Emc IP Holding Company LLC

Emc Corporation

Dixit Patel

Jean-Pierre Bono

William C Davenport

Srinivasa Rao Vempati

Miles Aram De Forest

Ming Yang

Morgan A Clark

Ajay Potnis

Kumari Bijayalaxmi Nanda

Miles Aram DeForest

Amit Dharmadhikari

Jean Pierre Bono

Sairam Iyer

Praveen S Solanki

Rahul Bhagwat

Techniques for performing online recovery of space accounting metadata. The techniques include reestablishing a plurality of counters for tracking amounts of physical storage space committed and/or unique to the volumes and/or volume families. The techniques include, while reestablishing the counters, conducting a tree walk through a mapping hierarchy of each volume and/or volume family, marking starting and ending points through the mapping hierarchy, and monitoring a logical offset from the starting point during the tree walk. Upon receipt of an IO request, the techniques determine whether the IO request and space accounting metadata recovery activities are attempting to access the same metadata based on the logical offset relative to the starting point, and update the counters and/or temporary versions of the counters, as appropriate. By providing such techniques, metadata inconsistencies can be addressed while allowing users of the data storage system full or at least partial access to their data.

Online recovery approach to space accounting

A separate family space is provisioned as a separate fault domain for each production device in a storage system. The space provisioned for each family has contiguous logical block addresses that do not overlap with the address space of any other family. Snaps and clones in a first family space are used for fault recovery without interrupting IOs to a second family space. If the first family space points to the same virtual block as other families as a result of deduplication then the virtual block is un-deduped. For example, another instance of the virtual block may be created and pointers updated such that only the first family space points to one of the virtual block instances.

Reducing the size of fault domains

A technique operable in a data storage system performs initialization of storage structures in multiple stages and allows hosts to request access to a data object even when some of the storage structures supporting the data object have not been initialized. Rather than initializing all storage structures for a data object before allowing host access, a storage processor instead initializes only a protocol stub structure. The protocol stub structure is configured to receive protocol requests from hosts and to initiate further processing should a storage request in that protocol be received. The further processing includes, in response to receiving a protocol request directed to a host-accessible data object, identifying storage structures involved in processing that protocol request and directing the storage processor to initialize those storage structures. Once the storage structures have been initialized, the storage processor may process the protocol request to completion via the initialized storage structures.

Demand-driven initialization of data storage structures

A method is used in managing file system checking. A logical space is created from a set of cylinder groups. The logical space includes respective unused blocks in each cylinder group of the set of cylinder groups. Each cylinder group of the set of cylinder groups is configured to store metadata and data of a file system. Information required to perform file system checking on the file system is stored in the logical space. File system checking is performed on the file system.

Managing file system checking

An improved technique involves providing read-only access to the file system while FSCK is running. After a storage processor detects corrupted data in the file system, the storage processor brings the file system offline and provides FSCK with read-only access to the file system. The storage processor then begins a repair process on the disk by starting to run FSCK, which writes corrupted blocks to a secondary storage device. Upon receiving a request from a user for access to a file in the file system, the storage processor mounts the file system with read-only access. Upon completion of FSCK, the storage processor brings the file system offline, applies the corrected blocks from secondary storage to the blocks in primary storage, and brings the file system online with full read and write access.

Accessing a file system during a file system check

A file server architecture decouples logical storage from physical storage and provides proactive detection and containment of faults, errors, and corruptions in a file system, in order to enable in place (online) and non-intrusive recovery. The file system is built upon a thinly provisioned logical volume, and there are stored three copies of the metadata defining the logical volume in order to provide quick, deterministic, and reliable recovery from a faulted system. A first copy of the metadata is distributed among all of the slices of physical storage allocated to the logical volume. A second copy of the metadata is stored in a root slice of the logical volume. A third copy of the metadata is stored separate from the slices of physical storage allocated to the logical volume.

Deterministic recovery of a file system built on a thinly provisioned logical volume having redundant metadata

To recover from partial writes of metadata transactions to a file system log, a redundancy check is added to each log record. Upon re-boot of the file server, the file system is recovered by finding the last valid record in the log, and then replaying records from the log into the on-disk file system. The replay is terminated prior to the last valid record upon reaching any record with a bad redundancy check. A 'fsck' is performed on the file system prior to granting client access to the file system only if the log recovery was terminated prior to the last valid record upon reaching a record with a bad redundancy check and the position of the records in the log indicates that this record with a bad redundancy check could not have been written to the log concurrently with the last valid record.

Growing community of inventors

Monroe, NJ, United States of America

Dixit Patel