use anyhow::{anyhow, Result}; use nom::{number::complete::*, IResult}; use std::path::Path; use std::sync::{Arc, Mutex}; use std::thread; use threadpool::ThreadPool; use crate::checksum; use crate::io_engine::{AsyncIoEngine, Block, IoEngine, SyncIoEngine}; use crate::pdata::btree::{btree_to_map, btree_to_map_with_sm, BTreeWalker, Node, NodeVisitor}; use crate::pdata::space_map::*; use crate::pdata::unpack::*; use crate::report::*; use crate::thin::superblock::*; //------------------------------------------ #[allow(dead_code)] struct BlockTime { block: u64, time: u32, } impl Unpack for BlockTime { fn disk_size() -> u32 { 8 } fn unpack(i: &[u8]) -> IResult<&[u8], BlockTime> { let (i, n) = le_u64(i)?; let block = n >> 24; let time = n & ((1 << 24) - 1); Ok(( i, BlockTime { block, time: time as u32, }, )) } } struct BottomLevelVisitor { data_sm: Arc>, } //------------------------------------------ impl NodeVisitor for BottomLevelVisitor { fn visit(&mut self, _w: &BTreeWalker, _b: &Block, node: &Node) -> Result<()> { // FIXME: do other checks if let Node::Leaf { header: _h, keys: _k, values, } = node { if values.len() == 0 { return Ok(()); } let mut data_sm = self.data_sm.lock().unwrap(); let mut start = values[0].block; let mut len = 1; for n in 1..values.len() { let block = values[n].block; if block == start + len { len += 1; } else { data_sm.inc(start, len)?; start = block; len = 1; } } data_sm.inc(start, len)?; } Ok(()) } } //------------------------------------------ #[derive(Clone)] struct DeviceDetail { mapped_blocks: u64, transaction_id: u64, creation_time: u32, snapshotted_time: u32, } impl Unpack for DeviceDetail { fn disk_size() -> u32 { 24 } fn unpack(i: &[u8]) -> IResult<&[u8], DeviceDetail> { let (i, mapped_blocks) = le_u64(i)?; let (i, transaction_id) = le_u64(i)?; let (i, creation_time) = le_u32(i)?; let (i, snapshotted_time) = le_u32(i)?; Ok(( i, DeviceDetail { mapped_blocks, transaction_id, creation_time, snapshotted_time, }, )) } } //------------------------------------------ struct OverflowChecker<'a> { data_sm: &'a dyn SpaceMap, } impl<'a> OverflowChecker<'a> { fn new(data_sm: &'a dyn SpaceMap) -> OverflowChecker<'a> { OverflowChecker { data_sm } } } impl<'a> NodeVisitor for OverflowChecker<'a> { fn visit(&mut self, _w: &BTreeWalker, _b: &Block, node: &Node) -> Result<()> { if let Node::Leaf { header: _h, keys, values, } = node { for n in 0..keys.len() { let k = keys[n]; let v = values[n]; let expected = self.data_sm.get(k)?; if expected != v { return Err(anyhow!("Bad reference count for data block {}. Expected {}, but space map contains {}.", k, expected, v)); } } } Ok(()) } } //------------------------------------------ fn check_space_map( kind: &str, engine: Arc, report: &Arc, entries: Vec, metadata_sm: Option>>, sm: Arc>, root: SMRoot, ) -> Result<()> { let sm = sm.lock().unwrap(); // overflow btree { let mut v = OverflowChecker::new(&*sm); let mut w; if metadata_sm.is_none() { w = BTreeWalker::new(engine.clone(), false); } else { w = BTreeWalker::new_with_sm(engine.clone(), metadata_sm.unwrap().clone(), false)?; } w.walk(&mut v, root.ref_count_root)?; } let mut blocks = Vec::new(); for i in &entries { blocks.push(Block::new(i.blocknr)); } // FIXME: we should do this in batches engine.read_many(&mut blocks)?; let mut leaks = 0; let mut fail = false; let mut blocknr = 0; for n in 0..entries.len() { let b = &blocks[n]; if checksum::metadata_block_type(&b.get_data()) != checksum::BT::BITMAP { return Err(anyhow!( "Index entry points to block ({}) that isn't a bitmap", b.loc )); } let bitmap = unpack::(b.get_data())?; for e in bitmap.entries { if blocknr >= root.nr_blocks { break; } match e { BitmapEntry::Small(actual) => { let expected = sm.get(blocknr)?; if actual == 1 && expected == 0 { leaks += 1; } else if actual != expected as u8 { report.fatal(&format!("Bad reference count for {} block {}. Expected {}, but space map contains {}.", kind, blocknr, expected, actual)); fail = true; } } BitmapEntry::Overflow => { let expected = sm.get(blocknr)?; if expected < 3 { report.fatal(&format!("Bad reference count for {} block {}. Expected {}, but space map says it's >= 3.", kind, blocknr, expected)); fail = true; } } } blocknr += 1; } } if leaks > 0 { report.non_fatal(&format!( "{} {} blocks have leaked. Use --auto-repair to fix.", leaks, kind )); } if fail { return Err(anyhow!("Inconsistent data space map")); } Ok(()) } //------------------------------------------ fn inc_entries(sm: &Arc>, entries: &[IndexEntry]) -> Result<()> { let mut sm = sm.lock().unwrap(); for ie in entries { sm.inc(ie.blocknr, 1)?; } Ok(()) } //------------------------------------------ const MAX_CONCURRENT_IO: u32 = 1024; pub struct ThinCheckOptions<'a> { pub dev: &'a Path, pub async_io: bool, pub report: Arc, } pub fn check(opts: &ThinCheckOptions) -> Result<()> { let report = opts.report.clone(); let engine: Arc; let nr_threads; if opts.async_io { nr_threads = std::cmp::min(4, num_cpus::get()); engine = Arc::new(AsyncIoEngine::new(opts.dev, MAX_CONCURRENT_IO)?); } else { nr_threads = num_cpus::get() * 2; engine = Arc::new(SyncIoEngine::new(opts.dev, nr_threads)?); } // superblock let sb = read_superblock(engine.as_ref(), SUPERBLOCK_LOCATION)?; let nr_allocated_metadata; { let root = unpack::(&sb.metadata_sm_root[0..])?; nr_allocated_metadata = root.nr_allocated; } // Device details. We read this once to get the number of thin devices, and hence the // maximum metadata ref count. Then create metadata space map, and reread to increment // the ref counts for that metadata. let devs = btree_to_map::(engine.clone(), false, sb.details_root)?; let nr_devs = devs.len(); let metadata_sm = core_sm(engine.get_nr_blocks(), nr_devs as u32); report.set_title("Checking thin metadata"); let tid; let stop_progress = Arc::new(Mutex::new(false)); { let report = report.clone(); let sm = metadata_sm.clone(); let stop_progress = stop_progress.clone(); tid = thread::spawn(move || { let interval = std::time::Duration::from_millis(250); loop { { let stop_progress = stop_progress.lock().unwrap(); if *stop_progress { break; } } let sm = sm.lock().unwrap(); let mut n = sm.get_nr_allocated().unwrap(); drop(sm); n *= 100; n /= nr_allocated_metadata; let _r = report.progress(n as u8); thread::sleep(interval); } }); } report.set_sub_title("device details tree"); let _devs = btree_to_map_with_sm::( engine.clone(), metadata_sm.clone(), false, sb.details_root, )?; // increment superblock { let mut sm = metadata_sm.lock().unwrap(); sm.inc(SUPERBLOCK_LOCATION, 1)?; } // mapping top level let roots = btree_to_map_with_sm::(engine.clone(), metadata_sm.clone(), false, sb.mapping_root)?; // Check the mappings filling in the data_sm as we go. report.set_sub_title("mapping tree"); let data_sm; { // FIXME: with a thread pool we need to return errors another way. let nr_workers = nr_threads; let pool = ThreadPool::new(nr_workers); let root = unpack::(&sb.data_sm_root[0..])?; data_sm = core_sm(root.nr_blocks, nr_devs as u32); for (_thin_id, root) in roots { let mut w = BTreeWalker::new_with_sm(engine.clone(), metadata_sm.clone(), false)?; let data_sm = data_sm.clone(); pool.execute(move || { let mut v = BottomLevelVisitor { data_sm }; // FIXME: return error match w.walk(&mut v, root) { Err(e) => { eprintln!("walk failed {:?}", e); std::process::abort(); } Ok(_result) => { //eprintln!("checked thin_dev {} -> {:?}", thin_id, result); } } }); } pool.join(); } report.set_sub_title("data space map"); let root = unpack::(&sb.data_sm_root[0..])?; let entries = btree_to_map_with_sm::( engine.clone(), metadata_sm.clone(), false, root.bitmap_root, )?; let entries: Vec = entries.values().cloned().collect(); inc_entries(&metadata_sm, &entries[0..])?; check_space_map( "data", engine.clone(), &report, entries, Some(metadata_sm.clone()), data_sm.clone(), root, )?; report.set_sub_title("metadata space map"); let root = unpack::(&sb.metadata_sm_root[0..])?; let mut b = Block::new(root.bitmap_root); engine.read(&mut b)?; metadata_sm.lock().unwrap().inc(root.bitmap_root, 1)?; let entries = unpack::(b.get_data())?.indexes; // Unused entries will point to block 0 let entries: Vec = entries .iter() .take_while(|e| e.blocknr != 0) .cloned() .collect(); inc_entries(&metadata_sm, &entries[0..])?; // We call this for the side effect of incrementing the ref counts // for the metadata that holds the tree. let _counts = btree_to_map_with_sm::( engine.clone(), metadata_sm.clone(), false, root.ref_count_root, )?; // Now the counts should be correct and we can check it. check_space_map( "metadata", engine.clone(), &report, entries, None, metadata_sm.clone(), root, )?; // Completing consumes the report. { let mut stop_progress = stop_progress.lock().unwrap(); *stop_progress = true; } tid.join(); Ok(()) } //------------------------------------------