]> git.proxmox.com Git - proxmox-backup.git/blobdiff - src/backup/datastore.rs
add pbs-tools subcrate
[proxmox-backup.git] / src / backup / datastore.rs
index be764836c576d2a4fa3610d03e4404009f47d195..55458de36af7533c12546bdc9d2b54c75e70dd2b 100644 (file)
@@ -12,6 +12,9 @@ use lazy_static::lazy_static;
 
 use proxmox::tools::fs::{replace_file, file_read_optional_string, CreateOptions, open_file_locked};
 
+use pbs_tools::format::HumanByte;
+use pbs_tools::fs::{lock_dir_noblock, DirLockGuard};
+
 use super::backup_info::{BackupGroup, BackupDir};
 use super::chunk_store::ChunkStore;
 use super::dynamic_index::{DynamicIndexReader, DynamicIndexWriter};
@@ -22,8 +25,6 @@ use super::{DataBlob, ArchiveType, archive_type};
 use crate::config::datastore::{self, DataStoreConfig};
 use crate::task::TaskState;
 use crate::tools;
-use crate::tools::format::HumanByte;
-use crate::tools::fs::{lock_dir_noblock, DirLockGuard};
 use crate::api2::types::{Authid, GarbageCollectionStatus};
 use crate::server::UPID;
 
@@ -37,7 +38,7 @@ lazy_static! {
 /// management interface for backup.
 pub struct DataStore {
     chunk_store: Arc<ChunkStore>,
-    gc_mutex: Mutex<bool>,
+    gc_mutex: Mutex<()>,
     last_gc_status: Mutex<GarbageCollectionStatus>,
     verify_new: bool,
 }
@@ -69,6 +70,18 @@ impl DataStore {
         Ok(datastore)
     }
 
+    /// removes all datastores that are not configured anymore
+    pub fn remove_unused_datastores() -> Result<(), Error>{
+        let (config, _digest) = datastore::config()?;
+
+        let mut map = DATASTORE_MAP.lock().unwrap();
+        // removes all elements that are not in the config
+        map.retain(|key, _| {
+            config.sections.contains_key(key)
+        });
+        Ok(())
+    }
+
     fn open_with_path(store_name: &str, path: &Path, config: DataStoreConfig) -> Result<Self, Error> {
         let chunk_store = ChunkStore::open(store_name, path)?;
 
@@ -89,7 +102,7 @@ impl DataStore {
 
         Ok(Self {
             chunk_store: Arc::new(chunk_store),
-            gc_mutex: Mutex::new(false),
+            gc_mutex: Mutex::new(()),
             last_gc_status: Mutex::new(gc_status),
             verify_new: config.verify_new.unwrap_or(false),
         })
@@ -98,7 +111,7 @@ impl DataStore {
     pub fn get_chunk_iterator(
         &self,
     ) -> Result<
-        impl Iterator<Item = (Result<tools::fs::ReadDirEntry, Error>, usize, bool)>,
+        impl Iterator<Item = (Result<pbs_tools::fs::ReadDirEntry, Error>, usize, bool)>,
         Error
     > {
         self.chunk_store.get_chunk_iterator()
@@ -153,6 +166,34 @@ impl DataStore {
         Ok(out)
     }
 
+    /// Fast index verification - only check if chunks exists
+    pub fn fast_index_verification(
+        &self,
+        index: &dyn IndexFile,
+        checked: &mut HashSet<[u8;32]>,
+    ) -> Result<(), Error> {
+
+        for pos in 0..index.index_count() {
+            let info = index.chunk_info(pos).unwrap();
+            if checked.contains(&info.digest) {
+                continue;
+            }
+
+            self.stat_chunk(&info.digest).
+                map_err(|err| {
+                    format_err!(
+                        "fast_index_verification error, stat_chunk {} failed - {}",
+                        proxmox::tools::digest_to_hex(&info.digest),
+                        err,
+                    )
+                })?;
+
+            checked.insert(info.digest);
+        }
+
+        Ok(())
+    }
+
     pub fn name(&self) -> &str {
         self.chunk_store.name()
     }
@@ -175,7 +216,7 @@ impl DataStore {
         wanted_files.insert(CLIENT_LOG_BLOB_NAME.to_string());
         manifest.files().iter().for_each(|item| { wanted_files.insert(item.filename.clone()); });
 
-        for item in tools::fs::read_subdir(libc::AT_FDCWD, &full_path)? {
+        for item in pbs_tools::fs::read_subdir(libc::AT_FDCWD, &full_path)? {
             if let Ok(item) = item {
                 if let Some(file_type) = item.file_type() {
                     if file_type != nix::dir::Type::File { continue; }
@@ -214,7 +255,7 @@ impl DataStore {
 
         let full_path = self.group_path(backup_group);
 
-        let _guard = tools::fs::lock_dir_noblock(&full_path, "backup group", "possible running backup")?;
+        let _guard = pbs_tools::fs::lock_dir_noblock(&full_path, "backup group", "possible running backup")?;
 
         log::info!("removing backup group {:?}", full_path);
 
@@ -395,16 +436,18 @@ impl DataStore {
         }
         let handle_entry_err = |err: walkdir::Error| {
             if let Some(inner) = err.io_error() {
-                let path = err.path().unwrap_or(Path::new(""));
-                match inner.kind() {
-                    io::ErrorKind::PermissionDenied => {
+                if let Some(path) = err.path() {
+                    if inner.kind() == io::ErrorKind::PermissionDenied {
                         // only allow to skip ext4 fsck directory, avoid GC if, for example,
                         // a user got file permissions wrong on datastore rsync to new server
                         if err.depth() > 1 || !path.ends_with("lost+found") {
-                            bail!("cannot continue garbage-collection safely, permission denied on: {}", path.display())
+                            bail!("cannot continue garbage-collection safely, permission denied on: {:?}", path)
                         }
-                    },
-                    _ => bail!("unexpected error on datastore traversal: {} - {}", inner, path.display()),
+                    } else {
+                        bail!("unexpected error on datastore traversal: {} - {:?}", inner, path)
+                    }
+                } else {
+                    bail!("unexpected error on datastore traversal: {}", inner)
                 }
             }
             Ok(())
@@ -443,13 +486,12 @@ impl DataStore {
             worker.check_abort()?;
             tools::fail_on_shutdown()?;
             let digest = index.index_digest(pos).unwrap();
-            if let Err(err) = self.chunk_store.touch_chunk(digest) {
+            if !self.chunk_store.cond_touch_chunk(digest, false)? {
                 crate::task_warn!(
                     worker,
-                    "warning: unable to access chunk {}, required by {:?} - {}",
+                    "warning: unable to access non-existent chunk {}, required by {:?}",
                     proxmox::tools::digest_to_hex(digest),
                     file_name,
-                    err,
                 );
 
                 // touch any corresponding .bad files to keep them around, meaning if a chunk is
@@ -476,12 +518,11 @@ impl DataStore {
         let image_list = self.list_images()?;
         let image_count = image_list.len();
 
-        let mut done = 0;
         let mut last_percentage: usize = 0;
 
         let mut strange_paths_count: u64 = 0;
 
-        for img in image_list {
+        for (i, img) in image_list.into_iter().enumerate() {
 
             worker.check_abort()?;
             tools::fail_on_shutdown()?;
@@ -514,15 +555,14 @@ impl DataStore {
                 Err(err) if err.kind() == io::ErrorKind::NotFound => (), // ignore vanished files
                 Err(err) => bail!("can't open index {} - {}", img.to_string_lossy(), err),
             }
-            done += 1;
 
-            let percentage = done*100/image_count;
+            let percentage = (i + 1) * 100 / image_count;
             if percentage > last_percentage {
                 crate::task_log!(
                     worker,
                     "marked {}% ({} of {} index files)",
                     percentage,
-                    done,
+                    i + 1,
                     image_count,
                 );
                 last_percentage = percentage;
@@ -546,7 +586,7 @@ impl DataStore {
     }
 
     pub fn garbage_collection_running(&self) -> bool {
-        if let Ok(_) = self.gc_mutex.try_lock() { false } else { true }
+        !matches!(self.gc_mutex.try_lock(), Ok(_))
     }
 
     pub fn garbage_collection(&self, worker: &dyn TaskState, upid: &UPID) -> Result<(), Error> {
@@ -687,6 +727,11 @@ impl DataStore {
     }
 
 
+    pub fn stat_chunk(&self, digest: &[u8; 32]) -> Result<std::fs::Metadata, Error> {
+        let (chunk_path, _digest_str) = self.chunk_store.chunk_path(digest);
+        std::fs::metadata(chunk_path).map_err(Error::from)
+    }
+
     pub fn load_chunk(&self, digest: &[u8; 32]) -> Result<DataBlob, Error> {
 
         let (chunk_path, digest_str) = self.chunk_store.chunk_path(digest);
@@ -781,5 +826,42 @@ impl DataStore {
     pub fn verify_new(&self) -> bool {
         self.verify_new
     }
-}
 
+    /// returns a list of chunks sorted by their inode number on disk
+    /// chunks that could not be stat'ed are at the end of the list
+    pub fn get_chunks_in_order<F, A>(
+        &self,
+        index: &Box<dyn IndexFile + Send>,
+        skip_chunk: F,
+        check_abort: A,
+    ) -> Result<Vec<(usize, u64)>, Error>
+    where
+        F: Fn(&[u8; 32]) -> bool,
+        A: Fn(usize) -> Result<(), Error>,
+    {
+        let index_count = index.index_count();
+        let mut chunk_list = Vec::with_capacity(index_count);
+        use std::os::unix::fs::MetadataExt;
+        for pos in 0..index_count {
+            check_abort(pos)?;
+
+            let info = index.chunk_info(pos).unwrap();
+
+            if skip_chunk(&info.digest) {
+                continue;
+            }
+
+            let ino = match self.stat_chunk(&info.digest) {
+                Err(_) => u64::MAX, // could not stat, move to end of list
+                Ok(metadata) => metadata.ino(),
+            };
+
+            chunk_list.push((pos, ino));
+        }
+
+        // sorting by inode improves data locality, which makes it lots faster on spinners
+        chunk_list.sort_unstable_by(|(_, ino_a), (_, ino_b)| ino_a.cmp(&ino_b));
+
+        Ok(chunk_list)
+    }
+}