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.

LockFile.java 15KB

Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
Fix atomic lock file creation on NFS FS_POSIX.createNewFile(File) failed to properly implement atomic file creation on NFS using the algorithm [1]: - name of the hard link must be unique to prevent that two processes using different NFS clients try to create the same link. This would render nlink useless to detect if there was a race. - the hard link must be retained for the lifetime of the file since we don't know when the state of the involved NFS clients will be synchronized. This depends on NFS configuration options. To fix these issues we need to change the signature of createNewFile which would break API. Hence deprecate the old method FS.createNewFile(File) and add a new method createNewFileAtomic(File). The new method returns a LockToken which needs to be retained by the caller (LockFile) until all involved NFS clients synchronized their state. Since we don't know when the NFS caches are synchronized we need to retain the token until the corresponding file is no longer needed. The LockToken must be closed after the LockFile using it has been committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile = false this will delete the hard link which guarded the atomic creation of the file. When acquiring the lock fails ensure that the hard link is removed. [1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html also see file creation flag O_EXCL in http://man7.org/linux/man-pages/man2/open.2.html Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563
  1. /*
  2. * Copyright (C) 2007, Robin Rosenberg <robin.rosenberg@dewire.com>
  3. * Copyright (C) 2006-2021, Shawn O. Pearce <spearce@spearce.org> and others
  4. *
  5. * This program and the accompanying materials are made available under the
  6. * terms of the Eclipse Distribution License v. 1.0 which is available at
  7. * https://www.eclipse.org/org/documents/edl-v10.php.
  8. *
  9. * SPDX-License-Identifier: BSD-3-Clause
  10. */
  11. package org.eclipse.jgit.internal.storage.file;
  12. import static org.eclipse.jgit.lib.Constants.LOCK_SUFFIX;
  13. import java.io.File;
  14. import java.io.FileInputStream;
  15. import java.io.FileNotFoundException;
  16. import java.io.FileOutputStream;
  17. import java.io.FilenameFilter;
  18. import java.io.IOException;
  19. import java.io.OutputStream;
  20. import java.nio.ByteBuffer;
  21. import java.nio.channels.Channels;
  22. import java.nio.channels.FileChannel;
  23. import java.nio.file.Files;
  24. import java.nio.file.StandardCopyOption;
  25. import java.nio.file.attribute.FileTime;
  26. import java.text.MessageFormat;
  27. import java.time.Instant;
  28. import java.util.concurrent.TimeUnit;
  29. import org.eclipse.jgit.internal.JGitText;
  30. import org.eclipse.jgit.lib.Constants;
  31. import org.eclipse.jgit.lib.ObjectId;
  32. import org.eclipse.jgit.util.FS;
  33. import org.eclipse.jgit.util.FS.LockToken;
  34. import org.eclipse.jgit.util.FileUtils;
  35. import org.slf4j.Logger;
  36. import org.slf4j.LoggerFactory;
  37. /**
  38. * Git style file locking and replacement.
  39. * <p>
  40. * To modify a ref file Git tries to use an atomic update approach: we write the
  41. * new data into a brand new file, then rename it in place over the old name.
  42. * This way we can just delete the temporary file if anything goes wrong, and
  43. * nothing has been damaged. To coordinate access from multiple processes at
  44. * once Git tries to atomically create the new temporary file under a well-known
  45. * name.
  46. */
  47. public class LockFile {
  48. private static final Logger LOG = LoggerFactory.getLogger(LockFile.class);
  49. /**
  50. * Unlock the given file.
  51. * <p>
  52. * This method can be used for recovering from a thrown
  53. * {@link org.eclipse.jgit.errors.LockFailedException} . This method does
  54. * not validate that the lock is or is not currently held before attempting
  55. * to unlock it.
  56. *
  57. * @param file
  58. * a {@link java.io.File} object.
  59. * @return true if unlocked, false if unlocking failed
  60. */
  61. public static boolean unlock(File file) {
  62. final File lockFile = getLockFile(file);
  63. final int flags = FileUtils.RETRY | FileUtils.SKIP_MISSING;
  64. try {
  65. FileUtils.delete(lockFile, flags);
  66. } catch (IOException ignored) {
  67. // Ignore and return whether lock file still exists
  68. }
  69. return !lockFile.exists();
  70. }
  71. /**
  72. * Get the lock file corresponding to the given file.
  73. *
  74. * @param file
  75. * @return lock file
  76. */
  77. static File getLockFile(File file) {
  78. return new File(file.getParentFile(),
  79. file.getName() + LOCK_SUFFIX);
  80. }
  81. /** Filter to skip over active lock files when listing a directory. */
  82. static final FilenameFilter FILTER = (File dir,
  83. String name) -> !name.endsWith(LOCK_SUFFIX);
  84. private final File ref;
  85. private final File lck;
  86. private boolean haveLck;
  87. private FileOutputStream os;
  88. private boolean needSnapshot;
  89. private boolean fsync;
  90. private boolean isAppend;
  91. private boolean written;
  92. private FileSnapshot commitSnapshot;
  93. private LockToken token;
  94. /**
  95. * Create a new lock for any file.
  96. *
  97. * @param f
  98. * the file that will be locked.
  99. */
  100. public LockFile(File f) {
  101. ref = f;
  102. lck = getLockFile(ref);
  103. }
  104. /**
  105. * Try to establish the lock.
  106. *
  107. * @return true if the lock is now held by the caller; false if it is held
  108. * by someone else.
  109. * @throws java.io.IOException
  110. * the temporary output file could not be created. The caller
  111. * does not hold the lock.
  112. */
  113. public boolean lock() throws IOException {
  114. if (haveLck) {
  115. throw new IllegalStateException(
  116. MessageFormat.format(JGitText.get().lockAlreadyHeld, ref));
  117. }
  118. FileUtils.mkdirs(lck.getParentFile(), true);
  119. try {
  120. token = FS.DETECTED.createNewFileAtomic(lck);
  121. } catch (IOException e) {
  122. LOG.error(JGitText.get().failedCreateLockFile, lck, e);
  123. throw e;
  124. }
  125. boolean obtainedLock = token.isCreated();
  126. if (obtainedLock) {
  127. haveLck = true;
  128. isAppend = false;
  129. written = false;
  130. } else {
  131. closeToken();
  132. }
  133. return obtainedLock;
  134. }
  135. /**
  136. * Try to establish the lock for appending.
  137. *
  138. * @return true if the lock is now held by the caller; false if it is held
  139. * by someone else.
  140. * @throws java.io.IOException
  141. * the temporary output file could not be created. The caller
  142. * does not hold the lock.
  143. */
  144. public boolean lockForAppend() throws IOException {
  145. if (!lock()) {
  146. return false;
  147. }
  148. copyCurrentContent();
  149. isAppend = true;
  150. written = false;
  151. return true;
  152. }
  153. // For tests only
  154. boolean isLocked() {
  155. return haveLck;
  156. }
  157. private FileOutputStream getStream() throws IOException {
  158. return new FileOutputStream(lck, isAppend);
  159. }
  160. /**
  161. * Copy the current file content into the temporary file.
  162. * <p>
  163. * This method saves the current file content by inserting it into the
  164. * temporary file, so that the caller can safely append rather than replace
  165. * the primary file.
  166. * <p>
  167. * This method does nothing if the current file does not exist, or exists
  168. * but is empty.
  169. *
  170. * @throws java.io.IOException
  171. * the temporary file could not be written, or a read error
  172. * occurred while reading from the current file. The lock is
  173. * released before throwing the underlying IO exception to the
  174. * caller.
  175. * @throws java.lang.RuntimeException
  176. * the temporary file could not be written. The lock is released
  177. * before throwing the underlying exception to the caller.
  178. */
  179. public void copyCurrentContent() throws IOException {
  180. requireLock();
  181. try (FileOutputStream out = getStream()) {
  182. try (FileInputStream fis = new FileInputStream(ref)) {
  183. if (fsync) {
  184. FileChannel in = fis.getChannel();
  185. long pos = 0;
  186. long cnt = in.size();
  187. while (0 < cnt) {
  188. long r = out.getChannel().transferFrom(in, pos, cnt);
  189. pos += r;
  190. cnt -= r;
  191. }
  192. } else {
  193. final byte[] buf = new byte[2048];
  194. int r;
  195. while ((r = fis.read(buf)) >= 0) {
  196. out.write(buf, 0, r);
  197. }
  198. }
  199. } catch (FileNotFoundException fnfe) {
  200. if (ref.exists()) {
  201. throw fnfe;
  202. }
  203. // Don't worry about a file that doesn't exist yet, it
  204. // conceptually has no current content to copy.
  205. }
  206. } catch (IOException | RuntimeException | Error ioe) {
  207. unlock();
  208. throw ioe;
  209. }
  210. }
  211. /**
  212. * Write an ObjectId and LF to the temporary file.
  213. *
  214. * @param id
  215. * the id to store in the file. The id will be written in hex,
  216. * followed by a sole LF.
  217. * @throws java.io.IOException
  218. * the temporary file could not be written. The lock is released
  219. * before throwing the underlying IO exception to the caller.
  220. * @throws java.lang.RuntimeException
  221. * the temporary file could not be written. The lock is released
  222. * before throwing the underlying exception to the caller.
  223. */
  224. public void write(ObjectId id) throws IOException {
  225. byte[] buf = new byte[Constants.OBJECT_ID_STRING_LENGTH + 1];
  226. id.copyTo(buf, 0);
  227. buf[Constants.OBJECT_ID_STRING_LENGTH] = '\n';
  228. write(buf);
  229. }
  230. /**
  231. * Write arbitrary data to the temporary file.
  232. *
  233. * @param content
  234. * the bytes to store in the temporary file. No additional bytes
  235. * are added, so if the file must end with an LF it must appear
  236. * at the end of the byte array.
  237. * @throws java.io.IOException
  238. * the temporary file could not be written. The lock is released
  239. * before throwing the underlying IO exception to the caller.
  240. * @throws java.lang.RuntimeException
  241. * the temporary file could not be written. The lock is released
  242. * before throwing the underlying exception to the caller.
  243. */
  244. public void write(byte[] content) throws IOException {
  245. requireLock();
  246. try (FileOutputStream out = getStream()) {
  247. if (written) {
  248. throw new IOException(MessageFormat
  249. .format(JGitText.get().lockStreamClosed, ref));
  250. }
  251. if (fsync) {
  252. FileChannel fc = out.getChannel();
  253. ByteBuffer buf = ByteBuffer.wrap(content);
  254. while (0 < buf.remaining()) {
  255. fc.write(buf);
  256. }
  257. fc.force(true);
  258. } else {
  259. out.write(content);
  260. }
  261. written = true;
  262. } catch (IOException | RuntimeException | Error ioe) {
  263. unlock();
  264. throw ioe;
  265. }
  266. }
  267. /**
  268. * Obtain the direct output stream for this lock.
  269. * <p>
  270. * The stream may only be accessed once, and only after {@link #lock()} has
  271. * been successfully invoked and returned true. Callers must close the
  272. * stream prior to calling {@link #commit()} to commit the change.
  273. *
  274. * @return a stream to write to the new file. The stream is unbuffered.
  275. */
  276. public OutputStream getOutputStream() {
  277. requireLock();
  278. if (written || os != null) {
  279. throw new IllegalStateException(MessageFormat
  280. .format(JGitText.get().lockStreamMultiple, ref));
  281. }
  282. return new OutputStream() {
  283. private OutputStream out;
  284. private boolean closed;
  285. private OutputStream get() throws IOException {
  286. if (written) {
  287. throw new IOException(MessageFormat
  288. .format(JGitText.get().lockStreamMultiple, ref));
  289. }
  290. if (out == null) {
  291. os = getStream();
  292. if (fsync) {
  293. out = Channels.newOutputStream(os.getChannel());
  294. } else {
  295. out = os;
  296. }
  297. }
  298. return out;
  299. }
  300. @Override
  301. public void write(byte[] b, int o, int n) throws IOException {
  302. get().write(b, o, n);
  303. }
  304. @Override
  305. public void write(byte[] b) throws IOException {
  306. get().write(b);
  307. }
  308. @Override
  309. public void write(int b) throws IOException {
  310. get().write(b);
  311. }
  312. @Override
  313. public void close() throws IOException {
  314. if (closed) {
  315. return;
  316. }
  317. closed = true;
  318. try {
  319. if (written) {
  320. throw new IOException(MessageFormat
  321. .format(JGitText.get().lockStreamClosed, ref));
  322. }
  323. if (out != null) {
  324. if (fsync) {
  325. os.getChannel().force(true);
  326. }
  327. out.close();
  328. os = null;
  329. }
  330. written = true;
  331. } catch (IOException | RuntimeException | Error ioe) {
  332. unlock();
  333. throw ioe;
  334. }
  335. }
  336. };
  337. }
  338. void requireLock() {
  339. if (!haveLck) {
  340. unlock();
  341. throw new IllegalStateException(MessageFormat.format(JGitText.get().lockOnNotHeld, ref));
  342. }
  343. }
  344. /**
  345. * Request that {@link #commit()} remember modification time.
  346. * <p>
  347. * This is an alias for {@code setNeedSnapshot(true)}.
  348. *
  349. * @param on
  350. * true if the commit method must remember the modification time.
  351. */
  352. public void setNeedStatInformation(boolean on) {
  353. setNeedSnapshot(on);
  354. }
  355. /**
  356. * Request that {@link #commit()} remember the
  357. * {@link org.eclipse.jgit.internal.storage.file.FileSnapshot}.
  358. *
  359. * @param on
  360. * true if the commit method must remember the FileSnapshot.
  361. */
  362. public void setNeedSnapshot(boolean on) {
  363. needSnapshot = on;
  364. }
  365. /**
  366. * Request that {@link #commit()} force dirty data to the drive.
  367. *
  368. * @param on
  369. * true if dirty data should be forced to the drive.
  370. */
  371. public void setFSync(boolean on) {
  372. fsync = on;
  373. }
  374. /**
  375. * Wait until the lock file information differs from the old file.
  376. * <p>
  377. * This method tests the last modification date. If both are the same, this
  378. * method sleeps until it can force the new lock file's modification date to
  379. * be later than the target file.
  380. *
  381. * @throws java.lang.InterruptedException
  382. * the thread was interrupted before the last modified date of
  383. * the lock file was different from the last modified date of
  384. * the target file.
  385. */
  386. public void waitForStatChange() throws InterruptedException {
  387. FileSnapshot o = FileSnapshot.save(ref);
  388. FileSnapshot n = FileSnapshot.save(lck);
  389. long fsTimeResolution = FS.getFileStoreAttributes(lck.toPath())
  390. .getFsTimestampResolution().toNanos();
  391. while (o.equals(n)) {
  392. TimeUnit.NANOSECONDS.sleep(fsTimeResolution);
  393. try {
  394. Files.setLastModifiedTime(lck.toPath(),
  395. FileTime.from(Instant.now()));
  396. } catch (IOException e) {
  397. n.waitUntilNotRacy();
  398. }
  399. n = FileSnapshot.save(lck);
  400. }
  401. }
  402. /**
  403. * Commit this change and release the lock.
  404. * <p>
  405. * If this method fails (returns false) the lock is still released.
  406. *
  407. * @return true if the commit was successful and the file contains the new
  408. * data; false if the commit failed and the file remains with the
  409. * old data.
  410. * @throws java.lang.IllegalStateException
  411. * the lock is not held.
  412. */
  413. public boolean commit() {
  414. if (os != null) {
  415. unlock();
  416. throw new IllegalStateException(MessageFormat.format(JGitText.get().lockOnNotClosed, ref));
  417. }
  418. saveStatInformation();
  419. try {
  420. FileUtils.rename(lck, ref, StandardCopyOption.ATOMIC_MOVE);
  421. haveLck = false;
  422. isAppend = false;
  423. written = false;
  424. closeToken();
  425. return true;
  426. } catch (IOException e) {
  427. unlock();
  428. return false;
  429. }
  430. }
  431. private void closeToken() {
  432. if (token != null) {
  433. token.close();
  434. token = null;
  435. }
  436. }
  437. private void saveStatInformation() {
  438. if (needSnapshot)
  439. commitSnapshot = FileSnapshot.save(lck);
  440. }
  441. /**
  442. * Get the modification time of the output file when it was committed.
  443. *
  444. * @return modification time of the lock file right before we committed it.
  445. * @deprecated use {@link #getCommitLastModifiedInstant()} instead
  446. */
  447. @Deprecated
  448. public long getCommitLastModified() {
  449. return commitSnapshot.lastModified();
  450. }
  451. /**
  452. * Get the modification time of the output file when it was committed.
  453. *
  454. * @return modification time of the lock file right before we committed it.
  455. */
  456. public Instant getCommitLastModifiedInstant() {
  457. return commitSnapshot.lastModifiedInstant();
  458. }
  459. /**
  460. * Get the {@link FileSnapshot} just before commit.
  461. *
  462. * @return get the {@link FileSnapshot} just before commit.
  463. */
  464. public FileSnapshot getCommitSnapshot() {
  465. return commitSnapshot;
  466. }
  467. /**
  468. * Update the commit snapshot {@link #getCommitSnapshot()} before commit.
  469. * <p>
  470. * This may be necessary if you need time stamp before commit occurs, e.g
  471. * while writing the index.
  472. */
  473. public void createCommitSnapshot() {
  474. saveStatInformation();
  475. }
  476. /**
  477. * Unlock this file and abort this change.
  478. * <p>
  479. * The temporary file (if created) is deleted before returning.
  480. */
  481. public void unlock() {
  482. if (os != null) {
  483. try {
  484. os.close();
  485. } catch (IOException e) {
  486. LOG.error(MessageFormat
  487. .format(JGitText.get().unlockLockFileFailed, lck), e);
  488. }
  489. os = null;
  490. }
  491. if (haveLck) {
  492. haveLck = false;
  493. try {
  494. FileUtils.delete(lck, FileUtils.RETRY);
  495. } catch (IOException e) {
  496. LOG.error(MessageFormat
  497. .format(JGitText.get().unlockLockFileFailed, lck), e);
  498. } finally {
  499. closeToken();
  500. }
  501. }
  502. isAppend = false;
  503. written = false;
  504. }
  505. /** {@inheritDoc} */
  506. @SuppressWarnings("nls")
  507. @Override
  508. public String toString() {
  509. return "LockFile[" + lck + ", haveLck=" + haveLck + "]";
  510. }
  511. }