Message ID | 20231101093325.30302-5-jgross@suse.com (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | tools: enable xenstore-stubdom to use 9pfs | expand |
On Wed, Nov 1, 2023 at 5:34 AM Juergen Gross <jgross@suse.com> wrote: > > Add the transport layer of 9pfs. This is basically the infrastructure > to receive requests from the frontend and to send the related answers > via the rings. > > In order to avoid unaligned accesses e.g. on Arm, add the definition of > __packed to the common-macros.h header. > > Signed-off-by: Juergen Gross <jgross@suse.com> > --- > diff --git a/tools/xenlogd/io.c b/tools/xenlogd/io.c > index ef0954d69d..590d06e906 100644 > --- a/tools/xenlogd/io.c > +++ b/tools/xenlogd/io.c > +static unsigned int get_request_bytes(device *device, unsigned int off, > + unsigned int len) > +{ > + unsigned int size; > + unsigned int out_data = ring_out_data(device); > + RING_IDX prod, cons; > + > + size = min(len - off, out_data); > + prod = xen_9pfs_mask(device->intf->out_prod, device->ring_size); > + cons = xen_9pfs_mask(device->cons_pvt_out, device->ring_size); > + xen_9pfs_read_packet(device->buffer + off, device->data.out, size, > + prod, &cons, device->ring_size); > + > + xen_rmb(); /* Read data out before setting visible consumer. */ > + device->cons_pvt_out += size; > + device->intf->out_cons = device->cons_pvt_out; > + > + /* Signal that more space is available now. */ > + xenevtchn_notify(xe, device->evtchn); > + > + return size; > +} > + > +static unsigned int put_request_bytes(device *device, unsigned int off, > + unsigned int len) Should this be named put_response_bytes? > +{ > + unsigned int size; > + unsigned int in_data = ring_in_free(device); > + RING_IDX prod, cons; > + > + size = min(len - off, in_data); IIUC, len is the total length of the outgoing data. Maybe total_len would be a better name? I at least read len as just a length for a particular call. Same comment applies to get_request_bytes() if you want to follow it. > + prod = xen_9pfs_mask(device->prod_pvt_in, device->ring_size); > + cons = xen_9pfs_mask(device->intf->in_cons, device->ring_size); > + xen_9pfs_write_packet(device->data.in, device->buffer + off, size, > + &prod, cons, device->ring_size); > + > + xen_wmb(); /* Write data out before setting visible producer. */ > + device->prod_pvt_in += size; > + device->intf->in_prod = device->prod_pvt_in; > + > + return size; > +} > + > static bool io_work_pending(device *device) > { > if ( device->stop_thread ) > return true; > - return false; > + if ( device->error ) > + return false; > + return device->handle_response ? ring_in_free(device) > + : ring_out_data(device); > } > > void *io_thread(void *arg) > { > device *device = arg; > + unsigned int count = 0; > + struct p9_header hdr; > + bool in_hdr = true; > + > + device->max_size = device->ring_size; > + device->buffer = malloc(device->max_size); > + if ( !device->buffer ) > + { > + syslog(LOG_CRIT, "memory allocation failure!"); > + return NULL; > + } > > while ( !device->stop_thread ) > { > @@ -36,9 +127,56 @@ void *io_thread(void *arg) > } > pthread_mutex_unlock(&device->mutex); > > - /* TODO: I/O handling. */ > + if ( device->stop_thread || device->error ) > + continue; > + > + if ( !device->handle_response ) > + { > + if ( in_hdr ) > + { > + count += get_request_bytes(device, count, sizeof(hdr)); > + if ( count != sizeof(hdr) ) > + continue; > + hdr = *(struct p9_header *)device->buffer; > + if ( hdr.size > device->max_size || hdr.size < sizeof(hdr) ) > + { > + syslog(LOG_ERR, "%u.%u specified illegal request length %u", > + device->domid, device->devid, hdr.size); > + device->error = true; When device->error is set, io_thread stops processing requests, but do we want to also tear down this backend? The event channel at least is left in place and unmasked. > + continue; > + } > + in_hdr = false; > + } > + > + count += get_request_bytes(device, count, hdr.size); > + if ( count < hdr.size ) > + continue; > + > + /* TODO: handle request. */ > + > + device->handle_response = true; > + hdr.size = ((struct p9_header *)device->buffer)->size; hdr.size is set during the struct copy above, so this isn't needed? Thanks, Jason
On 02.11.23 19:30, Jason Andryuk wrote: > On Wed, Nov 1, 2023 at 5:34 AM Juergen Gross <jgross@suse.com> wrote: >> >> Add the transport layer of 9pfs. This is basically the infrastructure >> to receive requests from the frontend and to send the related answers >> via the rings. >> >> In order to avoid unaligned accesses e.g. on Arm, add the definition of >> __packed to the common-macros.h header. >> >> Signed-off-by: Juergen Gross <jgross@suse.com> >> --- > >> diff --git a/tools/xenlogd/io.c b/tools/xenlogd/io.c >> index ef0954d69d..590d06e906 100644 >> --- a/tools/xenlogd/io.c >> +++ b/tools/xenlogd/io.c > >> +static unsigned int get_request_bytes(device *device, unsigned int off, >> + unsigned int len) >> +{ >> + unsigned int size; >> + unsigned int out_data = ring_out_data(device); >> + RING_IDX prod, cons; >> + >> + size = min(len - off, out_data); >> + prod = xen_9pfs_mask(device->intf->out_prod, device->ring_size); >> + cons = xen_9pfs_mask(device->cons_pvt_out, device->ring_size); >> + xen_9pfs_read_packet(device->buffer + off, device->data.out, size, >> + prod, &cons, device->ring_size); >> + >> + xen_rmb(); /* Read data out before setting visible consumer. */ >> + device->cons_pvt_out += size; >> + device->intf->out_cons = device->cons_pvt_out; >> + >> + /* Signal that more space is available now. */ >> + xenevtchn_notify(xe, device->evtchn); >> + >> + return size; >> +} >> + >> +static unsigned int put_request_bytes(device *device, unsigned int off, >> + unsigned int len) > > Should this be named put_response_bytes? Yes. > >> +{ >> + unsigned int size; >> + unsigned int in_data = ring_in_free(device); >> + RING_IDX prod, cons; >> + >> + size = min(len - off, in_data); > > IIUC, len is the total length of the outgoing data. Maybe total_len > would be a better name? I at least read len as just a length for a > particular call. Same comment applies to get_request_bytes() if you > want to follow it. Fine with me. > >> + prod = xen_9pfs_mask(device->prod_pvt_in, device->ring_size); >> + cons = xen_9pfs_mask(device->intf->in_cons, device->ring_size); >> + xen_9pfs_write_packet(device->data.in, device->buffer + off, size, >> + &prod, cons, device->ring_size); >> + >> + xen_wmb(); /* Write data out before setting visible producer. */ >> + device->prod_pvt_in += size; >> + device->intf->in_prod = device->prod_pvt_in; >> + >> + return size; >> +} >> + >> static bool io_work_pending(device *device) >> { >> if ( device->stop_thread ) >> return true; >> - return false; >> + if ( device->error ) >> + return false; >> + return device->handle_response ? ring_in_free(device) >> + : ring_out_data(device); >> } >> >> void *io_thread(void *arg) >> { >> device *device = arg; >> + unsigned int count = 0; >> + struct p9_header hdr; >> + bool in_hdr = true; >> + >> + device->max_size = device->ring_size; >> + device->buffer = malloc(device->max_size); >> + if ( !device->buffer ) >> + { >> + syslog(LOG_CRIT, "memory allocation failure!"); >> + return NULL; >> + } >> >> while ( !device->stop_thread ) >> { >> @@ -36,9 +127,56 @@ void *io_thread(void *arg) >> } >> pthread_mutex_unlock(&device->mutex); >> >> - /* TODO: I/O handling. */ >> + if ( device->stop_thread || device->error ) >> + continue; >> + >> + if ( !device->handle_response ) >> + { >> + if ( in_hdr ) >> + { >> + count += get_request_bytes(device, count, sizeof(hdr)); >> + if ( count != sizeof(hdr) ) >> + continue; >> + hdr = *(struct p9_header *)device->buffer; >> + if ( hdr.size > device->max_size || hdr.size < sizeof(hdr) ) >> + { >> + syslog(LOG_ERR, "%u.%u specified illegal request length %u", >> + device->domid, device->devid, hdr.size); >> + device->error = true; > > When device->error is set, io_thread stops processing requests, but do > we want to also tear down this backend? The event channel at least is > left in place and unmasked. I think tearing down the backend shouldn't be done, but you are right that the event channel should be kept masked. > >> + continue; >> + } >> + in_hdr = false; >> + } >> + >> + count += get_request_bytes(device, count, hdr.size); >> + if ( count < hdr.size ) >> + continue; >> + >> + /* TODO: handle request. */ >> + >> + device->handle_response = true; >> + hdr.size = ((struct p9_header *)device->buffer)->size; > > hdr.size is set during the struct copy above, so this isn't needed? It will be needed later when the "TODO" is filled in, as it will then be the response size. Juergen
diff --git a/tools/include/xen-tools/common-macros.h b/tools/include/xen-tools/common-macros.h index e5ed603904..c3fd7d2a30 100644 --- a/tools/include/xen-tools/common-macros.h +++ b/tools/include/xen-tools/common-macros.h @@ -79,6 +79,10 @@ #define __must_check __attribute__((__warn_unused_result__)) #endif +#ifndef __packed +#define __packed __attribute__((__packed__)) +#endif + #define container_of(ptr, type, member) ({ \ typeof(((type *)0)->member) *mptr__ = (ptr); \ (type *)((char *)mptr__ - offsetof(type, member)); \ diff --git a/tools/xenlogd/io.c b/tools/xenlogd/io.c index ef0954d69d..590d06e906 100644 --- a/tools/xenlogd/io.c +++ b/tools/xenlogd/io.c @@ -6,24 +6,115 @@ * Copyright (C) 2023 Juergen Gross <jgross@suse.com> * * I/O thread handling. + * + * Only handle one request at a time, pushing out the complete response + * before looking for the next request. */ #include <stdbool.h> +#include <stdlib.h> #include <string.h> #include <syslog.h> +#include <xenctrl.h> /* For cpu barriers. */ +#include <xen-tools/common-macros.h> #include "xenlogd.h" +/* + * Note that the ring names "in" and "out" are from the frontend's + * perspective, so the "in" ring will be used for responses to the frontend, + * while the "out" ring is used for requests from the frontend to the + * backend. + */ +static unsigned int ring_in_free(device *device) +{ + unsigned int queued; + + queued = xen_9pfs_queued(device->prod_pvt_in, device->intf->in_cons, + device->ring_size); + xen_rmb(); + + return device->ring_size - queued; +} + +static unsigned int ring_out_data(device *device) +{ + unsigned int queued; + + queued = xen_9pfs_queued(device->intf->out_prod, device->cons_pvt_out, + device->ring_size); + xen_rmb(); + + return queued; +} + +static unsigned int get_request_bytes(device *device, unsigned int off, + unsigned int len) +{ + unsigned int size; + unsigned int out_data = ring_out_data(device); + RING_IDX prod, cons; + + size = min(len - off, out_data); + prod = xen_9pfs_mask(device->intf->out_prod, device->ring_size); + cons = xen_9pfs_mask(device->cons_pvt_out, device->ring_size); + xen_9pfs_read_packet(device->buffer + off, device->data.out, size, + prod, &cons, device->ring_size); + + xen_rmb(); /* Read data out before setting visible consumer. */ + device->cons_pvt_out += size; + device->intf->out_cons = device->cons_pvt_out; + + /* Signal that more space is available now. */ + xenevtchn_notify(xe, device->evtchn); + + return size; +} + +static unsigned int put_request_bytes(device *device, unsigned int off, + unsigned int len) +{ + unsigned int size; + unsigned int in_data = ring_in_free(device); + RING_IDX prod, cons; + + size = min(len - off, in_data); + prod = xen_9pfs_mask(device->prod_pvt_in, device->ring_size); + cons = xen_9pfs_mask(device->intf->in_cons, device->ring_size); + xen_9pfs_write_packet(device->data.in, device->buffer + off, size, + &prod, cons, device->ring_size); + + xen_wmb(); /* Write data out before setting visible producer. */ + device->prod_pvt_in += size; + device->intf->in_prod = device->prod_pvt_in; + + return size; +} + static bool io_work_pending(device *device) { if ( device->stop_thread ) return true; - return false; + if ( device->error ) + return false; + return device->handle_response ? ring_in_free(device) + : ring_out_data(device); } void *io_thread(void *arg) { device *device = arg; + unsigned int count = 0; + struct p9_header hdr; + bool in_hdr = true; + + device->max_size = device->ring_size; + device->buffer = malloc(device->max_size); + if ( !device->buffer ) + { + syslog(LOG_CRIT, "memory allocation failure!"); + return NULL; + } while ( !device->stop_thread ) { @@ -36,9 +127,56 @@ void *io_thread(void *arg) } pthread_mutex_unlock(&device->mutex); - /* TODO: I/O handling. */ + if ( device->stop_thread || device->error ) + continue; + + if ( !device->handle_response ) + { + if ( in_hdr ) + { + count += get_request_bytes(device, count, sizeof(hdr)); + if ( count != sizeof(hdr) ) + continue; + hdr = *(struct p9_header *)device->buffer; + if ( hdr.size > device->max_size || hdr.size < sizeof(hdr) ) + { + syslog(LOG_ERR, "%u.%u specified illegal request length %u", + device->domid, device->devid, hdr.size); + device->error = true; + continue; + } + in_hdr = false; + } + + count += get_request_bytes(device, count, hdr.size); + if ( count < hdr.size ) + continue; + + /* TODO: handle request. */ + + device->handle_response = true; + hdr.size = ((struct p9_header *)device->buffer)->size; + count = 0; + } + + if ( device->handle_response ) + { + count += put_request_bytes(device, count, hdr.size); + + if ( count == hdr.size ) + { + /* Signal presence of response. */ + xenevtchn_notify(xe, device->evtchn); + + device->handle_response = false; + in_hdr = true; + count = 0; + } + } } + free(device->buffer); + device->thread_active = false; return NULL; diff --git a/tools/xenlogd/xenlogd.h b/tools/xenlogd/xenlogd.h index a8e9f9ca22..40aa7cd03a 100644 --- a/tools/xenlogd/xenlogd.h +++ b/tools/xenlogd/xenlogd.h @@ -13,6 +13,12 @@ #define MAX_OPEN_FILES_DEFAULT 5 +struct p9_header { + uint32_t size; + uint8_t cmd; + uint16_t tag; +} __attribute__((packed)); + typedef struct device device; struct device { /* Admin data. */ @@ -40,7 +46,17 @@ struct device { struct xen_9pfs_data_intf *intf; unsigned int ring_order; RING_IDX ring_size; + + /* Transport layer data. */ struct xen_9pfs_data data; + RING_IDX prod_pvt_in; + RING_IDX cons_pvt_out; + + /* Request and response handling. */ + uint32_t max_size; + bool error; /* Protocol error - stop processing. */ + bool handle_response; /* Main loop now handling response. */ + void *buffer; /* Request/response buffer. */ }; extern xenevtchn_handle *xe;
Add the transport layer of 9pfs. This is basically the infrastructure to receive requests from the frontend and to send the related answers via the rings. In order to avoid unaligned accesses e.g. on Arm, add the definition of __packed to the common-macros.h header. Signed-off-by: Juergen Gross <jgross@suse.com> --- tools/include/xen-tools/common-macros.h | 4 + tools/xenlogd/io.c | 142 +++++++++++++++++++++++- tools/xenlogd/xenlogd.h | 16 +++ 3 files changed, 160 insertions(+), 2 deletions(-)