gbmap: Renamed/moved lookahead.bmapped -> gbmap.known

And:

- Tweaked the behavior of gbmap.window/known to _not_ match disk.
  gbmap.known matching disk is what required a separate
  lookahead.bmapped in the first place, but we never use both fields.

- _Don't_ revert gbmap on failed mdir commits!

  This was broken! If we reverted we risked inheriting outdated
  in-flight block information.

  This could be fixed by also zeroing lookahead.bmapped, but would force
  a gbmap rebuild. And why? The only interaction between mdir commit and
  the gbmap is block allocation, which is intentionally allowed to go
  out-of-sync to relax issues like this.

  Note we still revert in lfs3_fs_grow, the new gbmap we create there is
  incompatible with the previous disk size.

As a part of these changes, gbmap.window now behaves roughly the same as
gbmap.known and updates eagerly on block allocation.

This makes lookahead.window and gbmap.window somewhat redundant, but
simplifies the relevant logic (especially due to how lookahead.window
lags behind lookahead.off).

---

A bunch of bugs fell out-of-this, the interactions with lfs3_fs_mkgbmap
and lfs3_fs_grow being especially tricky, but fortunately our testing is
doing a good job.

At least the code changes were minimal, saves a bit of RAM:

                       code          stack          ctx
  no-gbmap before:    37168           2352          684
  no-gbmap after:     37168 (+0.0%)   2352 (+0.0%)  684 (+0.0%)

                       code          stack          ctx
  maybe-gbmap before: 39688           2392          852
  maybe-gbmap after:  39720 (+0.1%)   2376 (-0.7%)  848 (-0.5%)

                       code          stack          ctx
  yes-gbmap before:   39156           2392          852
  yes-gbmap after:    39208 (+0.1%)   2376 (-0.7%)  848 (-0.5%)
This commit is contained in:
Christopher Haster
2025-10-11 16:10:54 -05:00
parent 67d3c6ea69
commit 61dc21ccb7
3 changed files with 91 additions and 55 deletions
+51 -43
View File
@@ -5197,12 +5197,16 @@ static void lfs3_fs_claimbtree(lfs3_t *lfs3, lfs3_btree_t *btree) {
lfs3_btree_claim(&lfs3->mtree);
}
// claim the gbmap
// claim gbmap snapshots
#ifdef LFS3_GBMAP
if (&lfs3->gbmap.b != btree
&& lfs3->gbmap.b.r.blocks[0] == btree->r.blocks[0]) {
lfs3_btree_claim(&lfs3->gbmap.b);
}
if (&lfs3->gbmap.b_p != btree
&& lfs3->gbmap.b_p.r.blocks[0] == btree->r.blocks[0]) {
lfs3_btree_claim(&lfs3->gbmap.b_p);
}
#endif
// claim file btrees/bshrubs
@@ -7817,17 +7821,9 @@ static void lfs3_fs_revertgdelta(lfs3_t *lfs3) {
LFS3_UNREACHABLE();
}
// revert to the on-disk gbmap
#ifdef LFS3_GBMAP
if (lfs3_f_isgbmap(lfs3->flags)) {
err = lfs3_data_readgbmap(lfs3,
&LFS3_DATA_BUF(lfs3->gbmap_p, LFS3_GBMAP_DSIZE),
&lfs3->gbmap);
if (err) {
LFS3_UNREACHABLE();
}
}
#endif
// note we do _not_ revert the on-disk gbmap
//
// if we did, any in-flight state would be lost
}
#endif
@@ -7880,11 +7876,13 @@ static int lfs3_rbyd_appendgdelta(lfs3_t *lfs3, lfs3_rbyd_t *rbyd) {
// pending gbmap state?
#ifdef LFS3_GBMAP
if (lfs3_f_isgbmap(lfs3->flags)) {
// TODO is this the right place?
// try to update to most recent lookahead window
lfs3->gbmap.window = (lfs3->lookahead.window + lfs3->lookahead.off)
% lfs3->block_count;
lfs3->gbmap.known = lfs3->lookahead.gbmapped;
// lookahead and gbmap window offsets should always be in sync
//
// we probably don't need the duplicate fields, but it certainly
// makes the code simpler
LFS3_ASSERT(lfs3->gbmap.window
== (lfs3->lookahead.window + lfs3->lookahead.off)
% lfs3->block_count);
uint8_t gbmapdelta_[LFS3_GBMAP_DSIZE];
lfs3_data_fromgbmap(&lfs3->gbmap, gbmapdelta_);
@@ -11013,7 +11011,7 @@ static inline int lfs3_alloc_ckpoint(lfs3_t *lfs3) {
#ifdef LFS3_GBMAP
// do we need to rebuild the gbmap?
if (lfs3_f_isgbmap(lfs3->flags)
&& lfs3->lookahead.gbmapped < lfs3_min(
&& lfs3->gbmap.known < lfs3_min(
lfs3->cfg->gbmap_rebuild_thresh,
lfs3->block_count)) {
int err = lfs3_alloc_rebuildgbmap(lfs3);
@@ -11041,9 +11039,13 @@ static inline void lfs3_alloc_discard(lfs3_t *lfs3) {
// discard lookahead state
lfs3->lookahead.known = 0;
lfs3_memset(lfs3->lookahead.buffer, 0, lfs3->cfg->lookahead_size);
// discard gbmap window
// discard/resync the gbmap window, the resync is necessary to avoid
// disk changes breaking our mod math
#ifdef LFS3_GBMAP
lfs3->lookahead.gbmapped = 0;
lfs3->gbmap.window = (lfs3->lookahead.window + lfs3->lookahead.off)
% lfs3->block_count;
lfs3->gbmap.known = 0;
#endif
}
#endif
@@ -11163,8 +11165,8 @@ static void lfs3_alloc_inc(lfs3_t *lfs3) {
lfs3->lookahead.off += 1;
if (lfs3->lookahead.off == 8*lfs3->cfg->lookahead_size) {
lfs3->lookahead.off = 0;
lfs3->lookahead.window = (lfs3->lookahead.window
+ 8*lfs3->cfg->lookahead_size)
lfs3->lookahead.window
= (lfs3->lookahead.window + 8*lfs3->cfg->lookahead_size)
% lfs3->block_count;
}
@@ -11172,9 +11174,13 @@ static void lfs3_alloc_inc(lfs3_t *lfs3) {
lfs3->lookahead.known -= 1;
// decrement ckpoint
lfs3->lookahead.ckpoint -= 1;
// decrement gbmap known window
#ifdef LFS3_GBMAP
lfs3->lookahead.gbmapped = lfs3_smax(lfs3->lookahead.gbmapped-1, 0);
if (lfs3_f_isgbmap(lfs3->flags)) {
lfs3->gbmap.window = (lfs3->gbmap.window + 1) % lfs3->block_count;
lfs3->gbmap.known = lfs3_smax(lfs3->gbmap.known-1, 0);
}
#endif
}
#endif
@@ -11261,10 +11267,9 @@ static lfs3_sblock_t lfs3_alloc(lfs3_t *lfs3, uint32_t flags) {
// known blocks in our gbmap?
#ifdef LFS3_GBMAP
if (lfs3_f_isgbmap(lfs3->flags)
&& lfs3->lookahead.gbmapped > 0) {
if (lfs3_f_isgbmap(lfs3->flags) && lfs3->gbmap.known > 0) {
int err = lfs3_alloc_markgbmap(lfs3, &lfs3->gbmap.b, lfs3_min(
lfs3->lookahead.gbmapped,
lfs3->gbmap.known,
lfs3->lookahead.ckpoint));
if (err) {
return err;
@@ -11273,7 +11278,7 @@ static lfs3_sblock_t lfs3_alloc(lfs3_t *lfs3, uint32_t flags) {
// with known blocks we can trust the block state to be
// exact
lfs3_alloc_markfree(lfs3, lfs3_min(
lfs3->lookahead.gbmapped,
lfs3->gbmap.known,
lfs3->lookahead.ckpoint));
continue;
}
@@ -11318,7 +11323,7 @@ static lfs3_sblock_t lfs3_alloc(lfs3_t *lfs3, uint32_t flags) {
static int lfs3_alloc_rebuildgbmap(lfs3_t *lfs3) {
LFS3_INFO("Rebuilding gbmap "
"(gbmap %"PRId32"/%"PRId32")",
lfs3->lookahead.gbmapped,
lfs3->gbmap.known,
lfs3->block_count);
// create a copy of the gbmap
@@ -11380,12 +11385,10 @@ static int lfs3_alloc_rebuildgbmap(lfs3_t *lfs3) {
//
// we don't commit this to disk immediately, instead we piggypack on
// the next mdir commit, most writes terminate in an mdir commit so
// this avoids extra writing at a risk of needing to reconstruct the
// this avoids extra writing at a risk of needing to rebuild the
// gbmap if we lose power
//
// don't worry about window/known, lfs3_mdir_commit updates these
// last minute before calculating gdeltas for a commit
lfs3->lookahead.gbmapped = lfs3->lookahead.ckpoint;
lfs3->gbmap.known = lfs3->lookahead.ckpoint;
lfs3->gbmap.b = gbmap_;
return 0;
@@ -15567,8 +15570,6 @@ static int lfs3_init(lfs3_t *lfs3, uint32_t flags,
// setup other global gbmap state
#ifdef LFS3_GBMAP
lfs3_gbmap_init(&lfs3->gbmap);
// TODO should this be in the gbmap struct?
lfs3->lookahead.gbmapped = 0;
lfs3_memset(lfs3->gbmap_p, 0, LFS3_GBMAP_DSIZE);
lfs3_memset(lfs3->gbmap_d, 0, LFS3_GBMAP_DSIZE);
#endif
@@ -16154,10 +16155,10 @@ static int lfs3_mountinited(lfs3_t *lfs3) {
// if we have a gbmap, position our lookahead buffer at the last
// known gbmap window
lfs3->lookahead.window = lfs3->gbmap.window;
lfs3->lookahead.gbmapped = lfs3->gbmap.known;
#endif
} else {
// if we _don't_ have a gbmap, position our lookahead buffer
// if we don't have a gbmap, position our lookahead buffer
// pseudo-randomly using our gcksum as a prng
//
// the purpose of this is to avoid bad wear patterns such as always
@@ -16335,7 +16336,7 @@ static int lfs3_formatgbmap(lfs3_t *lfs3) {
// to test block 3 when gbmap is present
//
// assume we can write gbmap to block 2
lfs3->gbmap.window = 3;
lfs3->gbmap.window = 3 % lfs3->block_count;
lfs3->gbmap.known = lfs3->block_count;
lfs3->gbmap.b.r.blocks[0] = 2;
lfs3->gbmap.b.r.trunk = 0;
@@ -17053,7 +17054,7 @@ int lfs3_fs_grow(lfs3_t *lfs3, lfs3_size_t block_count_) {
// we can use the new blocks immediately as long as the commit
// with the new block_count is atomic
lfs3->block_count = block_count_;
// discard stale lookahead buffer
// discard stale lookahead buffer/gbmap
lfs3_alloc_discard(lfs3);
int err;
@@ -17111,6 +17112,12 @@ failed:;
lfs3->block_count = block_count;
// discard clobbered lookahead buffer
lfs3_alloc_discard(lfs3);
// revert to the previous gbmap
#ifdef LFS3_GBMAP
if (lfs3_f_isgbmap(lfs3->flags)) {
lfs3->gbmap.b = lfs3->gbmap.b_p;
}
#endif
return err;
}
@@ -17139,8 +17146,6 @@ int lfs3_fs_mkgbmap(lfs3_t *lfs3) {
// create an empty gbmap
lfs3_gbmap_init(&lfs3->gbmap);
// TODO should this be in the gbmap struct?
lfs3->lookahead.gbmapped = 0;
// start with everything free, rebuilding the gbmap will populate it
err = lfs3_gbmap_commit(lfs3, &lfs3->gbmap.b, 0, LFS3_RATTRS(
@@ -17152,7 +17157,12 @@ int lfs3_fs_mkgbmap(lfs3_t *lfs3) {
// go ahead and mark gbmap as in-use internally
lfs3->flags |= LFS3_F_GBMAP;
// checkpoint the allocator, this should trigger a rebuild
// sync gbmap/lookahead windows, note this needs to happen after any
// block allocation in lfs3_gbmap_commit
lfs3->gbmap.window = (lfs3->lookahead.window + lfs3->lookahead.off)
% lfs3->block_count;
// checkpoint the allocator again, this should trigger a rebuild
err = lfs3_alloc_ckpoint(lfs3);
if (err) {
goto failed;
@@ -17175,8 +17185,6 @@ failed:;
// if we failed clear the gbmap bit and reset the gbmap to be safe
lfs3->flags &= ~LFS3_F_GBMAP;
lfs3_gbmap_init(&lfs3->gbmap);
// TODO should this be in the gbmap struct?
lfs3->lookahead.gbmapped = 0;
return err;
}
#endif
-3
View File
@@ -908,9 +908,6 @@ typedef struct lfs3 {
lfs3_block_t off;
lfs3_block_t known;
lfs3_block_t ckpoint;
#ifdef LFS3_GBMAP
lfs3_block_t gbmapped;
#endif
uint8_t *buffer;
} lookahead;
#endif
+40 -9
View File
@@ -9,7 +9,14 @@
#
after = ['test_mtree', 'test_gbmap', 'test_dirs', 'test_files']
defines.FORMAT_BLOCK_COUNT = 'LFS3_IFDEF_YES_GBMAP(3, 2)'
# Test both with and without the gbmap if available
defines.GBMAP = [false, true]
if = '''
LFS3_IFDEF_YES_GBMAP(
GBMAP,
LFS3_IFDEF_GBMAP(true, !GBMAP))
'''
defines.FORMAT_BLOCK_COUNT = '(GBMAP) ? 3 : 2'
# test that we can alloc
@@ -30,7 +37,10 @@ code = '''
struct lfs3_cfg cfg = *CFG;
cfg.block_count = COUNT;
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, &cfg) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
&cfg) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, &cfg) => 0;
// start allocating
@@ -77,7 +87,10 @@ code = '''
struct lfs3_cfg cfg = *CFG;
cfg.block_count = COUNT;
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, &cfg) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
&cfg) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, &cfg) => 0;
// start allocating
@@ -138,7 +151,10 @@ defines.REMOUNT = [false, true]
in = 'lfs3.c'
code = '''
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, CFG) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
CFG) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, CFG) => 0;
// create this many directories
@@ -305,7 +321,10 @@ in = 'lfs3.c'
if = '(SIZE*N)/BLOCK_SIZE <= 32'
code = '''
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, CFG) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
CFG) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, CFG) => 0;
// create this many files
@@ -480,7 +499,10 @@ in = 'lfs3.c'
if = '(SIZE*N)/BLOCK_SIZE <= 32'
code = '''
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, CFG) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
CFG) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, CFG) => 0;
// create this many files
@@ -670,7 +692,10 @@ code = '''
struct lfs3_cfg cfg = *CFG;
cfg.block_count = COUNT;
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, &cfg) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
&cfg) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, &cfg) => 0;
// create n files repeatedly until we're sure we've wrapped around
@@ -754,7 +779,10 @@ code = '''
struct lfs3_cfg cfg = *CFG;
cfg.block_count = COUNT;
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, &cfg) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
&cfg) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, &cfg) => 0;
// create directories until we run out of space
@@ -837,7 +865,10 @@ code = '''
struct lfs3_cfg cfg = *CFG;
cfg.block_count = COUNT;
lfs3_t lfs3;
lfs3_format(&lfs3, LFS3_F_RDWR, &cfg) => 0;
lfs3_format(&lfs3,
LFS3_F_RDWR
| ((GBMAP) ? LFS3_IFDEF_GBMAP(LFS3_F_GBMAP, -1) : 0),
&cfg) => 0;
lfs3_mount(&lfs3, LFS3_M_RDWR, &cfg) => 0;
// create files until we run out of space