kernel.git/fs/dax.c, branch linux-4.12.y

dax: fix deadlock due to misaligned PMD faults

2017-08-30T08:26:33Z

commit fffa281b48a91ad6dac1a18c5907ece58fa3879b upstream. In DAX there are two separate places where the 2MiB range of a PMD is defined. The first is in the page tables, where a PMD mapping inserted for a given address spans from (vmf->address & PMD_MASK) to ((vmf->address & PMD_MASK) + PMD_SIZE - 1). That is, from the 2MiB boundary below the address to the 2MiB boundary above the address. So, for example, a fault at address 3MiB (0x30 0000) falls within the PMD that ranges from 2MiB (0x20 0000) to 4MiB (0x40 0000). The second PMD range is in the mapping->page_tree, where a given file offset is covered by a radix tree entry that spans from one 2MiB aligned file offset to another 2MiB aligned file offset. So, for example, the file offset for 3MiB (pgoff 768) falls within the PMD range for the order 9 radix tree entry that ranges from 2MiB (pgoff 512) to 4MiB (pgoff 1024). This system works so long as the addresses and file offsets for a given mapping both have the same offsets relative to the start of each PMD. Consider the case where the starting address for a given file isn't 2MiB aligned - say our faulting address is 3 MiB (0x30 0000), but that corresponds to the beginning of our file (pgoff 0). Now all the PMDs in the mapping are misaligned so that the 2MiB range defined in the page tables never matches up with the 2MiB range defined in the radix tree. The current code notices this case for DAX faults to storage with the following test in dax_pmd_insert_mapping(): if (pfn_t_to_pfn(pfn) & PG_PMD_COLOUR) goto unlock_fallback; This test makes sure that the pfn we get from the driver is 2MiB aligned, and relies on the assumption that the 2MiB alignment of the pfn we get back from the driver matches the 2MiB alignment of the faulting address. However, faults to holes were not checked and we could hit the problem described above. This was reported in response to the NVML nvml/src/test/pmempool_sync TEST5: $ cd nvml/src/test/pmempool_sync $ make TEST5 You can grab NVML here: https://github.com/pmem/nvml/ The dmesg warning you see when you hit this error is: WARNING: CPU: 13 PID: 2900 at fs/dax.c:641 dax_insert_mapping_entry+0x2df/0x310 Where we notice in dax_insert_mapping_entry() that the radix tree entry we are about to replace doesn't match the locked entry that we had previously inserted into the tree. This happens because the initial insertion was done in grab_mapping_entry() using a pgoff calculated from the faulting address (vmf->address), and the replacement in dax_pmd_load_hole() => dax_insert_mapping_entry() is done using vmf->pgoff. In our failure case those two page offsets (one calculated from vmf->address, one using vmf->pgoff) point to different order 9 radix tree entries. This failure case can result in a deadlock because the radix tree unlock also happens on the pgoff calculated from vmf->address. This means that the locked radix tree entry that we swapped in to the tree in dax_insert_mapping_entry() using vmf->pgoff is never unlocked, so all future faults to that 2MiB range will block forever. Fix this by validating that the faulting address's PMD offset matches the PMD offset from the start of the file. This check is done at the very beginning of the fault and covers faults that would have mapped to storage as well as faults to holes. I left the COLOUR check in dax_pmd_insert_mapping() in place in case we ever hit the insanity condition where the alignment of the pfn we get from the driver doesn't match the alignment of the userspace address. Link: http://lkml.kernel.org/r/20170822222436.18926-1-ross.zwisler@linux.intel.com Signed-off-by: Ross Zwisler Reported-by: "Slusarz, Marcin" Reviewed-by: Jan Kara Cc: Alexander Viro Cc: Christoph Hellwig Cc: Dan Williams Cc: Dave Chinner Cc: Matthew Wilcox Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Greg Kroah-Hartman

fs/dax.c: fix inefficiency in dax_writeback_mapping_range()

2017-06-23T23:15:55Z

dax_writeback_mapping_range() fails to update iteration index when searching radix tree for entries needing cache flushing. Thus each pagevec worth of entries is searched starting from the start which is inefficient and prone to livelocks. Update index properly. Link: http://lkml.kernel.org/r/20170619124531.21491-1-jack@suse.cz Fixes: 9973c98ecfda3 ("dax: add support for fsync/sync") Signed-off-by: Jan Kara Reviewed-by: Ross Zwisler Cc: Dan Williams Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

dax: fix race between colliding PMD & PTE entries

2017-06-02T22:07:37Z

We currently have two related PMD vs PTE races in the DAX code. These can both be easily triggered by having two threads reading and writing simultaneously to the same private mapping, with the key being that private mapping reads can be handled with PMDs but private mapping writes are always handled with PTEs so that we can COW. Here is the first race: CPU 0 CPU 1 (private mapping write) __handle_mm_fault() create_huge_pmd() - FALLBACK handle_pte_fault() passes check for pmd_devmap() (private mapping read) __handle_mm_fault() create_huge_pmd() dax_iomap_pmd_fault() inserts PMD dax_iomap_pte_fault() does a PTE fault, but we already have a DAX PMD installed in our page tables at this spot. Here's the second race: CPU 0 CPU 1 (private mapping read) __handle_mm_fault() passes check for pmd_none() create_huge_pmd() dax_iomap_pmd_fault() inserts PMD (private mapping write) __handle_mm_fault() create_huge_pmd() - FALLBACK (private mapping read) __handle_mm_fault() passes check for pmd_none() create_huge_pmd() handle_pte_fault() dax_iomap_pte_fault() inserts PTE dax_iomap_pmd_fault() inserts PMD, but we already have a PTE at this spot. The core of the issue is that while there is isolation between faults to the same range in the DAX fault handlers via our DAX entry locking, there is no isolation between faults in the code in mm/memory.c. This means for instance that this code in __handle_mm_fault() can run: if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { ret = create_huge_pmd(&vmf); But by the time we actually get to run the fault handler called by create_huge_pmd(), the PMD is no longer pmd_none() because a racing PTE fault has installed a normal PMD here as a parent. This is the cause of the 2nd race. The first race is similar - there is the following check in handle_pte_fault(): } else { /* See comment in pte_alloc_one_map() */ if (pmd_devmap(*vmf->pmd) || pmd_trans_unstable(vmf->pmd)) return 0; So if a pmd_devmap() PMD (a DAX PMD) has been installed at vmf->pmd, we will bail and retry the fault. This is correct, but there is nothing preventing the PMD from being installed after this check but before we actually get to the DAX PTE fault handlers. In my testing these races result in the following types of errors: BUG: Bad rss-counter state mm:ffff8800a817d280 idx:1 val:1 BUG: non-zero nr_ptes on freeing mm: 15 Fix this issue by having the DAX fault handlers verify that it is safe to continue their fault after they have taken an entry lock to block other racing faults. [ross.zwisler@linux.intel.com: improve fix for colliding PMD & PTE entries] Link: http://lkml.kernel.org/r/20170526195932.32178-1-ross.zwisler@linux.intel.com Link: http://lkml.kernel.org/r/20170522215749.23516-2-ross.zwisler@linux.intel.com Signed-off-by: Ross Zwisler Reported-by: Pawel Lebioda Reviewed-by: Jan Kara Cc: "Darrick J. Wong" Cc: Alexander Viro Cc: Christoph Hellwig Cc: Dan Williams Cc: Dave Hansen Cc: Matthew Wilcox Cc: "Kirill A . Shutemov" Cc: Pawel Lebioda Cc: Dave Jiang Cc: Xiong Zhou Cc: Eryu Guan Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

Merge branch 'akpm' (patches from Andrew)

2017-05-13T16:49:35Z

Merge misc fixes from Andrew Morton: "15 fixes" * emailed patches from Andrew Morton : mm, docs: update memory.stat description with workingset* entries mm: vmscan: scan until it finds eligible pages mm, thp: copying user pages must schedule on collapse dax: fix PMD data corruption when fault races with write dax: fix data corruption when fault races with write ext4: return to starting transaction in ext4_dax_huge_fault() mm: fix data corruption due to stale mmap reads dax: prevent invalidation of mapped DAX entries Tigran has moved mm, vmalloc: fix vmalloc users tracking properly mm/khugepaged: add missed tracepoint for collapse_huge_page_swapin gcov: support GCC 7.1 mm, vmstat: Remove spurious WARN() during zoneinfo print time: delete current_fs_time() hwpoison, memcg: forcibly uncharge LRU pages

dax: fix PMD data corruption when fault races with write

2017-05-12T22:57:16Z

This is based on a patch from Jan Kara that fixed the equivalent race in the DAX PTE fault path. Currently DAX PMD read fault can race with write(2) in the following way: CPU1 - write(2) CPU2 - read fault dax_iomap_pmd_fault() ->iomap_begin() - sees hole dax_iomap_rw() iomap_apply() ->iomap_begin - allocates blocks dax_iomap_actor() invalidate_inode_pages2_range() - there's nothing to invalidate grab_mapping_entry() - we add huge zero page to the radix tree and map it to page tables The result is that hole page is mapped into page tables (and thus zeros are seen in mmap) while file has data written in that place. Fix the problem by locking exception entry before mapping blocks for the fault. That way we are sure invalidate_inode_pages2_range() call for racing write will either block on entry lock waiting for the fault to finish (and unmap stale page tables after that) or read fault will see already allocated blocks by write(2). Fixes: 9f141d6ef6258 ("dax: Call ->iomap_begin without entry lock during dax fault") Link: http://lkml.kernel.org/r/20170510172700.18991-1-ross.zwisler@linux.intel.com Signed-off-by: Ross Zwisler Reviewed-by: Jan Kara Cc: Dan Williams Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

dax: fix data corruption when fault races with write

2017-05-12T22:57:16Z

Currently DAX read fault can race with write(2) in the following way: CPU1 - write(2) CPU2 - read fault dax_iomap_pte_fault() ->iomap_begin() - sees hole dax_iomap_rw() iomap_apply() ->iomap_begin - allocates blocks dax_iomap_actor() invalidate_inode_pages2_range() - there's nothing to invalidate grab_mapping_entry() - we add zero page in the radix tree and map it to page tables The result is that hole page is mapped into page tables (and thus zeros are seen in mmap) while file has data written in that place. Fix the problem by locking exception entry before mapping blocks for the fault. That way we are sure invalidate_inode_pages2_range() call for racing write will either block on entry lock waiting for the fault to finish (and unmap stale page tables after that) or read fault will see already allocated blocks by write(2). Fixes: 9f141d6ef6258a3a37a045842d9ba7e68f368956 Link: http://lkml.kernel.org/r/20170510085419.27601-5-jack@suse.cz Signed-off-by: Jan Kara Reviewed-by: Ross Zwisler Cc: Dan Williams Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

mm: fix data corruption due to stale mmap reads

2017-05-12T22:57:15Z

Currently, we didn't invalidate page tables during invalidate_inode_pages2() for DAX. That could result in e.g. 2MiB zero page being mapped into page tables while there were already underlying blocks allocated and thus data seen through mmap were different from data seen by read(2). The following sequence reproduces the problem: - open an mmap over a 2MiB hole - read from a 2MiB hole, faulting in a 2MiB zero page - write to the hole with write(3p). The write succeeds but we incorrectly leave the 2MiB zero page mapping intact. - via the mmap, read the data that was just written. Since the zero page mapping is still intact we read back zeroes instead of the new data. Fix the problem by unconditionally calling invalidate_inode_pages2_range() in dax_iomap_actor() for new block allocations and by properly invalidating page tables in invalidate_inode_pages2_range() for DAX mappings. Fixes: c6dcf52c23d2d3fb5235cec42d7dd3f786b87d55 Link: http://lkml.kernel.org/r/20170510085419.27601-3-jack@suse.cz Signed-off-by: Jan Kara Signed-off-by: Ross Zwisler Cc: Dan Williams Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

dax: prevent invalidation of mapped DAX entries

2017-05-12T22:57:15Z

Patch series "mm,dax: Fix data corruption due to mmap inconsistency", v4. This series fixes data corruption that can happen for DAX mounts when page faults race with write(2) and as a result page tables get out of sync with block mappings in the filesystem and thus data seen through mmap is different from data seen through read(2). The series passes testing with t_mmap_stale test program from Ross and also other mmap related tests on DAX filesystem. This patch (of 4): dax_invalidate_mapping_entry() currently removes DAX exceptional entries only if they are clean and unlocked. This is done via: invalidate_mapping_pages() invalidate_exceptional_entry() dax_invalidate_mapping_entry() However, for page cache pages removed in invalidate_mapping_pages() there is an additional criteria which is that the page must not be mapped. This is noted in the comments above invalidate_mapping_pages() and is checked in invalidate_inode_page(). For DAX entries this means that we can can end up in a situation where a DAX exceptional entry, either a huge zero page or a regular DAX entry, could end up mapped but without an associated radix tree entry. This is inconsistent with the rest of the DAX code and with what happens in the page cache case. We aren't able to unmap the DAX exceptional entry because according to its comments invalidate_mapping_pages() isn't allowed to block, and unmap_mapping_range() takes a write lock on the mapping->i_mmap_rwsem. Since we essentially never have unmapped DAX entries to evict from the radix tree, just remove dax_invalidate_mapping_entry(). Fixes: c6dcf52c23d2 ("mm: Invalidate DAX radix tree entries only if appropriate") Link: http://lkml.kernel.org/r/20170510085419.27601-2-jack@suse.cz Signed-off-by: Ross Zwisler Signed-off-by: Jan Kara Reported-by: Jan Kara Cc: Dan Williams Cc: [4.10+] Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

Merge branch 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm

2017-05-12T22:43:10Z

Pull libnvdimm fixes from Dan Williams: "Incremental fixes and a small feature addition on top of the main libnvdimm 4.12 pull request: - Geert noticed that tinyconfig was bloated by BLOCK selecting DAX. The size regression is fixed by moving all dax helpers into the dax-core and only specifying "select DAX" for FS_DAX and dax-capable drivers. He also asked for clarification of the NR_DEV_DAX config option which, on closer look, does not need to be a config option at all. Mike also throws in a DEV_DAX_PMEM fixup for good measure. - Ben's attention to detail on -stable patch submissions caught a case where the recent fixes to arch_copy_from_iter_pmem() missed a condition where we strand dirty data in the cache. This is tagged for -stable and will also be included in the rework of the pmem api to a proposed {memcpy,copy_user}_flushcache() interface for 4.13. - Vishal adds a feature that missed the initial pull due to pending review feedback. It allows the kernel to clear media errors when initializing a BTT (atomic sector update driver) instance on a pmem namespace. - Ross noticed that the dax_device + dax_operations conversion broke __dax_zero_page_range(). The nvdimm unit tests fail to check this path, but xfstests immediately trips over it. No excuse for missing this before submitting the 4.12 pull request. These all pass the nvdimm unit tests and an xfstests spot check. The set has received a build success notification from the kbuild robot" * 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm: filesystem-dax: fix broken __dax_zero_page_range() conversion libnvdimm, btt: ensure that initializing metadata clears poison libnvdimm: add an atomic vs process context flag to rw_bytes x86, pmem: Fix cache flushing for iovec write < 8 bytes device-dax: kill NR_DEV_DAX block, dax: move "select DAX" from BLOCK to FS_DAX device-dax: Tell kbuild DEV_DAX_PMEM depends on DEV_DAX

filesystem-dax: fix broken __dax_zero_page_range() conversion

2017-05-11T04:46:55Z

The conversion of __dax_zero_page_range() to 'struct dax_operations' caused it to frequently fail. The mistake was treating the @size parameter as a dax mapping length rather than just a length of the clear_pmem() operation. The dax mapping length is assumed to be hard coded as PAGE_SIZE. Without this fix any page unaligned zeroing request will trigger a -EINVAL return from bdev_dax_pgoff(). Cc: Jan Kara Cc: Christoph Hellwig Reported-by: Ross Zwisler Tested-by: Ross Zwisler Fixes: cccbce671582 ("filesystem-dax: convert to dax_direct_access()") Signed-off-by: Dan Williams