RAID 6 Data Recovery Services
RAID 6 uses dual parity to survive two simultaneous drive failures. But a long rebuild on large drives puts heavy stress on survivors that share the failed drive's age and wear. We recover RAID 6 volumes by imaging each member through a write-blocked channel and reconstructing the array offline. The rebuild never runs. If you manage a degraded or failed RAID 6 array, start with our RAID data recovery service overview or contact us directly for a free evaluation. No data, no charge.
Nothing gets rebuilt in place. Every member is imaged independently through a write-blocked channel on PC-3000 Portable III or DeepSpar Disk Imager, and the array geometry is then reconstructed in software against those clone files with Data Extractor Express RAID Edition, so no write reaches the original drives.
Per-member imaging is billed on the published HDD tiers, from $250 to $900 per drive for logical & firmware work and $1,200–$1,500 for a member that needs a head swap, donor drive billed separately, plus $400-$800 for the array reconstruction itself. There is no diagnostic fee, & if we recover nothing you don't owe a recovery fee. Ship the members in from anywhere in the country; imaging through reconstruction runs in-house at the Austin, TX lab.
How Does Dual Parity Make RAID 6 Different?
- P parity is a bitwise XOR of all data blocks in a stripe. It is the same single-parity calculation used in RAID 5 and can reconstruct any one missing block per stripe.
- Because P and Q are linearly independent, the controller can solve a system of two equations to reconstruct two missing blocks per stripe.
- We read the parity rotation, block size and member order from the on-disk metadata, or work them out from the raw drive images. Get any one of them wrong and the reconstruction comes out garbled.
- Usable capacity equals (N-2) times the smallest member size. A six-drive array of 12 TB members yields 48 TB of usable space, with 24 TB dedicated to parity.
- Every data write updates both P and Q, so RAID 6 writes are slower than RAID 5.
Reed-Solomon Q Parity in Linux md
- P Parity (XOR)
- P = d_0 ⊕ d_1 ⊕ d_2 ⊕ ... ⊕ d_(n-1). XOR is linear and reversible, so any single missing block is recovered by XORing the remaining data blocks with P. XOR alone cannot solve for two unknowns; a second, independent equation is required.
- Galois Field GF(2^8)
- Linux md does its Q math in this 256-element field. Each byte is a polynomial with binary coefficients, and addition and subtraction are both XOR.
- Q Parity Equation
- In Linux md, Q = g^0·D_0 + g^1·D_1 + g^2·D_2 + ... + g^(n-1)·D_(n-1), where + is XOR and g is a generator of GF(2^8). Each data block gets a different multiplier, so Q stays independent of P. When two drives fail, md solves the two equations for the two unknowns.
- Parity Rotation Pattern
- mdadm's standard RAID 6 layouts rotate P and Q across the drives. Its conversion layouts, such as left-symmetric-6, keep Q on the last device. We read the rotation, block size and data start offset from the on-disk metadata, or work them out from the raw images. Get any of them wrong and you end up with a reconstruction that's mathematically valid and unreadable.
URE Risk During RAID 6 Rebuilds
- Say the array has already lost one drive and hits a URE on a surviving member during the rebuild. That stripe is now missing two elements, which is what dual parity is built to cover. P and Q together solve for both, and the rebuild keeps going. The array only goes past its tolerance when a second member is already fully gone, leaving no parity to cover the URE, or when the long rebuild load kills a third same-batch drive mechanically.
- Rebuild time adds to the risk, because the array runs degraded for the whole rebuild. Every hour it runs degraded is another hour of exposure to heat, vibration and workload stress on aging drives. Those drives have the same power-on hours as the one that already failed.
- A hot spare doesn't remove this risk. It cuts the delay before a rebuild starts, but the rebuild still has to read the surviving members. If the hot spare kicks off an automatic rebuild while a second member is already marginal, the rebuild can push that member over the edge.
Critical: If your degraded RAID 6 holds data you can't replace and you don't have a verified backup, don't force a rebuild. Power the system down, label each drive with its slot position, and contact us. Every rebuild attempt on marginal hardware raises the chance of losing the data for good.
URE Probability Scales with Array Size
Consumer SATA drives carry a manufacturer-specified URE rate of 1 sector per 1014 bits read; that's roughly 12.5 TB of data before a single unreadable sector is expected at the worst-case spec rate.
When a six-drive RAID 6 array of 8 TB consumer members loses one disk, the controller must read 40 TB (3.2 × 1014 bits) from the five surviving drives. Applied as a worst-case upper bound, the binomial probability of hitting at least one URE during that read approaches 96%; on a twelve-drive array of 16 TB members reading 176 TB it is higher still.
Real drives usually read far better than that spec. The risk climbs as they age.
One URE on a survivor during a one-disk-down RAID 6 rebuild isn't the end of the array. P and Q together cover that stripe, so the rebuild reconstructs it and keeps going.
The escalation to a full hard drive data recovery scenario happens when a second member is already gone (no parity left to cover the URE) or when the sustained rebuild load fails a third same-batch survivor mechanically. That correlated mechanical failure, not the first read error, is the real risk.
Punctured Stripes on PERC and MegaRAID Controllers
Say a rebuild hits an unreadable sector on a surviving drive and there's no redundancy left to fill it in. Dell PERC and LSI/Broadcom MegaRAID controllers don't stop there. They write a puncture for that stripe onto the target drive and keep going. The rebuild finishes and the virtual drive comes back online.
Cisco's support article on MegaRAID controllers says a patrol read can do this too, not only a rebuild. The point of the puncture is that the controller never uses that block's bad parity again. Dell calls the same thing rebuild with errors.
Does a New Drive or a Second Rebuild Clear a Puncture?
No. The data that belonged in that stripe isn't on any member anymore, so there's nothing for parity to rebuild it from.
Dell's documented fix for a punctured array is to delete the array, create it again and run a full initialization. That wipes the virtual drive. It's how you get a clean array to restore a backup onto. It isn't a way to get files back.
HP Smart Array P-series and E-series controllers don't puncture the way PERC and MegaRAID do. Either way, you don't need the original card to get at the array. PERC and MegaRAID write the geometry to the member drives as a SNIA DDF variant on the trailing sectors. HP Smart Array writes its own RAID Information Sector at the start of each drive.
Lab Route for a Punctured RAID 6
If the data is irreplaceable and there's no verified backup, don't run a second rebuild on the live drives. A rebuild writes to the target drive, and that's where the puncture goes.
We clone every member through a write-blocked PC-3000 Portable III or DeepSpar Disk Imager first. Then we rebuild the array offline from those images in Data Extractor Express RAID Edition. The original drives aren't written to at any point.
SMR Drives in RAID 6 Arrays: A Rebuild Hazard
A DM-SMR (Drive-Managed Shingled Magnetic Recording) drive overlaps its tracks to fit more data on each platter.
When a RAID 6 array rebuilds onto a DM-SMR member, the sustained writes fill the drive's small CMR media cache. The drive then stalls for tens of seconds while it works through it. NAS and enterprise drives cap internal error recovery at about 7 seconds through TLER/ERC. To the controller, a stall that long looks like a dead drive, so it ejects a physically healthy drive mid-rebuild.
Specific WD Red models shipped as DM-SMR:
- WD40EFAX (4 TB)
- WD60EFAX (6 TB)
- WD20EFAX (2 TB)
If your degraded RAID 6 has DM-SMR members and holds data you can't replace with no verified backup, don't attempt a rebuild. Ship us the drives. We image each member on its own with PC-3000, then reconstruct the array from the clones. A stalled or aborted resilver is its own failure mode, and our RAID rebuild failure recovery page covers it.
Our RAID 6 Recovery Process
- Free evaluation and configuration audit: We document the controller type (hardware RAID card, mdadm, ZFS, Btrfs, or Synology SHR-2, which is a thin management layer over standard Linux mdadm RAID plus LVM), member count, stripe size, and parity rotation direction.
- Write-blocked forensic imaging: Each member drive goes on PC-3000 or DeepSpar imaging hardware through a write-blocked channel. We clone the full LBA range of every member, including the trailing sectors where PERC and MegaRAID keep their DDF metadata. A drive that needs a head swap gets it on our 0.02 micron ULPA-filtered clean bench before imaging.
- RAID parameter detection: We work out the stripe block size, parity rotation pattern, member order and data start offset from the on-disk metadata or the raw member images.
- Virtual array assembly: The cloned images are loaded into forensic reconstruction software, which assembles the virtual stripe map from the detected parameters.
- Filesystem extraction: Once the virtual array is built, we mount or parse the filesystem (EXT4, XFS, Btrfs, ZFS) and copy the files out to target media.
Software Assembly From Cloned Images
Once we know the parameters, we assemble the virtual array entirely in software from the cloned images. Nothing we do writes to your original drives.
How Much Does RAID 6 Recovery Cost?
| Recovery factor | Price / detail |
|---|---|
| Per-member logical or firmware imaging | $250 to $900 per drive. Covers filesystem corruption and firmware module damage requiring PC-3000 terminal access. |
| Per-member mechanical (head swap), air-filled | $1,200–$1,500 per drive with a 50% deposit. We do the head swap on our 0.02 micron ULPA-filtered clean bench, then clone the drive write-blocked. |
| Helium-sealed member | $3,000–$4,500 head swap or $4,000–$5,000 surface damage, plus helium and donor costs. |
| Array reconstruction | $400-$800. See how we price NAS and RAID recovery. |
No Data = No Charge: If we recover nothing from your array, you owe nothing. Free evaluation, no obligation.
How Parity Rotation Affects RAID 6 Recovery
Recovery depends on getting the parity rotation right. We read it from the on-disk metadata or work it out from the raw member images. The wrong pattern gives you unreadable output.
- When a RAID 6 array runs on mdadm (Linux software RAID), each member's superblock stores the layout type, chunk size and member order. Where that superblock sits depends on the metadata version: 0.90 and 1.0 put it at the end of the device, 1.1 at the start, and 1.2 4 KiB from the start. With intact superblocks, parameter detection is fast. When they're damaged or overwritten, as happens in an accidental reinitialization, we work the parameters out from byte-level patterns across the raw member images.
- The btrfs documentation says its RAID56 profiles shouldn't be used in production, only for evaluation or testing. It also says power failure safety for RAID56 metadata isn't 100%.
NAS SSD Caching Tier Failures in RAID 6 Arrays
On Synology Plus units, an NVMe drive used as a read-write cache holds writes that haven't reached the hard drives yet. If it drops off the PCIe bus, those uncommitted writes are gone. The Btrfs volume on the hard drives is left incomplete, and Synology reports it as Volume Crashed.
Stale Drive Forced Online: Parity Overwrite and Silent Corruption
What Happens When a Stale Drive Is Forced Online?
mdadm's --assemble --force will assemble the array even when the metadata on some devices looks out of date.
Broadcom's MegaRAID guide says a consistency check that finds a difference between data and parity assumes the data is accurate and corrects the parity. With a stale member in the array, the check makes parity match the stale blocks. The file system can then end up pointing at blocks that hold older content.
Controller-Specific Failure Vectors
Dell PERC Controllers
A drive that's removed and reinserted later comes up as a foreign configuration. Broadcom's guide says an import automatically rebuilds the drives that were pulled before the virtual drive went offline. That rebuild writes to them, which is the last thing you want while the array is the only copy of your data.
Broadcom / LSI MegaRAID
MegaRAID marks a drive Unconfigured Bad when the firmware detects an unrecoverable error on it. A member that's pulled and reinserted comes back as a foreign configuration instead. Broadcom's guide warns that forcing a drive online that's part of a redundant array isn't recommended.
Linux mdadm
mdadm records array state in each member's superblock, so a member that dropped offline carries metadata that's out of date compared with the others.
mdadm's man page warns that an array which needed --force to start may contain data corruption.
Where the superblocks themselves are damaged or zeroed, our mdadm superblock reconstruction workflow recovers the layout from the raw member images.
How We Detect Stale Members
After imaging each drive through write-blocked channels on PC-3000, we analyze the raw hex data stripe by stripe. In a synchronized array, the XOR sum of all data blocks and the P-parity block equals zero.
A non-zero result flags a desynchronized stripe. Then we check the P-syndrome against the Q-syndrome to find which member contributed the stale data, and we leave that member out of the virtual reconstruction.
Do not run a consistency check or rebuild after forcing a stale drive online. A consistency check rewrites parity wherever it disagrees with the data. With a stale drive in the array, parity gets rewritten to match the stale blocks. Power down, write down the drive slot positions, and contact us. Our technical reference explains why rebuilds destroy degraded arrays.
Sequential vs. Simultaneous Double-Disk Failures
| Failure Mode | What Happens | Recovery Approach |
|---|---|---|
| Simultaneous failure (power surge, backplane short, or shared vibration event) | Both drives stop at the same logical point in time. | We image each surviving member through write-blocked channels on PC-3000, then use the two independent parity equations to algebraically reconstruct the two missing blocks per stripe. This is the scenario RAID 6 was designed for. |
| Sequential failure | Sequential failure is worse. The first drive drops offline and misses every write after that. If the array keeps running degraded for hours or days, the stale drive's blocks drift away from the live array. When the second drive fails, an administrator may try forcing the stale first drive back online as a stopgap. That puts outdated blocks back into the live array, and a consistency check run afterwards rewrites parity to match them. | We image all members (including the stale drive) independently. Working from the images, we compare the RAID metadata on each member and check parity stripe by stripe to find which blocks on the stale drive are outdated. We leave those blocks out of the virtual reconstruction and fill the gaps from the remaining valid parity. For failed servers in general, see our server data recovery page. |
Why Is a Two-Disk-Degraded RAID 6 Recoverable When a Two-Disk-Degraded RAID 5 Is Not?
The difference is linear algebra, not hardware. RAID 5 carries a single parity strip, P, computed as the XOR of every data block in the stripe. XOR resolves exactly one unknown per stripe, so a one-disk-down RAID 5 reconstructs cleanly. The moment a second member is gone, that stripe holds two unknowns and only one equation. No algebra resolves two unknowns from a single equation, so the second missing data column has no recovery path from parity alone.
A genuine double-disk failure is therefore a different recovery posture on RAID 5 than on RAID 6; see our RAID 5 recovery and two-drives-failed recovery pages.
RAID 6 carries a second parity strip, Q, that's independent of P. In Linux md, Q is a Reed-Solomon syndrome over GF(2^8), and each data block carries a distinct generator multiplier. P and Q together are two equations in two unknowns per stripe. When two members drop, the surviving data plus the surviving P and Q syndromes still solve for both missing blocks.
A two-down RAID 6 has zero remaining redundancy, so a third loss before reconstruction is unrecoverable, but the array is logically complete and reconstructable offline from member images rather than from a live rebuild.
- Two-down RAID 5 (one equation)
- Single parity P (XOR) covers one unknown per stripe. Losing two members leaves a full data column with no equation that defines it. Parity alone cannot reconstruct it. The array can sometimes still be partially recovered if one of the two members is only logically dropped and images cleanly, which is exactly why imaging precedes any rebuild attempt.
- Two-down RAID 6 (two equations)
- P and Q are linearly independent. Two equations in two unknowns solve for both missing blocks per stripe using finite-field algebra. The data is all there across the surviving members and parity. The job is to destripe it correctly, not to recover lost information.
Imaging Both Parity Strips Before Any Reconstruction
Because the two-down recoverability depends entirely on preserving the P and Q equations, every member is imaged through a write-blocked imager before any reconstruction is attempted. PC-3000 Express and DeepSpar Disk Imager read and clone the full LBA range of each member, capturing both parity strips exactly as they sat when the array stopped.
We image the members before any parity recalculation, so P and Q stay untainted. That keeps the exact two-equation system we need to destripe the array.
Then we run the reconstruction in Data Extractor Express RAID Edition against the image files, never against the live members.
You don't need the original controller. Dell PERC and Broadcom MegaRAID write the array geometry as a SNIA DDF variant on the trailing sectors of each member. HP Smart Array writes its RAID Information Sector (RIS) to the start of each member drive. Either way, the layout travels with the drives.
Each member is imaged and priced individually at the standard HDD tiers, then the virtual reassembly is performed in-house at our Austin lab. If we recover nothing, you owe nothing, and there is no diagnostic fee. Offline RAID reconstruction is the safe path when the data is irreplaceable and there's no verified backup.
RAID 6 Recovery Questions
How many drives can fail in RAID 6 before data is lost?
What is the difference between RAID 5 and RAID 6?
Why is RAID 6 rebuild so risky on large drives?
My RAID 6 has a hot spare and started an automatic rebuild. Should I let it finish?
What does offline RAID 6 reconstruction mean, and why is it safer than a rebuild?
My RAID controller shows a 'Foreign Configuration' with stale drives after a power failure. Is it safe to import?
Why does my RAID 6 array show valid parity but corrupt files after forcing a drive online?
Can an SMR drive like a WD Red be used to replace a failed drive in a RAID 6?
How do helium-filled enterprise drives complicate RAID 6 recovery?
Can data be recovered if a RAID 6 rebuild fails partway through?
What is the difference between sequential and simultaneous double-disk failure in RAID 6?
How do controller-specific parity rotations affect RAID 6 recovery?
What mathematical algorithm does RAID 6 use for its second parity block?
Why can a two-drive failure be recovered on RAID 6 but not on RAID 5?
Data Recovery Standards & Verification
Our Austin lab operates on a transparency-first model. We use industry-standard recovery tools, including PC-3000 and DeepSpar, combined with strict environmental controls to maintain drive integrity. This approach allows us to serve clients nationwide with consistent technical standards.
Localized Clean Zone
Open-drive work is performed in a 0.02 micron ULPA-filtered laminar clean bench.
Transparent History
Serving clients nationwide via mail-in service since 2008. Our lead engineer holds PC-3000 and HEX Akademia certifications for hard drive firmware repair and mechanical recovery.
Media Coverage
Our repair work has been covered by The Wall Street Journal and Business Insider, with CBC News reporting on our pricing transparency. Louis Rossmann has testified in Right to Repair hearings in multiple states and founded the Repair Preservation Group.
Aligned Incentives
Our "No Data, No Charge" policy means we assume the risk of the recovery attempt, not the client.
Technical Oversight
Louis Rossmann
Our engineers review all lab protocols to maintain technical accuracy and honest service. Since 2008, his focus has been on clear technical communication and accurate diagnostics rather than sales-driven explanations.
We believe in showing the bench rather than just describing it. Open-drive work runs on a 0.02 micron ULPA-filtered laminar clean bench, and we filmed it.
See the particle counter test at the benchSince 2008
Established
As Featured In
Related services
Need Recovery for Other Devices?
All RAID levels: 0, 1, 5, 6, 10, 60
Multi-span striped RAID 6 arrays
Dell, HP, IBM enterprise servers
Synology, QNAP, Buffalo NAS
Mechanical HDD recovery
Stalled or aborted array resilver
Double-disk failure in parity arrays
Degraded RAID 6? Stop the rebuild.
Free evaluation. Offline reconstruction from cloned images. No data = no charge. Mail-in from anywhere in the U.S.