migrations: Fix potential rare race of migration-test after yank
v3 cover migrations: Fix potential rare race of migration-test after yank
v3:
- Use WITH_QEMU_LOCK_GUARD() for patch 2 [Eric]
(potentially I can also replace other existing uses of qemu_file_lock into
WITH_QEMU_LOCK_GUARD, but I decided to took Dave's r-b first and leave that for later)
- Added r-bs for Dave on patch 2/4
- Add a comment in patch 5 to explain why it's safe to unregister yank without qemu_file_lock, in postcopy_pause() [Dave]
v2:
- Pick r-b for Dave on patch 1/3
- Move migration_file_get_ioc() from patch 5 to patch 4, meanwhile rename it to qemu_file_get_ioc(). [Dave]
- Patch 2 "migration: Shutdown src in await_return_path_close_on_source()" is replaced by patch "migration: Make from_dst_file accesses thread-safe" [Dave]
Patch 1 fixes a possible race that migration thread can accidentally skip join() of rp_thread even if the return thread is enabled. Patch 1 is suspected to also be the root cause of the recent hard-to-reproduce migration-test failure here reported by PMM:
https://lore.kernel.org/qemu-devel/YPamXAHwan%2FPPXLf@work-vm/
I didn't reproduce it myself; but after co-debugged with Dave it's suspected that the race of rp_thread could be the cause. It's not exposed before because yank is soo strict on releasing instances, while we're not that strict before, and didn't join() on rp_thread wasn't so dangerous after all when migration succeeded before.
Patch 2 fixes another theoretical race on accessing from_dst_file spotted by Dave. I don't think there's known issues with it, but may still worth fixing.
Patch 3 should be a cleanup on yank that I think would be nice to have.
Patch 4-5 are further cleanups to remove the ref==1 check in channel_close(), finally, as I always thought that's a bit hackish. So I used explicit unregister of the yank function at necessary places to replace that ref==1 one.
I still think having patch 3-5 altogether would be great, however I think patch 1 should still be the most important to be reviewed. Also it would be great to know whether patch 1 could fix the known yank crash.
Please review, thanks.
Peter Xu (5):
migration: Fix missing join() of rp_thread
migration: Make from_dst_file accesses thread-safe
migration: Introduce migration_ioc_[un]register_yank()
migration: Teach QEMUFile to be QIOChannel-aware
migration: Move the yank unregister of channel_close out
migration/channel.c | 15 ++------- migration/migration.c | 57 +++++++++++++++++++++++++++-------- migration/migration.h | 15 +++++++-- migration/multifd.c | 8 ++--- migration/qemu-file-channel.c | 11 ++----- migration/qemu-file.c | 17 ++++++++++- migration/qemu-file.h | 4 ++- migration/ram.c | 3 +- migration/savevm.c | 11 +++++-- migration/yank_functions.c | 42 ++++++++++++++++++++++++++ migration/yank_functions.h | 3 ++ 11 files changed, 138 insertions(+), 48 deletions(-)
Signature · 1 line
2.31.1
PATCH v3 1/5 migration: Fix missing join() of rp_thread
It's possible that the migration thread skip the join() of the rp_thread in below race and crash on src right at finishing migration:
migration_thread rp_thread
---------------- ---------
migration_completion()
(before rp_thread quits)
from_dst_file=NULL
[thread got scheduled out]
s->rp_state.from_dst_file==NULL
(skip join() of rp_thread)
migrate_fd_cleanup()
qemu_fclose(s->to_dst_file)
yank_unregister_instance()
assert(yank_find_entry()) <------- crashIt could mostly happen with postcopy, but that shouldn't be required, e.g., I think it could also trigger with MIGRATION_CAPABILITY_RETURN_PATH set.
It's suspected that above race could be the root cause of a recent (but rare) migration-test break reported by either Dave or PMM:
https://lore.kernel.org/qemu-devel/YPamXAHwan%2FPPXLf@work-vm/
The issue is: from_dst_file is reset in the rp_thread, so if the thread reset it to NULL fast enough then the migration thread will assume there's no rp_thread at all.
This could potentially cause more severe issue (e.g. crash) after the yank code.
Fix it by using a boolean to keep "whether we've created rp_thread".
--- migration/migration.c | 4 +++- migration/migration.h | 7 +++++++ 2 files changed, 10 insertions(+), 1 deletion(-)
migration/migration.c +3 -1
@@ -2867,6 +2867,7 @@ static int open_return_path_on_source(MigrationState *ms,qemu_thread_create (&ms->rp_state.rp_thread,"return path" , source_return_path_thread, ms,QEMU_THREAD_JOINABLE ); + ms->rp_state.rp_thread_created = true;trace_open_return_path_on_source_continue ();
@@ -2891,6 +2892,7 @@ static int await_return_path_close_on_source(MigrationState *ms) }trace_await_return_path_close_on_source_joining ();qemu_thread_join (&ms->rp_state.rp_thread); + ms->rp_state.rp_thread_created = false;trace_await_return_path_close_on_source_close ();return ms->rp_state.error; }
@@ -3170,7 +3172,7 @@ static void migration_completion(MigrationState *s)* it will wait for the destination to send it's status in * a SHUT command). */ -if (s->rp_state.from_dst_file) { +if (s->rp_state.rp_thread_created) {int rp_error;trace_migration_return_path_end_before (); rp_error =await_return_path_close_on_source (s);
migration/migration.h +7 -0
@@ -195,6 +195,13 @@ struct MigrationState {QEMUFile *from_dst_file;QemuThread rp_thread;bool error; +/* +* We can also check non-zero of rp_thread, but there's no "official" +* way to do this, so this bool makes it slightly more elegant. +* Checking from_dst_file for this is racy because from_dst_file will +* be cleared in the rp_thread! +*/ +bool rp_thread_created;QemuSemaphore rp_sem; } rp_state;
Signature · 1 line
2.31.1
PATCH v3 2/5 migration: Make from_dst_file accesses thread-safe
In reply to Peter XuAccessing from_dst_file is potentially racy in current code base like below:
if (s->from_dst_file)
do_something(s->from_dst_file);Because from_dst_file can be reset right after the check in another thread (rp_thread). One example is migrate_fd_cancel().
Use the same qemu_file_lock to protect it too, just like to_dst_file.
When it's safe to access without lock, comment it.
There's one special reference in migration_thread() that can be replaced by the newly introduced rp_thread_created flag.
--- migration/migration.c | 39 +++++++++++++++++++++++++++++---------- migration/migration.h | 8 +++++--- migration/ram.c | 1 + 3 files changed, 35 insertions(+), 13 deletions(-)
migration/migration.c +29 -10
@@ -1879,9 +1879,11 @@ static void migrate_fd_cancel(MigrationState *s)QEMUFile *f =migrate_get_current ()->to_dst_file;trace_migrate_fd_cancel (); -if (s->rp_state.from_dst_file) { -/* shutdown the rp socket, so causing the rp thread to shutdown */ -qemu_file_shutdown (s->rp_state.from_dst_file); + WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { +if (s->rp_state.from_dst_file) { +/* shutdown the rp socket, so causing the rp thread to shutdown */ +qemu_file_shutdown (s->rp_state.from_dst_file); + } } do {
@@ -2686,6 +2688,23 @@ static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)return 0 ; } +/* Release ms->rp_state.from_dst_file in a safe way */ +static void migration_release_from_dst_file (MigrationState *ms) +{ +QEMUFile *file; + + WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { +/* +* Reset the from_dst_file pointer first before releasing it, as we +* can't block within lock section +*/ + file = ms->rp_state.from_dst_file; + ms->rp_state.from_dst_file =NULL ; + } + +qemu_fclose (file); +} +/* * Handles messages sent on the return path towards the source VM *
@@ -2827,11 +2846,13 @@ out:* Maybe there is something we can do: it looks like a * network down issue, and we pause for a recovery. */ -qemu_fclose (rp); - ms->rp_state.from_dst_file =NULL ; +migration_release_from_dst_file (ms); rp =NULL ;if (postcopy_pause_return_path_thread (ms)) { -/* Reload rp, reset the rest */ +/* +* Reload rp, reset the rest. Referencing it is safe since +* it's reset only by us above, or when migration completes +*/ rp = ms->rp_state.from_dst_file; ms->rp_state.error = false; goto retry;
@@ -2843,8 +2864,7 @@ out: }trace_source_return_path_thread_end (); - ms->rp_state.from_dst_file =NULL ; -qemu_fclose (rp); +migration_release_from_dst_file (ms);rcu_unregister_thread ();return NULL ; }
@@ -2852,7 +2872,6 @@ out:static int open_return_path_on_source (MigrationState *ms,bool create_thread) { - ms->rp_state.from_dst_file =qemu_file_get_return_path (ms->to_dst_file);if (!ms->rp_state.from_dst_file) {return -1 ;
@@ -3746,7 +3765,7 @@ static void *migration_thread(void *opaque)* If we opened the return path, we need to make sure dst has it * opened as well. */ -if (s->rp_state.from_dst_file) { +if (s->rp_state.rp_thread_created) {/* Now tell the dest that it should open its end so it can reply */ qemu_savevm_send_open_return_path (s->to_dst_file);
migration/migration.h +5 -3
@@ -154,12 +154,13 @@ struct MigrationState {QemuThread thread;QEMUBH *vm_start_bh;QEMUBH *cleanup_bh; +/* Protected by qemu_file_lock */ QEMUFile *to_dst_file;QIOChannelBuffer *bioc;/* -* Protects to_dst_file pointer. We need to make sure we won't -* yield or hang during the critical section, since this lock will -* be used in OOB command handler. +* Protects to_dst_file/from_dst_file pointers. We need to make sure we +* won't yield or hang during the critical section, since this lock will be +* used in OOB command handler. */ QemuMutex qemu_file_lock;
@@ -192,6 +193,7 @@ struct MigrationState {/* State related to return path */ struct { +/* Protected by qemu_file_lock */ QEMUFile *from_dst_file;QemuThread rp_thread;bool error;
migration/ram.c +1 -0
@@ -4012,6 +4012,7 @@ static void ram_dirty_bitmap_reload_notify(MigrationState *s)int ram_dirty_bitmap_reload (MigrationState *s,RAMBlock *block) {int ret = -EINVAL ; +/* from_dst_file is always valid because we're within rp_thread */ QEMUFile *file = s->rp_state.from_dst_file;unsigned long *le_bitmap, nbits = block->used_length >>TARGET_PAGE_BITS ;uint64_t local_size =DIV_ROUND_UP (nbits,8 );
Signature · 1 line
2.31.1
Re: [PATCH v3 2/5] migration: Make from_dst_file accesses thread-safe
Quoted 12 lines · Peter Xu
Accessing
from_dst_fileis potentially racy in current code base like below:if (s->from_dst_file) do_something(s->from_dst_file);Because
from_dst_filecan be reset right after the check in another thread (rp_thread). One example ismigrate_fd_cancel().Use the same
qemu_file_lockto protect it too, just liketo_dst_file.When it's safe to access without lock, comment it.
Quoted 7 lines
There's one special reference in
migration_thread()that can be replaced by the newly introducedrp_thread_createdflag.Reported-by Dr. David Alan Gilbert <[email protected]>Reviewed-by Dr. David Alan Gilbert <[email protected]>Signed-off-by Peter Xu <[email protected]>---
(Dave should have helped fixing this which I appreciated a lot, but just make
it be together with the record..)
Below needs to be squashed into the patch:
migration/migration.c +1 -1
@@ -2696,7 +2696,7 @@ static void migration_release_from_dst_file(MigrationState *ms) {QEMUFile *file; - WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { + WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) {/* * Reset the from_dst_file pointer first before releasing it, as we * can't block within lock section
Signature · 1 line
Peter Xu
PATCH v3 3/5 migration: Introduce migration_ioc_[un]register_yank()
In reply to Peter XuThere're plenty of places in migration/* that checks against either socket or tls typed ioc for yank operations. Provide two helpers to hide all these information.
--- migration/channel.c | 15 ++------------- migration/multifd.c | 8 ++------ migration/qemu-file-channel.c | 8 ++------ migration/yank_functions.c | 28 ++++++++++++++++++++++++++++ migration/yank_functions.h | 2 ++ 5 files changed, 36 insertions(+), 25 deletions(-)
migration/channel.c +2 -13
@@ -44,13 +44,7 @@ void migration_channel_process_incoming(QIOChannel *ioc) TYPE_QIO_CHANNEL_TLS)) { migration_tls_channel_process_incoming(s, ioc, &local_err); } else { -if (object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_SOCKET ) || -object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) { -yank_register_function (MIGRATION_YANK_INSTANCE , - migration_yank_iochannel, -QIO_CHANNEL (ioc)); - } - +migration_ioc_register_yank (ioc);migration_ioc_process_incoming (ioc, &local_err); }
@@ -94,12 +88,7 @@ void migration_channel_connect(MigrationState *s, }else {QEMUFile *f =qemu_fopen_channel_output (ioc); -if (object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_SOCKET ) || -object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) { -yank_register_function (MIGRATION_YANK_INSTANCE , - migration_yank_iochannel, -QIO_CHANNEL (ioc)); - } +migration_ioc_register_yank (ioc);qemu_mutex_lock (&s->qemu_file_lock); s->to_dst_file = f;
migration/multifd.c +2 -6
@@ -987,12 +987,8 @@ int multifd_load_cleanup(Error **errp)for (i =0 ; i <migrate_multifd_channels (); i++) {MultiFDRecvParams *p = &multifd_recv_state->params[i]; -if ((object_dynamic_cast (OBJECT (p->c),TYPE_QIO_CHANNEL_SOCKET ) || -object_dynamic_cast (OBJECT (p->c),TYPE_QIO_CHANNEL_TLS )) - &&OBJECT (p->c)->ref ==1 ) { -yank_unregister_function (MIGRATION_YANK_INSTANCE , - migration_yank_iochannel, -QIO_CHANNEL (p->c)); +if (OBJECT (p->c)->ref ==1 ) { +migration_ioc_unregister_yank (p->c); }object_unref (OBJECT (p->c));
migration/qemu-file-channel.c +2 -6
@@ -107,12 +107,8 @@ static int channel_close(void *opaque, Error **errp)int ret;QIOChannel *ioc =QIO_CHANNEL (opaque); ret =qio_channel_close (ioc, errp); -if ((object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_SOCKET ) || -object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) - &&OBJECT (ioc)->ref ==1 ) { -yank_unregister_function (MIGRATION_YANK_INSTANCE , - migration_yank_iochannel, -QIO_CHANNEL (ioc)); +if (OBJECT (ioc)->ref ==1 ) { +migration_ioc_unregister_yank (ioc); }object_unref (OBJECT (ioc));return ret;
migration/yank_functions.c +28 -0
@@ -11,6 +11,9 @@#include "qapi/error.h" #include "io/channel.h" #include "yank_functions.h" +#include "qemu/yank.h" +#include "io/channel-socket.h" +#include "io/channel-tls.h" void migration_yank_iochannel (void *opaque) {
@@ -18,3 +21,28 @@ void migration_yank_iochannel(void *opaque)qio_channel_shutdown (ioc,QIO_CHANNEL_SHUTDOWN_BOTH ,NULL ); } + +/* Return whether yank is supported on this ioc */ +static bool migration_ioc_yank_supported (QIOChannel *ioc) +{ +return object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_SOCKET ) || +object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS ); +} + +void migration_ioc_register_yank (QIOChannel *ioc) +{ +if (migration_ioc_yank_supported (ioc)) { +yank_register_function (MIGRATION_YANK_INSTANCE , + migration_yank_iochannel, +QIO_CHANNEL (ioc)); + } +} + +void migration_ioc_unregister_yank (QIOChannel *ioc) +{ +if (migration_ioc_yank_supported (ioc)) { +yank_unregister_function (MIGRATION_YANK_INSTANCE , + migration_yank_iochannel, +QIO_CHANNEL (ioc)); + } +}
migration/yank_functions.h +2 -0
@@ -15,3 +15,5 @@* @opaque: QIOChannel to shutdown */ void migration_yank_iochannel (void *opaque); +void migration_ioc_register_yank (QIOChannel *ioc); +void migration_ioc_unregister_yank (QIOChannel *ioc);
Signature · 1 line
2.31.1
PATCH v3 4/5 migration: Teach QEMUFile to be QIOChannel-aware
In reply to Peter Xumigration uses QIOChannel typed qemufiles. In follow up patches, we'll need the capability to identify this fact, so that we can get the backing QIOChannel from a QEMUFile.
We can also define types for QEMUFile but so far since we only need to be able to identify QIOChannel, introduce a boolean which is simpler.
Introduce another helper qemu_file_get_ioc() to return the ioc backend of a qemufile if has_ioc is set.
No functional change.
--- migration/qemu-file-channel.c | 4 ++-- migration/qemu-file.c | 17 ++++++++++++++++- migration/qemu-file.h | 4 +++- migration/ram.c | 2 +- migration/savevm.c | 4 ++-- 5 files changed, 24 insertions(+), 7 deletions(-)
migration/qemu-file-channel.c +2 -2
@@ -187,11 +187,11 @@ static const QEMUFileOps channel_output_ops = {QEMUFile *qemu_fopen_channel_input (QIOChannel *ioc) {object_ref (OBJECT (ioc)); -return qemu_fopen_ops (ioc, &channel_input_ops); +return qemu_fopen_ops (ioc, &channel_input_ops, true); }QEMUFile *qemu_fopen_channel_output (QIOChannel *ioc) {object_ref (OBJECT (ioc)); -return qemu_fopen_ops (ioc, &channel_output_ops); +return qemu_fopen_ops (ioc, &channel_output_ops, true); }
migration/qemu-file.c +16 -1
@@ -55,6 +55,8 @@ struct QEMUFile {Error *last_error_obj;/* has the file has been shutdown */ bool shutdown; +/* Whether opaque points to a QIOChannel */ +bool has_ioc; };/*
@@ -101,7 +103,7 @@ bool qemu_file_mode_is_not_valid(const char *mode)return false; } -QEMUFile *qemu_fopen_ops (void *opaque,const QEMUFileOps *ops) +QEMUFile *qemu_fopen_ops (void *opaque,const QEMUFileOps *ops,bool has_ioc) {QEMUFile *f;
@@ -109,6 +111,7 @@ QEMUFile *qemu_fopen_ops(void *opaque, const QEMUFileOps *ops) f->opaque = opaque; f->ops = ops; + f->has_ioc = has_ioc;return f; }
@@ -851,3 +854,15 @@ void qemu_file_set_blocking(QEMUFile *f, bool block) f->ops->set_blocking (f->opaque, block,NULL ); } } + +/* +* Return the ioc object if it's a migration channel. Note: it can return NULL +* for callers passing in a non-migration qemufile. E.g. see qemu_fopen_bdrv() +* and its usage in e.g. load_snapshot(). So we need to check against NULL +* before using it. If without the check, migration_incoming_state_destroy() +* could fail for load_snapshot(). +*/ +QIOChannel *qemu_file_get_ioc (QEMUFile *file) +{ +return file->has_ioc ?QIO_CHANNEL (file->opaque) :NULL ; +}
migration/qemu-file.h +3 -1
@@ -27,6 +27,7 @@#include <zlib.h> #include "exec/cpu-common.h" +#include "io/channel.h" /* Read a chunk of data from a file at the given position. The pos argument * can be ignored if the file is only be used for streaming. The number of
@@ -119,7 +120,7 @@ typedef struct QEMUFileHooks {QEMURamSaveFunc *save_page; } QEMUFileHooks; -QEMUFile *qemu_fopen_ops (void *opaque,const QEMUFileOps *ops); +QEMUFile *qemu_fopen_ops (void *opaque,const QEMUFileOps *ops,bool has_ioc);void qemu_file_set_hooks (QEMUFile *f,const QEMUFileHooks *hooks);int qemu_get_fd (QEMUFile *f);int qemu_fclose (QEMUFile *f);
@@ -179,5 +180,6 @@ void ram_control_load_hook(QEMUFile *f, uint64_t flags, void *data);size_t ram_control_save_page (QEMUFile *f,ram_addr_t block_offset,ram_addr_t offset,size_t size,uint64_t *bytes_sent); +QIOChannel *qemu_file_get_ioc (QEMUFile *file);#endif
migration/ram.c +1 -1
@@ -550,7 +550,7 @@ static int compress_threads_save_setup(void)/* comp_param[i].file is just used as a dummy buffer to save data, * set its ops to empty. */ - comp_param[i].file =qemu_fopen_ops (NULL , &empty_ops); + comp_param[i].file =qemu_fopen_ops (NULL , &empty_ops, false); comp_param[i].done = true; comp_param[i].quit = false;qemu_mutex_init (&comp_param[i].mutex);
migration/savevm.c +2 -2
@@ -168,9 +168,9 @@ static const QEMUFileOps bdrv_write_ops = {static QEMUFile *qemu_fopen_bdrv (BlockDriverState *bs,int is_writable) {if (is_writable) { -return qemu_fopen_ops (bs, &bdrv_write_ops); +return qemu_fopen_ops (bs, &bdrv_write_ops, false); } -return qemu_fopen_ops (bs, &bdrv_read_ops); +return qemu_fopen_ops (bs, &bdrv_read_ops, false); }
Signature · 1 line
2.31.1
PATCH v3 5/5 migration: Move the yank unregister of channel_close out
In reply to Peter XuIt's efficient, but hackish to call yank unregister calls in channel_close(), especially it'll be hard to debug when qemu crashed with some yank function leaked.
Remove that hack, but instead explicitly unregister yank functions at the places where needed, they are:
(on src)
- migrate_fd_cleanup
- postcopy_pause
(on dst)
- migration_incoming_state_destroy
- postcopy_pause_incoming
--- migration/migration.c | 14 +++++++++++++- migration/qemu-file-channel.c | 3 --- migration/savevm.c | 7 +++++++ migration/yank_functions.c | 14 ++++++++++++++ migration/yank_functions.h | 1 + 5 files changed, 35 insertions(+), 4 deletions(-)
migration/migration.c +13 -1
@@ -59,6 +59,7 @@#include "multifd.h" #include "qemu/yank.h" #include "sysemu/cpus.h" +#include "yank_functions.h" #define MAX_THROTTLE (128 << 20)/* Migration transfer speed throttling */
@@ -273,6 +274,7 @@ void migration_incoming_state_destroy(void) }if (mis->from_src_file) { +migration_ioc_unregister_yank_from_file (mis->from_src_file);qemu_fclose (mis->from_src_file); mis->from_src_file =NULL ; }
@@ -1811,6 +1813,7 @@ static void migrate_fd_cleanup(MigrationState *s)* Close the file handle without the lock to make sure the * critical section won't block for long. */ +migration_ioc_unregister_yank_from_file (tmp);qemu_fclose (tmp); }
@@ -3351,8 +3354,17 @@ static MigThrError postcopy_pause(MigrationState *s)while (true) {QEMUFile *file; -/* Current channel is possibly broken. Release it. */ +/* +* Current channel is possibly broken. Release it. Note that this is +* guaranteed even without lock because to_dst_file should only be +* modified by the migration thread. That also guarantees that the +* unregister of yank is safe too without the lock. It should be safe +* even to be within the qemu_file_lock, but we didn't do that to avoid +* taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make +* the qemu_file_lock critical section as small as possible. +*/ assert (s->to_dst_file); +migration_ioc_unregister_yank_from_file (s->to_dst_file);qemu_mutex_lock (&s->qemu_file_lock); file = s->to_dst_file; s->to_dst_file =NULL ;
migration/qemu-file-channel.c +0 -3
@@ -107,9 +107,6 @@ static int channel_close(void *opaque, Error **errp)int ret;QIOChannel *ioc =QIO_CHANNEL (opaque); ret =qio_channel_close (ioc, errp); -if (OBJECT (ioc)->ref ==1 ) { -migration_ioc_unregister_yank (ioc); - }object_unref (OBJECT (ioc));return ret; }
migration/savevm.c +7 -0
@@ -65,6 +65,7 @@#include "qemu/bitmap.h" #include "net/announce.h" #include "qemu/yank.h" +#include "yank_functions.h" const unsigned int postcopy_ram_discard_version;
@@ -2568,6 +2569,12 @@ static bool postcopy_pause_incoming(MigrationIncomingState *mis)/* Clear the triggered bit to allow one recovery */ mis->postcopy_recover_triggered = false; +/* +* Unregister yank with either from/to src would work, since ioc behind it +* is the same +*/ +migration_ioc_unregister_yank_from_file (mis->from_src_file); +assert (mis->from_src_file);qemu_file_shutdown (mis->from_src_file);qemu_fclose (mis->from_src_file);
migration/yank_functions.c +14 -0
@@ -14,6 +14,7 @@#include "qemu/yank.h" #include "io/channel-socket.h" #include "io/channel-tls.h" +#include "qemu-file.h" void migration_yank_iochannel (void *opaque) {
@@ -46,3 +47,16 @@ void migration_ioc_unregister_yank(QIOChannel *ioc) QIO_CHANNEL(ioc)); } } + +void migration_ioc_unregister_yank_from_file (QEMUFile *file) +{ +QIOChannel *ioc =qemu_file_get_ioc (file); + +if (ioc) { +/* +* For migration qemufiles, we'll always reach here. Though we'll skip +* calls from e.g. savevm/loadvm as they don't use yank. +*/ +migration_ioc_unregister_yank (ioc); + } +}
migration/yank_functions.h +1 -0
@@ -17,3 +17,4 @@void migration_yank_iochannel (void *opaque);void migration_ioc_register_yank (QIOChannel *ioc);void migration_ioc_unregister_yank (QIOChannel *ioc); +void migration_ioc_unregister_yank_from_file (QEMUFile *file);
Signature · 1 line
2.31.1
Re: [PATCH v3 5/5] migration: Move the yank unregister of channel_close out
Quoted 11 lines · * Peter Xu ([email protected]) wrote:
It's efficient, but hackish to call yank unregister calls in
channel_close(), especially it'll be hard to debug when qemu crashed with some yank function leaked.Remove that hack, but instead explicitly unregister yank functions at the places where needed, they are:
(on src)
-migrate_fd_cleanup
-postcopy_pause
Quoted 5 lines
(on dst)
-migration_incoming_state_destroy
-postcopy_pause_incomingSigned-off-by Peter Xu <[email protected]>
Quoted 135 lines
--- migration/migration.c | 14 +++++++++++++- migration/qemu-file-channel.c | 3 --- migration/savevm.c | 7 +++++++ migration/yank_functions.c | 14 ++++++++++++++ migration/yank_functions.h | 1 + 5 files changed, 35 insertions(+), 4 deletions(-)migration/migration.c +13
-1@@ -59,6 +59,7 @@#include "multifd.h" #include "qemu/yank.h" #include "sysemu/cpus.h" +#include "yank_functions.h" #define MAX_THROTTLE (128 << 20)/* Migration transfer speed throttling */ @@ -273,6 +274,7 @@ void migration_incoming_state_destroy(void) }if (mis->from_src_file) { +migration_ioc_unregister_yank_from_file (mis->from_src_file);qemu_fclose (mis->from_src_file); mis->from_src_file =NULL ; }@@ -1811,6 +1813,7 @@ static void migrate_fd_cleanup(MigrationState *s)* Close the file handle without the lock to make sure the * critical section won't block for long. */ +migration_ioc_unregister_yank_from_file (tmp);qemu_fclose (tmp); }@@ -3351,8 +3354,17 @@ static MigThrError postcopy_pause(MigrationState *s)while (true) {QEMUFile *file; -/* Current channel is possibly broken. Release it. */ +/* +* Current channel is possibly broken. Release it. Note that this is +* guaranteed even without lock because to_dst_file should only be +* modified by the migration thread. That also guarantees that the +* unregister of yank is safe too without the lock. It should be safe +* even to be within the qemu_file_lock, but we didn't do that to avoid +* taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make +* the qemu_file_lock critical section as small as possible. +*/ assert (s->to_dst_file); +migration_ioc_unregister_yank_from_file (s->to_dst_file);qemu_mutex_lock (&s->qemu_file_lock); file = s->to_dst_file; s->to_dst_file =NULL ;migration/qemu-file-channel.c +0
-3@@ -107,9 +107,6 @@ static int channel_close(void *opaque, Error **errp)int ret;QIOChannel *ioc =QIO_CHANNEL (opaque); ret =qio_channel_close (ioc, errp); -if (OBJECT (ioc)->ref ==1 ) { -migration_ioc_unregister_yank (ioc); - }object_unref (OBJECT (ioc));return ret; }migration/savevm.c +7
-0@@ -65,6 +65,7 @@#include "qemu/bitmap.h" #include "net/announce.h" #include "qemu/yank.h" +#include "yank_functions.h" const unsigned int postcopy_ram_discard_version;@@ -2568,6 +2569,12 @@ static bool postcopy_pause_incoming(MigrationIncomingState *mis)/* Clear the triggered bit to allow one recovery */ mis->postcopy_recover_triggered = false; +/* +* Unregister yank with either from/to src would work, since ioc behind it +* is the same +*/ +migration_ioc_unregister_yank_from_file (mis->from_src_file); +assert (mis->from_src_file);qemu_file_shutdown (mis->from_src_file);qemu_fclose (mis->from_src_file);migration/yank_functions.c +14
-0@@ -14,6 +14,7 @@#include "qemu/yank.h" #include "io/channel-socket.h" #include "io/channel-tls.h" +#include "qemu-file.h" void migration_yank_iochannel (void *opaque) {@@ -46,3 +47,16 @@ void migration_ioc_unregister_yank(QIOChannel *ioc) QIO_CHANNEL(ioc)); } } + +void migration_ioc_unregister_yank_from_file (QEMUFile *file) +{ +QIOChannel *ioc =qemu_file_get_ioc (file); + +if (ioc) { +/* +* For migration qemufiles, we'll always reach here. Though we'll skip +* calls from e.g. savevm/loadvm as they don't use yank. +*/ +migration_ioc_unregister_yank (ioc); + } +}migration/yank_functions.h +1
-0@@ -17,3 +17,4 @@void migration_yank_iochannel (void *opaque);void migration_ioc_register_yank (QIOChannel *ioc);void migration_ioc_unregister_yank (QIOChannel *ioc); +void migration_ioc_unregister_yank_from_file (QEMUFile *file);Signature · 1 line
2.31.1
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Quoted 51 lines · On Thu, 22 Jul 2021 13:58:36 -0400 Peter Xu <[email protected]> wrote:
v3:
- UseWITH_QEMU_LOCK_GUARD()for patch 2 [Eric]
(potentially I can also replace other existing uses ofqemu_file_lockinto
WITH_QEMU_LOCK_GUARD, but I decided to took Dave's r-b first and leave that for later)
- Added r-bs for Dave on patch 2/4
- Add a comment in patch 5 to explain why it's safe to unregister yank withoutqemu_file_lock, inpostcopy_pause()[Dave]v2:
- Pick r-b for Dave on patch 1/3
- Movemigration_file_get_ioc()from patch 5 to patch 4, meanwhile rename it toqemu_file_get_ioc(). [Dave]
- Patch 2 "migration: Shutdown src inawait_return_path_close_on_source()" is replaced by patch "migration: Makefrom_dst_fileaccesses thread-safe" [Dave]Patch 1 fixes a possible race that migration thread can accidentally skip
join()ofrp_threadeven if the return thread is enabled. Patch 1 is suspected to also be the root cause of the recent hard-to-reproduce migration-test failure here reported by PMM:https://lore.kernel.org/qemu-devel/YPamXAHwan%2FPPXLf@work-vm/
I didn't reproduce it myself; but after co-debugged with Dave it's suspected that the race of
rp_threadcould be the cause. It's not exposed before because yank is soo strict on releasing instances, while we're not that strict before, and didn'tjoin()onrp_threadwasn't so dangerous after all when migration succeeded before.Patch 2 fixes another theoretical race on accessing
from_dst_filespotted by Dave. I don't think there's known issues with it, but may still worth fixing.Patch 3 should be a cleanup on yank that I think would be nice to have.
Patch 4-5 are further cleanups to remove the ref==1 check in
channel_close(), finally, as I always thought that's a bit hackish. So I used explicit unregister of the yank function at necessary places to replace that ref==1 one.I still think having patch 3-5 altogether would be great, however I think patch 1 should still be the most important to be reviewed. Also it would be great to know whether patch 1 could fix the known yank crash.
Please review, thanks.
Peter Xu (5):
migration: Fix missingjoin()ofrp_thread
migration: Makefrom_dst_fileaccesses thread-safe
migration: Introduce migration_ioc_[un]register_yank()
migration: Teach QEMUFile to be QIOChannel-aware
migration: Move the yank unregister ofchannel_closeout
Quoted 13 lines
migration/channel.c | 15 ++------- migration/migration.c | 57 +++++++++++++++++++++++++++-------- migration/migration.h | 15 +++++++-- migration/multifd.c | 8 ++--- migration/qemu-file-channel.c | 11 ++----- migration/qemu-file.c | 17 ++++++++++- migration/qemu-file.h | 4 ++- migration/ram.c | 3 +- migration/savevm.c | 11 +++++-- migration/yank_functions.c | 42 ++++++++++++++++++++++++++ migration/yank_functions.h | 3 ++ 11 files changed, 138 insertions(+), 48 deletions(-)
Looks good to me,
Quoted 38 lines · * Peter Xu ([email protected]) wrote:
v3:
- UseWITH_QEMU_LOCK_GUARD()for patch 2 [Eric]
(potentially I can also replace other existing uses ofqemu_file_lockinto
WITH_QEMU_LOCK_GUARD, but I decided to took Dave's r-b first and leave that for later)
- Added r-bs for Dave on patch 2/4
- Add a comment in patch 5 to explain why it's safe to unregister yank withoutqemu_file_lock, inpostcopy_pause()[Dave]v2:
- Pick r-b for Dave on patch 1/3
- Movemigration_file_get_ioc()from patch 5 to patch 4, meanwhile rename it toqemu_file_get_ioc(). [Dave]
- Patch 2 "migration: Shutdown src inawait_return_path_close_on_source()" is replaced by patch "migration: Makefrom_dst_fileaccesses thread-safe" [Dave]Patch 1 fixes a possible race that migration thread can accidentally skip
join()ofrp_threadeven if the return thread is enabled. Patch 1 is suspected to also be the root cause of the recent hard-to-reproduce migration-test failure here reported by PMM:https://lore.kernel.org/qemu-devel/YPamXAHwan%2FPPXLf@work-vm/
I didn't reproduce it myself; but after co-debugged with Dave it's suspected that the race of
rp_threadcould be the cause. It's not exposed before because yank is soo strict on releasing instances, while we're not that strict before, and didn'tjoin()onrp_threadwasn't so dangerous after all when migration succeeded before.Patch 2 fixes another theoretical race on accessing
from_dst_filespotted by Dave. I don't think there's known issues with it, but may still worth fixing.Patch 3 should be a cleanup on yank that I think would be nice to have.
Patch 4-5 are further cleanups to remove the ref==1 check in
channel_close(), finally, as I always thought that's a bit hackish. So I used explicit unregister of the yank function at necessary places to replace that ref==1 one.
Quoted 5 lines
I still think having patch 3-5 altogether would be great, however I think patch 1 should still be the most important to be reviewed. Also it would be great to know whether patch 1 could fix the known yank crash.
Please review, thanks.
Queued with the fix; it survived over a long weekend running about 50k times OK.
Dave
Quoted 25 lines
Peter Xu (5):
migration: Fix missingjoin()ofrp_thread
migration: Makefrom_dst_fileaccesses thread-safe
migration: Introduce migration_ioc_[un]register_yank()
migration: Teach QEMUFile to be QIOChannel-aware
migration: Move the yank unregister ofchannel_closeoutmigration/channel.c | 15 ++------- migration/migration.c | 57 +++++++++++++++++++++++++++-------- migration/migration.h | 15 +++++++-- migration/multifd.c | 8 ++--- migration/qemu-file-channel.c | 11 ++----- migration/qemu-file.c | 17 ++++++++++- migration/qemu-file.h | 4 ++- migration/ram.c | 3 +- migration/savevm.c | 11 +++++-- migration/yank_functions.c | 42 ++++++++++++++++++++++++++ migration/yank_functions.h | 3 ++ 11 files changed, 138 insertions(+), 48 deletions(-)Signature · 1 line
2.31.1
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK