use anyhow::{anyhow, Result}; use byteorder::{LittleEndian, ReadBytesExt, WriteBytesExt}; use rand::prelude::*; use std::collections::VecDeque; use std::fs::OpenOptions; use std::io::{Cursor, Read, Seek, SeekFrom, Write}; use std::ops::Range; use std::path::{Path, PathBuf}; use tempfile::tempdir; use thinp::file_utils; use thinp::thin::xml::{self, Visit}; //------------------------------------ #[derive(Debug)] struct ThinBlock { thin_id: u32, thin_block: u64, data_block: u64, block_size: usize, } struct ThinReadRef { pub data: Vec, } struct ThinWriteRef<'a, W: Write + Seek> { file: &'a mut W, block_byte: u64, pub data: Vec, } impl ThinBlock { fn read_ref(&self, r: &mut R) -> Result { let mut rr = ThinReadRef { data: vec![0; self.block_size * 512], }; let byte = self.data_block * (self.block_size as u64) * 512; r.seek(SeekFrom::Start(byte))?; r.read_exact(&mut rr.data)?; Ok(rr) } fn zero_ref<'a, W: Write + Seek>(&self, w: &'a mut W) -> ThinWriteRef<'a, W> { ThinWriteRef { file: w, block_byte: self.data_block * (self.block_size as u64) * 512, data: vec![0; self.block_size * 512], } } //fn write_ref<'a, W>(&self, w: &'a mut W) -> Result> //where //W: Read + Write + Seek, //{ //let mut data = vec![0; self.block_size]; //w.seek(SeekFrom::Start(self.data_block * (self.block_size as u64)))?; //w.read_exact(&mut data[0..])?; // //let wr = ThinWriteRef { //file: w, //block_byte: self.data_block * (self.block_size as u64), //data: vec![0; self.block_size], //}; // //Ok(wr) //} } impl<'a, W: Write + Seek> Drop for ThinWriteRef<'a, W> { fn drop(&mut self) { // FIXME: We shouldn't panic in a drop function, so any IO // errors will have to make their way back to the user // another way (eg, via a flush() method). self.file.seek(SeekFrom::Start(self.block_byte)).unwrap(); self.file.write_all(&self.data).unwrap(); } } //------------------------------------ trait ThinVisitor { fn thin_block(&mut self, tb: &ThinBlock) -> Result<()>; } struct ThinXmlVisitor<'a, V: ThinVisitor> { inner: &'a mut V, block_size: Option, thin_id: Option, } impl<'a, V: ThinVisitor> xml::MetadataVisitor for ThinXmlVisitor<'a, V> { fn superblock_b(&mut self, sb: &xml::Superblock) -> Result { self.block_size = Some(sb.data_block_size); Ok(Visit::Continue) } fn superblock_e(&mut self) -> Result { Ok(Visit::Continue) } fn device_b(&mut self, d: &xml::Device) -> Result { self.thin_id = Some(d.dev_id); Ok(Visit::Continue) } fn device_e(&mut self) -> Result { Ok(Visit::Continue) } fn map(&mut self, m: &xml::Map) -> Result { for i in 0..m.len { let block = ThinBlock { thin_id: self.thin_id.unwrap(), thin_block: m.thin_begin + i, data_block: m.data_begin + i, block_size: self.block_size.unwrap() as usize, }; self.inner.thin_block(&block)?; } Ok(Visit::Continue) } fn eof(&mut self) -> Result { Ok(Visit::Stop) } } fn thin_visit(input: R, visitor: &mut M) -> Result<()> where R: Read, M: ThinVisitor, { let mut xml_visitor = ThinXmlVisitor { inner: visitor, block_size: None, thin_id: None, }; xml::read(input, &mut xml_visitor) } //------------------------------------ // To test thin_shrink we'd like to stamp a known pattern across the // provisioned areas of the thins in the pool, do the shrink, verify // the patterns. // A simple linear congruence generator used to create the data to // go into the thin blocks. struct Generator { x: u64, a: u64, c: u64, } impl Generator { fn new() -> Generator { Generator { x: 0, a: 6364136223846793005, c: 1442695040888963407, } } fn step(&mut self) { self.x = self.a.wrapping_mul(self.x).wrapping_add(self.c) } fn fill_buffer(&mut self, seed: u64, bytes: &mut [u8]) -> Result<()> { self.x = seed; assert!(bytes.len() % 8 == 0); let nr_words = bytes.len() / 8; let mut out = Cursor::new(bytes); for _ in 0..nr_words { out.write_u64::(self.x)?; self.step(); } Ok(()) } fn verify_buffer(&mut self, seed: u64, bytes: &[u8]) -> Result { self.x = seed; assert!(bytes.len() % 8 == 0); let nr_words = bytes.len() / 8; let mut input = Cursor::new(bytes); for _ in 0..nr_words { let w = input.read_u64::()?; if w != self.x { eprintln!("{} != {}", w, self.x); return Ok(false); } self.step(); } Ok(true) } } //------------------------------------ struct Stamper<'a, W: Write + Seek> { data_file: &'a mut W, seed: u64, } impl<'a, W: Write + Seek> Stamper<'a, W> { fn new(w: &'a mut W, seed: u64) -> Stamper<'a, W> { Stamper { data_file: w, seed } } } impl<'a, W: Write + Seek> ThinVisitor for Stamper<'a, W> { fn thin_block(&mut self, b: &ThinBlock) -> Result<()> { let mut wr = b.zero_ref(self.data_file); let mut gen = Generator::new(); gen.fill_buffer(self.seed ^ (b.thin_id as u64) ^ b.thin_block, &mut wr.data)?; Ok(()) } } //------------------------------------ struct Verifier<'a, R: Read + Seek> { data_file: &'a mut R, seed: u64, } impl<'a, R: Read + Seek> Verifier<'a, R> { fn new(r: &'a mut R, seed: u64) -> Verifier<'a, R> { Verifier { data_file: r, seed } } } impl<'a, R: Read + Seek> ThinVisitor for Verifier<'a, R> { fn thin_block(&mut self, b: &ThinBlock) -> Result<()> { let rr = b.read_ref(self.data_file)?; let mut gen = Generator::new(); if !gen.verify_buffer(self.seed ^ (b.thin_id as u64) ^ b.thin_block, &rr.data)? { return Err(anyhow!("data verify failed for {:?}", b)); } Ok(()) } } //------------------------------------ fn mk_path(dir: &Path, file: &str) -> PathBuf { let mut p = PathBuf::new(); p.push(dir); p.push(PathBuf::from(file)); p } fn generate_xml(path: &Path, g: &mut dyn Scenario) -> Result<()> { let xml_out = OpenOptions::new() .read(false) .write(true) .create(true) .truncate(true) .open(path)?; let mut w = xml::XmlWriter::new(xml_out); g.generate_xml(&mut w) } fn create_data_file(data_path: &Path, xml_path: &Path) -> Result<()> { let input = OpenOptions::new().read(true).write(false).open(xml_path)?; let sb = xml::read_superblock(input)?; let nr_blocks = sb.nr_data_blocks as u64; let block_size = sb.data_block_size as u64 * 512; let _file = file_utils::create_sized_file(data_path, nr_blocks * block_size)?; Ok(()) } fn stamp(xml_path: &Path, data_path: &Path, seed: u64) -> Result<()> { let mut data = OpenOptions::new() .read(false) .write(true) .open(&data_path)?; let xml = OpenOptions::new().read(true).write(false).open(&xml_path)?; let mut stamper = Stamper::new(&mut data, seed); thin_visit(xml, &mut stamper) } fn verify(xml_path: &Path, data_path: &Path, seed: u64) -> Result<()> { let mut data = OpenOptions::new() .read(true) .write(false) .open(&data_path)?; let xml = OpenOptions::new().read(true).write(false).open(&xml_path)?; let mut verifier = Verifier::new(&mut data, seed); thin_visit(xml, &mut verifier) } trait Scenario { fn generate_xml(&mut self, v: &mut dyn xml::MetadataVisitor) -> Result<()>; fn get_new_nr_blocks(&self) -> u64; } fn test_shrink(scenario: &mut dyn Scenario) -> Result<()> { let dir = tempdir()?; let xml_before = mk_path(dir.path(), "before.xml"); let xml_after = mk_path(dir.path(), "after.xml"); let data_path = mk_path(dir.path(), "metadata.bin"); generate_xml(&xml_before, scenario)?; create_data_file(&data_path, &xml_before)?; let mut rng = rand::thread_rng(); let seed = rng.gen::(); stamp(&xml_before, &data_path, seed)?; verify(&xml_before, &data_path, seed)?; let new_nr_blocks = scenario.get_new_nr_blocks(); thinp::shrink::toplevel::shrink(&xml_before, &xml_after, &data_path, new_nr_blocks, true)?; verify(&xml_after, &data_path, seed)?; Ok(()) } //------------------------------------ fn common_sb(nr_blocks: u64) -> xml::Superblock { xml::Superblock { uuid: "".to_string(), time: 0, transaction: 0, flags: None, version: None, data_block_size: 32, nr_data_blocks: nr_blocks, metadata_snap: None, } } struct EmptyPoolS {} impl Scenario for EmptyPoolS { fn generate_xml(&mut self, v: &mut dyn xml::MetadataVisitor) -> Result<()> { v.superblock_b(&common_sb(1024))?; v.superblock_e()?; Ok(()) } fn get_new_nr_blocks(&self) -> u64 { 512 } } #[test] fn shrink_empty_pool() -> Result<()> { let mut s = EmptyPoolS {}; test_shrink(&mut s) } //------------------------------------ struct SingleThinS { offset: u64, len: u64, old_nr_data_blocks: u64, new_nr_data_blocks: u64, } impl SingleThinS { fn new(offset: u64, len: u64, old_nr_data_blocks: u64, new_nr_data_blocks: u64) -> Self { SingleThinS { offset, len, old_nr_data_blocks, new_nr_data_blocks, } } } impl Scenario for SingleThinS { fn generate_xml(&mut self, v: &mut dyn xml::MetadataVisitor) -> Result<()> { v.superblock_b(&common_sb(self.old_nr_data_blocks))?; v.device_b(&xml::Device { dev_id: 0, mapped_blocks: self.len, transaction: 0, creation_time: 0, snap_time: 0, })?; v.map(&xml::Map { thin_begin: 0, data_begin: self.offset, time: 0, len: self.len, })?; v.device_e()?; v.superblock_e()?; Ok(()) } fn get_new_nr_blocks(&self) -> u64 { self.new_nr_data_blocks } } #[test] fn shrink_single_no_move_1() -> Result<()> { let mut s = SingleThinS::new(0, 1024, 2048, 1280); test_shrink(&mut s) } #[test] fn shrink_single_no_move_2() -> Result<()> { let mut s = SingleThinS::new(100, 1024, 2048, 1280); test_shrink(&mut s) } #[test] fn shrink_single_no_move_3() -> Result<()> { let mut s = SingleThinS::new(1024, 1024, 2048, 2048); test_shrink(&mut s) } #[test] fn shrink_single_partial_move() -> Result<()> { let mut s = SingleThinS::new(1024, 1024, 2048, 1280); test_shrink(&mut s) } #[test] fn shrink_single_total_move() -> Result<()> { let mut s = SingleThinS::new(2048, 1024, 1024 + 2048, 1280); test_shrink(&mut s) } #[test] fn shrink_insufficient_space() -> Result<()> { let mut s = SingleThinS::new(0, 2048, 3000, 1280); match test_shrink(&mut s) { Ok(_) => Err(anyhow!("Shrink unexpectedly succeeded")), Err(_) => Ok(()), } } //------------------------------------ struct FragmentedS { nr_thins: u32, thin_size: u64, old_nr_data_blocks: u64, new_nr_data_blocks: u64, } impl FragmentedS { fn new(nr_thins: u32, thin_size: u64) -> Self { let old_size = (nr_thins as u64) * thin_size; FragmentedS { nr_thins, thin_size, old_nr_data_blocks: (nr_thins as u64) * thin_size, new_nr_data_blocks: old_size * 3 / 4, } } } #[derive(Clone)] struct ThinRun { thin_id: u32, thin_begin: u64, len: u64, } #[derive(Clone, Debug, Copy)] struct MappedRun { thin_id: u32, thin_begin: u64, data_begin: u64, len: u64, } fn mk_runs(thin_id: u32, total_len: u64, run_len: std::ops::Range) -> Vec { let mut runs = Vec::new(); let mut b = 0u64; while b < total_len { let len = u64::min( total_len - b, thread_rng().gen_range(run_len.start, run_len.end), ); runs.push(ThinRun { thin_id: thin_id, thin_begin: b, len, }); b += len; } runs } impl Scenario for FragmentedS { fn generate_xml(&mut self, v: &mut dyn xml::MetadataVisitor) -> Result<()> { // Allocate each thin fully, in runs between 1 and 16. let mut runs = Vec::new(); for thin in 0..self.nr_thins { runs.append(&mut mk_runs(thin, self.thin_size, 1..17)); } // Shuffle runs.shuffle(&mut rand::thread_rng()); // map across the data let mut maps = Vec::new(); let mut b = 0; for r in &runs { maps.push(MappedRun { thin_id: r.thin_id, thin_begin: r.thin_begin, data_begin: b, len: r.len, }); b += r.len; } // drop half the mappings, which leaves us free runs let mut dropped = Vec::new(); for i in 0..maps.len() { if i % 2 == 0 { dropped.push(maps[i].clone()); } } // Unshuffle. This isn't strictly necc. but makes the xml // more readable. use std::cmp::Ordering; maps.sort_by(|&l, &r| match l.thin_id.cmp(&r.thin_id) { Ordering::Equal => l.thin_begin.cmp(&r.thin_begin), o => o, }); // write the xml v.superblock_b(&common_sb(self.old_nr_data_blocks))?; for thin in 0..self.nr_thins { v.device_b(&xml::Device { dev_id: thin, mapped_blocks: self.thin_size, transaction: 0, creation_time: 0, snap_time: 0, })?; for m in &dropped { if m.thin_id != thin { continue; } v.map(&xml::Map { thin_begin: m.thin_begin, data_begin: m.data_begin, time: 0, len: m.len, })?; } v.device_e()?; } v.superblock_e()?; Ok(()) } fn get_new_nr_blocks(&self) -> u64 { self.new_nr_data_blocks } } #[test] fn shrink_fragmented_thin_1() -> Result<()> { let mut s = FragmentedS::new(1, 2048); test_shrink(&mut s) } #[test] fn shrink_fragmented_thin_2() -> Result<()> { let mut s = FragmentedS::new(2, 2048); test_shrink(&mut s) } #[test] fn shrink_fragmented_thin_8() -> Result<()> { let mut s = FragmentedS::new(2, 2048); test_shrink(&mut s) } #[test] fn shrink_fragmented_thin_64() -> Result<()> { let mut s = FragmentedS::new(2, 2048); test_shrink(&mut s) } //------------------------------------ struct Allocator { runs: VecDeque>, } impl Allocator { fn new_shuffled(total_len: u64, run_len: Range) -> Allocator { let mut runs = Vec::new(); let mut b = 0u64; while b < total_len { let len = u64::min( total_len - b, thread_rng().gen_range(run_len.start, run_len.end), ); runs.push(b..(b + len)); b += len; } runs.shuffle(&mut thread_rng()); let runs: VecDeque> = runs.iter().map(|r| r.clone()).collect(); Allocator { runs } } fn is_empty(&self) -> bool { self.runs.is_empty() } fn alloc(&mut self, len: u64) -> Result>> { let mut len = len; let mut runs = Vec::new(); while len > 0 { let r = self.runs.pop_front(); if r.is_none() { return Err(anyhow!("could not allocate; out of space")); } let mut r = r.unwrap(); let rlen = r.end - r.start; if len < rlen { runs.push(r.start..(r.start + len)); // We need to push something back. self.runs.push_front((r.start + len)..r.end); len = 0; } else { runs.push(r.start..r.end); len -= rlen; } } Ok(runs) } } // Having explicitly unmapped regions makes it easier to // apply snapshots. #[derive(Clone)] enum Run { Mapped { data_begin: u64, len: u64 }, UnMapped { len: u64 }, } impl Run { fn len(&self) -> u64 { match self { Run::Mapped { data_begin: _data_begin, len, } => *len, Run::UnMapped { len } => *len, } } fn split(&self, n: u64) -> (Option, Option) { if n == 0 { return (None, Some(self.clone())); } else { if self.len() <= n { return (Some(self.clone()), None); } else { match self { Run::Mapped { data_begin, len } => ( Some(Run::Mapped { data_begin: *data_begin, len: n, }), Some(Run::Mapped { data_begin: data_begin + n, len: len - n, }), ), Run::UnMapped { len } => ( Some(Run::UnMapped { len: n }), Some(Run::UnMapped { len: len - n }), ), } } } } } #[derive(Clone)] struct ThinDev { thin_id: u32, dev_size: u64, runs: Vec, } impl ThinDev { fn emit(&self, v: &mut dyn xml::MetadataVisitor) -> Result<()> { v.device_b(&xml::Device { dev_id: self.thin_id, mapped_blocks: self.dev_size, transaction: 0, creation_time: 0, snap_time: 0, })?; let mut b = 0; for r in &self.runs { match r { Run::Mapped { data_begin, len } => { v.map(&xml::Map { thin_begin: b, data_begin: *data_begin, time: 0, len: *len, })?; b += len; } Run::UnMapped { len } => { b += len; } } } v.device_e()?; Ok(()) } } #[derive(Clone)] enum SnapRunType { Same, Diff, Hole, } #[derive(Clone)] struct SnapRun(SnapRunType, u64); fn mk_origin(thin_id: u32, total_len: u64, allocator: &mut Allocator) -> Result { let mut runs = Vec::new(); let mut b = 0; while b < total_len { let len = u64::min(thread_rng().gen_range(16, 64), total_len - b); match thread_rng().gen_range(0, 2) { 0 => { for data in allocator.alloc(len)? { assert!(data.end >= data.start); runs.push(Run::Mapped { data_begin: data.start, len: data.end - data.start, }); } } 1 => { runs.push(Run::UnMapped { len }); } _ => { return Err(anyhow!("bad value returned from rng")); } }; b += len; } Ok(ThinDev { thin_id, dev_size: total_len, runs, }) } fn mk_snap_mapping( total_len: u64, run_len: Range, same_percent: usize, diff_percent: usize, ) -> Vec { let mut runs = Vec::new(); let mut b = 0u64; while b < total_len { let len = u64::min( total_len - b, thread_rng().gen_range(run_len.start, run_len.end), ); let n = thread_rng().gen_range(0, 100); if n < same_percent { runs.push(SnapRun(SnapRunType::Same, len)); } else if n < diff_percent { runs.push(SnapRun(SnapRunType::Diff, len)); } else { runs.push(SnapRun(SnapRunType::Hole, len)); } b += len; } runs } fn split_runs(mut n: u64, runs: &Vec) -> (Vec, Vec) { let mut before = Vec::new(); let mut after = Vec::new(); for r in runs { match r.split(n) { (Some(lhs), None) => { before.push(lhs); } (Some(lhs), Some(rhs)) => { before.push(lhs); after.push(rhs); } (None, Some(rhs)) => { after.push(rhs); } (None, None) => {} } n -= r.len(); } (before, after) } fn apply_snap_runs( origin: &Vec, snap: &Vec, allocator: &mut Allocator, ) -> Result> { let mut origin = origin.clone(); let mut runs = Vec::new(); for SnapRun(st, slen) in snap { let (os, rest) = split_runs(*slen, &origin); match st { SnapRunType::Same => { for o in os { runs.push(o); } } SnapRunType::Diff => { for data in allocator.alloc(*slen)? { runs.push(Run::Mapped { data_begin: data.start, len: data.end - data.start, }); } } SnapRunType::Hole => { runs.push(Run::UnMapped { len: *slen }); } } origin = rest; } Ok(runs) } // Snapshots share mappings, not neccessarily the entire ranges. struct SnapS { len: u64, nr_snaps: u32, // Snaps will differ from the origin by this percentage percent_change: usize, old_nr_data_blocks: u64, new_nr_data_blocks: u64, } impl SnapS { fn new(len: u64, nr_snaps: u32, percent_change: usize) -> Self { let delta = len * (nr_snaps as u64) * (percent_change as u64) / 100; let old_nr_data_blocks = len + 3 * delta; let new_nr_data_blocks = len + 2 * delta; SnapS { len, nr_snaps, percent_change, old_nr_data_blocks, new_nr_data_blocks, } } } impl Scenario for SnapS { fn generate_xml(&mut self, v: &mut dyn xml::MetadataVisitor) -> Result<()> { let mut allocator = Allocator::new_shuffled(self.old_nr_data_blocks, 64..512); let origin = mk_origin(0, self.len, &mut allocator)?; v.superblock_b(&common_sb(self.old_nr_data_blocks))?; origin.emit(v)?; v.superblock_e()?; Ok(()) } fn get_new_nr_blocks(&self) -> u64 { self.new_nr_data_blocks } } #[test] fn shrink_identical_snap() -> Result<()> { let mut s = SnapS::new(1024, 1, 0); test_shrink(&mut s) } //------------------------------------