Only update bitmaps when writing data in primary subops

Fixes fully degraded EC writes and recovering EC writes corrupting bitmaps
This commit is contained in:
Vitaliy Filippov
2026-06-15 01:31:19 +03:00
parent 6a4627b625
commit 51ecbadb12
4 changed files with 59 additions and 17 deletions
+15 -3
View File
@@ -1571,10 +1571,22 @@ int blockstore_heap_t::add_small_write(object_id oid, heap_entry_t **obj_ptr, ui
wr->small().location = location;
if (bitmap)
memcpy(wr->get_ext_bitmap(this), bitmap, dsk->clean_entry_bitmap_size);
else if (obj)
memcpy(wr->get_ext_bitmap(this), obj->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
else
memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
{
bool found = false;
iterate_with_stable(obj, UINT64_MAX, [&](heap_entry_t *old_wr, bool stable)
{
if (old_wr->get_ext_bitmap(this))
{
found = true;
memcpy(wr->get_ext_bitmap(this), old_wr->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
return false;
}
return true;
});
if (!found)
memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
}
calc_checksums(wr, (uint8_t*)data, true);
*obj_ptr = wr;
});
+20 -12
View File
@@ -192,8 +192,16 @@ void osd_t::submit_primary_subop(osd_op_t *cur_op, osd_op_t *subop,
subop_len = 0;
}
si->read_error = false;
subop->bitmap = si->bmp_buf;
subop->bitmap_len = clean_entry_bitmap_size;
if (!wr || subop_len)
{
subop->bitmap = si->bmp_buf;
subop->bitmap_len = clean_entry_bitmap_size;
}
else
{
subop->bitmap = 0;
subop->bitmap_len = 0;
}
// Using rmw_buf to pass pointer to stripes. Dirty but works
subop->rmw_buf = si;
if (si->osd_num == this->osd_num)
@@ -217,14 +225,14 @@ void osd_t::submit_primary_subop(osd_op_t *cur_op, osd_op_t *subop,
.len = subop_len,
} },
.buf = (uint8_t*)(wr ? si->write_buf : si->read_buf),
.bitmap = (uint8_t*)si->bmp_buf,
.bitmap = (uint8_t*)subop->bitmap,
});
#ifdef OSD_DEBUG
printf(
"Submit %s to local: %jx:%jx v%ju %u-%u\n", wr ? "write" : "read",
inode, op_data->oid.stripe | si->role, op_version,
subop->bs_op->offset, subop->bs_op->len
);
printf(
"Submit %s to local: %jx:%jx v%ju %u-%u bmp %08x\n", wr ? "write" : "read",
inode, cur_op->op_data->oid.stripe | si->role, op_version,
subop->bs_op->offset, subop->bs_op->len, *(uint32_t*)si->bmp_buf
);
#endif
bs->enqueue_op(subop->bs_op);
}
@@ -243,14 +251,14 @@ void osd_t::submit_primary_subop(osd_op_t *cur_op, osd_op_t *subop,
.version = op_version,
.offset = wr ? si->write_start : si->read_start,
.len = subop_len,
.attr_len = wr ? clean_entry_bitmap_size : 0,
.attr_len = !wr || subop_len ? clean_entry_bitmap_size : 0,
.flags = cur_op->client_id == SELF_CLIENT && cur_op->req.hdr.opcode != OSD_OP_SCRUB ? OSD_OP_RECOVERY_RELATED : 0,
};
#ifdef OSD_DEBUG
printf(
"Submit %s to osd %ju: %jx:%jx v%ju %u-%u\n", wr ? "write" : "read", si->osd_num,
inode, op_data->oid.stripe | si->role, op_version,
subop->req.sec_rw.offset, subop->req.sec_rw.len
"Submit %s to osd %ju: %jx:%jx v%ju %u-%u bmp %08x\n", wr ? "write" : "read", si->osd_num,
inode, cur_op->op_data->oid.stripe | si->role, op_version,
subop->req.sec_rw.offset, subop->req.sec_rw.len, *(uint32_t*)si->bmp_buf
);
#endif
if (wr)
+4 -1
View File
@@ -162,7 +162,10 @@ void osd_t::exec_secondary_real(osd_op_t *cur_op)
cur_op->bs_op->offset = cur_op->req.sec_rw.offset;
cur_op->bs_op->len = cur_op->req.sec_rw.len;
cur_op->bs_op->buf = (uint8_t*)cur_op->buf;
cur_op->bs_op->bitmap = (uint8_t*)cur_op->bitmap;
if (cur_op->req.sec_rw.attr_len)
cur_op->bs_op->bitmap = (uint8_t*)cur_op->bitmap;
else
cur_op->bs_op->bitmap = NULL;
#ifdef OSD_STUB
cur_op->bs_op->retval = cur_op->bs_op->len;
#endif
+20 -1
View File
@@ -230,7 +230,26 @@ void test_update(bool csum)
_test_small_write(heap, dsk, 1, 0, 2, 8192, 4096, 16384, true, buffer_area.data()+16384, false);
object_id oid = { .inode = INODE_WITH_POOL(1, 1), .stripe = 0 };
assert(count_writes(heap, heap.read_entry(oid)) == 2);
heap_entry_t *obj = heap.read_entry(oid);
uint32_t mblock = 999;
int res = heap.add_small_write(oid, &obj, BS_HEAP_SMALL_WRITE, 3, 0, 0, 0, NULL, NULL, &mblock);
assert(res == 0);
assert(mblock == 0);
// Check inherited bitmap
uint8_t ref_ext_bitmap[dsk.clean_entry_bitmap_size];
memset(ref_ext_bitmap, 0xff, dsk.clean_entry_bitmap_size);
assert(!memcmp(obj->get_ext_bitmap(&heap), ref_ext_bitmap, dsk.clean_entry_bitmap_size));
mblock = 999;
res = heap.add_small_write(oid, &obj, BS_HEAP_SMALL_WRITE, 4, 0, 0, 0, NULL, NULL, &mblock);
assert(res == 0);
assert(mblock == 0);
// Check again - inherited from an unstable write
assert(!memcmp(obj->get_ext_bitmap(&heap), ref_ext_bitmap, dsk.clean_entry_bitmap_size));
assert(count_writes(heap, heap.read_entry(oid)) == 4);
}
printf("OK test_update %s\n", csum ? "csum" : "no_csum");