You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

PackParser.java 49KB

maxObjectSizeLimit for receive-pack. ReceivePack (and PackParser) can be configured with the maxObjectSizeLimit in order to prevent users from pushing too large objects to Git. The limit check is applied to all object types although it is most likely that a BLOB will exceed the limit. In all cases the size of the object header is excluded from the object size which is checked against the limit as this is the size of which a BLOB object would take in the working tree when checked out as a file. When an object exceeds the maxObjectSizeLimit the receive-pack will abort immediately. Delta objects (both offset and ref delta) are also checked against the limit. However, for delta objects we will first check the size of the inflated delta block against the maxObjectSizeLimit and abort immediately if it exceeds the limit. In this case we even do not know the exact size of the resolved delta object but we assume it will be larger than the given maxObjectSizeLimit as delta is generally only chosen if the delta can copy more data from the base object than the delta needs to insert or needs to represent the copy ranges. Aborting early, in this case, avoids unnecessary inflating of the (huge) delta block. Unfortunately, it is too expensive (especially for a large delta) to compute SHA-1 of an object that causes the receive-pack to abort. This would decrease the value of this feature whose main purpose is to protect server resources from users pushing huge objects. Therefore we don't report the SHA-1 in the error message. Change-Id: I177ef24553faacda444ed5895e40ac8925ca0d1e Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
12 years ago
ObjectIdOwnerMap: More lightweight map for ObjectIds OwnerMap is about 200 ms faster than SubclassMap, more friendly to the GC, and uses less storage: testing the "Counting objects" part of PackWriter on 1886362 objects: ObjectIdSubclassMap: load factor 50% table: 4194304 (wasted 2307942) ms spent 36998 36009 34795 34703 34941 35070 34284 34511 34638 34256 ms avg 34800 (last 9 runs) ObjectIdOwnerMap: load factor 100% table: 2097152 (wasted 210790) directory: 1024 ms spent 36842 35112 34922 34703 34580 34782 34165 34662 34314 34140 ms avg 34597 (last 9 runs) The major difference with OwnerMap is entries must extend from ObjectIdOwnerMap.Entry, where the OwnerMap has injected its own private "next" field into each object. This allows the OwnerMap to use a singly linked list for chaining collisions within a bucket. By putting collisions in a linked list, we gain the entire table back for the SHA-1 bits to index their own "private" slot. Unfortunately this means that each object can appear in at most ONE OwnerMap, as there is only one "next" field within the object instance to thread into the map. For types that are very object map heavy like RevWalk (entity RevObject) and PackWriter (entity ObjectToPack) this is sufficient, these entity types are only put into one map by their container. By introducing a new map type, we don't break existing applications that might be trying to use ObjectIdSubclassMap to track RevCommits they obtained from a RevWalk. The OwnerMap uses less memory. Each object uses 1 reference more (so we're up 1,886,362 references), but the table is 1/2 the size (2^20 rather than 2^21). The table itself wastes only 210,790 slots, rather than 2,307,942. So OwnerMap is wasting 200k fewer references. OwnerMap is more friendly to the GC, because it hardly ever generates garbage. As the map reaches its 100% load factor target, it doubles in size by allocating additional segment arrays of 2048 entries. (So the first grow allocates 1 segment, second 2 segments, third 4 segments, etc.) These segments are hooked into the pre-allocated directory of 1024 spaces. This permits the map to grow to 2 million objects before the directory itself has to grow. By using segments of 2048 entries, we are asking the GC to acquire 8,204 bytes in a 32 bit JVM. This is easier to satisfy then 2,307,942 bytes (for the 512k table that is just an intermediate step in the SubclassMap). By reusing the previously allocated segments (they are re-hashed in-place) we don't release any memory during a table grow. When the directory grows, it does so by discarding the old one and using one that is 4x larger (so the directory goes to 4096 entries on its first grow). A directory of size 4096 can handle up to 8 millon objects. The second directory grow (16384) goes to 33 million objects. At that point we're starting to really push the limits of the JVM heap, but at least its many small arrays. Previously SubclassMap would need a table of 67108864 entries to handle that object count, which needs a single contiguous allocation of 256 MiB. That's hard to come by in a 32 bit JVM. Instead OwnerMap uses 8192 arrays of about 8 KiB each. This is much easier to fit into a fragmented heap. Change-Id: Ia4acf5cfbf7e9b71bc7faa0db9060f6a969c0c50 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
ObjectIdOwnerMap: More lightweight map for ObjectIds OwnerMap is about 200 ms faster than SubclassMap, more friendly to the GC, and uses less storage: testing the "Counting objects" part of PackWriter on 1886362 objects: ObjectIdSubclassMap: load factor 50% table: 4194304 (wasted 2307942) ms spent 36998 36009 34795 34703 34941 35070 34284 34511 34638 34256 ms avg 34800 (last 9 runs) ObjectIdOwnerMap: load factor 100% table: 2097152 (wasted 210790) directory: 1024 ms spent 36842 35112 34922 34703 34580 34782 34165 34662 34314 34140 ms avg 34597 (last 9 runs) The major difference with OwnerMap is entries must extend from ObjectIdOwnerMap.Entry, where the OwnerMap has injected its own private "next" field into each object. This allows the OwnerMap to use a singly linked list for chaining collisions within a bucket. By putting collisions in a linked list, we gain the entire table back for the SHA-1 bits to index their own "private" slot. Unfortunately this means that each object can appear in at most ONE OwnerMap, as there is only one "next" field within the object instance to thread into the map. For types that are very object map heavy like RevWalk (entity RevObject) and PackWriter (entity ObjectToPack) this is sufficient, these entity types are only put into one map by their container. By introducing a new map type, we don't break existing applications that might be trying to use ObjectIdSubclassMap to track RevCommits they obtained from a RevWalk. The OwnerMap uses less memory. Each object uses 1 reference more (so we're up 1,886,362 references), but the table is 1/2 the size (2^20 rather than 2^21). The table itself wastes only 210,790 slots, rather than 2,307,942. So OwnerMap is wasting 200k fewer references. OwnerMap is more friendly to the GC, because it hardly ever generates garbage. As the map reaches its 100% load factor target, it doubles in size by allocating additional segment arrays of 2048 entries. (So the first grow allocates 1 segment, second 2 segments, third 4 segments, etc.) These segments are hooked into the pre-allocated directory of 1024 spaces. This permits the map to grow to 2 million objects before the directory itself has to grow. By using segments of 2048 entries, we are asking the GC to acquire 8,204 bytes in a 32 bit JVM. This is easier to satisfy then 2,307,942 bytes (for the 512k table that is just an intermediate step in the SubclassMap). By reusing the previously allocated segments (they are re-hashed in-place) we don't release any memory during a table grow. When the directory grows, it does so by discarding the old one and using one that is 4x larger (so the directory goes to 4096 entries on its first grow). A directory of size 4096 can handle up to 8 millon objects. The second directory grow (16384) goes to 33 million objects. At that point we're starting to really push the limits of the JVM heap, but at least its many small arrays. Previously SubclassMap would need a table of 67108864 entries to handle that object count, which needs a single contiguous allocation of 256 MiB. That's hard to come by in a 32 bit JVM. Instead OwnerMap uses 8192 arrays of about 8 KiB each. This is much easier to fit into a fragmented heap. Change-Id: Ia4acf5cfbf7e9b71bc7faa0db9060f6a969c0c50 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
maxObjectSizeLimit for receive-pack. ReceivePack (and PackParser) can be configured with the maxObjectSizeLimit in order to prevent users from pushing too large objects to Git. The limit check is applied to all object types although it is most likely that a BLOB will exceed the limit. In all cases the size of the object header is excluded from the object size which is checked against the limit as this is the size of which a BLOB object would take in the working tree when checked out as a file. When an object exceeds the maxObjectSizeLimit the receive-pack will abort immediately. Delta objects (both offset and ref delta) are also checked against the limit. However, for delta objects we will first check the size of the inflated delta block against the maxObjectSizeLimit and abort immediately if it exceeds the limit. In this case we even do not know the exact size of the resolved delta object but we assume it will be larger than the given maxObjectSizeLimit as delta is generally only chosen if the delta can copy more data from the base object than the delta needs to insert or needs to represent the copy ranges. Aborting early, in this case, avoids unnecessary inflating of the (huge) delta block. Unfortunately, it is too expensive (especially for a large delta) to compute SHA-1 of an object that causes the receive-pack to abort. This would decrease the value of this feature whose main purpose is to protect server resources from users pushing huge objects. Therefore we don't report the SHA-1 in the error message. Change-Id: I177ef24553faacda444ed5895e40ac8925ca0d1e Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
12 years ago
maxObjectSizeLimit for receive-pack. ReceivePack (and PackParser) can be configured with the maxObjectSizeLimit in order to prevent users from pushing too large objects to Git. The limit check is applied to all object types although it is most likely that a BLOB will exceed the limit. In all cases the size of the object header is excluded from the object size which is checked against the limit as this is the size of which a BLOB object would take in the working tree when checked out as a file. When an object exceeds the maxObjectSizeLimit the receive-pack will abort immediately. Delta objects (both offset and ref delta) are also checked against the limit. However, for delta objects we will first check the size of the inflated delta block against the maxObjectSizeLimit and abort immediately if it exceeds the limit. In this case we even do not know the exact size of the resolved delta object but we assume it will be larger than the given maxObjectSizeLimit as delta is generally only chosen if the delta can copy more data from the base object than the delta needs to insert or needs to represent the copy ranges. Aborting early, in this case, avoids unnecessary inflating of the (huge) delta block. Unfortunately, it is too expensive (especially for a large delta) to compute SHA-1 of an object that causes the receive-pack to abort. This would decrease the value of this feature whose main purpose is to protect server resources from users pushing huge objects. Therefore we don't report the SHA-1 in the error message. Change-Id: I177ef24553faacda444ed5895e40ac8925ca0d1e Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
12 years ago
ObjectIdOwnerMap: More lightweight map for ObjectIds OwnerMap is about 200 ms faster than SubclassMap, more friendly to the GC, and uses less storage: testing the "Counting objects" part of PackWriter on 1886362 objects: ObjectIdSubclassMap: load factor 50% table: 4194304 (wasted 2307942) ms spent 36998 36009 34795 34703 34941 35070 34284 34511 34638 34256 ms avg 34800 (last 9 runs) ObjectIdOwnerMap: load factor 100% table: 2097152 (wasted 210790) directory: 1024 ms spent 36842 35112 34922 34703 34580 34782 34165 34662 34314 34140 ms avg 34597 (last 9 runs) The major difference with OwnerMap is entries must extend from ObjectIdOwnerMap.Entry, where the OwnerMap has injected its own private "next" field into each object. This allows the OwnerMap to use a singly linked list for chaining collisions within a bucket. By putting collisions in a linked list, we gain the entire table back for the SHA-1 bits to index their own "private" slot. Unfortunately this means that each object can appear in at most ONE OwnerMap, as there is only one "next" field within the object instance to thread into the map. For types that are very object map heavy like RevWalk (entity RevObject) and PackWriter (entity ObjectToPack) this is sufficient, these entity types are only put into one map by their container. By introducing a new map type, we don't break existing applications that might be trying to use ObjectIdSubclassMap to track RevCommits they obtained from a RevWalk. The OwnerMap uses less memory. Each object uses 1 reference more (so we're up 1,886,362 references), but the table is 1/2 the size (2^20 rather than 2^21). The table itself wastes only 210,790 slots, rather than 2,307,942. So OwnerMap is wasting 200k fewer references. OwnerMap is more friendly to the GC, because it hardly ever generates garbage. As the map reaches its 100% load factor target, it doubles in size by allocating additional segment arrays of 2048 entries. (So the first grow allocates 1 segment, second 2 segments, third 4 segments, etc.) These segments are hooked into the pre-allocated directory of 1024 spaces. This permits the map to grow to 2 million objects before the directory itself has to grow. By using segments of 2048 entries, we are asking the GC to acquire 8,204 bytes in a 32 bit JVM. This is easier to satisfy then 2,307,942 bytes (for the 512k table that is just an intermediate step in the SubclassMap). By reusing the previously allocated segments (they are re-hashed in-place) we don't release any memory during a table grow. When the directory grows, it does so by discarding the old one and using one that is 4x larger (so the directory goes to 4096 entries on its first grow). A directory of size 4096 can handle up to 8 millon objects. The second directory grow (16384) goes to 33 million objects. At that point we're starting to really push the limits of the JVM heap, but at least its many small arrays. Previously SubclassMap would need a table of 67108864 entries to handle that object count, which needs a single contiguous allocation of 256 MiB. That's hard to come by in a 32 bit JVM. Instead OwnerMap uses 8192 arrays of about 8 KiB each. This is much easier to fit into a fragmented heap. Change-Id: Ia4acf5cfbf7e9b71bc7faa0db9060f6a969c0c50 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
maxObjectSizeLimit for receive-pack. ReceivePack (and PackParser) can be configured with the maxObjectSizeLimit in order to prevent users from pushing too large objects to Git. The limit check is applied to all object types although it is most likely that a BLOB will exceed the limit. In all cases the size of the object header is excluded from the object size which is checked against the limit as this is the size of which a BLOB object would take in the working tree when checked out as a file. When an object exceeds the maxObjectSizeLimit the receive-pack will abort immediately. Delta objects (both offset and ref delta) are also checked against the limit. However, for delta objects we will first check the size of the inflated delta block against the maxObjectSizeLimit and abort immediately if it exceeds the limit. In this case we even do not know the exact size of the resolved delta object but we assume it will be larger than the given maxObjectSizeLimit as delta is generally only chosen if the delta can copy more data from the base object than the delta needs to insert or needs to represent the copy ranges. Aborting early, in this case, avoids unnecessary inflating of the (huge) delta block. Unfortunately, it is too expensive (especially for a large delta) to compute SHA-1 of an object that causes the receive-pack to abort. This would decrease the value of this feature whose main purpose is to protect server resources from users pushing huge objects. Therefore we don't report the SHA-1 in the error message. Change-Id: I177ef24553faacda444ed5895e40ac8925ca0d1e Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
12 years ago
Store Git on any DHT jgit.storage.dht is a storage provider implementation for JGit that permits storing the Git repository in a distributed hashtable, NoSQL system, or other database. The actual underlying storage system is undefined, and can be plugged in by implementing 7 small interfaces: * Database * RepositoryIndexTable * RepositoryTable * RefTable * ChunkTable * ObjectIndexTable * WriteBuffer The storage provider interface tries to assume very little about the underlying storage system, and requires only three key features: * key -> value lookup (a hashtable is suitable) * atomic updates on single rows * asynchronous operations (Java's ExecutorService is easy to use) Most NoSQL database products offer all 3 of these features in their clients, and so does any decent network based cache system like the open source memcache product. Relying only on key equality for data retrevial makes it simple for the storage engine to distribute across multiple machines. Traditional SQL systems could also be used with a JDBC based spi implementation. Before submitting this change I have implemented six storage systems for the spi layer: * Apache HBase[1] * Apache Cassandra[2] * Google Bigtable[3] * an in-memory implementation for unit testing * a JDBC implementation for SQL * a generic cache provider that can ride on top of memcache All six systems came in with an spi layer around 1000 lines of code to implement the above 7 interfaces. This is a huge reduction in size compared to prior attempts to implement a new JGit storage layer. As this package shows, a complete JGit storage implementation is more than 17,000 lines of fairly complex code. A simple cache is provided in storage.dht.spi.cache. Implementers can use CacheDatabase to wrap any other type of Database and perform fast reads against a network based cache service, such as the open source memcached[4]. An implementation of CacheService must be provided to glue this spi onto the network cache. [1] https://github.com/spearce/jgit_hbase [2] https://github.com/spearce/jgit_cassandra [3] http://labs.google.com/papers/bigtable.html [4] http://memcached.org/ Change-Id: I0aa4072781f5ccc019ca421c036adff2c40c4295 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
maxObjectSizeLimit for receive-pack. ReceivePack (and PackParser) can be configured with the maxObjectSizeLimit in order to prevent users from pushing too large objects to Git. The limit check is applied to all object types although it is most likely that a BLOB will exceed the limit. In all cases the size of the object header is excluded from the object size which is checked against the limit as this is the size of which a BLOB object would take in the working tree when checked out as a file. When an object exceeds the maxObjectSizeLimit the receive-pack will abort immediately. Delta objects (both offset and ref delta) are also checked against the limit. However, for delta objects we will first check the size of the inflated delta block against the maxObjectSizeLimit and abort immediately if it exceeds the limit. In this case we even do not know the exact size of the resolved delta object but we assume it will be larger than the given maxObjectSizeLimit as delta is generally only chosen if the delta can copy more data from the base object than the delta needs to insert or needs to represent the copy ranges. Aborting early, in this case, avoids unnecessary inflating of the (huge) delta block. Unfortunately, it is too expensive (especially for a large delta) to compute SHA-1 of an object that causes the receive-pack to abort. This would decrease the value of this feature whose main purpose is to protect server resources from users pushing huge objects. Therefore we don't report the SHA-1 in the error message. Change-Id: I177ef24553faacda444ed5895e40ac8925ca0d1e Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
12 years ago
Store Git on any DHT jgit.storage.dht is a storage provider implementation for JGit that permits storing the Git repository in a distributed hashtable, NoSQL system, or other database. The actual underlying storage system is undefined, and can be plugged in by implementing 7 small interfaces: * Database * RepositoryIndexTable * RepositoryTable * RefTable * ChunkTable * ObjectIndexTable * WriteBuffer The storage provider interface tries to assume very little about the underlying storage system, and requires only three key features: * key -> value lookup (a hashtable is suitable) * atomic updates on single rows * asynchronous operations (Java's ExecutorService is easy to use) Most NoSQL database products offer all 3 of these features in their clients, and so does any decent network based cache system like the open source memcache product. Relying only on key equality for data retrevial makes it simple for the storage engine to distribute across multiple machines. Traditional SQL systems could also be used with a JDBC based spi implementation. Before submitting this change I have implemented six storage systems for the spi layer: * Apache HBase[1] * Apache Cassandra[2] * Google Bigtable[3] * an in-memory implementation for unit testing * a JDBC implementation for SQL * a generic cache provider that can ride on top of memcache All six systems came in with an spi layer around 1000 lines of code to implement the above 7 interfaces. This is a huge reduction in size compared to prior attempts to implement a new JGit storage layer. As this package shows, a complete JGit storage implementation is more than 17,000 lines of fairly complex code. A simple cache is provided in storage.dht.spi.cache. Implementers can use CacheDatabase to wrap any other type of Database and perform fast reads against a network based cache service, such as the open source memcached[4]. An implementation of CacheService must be provided to glue this spi onto the network cache. [1] https://github.com/spearce/jgit_hbase [2] https://github.com/spearce/jgit_cassandra [3] http://labs.google.com/papers/bigtable.html [4] http://memcached.org/ Change-Id: I0aa4072781f5ccc019ca421c036adff2c40c4295 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
maxObjectSizeLimit for receive-pack. ReceivePack (and PackParser) can be configured with the maxObjectSizeLimit in order to prevent users from pushing too large objects to Git. The limit check is applied to all object types although it is most likely that a BLOB will exceed the limit. In all cases the size of the object header is excluded from the object size which is checked against the limit as this is the size of which a BLOB object would take in the working tree when checked out as a file. When an object exceeds the maxObjectSizeLimit the receive-pack will abort immediately. Delta objects (both offset and ref delta) are also checked against the limit. However, for delta objects we will first check the size of the inflated delta block against the maxObjectSizeLimit and abort immediately if it exceeds the limit. In this case we even do not know the exact size of the resolved delta object but we assume it will be larger than the given maxObjectSizeLimit as delta is generally only chosen if the delta can copy more data from the base object than the delta needs to insert or needs to represent the copy ranges. Aborting early, in this case, avoids unnecessary inflating of the (huge) delta block. Unfortunately, it is too expensive (especially for a large delta) to compute SHA-1 of an object that causes the receive-pack to abort. This would decrease the value of this feature whose main purpose is to protect server resources from users pushing huge objects. Therefore we don't report the SHA-1 in the error message. Change-Id: I177ef24553faacda444ed5895e40ac8925ca0d1e Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
12 years ago
Store Git on any DHT jgit.storage.dht is a storage provider implementation for JGit that permits storing the Git repository in a distributed hashtable, NoSQL system, or other database. The actual underlying storage system is undefined, and can be plugged in by implementing 7 small interfaces: * Database * RepositoryIndexTable * RepositoryTable * RefTable * ChunkTable * ObjectIndexTable * WriteBuffer The storage provider interface tries to assume very little about the underlying storage system, and requires only three key features: * key -> value lookup (a hashtable is suitable) * atomic updates on single rows * asynchronous operations (Java's ExecutorService is easy to use) Most NoSQL database products offer all 3 of these features in their clients, and so does any decent network based cache system like the open source memcache product. Relying only on key equality for data retrevial makes it simple for the storage engine to distribute across multiple machines. Traditional SQL systems could also be used with a JDBC based spi implementation. Before submitting this change I have implemented six storage systems for the spi layer: * Apache HBase[1] * Apache Cassandra[2] * Google Bigtable[3] * an in-memory implementation for unit testing * a JDBC implementation for SQL * a generic cache provider that can ride on top of memcache All six systems came in with an spi layer around 1000 lines of code to implement the above 7 interfaces. This is a huge reduction in size compared to prior attempts to implement a new JGit storage layer. As this package shows, a complete JGit storage implementation is more than 17,000 lines of fairly complex code. A simple cache is provided in storage.dht.spi.cache. Implementers can use CacheDatabase to wrap any other type of Database and perform fast reads against a network based cache service, such as the open source memcached[4]. An implementation of CacheService must be provided to glue this spi onto the network cache. [1] https://github.com/spearce/jgit_hbase [2] https://github.com/spearce/jgit_cassandra [3] http://labs.google.com/papers/bigtable.html [4] http://memcached.org/ Change-Id: I0aa4072781f5ccc019ca421c036adff2c40c4295 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
Store Git on any DHT jgit.storage.dht is a storage provider implementation for JGit that permits storing the Git repository in a distributed hashtable, NoSQL system, or other database. The actual underlying storage system is undefined, and can be plugged in by implementing 7 small interfaces: * Database * RepositoryIndexTable * RepositoryTable * RefTable * ChunkTable * ObjectIndexTable * WriteBuffer The storage provider interface tries to assume very little about the underlying storage system, and requires only three key features: * key -> value lookup (a hashtable is suitable) * atomic updates on single rows * asynchronous operations (Java's ExecutorService is easy to use) Most NoSQL database products offer all 3 of these features in their clients, and so does any decent network based cache system like the open source memcache product. Relying only on key equality for data retrevial makes it simple for the storage engine to distribute across multiple machines. Traditional SQL systems could also be used with a JDBC based spi implementation. Before submitting this change I have implemented six storage systems for the spi layer: * Apache HBase[1] * Apache Cassandra[2] * Google Bigtable[3] * an in-memory implementation for unit testing * a JDBC implementation for SQL * a generic cache provider that can ride on top of memcache All six systems came in with an spi layer around 1000 lines of code to implement the above 7 interfaces. This is a huge reduction in size compared to prior attempts to implement a new JGit storage layer. As this package shows, a complete JGit storage implementation is more than 17,000 lines of fairly complex code. A simple cache is provided in storage.dht.spi.cache. Implementers can use CacheDatabase to wrap any other type of Database and perform fast reads against a network based cache service, such as the open source memcached[4]. An implementation of CacheService must be provided to glue this spi onto the network cache. [1] https://github.com/spearce/jgit_hbase [2] https://github.com/spearce/jgit_cassandra [3] http://labs.google.com/papers/bigtable.html [4] http://memcached.org/ Change-Id: I0aa4072781f5ccc019ca421c036adff2c40c4295 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
Store Git on any DHT jgit.storage.dht is a storage provider implementation for JGit that permits storing the Git repository in a distributed hashtable, NoSQL system, or other database. The actual underlying storage system is undefined, and can be plugged in by implementing 7 small interfaces: * Database * RepositoryIndexTable * RepositoryTable * RefTable * ChunkTable * ObjectIndexTable * WriteBuffer The storage provider interface tries to assume very little about the underlying storage system, and requires only three key features: * key -> value lookup (a hashtable is suitable) * atomic updates on single rows * asynchronous operations (Java's ExecutorService is easy to use) Most NoSQL database products offer all 3 of these features in their clients, and so does any decent network based cache system like the open source memcache product. Relying only on key equality for data retrevial makes it simple for the storage engine to distribute across multiple machines. Traditional SQL systems could also be used with a JDBC based spi implementation. Before submitting this change I have implemented six storage systems for the spi layer: * Apache HBase[1] * Apache Cassandra[2] * Google Bigtable[3] * an in-memory implementation for unit testing * a JDBC implementation for SQL * a generic cache provider that can ride on top of memcache All six systems came in with an spi layer around 1000 lines of code to implement the above 7 interfaces. This is a huge reduction in size compared to prior attempts to implement a new JGit storage layer. As this package shows, a complete JGit storage implementation is more than 17,000 lines of fairly complex code. A simple cache is provided in storage.dht.spi.cache. Implementers can use CacheDatabase to wrap any other type of Database and perform fast reads against a network based cache service, such as the open source memcached[4]. An implementation of CacheService must be provided to glue this spi onto the network cache. [1] https://github.com/spearce/jgit_hbase [2] https://github.com/spearce/jgit_cassandra [3] http://labs.google.com/papers/bigtable.html [4] http://memcached.org/ Change-Id: I0aa4072781f5ccc019ca421c036adff2c40c4295 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
Store Git on any DHT jgit.storage.dht is a storage provider implementation for JGit that permits storing the Git repository in a distributed hashtable, NoSQL system, or other database. The actual underlying storage system is undefined, and can be plugged in by implementing 7 small interfaces: * Database * RepositoryIndexTable * RepositoryTable * RefTable * ChunkTable * ObjectIndexTable * WriteBuffer The storage provider interface tries to assume very little about the underlying storage system, and requires only three key features: * key -> value lookup (a hashtable is suitable) * atomic updates on single rows * asynchronous operations (Java's ExecutorService is easy to use) Most NoSQL database products offer all 3 of these features in their clients, and so does any decent network based cache system like the open source memcache product. Relying only on key equality for data retrevial makes it simple for the storage engine to distribute across multiple machines. Traditional SQL systems could also be used with a JDBC based spi implementation. Before submitting this change I have implemented six storage systems for the spi layer: * Apache HBase[1] * Apache Cassandra[2] * Google Bigtable[3] * an in-memory implementation for unit testing * a JDBC implementation for SQL * a generic cache provider that can ride on top of memcache All six systems came in with an spi layer around 1000 lines of code to implement the above 7 interfaces. This is a huge reduction in size compared to prior attempts to implement a new JGit storage layer. As this package shows, a complete JGit storage implementation is more than 17,000 lines of fairly complex code. A simple cache is provided in storage.dht.spi.cache. Implementers can use CacheDatabase to wrap any other type of Database and perform fast reads against a network based cache service, such as the open source memcached[4]. An implementation of CacheService must be provided to glue this spi onto the network cache. [1] https://github.com/spearce/jgit_hbase [2] https://github.com/spearce/jgit_cassandra [3] http://labs.google.com/papers/bigtable.html [4] http://memcached.org/ Change-Id: I0aa4072781f5ccc019ca421c036adff2c40c4295 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
Store Git on any DHT jgit.storage.dht is a storage provider implementation for JGit that permits storing the Git repository in a distributed hashtable, NoSQL system, or other database. The actual underlying storage system is undefined, and can be plugged in by implementing 7 small interfaces: * Database * RepositoryIndexTable * RepositoryTable * RefTable * ChunkTable * ObjectIndexTable * WriteBuffer The storage provider interface tries to assume very little about the underlying storage system, and requires only three key features: * key -> value lookup (a hashtable is suitable) * atomic updates on single rows * asynchronous operations (Java's ExecutorService is easy to use) Most NoSQL database products offer all 3 of these features in their clients, and so does any decent network based cache system like the open source memcache product. Relying only on key equality for data retrevial makes it simple for the storage engine to distribute across multiple machines. Traditional SQL systems could also be used with a JDBC based spi implementation. Before submitting this change I have implemented six storage systems for the spi layer: * Apache HBase[1] * Apache Cassandra[2] * Google Bigtable[3] * an in-memory implementation for unit testing * a JDBC implementation for SQL * a generic cache provider that can ride on top of memcache All six systems came in with an spi layer around 1000 lines of code to implement the above 7 interfaces. This is a huge reduction in size compared to prior attempts to implement a new JGit storage layer. As this package shows, a complete JGit storage implementation is more than 17,000 lines of fairly complex code. A simple cache is provided in storage.dht.spi.cache. Implementers can use CacheDatabase to wrap any other type of Database and perform fast reads against a network based cache service, such as the open source memcached[4]. An implementation of CacheService must be provided to glue this spi onto the network cache. [1] https://github.com/spearce/jgit_hbase [2] https://github.com/spearce/jgit_cassandra [3] http://labs.google.com/papers/bigtable.html [4] http://memcached.org/ Change-Id: I0aa4072781f5ccc019ca421c036adff2c40c4295 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
ObjectIdOwnerMap: More lightweight map for ObjectIds OwnerMap is about 200 ms faster than SubclassMap, more friendly to the GC, and uses less storage: testing the "Counting objects" part of PackWriter on 1886362 objects: ObjectIdSubclassMap: load factor 50% table: 4194304 (wasted 2307942) ms spent 36998 36009 34795 34703 34941 35070 34284 34511 34638 34256 ms avg 34800 (last 9 runs) ObjectIdOwnerMap: load factor 100% table: 2097152 (wasted 210790) directory: 1024 ms spent 36842 35112 34922 34703 34580 34782 34165 34662 34314 34140 ms avg 34597 (last 9 runs) The major difference with OwnerMap is entries must extend from ObjectIdOwnerMap.Entry, where the OwnerMap has injected its own private "next" field into each object. This allows the OwnerMap to use a singly linked list for chaining collisions within a bucket. By putting collisions in a linked list, we gain the entire table back for the SHA-1 bits to index their own "private" slot. Unfortunately this means that each object can appear in at most ONE OwnerMap, as there is only one "next" field within the object instance to thread into the map. For types that are very object map heavy like RevWalk (entity RevObject) and PackWriter (entity ObjectToPack) this is sufficient, these entity types are only put into one map by their container. By introducing a new map type, we don't break existing applications that might be trying to use ObjectIdSubclassMap to track RevCommits they obtained from a RevWalk. The OwnerMap uses less memory. Each object uses 1 reference more (so we're up 1,886,362 references), but the table is 1/2 the size (2^20 rather than 2^21). The table itself wastes only 210,790 slots, rather than 2,307,942. So OwnerMap is wasting 200k fewer references. OwnerMap is more friendly to the GC, because it hardly ever generates garbage. As the map reaches its 100% load factor target, it doubles in size by allocating additional segment arrays of 2048 entries. (So the first grow allocates 1 segment, second 2 segments, third 4 segments, etc.) These segments are hooked into the pre-allocated directory of 1024 spaces. This permits the map to grow to 2 million objects before the directory itself has to grow. By using segments of 2048 entries, we are asking the GC to acquire 8,204 bytes in a 32 bit JVM. This is easier to satisfy then 2,307,942 bytes (for the 512k table that is just an intermediate step in the SubclassMap). By reusing the previously allocated segments (they are re-hashed in-place) we don't release any memory during a table grow. When the directory grows, it does so by discarding the old one and using one that is 4x larger (so the directory goes to 4096 entries on its first grow). A directory of size 4096 can handle up to 8 millon objects. The second directory grow (16384) goes to 33 million objects. At that point we're starting to really push the limits of the JVM heap, but at least its many small arrays. Previously SubclassMap would need a table of 67108864 entries to handle that object count, which needs a single contiguous allocation of 256 MiB. That's hard to come by in a 32 bit JVM. Instead OwnerMap uses 8192 arrays of about 8 KiB each. This is much easier to fit into a fragmented heap. Change-Id: Ia4acf5cfbf7e9b71bc7faa0db9060f6a969c0c50 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
13 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689
  1. /*
  2. * Copyright (C) 2008-2011, Google Inc.
  3. * Copyright (C) 2007-2008, Robin Rosenberg <robin.rosenberg@dewire.com>
  4. * Copyright (C) 2008, Shawn O. Pearce <spearce@spearce.org>
  5. * and other copyright owners as documented in the project's IP log.
  6. *
  7. * This program and the accompanying materials are made available
  8. * under the terms of the Eclipse Distribution License v1.0 which
  9. * accompanies this distribution, is reproduced below, and is
  10. * available at http://www.eclipse.org/org/documents/edl-v10.php
  11. *
  12. * All rights reserved.
  13. *
  14. * Redistribution and use in source and binary forms, with or
  15. * without modification, are permitted provided that the following
  16. * conditions are met:
  17. *
  18. * - Redistributions of source code must retain the above copyright
  19. * notice, this list of conditions and the following disclaimer.
  20. *
  21. * - Redistributions in binary form must reproduce the above
  22. * copyright notice, this list of conditions and the following
  23. * disclaimer in the documentation and/or other materials provided
  24. * with the distribution.
  25. *
  26. * - Neither the name of the Eclipse Foundation, Inc. nor the
  27. * names of its contributors may be used to endorse or promote
  28. * products derived from this software without specific prior
  29. * written permission.
  30. *
  31. * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
  32. * CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
  33. * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  34. * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  35. * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
  36. * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  37. * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  38. * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  39. * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  40. * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  41. * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  42. * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
  43. * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  44. */
  45. package org.eclipse.jgit.transport;
  46. import java.io.EOFException;
  47. import java.io.IOException;
  48. import java.io.InputStream;
  49. import java.security.MessageDigest;
  50. import java.text.MessageFormat;
  51. import java.util.ArrayList;
  52. import java.util.Arrays;
  53. import java.util.Comparator;
  54. import java.util.List;
  55. import java.util.concurrent.TimeUnit;
  56. import java.util.zip.DataFormatException;
  57. import java.util.zip.Inflater;
  58. import org.eclipse.jgit.errors.CorruptObjectException;
  59. import org.eclipse.jgit.errors.MissingObjectException;
  60. import org.eclipse.jgit.errors.TooLargeObjectInPackException;
  61. import org.eclipse.jgit.internal.JGitText;
  62. import org.eclipse.jgit.lib.AnyObjectId;
  63. import org.eclipse.jgit.lib.BatchingProgressMonitor;
  64. import org.eclipse.jgit.lib.Constants;
  65. import org.eclipse.jgit.lib.InflaterCache;
  66. import org.eclipse.jgit.lib.MutableObjectId;
  67. import org.eclipse.jgit.lib.NullProgressMonitor;
  68. import org.eclipse.jgit.lib.ObjectChecker;
  69. import org.eclipse.jgit.lib.ObjectDatabase;
  70. import org.eclipse.jgit.lib.ObjectId;
  71. import org.eclipse.jgit.lib.ObjectIdOwnerMap;
  72. import org.eclipse.jgit.lib.ObjectIdSubclassMap;
  73. import org.eclipse.jgit.lib.ObjectInserter;
  74. import org.eclipse.jgit.lib.ObjectLoader;
  75. import org.eclipse.jgit.lib.ObjectReader;
  76. import org.eclipse.jgit.lib.ObjectStream;
  77. import org.eclipse.jgit.lib.ProgressMonitor;
  78. import org.eclipse.jgit.storage.file.PackLock;
  79. import org.eclipse.jgit.storage.pack.BinaryDelta;
  80. import org.eclipse.jgit.util.BlockList;
  81. import org.eclipse.jgit.util.IO;
  82. import org.eclipse.jgit.util.NB;
  83. /**
  84. * Parses a pack stream and imports it for an {@link ObjectInserter}.
  85. * <p>
  86. * Applications can acquire an instance of a parser from ObjectInserter's
  87. * {@link ObjectInserter#newPackParser(InputStream)} method.
  88. * <p>
  89. * Implementations of {@link ObjectInserter} should subclass this type and
  90. * provide their own logic for the various {@code on*()} event methods declared
  91. * to be abstract.
  92. */
  93. public abstract class PackParser {
  94. /** Size of the internal stream buffer. */
  95. private static final int BUFFER_SIZE = 8192;
  96. /** Location data is being obtained from. */
  97. public static enum Source {
  98. /** Data is read from the incoming stream. */
  99. INPUT,
  100. /** Data is read back from the database's buffers. */
  101. DATABASE;
  102. }
  103. /** Object database used for loading existing objects. */
  104. private final ObjectDatabase objectDatabase;
  105. private InflaterStream inflater;
  106. private byte[] tempBuffer;
  107. private byte[] hdrBuf;
  108. private final MessageDigest objectDigest;
  109. private final MutableObjectId tempObjectId;
  110. private InputStream in;
  111. private byte[] buf;
  112. /** Position in the input stream of {@code buf[0]}. */
  113. private long bBase;
  114. private int bOffset;
  115. private int bAvail;
  116. private ObjectChecker objCheck;
  117. private boolean allowThin;
  118. private boolean checkObjectCollisions;
  119. private boolean needBaseObjectIds;
  120. private boolean checkEofAfterPackFooter;
  121. private long objectCount;
  122. private PackedObjectInfo[] entries;
  123. /**
  124. * Every object contained within the incoming pack.
  125. * <p>
  126. * This is a subset of {@link #entries}, as thin packs can add additional
  127. * objects to {@code entries} by copying already existing objects from the
  128. * repository onto the end of the thin pack to make it self-contained.
  129. */
  130. private ObjectIdSubclassMap<ObjectId> newObjectIds;
  131. private int deltaCount;
  132. private int entryCount;
  133. private ObjectIdOwnerMap<DeltaChain> baseById;
  134. /**
  135. * Objects referenced by their name from deltas, that aren't in this pack.
  136. * <p>
  137. * This is the set of objects that were copied onto the end of this pack to
  138. * make it complete. These objects were not transmitted by the remote peer,
  139. * but instead were assumed to already exist in the local repository.
  140. */
  141. private ObjectIdSubclassMap<ObjectId> baseObjectIds;
  142. private LongMap<UnresolvedDelta> baseByPos;
  143. /** Blobs whose contents need to be double-checked after indexing. */
  144. private BlockList<PackedObjectInfo> deferredCheckBlobs;
  145. private MessageDigest packDigest;
  146. private ObjectReader readCurs;
  147. /** Message to protect the pack data from garbage collection. */
  148. private String lockMessage;
  149. /** Git object size limit */
  150. private long maxObjectSizeLimit;
  151. /**
  152. * Initialize a pack parser.
  153. *
  154. * @param odb
  155. * database the parser will write its objects into.
  156. * @param src
  157. * the stream the parser will read.
  158. */
  159. protected PackParser(final ObjectDatabase odb, final InputStream src) {
  160. objectDatabase = odb.newCachedDatabase();
  161. in = src;
  162. inflater = new InflaterStream();
  163. readCurs = objectDatabase.newReader();
  164. buf = new byte[BUFFER_SIZE];
  165. tempBuffer = new byte[BUFFER_SIZE];
  166. hdrBuf = new byte[64];
  167. objectDigest = Constants.newMessageDigest();
  168. tempObjectId = new MutableObjectId();
  169. packDigest = Constants.newMessageDigest();
  170. checkObjectCollisions = true;
  171. }
  172. /** @return true if a thin pack (missing base objects) is permitted. */
  173. public boolean isAllowThin() {
  174. return allowThin;
  175. }
  176. /**
  177. * Configure this index pack instance to allow a thin pack.
  178. * <p>
  179. * Thin packs are sometimes used during network transfers to allow a delta
  180. * to be sent without a base object. Such packs are not permitted on disk.
  181. *
  182. * @param allow
  183. * true to enable a thin pack.
  184. */
  185. public void setAllowThin(final boolean allow) {
  186. allowThin = allow;
  187. }
  188. /** @return if true received objects are verified to prevent collisions. */
  189. public boolean isCheckObjectCollisions() {
  190. return checkObjectCollisions;
  191. }
  192. /**
  193. * Enable checking for collisions with existing objects.
  194. * <p>
  195. * By default PackParser looks for each received object in the repository.
  196. * If the object already exists, the existing object is compared
  197. * byte-for-byte with the newly received copy to ensure they are identical.
  198. * The receive is aborted with an exception if any byte differs. This check
  199. * is necessary to prevent an evil attacker from supplying a replacement
  200. * object into this repository in the event that a discovery enabling SHA-1
  201. * collisions is made.
  202. * <p>
  203. * This check may be very costly to perform, and some repositories may have
  204. * other ways to segregate newly received object data. The check is enabled
  205. * by default, but can be explicitly disabled if the implementation can
  206. * provide the same guarantee, or is willing to accept the risks associated
  207. * with bypassing the check.
  208. *
  209. * @param check
  210. * true to enable collision checking (strongly encouraged).
  211. */
  212. public void setCheckObjectCollisions(boolean check) {
  213. checkObjectCollisions = check;
  214. }
  215. /**
  216. * Configure this index pack instance to keep track of new objects.
  217. * <p>
  218. * By default an index pack doesn't save the new objects that were created
  219. * when it was instantiated. Setting this flag to {@code true} allows the
  220. * caller to use {@link #getNewObjectIds()} to retrieve that list.
  221. *
  222. * @param b
  223. * {@code true} to enable keeping track of new objects.
  224. */
  225. public void setNeedNewObjectIds(boolean b) {
  226. if (b)
  227. newObjectIds = new ObjectIdSubclassMap<ObjectId>();
  228. else
  229. newObjectIds = null;
  230. }
  231. private boolean needNewObjectIds() {
  232. return newObjectIds != null;
  233. }
  234. /**
  235. * Configure this index pack instance to keep track of the objects assumed
  236. * for delta bases.
  237. * <p>
  238. * By default an index pack doesn't save the objects that were used as delta
  239. * bases. Setting this flag to {@code true} will allow the caller to use
  240. * {@link #getBaseObjectIds()} to retrieve that list.
  241. *
  242. * @param b
  243. * {@code true} to enable keeping track of delta bases.
  244. */
  245. public void setNeedBaseObjectIds(boolean b) {
  246. this.needBaseObjectIds = b;
  247. }
  248. /** @return true if the EOF should be read from the input after the footer. */
  249. public boolean isCheckEofAfterPackFooter() {
  250. return checkEofAfterPackFooter;
  251. }
  252. /**
  253. * Ensure EOF is read from the input stream after the footer.
  254. *
  255. * @param b
  256. * true if the EOF should be read; false if it is not checked.
  257. */
  258. public void setCheckEofAfterPackFooter(boolean b) {
  259. checkEofAfterPackFooter = b;
  260. }
  261. /** @return the new objects that were sent by the user */
  262. public ObjectIdSubclassMap<ObjectId> getNewObjectIds() {
  263. if (newObjectIds != null)
  264. return newObjectIds;
  265. return new ObjectIdSubclassMap<ObjectId>();
  266. }
  267. /** @return set of objects the incoming pack assumed for delta purposes */
  268. public ObjectIdSubclassMap<ObjectId> getBaseObjectIds() {
  269. if (baseObjectIds != null)
  270. return baseObjectIds;
  271. return new ObjectIdSubclassMap<ObjectId>();
  272. }
  273. /**
  274. * Configure the checker used to validate received objects.
  275. * <p>
  276. * Usually object checking isn't necessary, as Git implementations only
  277. * create valid objects in pack files. However, additional checking may be
  278. * useful if processing data from an untrusted source.
  279. *
  280. * @param oc
  281. * the checker instance; null to disable object checking.
  282. */
  283. public void setObjectChecker(final ObjectChecker oc) {
  284. objCheck = oc;
  285. }
  286. /**
  287. * Configure the checker used to validate received objects.
  288. * <p>
  289. * Usually object checking isn't necessary, as Git implementations only
  290. * create valid objects in pack files. However, additional checking may be
  291. * useful if processing data from an untrusted source.
  292. * <p>
  293. * This is shorthand for:
  294. *
  295. * <pre>
  296. * setObjectChecker(on ? new ObjectChecker() : null);
  297. * </pre>
  298. *
  299. * @param on
  300. * true to enable the default checker; false to disable it.
  301. */
  302. public void setObjectChecking(final boolean on) {
  303. setObjectChecker(on ? new ObjectChecker() : null);
  304. }
  305. /** @return the message to record with the pack lock. */
  306. public String getLockMessage() {
  307. return lockMessage;
  308. }
  309. /**
  310. * Set the lock message for the incoming pack data.
  311. *
  312. * @param msg
  313. * if not null, the message to associate with the incoming data
  314. * while it is locked to prevent garbage collection.
  315. */
  316. public void setLockMessage(String msg) {
  317. lockMessage = msg;
  318. }
  319. /**
  320. * Set the maximum allowed Git object size.
  321. * <p>
  322. * If an object is larger than the given size the pack-parsing will throw an
  323. * exception aborting the parsing.
  324. *
  325. * @param limit
  326. * the Git object size limit. If zero then there is not limit.
  327. */
  328. public void setMaxObjectSizeLimit(long limit) {
  329. maxObjectSizeLimit = limit;
  330. }
  331. /**
  332. * Get the number of objects in the stream.
  333. * <p>
  334. * The object count is only available after {@link #parse(ProgressMonitor)}
  335. * has returned. The count may have been increased if the stream was a thin
  336. * pack, and missing bases objects were appending onto it by the subclass.
  337. *
  338. * @return number of objects parsed out of the stream.
  339. */
  340. public int getObjectCount() {
  341. return entryCount;
  342. }
  343. /***
  344. * Get the information about the requested object.
  345. * <p>
  346. * The object information is only available after
  347. * {@link #parse(ProgressMonitor)} has returned.
  348. *
  349. * @param nth
  350. * index of the object in the stream. Must be between 0 and
  351. * {@link #getObjectCount()}-1.
  352. * @return the object information.
  353. */
  354. public PackedObjectInfo getObject(int nth) {
  355. return entries[nth];
  356. }
  357. /**
  358. * Get all of the objects, sorted by their name.
  359. * <p>
  360. * The object information is only available after
  361. * {@link #parse(ProgressMonitor)} has returned.
  362. * <p>
  363. * To maintain lower memory usage and good runtime performance, this method
  364. * sorts the objects in-place and therefore impacts the ordering presented
  365. * by {@link #getObject(int)}.
  366. *
  367. * @param cmp
  368. * comparison function, if null objects are stored by ObjectId.
  369. * @return sorted list of objects in this pack stream.
  370. */
  371. public List<PackedObjectInfo> getSortedObjectList(
  372. Comparator<PackedObjectInfo> cmp) {
  373. Arrays.sort(entries, 0, entryCount, cmp);
  374. List<PackedObjectInfo> list = Arrays.asList(entries);
  375. if (entryCount < entries.length)
  376. list = list.subList(0, entryCount);
  377. return list;
  378. }
  379. /**
  380. * Parse the pack stream.
  381. *
  382. * @param progress
  383. * callback to provide progress feedback during parsing. If null,
  384. * {@link NullProgressMonitor} will be used.
  385. * @return the pack lock, if one was requested by setting
  386. * {@link #setLockMessage(String)}.
  387. * @throws IOException
  388. * the stream is malformed, or contains corrupt objects.
  389. */
  390. public final PackLock parse(ProgressMonitor progress) throws IOException {
  391. return parse(progress, progress);
  392. }
  393. /**
  394. * Parse the pack stream.
  395. *
  396. * @param receiving
  397. * receives progress feedback during the initial receiving
  398. * objects phase. If null, {@link NullProgressMonitor} will be
  399. * used.
  400. * @param resolving
  401. * receives progress feedback during the resolving objects phase.
  402. * @return the pack lock, if one was requested by setting
  403. * {@link #setLockMessage(String)}.
  404. * @throws IOException
  405. * the stream is malformed, or contains corrupt objects.
  406. */
  407. public PackLock parse(ProgressMonitor receiving, ProgressMonitor resolving)
  408. throws IOException {
  409. if (receiving == null)
  410. receiving = NullProgressMonitor.INSTANCE;
  411. if (resolving == null)
  412. resolving = NullProgressMonitor.INSTANCE;
  413. if (receiving == resolving)
  414. receiving.start(2 /* tasks */);
  415. try {
  416. readPackHeader();
  417. entries = new PackedObjectInfo[(int) objectCount];
  418. baseById = new ObjectIdOwnerMap<DeltaChain>();
  419. baseByPos = new LongMap<UnresolvedDelta>();
  420. deferredCheckBlobs = new BlockList<PackedObjectInfo>();
  421. receiving.beginTask(JGitText.get().receivingObjects,
  422. (int) objectCount);
  423. try {
  424. for (int done = 0; done < objectCount; done++) {
  425. indexOneObject();
  426. receiving.update(1);
  427. if (receiving.isCancelled())
  428. throw new IOException(JGitText.get().downloadCancelled);
  429. }
  430. readPackFooter();
  431. endInput();
  432. } finally {
  433. receiving.endTask();
  434. }
  435. if (!deferredCheckBlobs.isEmpty())
  436. doDeferredCheckBlobs();
  437. if (deltaCount > 0) {
  438. if (resolving instanceof BatchingProgressMonitor) {
  439. ((BatchingProgressMonitor) resolving).setDelayStart(
  440. 1000,
  441. TimeUnit.MILLISECONDS);
  442. }
  443. resolving.beginTask(JGitText.get().resolvingDeltas, deltaCount);
  444. resolveDeltas(resolving);
  445. if (entryCount < objectCount) {
  446. if (!isAllowThin()) {
  447. throw new IOException(MessageFormat.format(JGitText
  448. .get().packHasUnresolvedDeltas,
  449. (objectCount - entryCount)));
  450. }
  451. resolveDeltasWithExternalBases(resolving);
  452. if (entryCount < objectCount) {
  453. throw new IOException(MessageFormat.format(JGitText
  454. .get().packHasUnresolvedDeltas,
  455. (objectCount - entryCount)));
  456. }
  457. }
  458. resolving.endTask();
  459. }
  460. packDigest = null;
  461. baseById = null;
  462. baseByPos = null;
  463. } finally {
  464. try {
  465. if (readCurs != null)
  466. readCurs.release();
  467. } finally {
  468. readCurs = null;
  469. }
  470. try {
  471. inflater.release();
  472. } finally {
  473. inflater = null;
  474. }
  475. }
  476. return null; // By default there is no locking.
  477. }
  478. private void resolveDeltas(final ProgressMonitor progress)
  479. throws IOException {
  480. final int last = entryCount;
  481. for (int i = 0; i < last; i++) {
  482. resolveDeltas(entries[i], progress);
  483. if (progress.isCancelled())
  484. throw new IOException(
  485. JGitText.get().downloadCancelledDuringIndexing);
  486. }
  487. }
  488. private void resolveDeltas(final PackedObjectInfo oe,
  489. ProgressMonitor progress) throws IOException {
  490. UnresolvedDelta children = firstChildOf(oe);
  491. if (children == null)
  492. return;
  493. DeltaVisit visit = new DeltaVisit();
  494. visit.nextChild = children;
  495. ObjectTypeAndSize info = openDatabase(oe, new ObjectTypeAndSize());
  496. switch (info.type) {
  497. case Constants.OBJ_COMMIT:
  498. case Constants.OBJ_TREE:
  499. case Constants.OBJ_BLOB:
  500. case Constants.OBJ_TAG:
  501. visit.data = inflateAndReturn(Source.DATABASE, info.size);
  502. visit.id = oe;
  503. break;
  504. default:
  505. throw new IOException(MessageFormat.format(
  506. JGitText.get().unknownObjectType, info.type));
  507. }
  508. if (!checkCRC(oe.getCRC())) {
  509. throw new IOException(MessageFormat.format(
  510. JGitText.get().corruptionDetectedReReadingAt, oe
  511. .getOffset()));
  512. }
  513. resolveDeltas(visit.next(), info.type, info, progress);
  514. }
  515. private void resolveDeltas(DeltaVisit visit, final int type,
  516. ObjectTypeAndSize info, ProgressMonitor progress)
  517. throws IOException {
  518. do {
  519. progress.update(1);
  520. info = openDatabase(visit.delta, info);
  521. switch (info.type) {
  522. case Constants.OBJ_OFS_DELTA:
  523. case Constants.OBJ_REF_DELTA:
  524. break;
  525. default:
  526. throw new IOException(MessageFormat.format(
  527. JGitText.get().unknownObjectType, info.type));
  528. }
  529. byte[] delta = inflateAndReturn(Source.DATABASE, info.size);
  530. checkIfTooLarge(type, BinaryDelta.getResultSize(delta));
  531. visit.data = BinaryDelta.apply(visit.parent.data, delta);
  532. delta = null;
  533. if (!checkCRC(visit.delta.crc))
  534. throw new IOException(MessageFormat.format(
  535. JGitText.get().corruptionDetectedReReadingAt,
  536. visit.delta.position));
  537. objectDigest.update(Constants.encodedTypeString(type));
  538. objectDigest.update((byte) ' ');
  539. objectDigest.update(Constants.encodeASCII(visit.data.length));
  540. objectDigest.update((byte) 0);
  541. objectDigest.update(visit.data);
  542. tempObjectId.fromRaw(objectDigest.digest(), 0);
  543. verifySafeObject(tempObjectId, type, visit.data);
  544. PackedObjectInfo oe;
  545. oe = newInfo(tempObjectId, visit.delta, visit.parent.id);
  546. oe.setOffset(visit.delta.position);
  547. onInflatedObjectData(oe, type, visit.data);
  548. addObjectAndTrack(oe);
  549. visit.id = oe;
  550. visit.nextChild = firstChildOf(oe);
  551. visit = visit.next();
  552. } while (visit != null);
  553. }
  554. private final void checkIfTooLarge(int typeCode, long size)
  555. throws IOException {
  556. if (0 < maxObjectSizeLimit && maxObjectSizeLimit < size)
  557. switch (typeCode) {
  558. case Constants.OBJ_COMMIT:
  559. case Constants.OBJ_TREE:
  560. case Constants.OBJ_BLOB:
  561. case Constants.OBJ_TAG:
  562. throw new TooLargeObjectInPackException(size, maxObjectSizeLimit);
  563. case Constants.OBJ_OFS_DELTA:
  564. case Constants.OBJ_REF_DELTA:
  565. throw new TooLargeObjectInPackException(maxObjectSizeLimit);
  566. default:
  567. throw new IOException(MessageFormat.format(
  568. JGitText.get().unknownObjectType, typeCode));
  569. }
  570. }
  571. /**
  572. * Read the header of the current object.
  573. * <p>
  574. * After the header has been parsed, this method automatically invokes
  575. * {@link #onObjectHeader(Source, byte[], int, int)} to allow the
  576. * implementation to update its internal checksums for the bytes read.
  577. * <p>
  578. * When this method returns the database will be positioned on the first
  579. * byte of the deflated data stream.
  580. *
  581. * @param info
  582. * the info object to populate.
  583. * @return {@code info}, after populating.
  584. * @throws IOException
  585. * the size cannot be read.
  586. */
  587. protected ObjectTypeAndSize readObjectHeader(ObjectTypeAndSize info)
  588. throws IOException {
  589. int hdrPtr = 0;
  590. int c = readFrom(Source.DATABASE);
  591. hdrBuf[hdrPtr++] = (byte) c;
  592. info.type = (c >> 4) & 7;
  593. long sz = c & 15;
  594. int shift = 4;
  595. while ((c & 0x80) != 0) {
  596. c = readFrom(Source.DATABASE);
  597. hdrBuf[hdrPtr++] = (byte) c;
  598. sz += ((long) (c & 0x7f)) << shift;
  599. shift += 7;
  600. }
  601. info.size = sz;
  602. switch (info.type) {
  603. case Constants.OBJ_COMMIT:
  604. case Constants.OBJ_TREE:
  605. case Constants.OBJ_BLOB:
  606. case Constants.OBJ_TAG:
  607. onObjectHeader(Source.DATABASE, hdrBuf, 0, hdrPtr);
  608. break;
  609. case Constants.OBJ_OFS_DELTA:
  610. c = readFrom(Source.DATABASE);
  611. hdrBuf[hdrPtr++] = (byte) c;
  612. while ((c & 128) != 0) {
  613. c = readFrom(Source.DATABASE);
  614. hdrBuf[hdrPtr++] = (byte) c;
  615. }
  616. onObjectHeader(Source.DATABASE, hdrBuf, 0, hdrPtr);
  617. break;
  618. case Constants.OBJ_REF_DELTA:
  619. System.arraycopy(buf, fill(Source.DATABASE, 20), hdrBuf, hdrPtr, 20);
  620. hdrPtr += 20;
  621. use(20);
  622. onObjectHeader(Source.DATABASE, hdrBuf, 0, hdrPtr);
  623. break;
  624. default:
  625. throw new IOException(MessageFormat.format(
  626. JGitText.get().unknownObjectType, info.type));
  627. }
  628. return info;
  629. }
  630. private UnresolvedDelta removeBaseById(final AnyObjectId id) {
  631. final DeltaChain d = baseById.get(id);
  632. return d != null ? d.remove() : null;
  633. }
  634. private static UnresolvedDelta reverse(UnresolvedDelta c) {
  635. UnresolvedDelta tail = null;
  636. while (c != null) {
  637. final UnresolvedDelta n = c.next;
  638. c.next = tail;
  639. tail = c;
  640. c = n;
  641. }
  642. return tail;
  643. }
  644. private UnresolvedDelta firstChildOf(PackedObjectInfo oe) {
  645. UnresolvedDelta a = reverse(removeBaseById(oe));
  646. UnresolvedDelta b = reverse(baseByPos.remove(oe.getOffset()));
  647. if (a == null)
  648. return b;
  649. if (b == null)
  650. return a;
  651. UnresolvedDelta first = null;
  652. UnresolvedDelta last = null;
  653. while (a != null || b != null) {
  654. UnresolvedDelta curr;
  655. if (b == null || (a != null && a.position < b.position)) {
  656. curr = a;
  657. a = a.next;
  658. } else {
  659. curr = b;
  660. b = b.next;
  661. }
  662. if (last != null)
  663. last.next = curr;
  664. else
  665. first = curr;
  666. last = curr;
  667. curr.next = null;
  668. }
  669. return first;
  670. }
  671. private void resolveDeltasWithExternalBases(final ProgressMonitor progress)
  672. throws IOException {
  673. growEntries(baseById.size());
  674. if (needBaseObjectIds)
  675. baseObjectIds = new ObjectIdSubclassMap<ObjectId>();
  676. final List<DeltaChain> missing = new ArrayList<DeltaChain>(64);
  677. for (final DeltaChain baseId : baseById) {
  678. if (baseId.head == null)
  679. continue;
  680. if (needBaseObjectIds)
  681. baseObjectIds.add(baseId);
  682. final ObjectLoader ldr;
  683. try {
  684. ldr = readCurs.open(baseId);
  685. } catch (MissingObjectException notFound) {
  686. missing.add(baseId);
  687. continue;
  688. }
  689. final DeltaVisit visit = new DeltaVisit();
  690. visit.data = ldr.getCachedBytes(Integer.MAX_VALUE);
  691. visit.id = baseId;
  692. final int typeCode = ldr.getType();
  693. final PackedObjectInfo oe = newInfo(baseId, null, null);
  694. if (onAppendBase(typeCode, visit.data, oe))
  695. entries[entryCount++] = oe;
  696. visit.nextChild = firstChildOf(oe);
  697. resolveDeltas(visit.next(), typeCode,
  698. new ObjectTypeAndSize(), progress);
  699. if (progress.isCancelled())
  700. throw new IOException(
  701. JGitText.get().downloadCancelledDuringIndexing);
  702. }
  703. for (final DeltaChain base : missing) {
  704. if (base.head != null)
  705. throw new MissingObjectException(base, "delta base");
  706. }
  707. onEndThinPack();
  708. }
  709. private void growEntries(int extraObjects) {
  710. final PackedObjectInfo[] ne;
  711. ne = new PackedObjectInfo[(int) objectCount + extraObjects];
  712. System.arraycopy(entries, 0, ne, 0, entryCount);
  713. entries = ne;
  714. }
  715. private void readPackHeader() throws IOException {
  716. final int hdrln = Constants.PACK_SIGNATURE.length + 4 + 4;
  717. final int p = fill(Source.INPUT, hdrln);
  718. for (int k = 0; k < Constants.PACK_SIGNATURE.length; k++)
  719. if (buf[p + k] != Constants.PACK_SIGNATURE[k])
  720. throw new IOException(JGitText.get().notAPACKFile);
  721. final long vers = NB.decodeUInt32(buf, p + 4);
  722. if (vers != 2 && vers != 3)
  723. throw new IOException(MessageFormat.format(
  724. JGitText.get().unsupportedPackVersion, vers));
  725. objectCount = NB.decodeUInt32(buf, p + 8);
  726. use(hdrln);
  727. onPackHeader(objectCount);
  728. }
  729. private void readPackFooter() throws IOException {
  730. sync();
  731. final byte[] actHash = packDigest.digest();
  732. final int c = fill(Source.INPUT, 20);
  733. final byte[] srcHash = new byte[20];
  734. System.arraycopy(buf, c, srcHash, 0, 20);
  735. use(20);
  736. // The input stream should be at EOF at this point. We do not support
  737. // yielding back any remaining buffered data after the pack footer, so
  738. // protocols that embed a pack stream are required to either end their
  739. // stream with the pack, or embed the pack with a framing system like
  740. // the SideBandInputStream does.
  741. if (bAvail != 0)
  742. throw new CorruptObjectException(MessageFormat.format(
  743. JGitText.get().expectedEOFReceived,
  744. "\\x" + Integer.toHexString(buf[bOffset] & 0xff)));
  745. if (isCheckEofAfterPackFooter()) {
  746. int eof = in.read();
  747. if (0 <= eof)
  748. throw new CorruptObjectException(MessageFormat.format(
  749. JGitText.get().expectedEOFReceived,
  750. "\\x" + Integer.toHexString(eof)));
  751. }
  752. if (!Arrays.equals(actHash, srcHash))
  753. throw new CorruptObjectException(
  754. JGitText.get().corruptObjectPackfileChecksumIncorrect);
  755. onPackFooter(srcHash);
  756. }
  757. // Cleanup all resources associated with our input parsing.
  758. private void endInput() {
  759. in = null;
  760. }
  761. // Read one entire object or delta from the input.
  762. private void indexOneObject() throws IOException {
  763. final long streamPosition = streamPosition();
  764. int hdrPtr = 0;
  765. int c = readFrom(Source.INPUT);
  766. hdrBuf[hdrPtr++] = (byte) c;
  767. final int typeCode = (c >> 4) & 7;
  768. long sz = c & 15;
  769. int shift = 4;
  770. while ((c & 0x80) != 0) {
  771. c = readFrom(Source.INPUT);
  772. hdrBuf[hdrPtr++] = (byte) c;
  773. sz += ((long) (c & 0x7f)) << shift;
  774. shift += 7;
  775. }
  776. checkIfTooLarge(typeCode, sz);
  777. switch (typeCode) {
  778. case Constants.OBJ_COMMIT:
  779. case Constants.OBJ_TREE:
  780. case Constants.OBJ_BLOB:
  781. case Constants.OBJ_TAG:
  782. onBeginWholeObject(streamPosition, typeCode, sz);
  783. onObjectHeader(Source.INPUT, hdrBuf, 0, hdrPtr);
  784. whole(streamPosition, typeCode, sz);
  785. break;
  786. case Constants.OBJ_OFS_DELTA: {
  787. c = readFrom(Source.INPUT);
  788. hdrBuf[hdrPtr++] = (byte) c;
  789. long ofs = c & 127;
  790. while ((c & 128) != 0) {
  791. ofs += 1;
  792. c = readFrom(Source.INPUT);
  793. hdrBuf[hdrPtr++] = (byte) c;
  794. ofs <<= 7;
  795. ofs += (c & 127);
  796. }
  797. final long base = streamPosition - ofs;
  798. onBeginOfsDelta(streamPosition, base, sz);
  799. onObjectHeader(Source.INPUT, hdrBuf, 0, hdrPtr);
  800. inflateAndSkip(Source.INPUT, sz);
  801. UnresolvedDelta n = onEndDelta();
  802. n.position = streamPosition;
  803. n.next = baseByPos.put(base, n);
  804. deltaCount++;
  805. break;
  806. }
  807. case Constants.OBJ_REF_DELTA: {
  808. c = fill(Source.INPUT, 20);
  809. final ObjectId base = ObjectId.fromRaw(buf, c);
  810. System.arraycopy(buf, c, hdrBuf, hdrPtr, 20);
  811. hdrPtr += 20;
  812. use(20);
  813. DeltaChain r = baseById.get(base);
  814. if (r == null) {
  815. r = new DeltaChain(base);
  816. baseById.add(r);
  817. }
  818. onBeginRefDelta(streamPosition, base, sz);
  819. onObjectHeader(Source.INPUT, hdrBuf, 0, hdrPtr);
  820. inflateAndSkip(Source.INPUT, sz);
  821. UnresolvedDelta n = onEndDelta();
  822. n.position = streamPosition;
  823. r.add(n);
  824. deltaCount++;
  825. break;
  826. }
  827. default:
  828. throw new IOException(MessageFormat.format(
  829. JGitText.get().unknownObjectType, typeCode));
  830. }
  831. }
  832. private void whole(final long pos, final int type, final long sz)
  833. throws IOException {
  834. objectDigest.update(Constants.encodedTypeString(type));
  835. objectDigest.update((byte) ' ');
  836. objectDigest.update(Constants.encodeASCII(sz));
  837. objectDigest.update((byte) 0);
  838. final byte[] data;
  839. boolean checkContentLater = false;
  840. if (type == Constants.OBJ_BLOB) {
  841. byte[] readBuffer = buffer();
  842. InputStream inf = inflate(Source.INPUT, sz);
  843. long cnt = 0;
  844. while (cnt < sz) {
  845. int r = inf.read(readBuffer);
  846. if (r <= 0)
  847. break;
  848. objectDigest.update(readBuffer, 0, r);
  849. cnt += r;
  850. }
  851. inf.close();
  852. tempObjectId.fromRaw(objectDigest.digest(), 0);
  853. checkContentLater = isCheckObjectCollisions()
  854. && readCurs.has(tempObjectId);
  855. data = null;
  856. } else {
  857. data = inflateAndReturn(Source.INPUT, sz);
  858. objectDigest.update(data);
  859. tempObjectId.fromRaw(objectDigest.digest(), 0);
  860. verifySafeObject(tempObjectId, type, data);
  861. }
  862. PackedObjectInfo obj = newInfo(tempObjectId, null, null);
  863. obj.setOffset(pos);
  864. onEndWholeObject(obj);
  865. if (data != null)
  866. onInflatedObjectData(obj, type, data);
  867. addObjectAndTrack(obj);
  868. if (checkContentLater)
  869. deferredCheckBlobs.add(obj);
  870. }
  871. private void verifySafeObject(final AnyObjectId id, final int type,
  872. final byte[] data) throws IOException {
  873. if (objCheck != null) {
  874. try {
  875. objCheck.check(type, data);
  876. } catch (CorruptObjectException e) {
  877. throw new IOException(MessageFormat.format(
  878. JGitText.get().invalidObject, Constants
  879. .typeString(type), id.name(), e.getMessage()));
  880. }
  881. }
  882. if (isCheckObjectCollisions()) {
  883. try {
  884. final ObjectLoader ldr = readCurs.open(id, type);
  885. final byte[] existingData = ldr.getCachedBytes(data.length);
  886. if (!Arrays.equals(data, existingData)) {
  887. throw new IOException(MessageFormat.format(
  888. JGitText.get().collisionOn, id.name()));
  889. }
  890. } catch (MissingObjectException notLocal) {
  891. // This is OK, we don't have a copy of the object locally
  892. // but the API throws when we try to read it as usually its
  893. // an error to read something that doesn't exist.
  894. }
  895. }
  896. }
  897. private void doDeferredCheckBlobs() throws IOException {
  898. final byte[] readBuffer = buffer();
  899. final byte[] curBuffer = new byte[readBuffer.length];
  900. ObjectTypeAndSize info = new ObjectTypeAndSize();
  901. for (PackedObjectInfo obj : deferredCheckBlobs) {
  902. info = openDatabase(obj, info);
  903. if (info.type != Constants.OBJ_BLOB)
  904. throw new IOException(MessageFormat.format(
  905. JGitText.get().unknownObjectType, info.type));
  906. ObjectStream cur = readCurs.open(obj, info.type).openStream();
  907. try {
  908. long sz = info.size;
  909. if (cur.getSize() != sz)
  910. throw new IOException(MessageFormat.format(
  911. JGitText.get().collisionOn, obj.name()));
  912. InputStream pck = inflate(Source.DATABASE, sz);
  913. while (0 < sz) {
  914. int n = (int) Math.min(readBuffer.length, sz);
  915. IO.readFully(cur, curBuffer, 0, n);
  916. IO.readFully(pck, readBuffer, 0, n);
  917. for (int i = 0; i < n; i++) {
  918. if (curBuffer[i] != readBuffer[i])
  919. throw new IOException(MessageFormat.format(JGitText
  920. .get().collisionOn, obj.name()));
  921. }
  922. sz -= n;
  923. }
  924. pck.close();
  925. } finally {
  926. cur.close();
  927. }
  928. }
  929. }
  930. /** @return current position of the input stream being parsed. */
  931. private long streamPosition() {
  932. return bBase + bOffset;
  933. }
  934. private ObjectTypeAndSize openDatabase(PackedObjectInfo obj,
  935. ObjectTypeAndSize info) throws IOException {
  936. bOffset = 0;
  937. bAvail = 0;
  938. return seekDatabase(obj, info);
  939. }
  940. private ObjectTypeAndSize openDatabase(UnresolvedDelta delta,
  941. ObjectTypeAndSize info) throws IOException {
  942. bOffset = 0;
  943. bAvail = 0;
  944. return seekDatabase(delta, info);
  945. }
  946. // Consume exactly one byte from the buffer and return it.
  947. private int readFrom(final Source src) throws IOException {
  948. if (bAvail == 0)
  949. fill(src, 1);
  950. bAvail--;
  951. return buf[bOffset++] & 0xff;
  952. }
  953. // Consume cnt bytes from the buffer.
  954. private void use(final int cnt) {
  955. bOffset += cnt;
  956. bAvail -= cnt;
  957. }
  958. // Ensure at least need bytes are available in in {@link #buf}.
  959. private int fill(final Source src, final int need) throws IOException {
  960. while (bAvail < need) {
  961. int next = bOffset + bAvail;
  962. int free = buf.length - next;
  963. if (free + bAvail < need) {
  964. switch (src) {
  965. case INPUT:
  966. sync();
  967. break;
  968. case DATABASE:
  969. if (bAvail > 0)
  970. System.arraycopy(buf, bOffset, buf, 0, bAvail);
  971. bOffset = 0;
  972. break;
  973. }
  974. next = bAvail;
  975. free = buf.length - next;
  976. }
  977. switch (src) {
  978. case INPUT:
  979. next = in.read(buf, next, free);
  980. break;
  981. case DATABASE:
  982. next = readDatabase(buf, next, free);
  983. break;
  984. }
  985. if (next <= 0)
  986. throw new EOFException(JGitText.get().packfileIsTruncated);
  987. bAvail += next;
  988. }
  989. return bOffset;
  990. }
  991. // Store consumed bytes in {@link #buf} up to {@link #bOffset}.
  992. private void sync() throws IOException {
  993. packDigest.update(buf, 0, bOffset);
  994. onStoreStream(buf, 0, bOffset);
  995. if (bAvail > 0)
  996. System.arraycopy(buf, bOffset, buf, 0, bAvail);
  997. bBase += bOffset;
  998. bOffset = 0;
  999. }
  1000. /** @return a temporary byte array for use by the caller. */
  1001. protected byte[] buffer() {
  1002. return tempBuffer;
  1003. }
  1004. /**
  1005. * Construct a PackedObjectInfo instance for this parser.
  1006. *
  1007. * @param id
  1008. * identity of the object to be tracked.
  1009. * @param delta
  1010. * if the object was previously an unresolved delta, this is the
  1011. * delta object that was tracking it. Otherwise null.
  1012. * @param deltaBase
  1013. * if the object was previously an unresolved delta, this is the
  1014. * ObjectId of the base of the delta. The base may be outside of
  1015. * the pack stream if the stream was a thin-pack.
  1016. * @return info object containing this object's data.
  1017. */
  1018. protected PackedObjectInfo newInfo(AnyObjectId id, UnresolvedDelta delta,
  1019. ObjectId deltaBase) {
  1020. PackedObjectInfo oe = new PackedObjectInfo(id);
  1021. if (delta != null)
  1022. oe.setCRC(delta.crc);
  1023. return oe;
  1024. }
  1025. /**
  1026. * Store bytes received from the raw stream.
  1027. * <p>
  1028. * This method is invoked during {@link #parse(ProgressMonitor)} as data is
  1029. * consumed from the incoming stream. Implementors may use this event to
  1030. * archive the raw incoming stream to the destination repository in large
  1031. * chunks, without paying attention to object boundaries.
  1032. * <p>
  1033. * The only component of the pack not supplied to this method is the last 20
  1034. * bytes of the pack that comprise the trailing SHA-1 checksum. Those are
  1035. * passed to {@link #onPackFooter(byte[])}.
  1036. *
  1037. * @param raw
  1038. * buffer to copy data out of.
  1039. * @param pos
  1040. * first offset within the buffer that is valid.
  1041. * @param len
  1042. * number of bytes in the buffer that are valid.
  1043. * @throws IOException
  1044. * the stream cannot be archived.
  1045. */
  1046. protected abstract void onStoreStream(byte[] raw, int pos, int len)
  1047. throws IOException;
  1048. /**
  1049. * Store (and/or checksum) an object header.
  1050. * <p>
  1051. * Invoked after any of the {@code onBegin()} events. The entire header is
  1052. * supplied in a single invocation, before any object data is supplied.
  1053. *
  1054. * @param src
  1055. * where the data came from
  1056. * @param raw
  1057. * buffer to read data from.
  1058. * @param pos
  1059. * first offset within buffer that is valid.
  1060. * @param len
  1061. * number of bytes in buffer that are valid.
  1062. * @throws IOException
  1063. * the stream cannot be archived.
  1064. */
  1065. protected abstract void onObjectHeader(Source src, byte[] raw, int pos,
  1066. int len) throws IOException;
  1067. /**
  1068. * Store (and/or checksum) a portion of an object's data.
  1069. * <p>
  1070. * This method may be invoked multiple times per object, depending on the
  1071. * size of the object, the size of the parser's internal read buffer, and
  1072. * the alignment of the object relative to the read buffer.
  1073. * <p>
  1074. * Invoked after {@link #onObjectHeader(Source, byte[], int, int)}.
  1075. *
  1076. * @param src
  1077. * where the data came from
  1078. * @param raw
  1079. * buffer to read data from.
  1080. * @param pos
  1081. * first offset within buffer that is valid.
  1082. * @param len
  1083. * number of bytes in buffer that are valid.
  1084. * @throws IOException
  1085. * the stream cannot be archived.
  1086. */
  1087. protected abstract void onObjectData(Source src, byte[] raw, int pos,
  1088. int len) throws IOException;
  1089. /**
  1090. * Invoked for commits, trees, tags, and small blobs.
  1091. *
  1092. * @param obj
  1093. * the object info, populated.
  1094. * @param typeCode
  1095. * the type of the object.
  1096. * @param data
  1097. * inflated data for the object.
  1098. * @throws IOException
  1099. * the object cannot be archived.
  1100. */
  1101. protected abstract void onInflatedObjectData(PackedObjectInfo obj,
  1102. int typeCode, byte[] data) throws IOException;
  1103. /**
  1104. * Provide the implementation with the original stream's pack header.
  1105. *
  1106. * @param objCnt
  1107. * number of objects expected in the stream.
  1108. * @throws IOException
  1109. * the implementation refuses to work with this many objects.
  1110. */
  1111. protected abstract void onPackHeader(long objCnt) throws IOException;
  1112. /**
  1113. * Provide the implementation with the original stream's pack footer.
  1114. *
  1115. * @param hash
  1116. * the trailing 20 bytes of the pack, this is a SHA-1 checksum of
  1117. * all of the pack data.
  1118. * @throws IOException
  1119. * the stream cannot be archived.
  1120. */
  1121. protected abstract void onPackFooter(byte[] hash) throws IOException;
  1122. /**
  1123. * Provide the implementation with a base that was outside of the pack.
  1124. * <p>
  1125. * This event only occurs on a thin pack for base objects that were outside
  1126. * of the pack and came from the local repository. Usually an implementation
  1127. * uses this event to compress the base and append it onto the end of the
  1128. * pack, so the pack stays self-contained.
  1129. *
  1130. * @param typeCode
  1131. * type of the base object.
  1132. * @param data
  1133. * complete content of the base object.
  1134. * @param info
  1135. * packed object information for this base. Implementors must
  1136. * populate the CRC and offset members if returning true.
  1137. * @return true if the {@code info} should be included in the object list
  1138. * returned by {@link #getSortedObjectList(Comparator)}, false if it
  1139. * should not be included.
  1140. * @throws IOException
  1141. * the base could not be included into the pack.
  1142. */
  1143. protected abstract boolean onAppendBase(int typeCode, byte[] data,
  1144. PackedObjectInfo info) throws IOException;
  1145. /**
  1146. * Event indicating a thin pack has been completely processed.
  1147. * <p>
  1148. * This event is invoked only if a thin pack has delta references to objects
  1149. * external from the pack. The event is called after all of those deltas
  1150. * have been resolved.
  1151. *
  1152. * @throws IOException
  1153. * the pack cannot be archived.
  1154. */
  1155. protected abstract void onEndThinPack() throws IOException;
  1156. /**
  1157. * Reposition the database to re-read a previously stored object.
  1158. * <p>
  1159. * If the database is computing CRC-32 checksums for object data, it should
  1160. * reset its internal CRC instance during this method call.
  1161. *
  1162. * @param obj
  1163. * the object position to begin reading from. This is from
  1164. * {@link #newInfo(AnyObjectId, UnresolvedDelta, ObjectId)}.
  1165. * @param info
  1166. * object to populate with type and size.
  1167. * @return the {@code info} object.
  1168. * @throws IOException
  1169. * the database cannot reposition to this location.
  1170. */
  1171. protected abstract ObjectTypeAndSize seekDatabase(PackedObjectInfo obj,
  1172. ObjectTypeAndSize info) throws IOException;
  1173. /**
  1174. * Reposition the database to re-read a previously stored object.
  1175. * <p>
  1176. * If the database is computing CRC-32 checksums for object data, it should
  1177. * reset its internal CRC instance during this method call.
  1178. *
  1179. * @param delta
  1180. * the object position to begin reading from. This is an instance
  1181. * previously returned by {@link #onEndDelta()}.
  1182. * @param info
  1183. * object to populate with type and size.
  1184. * @return the {@code info} object.
  1185. * @throws IOException
  1186. * the database cannot reposition to this location.
  1187. */
  1188. protected abstract ObjectTypeAndSize seekDatabase(UnresolvedDelta delta,
  1189. ObjectTypeAndSize info) throws IOException;
  1190. /**
  1191. * Read from the database's current position into the buffer.
  1192. *
  1193. * @param dst
  1194. * the buffer to copy read data into.
  1195. * @param pos
  1196. * position within {@code dst} to start copying data into.
  1197. * @param cnt
  1198. * ideal target number of bytes to read. Actual read length may
  1199. * be shorter.
  1200. * @return number of bytes stored.
  1201. * @throws IOException
  1202. * the database cannot be accessed.
  1203. */
  1204. protected abstract int readDatabase(byte[] dst, int pos, int cnt)
  1205. throws IOException;
  1206. /**
  1207. * Check the current CRC matches the expected value.
  1208. * <p>
  1209. * This method is invoked when an object is read back in from the database
  1210. * and its data is used during delta resolution. The CRC is validated after
  1211. * the object has been fully read, allowing the parser to verify there was
  1212. * no silent data corruption.
  1213. * <p>
  1214. * Implementations are free to ignore this check by always returning true if
  1215. * they are performing other data integrity validations at a lower level.
  1216. *
  1217. * @param oldCRC
  1218. * the prior CRC that was recorded during the first scan of the
  1219. * object from the pack stream.
  1220. * @return true if the CRC matches; false if it does not.
  1221. */
  1222. protected abstract boolean checkCRC(int oldCRC);
  1223. /**
  1224. * Event notifying the start of an object stored whole (not as a delta).
  1225. *
  1226. * @param streamPosition
  1227. * position of this object in the incoming stream.
  1228. * @param type
  1229. * type of the object; one of {@link Constants#OBJ_COMMIT},
  1230. * {@link Constants#OBJ_TREE}, {@link Constants#OBJ_BLOB}, or
  1231. * {@link Constants#OBJ_TAG}.
  1232. * @param inflatedSize
  1233. * size of the object when fully inflated. The size stored within
  1234. * the pack may be larger or smaller, and is not yet known.
  1235. * @throws IOException
  1236. * the object cannot be recorded.
  1237. */
  1238. protected abstract void onBeginWholeObject(long streamPosition, int type,
  1239. long inflatedSize) throws IOException;
  1240. /**
  1241. * Event notifying the the current object.
  1242. *
  1243. *@param info
  1244. * object information.
  1245. * @throws IOException
  1246. * the object cannot be recorded.
  1247. */
  1248. protected abstract void onEndWholeObject(PackedObjectInfo info)
  1249. throws IOException;
  1250. /**
  1251. * Event notifying start of a delta referencing its base by offset.
  1252. *
  1253. * @param deltaStreamPosition
  1254. * position of this object in the incoming stream.
  1255. * @param baseStreamPosition
  1256. * position of the base object in the incoming stream. The base
  1257. * must be before the delta, therefore {@code baseStreamPosition
  1258. * &lt; deltaStreamPosition}. This is <b>not</b> the position
  1259. * returned by a prior end object event.
  1260. * @param inflatedSize
  1261. * size of the delta when fully inflated. The size stored within
  1262. * the pack may be larger or smaller, and is not yet known.
  1263. * @throws IOException
  1264. * the object cannot be recorded.
  1265. */
  1266. protected abstract void onBeginOfsDelta(long deltaStreamPosition,
  1267. long baseStreamPosition, long inflatedSize) throws IOException;
  1268. /**
  1269. * Event notifying start of a delta referencing its base by ObjectId.
  1270. *
  1271. * @param deltaStreamPosition
  1272. * position of this object in the incoming stream.
  1273. * @param baseId
  1274. * name of the base object. This object may be later in the
  1275. * stream, or might not appear at all in the stream (in the case
  1276. * of a thin-pack).
  1277. * @param inflatedSize
  1278. * size of the delta when fully inflated. The size stored within
  1279. * the pack may be larger or smaller, and is not yet known.
  1280. * @throws IOException
  1281. * the object cannot be recorded.
  1282. */
  1283. protected abstract void onBeginRefDelta(long deltaStreamPosition,
  1284. AnyObjectId baseId, long inflatedSize) throws IOException;
  1285. /**
  1286. * Event notifying the the current object.
  1287. *
  1288. *@return object information that must be populated with at least the
  1289. * offset.
  1290. * @throws IOException
  1291. * the object cannot be recorded.
  1292. */
  1293. protected UnresolvedDelta onEndDelta() throws IOException {
  1294. return new UnresolvedDelta();
  1295. }
  1296. /** Type and size information about an object in the database buffer. */
  1297. public static class ObjectTypeAndSize {
  1298. /** The type of the object. */
  1299. public int type;
  1300. /** The inflated size of the object. */
  1301. public long size;
  1302. }
  1303. private void inflateAndSkip(final Source src, final long inflatedSize)
  1304. throws IOException {
  1305. final InputStream inf = inflate(src, inflatedSize);
  1306. IO.skipFully(inf, inflatedSize);
  1307. inf.close();
  1308. }
  1309. private byte[] inflateAndReturn(final Source src, final long inflatedSize)
  1310. throws IOException {
  1311. final byte[] dst = new byte[(int) inflatedSize];
  1312. final InputStream inf = inflate(src, inflatedSize);
  1313. IO.readFully(inf, dst, 0, dst.length);
  1314. inf.close();
  1315. return dst;
  1316. }
  1317. private InputStream inflate(final Source src, final long inflatedSize)
  1318. throws IOException {
  1319. inflater.open(src, inflatedSize);
  1320. return inflater;
  1321. }
  1322. private static class DeltaChain extends ObjectIdOwnerMap.Entry {
  1323. UnresolvedDelta head;
  1324. DeltaChain(final AnyObjectId id) {
  1325. super(id);
  1326. }
  1327. UnresolvedDelta remove() {
  1328. final UnresolvedDelta r = head;
  1329. if (r != null)
  1330. head = null;
  1331. return r;
  1332. }
  1333. void add(final UnresolvedDelta d) {
  1334. d.next = head;
  1335. head = d;
  1336. }
  1337. }
  1338. /** Information about an unresolved delta in this pack stream. */
  1339. public static class UnresolvedDelta {
  1340. long position;
  1341. int crc;
  1342. UnresolvedDelta next;
  1343. /** @return offset within the input stream. */
  1344. public long getOffset() {
  1345. return position;
  1346. }
  1347. /** @return the CRC-32 checksum of the stored delta data. */
  1348. public int getCRC() {
  1349. return crc;
  1350. }
  1351. /**
  1352. * @param crc32
  1353. * the CRC-32 checksum of the stored delta data.
  1354. */
  1355. public void setCRC(int crc32) {
  1356. crc = crc32;
  1357. }
  1358. }
  1359. private static class DeltaVisit {
  1360. final UnresolvedDelta delta;
  1361. ObjectId id;
  1362. byte[] data;
  1363. DeltaVisit parent;
  1364. UnresolvedDelta nextChild;
  1365. DeltaVisit() {
  1366. this.delta = null; // At the root of the stack we have a base.
  1367. }
  1368. DeltaVisit(DeltaVisit parent) {
  1369. this.parent = parent;
  1370. this.delta = parent.nextChild;
  1371. parent.nextChild = delta.next;
  1372. }
  1373. DeltaVisit next() {
  1374. // If our parent has no more children, discard it.
  1375. if (parent != null && parent.nextChild == null) {
  1376. parent.data = null;
  1377. parent = parent.parent;
  1378. }
  1379. if (nextChild != null)
  1380. return new DeltaVisit(this);
  1381. // If we have no child ourselves, our parent must (if it exists),
  1382. // due to the discard rule above. With no parent, we are done.
  1383. if (parent != null)
  1384. return new DeltaVisit(parent);
  1385. return null;
  1386. }
  1387. }
  1388. private void addObjectAndTrack(PackedObjectInfo oe) {
  1389. entries[entryCount++] = oe;
  1390. if (needNewObjectIds())
  1391. newObjectIds.add(oe);
  1392. }
  1393. private class InflaterStream extends InputStream {
  1394. private final Inflater inf;
  1395. private final byte[] skipBuffer;
  1396. private Source src;
  1397. private long expectedSize;
  1398. private long actualSize;
  1399. private int p;
  1400. InflaterStream() {
  1401. inf = InflaterCache.get();
  1402. skipBuffer = new byte[512];
  1403. }
  1404. void release() {
  1405. inf.reset();
  1406. InflaterCache.release(inf);
  1407. }
  1408. void open(Source source, long inflatedSize) throws IOException {
  1409. src = source;
  1410. expectedSize = inflatedSize;
  1411. actualSize = 0;
  1412. p = fill(src, 1);
  1413. inf.setInput(buf, p, bAvail);
  1414. }
  1415. @Override
  1416. public long skip(long toSkip) throws IOException {
  1417. long n = 0;
  1418. while (n < toSkip) {
  1419. final int cnt = (int) Math.min(skipBuffer.length, toSkip - n);
  1420. final int r = read(skipBuffer, 0, cnt);
  1421. if (r <= 0)
  1422. break;
  1423. n += r;
  1424. }
  1425. return n;
  1426. }
  1427. @Override
  1428. public int read() throws IOException {
  1429. int n = read(skipBuffer, 0, 1);
  1430. return n == 1 ? skipBuffer[0] & 0xff : -1;
  1431. }
  1432. @Override
  1433. public int read(byte[] dst, int pos, int cnt) throws IOException {
  1434. try {
  1435. int n = 0;
  1436. while (n < cnt) {
  1437. int r = inf.inflate(dst, pos + n, cnt - n);
  1438. if (r == 0) {
  1439. if (inf.finished())
  1440. break;
  1441. if (inf.needsInput()) {
  1442. onObjectData(src, buf, p, bAvail);
  1443. use(bAvail);
  1444. p = fill(src, 1);
  1445. inf.setInput(buf, p, bAvail);
  1446. } else {
  1447. throw new CorruptObjectException(
  1448. MessageFormat
  1449. .format(
  1450. JGitText.get().packfileCorruptionDetected,
  1451. JGitText.get().unknownZlibError));
  1452. }
  1453. } else {
  1454. n += r;
  1455. }
  1456. }
  1457. actualSize += n;
  1458. return 0 < n ? n : -1;
  1459. } catch (DataFormatException dfe) {
  1460. throw new CorruptObjectException(MessageFormat.format(JGitText
  1461. .get().packfileCorruptionDetected, dfe.getMessage()));
  1462. }
  1463. }
  1464. @Override
  1465. public void close() throws IOException {
  1466. // We need to read here to enter the loop above and pump the
  1467. // trailing checksum into the Inflater. It should return -1 as the
  1468. // caller was supposed to consume all content.
  1469. //
  1470. if (read(skipBuffer) != -1 || actualSize != expectedSize) {
  1471. throw new CorruptObjectException(MessageFormat.format(JGitText
  1472. .get().packfileCorruptionDetected,
  1473. JGitText.get().wrongDecompressedLength));
  1474. }
  1475. int used = bAvail - inf.getRemaining();
  1476. if (0 < used) {
  1477. onObjectData(src, buf, p, used);
  1478. use(used);
  1479. }
  1480. inf.reset();
  1481. }
  1482. }
  1483. }