean test QUIET=1 CFLAGS+="-DLFS_INLINE_MAX=0"

128
lfs.c
View File

@@ -265,6 +265,8 @@ static int32_t lfs_fs_parent(lfs_t *lfs, const lfs_block_t dir[2],
lfs_mdir_t *parent);
static int lfs_fs_relocate(lfs_t *lfs,
const lfs_block_t oldpair[2], lfs_block_t newpair[2]);
static int lfs_fs_deorphan(lfs_t *lfs);
static int lfs_fs_demove(lfs_t *lfs);
static int lfs_fs_forceconsistency(lfs_t *lfs);
static int lfs_deinit(lfs_t *lfs);
@@ -924,7 +926,27 @@ static int32_t lfs_dir_fetchmatch(lfs_t *lfs,
tempfoundtag -= LFS_MKTAG(0, 1, 0);
}
} else if ((tag & findmask) == (findtag & findmask)) {
int res = lfs_bd_cmp(lfs, dir->pair[0], off+sizeof(tag),
// found a match?
if (lfs_tag_type(findtag) == LFS_TYPE_DIRSTRUCT) {
lfs_block_t child[2];
err = lfs_bd_read(lfs, dir->pair[0], off+sizeof(tag),
child, sizeof(child));
if (err < 0) {
if (err == LFS_ERR_CORRUPT) {
dir->erased = false;
break;
}
return err;
}
lfs_pair_fromle32(child);
if (lfs_pair_cmp(child,
(const lfs_block_t *)findbuffer) == 0) {
tempfoundtag = tag;
}
} else if (lfs_tag_type(findtag) == LFS_TYPE_NAME) {
int res = lfs_bd_cmp(lfs,
dir->pair[0], off+sizeof(tag),
findbuffer, lfs_tag_size(findtag));
if (res < 0) {
if (res == LFS_ERR_CORRUPT) {
@@ -935,7 +957,9 @@ static int32_t lfs_dir_fetchmatch(lfs_t *lfs,
}
if (res) {
// found a match
tempfoundtag = tag;
}
} else {
tempfoundtag = tag;
}
}
@@ -1016,8 +1040,6 @@ static int lfs_dir_compact(lfs_t *lfs,
lfs_mdir_t *source, uint16_t begin, uint16_t end) {
// save some state in case block is bad
const lfs_block_t oldpair[2] = {dir->pair[1], dir->pair[0]};
int16_t ack;
bool expanding = false;
bool relocated = false;
// There's nothing special about our global delta, so feed it back
@@ -1025,11 +1047,16 @@ static int lfs_dir_compact(lfs_t *lfs,
lfs_global_xor(&lfs->locals, &dir->locals);
lfs_global_zero(&dir->locals);
while (true) {
// last complete id
dir->count = end - begin;
int16_t ack = -1;
bool exhausted = false;
// increment revision count
dir->rev += 1;
if (lfs_pair_cmp(dir->pair, (const lfs_block_t[2]){0, 1}) == 0 &&
dir->rev % 16 == 0) {
if (lfs->cfg->block_cycles && dir->rev % lfs->cfg->block_cycles == 0) {
if (lfs_pair_cmp(dir->pair, (const lfs_block_t[2]){0, 1}) == 0) {
// we're writing too much to the superblock, should we expand?
lfs_ssize_t res = lfs_fs_size(lfs);
if (res < 0) {
@@ -1038,18 +1065,18 @@ static int lfs_dir_compact(lfs_t *lfs,
// do we have enough space to expand?
if (res < lfs->cfg->block_count/2) {
expanding = (lfs_pair_cmp(dir->pair, lfs->root) != 0);
LFS_DEBUG("Expanding superblock at rev %"PRIu32, dir->rev);
ack = 0;
exhausted = (lfs_pair_cmp(dir->pair, lfs->root) != 0);
goto split;
}
} else {
// we're writing too much, time to relocate
exhausted = true;
goto relocate;
}
}
while (true) {
// last complete id
ack = -1;
dir->count = end - begin;
if (true) {
// erase block to write to
int err = lfs_bd_erase(lfs, dir->pair[1]);
if (err) {
@@ -1143,7 +1170,6 @@ static int lfs_dir_compact(lfs_t *lfs,
dir->off = commit.off;
dir->etag = commit.ptag;
dir->erased = true;
}
break;
split:
@@ -1157,7 +1183,7 @@ split:
}
lfs_mdir_t tail;
int err = lfs_dir_alloc(lfs, &tail);
err = lfs_dir_alloc(lfs, &tail);
if (err) {
return err;
}
@@ -1166,7 +1192,7 @@ split:
tail.tail[0] = dir->tail[0];
tail.tail[1] = dir->tail[1];
err = lfs_dir_compact(lfs, &tail, attrs, source, ack+1-expanding, end);
err = lfs_dir_compact(lfs, &tail, attrs, source, ack+1-exhausted, end);
if (err) {
return err;
}
@@ -1178,13 +1204,12 @@ split:
continue;
relocate:
//commit was corrupted
LFS_DEBUG("Bad block at %"PRIu32,
dir->pair[1]);
// drop caches and prepare to relocate block
// commit was corrupted, drop caches and prepare to relocate block
relocated = true;
lfs_cache_drop(lfs, &lfs->pcache);
if (!exhausted) {
LFS_DEBUG("Bad block at %"PRIu32, dir->pair[1]);
}
// can't relocate superblock, filesystem is now frozen
if (lfs_pair_cmp(oldpair, (const lfs_block_t[2]){0, 1}) == 0) {
@@ -1194,7 +1219,7 @@ relocate:
// relocate half of pair
err = lfs_alloc(lfs, &dir->pair[1]);
if (err) {
if (err && (err != LFS_ERR_NOSPC && !exhausted)) {
return err;
}
@@ -1273,7 +1298,11 @@ static int lfs_dir_commit(lfs_t *lfs, lfs_mdir_t *dir,
}
}
if (dir->erased) {
while (true) {
if (!dir->erased) {
goto compact;
}
// try to commit
struct lfs_commit commit = {
.block = dir->pair[0],
@@ -1332,14 +1361,17 @@ static int lfs_dir_commit(lfs_t *lfs, lfs_mdir_t *dir,
// successful commit, update globals
lfs_global_xor(&dir->locals, &lfs->locals);
lfs_global_zero(&lfs->locals);
} else {
break;
compact:
// fall back to compaction
lfs_cache_drop(lfs, &lfs->pcache);
int err = lfs_dir_compact(lfs, dir, attrs, dir, 0, dir->count);
err = lfs_dir_compact(lfs, dir, attrs, dir, 0, dir->count);
if (err) {
return err;
}
break;
}
// update globals that are affected
@@ -1762,7 +1794,6 @@ static int lfs_ctz_extend(lfs_t *lfs,
}
LFS_ASSERT(nblock >= 2 && nblock <= lfs->cfg->block_count);
if (true) {
err = lfs_bd_erase(lfs, nblock);
if (err) {
if (err == LFS_ERR_CORRUPT) {
@@ -1837,7 +1868,6 @@ static int lfs_ctz_extend(lfs_t *lfs,
*block = nblock;
*off = 4*skips;
return 0;
}
relocate:
LFS_DEBUG("Bad block at %"PRIu32, nblock);
@@ -2205,7 +2235,7 @@ int lfs_file_sync(lfs_t *lfs, lfs_file_t *file) {
type = LFS_TYPE_INLINESTRUCT;
buffer = file->cache.buffer;
size = file->ctz.size;
} else if (lfs_tole32(0x11223344) == 0x11223344) {
} else {
// update the ctz reference
type = LFS_TYPE_CTZSTRUCT;
// copy ctz so alloc will work during a relocate
@@ -3148,17 +3178,13 @@ static int32_t lfs_fs_parent(lfs_t *lfs, const lfs_block_t pair[2],
}
// search for both orderings so we can reuse the find function
lfs_block_t child[2] = {pair[0], pair[1]};
lfs_pair_tole32(child);
for (int i = 0; i < 2; i++) {
int32_t tag = lfs_dir_find(lfs, parent,
(const lfs_block_t[2]){0, 1}, true, 0x7fc00fff,
LFS_MKTAG(LFS_TYPE_DIRSTRUCT, 0, sizeof(child)), child);
LFS_MKTAG(LFS_TYPE_DIRSTRUCT, 0, 8), pair);
if (tag != LFS_ERR_NOENT) {
return tag;
}
lfs_pair_swap(child);
}
return LFS_ERR_NOENT;
@@ -3201,7 +3227,7 @@ static int lfs_fs_relocate(lfs_t *lfs,
}
// clean up bad block, which should now be a desync
return lfs_fs_forceconsistency(lfs);
return lfs_fs_deorphan(lfs);
}
// find pred
@@ -3227,11 +3253,7 @@ static int lfs_fs_relocate(lfs_t *lfs,
return 0;
}
static int lfs_fs_forceconsistency(lfs_t *lfs) {
if (!lfs->globals.s.deorphaned) {
LFS_DEBUG("Found orphans %"PRIu32,
lfs->globals.s.deorphaned);
static int lfs_fs_deorphan(lfs_t *lfs) {
// Fix any orphans
lfs_mdir_t pdir = {.split = true};
lfs_mdir_t dir = {.tail = {0, 1}};
@@ -3299,11 +3321,12 @@ static int lfs_fs_forceconsistency(lfs_t *lfs) {
memcpy(&pdir, &dir, sizeof(pdir));
}
// mark orphan as fixed
lfs_global_deorphaned(lfs, false);
}
// mark orphans as fixed
lfs_global_deorphaned(lfs, true);
return 0;
}
if (lfs->globals.s.moveid != 0x3ff) {
static int lfs_fs_demove(lfs_t *lfs) {
// Fix bad moves
LFS_DEBUG("Fixing move %"PRIu32" %"PRIu32" %"PRIu32,
lfs->globals.s.movepair[0],
@@ -3322,6 +3345,23 @@ static int lfs_fs_forceconsistency(lfs_t *lfs) {
if (err) {
return err;
}
return 0;
}
static int lfs_fs_forceconsistency(lfs_t *lfs) {
if (!lfs->globals.s.deorphaned) {
int err = lfs_fs_deorphan(lfs);
if (err) {
return err;
}
}
if (lfs->globals.s.moveid != 0x3ff) {
int err = lfs_fs_demove(lfs);
if (err) {
return err;
}
}
return 0;

17
lfs.h
View File

@@ -178,12 +178,6 @@ struct lfs_config {
// multiple of this value.
lfs_size_t prog_size;
// Size of block caches. Each cache buffers a portion of a block in RAM.
// This determines the size of the read cache, the program cache, and a
// cache per file. Larger caches can improve performance by storing more
// data. Must be a multiple of the read and program sizes.
lfs_size_t cache_size;
// Size of an erasable block. This does not impact ram consumption and
// may be larger than the physical erase size. However, this should be
// kept small as each file currently takes up an entire block.
@@ -193,6 +187,17 @@ struct lfs_config {
// Number of erasable blocks on the device.
lfs_size_t block_count;
// Number of erase cycles before we should move data to another block.
// May be zero to never move data, in which case no block-level
// wear-leveling is performed.
uint32_t block_cycles;
// Size of block caches. Each cache buffers a portion of a block in RAM.
// This determines the size of the read cache, the program cache, and a
// cache per file. Larger caches can improve performance by storing more
// data. Must be a multiple of the read and program sizes.
lfs_size_t cache_size;
// Number of blocks to lookahead during block allocation. A larger
// lookahead reduces the number of passes required to allocate a block.
// The lookahead buffer requires only 1 bit per block so it can be quite

15
tests/template.fmt
View File

@@ -69,10 +69,6 @@ uintmax_t test;
#define LFS_PROG_SIZE LFS_READ_SIZE
#endif
#ifndef LFS_CACHE_SIZE
#define LFS_CACHE_SIZE 64
#endif
#ifndef LFS_BLOCK_SIZE
#define LFS_BLOCK_SIZE 512
#endif
@@ -81,6 +77,14 @@ uintmax_t test;
#define LFS_BLOCK_COUNT 1024
#endif
#ifndef LFS_BLOCK_CYCLES
#define LFS_BLOCK_CYCLES 1024
#endif
#ifndef LFS_CACHE_SIZE
#define LFS_CACHE_SIZE 64
#endif
#ifndef LFS_LOOKAHEAD
#define LFS_LOOKAHEAD 128
#endif
@@ -94,9 +98,10 @@ const struct lfs_config cfg = {{
.read_size = LFS_READ_SIZE,
.prog_size = LFS_PROG_SIZE,
.cache_size = LFS_CACHE_SIZE,
.block_size = LFS_BLOCK_SIZE,
.block_count = LFS_BLOCK_COUNT,
.block_cycles = LFS_BLOCK_CYCLES,
.cache_size = LFS_CACHE_SIZE,
.lookahead = LFS_LOOKAHEAD,
}};