3 * Version 5.8.0 released xx February 2020
6 DragonFly version 5.8 brings ....
8 The details of all commits between the 5.6 and 5.8 branches are available in the associated commit messages for [5.8.0rc1](http://lists.dragonflybsd.org/pipermail/commits/2020-February/720529.html) and [5.8.0]().
12 * 'dsynth' written and added to base, making it more convenient for users to build their own binary repos for ports.
22 MD5 (dfly-x86_64-5.8.0_REL.img) =
23 MD5 (dfly-x86_64-5.8.0_REL.iso) =
24 MD5 (dfly-x86_64-5.8.0_REL.img.bz2) =
25 MD5 (dfly-x86_64-5.8.0_REL.iso.bz2) =
29 If you have an existing 5.x system and are running a generic kernel, the normal upgrade process, described below, will work.
31 Change your local `/usr/src` to 5.8:
35 git branch DragonFly_RELEASE_5_8 origin/DragonFly_RELEASE_5_8
36 git checkout DragonFly_RELEASE_5_8
39 And then rebuild: (in `/usr/src`)
51 `make initrd` is optional for systems with unencrypted disks, and has been possible for several releases; it's mentioned here for completeness. Look at the [initrd(7)](https://leaf.dragonflybsd.org/cgi/web-man?command=initrd§ion=ANY) man page for details on what it does.
53 Don't forget to upgrade your existing packages. 5.8 packages have already been built and are immediately available.
58 ## All other changes since prior release
62 * Fix a minor kernel stack disclosure bug and a few minor memory leaks.
64 * Fix a buffer overflow in libfetch (CVS-2020-7450)
68 * Fix two environment variable vulnerabilities in the telnet client.
72 * Scaling adjustments to improve performance on high-core-count machines. VM page lookup shortcuts, reduced contention when opening special devices like /dev/lpmap, reduced contention for open(... O_RDWR) calls, recoded namecache (mp, ncp) mount point lookup cache.
74 * Significantly reduce SMP contention in low-memory situation.
76 * Significantly improve page selection in the pageout scan. SMP work over the last few years marginalized the page selection algorithm, causing it to often select pages which were actually active. These changes significantly reduce the selection of such pages in favor of inactive pages. The result is the elimination of stuttering in applications like Chrome when system memory is tight.
78 * Do not shortcut VM faults of INACTIVE pages, require the long-path to ensure that the pages are moved to the ACTIVE queue first. This prevents premature pageout of heavily cached VM pages such as pages from highly-shared libraries.
80 * Improve pageout daemon pipe-lining by fixing some of the hysteresis code. Also improve the early-termination checks for the pageout daemon to stop paging more quickly when a large amount of memory is suddenly freed up.
82 * Fixed a rare wait*() deadlock.
84 * The vm.fault_quick sysctl now supports a burst count, but the default is still 1. We do not expect significant gains in performance until the APIs can be refactored all the way through to the low-level pmap code.
86 * Offset the stathz timer by 50% of the hz timer and change default frequency to (hz + 1). This is intended to reduce sampling errors that were causing phantom 'interrupt' time accumulation in system statistics.
88 * Refactor the random number generator in the kernel and also as supplied to userland to use per-cpu state. This gets rid of a significant source of SMP contention under heavy loads. Adjust the RDRAND driver to collect entropy on all CPUs, as well as to chain some of the entropy between CPUs. This ensures that no single CPU can accidentally become deficient.
90 * Reduce excessive rdrand harvesting. The rdrand harvesting rate was too high and causing high-core-count Zen CPUs to burn 2-3% of the system cpu at idle. The harvest size has been reduced to compensate. In addition, some of the entropy is chained between CPUs to ensure that entropy collection on one cpu is not only beneficial to that cpu.
92 * Fix a USB crash on quick plug/detach/plug/detach sequences.
94 * Add system call support for realpath(3), allowing the kernel to build the path for us instead of having to cycle tons of *stat() calls.
96 * Add system call support for getrandom(2). We already had a sysctl-based mechanism and a device mechanism.
98 * Fix a bug in the (rare) lockmgr cancel path.
100 * Fix 'indefinite wait buffer' live-locks related to encrypted disks. The encrypted I/O path had a low-memory deadlock that could prevent paging to encrypted-swap from happening, resulting in a locked up system.
102 * Reduce mpipe overhead, improving (primarily) encrypted I/O performance a bit.
104 * Fix a panic in the control messages processing for IPV6.
106 * Enable keyboard polling in shutdown -h to try to keep the keyboard operational at the 'press any key to reboot' message.
108 * Fix a dangling ref on filesystem fifo vnodes. This could prevent umount from succeeding indefinitely.
110 * Merge a m_pulldown() bug fix (discussed with multiple BSDs). See c24e3890e671.
112 * Enable busmaster operation on bridges (the BIOS might not).
114 * Make sure the %fs selector is properly loaded in the AP bootstrap. Otherwise the selector may be invalid on certain Zen 2 CPUs.
116 * Change the TCP keepalive options from milliseconds to seconds, making the units the same as linux, in order to improve ports compatibility.
118 * Fix a rare pmap placemarker timing race that could cause a process to get stuck in 'pvplw'.
120 * Implement SMAP/SMEP support.
124 * Improve nullfs mount/umount performance slightly.
126 * Improve TMPFS by not reserving an entire 16KB buffer cache buffer for files <= 8KB in size. Small files now eat 4KB, 8KB, or 12KB of physical memory before returning to multiples of 16KB.
128 * Reduce low-memory thrashing due to TMPFS. Normally TMPFS cycles pages through the VM page cache only and avoids flushing them to backing store (swap). In low-memory conditions, TMPFS now cycles dirtied pages through the buffer cache and bypasses the inactive queue upon completion to avoid interfering with memory pressure caused by processes.
130 * Enable write-clustering on TMPFS, improving paging performance when under memory pressure. Prior to this fix, physical I/O was limited to 16KB chunks.
132 * HAMMER2 - Fix inode & chain limits in order to avoid kmalloc pool assertions on low-memory machines.
134 * HAMMER2 - Significantly improve the flush pipeline and dirty inode count hysteresis. This fixes a long-standing issue where heavy filesystem activity was not making use of available I/O bandwidth due to a number of tsleep() stalls in the flusher loop. This also improves frontend-operation vs backend-flush concurrency.
136 * HAMMER2 - Fix a bug where a small file less than or equal to 512 bytes that is extended above 512 bytes can sometimes lose data in the first 512 bytes (replaced with 0x00 bytes).
138 * DEVFS - Fixed a missing lock that could result in a panic during heavy duplicate mount/umount activity.
140 * All filesystems now default to microseconds precision instead of seconds precision. The use of nanoseconds is problematic due to utimes()'s archaic API.
142 * NFS - Fix a rare NFS deadlock related to mmap()'s I/O.
148 * Further synchronization with linux
152 * sound/hda - use driver-specific taskqueue instead of per-cpu taskqueue.
154 * Fix memory corruption and panic in the PSM driver.
156 * AHCI - Add a quirk for the MCP73 AHCI controller.
158 * AHCI - Add a TPS balancing mode for storage devices, enabled by default. This feature brings write tps down to match read tps when both are pending. Normally both read and write I/Os get a few private tags, but because write I/O is heavily buffered by the device even a single tag can destroy read performance.
160 * AHCI - Enforce a 10-second poll of the chipset when idle to try to catch any potential lost interrupts. This is an attempt to work around a 'lost' timeout issue that we sometimes see.
162 * u4b/uaudio - Fix NULL pointer indirection panic.
164 * IWM - Full sync from FreeBSD, adding iwm-9000 and iwm-9260 support.
168 * Binutils234 added, and 225 removed. 2.27 is still the default for now.
170 * pctrack (%rip sampler) - Improved symbol table translation.
172 * Sync math.h, sincos(), and other math support from OpenBSD.
174 * Synt <sys/ttydefaults.h> from FreeBSD to improve ports compatibility.
176 * Synchronize improvements made to fsck_msdosfs from FreeBSD.
178 * Check if the context is invalid in setcontext() and swapcontext() to throw errors earlier in situations where too-small a stack size has been specified.
182 * Fix 'ps' and 'fstat' when used on kernel cores.
184 * vmstat now pulls 'ncpus' from the kernel core image, when used with a kernel core. Also make the verbose (-v) option work with (-m), dumping all memory pools whether they are empty or not.
186 * Adopt a more linux-friendly basename() and dirname() to improve ports compatibility.
188 * Fix bug in fparseln() (taken from NetBSD).
190 * libc/getaddrinfo() - Accept numeric servname for AF_INET / AF_INET6. Take from FreeBSD.
192 * Update zoneinfo to tzdata2019c.
194 * Fix an issue when rotating /var/log/wtmpx.
196 * iostat - fix column count bug.
198 * libdevstat - Fix a bug in DS_SELECT_REMOVE.
200 * Remove mrouted and related utilities from the tree. Newer, better versions can be found in dports and ours was getting crufy.
204 * rcorder-visualize.sh imported from FreeBSD
206 * Sync ACPI with Intel's version 20200214
208 * Unbreak the EFI build when WORLD_LDVER is overridden.
210 * The installer now chooses a more sane default for the /build partition.
212 ### Various tools have been upgraded in the base system:
214 * The build system is slowly being adapted to allow bootstrap utilities and tools to be compiled on a wide range of operating systems as part of a project to supply a 'build.sh' script to be able to build DragonFlyBSD on other OSs. This work is not yet complete.
216 * The build system now has a 'build-all' and 'install-all' target to conveniently do a full world + kernel build and install in one or two steps as desired.
224 * A full set of new binary packages has been built for 5.8, available through the pkg tool.