Skip to content

Est. 2011ยทMicrosoft Partner ID 7033487ยทDelivery under 3 minยทSupport 7 days a week

Your vault is empty.

License Error 0x00000025

NPFS_FILE_SYSTEM 0x00000025 and cache manager crashes on file servers

12 min read Updated October 5, 2026 Windows Crashes & Boot Recovery

Fix it now

Three of these four bug checks have one published cause: nonpaged pool memory has run out. The fourth, 0x000000EB, is a deadlock in the file system driver stack. Measure pool before you touch anything, because a scheduling change cannot fix a pool problem and a memory upgrade cannot fix a deadlock.

Run these on the server in an elevated PowerShell window, during the busy period

fltmc filters
Get-Counter '\Memory\Pool Nonpaged Bytes','\Memory\Available MBytes' -SampleInterval 10 -MaxSamples 30
Get-WinEvent -LogName System -MaxEvents 100 | Where-Object Id -eq 1001
  1. Watch Pool Nonpaged Bytes across the working day. A figure that climbs and never falls back is the documented cause of 0x25, 0x27 and 0x34, and Microsoft’s documented resolution is to add physical memory to the machine.
  2. Account for every third-party filter in the fltmc filters output. Each one belongs to a product that is consuming pool and inspecting the paths these bug checks live in.
  3. Bring the server up to the current cumulative update and update every storage, backup and endpoint agent to the build its vendor publishes for this Windows Server release.
  4. For 0x000000EB specifically, read the bug check parameters rather than the clock: parameter 1 is the total dirty pages, parameter 2 the dirty pages destined for the page file, and on current Windows parameter 4 is the most recent modified write error status.

Adding memory is Microsoft’s documented resolution for pool depletion, but it buys time rather than fixing a leak. If pool climbs steadily between restarts, find the driver that owns the growing pool tag first.

If pool now stays flat under load and the server survives a full backup cycle, you are done. If not, the next section separates the pool cases from the deadlock.

Why it happens

Named pipes look like files but are not stored anywhere. The named pipe file system driver presents them through the same interface as a disk file system so that ordinary file operations work against them, which is how remote procedure calls and a great deal of service communication are carried. The SMB redirector does the same trick for remote shares. Behind both sits the cache manager, which holds file data in memory and writes it back lazily.

All three are documented as failing the same way. Microsoft’s published cause for NPFS_FILE_SYSTEM, RDR_FILE_SYSTEM and CACHE_MANAGER is depletion of nonpaged pool memory – and it adds the detail that matters on a busy server: even when pool is not completely exhausted, a kernel-mode driver asking for nonpaged pool while very little is available can trigger the same stop. So the component named on the screen is often not the component at fault. It is simply the one that asked for memory at the wrong moment.

0x000000EB is different in kind and the difference is worth learning. DIRTY_MAPPED_PAGES_CONGESTION means no free pages are available to continue operations, and Microsoft’s stated cause is that the file system driver stack has deadlocked: most modified pages are destined for the file system, the file system is not operational, and none of those pages can be reused without losing data. Microsoft names two shapes for it – a driver that has blocked, deadlocking the modified or mapped page writers, and a storage driver that is not processing requests, with stranded queues or unresponsive drives – and says plainly that any file system or filter driver in the stack may be at fault.

That is why rescheduling the backup window is the wrong first move on this code. A deadlocked driver deadlocks whenever it is exercised; moving the work to three in the morning moves the crash to three in the morning.

Nonpaged pool is being exhausted

You have this one if 0x25, 0x27 or 0x34, and Pool Nonpaged Bytes climbs through the working day or across uptime without falling back.

  1. Measure first, with Performance Monitor or Get-Counter, across a full working cycle rather than at one moment.
  2. Identify the driver behind a growing pool tag using the pool monitoring tool from the Windows Driver Kit, and map the tag back to the driver that owns it.
  3. Update or remove that driver. A leaking third-party filter is the usual owner.
  4. Add physical memory where the workload has simply outgrown the machine. That is Microsoft’s documented resolution for pool depletion.

Confirm the paging file is present and sized so a dump can be written. Without one you get the same crash with nothing to read afterwards.

A third-party filter driver in the file or network path

You have this one if fltmc filters lists security, backup, replication or storage filters, and the crashes began after one was installed or upgraded.

  1. List the filters and identify the product behind each one.
  2. Install the build the vendor publishes for this exact Windows Server version; server support usually lags client support.
  3. Apply the vendor’s documented exclusions for file server roles, so the filter is not inspecting every client’s share traffic.
  4. Remove any filter belonging to a product that is no longer installed or no longer supported.

On a file server, a scanner that inspects every share access is inspecting every client’s traffic too. Exclusions here are about correctness, not only speed.

A driver in the stack has deadlocked or stalled

You have this one if 0x000000EB, with a large parameter 1 or parameter 2, and the crash arrives under sustained write load rather than at random.

  1. Read parameters 1 and 2 from the bug check. Microsoft ties large values there to a blocked driver or a storage driver that is not processing requests.
  2. Check the path to the storage for a degraded array, a failed multipath link or a saturated iSCSI network, since an unresponsive drive presents the same way as a blocked driver.
  3. Update the storage stack – HBA driver, multipath software, array firmware – to the versions the hardware vendor currently publishes.
  4. Where a virtual machine is involved, check the host’s storage queue as well as the guest’s; the guest can only see its own view.

The server is behind on updates

You have this one if No third-party filter stands out, pool is flat, and the machine has not been patched for several months.

  1. Apply the current cumulative update and any servicing stack update it depends on.
  2. Reboot into the patched build and watch for a full business cycle before concluding anything.
  3. Check the support pages for the agents on the machine too; their fixes often accompany a Windows one.

Full reference

The four codes side by side

Code Name Published meaning Published cause
0x00000025 NPFS_FILE_SYSTEM A problem occurred in the NPFS file system Depletion of nonpaged pool memory
0x00000027 RDR_FILE_SYSTEM A problem occurred in the SMB redirector file system Depletion of nonpaged pool memory
0x00000034 CACHE_MANAGER A problem occurred in the file system’s cache manager Depletion of nonpaged pool memory
0x000000EB DIRTY_MAPPED_PAGES_CONGESTION No free pages are available to continue operations The file system driver stack has deadlocked

On the first three, parameter 1 encodes a source file and line inside Windows: the high 16 bits identify the source file and the low 16 bits the line. That is useful to Microsoft support and to nobody else, which is another reason the answer is not in the parameters on those three.

Reading 0x000000EB’s parameters

Parameter Meaning
1 The total number of dirty pages
2 The number of dirty pages destined for the page file
3 Reserved on Windows Vista and later
4 The most recent modified write error status on Windows Vista and later

A large parameter 1 or 2 is what Microsoft points at a blocked driver or an unresponsive storage device. Parameter 4 is worth translating: a modified write error status names the failure the storage returned last, which turns a vague congestion story into a specific device.

Working out which driver owns the pool

  1. Add the Memory counters Pool Nonpaged Bytes and Available MBytes to a data collector set and let it run through a full working day.
  2. If nonpaged pool climbs and never falls back, enable pool tracking and use the pool monitoring tool from the Windows Driver Kit to list allocations by tag.
  3. Sort by bytes and watch which tag grows. The tag is four characters chosen by the driver author.
  4. Map the tag back to the driver that owns it by searching the driver binaries on the machine for that tag.
  5. Update or remove that driver, then repeat the measurement to confirm the growth has stopped.

What the filter list is telling you

fltmc is Microsoft’s filter manager control program, documented as the way to enumerate minifilter drivers, instances and volumes. On a file server the output is often longer than people expect, because backup, anti-malware, encryption, auditing, replication and storage products each install one. Every entry in that list is code sitting between the SMB redirector and the disk, consuming nonpaged pool and able to block the write-back path. That is the intersection of both failure modes in this article.

Do not remove a filter driver by deleting files or registry keys. Use the vendor’s own removal tool. A partially removed filter can leave an altitude registered with no driver behind it, and the machine will not start.

What will not help

  • Disabling the cache manager. There is no supported way to do it and the system depends on it for throughput.
  • Moving the backup window, on 0x000000EB. A deadlock recurs whenever the path is exercised.
  • Adding memory, on 0x000000EB. More pages do not unblock a driver that has stopped processing.
  • Rescheduling work, on 0x25, 0x27 and 0x34. A pool leak fills up on its own timetable.
  • Reinstalling Windows. None of the published causes for these four codes is Windows itself being damaged.

Before you conclude it is hardware

These four codes are the ones that most often send a team down the hardware route unnecessarily. Microsoft attributes all four to memory pressure or to drivers, not to failing components. Test hardware if you also see storage or memory errors in the System log, if other stop codes appear alongside these, or if the machine misbehaves under workloads that have nothing to do with file sharing. Otherwise the answer is in the pool counters and the filter list.

When a licence is the actual fix

Nothing in this article is fixed by buying something, with one exception: where the filter causing it belongs to an endpoint agent the vendor no longer builds for your Windows Server version, no amount of exclusion tuning helps, because the code in the kernel is the code that is wrong. Start with what you already have – Microsoft Defender Antivirus ships in Windows Server at no extra cost, and on a file server it is one fewer third-party filter in the stack. Move to a subscription only when you need central policy and reporting across an estate. If that is the case, note before you budget that Microsoft states plainly that onboarding a device running Windows Server needs a separate licence, such as Microsoft Defender for Business servers, on top of Defender for Business or Microsoft 365 Business Premium for the workstations. Arco supplies both parts and will count your servers separately from your desktops, because Microsoft does.

Every code this article covers

Code What it points at Source
0x00000025 NPFS_FILE_SYSTEM: a problem occurred in the NPFS file system. Documented cause is depletion of nonpaged pool memory Microsoft Learn
0x00000027 RDR_FILE_SYSTEM: a problem occurred in the SMB redirector file system. Same documented cause, nonpaged pool depletion Microsoft Learn
0x00000034 CACHE_MANAGER: a problem occurred in the file system’s cache manager. Same documented cause Microsoft Learn
0x000000EB DIRTY_MAPPED_PAGES_CONGESTION: no free pages are available to continue operations because the file system driver stack has deadlocked, or a storage driver is not processing requests Microsoft Learn

Confirm the fix worked

  1. Pool Nonpaged Bytes stays flat across a full working day rather than climbing.
  2. A complete backup cycle and a full scan run in a maintenance window without the server restarting.
  3. fltmc filters lists only filters belonging to products that are installed, supported and current for this Windows Server version.
  4. No new BugCheck events with ID 1001 appear in the System log over a fortnight.
  5. The paging file is present and large enough that a dump would be written if it happened again.

Questions people ask about this

Is this hardware failing?

Microsoft’s published causes for all four codes are memory pressure and drivers, not failing components. Check pool, the filter list and the patch level first. Hardware becomes worth testing if the System log also carries disk or memory errors, or if other stop codes appear alongside these.

Will adding RAM fix it?

For 0x25, 0x27 and 0x34, adding physical memory is Microsoft’s documented resolution, because it increases the nonpaged pool available to the kernel. For 0x000000EB it will not, because that is a deadlock rather than a shortage. And on any of them, a genuine leak will eventually consume whatever you add.

Can I stop the crash by turning the cache off?

No, and there is no supported way to do it. The cache manager is not optional. The workable levers are reducing pool pressure, fixing the driver that is leaking or blocking, and making the storage behind the cache respond.

Why does it only happen on the file server?

Because that is the machine exercising the paths. The same agent on a machine with light file access never reaches the code that fails. That is a good reason to treat file servers as their own configuration case rather than applying the workstation policy to them.

How do I tell a pool problem from a deadlock without a debugger?

By the counter and the code. If Pool Nonpaged Bytes climbs across uptime and the stop code is 0x25, 0x27 or 0x34, it is pool. If the code is 0x000000EB and the crash arrives under sustained write load with pool healthy, it is the driver stack, and the bug check parameters are where to look next.

Related error codes

Was this article helpful?

Your feedback helps us improve our documentation.

Related articles

License Error PAGE_FAULT_IN_NONPAGED_AREA 0x00000050 – what the stop code really means License Error NTFS_FILE_SYSTEM 0x00000024 – volume corruption behind the blue screen License Error RESOURCE_OWNER_POINTER_INVALID 0x00000132 and lock ownership crashes License Error ELAM_DRIVER_DETECTED_FATAL_ERROR 0x00000178 and security subsystem crashes
โ† Back to Knowledge Base