From ea28413eb20461fec76c31915bc828097ae94307 Mon Sep 17 00:00:00 2001
From: Christopher Haster <geky@geky.net>
Date: Wed, 10 May 2023 11:31:20 -0500
Subject: [PATCH] Added a bit of fuzz testing over mtree splits

This isn't the greatest coverage as we don't have a verifiable simulation.
Simulating the splitting-bucket-tree that is the mtree is tricky.

So right now this mostly just checks there's no internal assert failures and
if we have the expected number of entries afterwards.
---
 lfs.c                 |  7 +---
 tests/test_mtree.toml | 87 ++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 86 insertions(+), 8 deletions(-)

diff --git a/lfs.c b/lfs.c
index b849946f..0620c9e1 100644
--- a/lfs.c
+++ b/lfs.c
@@ -4523,12 +4523,7 @@ static inline int lfsr_mtree_isinlined(lfs_t *lfs) {
 }
 
 static inline lfs_ssize_t lfsr_mtree_weight(lfs_t *lfs) {
-    // inlined mdir?
-    if (lfsr_mtree_isinlined(lfs)) {
-        return lfs->mroot.rbyd.weight;
-    } else {
-        return lfsr_btree_weight(&lfs->mtree);
-    }
+    return lfsr_btree_weight(&lfs->mtree);
 }
 
 static int lfsr_mtree_lookup(lfs_t *lfs, lfs_ssize_t mid, lfsr_mdir_t *mdir_) {
diff --git a/tests/test_mtree.toml b/tests/test_mtree.toml
index e119f15c..696fef86 100644
--- a/tests/test_mtree.toml
+++ b/tests/test_mtree.toml
@@ -63,7 +63,7 @@ code = '''
 # TODO test many mroots
 
 # try creating a range of entries that may or may not split our mtree
-[cases.test_mtree_splitting]
+[cases.test_mtree_split]
 defines.N = [5, 10, 20, 40, 80, 160, 320]
 in = 'lfs.c'
 code = '''
@@ -71,6 +71,7 @@ code = '''
     lfs_t lfs;
     lfsr_format(&lfs, cfg) => 0;
 
+    // create entries
     lfsr_mount(&lfs, cfg) => 0;
     lfsr_mdir_t mdir;
     lfsr_mtree_lookup(&lfs, lfsr_mtree_weight(&lfs)-1, &mdir) => 0;
@@ -88,6 +89,7 @@ code = '''
     }
     lfsr_unmount(&lfs) => 0;
 
+    // try looking up each entry
     lfsr_mount(&lfs, cfg) => 0;
     lfs_ssize_t mid = -1;
     rid = 0;
@@ -112,7 +114,7 @@ code = '''
 # create a range of entries, and commit to the mroot several times,
 # this makes it more likely to force uninlining without necessarily
 # splitting
-[cases.test_mtree_uninlining]
+[cases.test_mtree_uninline]
 defines.N = [5, 10, 20, 40, 80, 160, 320]
 defines.M = [5, 5000]
 in = 'lfs.c'
@@ -121,6 +123,7 @@ code = '''
     lfs_t lfs;
     lfsr_format(&lfs, cfg) => 0;
 
+    // create entries
     lfsr_mount(&lfs, cfg) => 0;
     lfsr_mdir_t mdir;
     lfsr_mtree_lookup(&lfs, lfsr_mtree_weight(&lfs)-1, &mdir) => 0;
@@ -137,6 +140,8 @@ code = '''
         rid += 1;
     }
 
+    // also append a bunch of fs-level attributes, if the mdir should be
+    // uninlined this will force it to happen
     for (lfs_size_t i = 0; i < M; i++) {
         lfsr_mdir_commit(&lfs, &lfs.mroot, &(lfs_ssize_t){-1}, LFSR_ATTRS(
                 LFSR_ATTR(-1, UATTR(1), 0, &alphas[i % 26], 1))) => 0;
@@ -148,6 +153,7 @@ code = '''
     }
     lfsr_unmount(&lfs) => 0;
 
+    // try looking up each entry
     lfsr_mount(&lfs, cfg) => 0;
     lfs_ssize_t mid = -1;
     rid = 0;
@@ -174,3 +180,80 @@ code = '''
     lfsr_unmount(&lfs) => 0;
 '''
 
+# create random entries
+[cases.test_mtree_split_fuzz]
+defines.N = [5, 10, 20, 40, 80, 160, 320]
+defines.SAMPLES = 10
+# -1 => all pseudo-random seeds
+#  n => reproduce a specific seed
+defines.SEED = -1
+in = 'lfs.c'
+code = '''
+    const char *alphas = "abcdefghijklmnopqrstuvwxyz";
+
+    // iterate through severals seeds that we can reproduce easily
+    for (uint32_t seed = (SEED == -1 ? 1 : SEED);
+            (SEED == -1 ? seed < SAMPLES+1 : seed == SEED);
+            seed++) {
+        printf("--- seed: %d ---\n", seed);
+        // create lfs here since we need to reset each iteration, we're
+        // space constrained and we can't expect gc to work at this point
+        lfs_t lfs;
+        lfsr_format(&lfs, cfg) => 0;
+
+        lfsr_mount(&lfs, cfg) => 0;
+        // at least keep track of the number of entries we expect
+        lfs_size_t count = 0;
+
+        uint32_t prng = seed;
+        for (lfs_size_t i = 0; i < N; i++) {
+            // choose a pseudo-random mid
+            lfs_ssize_t mid = lfsr_mtree_weight(&lfs) == 0
+                    ? -1
+                    : (lfs_ssize_t)(TEST_PRNG(&prng) % lfsr_mtree_weight(&lfs));
+            // fetch mdir
+            lfsr_mdir_t mdir;
+            lfsr_mtree_lookup(&lfs, mid, &mdir) => 0;
+            // choose a pseudo-random rid
+            lfs_ssize_t rid = TEST_PRNG(&prng) % (mdir.rbyd.weight+1);
+
+            // add to rbyd, potentially splitting the mdir
+            lfsr_mdir_commit(&lfs, &mdir, &rid, LFSR_ATTRS(
+                    LFSR_ATTR(rid, MKINLINED, +1, &alphas[i % 26], 1))) => 0;
+
+            // make sure we can look up the new entry
+            uint8_t buffer[4];
+            lfsr_mdir_get(&lfs, &mdir, rid, LFSR_TAG_INLINED, buffer, 4) => 1;
+            assert(memcmp(buffer, &alphas[i % 26], 1) == 0);
+
+            count += 1;
+        }
+        lfsr_unmount(&lfs) => 0;
+
+        lfsr_mount(&lfs, cfg) => 0;
+        // try looking up each entry
+        lfs_size_t count_ = 0;
+
+        for (lfs_ssize_t mid = (lfsr_mtree_isinlined(&lfs) ? -1 : 0);
+                mid < lfsr_mtree_weight(&lfs);
+                mid++) {
+            lfsr_mdir_t mdir;
+            lfsr_mtree_lookup(&lfs, mid, &mdir) => 0;
+            for (lfs_ssize_t rid = 0;
+                    rid < (lfs_ssize_t)mdir.rbyd.weight;
+                    rid++) {
+                uint8_t buffer[4];
+                lfsr_mdir_get(&lfs, &mdir, rid, LFSR_TAG_INLINED,
+                        buffer, 4) => 1;
+
+                count_ += 1;
+            }
+        }
+
+        // the mtree is a bit difficult to simulate, but we can at least test
+        // we ended up with the right number of entries
+        assert(count_ == count);
+
+        lfsr_unmount(&lfs) => 0;
+    }
+'''