When use `mount_vboxvfs ...` it shows: link_elf_obj: symbol bcmp undefined linker_load_file: /boot/modules/vboxvfs.ko - unsupported file type FreeBSD 13-Current Additions: 5.2.34_2 (installed via pkg, not compiled)
Do versions in output of `sysctl kern.osreldate` and `pkg query %Av virtualbox-ose-additions` match? Try rebuilding from ports if they don't.
(In reply to Yuri Pankov from comment #1) Identical: 1300084
The kernel does not offer bcmp: $ nm /boot/kernel/kernel|grep -i bcmp ffffffff80cb4610 T timingsafe_bcmp The port links vboxvfs.ko with a library that references bcmp $ cat /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs.dep define vboxvfs_CMDS_PREV_ @kmk_builtin_rm -f -- /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs.dep /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs.ko /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs.map ld -d --warn-common -m elf_x86_64_fbsd -r -o /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs.ko /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs_vfsops.o /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs_vnops.o /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs_prov.o /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/lib/additions/VBoxGuestR0Lib.a /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/lib/additions/RuntimeGuestR0.a awk -f /usr/src/sys/conf/kmod_syms.awk /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs.ko /dev/null | xargs -J% objcopy % /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/vboxvfs/vboxvfs.ko endef $ nm /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/lib/additions/RuntimeGuestR0.a | grep -B13 bcmp strformatrt.o: U VBoxGuest_RTStrFormat U VBoxGuest_RTStrFormatNumber 0000000000000000 r _ZL13g_szHexDigits 0000000000002060 t _ZL15rtstrFormatIPv6PFmPvPKcmES_PK10RTUINT128U 0000000000000000 r _ZZ13rtstrFormatRtE12s_wszBigNull 0000000000000280 r _ZZ13rtstrFormatRtE8s_aTypes 0000000000000870 r _ZZ13rtstrFormatRtE8s_aTypes_0 0000000000000850 r _ZZ13rtstrFormatRtE8s_szNull 0000000000000857 r _ZZ13rtstrFormatRtE8s_szTrue 000000000000085d r _ZZ13rtstrFormatRtE9s_szFalse U __start_set_pcpu U __stop_set_pcpu U bcmp Someone with an understanding of FreeBSD kernel modules will have to explain where things went sideways. John groenveld@acm.org
There is no bcmp symbol on amd64 for almost 2 years now. In sys/systm.h there is: int bcmp(const void *b1, const void *b2, size_t len); [..] #define bcmp(b1, b2, len) __builtin_memcmp((b1), (b2), (len)) The first line should probably get removed, but it should not be affecting vbox. vbox likes to *disable* several warnings. Thus what most likely happens is that some place using bcmp *fails* to include the systm.h header, due to warns disabled nobody is even told and the compiler ends up generating a call to bcmp (instead of memcmp, or no call in the first place). Thus someone(tm) should make sure the header is included in files which end up doing bcmp. Also will not hurt to make sure the build environment is clean and up to date -- if thep ort pulls in *OLD* files it will only see bcmp declared, but no macro telling it to use __builtin_memcmp and then it will stick to generating the call.
Link to VB bugtracker: https://www.virtualbox.org/ticket/19420
(In reply to mikhail.rokhin from comment #5) Note the driver is ours, not theirs. https://svnweb.freebsd.org/changeset/ports/446145 Please close the upstream PR.
Mahdi, can you please take a look?
The invocation of bcmp(3) appears to be hidden in strformatrt.cpp in this switch: // /* // * Regular hex dump with dittoing. // */ // case 'D': John groenveld@acm.org
Quick grep reveals stuff like: src/VBox/Additions/x11/x11include/XFree86-4.3/X11/Xfuncs.h: #else /* else vanilla BSD */ #define memmove(dst,src,len) bcopy((char *)(src),(char *)(dst),(int)(len)) #define memcpy(dst,src,len) bcopy((char *)(src),(char *)(dst),(int)(len)) #define memcmp(b1,b2,len) bcmp((char *)(b1),(char *)(b2),(int)(len)) All defines of the sort should be eliminated from all files -- memmove, memcpy and memcmp are provided as first class citizens both in the kernel and libc. For kernel code, previously mentioned sys/systm.h header needs to be included.
Any thoughts on why when I compile strformatrt.cpp with -O2 I get the bcmp symbol, but with -O0 I don't? Would that be a clang10 misfeature? John groenveld@acm.org $ c++ -c -O0 -g -pipe -pedantic -Wshadow -Wall -Wextra -Wno-missing-field-initializers -Wno-unused -Wno-trigraphs -fdiagnostics-show-option -Wno-unused-parameter -Wno-language-extension-token -Wno-gnu-folding-constant -Wno-gnu-anonymous-struct -Wno-string-plus-int -Wno-nested-anon-types -Wno-variadic-macros -Wno-long-long -Wno-return-type-c-linkage -Wno-overloaded-virtual -Wno-variadic-macros -O0 -fno-omit-frame-pointer -fno-strict-aliasing -m64 -I/usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/src/VBox/Runtime/include -I/usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/src/libs/liblzf-3.4 -I/usr/local/include -I/usr/local/include/libxml2 -I/usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/include -I/usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release -DVBOX -DVBOX_OSE -DVBOX_WITH_64_BITS_GUESTS -DVBOX_WITH_REM -DVBOX_WITH_RAW_MODE -DRT_OS_FREEBSD -D__FREEBSD__ -DRT_ARCH_AMD64 -D__AMD64__ -DVBOX_WITH_DEBUGGER -DVBOX_WITH_HARDENING -DRTPATH_APP_PRIVATE=\"/usr/local/lib/virtualbox\" -DRTPATH_APP_PRIVATE_ARCH=\"/usr/local/lib/virtualbox\" -DRTPATH_SHARED_LIBS=\"/usr/local/lib/virtualbox\" -DRTPATH_APP_DOCS=\"/usr/local/lib/virtualbox\" -D_REENTRANT -DIN_RING3 -DLOG_DISABLED -DIN_BLD_PROG -DIN_RT_R3 -DIN_ADV_BLD_PROG -DIN_RT_R3 -DLDR_WITH_NATIVE -DLDR_WITH_ELF32 -DLDR_WITH_LX -DLDR_WITH_MACHO -DLDR_WITH_PE -DRT_WITH_VBOX -DRT_NO_GIP -DRT_WITHOUT_NOCRT_WRAPPERS -DNOFILEID -DRT_WITH_ICONV_CACHE -DIPRT_WITHOUT_LDR_VERIFY -DRT_NO_GIP -Wp,-MD,/usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/RuntimeBldProg/common/string/strformatrt.o.dep -Wp,-MT,/usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/RuntimeBldProg/common/string/strformatrt.o -Wp,-MP -o /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/RuntimeBldProg/common/string/strformatrt.o /usr/ports/emulators/virtualbox-ose-additions/work/VirtualBox-5.2.34/src/VBox/Runtime/common/string/strformatrt.cpp $ nm ./work/VirtualBox-5.2.34/out/freebsd.amd64/release/obj/RuntimeBldProg/common/string/strformatrt.o | tail -8 0000000000000add r _ZZ13rtstrFormatRtE9s_szFalse U memchr U memcmp U memset U rtStrFormatKernelAddress 0000000000000000 T rtstrFormatRt U strlen U strncmp
This is probably an optimization which is only there because of something wrong with headers. There is one fundamental discrepancy between bcmp and memcmp: the former only indicates there is a difference while the latter computes what it is on the first differing byte. Thus, if bcmp is provided and buffers differ, it has less work to do than memcmp and is a better pick if available. However, FreeBSD headers explicitly define bcmp as builtin_memcmp, therefore the above should not be happening if they are properly included and the defines are not overwritten. Once more this is just some build mess.
strformatrt.cpp compiles with gcc9 without the bcmp(3) reference but the package fails later due to a clang vs gcc flag conflict that I'll try to track down later. I would like to try to build the package with clang and -O0. The second invocation of -O2 is set in Config.kmk which I can patch. But the first one is a mystery. I tried setting the port Makefile with CXXFLAGS and CFLAGS but that didn't take. John groenveld@acm.org
Created attachment 213246 [details] stack trace screen snapshot
Comment on attachment 213246 [details] stack trace screen snapshot VBOXVFS[1]: sfprov_mount: path: [browser] sfprov_mount(browser): error=0 rc=0 panic: getnewvnode: not registered vector op 0xffffffff82a78418 cpuid = 0 time = 1586551831 KDB: stack backtrace: db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfffffe00250ee5c0 vpanic() at vpanic+0x182/frame 0xfffffe00250ee610 panic() at panic+0x43/frame 0xfffffe00250ee670 getnewvnode() at getnewvnode+0x254/frame 0xfffffe00250ee6b0 vboxfs_alloc_vp() at vboxfs_alloc_vp+0x470/frame 0xfffffe00250ee790 vboxfs_root() at vboxfs_root+0x32/frame 0xfffffe00250ee7c0 vfs_domount() at vfs_domount+0x85c/frame 0xfffffe00250ee9f0 vfs_donmount() at vfs_donmount+0x871/frame 0xfffffe00250eea90 sys_nmount() at sys_nmount+0x69/frame 0xfffffe00250eead0 amd64_syscall() at amd64_syscall+0x140/frame 0xfffffe00250eebf0 fast_syscall_common() at fast_syscall_common+0x101/frame 0xfffffe00250eebf0 --- syscall (378, FreeBSD ELF64, sys_nmount), rip = 0x8002f9fea, rsp = 0x7fffffffd3e8, rbp = 0x7fffffffd950 --- KDB: enter: panic [ thread pid 862 tid 100464 ] Stopped at kdb_enter+0x37: movq $0,0x10927d6(%rip) db>
This one should be easy: it needs VFS_VOP_VECTOR_REGISTER to be added for the argument passed to getnewvnode, grep will show you plenty of examples in the tree.
Thank you! vboxvfs_vnops.c + #if __FreeBSD_version > 1300000 + VFS_VOP_VECTOR_REGISTER(vboxfs_vnodeops); + #endif John groenveld@acm.org
John-Mark Gurney's patch worked for me with FreeBSD 13.0-CURRENT #0 r360723: <URL:https://lists.freebsd.org/pipermail/freebsd-virtualization/2020-May/008448.html> John groenveld@acm.org