summaryrefslogtreecommitdiffstats
path: root/src/guestfs.c
Commit message (Collapse)AuthorAgeFilesLines
...
* Mac OS X: Disable null vmchannel implementation.Richard Jones2010-03-221-1/+14
| | | | | | | | | As far as I can tell, Darwin has no way to check the peer euid of a _loopback_ TCP socket. This is required for the "null vmchannel" implementation to work securely. Therefore disable this - Darwin will use one of the other supported vmchannel implementations instead.
* Mac OS X: 'xdr_uint32_t' is 'xdr_u_int32_t'Richard Jones2010-03-221-0/+4
|
* Mac OS X: Fix HAVE_GNU_CALLOC so it works when __GLIBC__ is not defined.Richard Jones2010-03-221-0/+4
|
* Mac OS X: strerror_r on Macs is not like GNU strerror_r.Richard Jones2010-03-221-1/+1
| | | | | Really this should be turned into a configure-time test. Perhaps one exists already?
* Mac OS X: define MAX macro if not already defined.Richard Jones2010-03-221-0/+4
|
* Mac OS X: kill(2) requires <signal.h>Richard Jones2010-03-221-0/+1
|
* Run qemu with -nographic option.Richard Jones2010-03-221-2/+3
| | | | | On Mac OS X this prevents a short "flash" as qemu opens a toplevel window.
* add_drive_ro adds readonly=on option if available.Richard Jones2010-03-161-4/+18
| | | | | | | | | | | | | | Change the add_drive_ro call so it adds the readonly=on option if qemu supports that. This just means that qemu will not try to open the drive with O_RDWR, and should not otherwise change the behaviour of qemu or libguestfs. (In particular, writes to the read-only drive are still permitted, and are just discarded when the handle is closed). However it should alleviate RHBZ#571714 where udev was deciding to incorrectly relabel a device because we had opened the device for writing (even though we didn't actually write to it).
* Allow qemu_supports to run earlier.Richard Jones2010-03-161-9/+18
| | | | | | | Reimplement qemu_supports() internal function. Allow it to run before launch so we can test qemu features. Document that you should run guestfs_set_qemu as early as possible to make sure these tests are reliable.
* Rewrite libguestfs-supermin-helper in C.Richard Jones2010-03-121-1/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | libguestfs-supermin-helper was previously a shell script. Although we had steadily optimized it, there were a number of intractable hot spots: (1) cpio still reads input files in 512 byte chunks; this is *very* pessimal behaviour, particularly when SELinux is enabled. (2) the hostfiles globbing was done very inefficiently by the shell, with the shell rereading the same directory over and over again. This is a rewrite of this shell script in C. It is approximately 3 times faster without SELinux, and has an even greater speed difference with SELinux. The main features are: (a) It never frees memory, making it simpler. The program is designed to run and exit in sub-second times, so this is acceptable. (b) It caches directory reads, making the globbing of host files much faster (measured this as ~ 4 x speed up). (c) It doesn't use external cpio, but instead contains code to write newc format cpio files, which is all that the kernel can read. Unlike cpio, this code uses large buffers for reads and writes. (d) Ignores missing or unreadable hostfiles, whereas cpio gave a warning. (e) Checks all return values from system calls. (f) With --verbose flag, it will print messages timing itself. This passes all tests. Updated with feedback from Jim Meyering.
* Remove some unused variables.Richard Jones2010-02-151-2/+1
| | | | | | Since we have to compile with -Wno-unused-variables, we don't spot unused variables in code. I found these by compiling the code in Ubuntu.
* New APIs: add-drive{,-ro}-with-if allows you to set QEMU block emulation.Richard Jones2010-02-121-5/+19
| | | | | | | | | | | The default if=... comes from configure time (currently it defaults to if=virtio). This change allows you to set the QEMU block emulation. We don't think this will be used very often, but virt-v2v requires it in order to work around a subtle problem with running 'mkinitrd' in an appliance attached to a guest.
* Add -enable-kvm option to qemu command line.Richard Jones2010-02-091-0/+31
| | | | | | | | If the version of qemu being used supports -enable-kvm option, then check if /dev/kvm is openable and add this option. I have found this option makes no difference, although it is *supposed* to enable KVM (hardware virtualization) support.
* supermin: Prevent multilib corruption (RHBZ#558593).Richard Jones2010-01-271-2/+4
| | | | | | | | | | | | | | | | | | | On some combination of installing, upgrading and removing the base libguestfs package on x86_64, multilib can corrupt libguestfs by leaving a copy of /usr/bin/libguestfs-supermin-helper around which references the wrong architecture (usually, contains links to the i386-based appliance, when the x86_64 appliance should be constructed). This commit changes libguestfs-supermin-helper so that the script is the same on all architectures. Instead, the library passes the differences to the script (eg. $host_cpu). Because the i386 and x86_64 libraries should be at different locations (/usr/lib vs /usr/lib64) this should prevent multilib from screwing things up. Related links: https://bugzilla.redhat.com/show_bug.cgi?id=558593 http://rwmj.wordpress.com/2009/11/16/please-someone-shoot-multilib/#content https://bugzilla.redhat.com/show_bug.cgi?id=235752
* qemu: Upstream regression of -stdio serial option.Richard Jones2010-01-071-4/+17
| | | | | | | | | | | | | Best explained by the comment in the code: /* Newer versions of qemu (from around 2009/12) changed the * behaviour of monitors so that an implicit '-monitor stdio' is * assumed if we are in -nographic mode and there is no other * -monitor option. Only a single stdio device is allowed, so * this broke the '-serial stdio' option. There is a new flag * called -nodefaults which gets rid of all this default crud, so * let's use that to avoid this and any future surprises. */
* lib: Add thread-safety to global list of handles.Richard Jones2009-12-071-9/+10
| | | | | | | | | | | | | | | | This commit uses the Gnulib 'lock' module to implement a mutex on the global list of handles which is stored by the library. Note that Gnulib nicely avoids explicitly linking with -lpthread unless the application program itself links to -lpthread. Locks are only enabled in multithreaded applications. $ ldd src/.libs/libguestfs.so.0.217.0 linux-vdso.so.1 => (0x00007fffcb7ff000) libc.so.6 => /lib64/libc.so.6 (0x00007f96a4e6c000) /lib64/ld-linux-x86-64.so.2 (0x00007f96a544d000) Please enter the commit message for your changes. Lines starting
* Don't export STREQ and friends in <guestfs.h>Richard Jones2009-11-101-0/+1
| | | | | Move these to private header file(s) and other places as required since these aren't part of the public API.
* Fix problems found by 'make syntax check'Richard Jones2009-11-101-1/+1
|
* Print timestamped messages during appliance launch.Richard Jones2009-11-101-0/+54
| | | | | In verbose mode, print timestamped messages during guestfs_launch so we can see how long each step takes.
* Record time of guest launch.Richard Jones2009-11-101-6/+7
| | | | | | | | | | The guest handle field start_t was previously used (when we had the wait_ready call), but had fallen into disuse. Note that it could never be accessed through the API. Rename this field as launch_t, convert it to a timeval, and use it to measure the time since guestfs_launch was called so that we can start profiling guest launch.
* appliance: Prefix kernel messages with timestamp.Richard Jones2009-11-101-0/+1
|
* use STREQ, not strcmp: part 1Jim Meyering2009-11-091-9/+9
| | | | | git grep -l 'strcmp *([^=]*== *0'|xargs \ perl -pi -e 's/\bstrcmp( *\(.*?\)) *== *0/STREQ$1/g'
* Allow callers to disable the recovery process.Richard Jones2009-11-031-26/+45
|
* Set LC_ALL=C before running qemu.Richard Jones2009-10-211-2/+4
|
* avoid use of all ctype macrosJim Meyering2009-09-241-2/+2
| | | | | | | | | | | | | | | | | | | | * cfg.mk (disable_temporarily): Don't disable sc_avoid_ctype_macros. * fish/tilde.c: Remove unnecessary inclusion of ctype.h. * bootstrap: Add gnulib's c-ctype module to the list. * daemon/m4/gnulib-cache.m4: Likewise. * daemon/ext2.c: Include "c-ctype.h", not <ctype.h>. Use c_isspace, etc, rather than isspace. * daemon/guestfsd.c: Likewise. * daemon/lvm.c: Likewise. * daemon/proto.c: Likewise. * fish/fish.c: Likewise. * fish/tilde.c: Likewise. * src/generator.ml: Likewise. * src/guestfs.c: Likewise. * examples/to-xml.c: Likewise. * examples/Makefile.am (to_xml_CPPFLAGS): Add -I$(top_srcdir)/gnulib/lib so inclusion of "c-ctype.h" works. (to_xml_CPPFLAGS): Rename from to_xml_CFLAGS.
* Add direct appliance mode flag and implementation.Richard Jones2009-09-231-33/+69
| | | | | | | | When the g->direct flag is set, the appliance stdin/stdout are not connected to the library. Instead they inherit the stdin/stdout of the caller. This is used to implement virt-rescue.
* Change handling of spaces on Linux kernel command line.Richard Jones2009-09-221-8/+8
|
* Implement "null vmchannel" - no vmchannel needed!Richard Jones2009-09-221-64/+297
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This commit removes the requirement for vmchannel, although retaining support for it. In this configuration, known as "null vmchannel", the library listens on a random loopback port. It passes the number of this port to the appliance (guestfs_vmchannel=tcp:10.0.2.2:12345), and the daemon then connects back. The library, instead of connecting, listens and accepts the connection during guestfs_launch. QEMU SLIRP (user mode networking) is still required to make this work: SLIRP forwards the TCP connection transparently (instead of explicitly as with guestfwd) to 127.0.0.1:<port> There is a window in which any other local process on the machine could see the port number in the qemu command line and try to connect to it. This would be a devastating security hole, because any local process could pretend to be the daemon and feed back malicious replies to our remote procedure calls. To prevent this, we check the UID of the other side of the TCP connection. If the UID is different from the library's EUID, then we reject the connection. To do this we have to parse /proc/net/tcp. (On Solaris we could use getsockopt (SO_PEERCRED), but this doesn't work on Linux TCP sockets). Other vmchannel(s) are still supported. This is important, because we can't in general be sure the qemu will always support SLIRP. In particular, in recent versions of qemu it is possible to compile out SLIRP.
* Flexible guestfs_vmchannel parameter for future appliances.Richard Jones2009-09-221-0/+4
| | | | | | | | | | | | | | | | | | | | | | This reimplements parts of commit da0a4f8d1f6ddd302ceba028d87c6e009589e503 in a different, but compatible way. We pass guestfs_vmchannel=tcp:<ip>:<port> on the command line. This is intended to be used as follows (now and in future versions): tcp:10.0.2.4:6666 for guestfwd vmchannel tcp:10.0.2.2:<port> for future "no vmchannel" implementation /dev/vcon4 for future virtio-console vmchannel* It also accepts the old-style guestfs=10.0.2.4:6666 parameter which is sent by older libraries, and turns this transparently into the correct format above. If no guestfs_vmchannel is passed, then this defaults to the guestfwd vmchannel which older libraries would expect. * Maybe this last one should be dev:/dev/vcon4 or file:/dev/vcon4, but we don't need to decide that now.
* Make GUESTFWD_PORT into a string.Richard Jones2009-09-211-4/+4
|
* Remove unnecessary extra space from qemu command line.Richard Jones2009-09-211-1/+1
|
* Remove guestfs_wait_ready (turn it into a no-op).Richard Jones2009-09-211-32/+30
| | | | | | | | | | | | | | This commit changes guestfs_launch so that it both launches the appliance and waits until it is ready (ie. the daemon communicates back to us). Since we removed the pretence that we could implement a low-level asynchronous API, the need to call launch() followed by wait_ready() has looked a bit silly. Now guestfs_wait_ready() is basically a no-op. It is left in the API for backwards compatibility. Any calls to guestfs_wait_ready() can be removed from client code.
* Combine temporary buffers.Richard Jones2009-09-211-7/+5
| | | | | | | buf[], append[] and memsize_str[] were all temporary buffers used in non-overlapping code. Combine them to use a single buffer (buf[]).
* Rearrange qemu command line order (no functional change).Richard Jones2009-09-211-29/+30
|
* Enable new-style -chardev ... guestfwd command line.Richard Jones2009-09-181-20/+26
| | | | | | | | | | | | Newer versions of qemu have changed the command line format (again). '-net channel' is now deprecated. Instead we use: -chardev socket,id=guestfsvmc,path=/path/to/sock,server,nowait -net user,vlan=0,net=10.0.2.0/8,guestfwd=tcp:10.0.2.4:6666-chardev:guestfsvmc -net nic,model=ne2k_pci,vlan=0 The old style format is still used if the new style is not detected at runtime.
* Remove explicit guestfs=10.0.2.4:6666 kernel command line parameter.Richard Jones2009-09-171-6/+4
| | | | | | | | | | Since we control the appliance tightly, we can just specify that it will always use a particular host and port, and we don't need to pass it on the command line each time. Also the VMCHANNEL_* constants are only relevant to the particular guestfwd vmchannel implementation, so we rename them as GUESTFWD_*.
* Fix verbose packet dumping functions.Richard Jones2009-09-171-7/+7
| | | | | | | | | | Add the configure parameter --enable-packet-dump so that this code can be enabled without editing the source. This code is normally commented out, because it is too verbose unless you happen to be debugging the underlying protocol. Because it is normally commented out, I found it had bit-rotted slightly. This commit also fixes the obvious problems.
* Fix comparison between signed and unsigned (for RHEL 5).1.0.69Richard Jones2009-09-151-1/+1
|
* Remove main loop.Richard Jones2009-09-141-883/+478
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This commit removes the external main loop, which never worked and caused a number of bugs. Requests are now done synchronously, and if the user wants to have requests issued in the background or to have a responsive GUI, then they'll just have to use threads. The big change is to push all reads and writes through two functions called send_to_daemon (for writes) and recv_from_daemon (for reads) which operate synchronously. These functions read/write whole messages, and also handle checking for EOF (ie. daemon died) and asynchronous log message events from qemu (eg. from debug / dmesg printed by the guest). A more complete description of how these work can be found in the code. This code passes a complete run of the tests. Bugs believed to be fixed by this commit: https://bugzilla.redhat.com/show_bug.cgi?id=501888 internal error: reply callback called twice https://bugzilla.redhat.com/show_bug.cgi?id=504418 In virt-inspector: "download: guestfs_download reply failed, see earlier error messages" I have tried to avoid reintroducing this: https://bugzilla.redhat.com/show_bug.cgi?id=508713 libguestfs: error: write: Broken pipe (guestfish only) One other benefit of this is that 'set_busy/end_busy' calls no longer appear in traces.
* Add command trace functionality.Richard Jones2009-09-141-0/+17
| | | | | Enable this by calling guestfs_trace (handle, 1) or by setting the LIBGUESTFS_TRACE=1 environment variable.
* Non-daemon actions indirect through generated code.Richard Jones2009-09-141-32/+33
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Previously non-daemon actions were called directly by user code, eg: /* Non-generated */ int guestfs_set_verbose (guestfs_h *g, int v) { g->verbose = !!v; return 0; } This changes these actions so they go indirectly via some generated code, eg: /* Generated */ int guestfs_set_verbose (guestfs_h *g, int verbose) { return guestfs__set_verbose (g, verbose); } /* Non-generated */ int guestfs__set_verbose (guestfs_h *g, int v) { g->verbose = !!v; return 0; } The aim is to have a place in the generated code where we can add debug or tracing information for these non- daemon calls.
* maint: guestfs.c: remove unnecessary initializationJim Meyering2009-08-311-1/+1
| | | | | * src/guestfs.c (guestfs__receive_file_sync): Don't set "r", only to ignore it.
* build: don't define _GNU_SOURCE manuallyJim Meyering2009-08-241-1/+0
| | | | | | | | | | | | Now that we're using gnulib in earnest, any manual definition would provoke a redefinition warning. * fish/fish.c (_GNU_SOURCE): Don't define. * fish/destpaths.c (_GNU_SOURCE): Likewise. * src/guestfs.c (_GNU_SOURCE): Likewise. * bootstrap (modules): Add asprintf, strchrnul, strerror, strndup and vasprintf. * fish/fish.c (main): Set argv[0] to sanitized program_name, so functions like getopt_long that use argv[0] use the clean name.
* Add waitpid along guestfs_close path (RHBZ#518747).Richard Jones2009-08-221-0/+4
|
* guestfs_launch: Correct checks for dup failure.Richard Jones2009-08-191-7/+7
|
* build: Comment out some unused macros.Richard Jones2009-08-191-3/+3
| | | | | GCC 4.4.1 warns about some unused macros. Comment these out while they are not used.
* suppress warnings from -Wmissing-noreturnJim Meyering2009-08-181-0/+2
| | | | | | | Even though these functions are marked as "not implemented yet", and they will surely return a value once implemented, ... * src/guestfs.c (select_add_timeout): Declare with noreturn attribute. (select_remove_timeout): Likewise.
* suppress a warning from -Wswitch-defaultJim Meyering2009-08-181-0/+2
| | | | * src/guestfs.c (guestfs_end_busy): Add a "default:" label.
* suppress signed/unsigned-comparison warningsJim Meyering2009-08-181-2/+2
| | | | | * src/guestfs.c [struct guestfs_h] (msg_in_size, msg_in_allocated): (msg_out_size, msg_out_allocated): Change type from int to unsigned int.
* build: don't perform arithmetic on void* pointersJim Meyering2009-08-181-3/+5
| | | | * src/guestfs.c (receive_file_data_sync, xread, xwrite): Use char*.
generated by cgit (git 2.34.1) at 2025-09-01 05:40:45 +0000