Difference between revisions of "ChangeLog-2.2"

• fix fakeinit
• workaround Kconfig oddities
• fix OCFS2 deadlock

• disable AFS if CONFIG_VSERVER_SECURITY
• a fake init should not be signalled by kill -1
• map pids for kill
• don't let guests see RTNLGRP_IPV[46]_IFADDR messages

• add the task and its xid, nid, and tag to all warning messages
• use splice instead of modifying sendfile to work on regular files
• fix lots of COW issues, e.g. returning the new file when it has been replaced during the COW-operation

• fix oops in COW when the original file can't be looked up (e.g. due to tagging)
• check for use-counts and tasks before VXS_SHUTDOWN

ADDENDUM: The patch was re-released on 11 Oct 2007, which also included:

• fix reference counting of vxi's in POSIX timer code

• reset the tagging of loop devices when they're cleared
• fix oops in COW when the original file can't be looked up
• check for use-counts and tasks before VXS_SHUTDOWN
• fix reference counting of vxi's in POSIX timer code
• fix a bug in splice

• reset the tagging of loop devices when they're cleared

• fix a reference counting issue for nsproxies (2.6.22)
• fix an oops in JFS
• use ULL instead of UL where appropriate

• check for errors on COW-break, and remove the temporary file if an error occured
• remove tty poisoning and BUG_ONs
• remove commented code

• CodingStyle cleanups
• add vc_f[gs]et_iattr interfaces
• add disk I/O accounting
• fix an oops on COW-break
• fix disk limits on ext3/ext4

• fix copyright notices
• remove some commented code

• revert the change to do_each_task_pid, adding the check where necessary
• add some tty debugging

• fix an initpid issue in /proc which would make a guest's init take the host's init place

• handle COW on utime as well
• have __create_[vn]x_info claim it
• make the per-CPU structures configurable
• make CONFIG_KEYS depend on !CONFIG_VSERVER_SECURITY

• don't oops on reject routes

• backport the proc banner from 2.6.20 to 2.6.19
• fix vroot

• fix kill -HUP 1 (missing vx_rmap_pid in kill_proc_info)
• lots of cleanup
• add xid to OOM messages
• remove the xid checks in IPC code (should be safe with the namespace)
• remove the runtime tagging of inodes

• fix lock accounting
• fix dentry leak in proc
• fix process visibility issues in do_each_task_pid and do_each_pid_task
• make find_task_by_pid_type check the xid
• return correct link counts for /proc/virt{ual,net}
• fix do_set_sched to not set the new values on every CPU if force is used, and to flush any pending updates

• fix lock accounting
• fix dentry leak in proc

• fix process visibility issues in do_each_task_pid and do_each_pid_task
• make find_task_by_pid_type check the xid
• return correct link counts for /proc/virt{ual,net}
• fix do_set_sched to not set the new values on every CPU if force is used, and to flush any pending updates

• remove some debugging output from the reaper selection

• let guests access dm and loop devices created on the host
• cleanup FIXMEs
• clarify that CONFIG_VSERVER_LEGACY_VERSION is problematic for alpha utils
• add network flags and caps to proc
• fix problems with UDP socket binding
• use the next thread if the parent thread is dead for the reaper, and only the vx_child_reaper if the child is the new parent

• use the pid namespace reaper stuff instead of init_task

• fix oops on COW link breaking
• don't make HARDCPU or COWBL depend on EXPERIMENTAL
• make LEGACY{,NET} depend on EXPERIMENTAL
• make legacy guests enterable again
• get the nsproxy even if we're just returning the new one
• get the fs_struct even if we're just returning the old one

• rebase on 2.6.20

• make VSERVER select SYSVIPC

• fix a warning on 64-bit arches
• use long atomics for the socket accounting, to avoid wrap-arounds causing giant numbers on 64-bit arches

• fix a proc bug leading to oops in do_task_stat
• add global use counts for namespace proxies, mount, IPC and UTS spaces, plus filesystem structures
• add a backport of Serge E. Hallyn's initial fix to the NFS-causing-Oops-on-task-exit problem
• don't automatically unshare the UTS and IPC spaces if a mount namespace is requested on clone
• don't unhash the context if creating it fails
• VLIMIT_MAPPED is accounted
• add support for msecs and reporting idle time in vc_get_sched
• use SEND_SIG_PRIV for all vx_info_kills, making vkill work as expected again
• rewrite vx_{enter,set}_space

• "fix" ecryptfs
• use nx_check for all network related checks
• special-case migrate to nid 1
• install inode.h during headers_install

• compress ext[234]/jfs/ocfs2_sync_flags logic
• fixup limits before returning them in vc_rlimit_stat
• add vc_set_sched_v5 which has the interval2 and fill_rate2 values in the struct
• add vc_get_sched to retrieve the scheduler configuration
• add vc_sched_info to retrieve the user/sys/hold values, as well as the vavavoom and usecs per token
• make the priority bias and vavavoom per-CPU
• add VXSM_MSEC, allowing set_sched's intervals to be specified in msecs instead of tokens
• move the VCI_VERSION defines to vci_config.h
• use nx_check for TIME_WAIT sockets

• fix older versions of vc_enter_space
• rework the help texts for some of the options
• disable legacy by default

• fix on hold processes appearing as dead
• attempt to fix a proc bug by checking if the process is alive

• fix a proc bug which would lead to empty /proc/virtual/<xid>/* files

• fix xfs quota_ctl check
• add a space mask to the vxi
• change the vserver proc code to use different functions for the per-context entries and the global entries
• fix vx_{enter,set}_space to not potentially leak an fs_struct
• make vx_set_space sanity check the provided mask

• rename namespace.c to space.c
• only set the UTS and IPC spaces in the compat branch of vc_ctx_create
• remove unneeded vs_base.h include from net/ipv6/addrconf.c

• forked from 2.6.19-vs2.1.x-t6
• removed quota hashes

• rework the vserver includes to let vx_check use in_{interrupt,irq,softirq}
• remove the unneeded __{enter,leave}_vx_admin
• fix the RSS limit to include anonymous pages too
• mapped files are now in VLIMIT_MAPPED
• add a vx_badness, increasing the odds of an OOM kill for guests above their soft RSS limit
• start using nx_check where appropriate

• have __create claim the vxi
• restructure/simplify vc_{ctx,net}_create

• improve the __{enter,leave}_vx_admin usage
• let the spectator cross the barrier too
• xid 2 is a static xid
• take the hash spinlock in __unhash_[vn]x_info
• LOCK was supposed to be PRIVATE
• improved the supported rlimit code
• add the xid to OOM messages

• removed all legacy code

• forked from 2.1.1
• quota hashes removed