sapling/eden/fs/inodes/TreeInode.h

422 lines
14 KiB
C
Raw Normal View History

/*
* Copyright (c) 2016-present, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under the BSD-style license found in the
* LICENSE file in the root directory of this source tree. An additional grant
* of patent rights can be found in the PATENTS file in the same directory.
*
*/
#pragma once
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
#include <folly/Optional.h>
#include "eden/fs/inodes/InodeBase.h"
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
#include "eden/fs/model/Hash.h"
#include "eden/utils/PathMap.h"
namespace facebook {
namespace eden {
class CheckoutAction;
class CheckoutContext;
class EdenMount;
class FileHandle;
class InodeMap;
class ObjectStore;
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
class Overlay;
class RenameLock;
class Tree;
class TreeEntry;
/**
* Represents a directory in the file system.
*/
class TreeInode : public InodeBase {
public:
enum : int { WRONG_TYPE_ERRNO = ENOTDIR };
/**
* Represents a directory entry.
*
* A directory entry can be in one of several states:
*
* - An InodeBase object for the entry may or may not exist. If it does
* exist, it is the authoritative source of data for the entry.
*
* - If the child InodeBase object does not exist, we may or may not have an
* inode number already allocated for the child. An inode number can be
* allocated on-demand if necessary, without fully creating a child
* InodeBase object.
*
* - The child may or may not be materialized in the overlay.
*
* If the child contents are identical to an existing source control Tree
* or Blob then it does not need to be materialized, and the Entry may only
* contain the hash identifying the Tree/Blob.
*
* If the child is materialized in the overlay, then it must have an inode
* number allocated to it.
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
*/
struct Entry {
public:
/**
* Create a hash for a non-materialized entry.
*/
Entry(mode_t m, Hash hash) : mode(m), hash_{hash} {}
/**
* Create a hash for a materialized entry.
*/
Entry(mode_t m, fuse_ino_t number) : mode(m), inodeNumber_{number} {}
Entry(Entry&& e) = default;
Entry& operator=(Entry&& e) = default;
bool isMaterialized() const {
// TODO: In the future we should probably only allow callers to invoke
// this method when inode is not set. If inode is set it should be the
// authoritative source of data.
return !hash_.hasValue();
}
Hash getHash() const {
// TODO: In the future we should probably only allow callers to invoke
// this method when inode is not set. If inode is set it should be the
// authoritative source of data.
return hash_.value();
}
bool hasInodeNumber() const {
return inodeNumber_ != 0;
}
fuse_ino_t getInodeNumber() const {
DCHECK_NE(inodeNumber_, 0);
return inodeNumber_;
}
void setInodeNumber(fuse_ino_t number) {
DCHECK_EQ(inodeNumber_, 0);
DCHECK(!inode);
inodeNumber_ = number;
}
void setMaterialized(fuse_ino_t inode) {
DCHECK(inodeNumber_ == 0 || inode == inodeNumber_);
inodeNumber_ = inode;
hash_.clear();
}
void setUnmaterialized(Hash hash) {
hash_ = hash;
}
// TODO: Make mode private and provide an accessor method instead
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
/** The complete st_mode value for this entry */
mode_t mode{0};
private:
/**
* If the entry is not materialized, this contains the hash
* identifying the source control Tree (if this is a directory) or Blob
* (if this is a file) that contains the entry contents.
*
* If the entry is materialized, this field is not set.
*
* TODO: If inode is set, this field generally should not be used, and the
* child InodeBase should be consulted instead.
*/
folly::Optional<Hash> hash_;
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
/**
* The inode number, if one is allocated for this entry, or 0 if one is not
* allocated.
*
* An inode number is required for materialized entries, so this is always
* non-zero if hash_ is not set. (It may also be non-zero even when hash_
* is set.)
*/
fuse_ino_t inodeNumber_{0};
public:
// TODO: Make inode private and provide an accessor method instead
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
/**
* A pointer to the child inode, if it is loaded, or null if it is not
* loaded.
*
* Note that we store this as a raw pointer. Children inodes hold a
* reference to their parent TreeInode, not the other way around.
* Children inodes can be destroyed only in one of two ways:
* - Being unlinked, then having their last reference go away.
* In this case they will be removed from our entries list when they are
* unlinked.
* - Being unloaded (after their reference count is already 0). In this
* case the parent TreeInodes responsible for triggering unloading of its
* children, so it resets this pointer to null when it unloads the child.
*/
InodeBase* inode{nullptr};
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
};
/** Represents a directory in the overlay */
struct Dir {
/** The direct children of this directory */
PathMap<std::unique_ptr<Entry>> entries;
/** If the origin of this dir was a Tree, the hash of that tree */
folly::Optional<Hash> treeHash;
/** true if the dir has been materialized to the overlay.
* If the contents match the original tree, this is false. */
bool materialized{false};
};
/** Holds the results of a create operation.
*
* It is important that the file handle creation respect O_EXCL if
* it set in the flags parameter to TreeInode::create.
*/
struct CreateResult {
/// file attributes and cache ttls.
fusell::Dispatcher::Attr attr;
/// The newly created inode instance.
InodePtr inode;
/// The newly opened file handle.
std::shared_ptr<FileHandle> file;
explicit CreateResult(const fusell::MountPoint* mount) : attr(mount) {}
};
TreeInode(
fuse_ino_t ino,
TreeInodePtr parent,
PathComponentPiece name,
Entry* entry,
std::unique_ptr<Tree>&& tree);
/// Construct an inode that only has backing in the Overlay area
TreeInode(
fuse_ino_t ino,
TreeInodePtr parent,
PathComponentPiece name,
Entry* entry,
Dir&& dir);
/// Constructors for the root TreeInode
TreeInode(EdenMount* mount, std::unique_ptr<Tree>&& tree);
TreeInode(EdenMount* mount, Dir&& tree);
~TreeInode();
folly::Future<fusell::Dispatcher::Attr> getattr() override;
fusell::Dispatcher::Attr getAttrLocked(const Dir* contents);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
/** Implements the InodeBase method used by the Dispatcher
* to create the Inode instance for a given name */
folly::Future<InodePtr> getChildByName(PathComponentPiece namepiece);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
/**
* Get the inode object for a child of this directory.
*
* The Inode object will be loaded if it is not already loaded.
*/
folly::Future<InodePtr> getOrLoadChild(PathComponentPiece name);
folly::Future<TreeInodePtr> getOrLoadChildTree(PathComponentPiece name);
/**
* Recursively look up a child inode.
*
* The Inode object in question, and all intervening TreeInode objects,
* will be loaded if they are not already loaded.
*/
folly::Future<InodePtr> getChildRecursive(RelativePathPiece name);
fuse_ino_t getChildInodeNumber(PathComponentPiece name);
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
folly::Future<std::shared_ptr<fusell::DirHandle>> opendir(
const struct fuse_file_info& fi);
folly::Future<folly::Unit> rename(
PathComponentPiece name,
TreeInodePtr newParent,
PathComponentPiece newName);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
const folly::Synchronized<Dir>& getContents() const {
return contents_;
}
folly::Synchronized<Dir>& getContents() {
return contents_;
}
/**
* Get the InodeMap for this tree's EdenMount.
*
* The InodeMap is guaranteed to remain valid for at least the lifetime of
* the TreeInode object.
*/
InodeMap* getInodeMap() const;
/**
* Get the ObjectStore for this mount point.
*
* The ObjectStore is guaranteed to remain valid for at least the lifetime of
* the TreeInode object. (The ObjectStore is owned by the EdenMount.)
*/
ObjectStore* getStore() const;
const std::shared_ptr<Overlay>& getOverlay() const;
folly::Future<CreateResult>
create(PathComponentPiece name, mode_t mode, int flags);
FileInodePtr symlink(PathComponentPiece name, folly::StringPiece contents);
TreeInodePtr mkdir(PathComponentPiece name, mode_t mode);
folly::Future<folly::Unit> unlink(PathComponentPiece name);
folly::Future<folly::Unit> rmdir(PathComponentPiece name);
/**
* Update this directory so that it matches the specified source control Tree
* object.
*
* @param ctx The CheckoutContext for the current checkout operation.
* The caller guarantees that the CheckoutContext argument will remain
* valid until the returned Future completes.
* @param fromTree The Tree object that the checkout operation is moving
* from. This argument is necessary to detect conflicts between the
* current inode state and the expected previous source control state.
* This argument may be null when updating a TreeInode that did not exist
* in source control in the previous commit state.
* @param toTree The Tree object that the checkout operation is moving to.
* This argument must not be null. (In order to remove a file or tree
* that does not exist in the new commit, checkoutRemoveChild() must be
* invoked on the parent TreeInode instead.)
*
* @return Returns a future that will be fulfilled once this tree and all of
* its children have been updated.
*/
folly::Future<folly::Unit> checkout(
CheckoutContext* ctx,
std::unique_ptr<Tree> fromTree,
std::unique_ptr<Tree> toTree);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
/** Ensure that the overlay is tracking metadata for this inode
* This is required whenever we are about to make a structural change
* in the tree; renames, creation, deletion. */
void materializeDirAndParents();
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
/**
* Internal API only for use by InodeMap.
*
* InodeMap will call this API when a child inode needs to be loaded.
* The TreeInode will call InodeMap::inodeLoadComplete() or
* InodeMap::inodeLoadFailed() when the load finishes.
*/
void loadChildInode(PathComponentPiece name, fuse_ino_t number);
/**
* Unload all unreferenced children under this tree (recursively).
*
* This walks the children underneath this tree, unloading any inodes that
* are unreferenced.
*/
void unloadChildrenNow();
/**
* Load all materialized children underneath this TreeInode.
*
* This recursively descends into children directories.
*
* This method is intended to be called during the mount point initialization
* to trigger loading of materialized inodes. This allows other parts of the
* code to assume that materialized inodes are always loaded once the mount
* point has been initialized.
*
* Returns a Future that completes once all materialized inodes have been
* loaded.
*/
folly::Future<folly::Unit> loadMaterializedChildren();
// Helper functions to be used only by CheckoutAction
folly::Future<folly::Unit> checkoutReplaceEntry(
CheckoutContext* ctx,
InodePtr inode,
const TreeEntry& newScmEntry);
folly::Future<folly::Unit> checkoutRemoveChild(
CheckoutContext* ctx,
PathComponentPiece name,
InodePtr inode);
private:
class TreeRenameLocks;
class IncompleteInodeLoad;
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
void registerInodeLoadComplete(
folly::Future<std::unique_ptr<InodeBase>>& future,
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
PathComponentPiece name,
fuse_ino_t number);
void inodeLoadComplete(
PathComponentPiece childName,
std::unique_ptr<InodeBase> childInode);
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
folly::Future<std::unique_ptr<InodeBase>> startLoadingInodeNoThrow(
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
Entry* entry,
PathComponentPiece name,
fuse_ino_t number) noexcept;
folly::Future<std::unique_ptr<InodeBase>>
introduce a new InodeMap class Summary: This diff starts adding a new InodeMap class. This class will eventually consolidate the functionality of InodeNameMap plus the inode map stored in EdenDispatcher. This new class will bring several new benefits: - All inode mapping logic consolidated into a single class, with a single lock protecting the maps. - A well-defined model for loaded vs unloaded inodes. InodeMap explicitly tracks inodes that have InodeBase objects created for them vs inodes that have an inode number allocated (for FUSE) but do not have an InodeBase object in memory. This will make it possible to unload Inode objects on demand to reduce memory usage. - Tracking of pending loads, and de-duplicating load requests. This ensures that only one Inode object ever exists for a given inode number / path. If a second request to load an Inode arrives when a previous load request is still in progress, InodeMap deals with this situation properly. - Better support for using Inode objects without FUSE. With the old code, attempts to interact with Inode objects without going through the FUSE dispatch (say, when processing a thrift call) could result in inconsistent state. New inodes created would not be put into the EdenDispatcher map, which could result in problems. - More convenient child inode access from TreeInode. With this change, the TreeInode class can easily tell which of its children are loaded. This makes it easier to do tasks which only need to operate on existing loaded inode state (for instance, dirstate computation). - Support for saving and loading state, to implement graceful restart.. InodeMap provides a central place to write out inode state on shutdown and restoring it on startup. Saved inodes can easily be restored to an "unloaded" state on startup. This code is not implemented yet as part of this diff, but it should be straightforward to add in future diffs. Reviewed By: bolinfest Differential Revision: D4318060 fbshipit-source-id: d9b16430fc8367e3516e788d1e991e5163aa6997
2016-12-23 02:34:54 +03:00
startLoadingInode(Entry* entry, PathComponentPiece name, fuse_ino_t number);
folly::Future<folly::Unit> doRename(
TreeRenameLocks&& locks,
PathComponentPiece srcName,
PathMap<std::unique_ptr<Entry>>::iterator srcIter,
TreeInodePtr destParent,
PathComponentPiece destName);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
/** Translates a Tree object from our store into a Dir object
* used to track the directory in the inode */
static Dir buildDirFromTree(const Tree* tree);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
/**
* Get a TreeInodePtr to ourself.
*
* This uses TreeInodePtr::newPtrFromExisting() internally.
*
* This should only be called in contexts where we know an external caller
* already has an existing reference to us. (Which is most places--a caller
* has to have a reference to us in order to call any of our APIs.)
*/
TreeInodePtr inodePtrFromThis() {
return TreeInodePtr::newPtrFromExisting(this);
}
template <typename InodePtrType>
folly::Future<folly::Unit>
removeImpl(PathComponent name, InodePtr child, unsigned int attemptNum);
static void checkPreRemove(const TreeInodePtr& child);
static void checkPreRemove(const FileInodePtr& child);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
/**
* This helper function starts loading a currently unloaded child inode.
* It must be held with the contents_ lock held. (The Dir argument is only
* required as a parameter to ensure that the caller is actually holding the
* lock.)
*/
folly::Future<InodePtr> loadChildLocked(
Dir& dir,
PathComponentPiece name,
Entry* entry,
std::vector<IncompleteInodeLoad>* pendingLoads);
void computeCheckoutActions(
CheckoutContext* ctx,
const Tree* fromTree,
const Tree* toTree,
std::vector<std::unique_ptr<CheckoutAction>>* actions,
std::vector<IncompleteInodeLoad>* pendingLoads);
std::unique_ptr<CheckoutAction> processCheckoutEntry(
CheckoutContext* ctx,
Dir& contents,
const TreeEntry* oldScmEntry,
const TreeEntry* newScmEntry,
std::vector<IncompleteInodeLoad>* pendingLoads);
void saveOverlayPostCheckout(CheckoutContext* ctx, const Tree* tree);
eden: merge overlay into the inode objects Summary: It was starting to get pretty complex to manage locking across the inodes, filedata, overlay and soon the journal, so as a simplifying step, this folds data that was tracked by the overlay into the TreeInode itself. This is the first diff in a short series for this. This one: 1. Breaks the persistent overlay information, so shutting down eden and bringing it back up will lose your changes (to be restored in the following diff) 2. Allows deferring materialization of file data in more cases 3. Allows renaming dirs. The approach here is now to keep just one source of information about the directory contents; when we construct a TreeInode we import this data from the Tree and then apply mutations to it locally. Each inode can be mutated indepdently from others; we only need to lock the 1, 2 or 3 participating inodes in the various mutation operations. I'll tackle persistence of the mutations in the following diff, but the high level plan for that (to help understand this diff) is to always keep the directory inodes for mutations alive as inode objects. We make use of the canForget functionality introduced by D3774269 to ensure that these don't get evicted early. On startup we'll load this information from the overlay area. This model simplifies some of the processing around reading dirs and looking up children. Since the overlay data now tracks the appropriate tree or content hash we can be much more lazy at materializing data, especially in the rename case. For example, renaming "fbcode" to "fbcod" doesn't require us to recursively materialize the "fbcode" tree. Depends on D3653706 Reviewed By: simpkins Differential Revision: D3657894 fbshipit-source-id: d4561639845ca93b93487dc84bf11ad795927b1f
2016-09-10 02:56:00 +03:00
folly::Synchronized<Dir> contents_;
/** Can be nullptr for the root inode only, otherwise will be non-null */
TreeInode::Entry* entry_{nullptr};
};
}
}