Recently a customer workload encountered a data corruption in a specific multi-threaded write operation. The workload combined racing unaligned adjacent buffered writes with low memory conditions that caused both writeback and memory reclaim to race with the writes. The result of this was random partial blocks containing zeroes instead of the correct data. The underlying problem is that iomap caches the write iomap for the duration of the write() operation, but it fails to take into account that the extent underlying the iomap can change whilst the write is in progress. The short story is that an iomap can span mutliple folios, and so under low memory writeback can be cleaning folios the write() overlaps. Whilst the overlapping data is cached in memory, this isn't a problem, but because the folios are now clean they can be reclaimed. Once reclaimed, the write() does the wrong thing when re-instantiating partial folios because the iomap no longer reflects the underlying state of the extent. e.g. it thinks the extent is unwritten, so it zeroes the partial range, when in fact the underlying extent is now written and so it should have read the data from disk. This is how we get random zero ranges in the file instead of the correct data. The gory details of the race condition can be found here: https://lore.kernel.org/linux-xfs/20220817093627.GZ3600936@xxxxxxxxxxxxxxxxxxx/ Fixing the problem has two aspects. The first aspect of the problem is ensuring that iomap can detect a stale cached iomap during a write in a race-free manner. We already do this stale iomap detection in the writeback path, so we have a mechanism for detecting that the iomap backing the data range may have changed and needs to be remapped. In the case of the write() path, we have to ensure that the iomap is validated at a point in time when the page cache is stable and cannot be reclaimed from under us. We also need to validate the extent before we start performing any modifications to the folio state or contents. Combine these two requirements together, and the only "safe" place to validate the iomap is after we have looked up and locked the folio we are going to copy the data into, but before we've performed any initialisation operations on that folio. If the iomap fails validation, we then mark it stale, unlock the folio and end the write. This effectively means a stale iomap results in a short write. Filesystems should already be able to handle this, as write operations can end short for many reasons and need to iterate through another mapping cycle to be completed. Hence the iomap changes needed to detect and handle stale iomaps during write() operations is relatively simple.... However, the assumption is that filesystems should already be able to handle write failures safely, and that's where the second (first?) part of the problem exists. That is, handling a partial write is harder than just "punching out the unused delayed allocation extent". This is because mmap() based faults can race with writes, and if they land in the delalloc region that the write allocated, then punching out the delalloc region can cause data corruption. This data corruption problem is exposed by generic/346 when iomap is converted to detect stale iomaps during write() operations. Hence write failure in the filesytems needs to handle the fact that the write() in progress doesn't necessarily own the data in the page cache over the range of the delalloc extent it just allocated. As a result, we can't just truncate the page cache over the range the write() didn't reach and punch all the delalloc extent. We have to walk the page cache over the untouched range and skip over any dirty data region in the cache in that range. Which is .... non-trivial. That is, iterating the page cache has to handle partially populated folios (i.e. block size < page size) that contain data. The data might be discontiguous within a folio. Indeed, there might be *multiple* discontiguous data regions within a single folio. And to make matters more complex, multi-page folios mean we just don't know how many sub-folio regions we might have to iterate to find all these regions. All the corner cases between the conversions and rounding between filesystem block size, folio size and multi-page folio size combined with unaligned write offsets kept breaking my brain. Eventually, I realised that if the XFS code tracked the processed write regions by byte ranges instead of fileysetm block or page cache index, we could simply use mapping_seek_hole_data() to find the start and end of each discrete data region within the range we needed to scan. SEEK_DATA finds the start of the cached data region, SEEK_HOLE finds the end of the region. THese are byte based interfaces that understand partially uptodate folio regions, and so can iterate discrete sub-folio data regions directly. This largely solved the problem of discovering the dirty regions we need to keep the delalloc extent over. Of course, now xfs/196 fails. This is a error injection test that is supposed to exercise the delalloc extent recover code that the above fixes just completely reworked. the error injection assumes that it can just truncate the page cache over the write and then punch out the delalloc extent completely. This is fundamentally broken, and only has been working by chance - the chance is that writes are page aligned and page aligned writes don't install large folios in the page cache. IOWs, with sub-folio block size, and not know what size folios are in the cache, we can't actually guarantee that we can remove the cached dirty folios from the cache via truncation, and hence the new code will not remove the delalloc extents under those dirty folios. As a result the error injection results is writing zeroes to disk rather that removing the delalloc extents from memory. I can't make this error injection to work the way it was intended, so I removed it. The code that it is supposed to exercise is now exercised every time we detect a stale iomap, so we have much better coverage of the failed write error handling than the error injection provides us with, anyway.... So, this passes fstests on 1kb and 4kb block sizes and the data corruption reproducer does not detect data corruption, so this set of fixes is /finally/ something I'd consider ready for merge. Comments and testing welcome! -Dave. Version 4: - fix corner case where up to date blocks beyond EOF on a dirty folio don't get punched out. - add comment explaining the "scan for data only to EOF" logic that fixes the above corner case. - cleaned up some comments to be more clearer. - "loff_t pos, not offset" conversion. - use i_blocksize(inode) rather than open coding. - remove "if (seq)" check in xfs_iomap_write_direct() as seq is always set by callers. Version 3: - https://lore.kernel.org/linux-xfs/20221117055810.498014-1-david@xxxxxxxxxxxxx/ - Rearrange the deck chairs. - Remove mapping_seek_hole_data() export. - move code to iomap to allow mapping_seek_hole_data() not to be exported. - add export to iomap to allow filesystems access to functionality that uses mapping_seek_hole_data(). - add punch callback to iomap export to allow filesystem specific functionality for ranges found with mapping_seek_hole_data(). - call the new iomap export from from filesystem ->iomap_end callback so that iomap can call back into the filesystem again to do the stuff the filesystem needs to do. - Document that the iomap punch callback assumes that the filesystem must skip all extent types except for delalloc extents. - Document the lock order and limits on what the punch callback can actually do. - cleaned up xfs_iomap_valid(). Version 2: - https://lore.kernel.org/linux-xfs/20221115013043.360610-1-david@xxxxxxxxxxxxx/ - export mapping_seek_hole_data() - fix missing initialisation of the iomap sequence in xfs_fs_map_blocks() in the pnfs code. - move ->iomap_valid callback to the struct iomap_page_ops so that it is returned with the iomap rather than having the iomap_ops plumbed through the entire stack. - added a u64 validity_cookie to the struct iomap for carrying iomap verification information along with the iomap itself. - added IOMAP_F_XATTR for XFS to be able to tell the difference between iomaps that map attribute extents instead of file data extents. - converted the IOMAP_F_* flags to use the (1U << NN) definition pattern. - added patch to convert xfs_bmap_punch_delalloc_range() to take a byte range. Version 1: - https://lore.kernel.org/linux-xfs/20221101003412.3842572-1-david@xxxxxxxxxxxxx/ - complete rework of iomap stale detection - complete rework of XFS partial delalloc write error handling. Original RFC: - https://lore.kernel.org/linux-xfs/20220921082959.1411675-1-david@xxxxxxxxxxxxx/
