aboutsummaryrefslogtreecommitdiff
path: root/src/api/s3
diff options
context:
space:
mode:
Diffstat (limited to 'src/api/s3')
-rw-r--r--src/api/s3/api_server.rs15
-rw-r--r--src/api/s3/copy.rs5
-rw-r--r--src/api/s3/multipart.rs14
-rw-r--r--src/api/s3/post_object.rs26
-rw-r--r--src/api/s3/put.rs223
5 files changed, 164 insertions, 119 deletions
diff --git a/src/api/s3/api_server.rs b/src/api/s3/api_server.rs
index 08405923..51f19554 100644
--- a/src/api/s3/api_server.rs
+++ b/src/api/s3/api_server.rs
@@ -17,8 +17,7 @@ use garage_model::key_table::Key;
use crate::generic_server::*;
use crate::s3::error::*;
-use crate::signature::payload::check_payload_signature;
-use crate::signature::streaming::*;
+use crate::signature::verify_request;
use crate::helpers::*;
use crate::s3::bucket::*;
@@ -125,17 +124,7 @@ impl ApiHandler for S3ApiServer {
return Ok(options_res.map(|_empty_body: EmptyBody| empty_body()));
}
- let (api_key, mut content_sha256) = check_payload_signature(&garage, "s3", &req).await?;
- let api_key = api_key
- .ok_or_else(|| Error::forbidden("Garage does not support anonymous access yet"))?;
-
- let req = parse_streaming_body(
- &api_key,
- req,
- &mut content_sha256,
- &garage.config.s3_api.s3_region,
- "s3",
- )?;
+ let (req, api_key, content_sha256) = verify_request(&garage, req, "s3").await?;
let bucket_name = match bucket_name {
None => {
diff --git a/src/api/s3/copy.rs b/src/api/s3/copy.rs
index 7eb1fe60..880ce5f4 100644
--- a/src/api/s3/copy.rs
+++ b/src/api/s3/copy.rs
@@ -387,7 +387,10 @@ pub async fn handle_upload_part_copy(
// we need to insert that data as a new block.
async move {
if must_upload {
- garage2.block_manager.rpc_put_block(final_hash, data).await
+ garage2
+ .block_manager
+ .rpc_put_block(final_hash, data, None)
+ .await
} else {
Ok(())
}
diff --git a/src/api/s3/multipart.rs b/src/api/s3/multipart.rs
index b9d15b21..5959bcd6 100644
--- a/src/api/s3/multipart.rs
+++ b/src/api/s3/multipart.rs
@@ -6,7 +6,6 @@ use hyper::{Request, Response};
use md5::{Digest as Md5Digest, Md5};
use garage_table::*;
-use garage_util::async_hash::*;
use garage_util::data::*;
use garage_model::bucket_table::Bucket;
@@ -135,17 +134,8 @@ pub async fn handle_put_part(
garage.version_table.insert(&version).await?;
// Copy data to version
- let first_block_hash = async_blake2sum(first_block.clone()).await;
-
- let (total_size, data_md5sum, data_sha256sum) = read_and_put_blocks(
- &garage,
- &version,
- part_number,
- first_block,
- first_block_hash,
- &mut chunker,
- )
- .await?;
+ let (total_size, data_md5sum, data_sha256sum, _) =
+ read_and_put_blocks(&garage, &version, part_number, first_block, &mut chunker).await?;
// Verify that checksums map
ensure_checksum_matches(
diff --git a/src/api/s3/post_object.rs b/src/api/s3/post_object.rs
index bca8d6c6..b542cc1a 100644
--- a/src/api/s3/post_object.rs
+++ b/src/api/s3/post_object.rs
@@ -21,7 +21,7 @@ use crate::s3::cors::*;
use crate::s3::error::*;
use crate::s3::put::{get_headers, save_stream};
use crate::s3::xml as s3_xml;
-use crate::signature::payload::{parse_date, verify_v4};
+use crate::signature::payload::{verify_v4, Authorization};
pub async fn handle_post_object(
garage: Arc<Garage>,
@@ -88,22 +88,11 @@ pub async fn handle_post_object(
.get("key")
.ok_or_bad_request("No key was provided")?
.to_str()?;
- let credential = params
- .get("x-amz-credential")
- .ok_or_else(|| Error::forbidden("Garage does not support anonymous access yet"))?
- .to_str()?;
let policy = params
.get("policy")
.ok_or_bad_request("No policy was provided")?
.to_str()?;
- let signature = params
- .get("x-amz-signature")
- .ok_or_bad_request("No signature was provided")?
- .to_str()?;
- let date = params
- .get("x-amz-date")
- .ok_or_bad_request("No date was provided")?
- .to_str()?;
+ let authorization = Authorization::parse_form(&params)?;
let key = if key.contains("${filename}") {
// if no filename is provided, don't replace. This matches the behavior of AWS.
@@ -116,16 +105,7 @@ pub async fn handle_post_object(
key.to_owned()
};
- let date = parse_date(date)?;
- let api_key = verify_v4(
- &garage,
- "s3",
- credential,
- &date,
- signature,
- policy.as_bytes(),
- )
- .await?;
+ let api_key = verify_v4(&garage, "s3", &authorization, policy.as_bytes()).await?;
let bucket_id = garage
.bucket_helper()
diff --git a/src/api/s3/put.rs b/src/api/s3/put.rs
index fdfa567d..489f1136 100644
--- a/src/api/s3/put.rs
+++ b/src/api/s3/put.rs
@@ -3,10 +3,13 @@ use std::sync::Arc;
use base64::prelude::*;
use futures::prelude::*;
+use futures::stream::FuturesOrdered;
use futures::try_join;
use md5::{digest::generic_array::*, Digest as Md5Digest, Md5};
use sha2::Sha256;
+use tokio::sync::mpsc;
+
use hyper::body::Bytes;
use hyper::header::{HeaderMap, HeaderValue};
use hyper::{Request, Response};
@@ -17,6 +20,7 @@ use opentelemetry::{
};
use garage_net::bytes_buf::BytesBuf;
+use garage_rpc::rpc_helper::OrderTag;
use garage_table::*;
use garage_util::async_hash::*;
use garage_util::data::*;
@@ -35,6 +39,8 @@ use crate::helpers::*;
use crate::s3::api_server::{ReqBody, ResBody};
use crate::s3::error::*;
+const PUT_BLOCKS_MAX_PARALLEL: usize = 3;
+
pub async fn handle_put(
garage: Arc<Garage>,
req: Request<ReqBody>,
@@ -168,17 +174,8 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
garage.version_table.insert(&version).await?;
// Transfer data and verify checksum
- let first_block_hash = async_blake2sum(first_block.clone()).await;
-
- let (total_size, data_md5sum, data_sha256sum) = read_and_put_blocks(
- &garage,
- &version,
- 1,
- first_block,
- first_block_hash,
- &mut chunker,
- )
- .await?;
+ let (total_size, data_md5sum, data_sha256sum, first_block_hash) =
+ read_and_put_blocks(&garage, &version, 1, first_block, &mut chunker).await?;
ensure_checksum_matches(
data_md5sum.as_slice(),
@@ -299,84 +296,164 @@ pub(crate) async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> +
version: &Version,
part_number: u64,
first_block: Bytes,
- first_block_hash: Hash,
chunker: &mut StreamChunker<S>,
-) -> Result<(u64, GenericArray<u8, typenum::U16>, Hash), Error> {
+) -> Result<(u64, GenericArray<u8, typenum::U16>, Hash, Hash), Error> {
let tracer = opentelemetry::global::tracer("garage");
- let md5hasher = AsyncHasher::<Md5>::new();
- let sha256hasher = AsyncHasher::<Sha256>::new();
+ let (block_tx, mut block_rx) = mpsc::channel::<Result<Bytes, Error>>(2);
+ let read_blocks = async {
+ block_tx.send(Ok(first_block)).await?;
+ loop {
+ let res = chunker
+ .next()
+ .with_context(Context::current_with_span(
+ tracer.start("Read block from client"),
+ ))
+ .await;
+ match res {
+ Ok(Some(block)) => block_tx.send(Ok(block)).await?,
+ Ok(None) => break,
+ Err(e) => {
+ block_tx.send(Err(e)).await?;
+ break;
+ }
+ }
+ }
+ drop(block_tx);
+ Ok::<_, mpsc::error::SendError<_>>(())
+ };
- futures::future::join(
- md5hasher.update(first_block.clone()),
- sha256hasher.update(first_block.clone()),
- )
- .with_context(Context::current_with_span(
- tracer.start("Hash first block (md5, sha256)"),
- ))
- .await;
+ let (block_tx2, mut block_rx2) = mpsc::channel::<Result<Bytes, Error>>(1);
+ let hash_stream = async {
+ let md5hasher = AsyncHasher::<Md5>::new();
+ let sha256hasher = AsyncHasher::<Sha256>::new();
+ while let Some(next) = block_rx.recv().await {
+ match next {
+ Ok(block) => {
+ block_tx2.send(Ok(block.clone())).await?;
+ futures::future::join(
+ md5hasher.update(block.clone()),
+ sha256hasher.update(block.clone()),
+ )
+ .with_context(Context::current_with_span(
+ tracer.start("Hash block (md5, sha256)"),
+ ))
+ .await;
+ }
+ Err(e) => {
+ block_tx2.send(Err(e)).await?;
+ break;
+ }
+ }
+ }
+ drop(block_tx2);
+ Ok::<_, mpsc::error::SendError<_>>(futures::join!(
+ md5hasher.finalize(),
+ sha256hasher.finalize()
+ ))
+ };
- let mut next_offset = first_block.len();
- let mut put_curr_version_block = put_block_meta(
- garage,
- version,
- part_number,
- 0,
- first_block_hash,
- first_block.len() as u64,
- );
- let mut put_curr_block = garage
- .block_manager
- .rpc_put_block(first_block_hash, first_block);
-
- loop {
- let (_, _, next_block) = futures::try_join!(
- put_curr_block.map_err(Error::from),
- put_curr_version_block.map_err(Error::from),
- chunker.next(),
- )?;
- if let Some(block) = next_block {
- let (_, _, block_hash) = futures::future::join3(
- md5hasher.update(block.clone()),
- sha256hasher.update(block.clone()),
- async_blake2sum(block.clone()),
- )
- .with_context(Context::current_with_span(
- tracer.start("Hash block (md5, sha256, blake2)"),
- ))
- .await;
- let block_len = block.len();
- put_curr_version_block = put_block_meta(
+ let (block_tx3, mut block_rx3) = mpsc::channel::<Result<(Bytes, Hash), Error>>(1);
+ let hash_blocks = async {
+ let mut first_block_hash = None;
+ while let Some(next) = block_rx2.recv().await {
+ match next {
+ Ok(block) => {
+ let hash = async_blake2sum(block.clone())
+ .with_context(Context::current_with_span(
+ tracer.start("Hash block (blake2)"),
+ ))
+ .await;
+ if first_block_hash.is_none() {
+ first_block_hash = Some(hash);
+ }
+ block_tx3.send(Ok((block, hash))).await?;
+ }
+ Err(e) => {
+ block_tx3.send(Err(e)).await?;
+ break;
+ }
+ }
+ }
+ drop(block_tx3);
+ Ok::<_, mpsc::error::SendError<_>>(first_block_hash.unwrap())
+ };
+
+ let put_blocks = async {
+ // Structure for handling several concurrent writes to storage nodes
+ let order_stream = OrderTag::stream();
+ let mut write_futs = FuturesOrdered::new();
+ let mut written_bytes = 0u64;
+ loop {
+ // Simultaneously write blocks to storage nodes & await for next block to be written
+ let currently_running = write_futs.len();
+ let write_futs_next = async {
+ if write_futs.is_empty() {
+ futures::future::pending().await
+ } else {
+ write_futs.next().await.unwrap()
+ }
+ };
+ let recv_next = async {
+ // If more than a maximum number of writes are in progress, don't add more for now
+ if currently_running >= PUT_BLOCKS_MAX_PARALLEL {
+ futures::future::pending().await
+ } else {
+ block_rx3.recv().await
+ }
+ };
+ let (block, hash) = tokio::select! {
+ result = write_futs_next => {
+ result?;
+ continue;
+ },
+ recv = recv_next => match recv {
+ Some(next) => next?,
+ None => break,
+ },
+ };
+
+ // For next block to be written: count its size and spawn future to write it
+ let offset = written_bytes;
+ written_bytes += block.len() as u64;
+ write_futs.push_back(put_block_and_meta(
garage,
version,
part_number,
- next_offset as u64,
- block_hash,
- block_len as u64,
- );
- put_curr_block = garage.block_manager.rpc_put_block(block_hash, block);
- next_offset += block_len;
- } else {
- break;
+ offset,
+ hash,
+ block,
+ order_stream.order(written_bytes),
+ ));
}
- }
+ while let Some(res) = write_futs.next().await {
+ res?;
+ }
+ Ok::<_, Error>(written_bytes)
+ };
+
+ let (_, stream_hash_result, block_hash_result, final_result) =
+ futures::join!(read_blocks, hash_stream, hash_blocks, put_blocks);
- let total_size = next_offset as u64;
- let data_md5sum = md5hasher.finalize().await;
+ let total_size = final_result?;
+ // unwrap here is ok, because if hasher failed, it is because something failed
+ // later in the pipeline which already caused a return at the ? on previous line
+ let (data_md5sum, data_sha256sum) = stream_hash_result.unwrap();
+ let first_block_hash = block_hash_result.unwrap();
- let data_sha256sum = sha256hasher.finalize().await;
let data_sha256sum = Hash::try_from(&data_sha256sum[..]).unwrap();
- Ok((total_size, data_md5sum, data_sha256sum))
+ Ok((total_size, data_md5sum, data_sha256sum, first_block_hash))
}
-async fn put_block_meta(
+async fn put_block_and_meta(
garage: &Garage,
version: &Version,
part_number: u64,
offset: u64,
hash: Hash,
- size: u64,
+ block: Bytes,
+ order_tag: OrderTag,
) -> Result<(), GarageError> {
let mut version = version.clone();
version.blocks.put(
@@ -384,7 +461,10 @@ async fn put_block_meta(
part_number,
offset,
},
- VersionBlock { hash, size },
+ VersionBlock {
+ hash,
+ size: block.len() as u64,
+ },
);
let block_ref = BlockRef {
@@ -394,6 +474,9 @@ async fn put_block_meta(
};
futures::try_join!(
+ garage
+ .block_manager
+ .rpc_put_block(hash, block, Some(order_tag)),
garage.version_table.insert(&version),
garage.block_ref_table.insert(&block_ref),
)?;
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-18 02:12:22 +0000