aboutsummaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* Remove gratuitous Linux only include in ztest & zdbMatthew Macy2019-10-192-2/+0
| | | | | | | | We don't need to include stdio_ext.h Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9483
* ZTS: Written props test fails with 4k disksJohn Wren Kennedy2019-10-181-5/+5
| | | | | | | | | | | | With 4k disks, this test will fail in the last section because the expected human readable value of 20.0M is reported as 20.1M. Rather than use the human readable property, switch to the parsable property and verify that the values are reasonably close. Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Tony Hutter <[email protected]> Signed-off-by: John Kennedy <[email protected]> Closes #9477
* Name anonymous enum of KMC_BIT constantsSerapheim Dimitropoulos2019-10-181-2/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Giving a name to this enum makes it discoverable from debugging tools like DRGN and SDB. For example, with the name proposed on this patch we can iterate over these values in DRGN: ``` >>> prog.type('enum kmc_bit').enumerators (('KMC_BIT_NOTOUCH', 0), ('KMC_BIT_NODEBUG', 1), ('KMC_BIT_NOMAGAZINE', 2), ('KMC_BIT_NOHASH', 3), ('KMC_BIT_QCACHE', 4), ('KMC_BIT_KMEM', 5), ('KMC_BIT_VMEM', 6), ('KMC_BIT_SLAB', 7), ... ``` This enables SDB to easily pretty-print the flags of the spl_kmem_caches in the system like this: ``` > spl_kmem_caches -o "name,flags,total_memory" name flags total_memory ------------------------ ----------------------- ------------ abd_t KMC_NOMAGAZINE|KMC_SLAB 4.5MB arc_buf_hdr_t_full KMC_NOMAGAZINE|KMC_SLAB 12.3MB ... <cropped> ... ddt_cache KMC_VMEM 583.7KB ddt_entry_cache KMC_NOMAGAZINE|KMC_SLAB 0.0B ... <cropped> ... zio_buf_1048576 KMC_NODEBUG|KMC_VMEM 0.0B ... <cropped> ... ``` Reviewed-by: Matt Ahrens <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Serapheim Dimitropoulos <[email protected]> Closes #9478
* Update skc_obj_alloc for spl kmem caches that are backed by LinuxSerapheim Dimitropoulos2019-10-182-6/+36
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Currently, for certain sizes and classes of allocations we use SPL caches that are backed by caches in the Linux Slab allocator to reduce fragmentation and increase utilization of memory. The way things are implemented for these caches as of now though is that we don't keep any statistics of the allocations that we make from these caches. This patch enables the tracking of allocated objects in those SPL caches by making the trade-off of grabbing the cache lock at every object allocation and free to update the respective counter. Additionally, this patch makes those caches visible in the /proc/spl/kmem/slab special file. As a side note, enabling the specific counter for those caches enables SDB to create a more user-friendly interface than /proc/spl/kmem/slab that can also cross-reference data from slabinfo. Here is for example the output of one of those caches in SDB that outputs the name of the underlying Linux cache, the memory of SPL objects allocated in that cache, and the percentage of those objects compared to all the objects in it: ``` > spl_kmem_caches | filter obj.skc_name == "zio_buf_512" | pp name ... source total_memory util ----------- ... ----------------- ------------ ---- zio_buf_512 ... kmalloc-512[SLUB] 16.9MB 8 ``` Reviewed-by: Matt Ahrens <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Serapheim Dimitropoulos <[email protected]> Closes #9474
* OpenZFS restructuring - ARC memory pressureMatthew Macy2019-10-185-438/+527
| | | | | | | | | | | Factor Linux specific memory pressure handling out of ARC. Each platform will have different available interfaces for managing memory pressure. Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: Ryan Moeller <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9472
* Detect if sed supports --in-placeRyan Moeller2019-10-165-3/+20
| | | | | | | | | | | | | | Not all versions of sed have the --in-place flag. Detect support for the flag during ./configure and provide a fallback mechanism for those systems where sed's behavior differs. The autoconf variable ${ac_inplace} can be used to choose the correct flags for editing a file in place with sed. Replace violating usages in Makefile.am with ${ac_inplace}. Reviewed-by: Chris Dunlop <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Ryan Moeller <[email protected]> Closes #9463
* Make zfsdev_getminor signature cross platformMatthew Macy2019-10-164-16/+10
| | | | | | | | Only pass the file descriptor to make zfsdev_get_miror() portable. Reviewed-by: George Melikov <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9466
* Move linux specific mmp module_param_call handler to platform codeMatthew Macy2019-10-163-28/+47
| | | | | | Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9465
* Add default case to lua kernel codeMatthew Macy2019-10-161-0/+6
| | | | | | | | | | Some platforms, e.g. FreeBSD, support user space setjmp semantics in kernel. Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9450
* Fix signature for private functions without header declarationsMatthew Macy2019-10-162-2/+2
| | | | | | | | Clang will complain if a function has no prior declaration Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9467
* Remove dead code and cleanup scoping in dmu_send.cMatthew Macy2019-10-131-16/+5
| | | | | | | | | | | | | This addresses a number of problems with dmu_send.c: * bp_span is unused which makes clang complain * dump_write conflicts with FreeBSD's existing core dump code * range_alloc is private to the file and not declared in any headers causing clang to complain Reviewed-by: Matt Ahrens <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9432
* Don't call sizeof on voidPaul Dagnelie2019-10-132-7/+7
| | | | | | | | | We get the sizeof the appropriate type, and don't cast away const. Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Matt Macy <[email protected]> Signed-off-by: Paul Dagnelie <[email protected]> Closes #9455
* Move zfs_onexit_fd_hold to platform codeMatthew Macy2019-10-133-33/+70
| | | | | | | | FreeBSD has a very different implementation. Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9442
* Move zio_delay_interrupt to platform codeMatthew Macy2019-10-134-71/+105
| | | | | | | | FreeBSD has its own implementation as do other platforms. Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Ryan Moeller <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9439
* Modify sharenfs=on default behaviorBrian Behlendorf2019-10-132-3/+4
| | | | | | | | | | | | While it may sometimes be convenient to export an NFS filesystem with no_root_squash it should not be the default behavior. Align the default behavior with the Linux NFS server defaults. To restore the previous behavior use 'zfs set sharenfs="no_root_squash,..."'. Reviewed-by: loli10K <[email protected]> Reviewed-by: Richard Laager <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #9397 Closes #9425
* Typo fix in comment: dso_dryrunchrisrd2019-10-111-1/+1
| | | | | | Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Chris Dunlop <[email protected]> Closes #9452
* ZTS: Fix zpool_status_-sBrian Behlendorf2019-10-112-2/+3
| | | | | | | | | | | | | | | | | | | | | | | After commit 5e74ac51 which split and reordered the run files the `zpool_status_-s` test began failing. The new ordering placed the test after a previous test which used `zpool replace` to replace a disk but did not clear its label. This resulted in the next test, `zpool_status_-s`, failing because of the potentially active pool being detected on the replaced vdev. /dev/loop0 is part of potentially active pool 'testpool' Use the default_mirror_setup_noexit() and default_cleanup_noexit() functions to create the pool in `zpool_status_-s`. They use the -f flag by default. In the `scrub_after_resilver` test wipe the label during cleanup to prevent future failures if the tests are again reordered. Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: John Kennedy <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #9451
* Function name and comment updatesPaul Dagnelie2019-10-114-16/+40
| | | | | | | | | | | Rename certain functions for more consistency when they share common features. Make comments clearer about what arguments should be passed to the insert and add functions. Reviewed by: Sara Hartse <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Matt Ahrens <[email protected]> Signed-off-by: Paul Dagnelie <[email protected]> Closes #9441
* Remove linux/mod_compat.h from common codeMatthew Macy2019-10-113-3/+0
| | | | | | | | | It is no longer necessary; mod_compat.h is included from zfs_context.h. Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Igor Kozhukhov <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9449
* Expose dmu_buf_hold_array_by_dnode to platform codeMatthew Macy2019-10-112-2/+4
| | | | | | | | FreeBSD uses this in its pager ops routines Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Ryan Moeller <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9431
* Implement ZPOOL_IMPORT_UDEV_TIMEOUT_MSRichard Yao2019-10-112-1/+16
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Since 0.7.0, zpool import would unconditionally block on udev for 30 seconds. This introduced a regression in initramfs environments that lack udev (particularly mdev based environments), yet use a zfs userland tools intended for the system that had been built against udev. Gentoo's genkernel is the main example, although custom user initramfs environments would be similarly impacted unless special builds of the ZFS userland utilities were done for them. Such environments already have their own mechanisms for blocking until device nodes are ready (such as genkernel's scandelay parameter), so it is unnecessary for zpool import to block on a non-existent udev until a timeout is reached inside of them. Rather than trying to intelligently determine whether udev is available on the system to avoid unnecessarily blocking in such environments, it seems best to just allow the environment to override the timeout. I propose that we add an environment variable called ZPOOL_IMPORT_UDEV_TIMEOUT_MS. Setting it to 0 would restore the 0.6.x behavior that was more desirable in mdev based initramfs environments. This allows the system user land utilities to be reused when building mdev-based initramfs archives. Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Georgy Yakovlev <[email protected]> Signed-off-by: Richard Yao <[email protected]> Closes #9436
* Clarify loop variable name in zfs copies testRyan Moeller2019-10-111-2/+2
| | | | | | Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: John Kennedy <[email protected]> Signed-off-by: Ryan Moeller <[email protected]> Closes #9445
* Fix some style nits in testsRyan Moeller2019-10-117-17/+15
| | | | | | | | Mostly whitespace changes, no functional changes intended. Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: John Kennedy <[email protected]> Signed-off-by: Ryan Moeller <[email protected]> Closes #9447
* Fix pool creation with feature@allocation_classes disabledloli10K2019-10-104-1/+44
| | | | | | | | | | | When "feature@allocation_classes" is not enabled on the pool no vdev with "special" or "dedup" allocation type should be allowed to exist in the vdev tree. Reviewed-by: Pavel Zakharov <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: loli10K <[email protected]> Closes #9427 Closes #9429
* Move get_temporary_prop to platform codeMatthew Macy2019-10-103-81/+84
| | | | | | | | | | Temporary property handling at the VFS layer requires platform specific code. Reviewed-by: Sean Eric Fagan <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Jorgen Lundman <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9401
* Add kmem cache accessorsMatthew Macy2019-10-103-4/+21
| | | | | | | | | | | | Make the metaslab platform agnostic again by adding accessor functions which can be implemented by each platform. Reviewed-by: Paul Dagnelie <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: Ryan Moeller <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9404
* Make `zil_async_to_sync` visible to platform codeMatthew Macy2019-10-102-3/+2
| | | | | | | | | FreeBSD's zvol platform code requires access to the zil_async_to_sync() function. Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9440
* Update `zfs program` command usageBrian Behlendorf2019-10-101-2/+4
| | | | | | | | | | | | | | | | | Update the zfs(8) man page to clearly describe that arguments for channel programs are to be listed after the -- sentinel which terminates argument processing. This behavior is supported by getopt on Linux, FreeBSD, and Illumos according to each platforms respective man pages. zfs program [-jn] [-t instruction-limit] [-m memory-limit] pool script [--] arg1 ... Reviewed-by: Clint Armstrong <[email protected]> Reviewed-by: George Melikov <[email protected]> Reviewed-by: loli10K <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #9056 Closes #9428
* Fix strdup conflict on other platformsMatthew Macy2019-10-1021-75/+76
| | | | | | | | | | | | | | | | In the FreeBSD kernel the strdup signature is: ``` char *strdup(const char *__restrict, struct malloc_type *); ``` It's unfortunate that the developers have chosen to change the signature of libc functions - but it's what I have to deal with. Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9433
* Make clang happy with vdev_raidz_ codeMatthew Macy2019-10-102-0/+13
| | | | | | | | | | | | | | | | | | The macros are used to generate code for conditions without a corresponding branch. This is not a problem in practice, but clang has no way of knowing that. Add a default branch with a VERIFY(0) to indicate that it "can't happen" ``` In file included from \ /usr/home/mmacy/devel/ZoF/module/zfs/vdev_raidz_math_sse2.c:607: /usr/home/mmacy/devel/ZoF/module/zfs/vdev_raidz_math_impl.h:281:3: \ error: no case matching constant switch condition '3' [-Werror] ``` Reviewed-by: George Melikov <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9434
* Sort list of generated files in configure.acRyan Moeller2019-10-091-91/+91
| | | | | | | No functional change, just a quality of life improvement. Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Ryan Moeller <[email protected]> Closes #9426
* Move platform independent tests to a shared runfileRyan Moeller2019-10-096-949/+1036
| | | | | | | | | | | | | | | | | | | Tests that aren't limited to running on Linux can be moved to a common runfile to be shared with other platforms. The test runner and wrapper script are enhanced to allow specifying multiple runfiles as a comma-separated list. The default runfiles are now "common.run,PLATFORM.run" where PLATFORM is determined at run time. Sections in runfiles that share a path with another runfile can append a colon separator and an identifier to the path in the section name, ie `[tests/functional/atime:Linux]`, to avoid overriding the tests specified by other runfiles. Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: John Kennedy <[email protected]> Signed-off-by: Ryan Moeller <[email protected]> Closes #9391
* Reduce loaded range tree memory usagePaul Dagnelie2019-10-0950-638/+3722
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This patch implements a new tree structure for ZFS, and uses it to store range trees more efficiently. The new structure is approximately a B-tree, though there are some small differences from the usual characterizations. The tree has core nodes and leaf nodes; each contain data elements, which the elements in the core nodes acting as separators between its children. The difference between core and leaf nodes is that the core nodes have an array of children, while leaf nodes don't. Every node in the tree may be only partially full; in most cases, they are all at least 50% full (in terms of element count) except for the root node, which can be less full. Underfull nodes will steal from their neighbors or merge to remain full enough, while overfull nodes will split in two. The data elements are contained in tree-controlled buffers; they are copied into these on insertion, and overwritten on deletion. This means that the elements are not independently allocated, which reduces overhead, but also means they can't be shared between trees (and also that pointers to them are only valid until a side-effectful tree operation occurs). The overhead varies based on how dense the tree is, but is usually on the order of about 50% of the element size; the per-node overheads are very small, and so don't make a significant difference. The trees can accept arbitrary records; they accept a size and a comparator to allow them to be used for a variety of purposes. The new trees replace the AVL trees used in the range trees today. Currently, the range_seg_t structure contains three 8 byte integers of payload and two 24 byte avl_tree_node_ts to handle its storage in both an offset-sorted tree and a size-sorted tree (total size: 64 bytes). In the new model, the range seg structures are usually two 4 byte integers, but a separate one needs to exist for the size-sorted and offset-sorted tree. Between the raw size, the 50% overhead, and the double storage, the new btrees are expected to use 8*1.5*2 = 24 bytes per record, or 33.3% as much memory as the AVL trees (this is for the purposes of storing metaslab range trees; for other purposes, like scrubs, they use ~50% as much memory). We reduced the size of the payload in the range segments by teaching range trees about starting offsets and shifts; since metaslabs have a fixed starting offset, and they all operate in terms of disk sectors, we can store the ranges using 4-byte integers as long as the size of the metaslab divided by the sector size is less than 2^32. For 512-byte sectors, this is a 2^41 (or 2TB) metaslab, which with the default settings corresponds to a 256PB disk. 4k sector disks can handle metaslabs up to 2^46 bytes, or 2^63 byte disks. Since we do not anticipate disks of this size in the near future, there should be almost no cases where metaslabs need 64-byte integers to store their ranges. We do still have the capability to store 64-byte integer ranges to account for cases where we are storing per-vdev (or per-dnode) trees, which could reasonably go above the limits discussed. We also do not store fill information in the compact version of the node, since it is only used for sorted scrub. We also optimized the metaslab loading process in various other ways to offset some inefficiencies in the btree model. While individual operations (find, insert, remove_from) are faster for the btree than they are for the avl tree, remove usually requires a find operation, while in the AVL tree model the element itself suffices. Some clever changes actually caused an overall speedup in metaslab loading; we use approximately 40% less cpu to load metaslabs in our tests on Illumos. Another memory and performance optimization was achieved by changing what is stored in the size-sorted trees. When a disk is heavily fragmented, the df algorithm used by default in ZFS will almost always find a number of small regions in its initial cursor-based search; it will usually only fall back to the size-sorted tree to find larger regions. If we increase the size of the cursor-based search slightly, and don't store segments that are smaller than a tunable size floor in the size-sorted tree, we can further cut memory usage down to below 20% of what the AVL trees store. This also results in further reductions in CPU time spent loading metaslabs. The 16KiB size floor was chosen because it results in substantial memory usage reduction while not usually resulting in situations where we can't find an appropriate chunk with the cursor and are forced to use an oversized chunk from the size-sorted tree. In addition, even if we do have to use an oversized chunk from the size-sorted tree, the chunk would be too small to use for ZIL allocations, so it isn't as big of a loss as it might otherwise be. And often, more small allocations will follow the initial one, and the cursor search will now find the remainder of the chunk we didn't use all of and use it for subsequent allocations. Practical testing has shown little or no change in fragmentation as a result of this change. If the size-sorted tree becomes empty while the offset sorted one still has entries, it will load all the entries from the offset sorted tree and disregard the size floor until it is unloaded again. This operation occurs rarely with the default setting, only on incredibly thoroughly fragmented pools. There are some other small changes to zdb to teach it to handle btrees, but nothing major. Reviewed-by: George Wilson <[email protected]> Reviewed-by: Matt Ahrens <[email protected]> Reviewed by: Sebastien Roy [email protected] Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Paul Dagnelie <[email protected]> Closes #9181
* ZTS: Fix mmp_hostid testIgor K2019-10-081-1/+5
| | | | | | | | Correctly use the `mntpnt_fs` variable, and include additional logic to ensure the /etc/hostid is correct set up and cleaned up. Signed-off-by: Brian Behlendorf <[email protected]> Signed-off-by: Igor Kozhukhov <[email protected]> Closes #9349
* module/Makefile.in: don't run xargs if emptyGeorge Melikov2019-10-081-1/+1
| | | | | | | | | | If stdin if empty - don't run xargs command, otherwise we can get `cp: missing file operand` error. Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: George Melikov <[email protected]> Closes #9418
* ZTS: Fix trim/trim_config and trim/autotrim_configBrian Behlendorf2019-10-042-2/+2
| | | | | | | | | | | | There have been occasional CI failures which occur when the trimmed vdev size exactly matches the target size. Resolve this by slightly relaxing the conditional and checking for -ge rather than -gt. In all of the cases observer, the values match exactly. For example: Failure /mnt/trim-vdev1 is 768 MB which is not -gt than 768 MB Reviewed-by: Ryan Moeller <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #9399
* Fix automount for root filesystemsBrian Behlendorf2019-10-041-1/+2
| | | | | | | | | | | Commit 093bb64 resolved an automount failures for chroot'd processes but inadvertently broke automounting for root filesystems where the vfs_mntpoint is NULL. Resolve the issue by checking for NULL in order to generate the correct path. Reviewed-by: Tom Caputi <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #9381 Closes #9384
* Rename rangelock_ functions to zfs_rangelock_Matthew Macy2019-10-036-82/+84
| | | | | | | | | A rangelock KPI already exists on FreeBSD. Add a zfs_ prefix as per our convention to prevent any conflict with existing symbols. Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9402
* dbuf_hold_impl() cleanup to improve cached read performanceTony Nguyen2019-10-031-154/+69
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Currently every dbuf_hold_impl() incurs kmem_alloc() and kmem_free() which can be costly for cached read performance. This change reverts the dbuf_hold_impl() fix stack commit, i.e. fc5bb51f08a6c91ff9ad3559d0266eeeab0b1f61 to eliminate the extra kmem_alloc() and kmem_free() operations and improve cached read performance. With the change, each dbuf_hold_impl() frame uses 40 bytes more, total of 800 for 20 recursive levels. Linux kernel stack sizes are 8K and 16K for 32bit and 64bit, respectively, so stack overrun risk is limited. Sample stack output comparisons with 50 PB file and recordsize=512 Current code 11) 2240 64 arc_alloc_buf+0x4a/0xd0 [zfs] 12) 2176 264 dbuf_read_impl.constprop.16+0x2e3/0x7f0 [zfs] 13) 1912 120 dbuf_read+0xe5/0x520 [zfs] 14) 1792 56 dbuf_hold_impl_arg+0x572/0x630 [zfs] 15) 1736 64 dbuf_hold_impl_arg+0x508/0x630 [zfs] 16) 1672 64 dbuf_hold_impl_arg+0x508/0x630 [zfs] 17) 1608 40 dbuf_hold_impl+0x23/0x40 [zfs] 18) 1568 40 dbuf_hold_level+0x32/0x60 [zfs] 19) 1528 16 dbuf_hold+0x16/0x20 [zfs] dbuf_hold_impl() cleanup 11) 2320 64 arc_alloc_buf+0x4a/0xd0 [zfs] 12) 2256 264 dbuf_read_impl.constprop.17+0x2e3/0x7f0 [zfs] 13) 1992 120 dbuf_read+0xe5/0x520 [zfs] 14) 1872 96 dbuf_hold_impl+0x50f/0x5e0 [zfs] 15) 1776 104 dbuf_hold_impl+0x4df/0x5e0 [zfs] 16) 1672 104 dbuf_hold_impl+0x4df/0x5e0 [zfs] 17) 1568 40 dbuf_hold_level+0x32/0x60 [zfs] 18) 1528 16 dbuf_hold+0x16/0x20 [zfs] Performance observations on 8K recordsize filesystem: - 8/128/1024K at 1-128 sequential cached read, ~3% improvement Testing done on Ubuntu 18.04 with 4.15 kernel, 8vCPUs and SSD storage on VMware ESX. Reviewed-by: Matt Ahrens <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Tony Nguyen <[email protected]> Closes #9351
* OpenZFS restructuring - libzfsMatthew Macy2019-10-0311-797/+1026
| | | | | | | | | Factor Linux specific functionality out of libzfs. Reviewed-by: Allan Jude <[email protected]> Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matthew Macy <[email protected]> Closes #9377
* OpenZFS restructuring - libzutilMatthew Macy2019-10-038-1337/+1508
| | | | | | | | Factor Linux specific functionality out of libzutil. Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Signed-off-by: Ryan Moeller <[email protected]> Closes #9356
* ZTS: Fix upgrade_readonly_poolBrian Behlendorf2019-10-031-4/+4
| | | | | | | | | | | | | | | | | | Update cleanup_upgrade to use destroy_dataset and destroy_pool when performing cleanup. These wrappers retry if the pool is busy preventing occasional failures like those observed when running tests upgrade_readonly_pool. For example: SUCCESS: test enabled == enabled User accounting upgrade is not executed on readonly pool NOTE: Performing local cleanup via log_onexit (cleanup_upgrade) cannot destroy 'testpool': pool is busy ERROR: zpool destroy testpool exited 1 Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: John Kennedy <[email protected]> Reviewed-by: Igor Kozhukhov <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #9400
* Workaround to avoid a race when /var/lib is a persistent datasetDidier Roche2019-10-021-0/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | If /var/lib is a dataset not under <pool>/ROOT/<root_dataset>, as proposed in the ubuntu root on zfs upstream guide (https://github.com/zfsonlinux/zfs/wiki/Ubuntu-18.04-Root-on-ZFS), we end up with a race where some services, like systemd-random-seed are writing under /var/lib, while zfs-mount is called. zfs mount will then potentially fail because of /var/lib isn't empty and so, can't be mounted. Order those 2 units for now (more may be needed) as we can't declare virtually a provide mount point to match "RequiresMountsFor=/var/lib/systemd/random-seed" from systemd-random-seed.service. The optional generator for zfs 0.8 fixes it, but it's not enabled by default nor necessarily required. Example: - rpool/ROOT/ubuntu (mountpoint = /) - rpool/var/ (mountpoint = /var) - rpool/var/lib (mountpoint = /var/lib) Both zfs-mount.service and systemd-random-seed.service are starting After=systemd-remount-fs.service. zfs-mount.service should be done before local-fs.target while systemd-random-seed.service should finish before sysinit.target (which is a later target). Ideally, we would have a way for zfs mount -a unit to declare all paths or move systemd-random-seed after local-fs.target. Reviewed-by: Antonio Russo <[email protected]> Reviewed-by: Richard Laager <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Didier Roche <[email protected]> Closes #9360
* OpenZFS restructuring - libsplMatthew Macy2019-10-0265-443/+127
| | | | | | | | | Factor Linux specific pieces out of libspl. Reviewed-by: Ryan Moeller <[email protected]> Reviewed-by: Sean Eric Fagan <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9336
* Add inode accessors to common codeMatthew Macy2019-10-025-15/+23
| | | | | | Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Jorgen Lundman <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9389
* OpenZFS restructuring - arc_statsMatthew Macy2019-10-012-281/+282
| | | | | | | | Make arc_stats visible to platform code. Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Igor Kozhukhov <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9386
* Enable clang to use intrinsics for lz4Matthew Macy2019-10-011-3/+3
| | | | | | | | Reviewed-by: Igor Kozhukhov <[email protected]> Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: George Melikov <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Matt Macy <[email protected]> Closes #9385
* Fix for zfs-dracut regressiondacianstremtan2019-10-011-6/+7
| | | | | | | | | | | | Line 31 and 32 overwrote the ${root} variable which broke mount-zfs.sh We have create a new variable for the dataset instead of overwriting the ${root} variable in zfs-load-key.sh${root} variable in zfs-load-key.sh Reviewed-by: Kash Pande <[email protected]> Reviewed-by: Garrett Fields <[email protected]> Reviewed-by: Brian Behlendorf <[email protected]> Signed-off-by: Dacian Reece-Stremtan <[email protected]> Closes #8913 Closes #9379
* Perform KABI checks in parallelBrian Behlendorf2019-10-01118-2127/+3245
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Reduce the time required for ./configure to perform the needed KABI checks by allowing kbuild to compile multiple test cases in parallel. This was accomplished by splitting each test's source code from the logic handling whether that code could be compiled or not. By introducing this split it's possible to minimize the number of times kbuild needs to be invoked. As importantly, it means all of the tests can be built in parallel. This does require a little extra care since we expect some tests to fail, so the --keep-going (-k) option must be provided otherwise some tests may not get compiled. Furthermore, since a failure during the kbuild modpost phase will result in an early exit; the final linking phase is limited to tests which passed the initial compilation and produced an object file. Once everything has been built the configure script proceeds as previously. The only significant difference is that it now merely needs to test for the existence of a .ko file to determine the result of a given test. This vastly speeds up the entire process. New test cases should use ZFS_LINUX_TEST_SRC to declare their test source code and ZFS_LINUX_TEST_RESULT to check the result. All of the existing kernel-*.m4 files have been updated accordingly, see config/kernel-current-time.m4 for a basic example. The legacy ZFS_LINUX_TRY_COMPILE macro has been kept to handle special cases but it's use is not encouraged. master (secs) patched (secs) ------------- ---------------- autogen.sh 61 68 configure 137 24 (~17% of current run time) make -j $(nproc) 44 44 make rpms 287 150 Reviewed-by: Tony Hutter <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #8547 Closes #9132 Closes #9341
* Timeout waiting for ZVOL device to be createdPrakash Surya2019-10-012-6/+14
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | We've seen cases where after creating a ZVOL, the ZVOL device node in "/dev" isn't generated after 20 seconds of waiting, which is the point at which our applications gives up on waiting and reports an error. The workload when this occurs is to "refresh" 400+ ZVOLs roughly at the same time, based on a policy set by the user. This refresh operation will destroy the ZVOL, and re-create it based on a snapshot. When this occurs, we see many hundreds of entries on the "z_zvol" taskq (based on inspection of the /proc/spl/taskq-all file). Many of the entries on the taskq end up in the "zvol_remove_minors_impl" function, and I've measured the latency of that function: Function = zvol_remove_minors_impl msecs : count distribution 0 -> 1 : 0 | | 2 -> 3 : 0 | | 4 -> 7 : 1 | | 8 -> 15 : 0 | | 16 -> 31 : 0 | | 32 -> 63 : 0 | | 64 -> 127 : 1 | | 128 -> 255 : 45 |****************************************| 256 -> 511 : 5 |**** | That data is from a 10 second sample, using the BCC "funclatency" tool. As we can see, in this 10 second sample, most calls took 128ms at a minimum. Thus, some basic math tells us that in any 20 second interval, we could only process at most about 150 removals, which is much less than the 400+ that'll occur based on the workload. As a result of this, and since all ZVOL minor operations will go through the single threaded "z_zvol" taskq, the latency for creating a single ZVOL device can be unreasonably large due to other ZVOL activity on the system. In our case, it's large enough to cause the application to generate an error and fail the operation. When profiling the "zvol_remove_minors_impl" function, I saw that most of the time in the function was spent off-cpu, blocked in the function "taskq_wait_outstanding". How this works, is "zvol_remove_minors_impl" will dispatch calls to "zvol_free" using the "system_taskq", and then the "taskq_wait_outstanding" function is used to wait for all of those dispatched calls to occur before "zvol_remove_minors_impl" will return. As far as I can tell, "zvol_remove_minors_impl" doesn't necessarily have to wait for all calls to "zvol_free" to occur before it returns. Thus, this change removes the call to "taskq_wait_oustanding", so that calls to "zvol_free" don't affect the latency of "zvol_remove_minors_impl". Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: John Gallagher <[email protected]> Signed-off-by: Prakash Surya <[email protected]> Closes #9380
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-21 07:22:01 +0000