/*
* Copyright 2000, International Business Machines Corporation and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
*
- * Portions Copyright (c) 2006-2008 Sine Nomine Associates
+ * Portions Copyright (c) 2006-2010 Sine Nomine Associates
*/
/*
#include <afsconfig.h>
#include <afs/param.h>
+#include <roken.h>
#include <sys/types.h>
#include <stdio.h>
+#ifdef HAVE_STDINT_H
+# include <stdint.h>
+#endif
#ifdef AFS_NT40_ENV
#include <winsock2.h>
#include <time.h>
#include <unistd.h>
#endif
#include <errno.h>
-#ifdef AFS_PTHREAD_ENV
-#include <assert.h>
-#else /* AFS_PTHREAD_ENV */
-#include <afs/assert.h>
-#endif /* AFS_PTHREAD_ENV */
+#include <afs/afs_assert.h>
#include <signal.h>
#include <string.h>
#include "nfs.h"
#include <afs/errors.h>
#include "daemon_com.h"
+#include "daemon_com_inline.h"
#include "fssync.h"
+#include "fssync_inline.h"
+#include "salvsync.h"
#include "lwp.h"
#include "lock.h"
#include <afs/afssyscalls.h>
#include "volume.h"
#include "volume_inline.h"
#include "partition.h"
+#include "vg_cache.h"
+#include "common.h"
#ifdef HAVE_POLL
#include <sys/poll.h>
#ifdef FSSYNC_BUILD_SERVER
-/*@printflike@*/ extern void Log(const char *format, ...);
-
int (*V_BreakVolumeCallbacks) (VolumeId volume);
#define MAXHANDLERS 4 /* Up to 4 clients; must be at least 2, so that
/**
* fssync server socket handle.
*/
-static SYNC_server_state_t fssync_server_state =
- { -1, /* file descriptor */
+static SYNC_server_state_t fssync_server_state =
+ { OSI_NULLSOCKET, /* file descriptor */
FSSYNC_ENDPOINT_DECL, /* server endpoint */
FSYNC_PROTO_VERSION, /* protocol version */
5, /* bind() retry limit */
"FSSYNC", /* protocol name string */
};
+#ifdef AFS_DEMAND_ATTACH_FS
+/**
+ * a queue of volume pointers to salvage in the background.
+ */
+struct fsync_salv_node {
+ struct rx_queue q;
+ Volume *vp; /**< volume to salvage */
+ unsigned char update_salv_prio; /**< whether we should update the salvage priority or not */
+};
+static struct {
+ struct rx_queue head;
+ pthread_cond_t cv;
+} fsync_salv;
+
+static void * FSYNC_salvageThread(void *);
+static void FSYNC_backgroundSalvage(Volume *vp);
+#endif /* AFS_DEMAND_ATTACH_FS */
/* Forward declarations */
static void * FSYNC_sync(void *);
static afs_int32 FSYNC_com_VolQuery(FSSYNC_VolOp_command * com, SYNC_response * res);
static afs_int32 FSYNC_com_VolHdrQuery(FSSYNC_VolOp_command * com, SYNC_response * res);
#ifdef AFS_DEMAND_ATTACH_FS
+static afs_int32 FSYNC_com_VGUpdate(osi_socket fd, SYNC_command * com, SYNC_response * res);
static afs_int32 FSYNC_com_VolOpQuery(FSSYNC_VolOp_command * com, SYNC_response * res);
+static afs_int32 FSYNC_com_VGQuery(FSSYNC_VolOp_command * com, SYNC_response * res);
+static afs_int32 FSYNC_com_VGScan(FSSYNC_VolOp_command * com, SYNC_response * res);
+static afs_int32 FSYNC_com_VGScanAll(FSSYNC_VolOp_command * com, SYNC_response * res);
#endif /* AFS_DEMAND_ATTACH_FS */
static afs_int32 FSYNC_com_VnQry(osi_socket fd, SYNC_command * com, SYNC_response * res);
Lock_Init(&FSYNC_handler_lock);
#ifdef AFS_PTHREAD_ENV
- assert(pthread_attr_init(&tattr) == 0);
- assert(pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_DETACHED) == 0);
- assert(pthread_create(&tid, &tattr, FSYNC_sync, NULL) == 0);
+ osi_Assert(pthread_attr_init(&tattr) == 0);
+ osi_Assert(pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_DETACHED) == 0);
+ osi_Assert(pthread_create(&tid, &tattr, FSYNC_sync, NULL) == 0);
#else /* AFS_PTHREAD_ENV */
- assert(LWP_CreateProcess
+ osi_Assert(LWP_CreateProcess
(FSYNC_sync, USUAL_STACK_SIZE, USUAL_PRIORITY, (void *)0,
"FSYNC_sync", &pid) == LWP_SUCCESS);
#endif /* AFS_PTHREAD_ENV */
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ queue_Init(&fsync_salv.head);
+ CV_INIT(&fsync_salv.cv, "fsync salv", CV_DEFAULT, 0);
+ osi_Assert(pthread_create(&tid, &tattr, FSYNC_salvageThread, NULL) == 0);
+#endif /* AFS_DEMAND_ATTACH_FS */
}
#if defined(HAVE_POLL) && defined(AFS_PTHREAD_ENV)
SYNC_server_state_t * state = &fssync_server_state;
#ifdef AFS_DEMAND_ATTACH_FS
VThreadOptions_t * thread_opts;
-#endif
+ int min_vinit = 2;
+#else
+ /*
+ * For non-DAFS, only wait until we begin attaching volumes (instead
+ * of waiting until all volumes are attached), since it can take
+ * awhile until VInit == 2.
+ */
+ int min_vinit = 1;
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ /* we must not be called before vol package initialization, since we use
+ * vol package mutexes and conds etc */
+ osi_Assert(VInit);
SYNC_getAddr(&state->endpoint, &state->addr);
SYNC_cleanupSock(state);
#ifdef AFS_PTHREAD_ENV
/* set our 'thread-id' so that the host hold table works */
- MUTEX_ENTER(&rx_stats_mutex); /* protects rxi_pthread_hinum */
- tid = ++rxi_pthread_hinum;
- MUTEX_EXIT(&rx_stats_mutex);
+ tid = rx_NewThreadId();
pthread_setspecific(rx_thread_id_key, (void *)(intptr_t)tid);
Log("Set thread id %d for FSYNC_sync\n", tid);
#endif /* AFS_PTHREAD_ENV */
- while (!VInit) {
- /* Let somebody else run until level > 0. That doesn't mean that
- * all volumes have been attached. */
+ VOL_LOCK;
+
+ while (VInit < min_vinit) {
+ /* Let somebody else run until all volumes have been preattached
+ * (DAFS), or we have started attaching volumes (non-DAFS). This
+ * doesn't mean that all volumes have been attached.
+ */
#ifdef AFS_PTHREAD_ENV
- pthread_yield();
+ VOL_CV_WAIT(&vol_vinit_cond);
#else /* AFS_PTHREAD_ENV */
LWP_DispatchProcess();
#endif /* AFS_PTHREAD_ENV */
}
+
+ VOL_UNLOCK;
+
state->fd = SYNC_getSock(&state->endpoint);
code = SYNC_bindSock(state);
- assert(!code);
+ osi_Assert(!code);
#ifdef AFS_DEMAND_ATTACH_FS
/*
}
memcpy(thread_opts, &VThread_defaults, sizeof(VThread_defaults));
thread_opts->disallow_salvsync = 1;
- assert(pthread_setspecific(VThread_key, thread_opts) == 0);
+ osi_Assert(pthread_setspecific(VThread_key, thread_opts) == 0);
+
+ code = VVGCache_PkgInit();
+ osi_Assert(code == 0);
#endif
InitHandler();
return NULL; /* hush now, little gcc */
}
+#ifdef AFS_DEMAND_ATTACH_FS
+/**
+ * thread for salvaging volumes in the background.
+ *
+ * Since FSSYNC handlers cannot issue SALVSYNC requests in order to avoid
+ * deadlock issues, this thread exists so code in the FSSYNC handler thread
+ * can hand off volumes to be salvaged in the background.
+ *
+ * @param[in] args unused
+ *
+ * @note DEMAND_ATTACH_FS only
+ */
+static void *
+FSYNC_salvageThread(void * args)
+{
+ Volume *vp;
+ struct fsync_salv_node *node;
+
+ VOL_LOCK;
+
+ for (;;) {
+ while (queue_IsEmpty(&fsync_salv.head)) {
+ VOL_CV_WAIT(&fsync_salv.cv);
+ }
+
+ node = queue_First(&fsync_salv.head, fsync_salv_node);
+ queue_Remove(node);
+
+ vp = node->vp;
+ if (node->update_salv_prio) {
+ if (VUpdateSalvagePriority_r(vp)) {
+ ViceLog(0, ("FSYNC_salvageThread: unable to raise salvage priority "
+ "for volume %lu\n", afs_printable_uint32_lu(vp->hashid)));
+ }
+ }
+
+ free(node);
+ node = NULL;
+
+ VCancelReservation_r(vp);
+ }
+
+ VOL_UNLOCK;
+
+ return NULL;
+}
+
+/**
+ * salvage a volume in the background.
+ *
+ * Salvages cannot be scheduled directly from the main FSYNC thread, so
+ * instead call this function to schedule a salvage asynchronously in the
+ * FSYNC_salvageThread thread.
+ *
+ * @param[in] vp volume to pointer to salvage
+ *
+ * @pre VOL_LOCK held
+ *
+ * @note DEMAND_ATTACH_FS only
+ */
+static void
+FSYNC_backgroundSalvage(Volume *vp)
+{
+ struct fsync_salv_node *node;
+ Error ec;
+
+ VCreateReservation_r(vp);
+
+ node = malloc(sizeof(struct fsync_salv_node));
+ node->vp = vp;
+
+ /* Save this value, to know if we should VUpdateSalvagePriority_r.
+ * We need to save it here, snce VRequestSalvage_r will change it. */
+ node->update_salv_prio = vp->salvage.requested;
+
+ if (VRequestSalvage_r(&ec, vp, SALVSYNC_ERROR, 0)) {
+ ViceLog(0, ("FSYNC_backgroundSalvage: unable to request salvage for volume %lu\n",
+ afs_printable_uint32_lu(vp->hashid)));
+ }
+
+ queue_Append(&fsync_salv.head, node);
+ CV_BROADCAST(&fsync_salv.cv);
+}
+#endif /* AFS_DEMAND_ATTACH_FS */
+
static void
FSYNC_newconnection(osi_socket afd)
{
socklen_t junk;
junk = sizeof(other);
fd = accept(afd, (struct sockaddr *)&other, &junk);
- if (fd == -1) {
+ if (fd == OSI_NULLSOCKET) {
Log("FSYNC_newconnection: accept failed, errno==%d\n", errno);
- assert(1 == 2);
+ osi_Assert(1 == 2);
} else if (!AddHandler(fd, FSYNC_com)) {
AcceptOff();
- assert(AddHandler(fd, FSYNC_com));
+ osi_Assert(AddHandler(fd, FSYNC_com));
}
}
goto done;
}
+ ViceLog(125, ("FSYNC_com: from fd %d got command %ld (%s) reason %ld (%s) "
+ "pt %ld (%s) pid %ld\n", (int)fd,
+ afs_printable_int32_ld(com.hdr.command),
+ FSYNC_com2string(com.hdr.command),
+ afs_printable_int32_ld(com.hdr.reason),
+ FSYNC_reason2string(com.hdr.reason),
+ afs_printable_int32_ld(com.hdr.programType),
+ VPTypeToString(com.hdr.programType),
+ afs_printable_int32_ld(com.hdr.pid)));
+
res.hdr.com_seq = com.hdr.com_seq;
VOL_LOCK;
case FSYNC_VOL_DONE:
case FSYNC_VOL_QUERY:
case FSYNC_VOL_QUERY_HDR:
+#ifdef AFS_DEMAND_ATTACH_FS
case FSYNC_VOL_QUERY_VOP:
+ case FSYNC_VG_QUERY:
+ case FSYNC_VG_SCAN:
+ case FSYNC_VG_SCAN_ALL:
+#endif
res.hdr.response = FSYNC_com_VolOp(fd, &com, &res);
break;
case FSYNC_VOL_STATS_GENERAL:
case FSYNC_VOL_QUERY_VNODE:
res.hdr.response = FSYNC_com_VnQry(fd, &com, &res);
break;
+#ifdef AFS_DEMAND_ATTACH_FS
+ case FSYNC_VG_ADD:
+ case FSYNC_VG_DEL:
+ res.hdr.response = FSYNC_com_VGUpdate(fd, &com, &res);
+ break;
+#endif
default:
res.hdr.response = SYNC_BAD_COMMAND;
break;
}
VOL_UNLOCK;
+ ViceLog(125, ("FSYNC_com: fd %d responding with code %ld (%s) reason %ld "
+ "(%s)\n", (int)fd,
+ afs_printable_int32_ld(res.hdr.response),
+ SYNC_res2string(res.hdr.response),
+ afs_printable_int32_ld(res.hdr.reason),
+ FSYNC_reason2string(res.hdr.reason)));
+
respond:
SYNC_putRes(&fssync_server_state, fd, &res);
}
}
+ ViceLog(125, ("FSYNC_com_VolOp: fd %d got command for vol %lu part %.16s\n",
+ (int)fd, afs_printable_uint32_lu(vcom.vop->volume),
+ vcom.vop->partName));
+
switch (com->hdr.command) {
case FSYNC_VOL_ON:
case FSYNC_VOL_ATTACH:
case FSYNC_VOL_QUERY_VOP:
code = FSYNC_com_VolOpQuery(&vcom, res);
break;
+ case FSYNC_VG_QUERY:
+ code = FSYNC_com_VGQuery(&vcom, res);
+ break;
+ case FSYNC_VG_SCAN:
+ code = FSYNC_com_VGScan(&vcom, res);
+ break;
+ case FSYNC_VG_SCAN_ALL:
+ code = FSYNC_com_VGScanAll(&vcom, res);
+ break;
#endif /* AFS_DEMAND_ATTACH_FS */
default:
code = SYNC_BAD_COMMAND;
FSYNC_com_VolOn(FSSYNC_VolOp_command * vcom, SYNC_response * res)
{
afs_int32 code = SYNC_OK;
+#ifndef AFS_DEMAND_ATTACH_FS
char tvolName[VMAXPATHLEN];
+#endif
Volume * vp;
Error error;
vp = VLookupVolume_r(&error, vcom->vop->volume, NULL);
if (vp &&
FSYNC_partMatch(vcom, vp, 1) &&
- vp->pending_vol_op &&
+ vp->pending_vol_op &&
(vcom->hdr->programType != vp->pending_vol_op->com.programType)) {
/* a different program has this volume checked out. deny. */
Log("FSYNC_VolOn: WARNING: program type %u has attempted to manipulate "
- "state for volume %u using command code %u while the volume is "
+ "state for volume %u using command code %u while the volume is "
"checked out by program type %u for command code %u.\n",
vcom->hdr->programType,
vcom->vop->volume,
if (vcom->hdr->command == FSYNC_VOL_LEAVE_OFF) {
/* nothing much to do if we're leaving the volume offline */
#ifdef AFS_DEMAND_ATTACH_FS
- if (vp) {
+ if (vp && V_attachState(vp) != VOL_STATE_DELETED) {
if (FSYNC_partMatch(vcom, vp, 1)) {
if ((V_attachState(vp) == VOL_STATE_UNATTACHED) ||
(V_attachState(vp) == VOL_STATE_PREATTACHED)) {
res->hdr.reason = FSYNC_WRONG_PART;
}
} else {
- code = SYNC_DENIED;
+ code = SYNC_FAILED;
res->hdr.reason = FSYNC_UNKNOWN_VOLID;
}
#endif
VDeregisterVolOp_r(vp);
}
#else /* !AFS_DEMAND_ATTACH_FS */
- tvolName[0] = '/';
+ tvolName[0] = OS_DIRSEPC;
snprintf(&tvolName[1], sizeof(tvolName)-1, VFORMAT, afs_printable_uint32_lu(vcom->vop->volume));
tvolName[sizeof(tvolName)-1] = '\0';
* @note this is an FSYNC RPC server stub
*
* @note this procedure handles the following FSSYNC command codes:
- * - FSYNC_VOL_OFF
+ * - FSYNC_VOL_OFF
* - FSYNC_VOL_NEEDVOLUME
*
* @note the supplementary reason code contains additional details.
Volume * vp;
Error error;
#ifdef AFS_DEMAND_ATTACH_FS
- int reserved = 0;
- Volume *nvp;
+ Volume *nvp, *rvp = NULL;
#endif
if (SYNC_verifyProtocolString(vcom->vop->partName, sizeof(vcom->vop->partName))) {
/* enforce mutual exclusion for volume ops */
if (vp->pending_vol_op) {
if (vp->pending_vol_op->com.programType != type) {
+ if (vp->pending_vol_op->com.command == FSYNC_VOL_OFF &&
+ vp->pending_vol_op->com.reason == FSYNC_SALVAGE) {
+
+ Log("denying offline request for volume %lu; volume is salvaging\n",
+ afs_printable_uint32_lu(vp->hashid));
+
+ res->hdr.reason = FSYNC_SALVAGE;
+ goto deny;
+ }
Log("volume %u already checked out\n", vp->hashid);
/* XXX debug */
Log("vp->vop = { com = { ver=%u, prog=%d, com=%d, reason=%d, len=%u, flags=0x%x }, vop = { vol=%u, part='%s' } }\n",
- vp->pending_vol_op->com.proto_version,
+ vp->pending_vol_op->com.proto_version,
vp->pending_vol_op->com.programType,
vp->pending_vol_op->com.command,
vp->pending_vol_op->com.reason,
}
}
+ /* wait for exclusive ops, so we have an accurate picture of the
+ * vol attach state */
+ VCreateReservation_r(vp);
+ VWaitExclusiveState_r(vp);
+ rvp = vp;
+
/* filter based upon requestor
*
- * volume utilities are not allowed to check out volumes
- * which are in an error state
+ * volume utilities / volserver are not allowed to check out
+ * volumes which are in an error state
*
* unknown utility programs will be denied on principal
*/
switch (type) {
case salvageServer:
- /* it is possible for the salvageserver to checkout a
+ /* it is possible for the salvageserver to checkout a
* volume for salvage before its scheduling request
* has been sent to the salvageserver */
if (vp->salvage.requested && !vp->salvage.scheduled) {
vp->salvage.scheduled = 1;
}
+
+ /* If the volume is in VOL_STATE_SALVAGE_REQ, we need to wait
+ * for the vol to go offline before we can give it away. Also
+ * make sure we don't come out with vp in an excl state. */
+ while (V_attachState(vp) == VOL_STATE_SALVAGE_REQ ||
+ VIsExclusiveState(V_attachState(vp))) {
+
+ VOL_CV_WAIT(&V_attachCV(vp));
+ }
+
case debugUtility:
break;
case volumeUtility:
+ case volumeServer:
+ if (VIsSalvaging(vp)) {
+ Log("denying offline request for volume %lu; volume is in salvaging state\n",
+ afs_printable_uint32_lu(vp->hashid));
+ res->hdr.reason = FSYNC_SALVAGE;
+
+ /* the volume hasn't been checked out yet by the salvager,
+ * but we think the volume is salvaging; schedule a
+ * a salvage to update the salvage priority */
+ FSYNC_backgroundSalvage(vp);
+
+ goto deny;
+ }
if (VIsErrorState(V_attachState(vp))) {
goto deny;
}
- if (vp->salvage.requested) {
- goto deny;
- }
break;
default:
case VOL_STATE_PREATTACHED:
case VOL_STATE_SALVAGING:
case VOL_STATE_ERROR:
+ case VOL_STATE_DELETED:
/* register the volume operation metadata with the volume
*
* if the volume is currently pre-attached, attach2()
/* convert to heavyweight ref */
nvp = VGetVolumeByVp_r(&error, vp);
-
if (!nvp) {
- Log("FSYNC_com_VolOff: failed to get heavyweight reference to volume %u\n",
- vcom->vop->volume);
+ /*
+ * It's possible for VGetVolumeByVp_r to have dropped and
+ * re-acquired VOL_LOCK, so volume state may have changed
+ * back to one of the states we tested for above. Since
+ * GetVolume can return NULL in some of those states, just
+ * test for the states again here.
+ */
+ switch (V_attachState(vp)) {
+ case VOL_STATE_UNATTACHED:
+ case VOL_STATE_PREATTACHED:
+ case VOL_STATE_SALVAGING:
+ case VOL_STATE_ERROR:
+ case VOL_STATE_DELETED:
+ /* register the volume operation metadata with the volume
+ *
+ * if the volume is currently pre-attached, attach2()
+ * will evaluate the vol op metadata to determine whether
+ * attaching the volume would be safe */
+ VRegisterVolOp_r(vp, &info);
+ vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningUnknown;
+ goto done;
+ default:
+ break;
+ }
+
+ Log("FSYNC_com_VolOff: failed to get heavyweight reference to volume %u (state=%u, flags=0x%x)\n",
+ vcom->vop->volume, V_attachState(vp), V_attachFlags(vp));
res->hdr.reason = FSYNC_VOL_PKG_ERROR;
goto deny;
} else if (nvp != vp) {
vp = nvp;
}
+ /* kill off lightweight ref to ensure we can't deadlock against ourselves later... */
+ VCancelReservation_r(rvp);
+ rvp = NULL;
+
/* register the volume operation metadata with the volume */
VRegisterVolOp_r(vp, &info);
if (VVolOpLeaveOnline_r(vp, &info)) {
VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT); /* At least get volume stats right */
if (LogLevel) {
- Log("FSYNC: Volume %u (%s) was left on line for an external %s request\n",
- V_id(vp), V_name(vp),
- vcom->hdr->reason == V_CLONE ? "clone" :
- vcom->hdr->reason == V_READONLY ? "readonly" :
- vcom->hdr->reason == V_DUMP ? "dump" :
+ Log("FSYNC: Volume %u (%s) was left on line for an external %s request\n",
+ V_id(vp), V_name(vp),
+ vcom->hdr->reason == V_CLONE ? "clone" :
+ vcom->hdr->reason == V_READONLY ? "readonly" :
+ vcom->hdr->reason == V_DUMP ? "dump" :
+ vcom->hdr->reason == FSYNC_SALVAGE ? "salvage" :
"UNKNOWN");
}
#ifdef AFS_DEMAND_ATTACH_FS
#ifdef AFS_DEMAND_ATTACH_FS
VOfflineForVolOp_r(&error, vp, "A volume utility is running.");
if (error==0) {
- assert(vp->nUsers==0);
- vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
+ osi_Assert(vp->nUsers==0);
+ vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
}
else {
VDeregisterVolOp_r(vp);
vp = NULL;
}
}
+ goto done;
+
+ deny:
+ code = SYNC_DENIED;
done:
+#ifdef AFS_DEMAND_ATTACH_FS
+ if (rvp) {
+ VCancelReservation_r(rvp);
+ }
+#endif
return code;
-
- deny:
- return SYNC_DENIED;
}
/**
FSYNC_com_VolDone(FSSYNC_VolOp_command * vcom, SYNC_response * res)
{
afs_int32 code = SYNC_FAILED;
-#ifdef AFS_DEMAND_ATTACH_FS
Error error;
Volume * vp;
-#endif
if (SYNC_verifyProtocolString(vcom->vop->partName, sizeof(vcom->vop->partName))) {
res->hdr.reason = SYNC_REASON_MALFORMED_PACKET;
if (vcom->v)
vcom->v->volumeID = 0;
-#ifdef AFS_DEMAND_ATTACH_FS
vp = VLookupVolume_r(&error, vcom->vop->volume, NULL);
if (vp) {
if (FSYNC_partMatch(vcom, vp, 1)) {
+#ifdef AFS_DEMAND_ATTACH_FS
if ((V_attachState(vp) == VOL_STATE_UNATTACHED) ||
(V_attachState(vp) == VOL_STATE_PREATTACHED)) {
- VChangeState_r(vp, VOL_STATE_UNATTACHED);
+
+ /* Change state to DELETED, not UNATTACHED, so clients get
+ * a VNOVOL error when they try to access from now on. */
+
+ VChangeState_r(vp, VOL_STATE_DELETED);
VDeregisterVolOp_r(vp);
+
+ /* Someday we should free the vp, too, after about 2 hours,
+ * possibly by putting the vp back on the VLRU. */
+
code = SYNC_OK;
} else {
code = SYNC_DENIED;
res->hdr.reason = FSYNC_BAD_STATE;
}
+#else /* AFS_DEMAND_ATTACH_FS */
+ if (!vp->specialStatus) {
+ vp->specialStatus = VNOVOL;
+ }
+ code = SYNC_OK;
+#endif /* !AFS_DEMAND_ATTACH_FS */
} else {
code = SYNC_OK; /* XXX is this really a good idea? */
res->hdr.reason = FSYNC_WRONG_PART;
} else {
res->hdr.reason = FSYNC_UNKNOWN_VOLID;
}
-#endif
done:
return code;
}
vp = VLookupVolume_r(&error, vcom->vop->volume, NULL);
+
+ if (!vp && vcom->hdr->reason == FSYNC_SALVAGE) {
+ /* The requested volume doesn't seem to exist. However, it is possible
+ * that this is triggered by trying to create or clone a volume that
+ * was prevented from succeeding by a half-created volume in the way.
+ * (e.g. we tried to create volume X, but volume X exists except that
+ * its .vol header was deleted for some reason) So, still try to
+ * a salvage for that volume ID. */
+
+ Log("FSYNC_com_VolError: attempting to schedule salvage for unknown "
+ "volume %lu part %s\n", afs_printable_uint32_lu(vcom->vop->volume),
+ vcom->vop->partName);
+ vp = VPreAttachVolumeById_r(&error, vcom->vop->partName,
+ vcom->vop->volume);
+ }
+
if (vp) {
if (FSYNC_partMatch(vcom, vp, 0)) {
/* null out salvsync control state, as it's no longer relevant */
memset(&vp->salvage, 0, sizeof(vp->salvage));
- VChangeState_r(vp, VOL_STATE_ERROR);
+ VDeregisterVolOp_r(vp);
+
+ if (vcom->hdr->reason == FSYNC_SALVAGE) {
+ FSYNC_backgroundSalvage(vp);
+ } else {
+ VChangeState_r(vp, VOL_STATE_ERROR);
+ }
+
code = SYNC_OK;
} else {
res->hdr.reason = FSYNC_WRONG_PART;
vp = VLookupVolume_r(&error, vcom->vop->volume, NULL);
if (vp && vp->pending_vol_op) {
- assert(sizeof(FSSYNC_VolOp_info) <= res->payload.len);
+ osi_Assert(sizeof(FSSYNC_VolOp_info) <= res->payload.len);
memcpy(res->payload.buf, vp->pending_vol_op, sizeof(FSSYNC_VolOp_info));
res->hdr.response_len += sizeof(FSSYNC_VolOp_info);
} else {
- if (vp) {
- res->hdr.reason = FSYNC_NO_PENDING_VOL_OP;
- } else {
+ if (!vp || V_attachState(vp) == VOL_STATE_DELETED) {
res->hdr.reason = FSYNC_UNKNOWN_VOLID;
+ } else if (!FSYNC_partMatch(vcom, vp, 1)) {
+ res->hdr.reason = FSYNC_WRONG_PART;
+ } else {
+ res->hdr.reason = FSYNC_NO_PENDING_VOL_OP;
}
code = SYNC_FAILED;
}
return code;
}
+
+static afs_int32
+FSYNC_com_VGQuery(FSSYNC_VolOp_command * vcom, SYNC_response * res)
+{
+ afs_int32 code = SYNC_FAILED;
+ int rc;
+ struct DiskPartition64 * dp;
+
+ if (SYNC_verifyProtocolString(vcom->vop->partName, sizeof(vcom->vop->partName))) {
+ res->hdr.reason = SYNC_REASON_MALFORMED_PACKET;
+ goto done;
+ }
+
+ dp = VGetPartition_r(vcom->vop->partName, 0);
+ if (dp == NULL) {
+ res->hdr.reason = FSYNC_BAD_PART;
+ goto done;
+ }
+
+ osi_Assert(sizeof(FSSYNC_VGQry_response_t) <= res->payload.len);
+
+ rc = VVGCache_query_r(dp, vcom->vop->volume, res->payload.buf);
+ switch (rc) {
+ case 0:
+ res->hdr.response_len += sizeof(FSSYNC_VGQry_response_t);
+ code = SYNC_OK;
+ break;
+ case EAGAIN:
+ res->hdr.reason = FSYNC_PART_SCANNING;
+ break;
+ case ENOENT:
+ res->hdr.reason = FSYNC_UNKNOWN_VOLID;
+ break;
+ default:
+ break;
+ }
+
+ done:
+ return code;
+}
+
+static afs_int32
+FSYNC_com_VGUpdate(osi_socket fd, SYNC_command * com, SYNC_response * res)
+{
+ afs_int32 code = SYNC_FAILED;
+ struct DiskPartition64 * dp;
+ FSSYNC_VGUpdate_command_t * vgucom;
+ int rc;
+
+ if (com->recv_len != (sizeof(com->hdr) + sizeof(*vgucom))) {
+ res->hdr.reason = SYNC_REASON_MALFORMED_PACKET;
+ res->hdr.flags |= SYNC_FLAG_CHANNEL_SHUTDOWN;
+ code = SYNC_COM_ERROR;
+ goto done;
+ }
+
+ vgucom = com->payload.buf;
+
+ ViceLog(125, ("FSYNC_com_VGUpdate: fd %d got command for parent %lu child "
+ "%lu partName %.16s\n", (int)fd,
+ afs_printable_uint32_lu(vgucom->parent),
+ afs_printable_uint32_lu(vgucom->child),
+ vgucom->partName));
+
+ if (SYNC_verifyProtocolString(vgucom->partName, sizeof(vgucom->partName))) {
+ res->hdr.reason = SYNC_REASON_MALFORMED_PACKET;
+ goto done;
+ }
+
+ dp = VGetPartition_r(vgucom->partName, 0);
+ if (dp == NULL) {
+ res->hdr.reason = FSYNC_BAD_PART;
+ goto done;
+ }
+
+ switch(com->hdr.command) {
+ case FSYNC_VG_ADD:
+ rc = VVGCache_entry_add_r(dp, vgucom->parent, vgucom->child, NULL);
+ break;
+
+ case FSYNC_VG_DEL:
+ rc = VVGCache_entry_del_r(dp, vgucom->parent, vgucom->child);
+ break;
+
+ default:
+ Log("FSYNC_com_VGUpdate called improperly\n");
+ rc = -1;
+ break;
+ }
+
+ /* EINVAL means the partition VGC doesn't exist at all; not really
+ * an error */
+ if (rc == 0 || rc == EINVAL) {
+ code = SYNC_OK;
+ }
+
+ if (rc == ENOENT) {
+ res->hdr.reason = FSYNC_UNKNOWN_VOLID;
+ } else {
+ res->hdr.reason = FSYNC_WHATEVER;
+ }
+
+ done:
+ return code;
+}
+
+static afs_int32
+FSYNC_com_VGScan(FSSYNC_VolOp_command * vcom, SYNC_response * res)
+{
+ afs_int32 code = SYNC_FAILED;
+ struct DiskPartition64 * dp;
+
+ if (SYNC_verifyProtocolString(vcom->vop->partName, sizeof(vcom->vop->partName))) {
+ res->hdr.reason = SYNC_REASON_MALFORMED_PACKET;
+ goto done;
+ }
+
+ dp = VGetPartition_r(vcom->vop->partName, 0);
+ if (dp == NULL) {
+ res->hdr.reason = FSYNC_BAD_PART;
+ goto done;
+ }
+
+ if (VVGCache_scanStart_r(dp) == 0) {
+ code = SYNC_OK;
+ }
+
+ done:
+ return code;
+}
+
+static afs_int32
+FSYNC_com_VGScanAll(FSSYNC_VolOp_command * com, SYNC_response * res)
+{
+ afs_int32 code = SYNC_FAILED;
+
+ if (VVGCache_scanStart_r(NULL) == 0) {
+ code = SYNC_OK;
+ }
+
+ return code;
+}
#endif /* AFS_DEMAND_ATTACH_FS */
static afs_int32
return SYNC_COM_ERROR;
}
+ ViceLog(125, ("FSYNC_com_VnQry: fd %d got command for vol %lu vnode %lu "
+ "uniq %lu spare %lu partName %.16s\n", (int)fd,
+ afs_printable_uint32_lu(qry->volume),
+ afs_printable_uint32_lu(qry->vnode),
+ afs_printable_uint32_lu(qry->unique),
+ afs_printable_uint32_lu(qry->spare),
+ qry->partName));
+
#ifdef AFS_DEMAND_ATTACH_FS
vp = VLookupVolume_r(&error, qry->volume, NULL);
#else /* !AFS_DEMAND_ATTACH_FS */
scom.sop = (FSSYNC_StatsOp_hdr *) com->payload.buf;
scom.com = com;
+ ViceLog(125, ("FSYNC_com_StatsOp: fd %d got command for stats: "
+ "{vlru_generation = %lu, hash_bucket = %lu, partName = "
+ "%.16s}\n", (int)fd,
+ afs_printable_uint32_lu(scom.sop->args.vlru_generation),
+ afs_printable_uint32_lu(scom.sop->args.hash_bucket),
+ scom.sop->args.partName));
+
switch (com->hdr.command) {
case FSYNC_VOL_STATS_GENERAL:
code = FSYNC_com_StatsOpGeneral(&scom, res);
stats->minFree = dp->minFree;
stats->f_files = dp->f_files;
stats->vol_list_len = dp->vol_list.len;
-
+
res->hdr.response_len += sizeof(struct DiskPartitionStats64);
}
AssignInt64(head->reorders, &stats->chain_reorders);
res->hdr.response_len += sizeof(struct VolumeHashChainStats);
-
+
return code;
}
}
/**
- * check whether command packet partition name matches volume
+ * check whether command packet partition name matches volume
* object's partition name.
*
* @param[in] vcom pointer to command packet
*
* @internal
*/
-static int
+static int
FSYNC_partMatch(FSSYNC_VolOp_command * vcom, Volume * vp, int match_anon)
{
return ((match_anon && vcom->vop->partName[0] == 0) ||
- (strncmp(vcom->vop->partName, V_partition(vp)->name,
+ (strncmp(vcom->vop->partName, V_partition(vp)->name,
sizeof(vcom->vop->partName)) == 0));
}
Volume *vp;
- tvolName[0] = '/';
+ tvolName[0] = OS_DIRSEPC;
sprintf(&tvolName[1], VFORMAT, afs_printable_uint32_lu(p[i].volumeID));
vp = VAttachVolumeByName_r(&error, p[i].partName, tvolName,
V_VOLUPD);
AcceptOn(void)
{
if (AcceptHandler == -1) {
- assert(AddHandler(fssync_server_state.fd, FSYNC_newconnection));
+ osi_Assert(AddHandler(fssync_server_state.fd, FSYNC_newconnection));
AcceptHandler = FindHandler(fssync_server_state.fd);
}
}
AcceptOff(void)
{
if (AcceptHandler != -1) {
- assert(RemoveHandler(fssync_server_state.fd));
+ osi_Assert(RemoveHandler(fssync_server_state.fd));
AcceptHandler = -1;
}
}
static void
InitHandler(void)
{
- register int i;
+ int i;
ObtainWriteLock(&FSYNC_handler_lock);
for (i = 0; i < MAXHANDLERS; i++) {
- HandlerFD[i] = -1;
+ HandlerFD[i] = OSI_NULLSOCKET;
HandlerProc[i] = 0;
}
ReleaseWriteLock(&FSYNC_handler_lock);
static void
CallHandler(fd_set * fdsetp)
{
- register int i;
+ int i;
ObtainReadLock(&FSYNC_handler_lock);
for (i = 0; i < MAXHANDLERS; i++) {
if (HandlerFD[i] >= 0 && FD_ISSET(HandlerFD[i], fdsetp)) {
static int
AddHandler(osi_socket afd, void (*aproc) (osi_socket))
{
- register int i;
+ int i;
ObtainWriteLock(&FSYNC_handler_lock);
for (i = 0; i < MAXHANDLERS; i++)
- if (HandlerFD[i] == -1)
+ if (HandlerFD[i] == OSI_NULLSOCKET)
break;
if (i >= MAXHANDLERS) {
ReleaseWriteLock(&FSYNC_handler_lock);
}
static int
-FindHandler(register osi_socket afd)
+FindHandler(osi_socket afd)
{
- register int i;
+ int i;
ObtainReadLock(&FSYNC_handler_lock);
for (i = 0; i < MAXHANDLERS; i++)
if (HandlerFD[i] == afd) {
return i;
}
ReleaseReadLock(&FSYNC_handler_lock); /* just in case */
- assert(1 == 2);
+ osi_Assert(1 == 2);
return -1; /* satisfy compiler */
}
static int
-FindHandler_r(register osi_socket afd)
+FindHandler_r(osi_socket afd)
{
- register int i;
+ int i;
for (i = 0; i < MAXHANDLERS; i++)
if (HandlerFD[i] == afd) {
return i;
}
- assert(1 == 2);
+ osi_Assert(1 == 2);
return -1; /* satisfy compiler */
}
static int
-RemoveHandler(register osi_socket afd)
+RemoveHandler(osi_socket afd)
{
ObtainWriteLock(&FSYNC_handler_lock);
- HandlerFD[FindHandler_r(afd)] = -1;
+ HandlerFD[FindHandler_r(afd)] = OSI_NULLSOCKET;
ReleaseWriteLock(&FSYNC_handler_lock);
return 1;
}
int fdi = 0;
ObtainReadLock(&FSYNC_handler_lock);
for (i = 0; i < MAXHANDLERS; i++)
- if (HandlerFD[i] != -1) {
- assert(fdi<maxfds);
+ if (HandlerFD[i] != OSI_NULLSOCKET) {
+ osi_Assert(fdi<maxfds);
fds[fdi].fd = HandlerFD[i];
fds[fdi].events = events;
fds[fdi].revents = 0;
static void
GetHandler(fd_set * fdsetp, int *maxfdp)
{
- register int i;
- register int maxfd = -1;
+ int i;
+ int maxfd = -1;
FD_ZERO(fdsetp);
ObtainReadLock(&FSYNC_handler_lock); /* just in case */
for (i = 0; i < MAXHANDLERS; i++)
- if (HandlerFD[i] != -1) {
+ if (HandlerFD[i] != OSI_NULLSOCKET) {
FD_SET(HandlerFD[i], fdsetp);
#ifndef AFS_NT40_ENV
/* On Windows the nfds parameter to select() is ignored */