virtiofsd, vhost-user-fs: Add support for notification queue
cover virtiofsd, vhost-user-fs: Add support for notification queue
Hi,
Here are RFC patches for adding a notification queue to virtio-fs to send notifications from host to guest. It also has patches to support remote posix locks which make use of this newly introduced notification queue.
These patches apply on top of following.
https://gitlab.com/virtio-fs/qemu/tree/virtio-fs-dev
These change require changes in spec also. I have yet to create spec changes and that's why this is still an RFC patch series.
Any feedback is appreciated.
Thanks
Vivek
Vivek Goyal (4):
virtiofsd: Release file locks using F_UNLCK
virtiofd: Create a notification queue
virtiofsd: Specify size of notification buffer using config space
virtiofsd: Implement blocking posix locks
contrib/virtiofsd/fuse_i.h | 1 + contrib/virtiofsd/fuse_kernel.h | 7 + contrib/virtiofsd/fuse_lowlevel.c | 23 ++- contrib/virtiofsd/fuse_lowlevel.h | 25 +++ contrib/virtiofsd/fuse_virtio.c | 226 +++++++++++++++++++-- contrib/virtiofsd/passthrough_ll.c | 79 +++++-- hw/virtio/vhost-user-fs-pci.c | 2 +- hw/virtio/vhost-user-fs.c | 63 +++++- include/hw/virtio/vhost-user-fs.h | 3 + include/standard-headers/linux/virtio_fs.h | 5 + 10 files changed, 386 insertions(+), 48 deletions(-)
Signature · 1 line
2.20.1
PATCH 1/4 virtiofsd: Release file locks using F_UNLCK
We are emulating posix locks for guest using open file description locks in virtiofsd. When any of the fd is closed in guest, we find associated OFD lock fd (if there is one) and close it to release all the locks.
Assumption here is that there is no other thread using lo_inode_plock structure or plock->fd, hence it is safe to do so.
But now we are about to introduce blocking variant of locks (SETLKW), and that means we might be waiting to a lock to be available and using plock->fd. And that means there are still users of plock structure.
So release locks using fcntl(SETLK, F_UNLCK) instead and plock will be freed later.
--- contrib/virtiofsd/passthrough_ll.c | 30 ++++++++++++++++++++---------- 1 file changed, 20 insertions(+), 10 deletions(-)
contrib/virtiofsd/passthrough_ll.c +20 -10
@@ -936,6 +936,14 @@ static void put_shared(struct lo_data *lo, struct lo_inode *inode) } } +static void release_plock (gpointer data) +{ +struct lo_inode_plock *plock = data; + +close (plock->fd); +free (plock); +} +/* Increments nlookup and caller must release refcount using * lo_inode_put(&parent). */
@@ -994,7 +1002,8 @@ static int lo_do_lookup(fuse_req_t req, fuse_ino_t parent, const char *name, inode->key.ino = e->attr.st_ino; inode->key.dev = e->attr.st_dev;pthread_mutex_init (&inode->plock_mutex,NULL ); - inode->posix_locks =g_hash_table_new (g_direct_hash, g_direct_equal); + inode->posix_locks =g_hash_table_new_full (g_direct_hash, + g_direct_equal,NULL , release_plock);get_shared (lo, inode);
@@ -1436,9 +1445,6 @@ static void unref_inode(struct lo_data *lo, struct lo_inode *inode, uint64_t n)if (!inode->nlookup) {lo_map_remove (&lo->ino_map, inode->fuse_ino);g_hash_table_remove (lo->inodes, &inode->key); -if (g_hash_table_size (inode->posix_locks)) { -fuse_log (FUSE_LOG_WARNING ,"Hash table is not empty\n" ); - }g_hash_table_destroy (inode->posix_locks);pthread_mutex_destroy (&inode->plock_mutex);
@@ -1868,6 +1874,7 @@ static struct lo_inode_plock *lookup_create_plock_ctx(struct lo_data *lo, plock->fd = fd;g_hash_table_insert (inode->posix_locks,GUINT_TO_POINTER (plock->lock_owner), plock); +fuse_log (FUSE_LOG_DEBUG ,"lookup_create_plock_ctx(): Inserted element in posix_locks hash table with value pointer %p\n" , plock);return plock; }
@@ -2046,6 +2053,7 @@ static void lo_flush(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi) (void ) ino;struct lo_inode *inode;struct lo_inode_plock *plock; +struct flock flock; inode =lo_inode (req, ino);if (!inode) {
@@ -2058,14 +2066,16 @@ static void lo_flush(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi) plock =g_hash_table_lookup (inode->posix_locks,GUINT_TO_POINTER (fi->lock_owner));if (plock) { -g_hash_table_remove (inode->posix_locks, -GUINT_TO_POINTER (fi->lock_owner));/* -* We had used open() for locks and had only one fd. So -* closing this fd should release all OFD locks. +* An fd is being closed. For posix locks, this means +* drop all the associated locks. */ -close (plock->fd); -free (plock); +memset (&flock,0 ,sizeof (struct flock )); + flock.l_type =F_UNLCK ; + flock.l_whence =SEEK_SET ; +/* Unlock whole file */ + flock.l_start = flock.l_len =0 ; +fcntl (plock->fd,F_SETLK , &flock); }pthread_mutex_unlock (&inode->plock_mutex);
Signature · 1 line
2.20.1
Re: [PATCH 1/4] virtiofsd: Release file locks using F_UNLCK
contrib/virtiofsd/passthrough_ll.c static void put_shared(struct lo_data *lo, struct lo_inode *inode) · line 939
Quoted 9 lines · Vivek Goyal
contrib/virtiofsd/passthrough_ll.c +1
-0@@ -936,6 +936,14 @@ static void put_shared(struct lo_data *lo, struct lo_inode *inode) } } +static void release_plock (gpointer data)
The name posix_locks_value_destroy() would be clearer because it matches g_hash_table_new_full() terminology and the function cannot be confused with a lock acquire/release operation.
This patch conflicts with the cleanups that are currently being made to virtiofsd:
https://github.com/stefanha/qemu/commit/1e493175feca58a81a2d0cbdac93b92e5425d850#diff-ca2dea995d1e6cdb95c8a47c7cca51ceR773
Stefan
Re: [PATCH 1/4] virtiofsd: Release file locks using F_UNLCK
Quoted 9 lines · Stefan Hajnoczi
Quoted 8 lines · Vivek Goyal
contrib/virtiofsd/passthrough_ll.c +0
-0@@ -936,6 +936,14 @@ static void put_shared(struct lo_data *lo, struct lo_inode *inode) } }
Quoted 5 lines
contrib/virtiofsd/passthrough_ll.c static void put_shared(struct lo_data *lo, struct lo_inode *inode) · line 939
Quoted 1 line
contrib/virtiofsd/passthrough_ll.c +1
-0+static void release_plock (gpointer data)The name
posix_locks_value_destroy()would be clearer because it matchesg_hash_table_new_full()terminology and the function cannot be confused with a lock acquire/release operation.
Ok, will use this name.
Quoted 4 lines
This patch conflicts with the cleanups that are currently being made to virtiofsd:
https://github.com/stefanha/qemu/commit/1e493175feca58a81a2d0cbdac93b92e5425d850#diff-ca2dea995d1e6cdb95c8a47c7cca51ceR773
Yes it will. I see you are removing element from hash table on lo_flush(). This works fine today but with waiting locks, we drop the inode->plock_mutex lock and then wait for the lock and expect "lo_inode_plock" to not go away.
So I don't think you can remove the element from hash table upon lo_flush(). May be we can refcount lo_inode_plock structure and first release all the locks using setlk(UNLCK) and then drop the reference. If this is last refernce, it will be freed.
And waiting lock code, will obtain a refernce under inode->posix_locks and then wait for lock outside the lock.
IOW, I will say don't do this optimization of lookup + remove because it will not work with blocking locks.
Thanks
Vivek
PATCH 2/4 virtiofd: Create a notification queue
In reply to Vivek GoyalAdd a notification queue which will be used to send async notifications for file lock availability.
--- contrib/virtiofsd/fuse_i.h | 1 + contrib/virtiofsd/fuse_virtio.c | 108 ++++++++++++++++++--- hw/virtio/vhost-user-fs-pci.c | 2 +- hw/virtio/vhost-user-fs.c | 37 +++++-- include/hw/virtio/vhost-user-fs.h | 1 + include/standard-headers/linux/virtio_fs.h | 3 + 6 files changed, 130 insertions(+), 22 deletions(-)
contrib/virtiofsd/fuse_i.h +1 -0
@@ -74,6 +74,7 @@ struct fuse_session {char *vu_socket_lock;struct fv_VuDev *virtio_dev;int thread_pool_size; +bool notify_enabled; };struct fuse_chan {
contrib/virtiofsd/fuse_virtio.c +94 -14
161 lines not on this page · raw message
hw/virtio/vhost-user-fs-pci.c +1 -1
@@ -44,7 +44,7 @@ static void vhost_user_fs_pci_realize(VirtIOPCIProxy *vpci_dev, Error **errp)uint64_t totalsize;if (vpci_dev->nvectors ==DEV_NVECTORS_UNSPECIFIED ) { - vpci_dev->nvectors = dev->vdev.conf.num_request_queues +1 ; + vpci_dev->nvectors = dev->vdev.conf.num_request_queues +2 ; }qdev_set_parent_bus (vdev,BUS (&vpci_dev->bus));
hw/virtio/vhost-user-fs.c +30 -7
@@ -24,6 +24,10 @@#include "exec/address-spaces.h" #include "trace.h" +static const int user_feature_bits[] = { +VIRTIO_FS_F_NOTIFICATION , +}; +uint64_t vhost_user_fs_slave_map (struct vhost_dev *dev,VhostUserFSSlaveMsg *sm,int fd) {
@@ -378,12 +382,23 @@ static void vuf_set_status(VirtIODevice *vdev, uint8_t status) } } -static uint64_t vuf_get_features (VirtIODevice *vdev, -uint64_t requested_features, -Error **errp) +static uint64_t vuf_get_features (VirtIODevice *vdev,uint64_t features, +Error **errp) { -/* No feature bits used yet */ -return requested_features; +VHostUserFS *fs =VHOST_USER_FS (vdev); + +virtio_add_feature (&features,VIRTIO_FS_F_NOTIFICATION ); + +return vhost_get_features (&fs->vhost_dev, user_feature_bits, features); +} + +static void vuf_set_features (VirtIODevice *vdev,uint64_t features) +{ +VHostUserFS *fs =VHOST_USER_FS (vdev); + +if (virtio_has_feature (features,VIRTIO_FS_F_NOTIFICATION )) { + fs->notify_enabled = true; + } }static void vuf_handle_output (VirtIODevice *vdev,VirtQueue *vq)
@@ -515,13 +530,20 @@ static void vuf_device_realize(DeviceState *dev, Error **errp)/* Hiprio queue */ virtio_add_queue (vdev, fs->conf.queue_size, vuf_handle_output); +/* Notification queue. Feature negotiation happens later. So at this +* point of time we don't know if driver will use notification queue +* or not. +*/ +virtio_add_queue (vdev, fs->conf.queue_size, vuf_handle_output); +/* Request queues */ for (i =0 ; i < fs->conf.num_request_queues; i++) {virtio_add_queue (vdev, fs->conf.queue_size, vuf_handle_output); } -/* 1 high prio queue, plus the number configured */ - fs->vhost_dev.nvqs =1 + fs->conf.num_request_queues; +/* 1 high prio queue, 1 notification queue plus the number configured */ + fs->vhost_dev.nvqs =2 + fs->conf.num_request_queues; + fs->vhost_dev.vqs = g_new0(struct vhost_virtqueue, fs->vhost_dev.nvqs); ret =vhost_dev_init (&fs->vhost_dev, &fs->vhost_user,VHOST_BACKEND_TYPE_USER ,0 );
@@ -584,6 +606,7 @@ static void vuf_class_init(ObjectClass *klass, void *data) vdc->realize = vuf_device_realize; vdc->unrealize = vuf_device_unrealize; vdc->get_features = vuf_get_features; + vdc->set_features = vuf_set_features; vdc->get_config = vuf_get_config; vdc->set_status = vuf_set_status; vdc->guest_notifier_mask = vuf_guest_notifier_mask;
include/hw/virtio/vhost-user-fs.h +1 -0
@@ -64,6 +64,7 @@ typedef struct {/* Metadata version table */ size_t mdvt_size;MemoryRegion mdvt; +bool notify_enabled; } VHostUserFS;/* Callbacks from the vhost-user code for slave commands */
include/standard-headers/linux/virtio_fs.h +3 -0
@@ -8,6 +8,9 @@#include "standard-headers/linux/virtio_config.h" #include "standard-headers/linux/virtio_types.h" +/* Feature bits */ +#define VIRTIO_FS_F_NOTIFICATION 0/* Notification queue supported */ +struct virtio_fs_config {/* Filesystem name (UTF-8, not NUL-terminated, padded with NULs) */ uint8_t tag[36 ];
Signature · 1 line
2.20.1
Re: [PATCH 2/4] virtiofd: Create a notification queue
contrib/virtiofsd/fuse_virtio.c struct fv_VuDev { · line 123
Quoted 7 lines · Vivek Goyal
/* Callback from libvhost-user */
static voidfv_set_features(VuDev *dev, uint64_t features){contrib/virtiofsd/fuse_virtio.c +4
-0+ struct fv_VuDev *vud = container_of(dev, struct fv_VuDev, dev); +struct fuse_session *se = vud->se; + +if ((1 <<VIRTIO_FS_F_NOTIFICATION ) & features) {
For consistency 1ull should be used. That way the reader does not have to check the bit position to verify that the bitmap isn't truncated at 32 bits.
contrib/virtiofsd/fuse_virtio.c struct fv_VuDev { · line 125
Quoted 2 lines
contrib/virtiofsd/fuse_virtio.c +2
-0+ vud->notify_enabled = true; + se->notify_enabled = true;
Only one copy of this field is needed. vud has a pointer to se.
Quoted 56 lines
contrib/virtiofsd/fuse_virtio.c +49
-0+ } }/* @@ -662,6 +671,65 @@ static void fv_queue_worker(gpointer data, gpointer user_data)free (req); } +static void *fv_queue_notify_thread (void *opaque) +{ +struct fv_QueueInfo *qi = opaque; + +fuse_log (FUSE_LOG_INFO ,"%s: Start for queue %d kick_fd %d\n" , __func__, + qi->qidx, qi->kick_fd); + +while (1 ) { +struct pollfd pf[2 ]; + + pf[0 ].fd = qi->kick_fd; + pf[0 ].events =POLLIN ; + pf[0 ].revents =0 ; + pf[1 ].fd = qi->kill_fd; + pf[1 ].events =POLLIN ; + pf[1 ].revents =0 ; + +fuse_log (FUSE_LOG_DEBUG ,"%s: Waiting for Queue %d event\n" , __func__, + qi->qidx); +int poll_res =ppoll (pf,2 ,NULL ,NULL ); + +if (poll_res ==-1 ) { +if (errno ==EINTR ) { +fuse_log (FUSE_LOG_INFO ,"%s: ppoll interrupted, going around\n" , + __func__); +continue ; + } +fuse_log (FUSE_LOG_ERR ,"fv_queue_thread ppoll: %m\n" ); +break ; + } +assert (poll_res >=1 ); +if (pf[0 ].revents & (POLLERR |POLLHUP |POLLNVAL )) { +fuse_log (FUSE_LOG_ERR ,"%s: Unexpected poll revents %x Queue %d\n" , + __func__, pf[0 ].revents, qi->qidx); +break ; + } +if (pf[1 ].revents & (POLLERR |POLLHUP |POLLNVAL )) { +fuse_log (FUSE_LOG_ERR ,"%s: Unexpected poll revents %x Queue %d" +"killfd\n" , __func__, pf[1 ].revents, qi->qidx); +break ; + } +if (pf[1 ].revents) { +fuse_log (FUSE_LOG_INFO ,"%s: kill event on queue %d - quitting\n" , + __func__, qi->qidx); +break ; + } +assert (pf[0 ].revents &POLLIN ); + fuse_log(FUSE_LOG_DEBUG, "%s: Got queue event on Queue %d\n", __func__,
contrib/virtiofsd/fuse_virtio.c static void fv_queue_worker(gpointer data, gpointer user_data) · line 731
Quoted 10 lines
contrib/virtiofsd/fuse_virtio.c +10
-0+ qi->qidx); + +eventfd_t evalue; +if (eventfd_read (qi->kick_fd, &evalue)) { +fuse_log (FUSE_LOG_ERR ,"Eventfd_read for queue: %m\n" ); +break ; + } + } +return NULL ; +}
It's difficult to review function without any actual functionality using the virtqueue. I'm not sure a thread is even needed since the device only needs to get a buffer when it has a notification for the driver. I'll have to wait for the following patches to see what happens here...
Quoted 15 lines
hw/virtio/vhost-user-fs.c +5
-5@@ -378,12 +382,23 @@ static void vuf_set_status(VirtIODevice *vdev, uint8_t status) } } -static uint64_t vuf_get_features (VirtIODevice *vdev, -uint64_t requested_features, -Error **errp) +static uint64_t vuf_get_features (VirtIODevice *vdev,uint64_t features, +Error **errp) { -/* No feature bits used yet */ -return requested_features; +VHostUserFS *fs =VHOST_USER_FS (vdev); + +virtio_add_feature (&features,VIRTIO_FS_F_NOTIFICATION );
hw/virtio/vhost-user-fs.c static void vuf_set_status(VirtIODevice *vdev, uint8_t status) · line 400
Quoted 10 lines
hw/virtio/vhost-user-fs.c +10
-0+ +return vhost_get_features (&fs->vhost_dev, user_feature_bits, features); +} + +static void vuf_set_features (VirtIODevice *vdev,uint64_t features) +{ +VHostUserFS *fs =VHOST_USER_FS (vdev); + +if (virtio_has_feature (features,VIRTIO_FS_F_NOTIFICATION )) { + fs->notify_enabled = true;
This field is unused, please remove it.
Re: [PATCH 2/4] virtiofd: Create a notification queue
Quoted 9 lines · Stefan Hajnoczi
contrib/virtiofsd/fuse_virtio.c struct fv_VuDev { · line 123
Quoted 7 lines · Vivek Goyal
/* Callback from libvhost-user */
static voidfv_set_features(VuDev *dev, uint64_t features){contrib/virtiofsd/fuse_virtio.c +4
-0+ struct fv_VuDev *vud = container_of(dev, struct fv_VuDev, dev); +struct fuse_session *se = vud->se; + +if ((1 <<VIRTIO_FS_F_NOTIFICATION ) & features) {
Quoted 3 lines
For consistency 1ull should be used. That way the reader does not have to check the bit position to verify that the bitmap isn't truncated at 32 bits.
Ok, will do.
Quoted 5 lines
contrib/virtiofsd/fuse_virtio.c struct fv_VuDev { · line 125
Quoted 3 lines
contrib/virtiofsd/fuse_virtio.c +2
-0+ vud->notify_enabled = true; + se->notify_enabled = true;Only one copy of this field is needed. vud has a pointer to se.
I need to access ->notify_enabled in passthrough_ll.c to determine if notification queue is enabled or not. That determines if async locks are supported or not. And based on that either -EOPNOTSUPP is returned or a response to wait is returned.
I did not see passthrough_ll.c accessing vud. I did see it having access to session object though. So I created a copy there.
But I am open to suggestions on what's the best way to access this information in passthrough_ll.c
Quoted 64 lines
Quoted 64 lines
contrib/virtiofsd/fuse_virtio.c +56
-0+ } }/* @@ -662,6 +671,65 @@ static void fv_queue_worker(gpointer data, gpointer user_data)free (req); } +static void *fv_queue_notify_thread (void *opaque) +{ +struct fv_QueueInfo *qi = opaque; + +fuse_log (FUSE_LOG_INFO ,"%s: Start for queue %d kick_fd %d\n" , __func__, + qi->qidx, qi->kick_fd); + +while (1 ) { +struct pollfd pf[2 ]; + + pf[0 ].fd = qi->kick_fd; + pf[0 ].events =POLLIN ; + pf[0 ].revents =0 ; + pf[1 ].fd = qi->kill_fd; + pf[1 ].events =POLLIN ; + pf[1 ].revents =0 ; + +fuse_log (FUSE_LOG_DEBUG ,"%s: Waiting for Queue %d event\n" , __func__, + qi->qidx); +int poll_res =ppoll (pf,2 ,NULL ,NULL ); + +if (poll_res ==-1 ) { +if (errno ==EINTR ) { +fuse_log (FUSE_LOG_INFO ,"%s: ppoll interrupted, going around\n" , + __func__); +continue ; + } +fuse_log (FUSE_LOG_ERR ,"fv_queue_thread ppoll: %m\n" ); +break ; + } +assert (poll_res >=1 ); +if (pf[0 ].revents & (POLLERR |POLLHUP |POLLNVAL )) { +fuse_log (FUSE_LOG_ERR ,"%s: Unexpected poll revents %x Queue %d\n" , + __func__, pf[0 ].revents, qi->qidx); +break ; + } +if (pf[1 ].revents & (POLLERR |POLLHUP |POLLNVAL )) { +fuse_log (FUSE_LOG_ERR ,"%s: Unexpected poll revents %x Queue %d" +"killfd\n" , __func__, pf[1 ].revents, qi->qidx); +break ; + } +if (pf[1 ].revents) { +fuse_log (FUSE_LOG_INFO ,"%s: kill event on queue %d - quitting\n" , + __func__, qi->qidx); +break ; + } +assert (pf[0 ].revents &POLLIN ); +fuse_log (FUSE_LOG_DEBUG ,"%s: Got queue event on Queue %d\n" , __func__, + qi->qidx); + +eventfd_t evalue; +if (eventfd_read (qi->kick_fd, &evalue)) { +fuse_log (FUSE_LOG_ERR ,"Eventfd_read for queue: %m\n" ); +break ; + }
Quoted 8 lines
contrib/virtiofsd/fuse_virtio.c static void fv_queue_worker(gpointer data, gpointer user_data) · line 731
Quoted 3 lines
contrib/virtiofsd/fuse_virtio.c +3
-0+ } +return NULL ; +}It's difficult to review function without any actual functionality using the virtqueue. I'm not sure a thread is even needed since the device only needs to get a buffer when it has a notification for the driver. I'll have to wait for the following patches to see what happens here...
This might very well be redundant. I am not sure. Can get rid of this thread if not needed at all. So we don't need to monitor even kill_fd and take any special action?
Quoted 20 lines
Quoted 20 lines
hw/virtio/vhost-user-fs.c +9
-5@@ -378,12 +382,23 @@ static void vuf_set_status(VirtIODevice *vdev, uint8_t status) } } -static uint64_t vuf_get_features (VirtIODevice *vdev, -uint64_t requested_features, -Error **errp) +static uint64_t vuf_get_features (VirtIODevice *vdev,uint64_t features, +Error **errp) { -/* No feature bits used yet */ -return requested_features; +VHostUserFS *fs =VHOST_USER_FS (vdev); + +virtio_add_feature (&features,VIRTIO_FS_F_NOTIFICATION ); + +return vhost_get_features (&fs->vhost_dev, user_feature_bits, features); +} +
Quoted 8 lines
hw/virtio/vhost-user-fs.c static void vuf_set_status(VirtIODevice *vdev, uint8_t status) · line 400
Quoted 6 lines
hw/virtio/vhost-user-fs.c +6
-0+static void vuf_set_features (VirtIODevice *vdev,uint64_t features) +{ +VHostUserFS *fs =VHOST_USER_FS (vdev); + +if (virtio_has_feature (features,VIRTIO_FS_F_NOTIFICATION )) { + fs->notify_enabled = true;This field is unused, please remove it.
vuf_get_config() uses it.
Thanks
Vivek
Re: [PATCH 2/4] virtiofd: Create a notification queue
Quoted 106 lines · * Vivek Goyal ([email protected]) wrote:
Quoted 9 lines · Stefan Hajnoczi
contrib/virtiofsd/fuse_virtio.c struct fv_VuDev { · line 123
Quoted 7 lines · Vivek Goyal
/* Callback from libvhost-user */
static voidfv_set_features(VuDev *dev, uint64_t features){contrib/virtiofsd/fuse_virtio.c +4
-0+ struct fv_VuDev *vud = container_of(dev, struct fv_VuDev, dev); +struct fuse_session *se = vud->se; + +if ((1 <<VIRTIO_FS_F_NOTIFICATION ) & features) {Quoted 3 lines
For consistency 1ull should be used. That way the reader does not have to check the bit position to verify that the bitmap isn't truncated at 32 bits.
Ok, will do.
Quoted 5 lines
contrib/virtiofsd/fuse_virtio.c struct fv_VuDev { · line 125
Quoted 3 lines
contrib/virtiofsd/fuse_virtio.c +2
-0+ vud->notify_enabled = true; + se->notify_enabled = true;Only one copy of this field is needed. vud has a pointer to se.
I need to access ->notify_enabled in
passthrough_ll.cto determine if notification queue is enabled or not. That determines if async locks are supported or not. And based on that either-EOPNOTSUPPis returned or a response to wait is returned.I did not see
passthrough_ll.caccessing vud. I did see it having access to session object though. So I created a copy there.But I am open to suggestions on what's the best way to access this information in
passthrough_ll.cQuoted 64 lines
Quoted 64 lines
contrib/virtiofsd/fuse_virtio.c +56
-0+ } }/* @@ -662,6 +671,65 @@ static void fv_queue_worker(gpointer data, gpointer user_data)free (req); } +static void *fv_queue_notify_thread (void *opaque) +{ +struct fv_QueueInfo *qi = opaque; + +fuse_log (FUSE_LOG_INFO ,"%s: Start for queue %d kick_fd %d\n" , __func__, + qi->qidx, qi->kick_fd); + +while (1 ) { +struct pollfd pf[2 ]; + + pf[0 ].fd = qi->kick_fd; + pf[0 ].events =POLLIN ; + pf[0 ].revents =0 ; + pf[1 ].fd = qi->kill_fd; + pf[1 ].events =POLLIN ; + pf[1 ].revents =0 ; + +fuse_log (FUSE_LOG_DEBUG ,"%s: Waiting for Queue %d event\n" , __func__, + qi->qidx); +int poll_res =ppoll (pf,2 ,NULL ,NULL ); + +if (poll_res ==-1 ) { +if (errno ==EINTR ) { +fuse_log (FUSE_LOG_INFO ,"%s: ppoll interrupted, going around\n" , + __func__); +continue ; + } +fuse_log (FUSE_LOG_ERR ,"fv_queue_thread ppoll: %m\n" ); +break ; + } +assert (poll_res >=1 ); +if (pf[0 ].revents & (POLLERR |POLLHUP |POLLNVAL )) { +fuse_log (FUSE_LOG_ERR ,"%s: Unexpected poll revents %x Queue %d\n" , + __func__, pf[0 ].revents, qi->qidx); +break ; + } +if (pf[1 ].revents & (POLLERR |POLLHUP |POLLNVAL )) { +fuse_log (FUSE_LOG_ERR ,"%s: Unexpected poll revents %x Queue %d" +"killfd\n" , __func__, pf[1 ].revents, qi->qidx); +break ; + } +if (pf[1 ].revents) { +fuse_log (FUSE_LOG_INFO ,"%s: kill event on queue %d - quitting\n" , + __func__, qi->qidx); +break ; + } +assert (pf[0 ].revents &POLLIN ); +fuse_log (FUSE_LOG_DEBUG ,"%s: Got queue event on Queue %d\n" , __func__, + qi->qidx); + +eventfd_t evalue; +if (eventfd_read (qi->kick_fd, &evalue)) { +fuse_log (FUSE_LOG_ERR ,"Eventfd_read for queue: %m\n" ); +break ; + }Quoted 8 lines
contrib/virtiofsd/fuse_virtio.c static void fv_queue_worker(gpointer data, gpointer user_data) · line 731
Quoted 3 lines
contrib/virtiofsd/fuse_virtio.c +3
-0+ } +return NULL ; +}It's difficult to review function without any actual functionality using the virtqueue. I'm not sure a thread is even needed since the device only needs to get a buffer when it has a notification for the driver. I'll have to wait for the following patches to see what happens here...
Quoted 3 lines
This might very well be redundant. I am not sure. Can get rid of this thread if not needed at all. So we don't need to monitor even
kill_fdand take any special action?
The kill_fd is internal to virtiofsd; it's only used as a way for the main thread to cause the queue thread to exit; if you've not got the thread, you don't need the kill_fd.
Dave
Quoted 29 lines
Quoted 20 lines
Quoted 20 lines
hw/virtio/vhost-user-fs.c +9
-5@@ -378,12 +382,23 @@ static void vuf_set_status(VirtIODevice *vdev, uint8_t status) } } -static uint64_t vuf_get_features (VirtIODevice *vdev, -uint64_t requested_features, -Error **errp) +static uint64_t vuf_get_features (VirtIODevice *vdev,uint64_t features, +Error **errp) { -/* No feature bits used yet */ -return requested_features; +VHostUserFS *fs =VHOST_USER_FS (vdev); + +virtio_add_feature (&features,VIRTIO_FS_F_NOTIFICATION ); + +return vhost_get_features (&fs->vhost_dev, user_feature_bits, features); +} +Quoted 8 lines
hw/virtio/vhost-user-fs.c static void vuf_set_status(VirtIODevice *vdev, uint8_t status) · line 400
Quoted 6 lines
hw/virtio/vhost-user-fs.c +6
-0+static void vuf_set_features (VirtIODevice *vdev,uint64_t features) +{ +VHostUserFS *fs =VHOST_USER_FS (vdev); + +if (virtio_has_feature (features,VIRTIO_FS_F_NOTIFICATION )) { + fs->notify_enabled = true;This field is unused, please remove it.
Quoted 4 lines
vuf_get_config()uses it.Thanks
Vivek
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK
PATCH 3/4 virtiofsd: Specify size of notification buffer using config space
In reply to Vivek GoyalDaemon specifies size of notification buffer needed and that should be done using config space.
Only ->notify_buf_size value of config space comes from daemon. Rest of it is filled by qemu device emulation code.
--- contrib/virtiofsd/fuse_virtio.c | 26 +++++++++++++++++++++- hw/virtio/vhost-user-fs.c | 26 ++++++++++++++++++++++ include/hw/virtio/vhost-user-fs.h | 2 ++ include/standard-headers/linux/virtio_fs.h | 2 ++ 4 files changed, 55 insertions(+), 1 deletion(-)
contrib/virtiofsd/fuse_virtio.c +25 -1
@@ -109,7 +109,8 @@ static uint64_t fv_get_features(VuDev *dev)uint64_t features; features =1ull <<VIRTIO_F_VERSION_1 | -1ull <<VIRTIO_FS_F_NOTIFICATION ; +1ull <<VIRTIO_FS_F_NOTIFICATION | +1ull <<VHOST_USER_F_PROTOCOL_FEATURES ;return features; }
@@ -927,6 +928,27 @@ static bool fv_queue_order(VuDev *dev, int qidx)return false; } +static uint64_t fv_get_protocol_features (VuDev *dev) +{ +return 1ull <<VHOST_USER_PROTOCOL_F_CONFIG ; +} + +static int fv_get_config (VuDev *dev,uint8_t *config,uint32_t len) +{ +struct virtio_fs_config fscfg = {}; + +fuse_log (FUSE_LOG_DEBUG ,"%s:Setting notify_buf_size=%d\n" , __func__, +sizeof (struct fuse_notify_lock_out )); +/* +* As of now only notification related to lock is supported. As more +* notification types are supported, bump up the size accordingly. +*/ + fscfg.notify_buf_size =sizeof (struct fuse_notify_lock_out ); + +memcpy (config, &fscfg, len); +return 0 ; +} +static const VuDevIface fv_iface = { .get_features = fv_get_features, .set_features = fv_set_features,
@@ -935,6 +957,8 @@ static const VuDevIface fv_iface = { .queue_set_started = fv_queue_set_started, .queue_is_processed_in_order = fv_queue_order, + .get_protocol_features = fv_get_protocol_features, + .get_config = fv_get_config, };/*
hw/virtio/vhost-user-fs.c +26 -0
@@ -277,16 +277,40 @@ uint64_t vhost_user_fs_slave_io(struct vhost_dev *dev, VhostUserFSSlaveMsg *sm,return (uint64_t )done; } +static int vhost_user_fs_handle_config_change (struct vhost_dev *dev) +{ +return 0 ; +} + +const VhostDevConfigOps fs_ops = { + .vhost_dev_config_notifier = vhost_user_fs_handle_config_change, +};static void vuf_get_config (VirtIODevice *vdev,uint8_t *config) {VHostUserFS *fs =VHOST_USER_FS (vdev);struct virtio_fs_config fscfg = {}; +int ret; + +/* +* As of now we only get notification buffer size from device. And that's +* needed only if notification queue is enabled. +*/ +if (fs->notify_enabled) { + ret =vhost_dev_get_config (&fs->vhost_dev, (uint8_t *)&fs->fscfg, +sizeof (struct virtio_fs_config )); +if (ret <0 ) { +error_report ("vhost-user-fs: get device config space failed." +" ret=%d\n" , ret); +return ; + } + }memcpy ((char *)fscfg.tag, fs->conf.tag,MIN (strlen (fs->conf.tag) +1 ,sizeof (fscfg.tag)));virtio_stl_p (vdev, &fscfg.num_request_queues, fs->conf.num_request_queues); +virtio_stl_p (vdev, &fscfg.notify_buf_size, fs->fscfg.notify_buf_size);memcpy (config, &fscfg,sizeof (fscfg)); }
@@ -545,6 +569,8 @@ static void vuf_device_realize(DeviceState *dev, Error **errp) fs->vhost_dev.nvqs =2 + fs->conf.num_request_queues; fs->vhost_dev.vqs = g_new0(struct vhost_virtqueue, fs->vhost_dev.nvqs); + +vhost_dev_set_config_notifier (&fs->vhost_dev, &fs_ops); ret =vhost_dev_init (&fs->vhost_dev, &fs->vhost_user,VHOST_BACKEND_TYPE_USER ,0 );if (ret <0 ) {
include/hw/virtio/vhost-user-fs.h +2 -0
@@ -14,6 +14,7 @@#ifndef _QEMU_VHOST_USER_FS_H#define _QEMU_VHOST_USER_FS_H +#include "standard-headers/linux/virtio_fs.h" #include "hw/virtio/virtio.h" #include "hw/virtio/vhost.h" #include "hw/virtio/vhost-user.h"
@@ -58,6 +59,7 @@ typedef struct {struct vhost_virtqueue *vhost_vqs;struct vhost_dev vhost_dev;VhostUserState vhost_user; +struct virtio_fs_config fscfg;/*< public >*/ MemoryRegion cache;
include/standard-headers/linux/virtio_fs.h +2 -0
@@ -17,6 +17,8 @@ struct virtio_fs_config {/* Number of request queues */ uint32_t num_request_queues; +/* Size of notification buffer */ +uint32_t notify_buf_size; }QEMU_PACKED ;#define VIRTIO_FS_PCI_CACHE_BAR 2
Signature · 1 line
2.20.1
Re: [PATCH 3/4] virtiofsd: Specify size of notification buffer using config space
contrib/virtiofsd/fuse_virtio.c static uint64_t fv_get_features(VuDev *dev) · line 113
Quoted 11 lines · Vivek Goyal
contrib/virtiofsd/fuse_virtio.c +2
-1@@ -109,7 +109,8 @@ static uint64_t fv_get_features(VuDev *dev)uint64_t features; features =1ull <<VIRTIO_F_VERSION_1 | -1ull <<VIRTIO_FS_F_NOTIFICATION ; +1ull <<VIRTIO_FS_F_NOTIFICATION | +1ull <<VHOST_USER_F_PROTOCOL_FEATURES ;
This is not needed since VHOST_USER_F_PROTOCOL_FEATURES is already added by vu_get_features_exec():
vu_get_features_exec(VuDev *dev, VhostUserMsg *vmsg)
{
vmsg->payload.u64 =
1ULL << VHOST_F_LOG_ALL |
1ULL << VHOST_USER_F_PROTOCOL_FEATURES; if (dev->iface->get_features) {
vmsg->payload.u64 |= dev->iface->get_features(dev);
}contrib/virtiofsd/fuse_virtio.c static bool fv_queue_order(VuDev *dev, int qidx) · line 934
Quoted 11 lines
contrib/virtiofsd/fuse_virtio.c +4
-0return features; }@@ -927,6 +928,27 @@ static bool fv_queue_order(VuDev *dev, int qidx)return false; } +static uint64_t fv_get_protocol_features (VuDev *dev) +{ +return 1ull <<VHOST_USER_PROTOCOL_F_CONFIG ; +}
Please change vu_get_protocol_features_exec() in a separate patch so that devices don't need this boilerplate .get_protocol_features() code:
static bool
vu_get_protocol_features_exec(VuDev *dev, VhostUserMsg *vmsg)
{
...
- if (dev->iface->get_config && dev->iface->set_config) {
+ if (dev->iface->get_config || dev->iface->set_config) {
features |= 1ULL << VHOST_USER_PROTOCOL_F_CONFIG;
}contrib/virtiofsd/fuse_virtio.c static bool fv_queue_order(VuDev *dev, int qidx) · line 946
Quoted 12 lines
contrib/virtiofsd/fuse_virtio.c +12
-0+ +static int fv_get_config (VuDev *dev,uint8_t *config,uint32_t len) +{ +struct virtio_fs_config fscfg = {}; + +fuse_log (FUSE_LOG_DEBUG ,"%s:Setting notify_buf_size=%d\n" , __func__, +sizeof (struct fuse_notify_lock_out )); +/* +* As of now only notification related to lock is supported. As more +* notification types are supported, bump up the size accordingly. +*/ + fscfg.notify_buf_size =sizeof (struct fuse_notify_lock_out );
Missing cpu_to_le32().
I'm not sure about specifying the size precisely down to the last byte because any change to guest-visible aspects of the device (like VIRTIO Configuration Space) are not compatible across live migration. It will be necessary to introduce a device version command-line option for live migration compatibility so that existing guests can be migrated to a new virtiofsd without the device changing underneath them.
How about rounding this up to 4 KB?
Quoted 4 lines
contrib/virtiofsd/fuse_virtio.c +0
-0static void vuf_get_config (VirtIODevice *vdev,uint8_t *config) {VHostUserFS *fs =VHOST_USER_FS (vdev);struct virtio_fs_config fscfg = {};
contrib/virtiofsd/fuse_virtio.c static bool fv_queue_order(VuDev *dev, int qidx)
Quoted 10 lines
contrib/virtiofsd/fuse_virtio.c +10
-0+int ret; + +/* +* As of now we only get notification buffer size from device. And that's +* needed only if notification queue is enabled. +*/ +if (fs->notify_enabled) { + ret =vhost_dev_get_config (&fs->vhost_dev, (uint8_t *)&fs->fscfg, +sizeof (struct virtio_fs_config )); +if (ret <0 ) {
Indentation.
contrib/virtiofsd/fuse_virtio.c
Quoted 4 lines
contrib/virtiofsd/fuse_virtio.c +4
-0+error_report ("vhost-user-fs: get device config space failed." +" ret=%d\n" , ret); +return ; + }
Missing le32_to_cpu() for notify_buf_size.
Quoted 11 lines
contrib/virtiofsd/fuse_virtio.c +2
-0+ }memcpy ((char *)fscfg.tag, fs->conf.tag,MIN (strlen (fs->conf.tag) +1 ,sizeof (fscfg.tag)));virtio_stl_p (vdev, &fscfg.num_request_queues, fs->conf.num_request_queues); +virtio_stl_p (vdev, &fscfg.notify_buf_size, fs->fscfg.notify_buf_size);memcpy (config, &fscfg,sizeof (fscfg)); }
contrib/virtiofsd/fuse_virtio.c static void vuf_device_realize(DeviceState *dev, Error **errp) · line 573
Quoted 5 lines
contrib/virtiofsd/fuse_virtio.c +2
-0@@ -545,6 +569,8 @@ static void vuf_device_realize(DeviceState *dev, Error **errp) fs->vhost_dev.nvqs =2 + fs->conf.num_request_queues; fs->vhost_dev.vqs = g_new0(struct vhost_virtqueue, fs->vhost_dev.nvqs); + +vhost_dev_set_config_notifier (&fs->vhost_dev, &fs_ops);
Is this really needed since vhost_user_fs_handle_config_change() ignores it?
Re: [PATCH 3/4] virtiofsd: Specify size of notification buffer using config space
Quoted 8 lines · Stefan Hajnoczi
Quoted 7 lines · Vivek Goyal
contrib/virtiofsd/fuse_virtio.c +0
-0@@ -109,7 +109,8 @@ static uint64_t fv_get_features(VuDev *dev)uint64_t features;
Quoted 7 lines
contrib/virtiofsd/fuse_virtio.c static uint64_t fv_get_features(VuDev *dev) · line 113
Quoted 4 lines
contrib/virtiofsd/fuse_virtio.c +2
-1features =1ull <<VIRTIO_F_VERSION_1 | -1ull <<VIRTIO_FS_F_NOTIFICATION ; +1ull <<VIRTIO_FS_F_NOTIFICATION | +1ull <<VHOST_USER_F_PROTOCOL_FEATURES ;This is not needed since
VHOST_USER_F_PROTOCOL_FEATURESis already added byvu_get_features_exec():
Will do.
Quoted 26 lines
vu_get_features_exec(VuDev *dev, VhostUserMsg *vmsg) { vmsg->payload.u64 = 1ULL << VHOST_F_LOG_ALL | 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;if (dev->iface->get_features) { vmsg->payload.u64 |= dev->iface->get_features(dev); }contrib/virtiofsd/fuse_virtio.c static bool fv_queue_order(VuDev *dev, int qidx) · line 934
Quoted 11 lines
contrib/virtiofsd/fuse_virtio.c +4
-0return features; }@@ -927,6 +928,27 @@ static bool fv_queue_order(VuDev *dev, int qidx)return false; } +static uint64_t fv_get_protocol_features (VuDev *dev) +{ +return 1ull <<VHOST_USER_PROTOCOL_F_CONFIG ; +}Please change
vu_get_protocol_features_exec()in a separate patch so that devices don't need this boilerplate .get_protocol_features() code:
Quoted 7 lines
static bool
vu_get_protocol_features_exec(VuDev *dev, VhostUserMsg *vmsg){ ... - if (dev->iface->get_config && dev->iface->set_config) { + if (dev->iface->get_config || dev->iface->set_config) { features |= 1ULL << VHOST_USER_PROTOCOL_F_CONFIG;
This seems more like a nice to have thing. Can we leave it for sometime later.
Quoted 16 lines
}contrib/virtiofsd/fuse_virtio.c static bool fv_queue_order(VuDev *dev, int qidx) · line 946
Quoted 12 lines
contrib/virtiofsd/fuse_virtio.c +12
-0+ +static int fv_get_config (VuDev *dev,uint8_t *config,uint32_t len) +{ +struct virtio_fs_config fscfg = {}; + +fuse_log (FUSE_LOG_DEBUG ,"%s:Setting notify_buf_size=%d\n" , __func__, +sizeof (struct fuse_notify_lock_out )); +/* +* As of now only notification related to lock is supported. As more +* notification types are supported, bump up the size accordingly. +*/ + fscfg.notify_buf_size =sizeof (struct fuse_notify_lock_out );Missing
cpu_to_le32().
Not sure. Deivce converts to le32 when guests asks for it. So there should not be any need to do this conversion between vhost-user daemon and device. I am assuming that both daemon and qemu are using same endianess and if that's the case, first converting it to le32 and undoing this operation on other end (if we are running on an architecture with big endian), seems unnecessary and confusing.
static void vuf_get_config(VirtIODevice *vdev, uint8_t *config)
{
...
...
virtio_stl_p(vdev, &fscfg.notify_buf_size, fs->fscfg.notify_buf_size);
}Quoted 7 lines
I'm not sure about specifying the size precisely down to the last byte because any change to guest-visible aspects of the device (like VIRTIO Configuration Space) are not compatible across live migration. It will be necessary to introduce a device version command-line option for live migration compatibility so that existing guests can be migrated to a new virtiofsd without the device changing underneath them.
I am not sure I understand this point. If we were to support live migration, will we not have to reset the queue and regoniate with device again on destination host.
Quoted 2 lines
How about rounding this up to 4 KB?
Not sure how will that help. Right now it feels just wasteful of memory.
Quoted 17 lines
Quoted 5 lines
contrib/virtiofsd/fuse_virtio.c +0
-0static void vuf_get_config (VirtIODevice *vdev,uint8_t *config) {VHostUserFS *fs =VHOST_USER_FS (vdev);struct virtio_fs_config fscfg = {};contrib/virtiofsd/fuse_virtio.c static bool fv_queue_order(VuDev *dev, int qidx)
Quoted 10 lines
contrib/virtiofsd/fuse_virtio.c +10
-0+int ret; + +/* +* As of now we only get notification buffer size from device. And that's +* needed only if notification queue is enabled. +*/ +if (fs->notify_enabled) { + ret =vhost_dev_get_config (&fs->vhost_dev, (uint8_t *)&fs->fscfg, +sizeof (struct virtio_fs_config )); +if (ret <0 ) {Indentation.
Will fix.
Quoted 7 lines
contrib/virtiofsd/fuse_virtio.c
Quoted 5 lines
contrib/virtiofsd/fuse_virtio.c +4
-0+error_report ("vhost-user-fs: get device config space failed." +" ret=%d\n" , ret); +return ; + }Missing
le32_to_cpu()fornotify_buf_size.
See above.
[..]
Quoted 9 lines
contrib/virtiofsd/fuse_virtio.c static void vuf_device_realize(DeviceState *dev, Error **errp) · line 573
Quoted 6 lines
contrib/virtiofsd/fuse_virtio.c +2
-0@@ -545,6 +569,8 @@ static void vuf_device_realize(DeviceState *dev, Error **errp) fs->vhost_dev.nvqs =2 + fs->conf.num_request_queues; fs->vhost_dev.vqs = g_new0(struct vhost_virtqueue, fs->vhost_dev.nvqs); + +vhost_dev_set_config_notifier (&fs->vhost_dev, &fs_ops);Is this really needed since
vhost_user_fs_handle_config_change()ignores it?
Initially I did not introduce it but code did not work. Looked little closer and noticed following code in vhost_user_backend_init().
if (!dev->config_ops || !dev->config_ops->vhost_dev_config_notifier) {
/* Don't acknowledge CONFIG feature if device doesn't support it */
dev->protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG);So if dev->config_ops->vhost_dev_config_notifier is not provided, feature VHOST_USER_PROTOCOL_F_CONFIG will be reset.
Its kind of odd that its a hard requirement. Anyway, that's the reason I added it so that VHOST_USER_PROTOCOL_F_CONFIG continues to work.
Thanks
Vivek
PATCH 4/4 virtiofsd: Implement blocking posix locks
In reply to Vivek GoyalAs of now we don't support fcntl(F_SETLKW) and if we see one, we return -EOPNOTSUPP.
Change that by accepting these requests and returning a reply immediately asking caller to wait. Once lock is available, send a notification to the waiter indicating lock is available.
--- contrib/virtiofsd/fuse_kernel.h | 7 +++ contrib/virtiofsd/fuse_lowlevel.c | 23 +++++++- contrib/virtiofsd/fuse_lowlevel.h | 25 ++++++++ contrib/virtiofsd/fuse_virtio.c | 94 ++++++++++++++++++++++++++++-- contrib/virtiofsd/passthrough_ll.c | 49 +++++++++++++--- 5 files changed, 182 insertions(+), 16 deletions(-)
contrib/virtiofsd/fuse_kernel.h +7 -0
@@ -444,6 +444,7 @@ enum fuse_notify_code {FUSE_NOTIFY_STORE =4 ,FUSE_NOTIFY_RETRIEVE =5 ,FUSE_NOTIFY_DELETE =6 , +FUSE_NOTIFY_LOCK =7 , FUSE_NOTIFY_CODE_MAX, };
@@ -836,6 +837,12 @@ struct fuse_notify_retrieve_in {uint64_t dummy4; }; +struct fuse_notify_lock_out { +uint64_t id; +int32_t error; +int32_t padding; +}; +/* Device ioctls: */ #define FUSE_DEV_IOC_CLONE _IOR(229, 0, uint32_t)
contrib/virtiofsd/fuse_lowlevel.c +22 -1
@@ -183,7 +183,8 @@ int fuse_send_reply_iov_nofree(fuse_req_t req, int error, struct iovec *iov, {struct fuse_out_header out; -if (error <=-1000 || error >0 ) { +/* error = 1 has been used to signal client to wait for notificaiton */ +if (error <=-1000 || error >1 ) {fuse_log (FUSE_LOG_ERR ,"fuse: bad error value: %i\n" , error); error = -ERANGE ; }
@@ -291,6 +292,12 @@ int fuse_reply_err(fuse_req_t req, int err)return send_reply (req, -err,NULL ,0 ); } +int fuse_reply_wait (fuse_req_t req) +{ +/* TODO: This is a hack. Fix it */ +return send_reply (req,1 ,NULL ,0 ); +} +void fuse_reply_none (fuse_req_t req) {fuse_free_req (req);
@@ -2207,6 +2214,20 @@ static int send_notify_iov(struct fuse_session *se, int notify_code,return fuse_send_msg (se,NULL , iov, count); } +int fuse_lowlevel_notify_lock (struct fuse_session *se,uint64_t req_id, +int32_t error) +{ +struct fuse_notify_lock_out outarg; +struct iovec iov[2 ]; + + outarg.id = req_id; + outarg.error = -error; + + iov[1 ].iov_base = &outarg; + iov[1 ].iov_len =sizeof (outarg); +return send_notify_iov (se,FUSE_NOTIFY_LOCK , iov,2 ); +} +int fuse_lowlevel_notify_poll (struct fuse_pollhandle *ph) {if (ph !=NULL ) {
contrib/virtiofsd/fuse_lowlevel.h +25 -0
@@ -1251,6 +1251,22 @@ struct fuse_lowlevel_ops {*/ int fuse_reply_err (fuse_req_t req,int err); +/** +* Ask caller to wait for lock. +* +* Possible requests: +* setlkw +* +* If caller sends a blocking lock request (setlkw), then reply to caller +* that wait for lock to be available. Once lock is available caller will +* receive a notification with request's unique id. Notification will +* carry info whether lock was successfully obtained or not. +* +* @param req request handle +* @return zero for success, -errno for failure to send reply +*/ +int fuse_reply_wait (fuse_req_t req); +/** * Don't send reply *
@@ -1704,6 +1720,15 @@ int fuse_lowlevel_notify_delete(struct fuse_session *se,int fuse_lowlevel_notify_store (struct fuse_session *se,fuse_ino_t ino,off_t offset,struct fuse_bufvec *bufv,enum fuse_buf_copy_flags flags); +/** +* Notify event related to previous lock request +* +* @param se the session object +* @param req_id the id of the request which requested setlkw +* @param error zero for success, -errno for the failure +*/ +int fuse_lowlevel_notify_lock (struct fuse_session *se,uint64_t req_id, +int32_t error);/* ----------------------------------------------------------- * * Utility functions *
contrib/virtiofsd/fuse_virtio.c +88 -6
113 lines not on this page · raw message
contrib/virtiofsd/passthrough_ll.c +40 -9
79 lines not on this page · raw message
Signature · 1 line
2.20.1
Re: [PATCH 4/4] virtiofsd: Implement blocking posix locks
Quoted 5 lines · Vivek Goyal
contrib/virtiofsd/fuse_lowlevel.c +0
-0
contrib/virtiofsd/fuse_lowlevel.c int fuse_send_reply_iov_nofree(fuse_req_t req, int error, struct iovec *iov, · line 190
Quoted 9 lines
contrib/virtiofsd/fuse_lowlevel.c +2
-1@@ -183,7 +183,8 @@ int fuse_send_reply_iov_nofree(fuse_req_t req, int error, struct iovec *iov, {struct fuse_out_header out; -if (error <=-1000 || error >0 ) { +/* error = 1 has been used to signal client to wait for notificaiton */ +if (error <=-1000 || error >1 ) {fuse_log (FUSE_LOG_ERR ,"fuse: bad error value: %i\n" , error); error = -ERANGE ; }
What is this?
contrib/virtiofsd/fuse_lowlevel.c static int send_notify_iov(struct fuse_session *se, int notify_code, · line 2220
Quoted 4 lines
contrib/virtiofsd/fuse_lowlevel.c +4
-0+int fuse_lowlevel_notify_lock (struct fuse_session *se,uint64_t req_id, +int32_t error) +{ +struct fuse_notify_lock_out outarg;
Missing = {} initialization to avoid information leaks to the guest.
contrib/virtiofsd/fuse_lowlevel.c int fuse_lowlevel_notify_delete(struct fuse_session *se, · line 1727
Quoted 9 lines
contrib/virtiofsd/fuse_lowlevel.c +5
-0@@ -1704,6 +1720,15 @@ int fuse_lowlevel_notify_delete(struct fuse_session *se,int fuse_lowlevel_notify_store (struct fuse_session *se,fuse_ino_t ino,off_t offset,struct fuse_bufvec *bufv,enum fuse_buf_copy_flags flags); +/** +* Notify event related to previous lock request +* +* @param se the session object +* @param req_id the id of the request which requested setlkw
The rest of the code calls this id "unique":
+ * @param req_unique the unique id of the setlkw request
contrib/virtiofsd/fuse_lowlevel.c int fuse_lowlevel_notify_delete(struct fuse_session *se,
Quoted 8 lines
contrib/virtiofsd/fuse_lowlevel.c +8
-0+/* Pop an element from queue */ + req =vu_queue_pop (dev, q,sizeof (FVRequest), &bad_in_num, &bad_out_num); +if (!req) { +/* TODO: Implement some sort of ring buffer and queue notifications +* on that and send these later when notification queue has space +* available. +*/ +return -ENOSPC ;
Ah, I thought the point of the notifications processing thread was exactly this case. It could wake any threads waiting for buffers.
This wakeup could be implemented with a condvar - no ring buffer necessary.
Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Implement blocking posix locks
Quoted 9 lines · Stefan Hajnoczi
Quoted 8 lines · Vivek Goyal
contrib/virtiofsd/fuse_lowlevel.c +0
-0@@ -183,7 +183,8 @@ int fuse_send_reply_iov_nofree(fuse_req_t req, int error, struct iovec *iov, {struct fuse_out_header out;
Quoted 8 lines
contrib/virtiofsd/fuse_lowlevel.c int fuse_send_reply_iov_nofree(fuse_req_t req, int error, struct iovec *iov, · line 190
Quoted 6 lines
contrib/virtiofsd/fuse_lowlevel.c +2
-1-if (error <=-1000 || error >0 ) { +/* error = 1 has been used to signal client to wait for notificaiton */ +if (error <=-1000 || error >1 ) {fuse_log (FUSE_LOG_ERR ,"fuse: bad error value: %i\n" , error); error = -ERANGE ; }What is this?
When a waiting lock request comes in, we need a way to reply back saying wait for the notification. So I used value "1" for the fuse_out_header->error field for this purpose. As of now, 0 is returned for success and negative values for error code. So positive values seem to be unused.
Quoted 7 lines
contrib/virtiofsd/fuse_lowlevel.c static int send_notify_iov(struct fuse_session *se, int notify_code, · line 2220
Quoted 5 lines
contrib/virtiofsd/fuse_lowlevel.c +4
-0+int fuse_lowlevel_notify_lock (struct fuse_session *se,uint64_t req_id, +int32_t error) +{ +struct fuse_notify_lock_out outarg;Missing = {} initialization to avoid information leaks to the guest.
Will do.
Quoted 12 lines
contrib/virtiofsd/fuse_lowlevel.c int fuse_lowlevel_notify_delete(struct fuse_session *se, · line 1727
Quoted 10 lines
contrib/virtiofsd/fuse_lowlevel.c +5
-0@@ -1704,6 +1720,15 @@ int fuse_lowlevel_notify_delete(struct fuse_session *se,int fuse_lowlevel_notify_store (struct fuse_session *se,fuse_ino_t ino,off_t offset,struct fuse_bufvec *bufv,enum fuse_buf_copy_flags flags); +/** +* Notify event related to previous lock request +* +* @param se the session object +* @param req_id the id of the request which requested setlkw The rest of the code calls this id "unique":
Will change it.
Quoted 15 lines
+ * @param
req_uniquethe unique id of the setlkw requestcontrib/virtiofsd/fuse_lowlevel.c int fuse_lowlevel_notify_delete(struct fuse_session *se,
Quoted 8 lines
contrib/virtiofsd/fuse_lowlevel.c +8
-0+/* Pop an element from queue */ + req =vu_queue_pop (dev, q,sizeof (FVRequest), &bad_in_num, &bad_out_num); +if (!req) { +/* TODO: Implement some sort of ring buffer and queue notifications +* on that and send these later when notification queue has space +* available. +*/ +return -ENOSPC ;Ah, I thought the point of the notifications processing thread was exactly this case. It could wake any threads waiting for buffers.
Quoted 2 lines
This wakeup could be implemented with a condvar - no ring buffer necessary.
I was thinking that thread sending notification should not block. It can just queue the notification reuqest and some other thread (including notification thread could send it later). Number of pre-allocated buffers could be of fixed and we will drop notifications if guest is not responding. This will also take care of concerns w.r.t rogue guest blocking filesystem code in daemon.
Anyway, this is a TODO item and not implemented yet.
Thanks
Vivek
Re: [PATCH 4/4] virtiofsd: Implement blocking posix locks
In reply to Vivek GoyalQuoted 158 lines · * Vivek Goyal ([email protected]) wrote:
As of now we don't support
fcntl(F_SETLKW)and if we see one, we return-EOPNOTSUPP.Change that by accepting these requests and returning a reply immediately asking caller to wait. Once lock is available, send a notification to the waiter indicating lock is available.
Signed-off-by Vivek Goyal <[email protected]>--- contrib/virtiofsd/fuse_kernel.h | 7 +++ contrib/virtiofsd/fuse_lowlevel.c | 23 +++++++- contrib/virtiofsd/fuse_lowlevel.h | 25 ++++++++ contrib/virtiofsd/fuse_virtio.c | 94 ++++++++++++++++++++++++++++-- contrib/virtiofsd/passthrough_ll.c | 49 +++++++++++++--- 5 files changed, 182 insertions(+), 16 deletions(-)contrib/virtiofsd/fuse_kernel.h +7
-0@@ -444,6 +444,7 @@ enum fuse_notify_code {FUSE_NOTIFY_STORE =4 ,FUSE_NOTIFY_RETRIEVE =5 ,FUSE_NOTIFY_DELETE =6 , +FUSE_NOTIFY_LOCK =7 , FUSE_NOTIFY_CODE_MAX, };@@ -836,6 +837,12 @@ struct fuse_notify_retrieve_in {uint64_t dummy4; }; +struct fuse_notify_lock_out { +uint64_t id; +int32_t error; +int32_t padding; +}; +/* Device ioctls: */ #define FUSE_DEV_IOC_CLONE _IOR(229, 0, uint32_t)contrib/virtiofsd/fuse_lowlevel.c +22
-1@@ -183,7 +183,8 @@ int fuse_send_reply_iov_nofree(fuse_req_t req, int error, struct iovec *iov, {struct fuse_out_header out; -if (error <=-1000 || error >0 ) { +/* error = 1 has been used to signal client to wait for notificaiton */ +if (error <=-1000 || error >1 ) {fuse_log (FUSE_LOG_ERR ,"fuse: bad error value: %i\n" , error); error = -ERANGE ; }@@ -291,6 +292,12 @@ int fuse_reply_err(fuse_req_t req, int err)return send_reply (req, -err,NULL ,0 ); } +int fuse_reply_wait (fuse_req_t req) +{ +/* TODO: This is a hack. Fix it */ +return send_reply (req,1 ,NULL ,0 ); +} +void fuse_reply_none (fuse_req_t req) {fuse_free_req (req);@@ -2207,6 +2214,20 @@ static int send_notify_iov(struct fuse_session *se, int notify_code,return fuse_send_msg (se,NULL , iov, count); } +int fuse_lowlevel_notify_lock (struct fuse_session *se,uint64_t req_id, +int32_t error) +{ +struct fuse_notify_lock_out outarg; +struct iovec iov[2 ]; + + outarg.id = req_id; + outarg.error = -error; + + iov[1 ].iov_base = &outarg; + iov[1 ].iov_len =sizeof (outarg); +return send_notify_iov (se,FUSE_NOTIFY_LOCK , iov,2 ); +} +int fuse_lowlevel_notify_poll (struct fuse_pollhandle *ph) {if (ph !=NULL ) {contrib/virtiofsd/fuse_lowlevel.h +25
-0@@ -1251,6 +1251,22 @@ struct fuse_lowlevel_ops {*/ int fuse_reply_err (fuse_req_t req,int err); +/** +* Ask caller to wait for lock. +* +* Possible requests: +* setlkw +* +* If caller sends a blocking lock request (setlkw), then reply to caller +* that wait for lock to be available. Once lock is available caller will +* receive a notification with request's unique id. Notification will +* carry info whether lock was successfully obtained or not. +* +* @param req request handle +* @return zero for success, -errno for failure to send reply +*/ +int fuse_reply_wait (fuse_req_t req); +/** * Don't send reply * @@ -1704,6 +1720,15 @@ int fuse_lowlevel_notify_delete(struct fuse_session *se,int fuse_lowlevel_notify_store (struct fuse_session *se,fuse_ino_t ino,off_t offset,struct fuse_bufvec *bufv,enum fuse_buf_copy_flags flags); +/** +* Notify event related to previous lock request +* +* @param se the session object +* @param req_id the id of the request which requested setlkw +* @param error zero for success, -errno for the failure +*/ +int fuse_lowlevel_notify_lock (struct fuse_session *se,uint64_t req_id, +int32_t error);/* ----------------------------------------------------------- * * Utility functions * contrib/virtiofsd/fuse_virtio.c +18
-0@@ -215,6 +215,81 @@ static void copy_iov(struct iovec *src_iov, int src_count, } } +static int virtio_send_notify_msg (struct fuse_session *se,struct iovec *iov, +int count) +{ +struct fv_QueueInfo *qi; +VuDev *dev = &se->virtio_dev->dev; +VuVirtq *q; +FVRequest *req; +VuVirtqElement *elem; +unsigned int in_num, bad_in_num =0 , bad_out_num =0 ; +struct fuse_out_header *out = iov[0 ].iov_base; +size_t in_len, tosend_len =iov_size (iov, count); +struct iovec *in_sg; +int ret =0 ; + +/* Notifications have unique == 0 */ +assert (!out->unique); + +if (!se->notify_enabled)
contrib/virtiofsd/fuse_virtio.c static void copy_iov(struct iovec *src_iov, int src_count, · line 245
Quoted 10 lines
contrib/virtiofsd/fuse_virtio.c +10
-0+return -EOPNOTSUPP ; + +/* If notifications are enabled, queue index 1 is notification queue */ + qi = se->virtio_dev->qi[1 ]; + q =vu_get_queue (dev, qi->qidx); + +pthread_rwlock_rdlock (&qi->virtio_dev->vu_dispatch_rwlock); +pthread_mutex_lock (&qi->vq_lock); +/* Pop an element from queue */ + req =vu_queue_pop (dev, q,sizeof (FVRequest), &bad_in_num, &bad_out_num);
You don't need bad_in_num/bad_out_num - just pass NULL for both; they're only needed if you expect to read/write data that's not mappable (i.e. in our direct write case).
Quoted 143 lines
contrib/virtiofsd/fuse_virtio.c +60
-6+if (!req) { +/* TODO: Implement some sort of ring buffer and queue notifications +* on that and send these later when notification queue has space +* available. +*/ +return -ENOSPC ; + } +pthread_mutex_unlock (&qi->vq_lock); +pthread_rwlock_unlock (&qi->virtio_dev->vu_dispatch_rwlock); + + out->len = tosend_len; + elem = &req->elem; + in_num = elem->in_num; + in_sg = elem->in_sg; + in_len =iov_size (in_sg, in_num); +fuse_log (FUSE_LOG_DEBUG ,"%s: elem %d: with %d in desc of length %zd\n" , + __func__, elem->index, in_num, in_len); + +if (in_len <sizeof (struct fuse_out_header )) { +fuse_log (FUSE_LOG_ERR ,"%s: elem %d too short for out_header\n" , + __func__, elem->index); + ret = -E2BIG ; + goto out; + } + +if (in_len < tosend_len) { +fuse_log (FUSE_LOG_ERR ,"%s: elem %d too small for data len" +" %zd\n" , __func__, elem->index, tosend_len); + ret = -E2BIG ; + goto out; + } + +/* First copy the header data from iov->in_sg */ +copy_iov (iov, count, in_sg, in_num, tosend_len); + +/* TODO: Add bad_innum handling */ +pthread_rwlock_rdlock (&qi->virtio_dev->vu_dispatch_rwlock); +pthread_mutex_lock (&qi->vq_lock); +vu_queue_push (dev, q, elem, tosend_len); +vu_queue_notify (dev, q); +pthread_mutex_unlock (&qi->vq_lock); +pthread_rwlock_unlock (&qi->virtio_dev->vu_dispatch_rwlock); +out: +free (req); +return ret; +} +/* * Called back by ll whenever it wants to send a reply/message back * The 1st element of the iov starts with the fuse_out_header @@ -223,11 +298,11 @@ static void copy_iov(struct iovec *src_iov, int src_count,int virtio_send_msg (struct fuse_session *se,struct fuse_chan *ch,struct iovec *iov,int count) { -FVRequest *req =container_of (ch, FVRequest, ch); -struct fv_QueueInfo *qi = ch->qi; +FVRequest *req; +struct fv_QueueInfo *qi;VuDev *dev = &se->virtio_dev->dev; -VuVirtq *q =vu_get_queue (dev, qi->qidx); -VuVirtqElement *elem = &req->elem; +VuVirtq *q; +VuVirtqElement *elem;int ret =0 ;assert (count >=1 );@@ -238,8 +313,15 @@ int virtio_send_msg(struct fuse_session *se, struct fuse_chan *ch,size_t tosend_len =iov_size (iov, count); -/* unique == 0 is notification, which we don't support */ -assert (out->unique); +/* unique == 0 is notification */ +if (!out->unique) +return virtio_send_notify_msg (se, iov, count); + +assert (ch); + req =container_of (ch, FVRequest, ch); + elem = &req->elem; + qi = ch->qi; + q =vu_get_queue (dev, qi->qidx);assert (!req->reply_sent);/* The 'in' part of the elem is to qemu */ contrib/virtiofsd/passthrough_ll.c +28
-6@@ -1925,7 +1925,10 @@ static void lo_setlk(fuse_req_t req, fuse_ino_t ino,struct lo_data *lo =lo_data (req);struct lo_inode *inode;struct lo_inode_plock *plock; -int ret, saverr =0 ; +int ret, saverr =0 , ofd; +uint64_t unique; +struct fuse_session *se = req->se; +bool async_lock = false;fuse_log (FUSE_LOG_DEBUG ,"lo_setlk(ino=%" PRIu64", flags=%d)" " cmd=%d pid=%d owner=0x%lx sleep=%d l_whence=%d"@@ -1933,11 +1936,6 @@ static void lo_setlk(fuse_req_t req, fuse_ino_t ino, lock->l_type, lock->l_pid, fi->lock_owner, sleep, lock->l_whence, lock->l_start, lock->l_len); -if (sleep) { -fuse_reply_err (req,EOPNOTSUPP ); -return ; - } - inode =lo_inode (req, ino);if (!inode) {fuse_reply_err (req,EBADF );@@ -1950,21 +1948,54 @@ static void lo_setlk(fuse_req_t req, fuse_ino_t ino,if (!plock) { saverr = ret; +pthread_mutex_unlock (&inode->plock_mutex); goto out; } +/* +* plock is now released when inode is going away. We already have +* a reference on inode, so it is guaranteed that plock->fd is +* still around even after dropping inode->plock_mutex lock +*/ + ofd = plock->fd; +pthread_mutex_unlock (&inode->plock_mutex); + +/* +* If this lock request can block, request caller to wait for +* notification. Do not access req after this. Once lock is +* available, send a notification instead. +*/ +if (sleep && lock->l_type !=F_UNLCK ) { +/* +* If notification queue is not enabled, can't support async +* locks. +*/ +if (!se->notify_enabled) { + saverr =EOPNOTSUPP ; + goto out; + } + async_lock = true;
contrib/virtiofsd/passthrough_ll.c static void lo_setlk(fuse_req_t req, fuse_ino_t ino, · line 1986
Quoted 10 lines
contrib/virtiofsd/passthrough_ll.c +7
-1+ unique = req->unique; +fuse_reply_wait (req); + }/* TODO: Is it alright to modify flock? */ lock->l_pid =0 ; - ret =fcntl (plock->fd,F_OFD_SETLK , lock); +if (async_lock) + ret =fcntl (ofd,F_OFD_SETLKW , lock); +else + ret =fcntl (ofd,F_OFD_SETLK , lock);
What happens if the guest is rebooted after it's asked for, but not been granted a lock?
Dave
Quoted 20 lines
contrib/virtiofsd/passthrough_ll.c +5
-2if (ret ==-1 ) { saverr = errno; } out: -pthread_mutex_unlock (&inode->plock_mutex);lo_inode_put (lo, &inode); -fuse_reply_err (req, saverr); +if (!async_lock) +fuse_reply_err (req, saverr); +else { +fuse_lowlevel_notify_lock (se, unique, saverr); + } } static void lo_fsyncdir(fuse_req_t req, fuse_ino_t ino, int datasync,Signature · 1 line
2.20.1
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Implement blocking posix locks
On Fri, Nov 22, 2019 at 05:47:32PM +0000, Dr. David Alan Gilbert wrote:
[..]
Quoted 24 lines
Quoted 24 lines
contrib/virtiofsd/fuse_virtio.c +24
-0+static int virtio_send_notify_msg (struct fuse_session *se,struct iovec *iov, +int count) +{ +struct fv_QueueInfo *qi; +VuDev *dev = &se->virtio_dev->dev; +VuVirtq *q; +FVRequest *req; +VuVirtqElement *elem; +unsigned int in_num, bad_in_num =0 , bad_out_num =0 ; +struct fuse_out_header *out = iov[0 ].iov_base; +size_t in_len, tosend_len =iov_size (iov, count); +struct iovec *in_sg; +int ret =0 ; + +/* Notifications have unique == 0 */ +assert (!out->unique); + +if (!se->notify_enabled) +return -EOPNOTSUPP ; + +/* If notifications are enabled, queue index 1 is notification queue */ + qi = se->virtio_dev->qi[1 ]; + q =vu_get_queue (dev, qi->qidx); +
Quoted 8 lines
contrib/virtiofsd/fuse_virtio.c static void copy_iov(struct iovec *src_iov, int src_count, · line 245
Quoted 4 lines
contrib/virtiofsd/fuse_virtio.c +4
-0+pthread_rwlock_rdlock (&qi->virtio_dev->vu_dispatch_rwlock); +pthread_mutex_lock (&qi->vq_lock); +/* Pop an element from queue */ + req =vu_queue_pop (dev, q,sizeof (FVRequest), &bad_in_num, &bad_out_num);You don't need bad_in_num/bad_out_num - just pass NULL for both; they're only needed if you expect to read/write data that's not mappable (i.e. in our direct write case).
Will do.
[..]
Quoted 36 lines
Quoted 36 lines
contrib/virtiofsd/passthrough_ll.c +27
-0@@ -1950,21 +1948,54 @@ static void lo_setlk(fuse_req_t req, fuse_ino_t ino,if (!plock) { saverr = ret; +pthread_mutex_unlock (&inode->plock_mutex); goto out; } +/* +* plock is now released when inode is going away. We already have +* a reference on inode, so it is guaranteed that plock->fd is +* still around even after dropping inode->plock_mutex lock +*/ + ofd = plock->fd; +pthread_mutex_unlock (&inode->plock_mutex); + +/* +* If this lock request can block, request caller to wait for +* notification. Do not access req after this. Once lock is +* available, send a notification instead. +*/ +if (sleep && lock->l_type !=F_UNLCK ) { +/* +* If notification queue is not enabled, can't support async +* locks. +*/ +if (!se->notify_enabled) { + saverr =EOPNOTSUPP ; + goto out; + } + async_lock = true; + unique = req->unique; +fuse_reply_wait (req); + }/* TODO: Is it alright to modify flock? */ lock->l_pid =0 ;
Quoted 8 lines
contrib/virtiofsd/passthrough_ll.c static void lo_setlk(fuse_req_t req, fuse_ino_t ino, · line 1986
Quoted 5 lines
contrib/virtiofsd/passthrough_ll.c +4
-1- ret =fcntl (plock->fd,F_OFD_SETLK , lock); +if (async_lock) + ret =fcntl (ofd,F_OFD_SETLKW , lock); +else + ret =fcntl (ofd,F_OFD_SETLK , lock);What happens if the guest is rebooted after it's asked for, but not been granted a lock?
I think a regular reboot can't be done till a request is pending, because virtio-fs can't be unmounted and unmount will wait for all pending requests to finish.
Destroying qemu will destroy deamon too.
Are there any other reboot paths I have missed.
Thanks
Vivek
Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Implement blocking posix locks
Quoted 84 lines · * Vivek Goyal ([email protected]) wrote:
On Fri, Nov 22, 2019 at 05:47:32PM +0000, Dr. David Alan Gilbert wrote:
[..]
Quoted 24 lines
Quoted 24 lines
contrib/virtiofsd/fuse_virtio.c +24
-0+static int virtio_send_notify_msg (struct fuse_session *se,struct iovec *iov, +int count) +{ +struct fv_QueueInfo *qi; +VuDev *dev = &se->virtio_dev->dev; +VuVirtq *q; +FVRequest *req; +VuVirtqElement *elem; +unsigned int in_num, bad_in_num =0 , bad_out_num =0 ; +struct fuse_out_header *out = iov[0 ].iov_base; +size_t in_len, tosend_len =iov_size (iov, count); +struct iovec *in_sg; +int ret =0 ; + +/* Notifications have unique == 0 */ +assert (!out->unique); + +if (!se->notify_enabled) +return -EOPNOTSUPP ; + +/* If notifications are enabled, queue index 1 is notification queue */ + qi = se->virtio_dev->qi[1 ]; + q =vu_get_queue (dev, qi->qidx); +Quoted 8 lines
contrib/virtiofsd/fuse_virtio.c static void copy_iov(struct iovec *src_iov, int src_count, · line 245
Quoted 4 lines
contrib/virtiofsd/fuse_virtio.c +4
-0+pthread_rwlock_rdlock (&qi->virtio_dev->vu_dispatch_rwlock); +pthread_mutex_lock (&qi->vq_lock); +/* Pop an element from queue */ + req =vu_queue_pop (dev, q,sizeof (FVRequest), &bad_in_num, &bad_out_num);You don't need bad_in_num/bad_out_num - just pass NULL for both; they're only needed if you expect to read/write data that's not mappable (i.e. in our direct write case).
Will do.
[..]
Quoted 36 lines
Quoted 36 lines
contrib/virtiofsd/passthrough_ll.c +27
-0@@ -1950,21 +1948,54 @@ static void lo_setlk(fuse_req_t req, fuse_ino_t ino,if (!plock) { saverr = ret; +pthread_mutex_unlock (&inode->plock_mutex); goto out; } +/* +* plock is now released when inode is going away. We already have +* a reference on inode, so it is guaranteed that plock->fd is +* still around even after dropping inode->plock_mutex lock +*/ + ofd = plock->fd; +pthread_mutex_unlock (&inode->plock_mutex); + +/* +* If this lock request can block, request caller to wait for +* notification. Do not access req after this. Once lock is +* available, send a notification instead. +*/ +if (sleep && lock->l_type !=F_UNLCK ) { +/* +* If notification queue is not enabled, can't support async +* locks. +*/ +if (!se->notify_enabled) { + saverr =EOPNOTSUPP ; + goto out; + } + async_lock = true; + unique = req->unique; +fuse_reply_wait (req); + }/* TODO: Is it alright to modify flock? */ lock->l_pid =0 ;Quoted 8 lines
contrib/virtiofsd/passthrough_ll.c static void lo_setlk(fuse_req_t req, fuse_ino_t ino, · line 1986
Quoted 5 lines
contrib/virtiofsd/passthrough_ll.c +4
-1- ret =fcntl (plock->fd,F_OFD_SETLK , lock); +if (async_lock) + ret =fcntl (ofd,F_OFD_SETLKW , lock); +else + ret =fcntl (ofd,F_OFD_SETLK , lock);What happens if the guest is rebooted after it's asked for, but not been granted a lock?
Quoted 7 lines
I think a regular reboot can't be done till a request is pending, because virtio-fs can't be unmounted and unmount will wait for all pending requests to finish.
Destroying qemu will destroy deamon too.
Are there any other reboot paths I have missed.
Yes, there are a few other ways the guest can reboot:
a) A echo b > /proc/sysrq-trigger
b) Telling qemu to do a reset
probably a few more as well; but they all end up with the daemon
still running over the same connection. See
'virtiofsd: Handle hard reboot' where I handle that case where a FUSE_INIT turns up unexpectedly.
Dave
Quoted 2 lines
Thanks
Vivek
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Implement blocking posix locks
On Tue, Nov 26, 2019 at 01:02:29PM +0000, Dr. David Alan Gilbert wrote:
[..]
Quoted 49 lines
Quoted 49 lines
Quoted 36 lines
Quoted 36 lines
contrib/virtiofsd/passthrough_ll.c +27
-0@@ -1950,21 +1948,54 @@ static void lo_setlk(fuse_req_t req, fuse_ino_t ino,if (!plock) { saverr = ret; +pthread_mutex_unlock (&inode->plock_mutex); goto out; } +/* +* plock is now released when inode is going away. We already have +* a reference on inode, so it is guaranteed that plock->fd is +* still around even after dropping inode->plock_mutex lock +*/ + ofd = plock->fd; +pthread_mutex_unlock (&inode->plock_mutex); + +/* +* If this lock request can block, request caller to wait for +* notification. Do not access req after this. Once lock is +* available, send a notification instead. +*/ +if (sleep && lock->l_type !=F_UNLCK ) { +/* +* If notification queue is not enabled, can't support async +* locks. +*/ +if (!se->notify_enabled) { + saverr =EOPNOTSUPP ; + goto out; + } + async_lock = true; + unique = req->unique; +fuse_reply_wait (req); + }/* TODO: Is it alright to modify flock? */ lock->l_pid =0 ;Quoted 8 lines
contrib/virtiofsd/passthrough_ll.c static void lo_setlk(fuse_req_t req, fuse_ino_t ino, · line 1986
Quoted 5 lines
contrib/virtiofsd/passthrough_ll.c +4
-1- ret =fcntl (plock->fd,F_OFD_SETLK , lock); +if (async_lock) + ret =fcntl (ofd,F_OFD_SETLKW , lock); +else + ret =fcntl (ofd,F_OFD_SETLK , lock);What happens if the guest is rebooted after it's asked for, but not been granted a lock?
I think a regular reboot can't be done till a request is pending, because virtio-fs can't be unmounted and unmount will wait for all pending requests to finish.
Quoted 6 lines
Quoted 3 lines
Destroying qemu will destroy deamon too.
Are there any other reboot paths I have missed.
Yes, there are a few other ways the guest can reboot:
a) A echo b >/proc/sysrq-trigger
I tried it. Both qemu and virtiofsd hang. virtiofsd wants to stop a queue. And that tries to stop thrad pool. But one of the threads in thread pool is blocked on setlkw. So g_thread_pool_free() hangs.
I am not seeing any option in glib thread pool API to stop or send signal to threads which are blocked.
Thanks
Vivek
Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Implement blocking posix locks
Quoted 59 lines · * Vivek Goyal ([email protected]) wrote:
On Tue, Nov 26, 2019 at 01:02:29PM +0000, Dr. David Alan Gilbert wrote:
[..]
Quoted 49 lines
Quoted 49 lines
Quoted 36 lines
Quoted 36 lines
contrib/virtiofsd/passthrough_ll.c +27
-0@@ -1950,21 +1948,54 @@ static void lo_setlk(fuse_req_t req, fuse_ino_t ino,if (!plock) { saverr = ret; +pthread_mutex_unlock (&inode->plock_mutex); goto out; } +/* +* plock is now released when inode is going away. We already have +* a reference on inode, so it is guaranteed that plock->fd is +* still around even after dropping inode->plock_mutex lock +*/ + ofd = plock->fd; +pthread_mutex_unlock (&inode->plock_mutex); + +/* +* If this lock request can block, request caller to wait for +* notification. Do not access req after this. Once lock is +* available, send a notification instead. +*/ +if (sleep && lock->l_type !=F_UNLCK ) { +/* +* If notification queue is not enabled, can't support async +* locks. +*/ +if (!se->notify_enabled) { + saverr =EOPNOTSUPP ; + goto out; + } + async_lock = true; + unique = req->unique; +fuse_reply_wait (req); + }/* TODO: Is it alright to modify flock? */ lock->l_pid =0 ;Quoted 8 lines
contrib/virtiofsd/passthrough_ll.c static void lo_setlk(fuse_req_t req, fuse_ino_t ino, · line 1986
Quoted 5 lines
contrib/virtiofsd/passthrough_ll.c +4
-1- ret =fcntl (plock->fd,F_OFD_SETLK , lock); +if (async_lock) + ret =fcntl (ofd,F_OFD_SETLKW , lock); +else + ret =fcntl (ofd,F_OFD_SETLK , lock);What happens if the guest is rebooted after it's asked for, but not been granted a lock?
I think a regular reboot can't be done till a request is pending, because virtio-fs can't be unmounted and unmount will wait for all pending requests to finish.
Quoted 6 lines
Quoted 3 lines
Destroying qemu will destroy deamon too.
Are there any other reboot paths I have missed.
Yes, there are a few other ways the guest can reboot:
a) A echo b >/proc/sysrq-trigger
Quoted 6 lines
I tried it. Both qemu and virtiofsd hang. virtiofsd wants to stop a queue. And that tries to stop thrad pool. But one of the threads in thread pool is blocked on setlkw. So
g_thread_pool_free()hangs.I am not seeing any option in glib thread pool API to stop or send signal to threads which are blocked.
Is there a way to setup pthread_cancel ? The upstream libfuse code has somec ases where it enables cancellation very carefully around something that might block, does it, then disables cancellation.
Dave
Quoted 2 lines
Thanks
Vivek
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK