fix(ecstore): reject poisoned format heal candidates

This commit is contained in:
马登山
2026-07-30 09:23:55 +08:00
parent 709f333171
commit 84418df104
4 changed files with 230 additions and 51 deletions
+2 -1
View File
@@ -111,7 +111,8 @@ use crate::{
// event::name::EventName,
services::event_notification::{EventArgs, send_event},
store::init_format::{
format_disk_id_matches_slot, get_format_erasure_in_quorum, load_format_erasure, load_format_erasure_all, save_format_file,
formats_match_reference_slots, get_format_erasure_in_quorum, load_format_erasure, load_format_erasure_all,
save_format_file,
},
};
use bytes::Bytes;
+13 -9
View File
@@ -1373,6 +1373,14 @@ impl crate::storage_api_contracts::heal::HealOperations for SetDisks {
async fn heal_format(&self, dry_run: bool) -> Result<(HealResultItem, Option<Error>)> {
let disks = self.disks.read().await.clone();
let (formats, errs) = load_format_erasure_all(&disks, true).await;
if errs.iter().any(|err| {
matches!(
err,
Some(DiskError::InconsistentDisk | DiskError::CorruptedFormat | DiskError::CorruptedBackend)
)
}) {
return Ok((HealResultItem::default(), Some(StorageError::CorruptedFormat)));
}
let slot_offset = self
.set_index
.checked_mul(self.set_drive_count)
@@ -1382,14 +1390,7 @@ impl crate::storage_api_contracts::heal::HealOperations for SetDisks {
Ok(_) => return Ok((HealResultItem::default(), Some(StorageError::CorruptedFormat))),
Err(err) => {
let can_use_cached_layout = count_errs(&errs, &DiskError::UnformattedDisk) > 0
&& formats.iter().enumerate().all(|(index, format)| {
format.as_ref().is_none_or(|format| {
self.format.shared_identity() == format.shared_identity()
&& slot_offset
.checked_add(index)
.is_some_and(|slot| format_disk_id_matches_slot(format, slot))
})
})
&& formats_match_reference_slots(&formats, &self.format, slot_offset)
&& errs
.iter()
.all(|err| err.is_none() || matches!(err, Some(DiskError::UnformattedDisk)));
@@ -1400,6 +1401,9 @@ impl crate::storage_api_contracts::heal::HealOperations for SetDisks {
}
}
};
if !formats_match_reference_slots(&formats, &ref_format, slot_offset) {
return Ok((HealResultItem::default(), Some(StorageError::CorruptedFormat)));
}
let endpoints = crate::layout::endpoints::Endpoints::from(self.set_endpoints.clone());
let before_drives = crate::layout::set_heal::formats_to_drives_info(&endpoints, &formats, &errs);
@@ -1911,7 +1915,7 @@ mod heal_result_report_tests {
.await
.expect("format heal should report the quorum failure in its result");
assert!(matches!(heal_err, Some(Error::ErasureReadQuorum)));
assert!(matches!(heal_err, Some(Error::CorruptedFormat)));
let unformatted = load_format_erasure(disks[1].as_ref().expect("second disk should be online"), true)
.await
.expect_err("a rejected fallback must not format the missing slot");