/**
* create and write a volume disk header to disk.
*
* @param[in] hdr volume disk header
* @param[in] dp disk partition object
*
* @return operation status
* @retval 0 success
* @retval EEXIST volume header already exists
* @retval EIO failed to write volume header
*
* @internal
*/
afs_int32
VCreateVolumeDiskHeader(VolumeDiskHeader_t * hdr,
struct DiskPartition64 * dp)
{
afs_int32 code = 0;
#ifdef AFS_DEMAND_ATTACH_FS
SYNC_response res;
#endif /* AFS_DEMAND_ATTACH_FS */
code = _VWriteVolumeDiskHeader(hdr, dp, O_CREAT | O_EXCL);
if (code) {
goto done;
}
#ifdef AFS_DEMAND_ATTACH_FS
memset(&res, 0, sizeof(res));
code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
if (code) {
Log("VCreateVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) failed "
"with code %ld reason %ld\n", dp->name,
afs_printable_uint32_lu(hdr->parent),
afs_printable_uint32_lu(hdr->id),
afs_printable_int32_ld(code),
afs_printable_int32_ld(res.hdr.reason));
}
#endif /* AFS_DEMAND_ATTACH_FS */
done:
return code;
}
/**
* read an existing volume disk header.
*
* @param[in] volid volume id
* @param[in] dp disk partition object
* @param[out] hdr volume disk header or NULL
*
* @note if hdr is NULL, this is essentially an existence test for the vol
* header
*
* @return operation status
* @retval 0 success
* @retval -1 volume header doesn't exist
* @retval EIO failed to read volume header
*
* @internal
*/
afs_int32
VReadVolumeDiskHeader(VolumeId volid,
struct DiskPartition64 * dp,
VolumeDiskHeader_t * hdr)
{
afs_int32 code = 0;
int fd;
char path[MAXPATHLEN];
snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
VPartitionPath(dp), afs_printable_uint32_lu(volid));
fd = open(path, O_RDONLY);
if (fd < 0) {
Log("VReadVolumeDiskHeader: Couldn't open header for volume %lu (errno %d).\n",
afs_printable_uint32_lu(volid), errno);
code = -1;
} else if (hdr && read(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
Log("VReadVolumeDiskHeader: Couldn't read header for volume %lu.\n",
afs_printable_uint32_lu(volid));
code = EIO;
}
if (fd >= 0) {
close(fd);
}
return code;
}
int
afscp_GetStatus(const struct afscp_venusfid *fid, struct AFSFetchStatus *s)
{
struct afscp_volume *v;
struct afscp_server *server;
struct AFSCallBack cb;
struct AFSVolSync vs;
struct AFSFid tf = fid->fid;
struct afscp_statent *stored, key;
void *cached;
int code, i, j;
time_t now;
v = afscp_VolumeById(fid->cell, fid->fid.Volume);
if (v == NULL) {
return -1;
}
memset(&key, 0, sizeof(key));
memcpy(&key.me, fid, sizeof(*fid));
cached = tfind(&key, &v->statcache, statcompare);
if (cached != NULL) {
stored = *(struct afscp_statent **)cached;
pthread_mutex_lock(&(stored->mtx));
memmove(s, &stored->status, sizeof(*s));
afs_dprintf(("Stat %u.%lu.%lu.%lu returning cached result\n",
fid->cell->id, fid->fid.Volume, fid->fid.Vnode,
fid->fid.Unique));
if (stored->nwaiters)
pthread_cond_broadcast(&(stored->cv));
pthread_mutex_unlock(&(stored->mtx));
return 0;
}
code = ENOENT;
for (i = 0; i < v->nservers; i++) {
server = afscp_ServerByIndex(v->servers[i]);
if (server && server->naddrs > 0) {
for (j = 0; j < server->naddrs; j++) {
time(&now);
code = RXAFS_FetchStatus(server->conns[j], &tf, s, &cb, &vs);
if (code == 0) {
afscp_AddCallBack(server, &fid->fid, s, &cb, now); /* calls _StatStuff */
afs_dprintf(("Stat %d.%lu.%lu.%lu"
" ok: type %ld size %ld\n",
fid->cell->id,
afs_printable_uint32_lu(fid->fid.Volume),
afs_printable_uint32_lu(fid->fid.Vnode),
afs_printable_uint32_lu(fid->fid.Unique),
afs_printable_int32_ld(s->FileType),
afs_printable_int32_ld(s->Length)));
return 0;
}
}
}
}
afscp_errno = code;
return -1;
}
/**
* destroy a volume disk header.
*
* @param[in] dp disk partition object
* @param[in] volid volume id
* @param[in] parent parent's volume id, 0 if unknown
*
* @return operation status
* @retval 0 success
*
* @note if parent is 0, the parent volume ID will be looked up from the
* fileserver
*
* @note for non-DAFS, parent is currently ignored
*/
afs_int32
VDestroyVolumeDiskHeader(struct DiskPartition64 * dp,
VolumeId volid,
VolumeId parent)
{
afs_int32 code = 0;
char path[MAXPATHLEN];
#ifdef AFS_DEMAND_ATTACH_FS
SYNC_response res;
#endif /* AFS_DEMAND_ATTACH_FS */
(void)afs_snprintf(path, sizeof(path),
"%s/" VFORMAT,
VPartitionPath(dp), afs_printable_uint32_lu(volid));
code = unlink(path);
if (code) {
Log("VDestroyVolumeDiskHeader: Couldn't unlink disk header, error = %d\n", errno);
goto done;
}
#ifdef AFS_DEMAND_ATTACH_FS
memset(&res, 0, sizeof(res));
if (!parent) {
FSSYNC_VGQry_response_t q_res;
code = FSYNC_VGCQuery(dp->name, volid, &q_res, &res);
if (code) {
Log("VDestroyVolumeDiskHeader: FSYNC_VGCQuery(%s, %lu) failed "
"with code %ld, reason %ld\n", dp->name,
afs_printable_uint32_lu(volid), afs_printable_int32_ld(code),
afs_printable_int32_ld(res.hdr.reason));
goto done;
}
parent = q_res.rw;
}
code = FSYNC_VGCDel(dp->name, parent, volid, FSYNC_WHATEVER, &res);
if (code) {
Log("VDestroyVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) failed "
"with code %ld reason %ld\n", dp->name,
afs_printable_uint32_lu(parent),
afs_printable_uint32_lu(volid),
afs_printable_int32_ld(code),
afs_printable_int32_ld(res.hdr.reason));
}
#endif /* AFS_DEMAND_ATTACH_FS */
done:
return code;
}
/**
* add an entry to the hash table.
*
* @param[in] dp disk partition object
* @param[in] volid volume id
* @param[in] ent volume group object
* @param[out] hash_out address in which to store pointer to hash entry
*
* @pre VOL_LOCK held
*
* @return operation status
* @retval 0 success
* @retval EEXIST hash entry for volid already exists, and it points to
* a different VG entry
*
* @internal
*/
static int
_VVGC_hash_entry_add(struct DiskPartition64 * dp,
VolumeId volid,
VVGCache_entry_t * ent,
VVGCache_hash_entry_t ** hash_out)
{
int code = 0;
VVGCache_hash_entry_t * hent;
int hash = VVGC_HASH(volid);
VVGCache_entry_t *nent;
code = _VVGC_lookup(dp, volid, &nent, hash_out);
if (!code) {
if (ent != nent) {
ViceLog(0, ("_VVGC_hash_entry_add: tried to add a duplicate "
" nonmatching entry for vol %lu: original "
"(%"AFS_PTR_FMT",%lu) new (%"AFS_PTR_FMT",%lu)\n",
afs_printable_uint32_lu(volid),
nent, afs_printable_uint32_lu(nent->rw),
ent, afs_printable_uint32_lu(ent->rw)));
return EEXIST;
}
ViceLog(1, ("_VVGC_hash_entry_add: tried to add duplicate "
"hash entry for vol %lu, VG %lu",
afs_printable_uint32_lu(volid),
afs_printable_uint32_lu(ent->rw)));
/* accept attempts to add matching duplicate entries; just
* pretend we added it */
return 0;
}
code = _VVGC_hash_entry_alloc(&hent);
if (code) {
goto done;
}
hent->entry = ent;
hent->dp = dp;
hent->volid = volid;
queue_Append(&VVGCache_hash_table.hash_buckets[hash],
hent);
done:
if (hash_out) {
*hash_out = hent;
}
return code;
}
/**
* flush all cache entries for a given disk partition.
*
* @param[in] part disk partition object
*
* @pre VOL_LOCK held
*
* @return operation status
* @retval 0 success
*
* @internal
*/
int
_VVGC_flush_part_r(struct DiskPartition64 * part)
{
int code = 0, res;
int i;
VVGCache_hash_entry_t * ent, * nent;
for (i = 0; i < VolumeHashTable.Size; i++) {
for (queue_Scan(&VVGCache_hash_table.hash_buckets[i],
ent,
nent,
VVGCache_hash_entry)) {
if (ent->dp == part) {
VolumeId volid = ent->volid;
res = _VVGC_hash_entry_del(ent);
if (res) {
ViceLog(0, ("_VVGC_flush_part_r: error %d deleting hash entry for %lu\n",
res, afs_printable_uint32_lu(volid)));
code = res;
}
}
}
}
return code;
}
/**
* write an existing volume disk header.
*
* @param[in] hdr volume disk header
* @param[in] dp disk partition object
* @param[in] cr assert if O_CREAT | O_EXCL should be passed to open()
*
* @return operation status
* @retval 0 success
* @retval -1 volume header doesn't exist
* @retval EIO failed to write volume header
*
* @internal
*/
static afs_int32
_VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
struct DiskPartition64 * dp,
int flags)
{
afs_int32 code = 0;
int fd;
char path[MAXPATHLEN];
#ifdef AFS_DEMAND_ATTACH_FS
/* prevent racing with VGC scanners reading the vol header while we are
* writing it */
code = VPartHeaderLock(dp, READ_LOCK);
if (code) {
return EIO;
}
#endif /* AFS_DEMAND_ATTACH_FS */
flags |= O_RDWR;
(void)afs_snprintf(path, sizeof(path),
"%s/" VFORMAT,
VPartitionPath(dp), afs_printable_uint32_lu(hdr->id));
fd = open(path, flags, 0644);
if (fd < 0) {
code = errno;
Log("_VWriteVolumeDiskHeader: Couldn't open header for volume %lu, "
"error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
} else if (write(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
Log("_VWriteVolumeDiskHeader: Couldn't write header for volume %lu, "
"error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
code = EIO;
}
if (fd >= 0) {
if (close(fd) != 0) {
Log("_VWriteVolumeDiskHeader: Error closing header for volume "
"%lu, errno %d\n", afs_printable_uint32_lu(hdr->id), errno);
}
}
#ifdef AFS_DEMAND_ATTACH_FS
VPartHeaderUnlock(dp, READ_LOCK);
#endif /* AFS_DEMAND_ATTACH_FS */
return code;
}
static int
VolOpQuery(struct cmd_syndesc * as, void * rock)
{
struct fssync_state state;
SYNC_PROTO_BUF_DECL(res_buf);
SYNC_response res;
FSSYNC_VolOp_info vop;
res.hdr.response_len = sizeof(res.hdr);
res.payload.buf = res_buf;
res.payload.len = SYNC_PROTO_MAX_LEN;
common_prolog(as, &state);
common_volop_prolog(as, &state);
do_volop(&state, FSYNC_VOL_QUERY_VOP, &res);
if (!(res.hdr.flags & SYNC_FLAG_DAFS_EXTENSIONS)) {
printf("*** file server not compiled with demand attach extensions.\n");
printf("*** pending volume operation metadata not available.\n");
}
if (res.hdr.response == SYNC_OK) {
memcpy(&vop, res.payload.buf, sizeof(FSSYNC_VolOp_info));
printf("pending_vol_op = {\n");
printf("\tcom = {\n");
printf("\t\tproto_version = %u\n", vop.com.proto_version);
printf("\t\tpkt_seq = %u\n", vop.com.pkt_seq);
printf("\t\tcom_seq = %u\n", vop.com.com_seq);
printf("\t\tprogramType = %d (%s)\n",
vop.com.programType, VPTypeToString(vop.com.programType));
printf("\t\tpid = %d\n", vop.com.pid);
printf("\t\ttid = %d\n", vop.com.tid);
printf("\t\tcommand = %d (%s)\n",
vop.com.command, FSYNC_com2string(vop.com.command));
printf("\t\treason = %d (%s)\n",
vop.com.reason, FSYNC_reason2string(vop.com.reason));
printf("\t\tcommand_len = %u\n", vop.com.command_len);
printf("\t\tflags = 0x%lux\n", afs_printable_uint32_lu(vop.com.flags));
printf("\t}\n");
printf("\tvop = {\n");
printf("\t\tvolume = %" AFS_VOLID_FMT "\n", afs_printable_VolumeId_lu(vop.vop.volume));
if (strnlen(vop.vop.partName, sizeof(vop.vop.partName)) <
sizeof(vop.vop.partName)) {
printf("\t\tpartName = '%s'\n", vop.vop.partName);
} else {
printf("\t\tpartName = (illegal string)\n");
}
printf("\t}\n");
printf("}\n");
}
return 0;
}
/**
* put back a reference to an entry.
*
* @param[in] dp disk partition object
* @param[in] entry cache entry
*
* @pre VOL_LOCK held
*
* @warning do not attempt to deref pointer after calling this interface
*
* @return operation status
* @retval 0 success
*
* @note dp is needed to lookup the RW hash entry to unlink, if we are
* putting back the final reference and freeing
*
* @internal
*/
static int
_VVGC_entry_put(struct DiskPartition64 * dp, VVGCache_entry_t * entry)
{
int code = 0;
osi_Assert(entry->refcnt > 0);
if (--entry->refcnt == 0) {
VVGCache_entry_t *nentry;
VVGCache_hash_entry_t *hentry;
/* first, try to delete the RW id hash entry pointing to this
* entry */
code = _VVGC_lookup(dp, entry->rw, &nentry, &hentry);
if (!code) {
if (nentry != entry) {
/* looking up the rw of this entry points to a different
* entry; should not happen */
ViceLog(0, ("VVGC_entry_put: error: entry lookup for entry %lu "
"found different entry than was passed",
afs_printable_uint32_lu(entry->rw)));
code = -1;
} else {
code = _VVGC_hash_entry_unlink(hentry);
hentry = NULL;
}
} else if (code == ENOENT) {
/* ignore ENOENT; this shouldn't happen, since the RW hash
* entry should always exist if the entry does... but we
* were going to delete it anyway, so try to continue */
ViceLog(0, ("VVGC_entry_put: warning: tried to unlink entry for "
"vol %lu, but RW hash entry doesn't exist; continuing "
"anyway...\n", afs_printable_uint32_lu(entry->rw)));
code = 0;
}
/* now, just free the entry itself */
if (!code) {
code = _VVGC_entry_free(entry);
}
}
return code;
}
/**
* lock an offset in a file descriptor.
*
* @param[in] fd file descriptor to lock
* @param[in] offset offset in file to lock
* @param[in] locktype READ_LOCK or WRITE_LOCK
* @param[in] nonblock 1 to fail immediately, 0 to wait to acquire lock
*
* @return operation status
* @retval 0 success
* @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
* specified
* @retval EIO error acquiring file lock
*/
static_inline int
_VLockFd(FD_t fd, afs_uint32 offset, int locktype, int nonblock)
{
int l_type = F_WRLCK;
int cmd = AFS_SETLKW;
struct afs_st_flock sf;
if (locktype == READ_LOCK) {
l_type = F_RDLCK;
}
if (nonblock) {
cmd = AFS_SETLK;
}
sf.l_start = offset;
sf.l_len = 1;
sf.l_type = l_type;
sf.l_whence = SEEK_SET;
if (fcntl(fd, cmd, &sf)) {
if (nonblock && (errno == EACCES || errno == EAGAIN)) {
/* We asked for a nonblocking lock, and it was already locked */
sf.l_pid = 0;
if (fcntl(fd, F_GETLK, &sf) != 0 || sf.l_pid == 0) {
Log("_VLockFd: fcntl failed with error %d when trying to "
"query the conflicting lock for fd %d (locktype=%d, "
"offset=%lu)\n", errno, fd, locktype,
afs_printable_uint32_lu(offset));
} else {
Log("_VLockFd: conflicting lock held on fd %d, offset %lu by "
"pid %ld (locktype=%d)\n", fd,
afs_printable_uint32_lu(offset), (long int)sf.l_pid,
locktype);
}
return EBUSY;
}
Log("_VLockFd: fcntl failed with error %d when trying to lock "
"fd %d (locktype=%d, offset=%lu)\n", errno, fd, locktype,
afs_printable_uint32_lu(offset));
return EIO;
}
return 0;
}
/**
* add a volid to the entry's child list.
*
* @param[in] ent volume group object
* @param[in] volid volume id
*
* @return operation status
* @retval 0 success
* @retval -1 child table is full
*
* @internal
*/
static int
_VVGC_entry_cl_add(VVGCache_entry_t * ent,
VolumeId volid)
{
int code = 0, i;
int empty_idx = -1;
/* search table to avoid duplicates */
for (i = 0; i < VOL_VG_MAX_VOLS; i++) {
if (ent->children[i] == volid) {
ViceLog(1, ("VVGC_entry_cl_add: tried to add duplicate vol "
"%lu to VG %lu\n",
afs_printable_uint32_lu(volid),
afs_printable_uint32_lu(ent->rw)));
goto done;
}
if (empty_idx == -1 && !ent->children[i]) {
empty_idx = i;
/* don't break; make sure we go through all children so we don't
* add a duplicate entry */
}
}
/* verify table isn't full */
if (empty_idx == -1) {
code = -1;
ViceLog(0, ("VVGC_entry_cl_add: tried to add vol %lu to VG %lu, but VG "
"is full\n", afs_printable_uint32_lu(volid),
afs_printable_uint32_lu(ent->rw)));
goto done;
}
/* add entry */
ent->children[empty_idx] = volid;
/* inc refcount */
code = _VVGC_entry_get(ent);
done:
return code;
}
/**
* delete an entry from the volume group cache.
*
* @param[in] dp disk partition object
* @param[in] parent parent volume id
* @param[in] child child volume id
*
* @pre VOL_LOCK held
*
* @internal
*
* @return operation status
* @retval 0 success
*/
int
_VVGC_entry_purge_r(struct DiskPartition64 * dp,
VolumeId parent, VolumeId child)
{
int code = 0, res;
VVGCache_entry_t * parent_ent, * child_ent;
VVGCache_hash_entry_t * child_hent;
/* check mappings for each volid */
res = _VVGC_lookup(dp, parent, &parent_ent, NULL);
if (res) {
code = res;
goto done;
}
res = _VVGC_lookup(dp, child, &child_ent, &child_hent);
if (res) {
code = res;
goto done;
}
/* if the mappings don't match, we have a serious error */
if (parent_ent != child_ent) {
ViceLog(0, ("VVGCache_entry_del: trying to delete vol %lu from VG %lu, "
"but vol %lu points to VGC entry %"AFS_PTR_FMT" and VG %lu "
"points to VGC entry %"AFS_PTR_FMT"\n",
afs_printable_uint32_lu(child),
afs_printable_uint32_lu(parent),
afs_printable_uint32_lu(child),
child_ent, afs_printable_uint32_lu(parent), parent_ent));
code = -1;
goto done;
}
code = _VVGC_hash_entry_del(child_hent);
done:
return code;
}
static int
handleit(struct cmd_syndesc *as, void *arock)
{
struct cmd_item *ti;
int err = 0;
afs_uint32 volumeId = 0;
char *partName = 0;
struct DiskPartition64 *partP = NULL;
#ifndef AFS_NT40_ENV
if (geteuid() != 0) {
printf("vol-info must be run as root; sorry\n");
exit(1);
}
#endif
if (as->parms[0].items)
online = 1;
else
online = 0;
if (as->parms[1].items)
DumpVnodes = 1;
else
DumpVnodes = 0;
if (as->parms[2].items)
DumpDate = 1;
else
DumpDate = 0;
if (as->parms[3].items)
DumpInodeNumber = 1;
else
DumpInodeNumber = 0;
if (as->parms[4].items)
InodeTimes = 1;
else
InodeTimes = 0;
if ((ti = as->parms[5].items))
partName = ti->data;
if ((ti = as->parms[6].items))
volumeId = strtoul(ti->data, NULL, 10);
if (as->parms[7].items)
dheader = 1;
else
dheader = 0;
if (as->parms[8].items) {
dsizeOnly = 1;
dheader = 1;
DumpVnodes = 1;
} else
dsizeOnly = 0;
if (as->parms[9].items) {
fixheader = 1;
} else
fixheader = 0;
if (as->parms[10].items) {
saveinodes = 1;
dheader = 1;
DumpVnodes = 1;
} else
saveinodes = 0;
if (as->parms[11].items) {
orphaned = 1;
DumpVnodes = 1;
} else
#if defined(AFS_NAMEI_ENV)
if (as->parms[12].items) {
PrintFileNames = 1;
DumpVnodes = 1;
} else
#endif
orphaned = 0;
DInit(10);
err = VAttachPartitions();
if (err) {
printf("%d partitions had errors during attach.\n", err);
}
if (partName) {
partP = VGetPartition(partName, 0);
if (!partP) {
printf("%s is not an AFS partition name on this server.\n",
partName);
exit(1);
}
}
if (!volumeId) {
if (!partP) {
HandleAllPart();
} else {
HandlePart(partP);
}
} else {
char name1[128];
if (!partP) {
partP = FindCurrentPartition();
if (!partP) {
printf("Current partition is not a vice partition.\n");
exit(1);
}
}
(void)afs_snprintf(name1, sizeof name1, VFORMAT,
afs_printable_uint32_lu(volumeId));
if (dsizeOnly && !saveinodes)
printf
("Volume-Id\t Volsize Auxsize Inodesize AVolsize SizeDiff (VolName)\n");
HandleVolume(partP, name1);
}
return 0;
}
/**
* verify that the fileserver still thinks we have a volume checked out.
*
* In DAFS, a non-fileserver program accesses a volume by checking it out from
* the fileserver (FSYNC_VOL_OFF or FSYNC_VOL_NEEDVOLUME), and then locks the
* volume. There is a possibility that the fileserver crashes or restarts for
* some reason between volume checkout and locking; if this happens, the
* fileserver could attach the volume before we had a chance to lock it. This
* function serves to detect if this has happened; it must be called after
* volume checkout and locking to make sure the fileserver still thinks we
* have the volume. (If it doesn't, we should try to check it out again.)
*
* @param[in] volume volume ID
* @param[in] partition partition name string
* @param[in] command the command that was used to checkout the volume
* @param[in] reason the reason code used to checkout the volume
*
* @return operation status
* @retval SYNC_OK the fileserver could not have attached the volume since
* it was checked out (either it thinks it is still checked
* out, or it doesn't know about the volume)
* @retval SYNC_DENIED fileserver may have restarted since checkout; checkout
* should be reattempted
* @retval SYNC_COM_ERROR internal/fatal error
*/
afs_int32
FSYNC_VerifyCheckout(VolumeId volume, char * partition,
afs_int32 command, afs_int32 reason)
{
SYNC_response res;
FSSYNC_VolOp_info vop;
afs_int32 code;
afs_int32 pid;
res.hdr.response_len = sizeof(res.hdr);
res.payload.buf = &vop;
res.payload.len = sizeof(vop);
code = FSYNC_VolOp(volume, partition, FSYNC_VOL_QUERY_VOP, FSYNC_WHATEVER, &res);
if (code != SYNC_OK) {
if (res.hdr.reason == FSYNC_NO_PENDING_VOL_OP) {
Log("FSYNC_VerifyCheckout: fileserver claims no vop for vol %lu "
"part %s; fileserver may have restarted since checkout\n",
afs_printable_uint32_lu(volume), partition);
return SYNC_DENIED;
}
if (res.hdr.reason == FSYNC_UNKNOWN_VOLID ||
res.hdr.reason == FSYNC_WRONG_PART) {
/* if the fileserver does not know about this volume on this
* partition, there's no way it could have attached it, so we're
* fine */
return SYNC_OK;
}
Log("FSYNC_VerifyCheckout: FSYNC_VOL_QUERY_VOP failed for vol %lu "
"part %s with code %ld reason %ld\n",
afs_printable_uint32_lu(volume), partition,
afs_printable_int32_ld(code),
afs_printable_int32_ld(res.hdr.reason));
return SYNC_COM_ERROR;
}
pid = getpid();
/* Check if the current vol op is us. Checking pid is probably enough, but
* be a little bit paranoid. We could also probably check tid, but I'm not
* completely confident of its reliability on all platforms (on pthread
* envs, we coerce a pthread_t to an afs_int32, which is not guaranteed
* to mean anything significant). */
if (vop.com.programType == programType && vop.com.pid == pid &&
vop.com.command == command && vop.com.reason == reason) {
/* looks like the current pending vol op is the same one as the one
* with which we checked it out. success. */
return SYNC_OK;
}
Log("FSYNC_VerifyCheckout: vop for vol %lu part %s does not match "
"expectations (got pt %ld pid %ld cmd %ld reason %ld, but expected "
"pt %ld pid %ld cmd %ld reason %ld); fileserver may have restarted "
"since checkout\n", afs_printable_uint32_lu(volume), partition,
afs_printable_int32_ld(vop.com.programType),
afs_printable_int32_ld(vop.com.pid),
afs_printable_int32_ld(vop.com.command),
afs_printable_int32_ld(vop.com.reason),
afs_printable_int32_ld(programType),
afs_printable_int32_ld(pid),
afs_printable_int32_ld(command),
afs_printable_int32_ld(reason));
return SYNC_DENIED;
}
struct afscp_volume *
afscp_VolumeByName(struct afscp_cell *cell, const char *vname,
afs_int32 intype)
{
union allvldbentry u;
struct afscp_volume *ret, key;
struct afscp_server *server;
afs_int32 code, vtype, type, srv;
void *s;
#ifdef AFSCP_DEBUG
struct in_addr i;
#endif
if (intype == RWVOL)
vtype = VLSF_RWVOL;
else if (intype == ROVOL)
vtype = VLSF_ROVOL;
else if (intype == BACKVOL)
vtype = VLSF_BACKVOL;
else {
afscp_errno = EINVAL;
return NULL;
}
memset(&key, 0, sizeof(key));
strlcpy(key.name, vname, sizeof(key.name));
key.voltype = vtype;
s = tfind(&key, &cell->volsbyname, ncompare);
if (s) {
ret = *(struct afscp_volume **)s;
return ret;
}
type = 0;
code = ubik_VL_GetEntryByNameU(cell->vlservers, 0, (char *)vname, &u.u);
if (code == RXGEN_OPCODE) {
type = 1;
code =
ubik_VL_GetEntryByNameN(cell->vlservers, 0, (char *)vname, &u.n);
if (code == RXGEN_OPCODE) {
type = 2;
code = ubik_VL_GetEntryByNameO(cell->vlservers, 0, (char *)vname,
&u.o);
}
}
if (code != 0) {
afscp_errno = code;
return NULL;
}
ret = calloc(1, sizeof(struct afscp_volume));
if (ret == NULL) {
afscp_errno = ENOMEM;
return NULL;
}
strlcpy(ret->name, u.u.name, sizeof(ret->name));
ret->nservers = 0;
ret->cell = cell;
switch (type) {
case 0:
ret->id = u.u.volumeId[intype];
for (srv = 0; srv < u.u.nServers; srv++) {
if ((u.u.serverFlags[srv] & vtype) == 0)
continue;
afs_dprintf(("uvldbentry server %d flags: %x\n", srv,
u.u.serverFlags[srv]));
if ((u.u.serverFlags[srv] & VLSF_UUID) == 0)
server =
afscp_ServerByAddr(cell, u.u.serverNumber[srv].time_low);
else
server = afscp_ServerById(cell, &u.u.serverNumber[srv]);
if (!server)
continue;
ret->servers[ret->nservers++] = server->index;
}
break;
case 1:
ret->id = u.n.volumeId[intype];
for (srv = 0; srv < u.n.nServers; srv++) {
if ((u.n.serverFlags[srv] & vtype) == 0)
continue;
server = afscp_ServerByAddr(cell, u.n.serverNumber[srv]);
if (!server)
continue;
ret->servers[ret->nservers++] = server->index;
}
break;
case 2:
ret->id = u.o.volumeId[intype];
for (srv = 0; srv < u.o.nServers; srv++) {
if ((u.o.serverFlags[srv] & vtype) == 0)
continue;
server = afscp_ServerByAddr(cell, u.o.serverNumber[srv]);
if (!server)
continue;
ret->servers[ret->nservers++] = server->index;
}
break;
}
if (!ret->nservers || !ret->id) {
free(ret);
return NULL;
}
ret->voltype = intype;
#ifdef AFSCP_DEBUG
server = afscp_ServerByIndex(ret->servers[0]);
if (server != NULL)
i.s_addr = server->addrs[0];
else
i.s_addr = 0;
#endif
afs_dprintf(("New volume BYNAME %s (%lu) on %s (%d)\n", ret->name,
afs_printable_uint32_lu(ret->id),
inet_ntoa(i), ret->servers[0]));
s = tsearch(&key, &cell->volsbyname, ncompare);
if (s)
*(struct afscp_volume **)s = ret;
key.id = ret->id;
s = tsearch(&key, &cell->volsbyid, icompare);
if (s)
*(struct afscp_volume **)s = ret;
return ret;
}
struct afscp_volume *
afscp_VolumeById(struct afscp_cell *cell, afs_uint32 id)
{
union allvldbentry u;
struct afscp_volume *ret, key;
struct afscp_server *server;
afs_int32 code, vtype, type, srv;
int voltype = -1;
char idbuffer[16];
void *s;
#ifdef AFSCP_DEBUG
struct in_addr i;
#endif
memset(&key, 0, sizeof(key));
key.id = id;
s = tfind(&key, &cell->volsbyid, icompare);
if (s) {
ret = *(struct afscp_volume **)s;
return ret;
}
snprintf(idbuffer, sizeof(idbuffer), "%lu", afs_printable_uint32_lu(id));
type = 0;
code = ubik_VL_GetEntryByNameU(cell->vlservers, 0, idbuffer, &u.u);
if (code == RXGEN_OPCODE) {
type = 1;
code = ubik_VL_GetEntryByIDN(cell->vlservers, 0, id, -1, &u.n);
if (code == RXGEN_OPCODE) {
type = 2;
code = ubik_VL_GetEntryByID(cell->vlservers, 0, id, -1, &u.o);
}
}
if (code != 0) {
afscp_errno = code;
return NULL;
}
ret = calloc(1, sizeof(struct afscp_volume));
if (ret == NULL) {
afscp_errno = ENOMEM;
return NULL;
}
strlcpy(ret->name, u.u.name, sizeof(ret->name));
ret->nservers = 0;
ret->cell = cell;
switch (type) {
case 0:
if (id == u.u.volumeId[RWVOL]) {
vtype = VLSF_RWVOL;
voltype = RWVOL;
} else if (id == u.u.volumeId[ROVOL]) {
vtype = VLSF_ROVOL;
voltype = ROVOL;
} else if (id == u.u.volumeId[BACKVOL]) {
vtype = VLSF_BACKVOL;
voltype = BACKVOL;
} else {
vtype = 0;
voltype = -1;
}
for (srv = 0; srv < u.u.nServers; srv++) {
if ((u.u.serverFlags[srv] & vtype) == 0)
continue;
if ((u.u.serverFlags[srv] & VLSF_UUID) == 0)
server =
afscp_ServerByAddr(cell, u.u.serverNumber[srv].time_low);
else
server = afscp_ServerById(cell, &u.u.serverNumber[srv]);
if (!server)
continue;
ret->servers[ret->nservers++] = server->index;
}
break;
case 1:
if (id == u.n.volumeId[RWVOL]) {
vtype = VLSF_RWVOL;
voltype = RWVOL;
} else if (id == u.n.volumeId[ROVOL]) {
vtype = VLSF_ROVOL;
voltype = ROVOL;
} else if (id == u.n.volumeId[BACKVOL]) {
vtype = VLSF_BACKVOL;
voltype = BACKVOL;
} else {
vtype = 0;
voltype = -1;
}
for (srv = 0; srv < u.n.nServers; srv++) {
if ((u.n.serverFlags[srv] & vtype) == 0)
continue;
server = afscp_ServerByAddr(cell, u.n.serverNumber[srv]);
if (server == NULL)
continue;
ret->servers[ret->nservers++] = server->index;
}
break;
case 2:
if (id == u.o.volumeId[RWVOL]) {
vtype = VLSF_RWVOL;
voltype = RWVOL;
} else if (id == u.o.volumeId[ROVOL]) {
vtype = VLSF_ROVOL;
voltype = ROVOL;
} else if (id == u.o.volumeId[BACKVOL]) {
vtype = VLSF_BACKVOL;
voltype = BACKVOL;
} else {
vtype = 0;
voltype = -1;
}
for (srv = 0; srv < u.o.nServers; srv++) {
if ((u.o.serverFlags[srv] & vtype) == 0)
continue;
server = afscp_ServerByAddr(cell, u.o.serverNumber[srv]);
if (server == NULL)
continue;
ret->servers[ret->nservers++] = server->index;
}
break;
}
ret->voltype = voltype;
#ifdef AFSCP_DEBUG
server = afscp_ServerByIndex(ret->servers[0]);
if (server)
i.s_addr = server->addrs[0];
else
i.s_addr = 0;
#endif
afs_dprintf(("New volume BYID %s (%lu) on %s (%d)\n", ret->name,
afs_printable_uint32_lu(ret->id), inet_ntoa(i),
ret->servers[0]));
s = tsearch(&key, &cell->volsbyid, icompare);
if (s)
*(struct afscp_volume **)s = ret;
strlcpy(key.name, ret->name, sizeof(key.name));
s = tsearch(&key, &cell->volsbyname, ncompare);
if (s)
*(struct afscp_volume **)s = ret;
return ret;
}
/**
* write an existing volume disk header.
*
* @param[in] hdr volume disk header
* @param[in] dp disk partition object
*
* @return operation status
* @retval 0 success
* @retval ENOENT volume header doesn't exist
* @retval EIO failed to write volume header
*/
afs_int32
VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
struct DiskPartition64 * dp)
{
afs_int32 code;
#ifdef AFS_DEMAND_ATTACH_FS
VolumeDiskHeader_t oldhdr;
int delvgc = 0, addvgc = 0;
SYNC_response res;
/* first, see if anything with the volume IDs have changed; if so, we
* need to update the VGC */
code = VReadVolumeDiskHeader(hdr->id, dp, &oldhdr);
if (code == 0 && (oldhdr.id != hdr->id || oldhdr.parent != hdr->parent)) {
/* the vol id or parent vol id changed; need to delete the VGC entry
* for the old vol id/parent, and add the new one */
delvgc = 1;
addvgc = 1;
} else if (code) {
/* couldn't get the old header info; add the new header info to the
* VGC in case it hasn't been added yet */
addvgc = 1;
}
#endif /* AFS_DEMAND_ATTACH_FS */
code = _VWriteVolumeDiskHeader(hdr, dp, 0);
if (code) {
goto done;
}
#ifdef AFS_DEMAND_ATTACH_FS
if (delvgc) {
memset(&res, 0, sizeof(res));
code = FSYNC_VGCDel(dp->name, oldhdr.parent, oldhdr.id, FSYNC_WHATEVER, &res);
/* unknown vol id is okay; it just further suggests the old header
* data was bogus, which is fine since we're trying to fix it */
if (code && res.hdr.reason != FSYNC_UNKNOWN_VOLID) {
Log("VWriteVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) "
"failed with code %ld reason %ld\n", dp->name,
afs_printable_uint32_lu(oldhdr.parent),
afs_printable_uint32_lu(oldhdr.id),
afs_printable_int32_ld(code),
afs_printable_int32_ld(res.hdr.reason));
}
}
if (addvgc) {
memset(&res, 0, sizeof(res));
code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
if (code) {
Log("VWriteVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) "
"failed with code %ld reason %ld\n", dp->name,
afs_printable_uint32_lu(hdr->parent),
afs_printable_uint32_lu(hdr->id),
afs_printable_int32_ld(code),
afs_printable_int32_ld(res.hdr.reason));
}
}
#endif /* AFS_DEMAND_ATTACH_FS */
done:
return code;
}
Volume *
VCreateVolume_r(Error * ec, char *partname, VolumeId volumeId, VolumeId parentId)
{ /* Should be the same as volumeId if there is
* no parent */
VolumeDiskData vol;
int i, rc;
char headerName[VMAXPATHLEN], volumePath[VMAXPATHLEN];
Device device;
struct DiskPartition64 *partition;
struct VolumeDiskHeader diskHeader;
IHandle_t *handle;
FdHandle_t *fdP;
Inode nearInode AFS_UNUSED = 0;
char *part, *name;
struct stat st;
struct VolumeHeader tempHeader;
struct afs_inode_info stuff[MAXINODETYPE];
afs_ino_str_t stmp;
# ifdef AFS_DEMAND_ATTACH_FS
int locktype = 0;
# endif /* AFS_DEMAND_ATTACH_FS */
init_inode_info(&tempHeader, stuff);
*ec = 0;
memset(&vol, 0, sizeof(vol));
vol.id = volumeId;
vol.parentId = parentId;
vol.copyDate = time(0); /* The only date which really means when this
* @i(instance) of this volume was created.
* Creation date does not mean this */
/* Initialize handle for error case below. */
handle = NULL;
/* Verify that the parition is valid before writing to it. */
if (!(partition = VGetPartition_r(partname, 0))) {
Log("VCreateVolume: partition %s is not in service.\n", partname);
*ec = VNOVOL;
return NULL;
}
#if defined(NEARINODE_HINT)
nearInodeHash(volumeId, nearInode);
nearInode %= partition->f_files;
#endif
VGetVolumePath(ec, vol.id, &part, &name);
if (*ec == VNOVOL || !strcmp(partition->name, part)) {
/* this case is ok */
} else {
/* return EXDEV if it's a clone to an alternate partition
* otherwise assume it's a move */
if (vol.parentId != vol.id) {
*ec = EXDEV;
return NULL;
}
}
*ec = 0;
# ifdef AFS_DEMAND_ATTACH_FS
/* volume doesn't exist yet, but we must lock it to try to prevent something
* else from reading it when we're e.g. half way through creating it (or
* something tries to create the same volume at the same time) */
locktype = VVolLockType(V_VOLUPD, 1);
rc = VLockVolumeByIdNB(volumeId, partition, locktype);
if (rc) {
Log("VCreateVolume: vol %lu already locked by someone else\n",
afs_printable_uint32_lu(volumeId));
*ec = VNOVOL;
return NULL;
}
# else /* AFS_DEMAND_ATTACH_FS */
VLockPartition_r(partname);
# endif /* !AFS_DEMAND_ATTACH_FS */
memset(&tempHeader, 0, sizeof(tempHeader));
tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
tempHeader.stamp.version = VOLUMEHEADERVERSION;
tempHeader.id = vol.id;
tempHeader.parent = vol.parentId;
vol.stamp.magic = VOLUMEINFOMAGIC;
vol.stamp.version = VOLUMEINFOVERSION;
vol.destroyMe = DESTROY_ME;
snprintf(headerName, sizeof headerName, VFORMAT,
afs_printable_VolumeId_lu(vol.id));
snprintf(volumePath, sizeof volumePath, "%s" OS_DIRSEP "%s",
VPartitionPath(partition), headerName);
rc = stat(volumePath, &st);
if (rc == 0 || errno != ENOENT) {
if (rc == 0) {
Log("VCreateVolume: Header file %s already exists!\n",
volumePath);
*ec = VVOLEXISTS;
} else {
Log("VCreateVolume: Error %d trying to stat header file %s\n",
errno, volumePath);
*ec = VNOVOL;
}
goto bad_noheader;
}
device = partition->device;
for (i = 0; i < MAXINODETYPE; i++) {
struct afs_inode_info *p = &stuff[i];
if (p->obsolete)
continue;
#ifdef AFS_NAMEI_ENV
*(p->inode) =
IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
(p->inodeType == VI_LINKTABLE) ? vol.parentId : vol.id,
INODESPECIAL, p->inodeType, vol.parentId);
if (!(VALID_INO(*(p->inode)))) {
if (errno == EEXIST && (p->inodeType == VI_LINKTABLE)) {
/* Increment the reference count instead. */
IHandle_t *lh;
int code;
*(p->inode) = namei_MakeSpecIno(vol.parentId, VI_LINKTABLE);
IH_INIT(lh, device, parentId, *(p->inode));
fdP = IH_OPEN(lh);
if (fdP == NULL) {
IH_RELEASE(lh);
goto bad;
}
code = IH_INC(lh, *(p->inode), parentId);
FDH_REALLYCLOSE(fdP);
IH_RELEASE(lh);
if (code < 0)
goto bad;
continue;
}
}
#else
*(p->inode) =
IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
vol.id, INODESPECIAL, p->inodeType, vol.parentId);
#endif
if (!VALID_INO(*(p->inode))) {
Log("VCreateVolume: Problem creating %s file associated with volume header %s\n", p->description, volumePath);
bad:
if (handle)
IH_RELEASE(handle);
RemoveInodes(stuff, device, vol.parentId, vol.id);
if (!*ec) {
*ec = VNOVOL;
}
VDestroyVolumeDiskHeader(partition, volumeId, parentId);
bad_noheader:
# ifdef AFS_DEMAND_ATTACH_FS
if (locktype) {
VUnlockVolumeById(volumeId, partition);
}
# endif /* AFS_DEMAND_ATTACH_FS */
return NULL;
}
IH_INIT(handle, device, vol.parentId, *(p->inode));
fdP = IH_OPEN(handle);
if (fdP == NULL) {
Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
PrintInode(stmp, *(p->inode)), errno);
goto bad;
}
if (FDH_PWRITE(fdP, (char *)&p->stamp, sizeof(p->stamp), 0) !=
sizeof(p->stamp)) {
Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
PrintInode(stmp, *(p->inode)), errno);
FDH_REALLYCLOSE(fdP);
goto bad;
}
FDH_REALLYCLOSE(fdP);
IH_RELEASE(handle);
nearInode = *(p->inode);
}
IH_INIT(handle, device, vol.parentId, tempHeader.volumeInfo);
fdP = IH_OPEN(handle);
if (fdP == NULL) {
Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
PrintInode(stmp, tempHeader.volumeInfo), errno);
goto bad;
}
if (FDH_PWRITE(fdP, (char *)&vol, sizeof(vol), 0) != sizeof(vol)) {
Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
PrintInode(stmp, tempHeader.volumeInfo), errno);
FDH_REALLYCLOSE(fdP);
goto bad;
}
FDH_CLOSE(fdP);
IH_RELEASE(handle);
VolumeHeaderToDisk(&diskHeader, &tempHeader);
rc = VCreateVolumeDiskHeader(&diskHeader, partition);
if (rc) {
Log("VCreateVolume: Error %d trying to write volume header for "
"volume %" AFS_VOLID_FMT " on partition %s; volume not created\n", rc,
afs_printable_VolumeId_lu(vol.id), VPartitionPath(partition));
if (rc == EEXIST) {
*ec = VVOLEXISTS;
}
goto bad;
}
# ifdef AFS_DEMAND_ATTACH_FS
if (locktype) {
VUnlockVolumeById(volumeId, partition);
}
# endif /* AFS_DEMAND_ATTACH_FS */
return (VAttachVolumeByName_r(ec, partname, headerName, V_SECRETLY));
}
/**
* add an entry to the volume group cache.
*
* @param[in] dp disk partition object
* @param[in] parent parent volume id
* @param[in] child child volume id
* @param[out] newvg if non-NULL, *newvg is 1 if adding this added a
* new VG, 0 if we added to an existing VG
*
* @pre VOL_LOCK held
*
* @return operation status
* @retval 0 success
* @retval -1 parent and child are already registered in
* different VGs
*/
int
VVGCache_entry_add_r(struct DiskPartition64 * dp,
VolumeId parent,
VolumeId child,
afs_int32 *newvg)
{
int code = 0, res;
VVGCache_entry_t * child_ent, * parent_ent;
if (newvg) {
*newvg = 0;
}
/* check for existing entries */
res = _VVGC_lookup(dp, child, &child_ent, NULL);
if (res && res != ENOENT) {
code = res;
goto done;
}
res = _VVGC_lookup(dp, parent, &parent_ent, NULL);
if (res && res != ENOENT) {
code = res;
goto done;
}
/*
* branch based upon existence of parent and child nodes
*/
if (parent_ent && child_ent) {
/* both exist. we're done.
* if they point different places, then report the error. */
if (child_ent != parent_ent) {
code = -1;
}
if (parent == child) {
/* if we're adding the RW entry as a child, the RW id may
* not be in the child array yet, so make sure not to skip
* over that */
goto cladd;
}
goto done;
} else if (!parent_ent && child_ent) {
/* child exists.
* update vg root volid, and add hash entry. */
parent_ent = child_ent;
parent_ent->rw = parent;
code = _VVGC_hash_entry_add(dp,
parent,
parent_ent,
NULL);
goto done;
} else if (!child_ent && !parent_ent) {
code = _VVGC_entry_add(dp,
parent,
&parent_ent,
NULL);
if (code) {
goto done;
}
if (newvg) {
*newvg = 1;
}
if (child == parent) {
/* if we're the RW, skip over adding the child hash entry;
* we already added the hash entry when creating the entry */
child_ent = parent_ent;
goto cladd;
}
}
osi_Assert(!child_ent);
child_ent = parent_ent;
code = _VVGC_hash_entry_add(dp,
child,
child_ent,
NULL);
if (code) {
goto done;
}
cladd:
code = _VVGC_entry_cl_add(child_ent, child);
done:
if (code && code != EINVAL) {
ViceLog(0, ("VVGCache_entry_add: error %d trying to add vol %lu to VG"
" %lu on partition %s", code, afs_printable_uint32_lu(child),
afs_printable_uint32_lu(parent), VPartitionPath(dp)));
}
if (code == 0 && VVGCache.part[dp->index].state == VVGC_PART_STATE_UPDATING) {
/* we successfully added the entry; make sure it's not on the
* to-delete list, so it doesn't get deleted later */
code = _VVGC_dlist_del_r(dp, parent, child);
if (code && code != ENOENT) {
ViceLog(0, ("VVGCache_entry_add: error %d trying to remove vol "
"%lu (parent %lu) from the to-delete list for part "
"%s.\n", code, afs_printable_uint32_lu(child),
afs_printable_uint32_lu(parent),
VPartitionPath(dp)));
} else {
code = 0;
}
}
return code;
}
