diff options
author | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 15:20:36 -0700 |
---|---|---|
committer | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 15:20:36 -0700 |
commit | 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch) | |
tree | 0bba044c4ce775e45a88a51686b5d9f90697ea9d /fs/afs/fsclient.c | |
download | kernel-crypto-1da177e4c3f41524e886b7f1b8a0c1fc7321cac2.tar.gz kernel-crypto-1da177e4c3f41524e886b7f1b8a0c1fc7321cac2.tar.xz kernel-crypto-1da177e4c3f41524e886b7f1b8a0c1fc7321cac2.zip |
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Diffstat (limited to 'fs/afs/fsclient.c')
-rw-r--r-- | fs/afs/fsclient.c | 837 |
1 files changed, 837 insertions, 0 deletions
diff --git a/fs/afs/fsclient.c b/fs/afs/fsclient.c new file mode 100644 index 00000000000..61bc371532a --- /dev/null +++ b/fs/afs/fsclient.c @@ -0,0 +1,837 @@ +/* fsclient.c: AFS File Server client stubs + * + * Copyright (C) 2002 Red Hat, Inc. All Rights Reserved. + * Written by David Howells (dhowells@redhat.com) + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version + * 2 of the License, or (at your option) any later version. + */ + +#include <linux/init.h> +#include <linux/sched.h> +#include <rxrpc/rxrpc.h> +#include <rxrpc/transport.h> +#include <rxrpc/connection.h> +#include <rxrpc/call.h> +#include "fsclient.h" +#include "cmservice.h" +#include "vnode.h" +#include "server.h" +#include "errors.h" +#include "internal.h" + +#define FSFETCHSTATUS 132 /* AFS Fetch file status */ +#define FSFETCHDATA 130 /* AFS Fetch file data */ +#define FSGIVEUPCALLBACKS 147 /* AFS Discard callback promises */ +#define FSGETVOLUMEINFO 148 /* AFS Get root volume information */ +#define FSGETROOTVOLUME 151 /* AFS Get root volume name */ +#define FSLOOKUP 161 /* AFS lookup file in directory */ + +/*****************************************************************************/ +/* + * map afs abort codes to/from Linux error codes + * - called with call->lock held + */ +static void afs_rxfs_aemap(struct rxrpc_call *call) +{ + switch (call->app_err_state) { + case RXRPC_ESTATE_LOCAL_ABORT: + call->app_abort_code = -call->app_errno; + break; + case RXRPC_ESTATE_PEER_ABORT: + call->app_errno = afs_abort_to_error(call->app_abort_code); + break; + default: + break; + } +} /* end afs_rxfs_aemap() */ + +/*****************************************************************************/ +/* + * get the root volume name from a fileserver + * - this operation doesn't seem to work correctly in OpenAFS server 1.2.2 + */ +#if 0 +int afs_rxfs_get_root_volume(struct afs_server *server, + char *buf, size_t *buflen) +{ + struct rxrpc_connection *conn; + struct rxrpc_call *call; + struct kvec piov[2]; + size_t sent; + int ret; + u32 param[1]; + + DECLARE_WAITQUEUE(myself, current); + + kenter("%p,%p,%u",server, buf, *buflen); + + /* get hold of the fileserver connection */ + ret = afs_server_get_fsconn(server, &conn); + if (ret < 0) + goto out; + + /* create a call through that connection */ + ret = rxrpc_create_call(conn, NULL, NULL, afs_rxfs_aemap, &call); + if (ret < 0) { + printk("kAFS: Unable to create call: %d\n", ret); + goto out_put_conn; + } + call->app_opcode = FSGETROOTVOLUME; + + /* we want to get event notifications from the call */ + add_wait_queue(&call->waitq, &myself); + + /* marshall the parameters */ + param[0] = htonl(FSGETROOTVOLUME); + + piov[0].iov_len = sizeof(param); + piov[0].iov_base = param; + + /* send the parameters to the server */ + ret = rxrpc_call_write_data(call, 1, piov, RXRPC_LAST_PACKET, GFP_NOFS, + 0, &sent); + if (ret < 0) + goto abort; + + /* wait for the reply to completely arrive */ + for (;;) { + set_current_state(TASK_INTERRUPTIBLE); + if (call->app_call_state != RXRPC_CSTATE_CLNT_RCV_REPLY || + signal_pending(current)) + break; + schedule(); + } + set_current_state(TASK_RUNNING); + + ret = -EINTR; + if (signal_pending(current)) + goto abort; + + switch (call->app_call_state) { + case RXRPC_CSTATE_ERROR: + ret = call->app_errno; + kdebug("Got Error: %d", ret); + goto out_unwait; + + case RXRPC_CSTATE_CLNT_GOT_REPLY: + /* read the reply */ + kdebug("Got Reply: qty=%d", call->app_ready_qty); + + ret = -EBADMSG; + if (call->app_ready_qty <= 4) + goto abort; + + ret = rxrpc_call_read_data(call, NULL, call->app_ready_qty, 0); + if (ret < 0) + goto abort; + +#if 0 + /* unmarshall the reply */ + bp = buffer; + for (loop = 0; loop < 65; loop++) + entry->name[loop] = ntohl(*bp++); + entry->name[64] = 0; + + entry->type = ntohl(*bp++); + entry->num_servers = ntohl(*bp++); + + for (loop = 0; loop < 8; loop++) + entry->servers[loop].addr.s_addr = *bp++; + + for (loop = 0; loop < 8; loop++) + entry->servers[loop].partition = ntohl(*bp++); + + for (loop = 0; loop < 8; loop++) + entry->servers[loop].flags = ntohl(*bp++); + + for (loop = 0; loop < 3; loop++) + entry->volume_ids[loop] = ntohl(*bp++); + + entry->clone_id = ntohl(*bp++); + entry->flags = ntohl(*bp); +#endif + + /* success */ + ret = 0; + goto out_unwait; + + default: + BUG(); + } + + abort: + set_current_state(TASK_UNINTERRUPTIBLE); + rxrpc_call_abort(call, ret); + schedule(); + out_unwait: + set_current_state(TASK_RUNNING); + remove_wait_queue(&call->waitq, &myself); + rxrpc_put_call(call); + out_put_conn: + afs_server_release_fsconn(server, conn); + out: + kleave(""); + return ret; +} /* end afs_rxfs_get_root_volume() */ +#endif + +/*****************************************************************************/ +/* + * get information about a volume + */ +#if 0 +int afs_rxfs_get_volume_info(struct afs_server *server, + const char *name, + struct afs_volume_info *vinfo) +{ + struct rxrpc_connection *conn; + struct rxrpc_call *call; + struct kvec piov[3]; + size_t sent; + int ret; + u32 param[2], *bp, zero; + + DECLARE_WAITQUEUE(myself, current); + + _enter("%p,%s,%p", server, name, vinfo); + + /* get hold of the fileserver connection */ + ret = afs_server_get_fsconn(server, &conn); + if (ret < 0) + goto out; + + /* create a call through that connection */ + ret = rxrpc_create_call(conn, NULL, NULL, afs_rxfs_aemap, &call); + if (ret < 0) { + printk("kAFS: Unable to create call: %d\n", ret); + goto out_put_conn; + } + call->app_opcode = FSGETVOLUMEINFO; + + /* we want to get event notifications from the call */ + add_wait_queue(&call->waitq, &myself); + + /* marshall the parameters */ + piov[1].iov_len = strlen(name); + piov[1].iov_base = (char *) name; + + zero = 0; + piov[2].iov_len = (4 - (piov[1].iov_len & 3)) & 3; + piov[2].iov_base = &zero; + + param[0] = htonl(FSGETVOLUMEINFO); + param[1] = htonl(piov[1].iov_len); + + piov[0].iov_len = sizeof(param); + piov[0].iov_base = param; + + /* send the parameters to the server */ + ret = rxrpc_call_write_data(call, 3, piov, RXRPC_LAST_PACKET, GFP_NOFS, + 0, &sent); + if (ret < 0) + goto abort; + + /* wait for the reply to completely arrive */ + bp = rxrpc_call_alloc_scratch(call, 64); + + ret = rxrpc_call_read_data(call, bp, 64, + RXRPC_CALL_READ_BLOCK | + RXRPC_CALL_READ_ALL); + if (ret < 0) { + if (ret == -ECONNABORTED) { + ret = call->app_errno; + goto out_unwait; + } + goto abort; + } + + /* unmarshall the reply */ + vinfo->vid = ntohl(*bp++); + vinfo->type = ntohl(*bp++); + + vinfo->type_vids[0] = ntohl(*bp++); + vinfo->type_vids[1] = ntohl(*bp++); + vinfo->type_vids[2] = ntohl(*bp++); + vinfo->type_vids[3] = ntohl(*bp++); + vinfo->type_vids[4] = ntohl(*bp++); + + vinfo->nservers = ntohl(*bp++); + vinfo->servers[0].addr.s_addr = *bp++; + vinfo->servers[1].addr.s_addr = *bp++; + vinfo->servers[2].addr.s_addr = *bp++; + vinfo->servers[3].addr.s_addr = *bp++; + vinfo->servers[4].addr.s_addr = *bp++; + vinfo->servers[5].addr.s_addr = *bp++; + vinfo->servers[6].addr.s_addr = *bp++; + vinfo->servers[7].addr.s_addr = *bp++; + + ret = -EBADMSG; + if (vinfo->nservers > 8) + goto abort; + + /* success */ + ret = 0; + + out_unwait: + set_current_state(TASK_RUNNING); + remove_wait_queue(&call->waitq, &myself); + rxrpc_put_call(call); + out_put_conn: + afs_server_release_fsconn(server, conn); + out: + _leave(""); + return ret; + + abort: + set_current_state(TASK_UNINTERRUPTIBLE); + rxrpc_call_abort(call, ret); + schedule(); + goto out_unwait; + +} /* end afs_rxfs_get_volume_info() */ +#endif + +/*****************************************************************************/ +/* + * fetch the status information for a file + */ +int afs_rxfs_fetch_file_status(struct afs_server *server, + struct afs_vnode *vnode, + struct afs_volsync *volsync) +{ + struct afs_server_callslot callslot; + struct rxrpc_call *call; + struct kvec piov[1]; + size_t sent; + int ret; + __be32 *bp; + + DECLARE_WAITQUEUE(myself, current); + + _enter("%p,{%u,%u,%u}", + server, vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique); + + /* get hold of the fileserver connection */ + ret = afs_server_request_callslot(server, &callslot); + if (ret < 0) + goto out; + + /* create a call through that connection */ + ret = rxrpc_create_call(callslot.conn, NULL, NULL, afs_rxfs_aemap, + &call); + if (ret < 0) { + printk("kAFS: Unable to create call: %d\n", ret); + goto out_put_conn; + } + call->app_opcode = FSFETCHSTATUS; + + /* we want to get event notifications from the call */ + add_wait_queue(&call->waitq, &myself); + + /* marshall the parameters */ + bp = rxrpc_call_alloc_scratch(call, 16); + bp[0] = htonl(FSFETCHSTATUS); + bp[1] = htonl(vnode->fid.vid); + bp[2] = htonl(vnode->fid.vnode); + bp[3] = htonl(vnode->fid.unique); + + piov[0].iov_len = 16; + piov[0].iov_base = bp; + + /* send the parameters to the server */ + ret = rxrpc_call_write_data(call, 1, piov, RXRPC_LAST_PACKET, GFP_NOFS, + 0, &sent); + if (ret < 0) + goto abort; + + /* wait for the reply to completely arrive */ + bp = rxrpc_call_alloc_scratch(call, 120); + + ret = rxrpc_call_read_data(call, bp, 120, + RXRPC_CALL_READ_BLOCK | + RXRPC_CALL_READ_ALL); + if (ret < 0) { + if (ret == -ECONNABORTED) { + ret = call->app_errno; + goto out_unwait; + } + goto abort; + } + + /* unmarshall the reply */ + vnode->status.if_version = ntohl(*bp++); + vnode->status.type = ntohl(*bp++); + vnode->status.nlink = ntohl(*bp++); + vnode->status.size = ntohl(*bp++); + vnode->status.version = ntohl(*bp++); + vnode->status.author = ntohl(*bp++); + vnode->status.owner = ntohl(*bp++); + vnode->status.caller_access = ntohl(*bp++); + vnode->status.anon_access = ntohl(*bp++); + vnode->status.mode = ntohl(*bp++); + vnode->status.parent.vid = vnode->fid.vid; + vnode->status.parent.vnode = ntohl(*bp++); + vnode->status.parent.unique = ntohl(*bp++); + bp++; /* seg size */ + vnode->status.mtime_client = ntohl(*bp++); + vnode->status.mtime_server = ntohl(*bp++); + bp++; /* group */ + bp++; /* sync counter */ + vnode->status.version |= ((unsigned long long) ntohl(*bp++)) << 32; + bp++; /* spare2 */ + bp++; /* spare3 */ + bp++; /* spare4 */ + + vnode->cb_version = ntohl(*bp++); + vnode->cb_expiry = ntohl(*bp++); + vnode->cb_type = ntohl(*bp++); + + if (volsync) { + volsync->creation = ntohl(*bp++); + bp++; /* spare2 */ + bp++; /* spare3 */ + bp++; /* spare4 */ + bp++; /* spare5 */ + bp++; /* spare6 */ + } + + /* success */ + ret = 0; + + out_unwait: + set_current_state(TASK_RUNNING); + remove_wait_queue(&call->waitq, &myself); + rxrpc_put_call(call); + out_put_conn: + afs_server_release_callslot(server, &callslot); + out: + _leave(""); + return ret; + + abort: + set_current_state(TASK_UNINTERRUPTIBLE); + rxrpc_call_abort(call, ret); + schedule(); + goto out_unwait; +} /* end afs_rxfs_fetch_file_status() */ + +/*****************************************************************************/ +/* + * fetch the contents of a file or directory + */ +int afs_rxfs_fetch_file_data(struct afs_server *server, + struct afs_vnode *vnode, + struct afs_rxfs_fetch_descriptor *desc, + struct afs_volsync *volsync) +{ + struct afs_server_callslot callslot; + struct rxrpc_call *call; + struct kvec piov[1]; + size_t sent; + int ret; + __be32 *bp; + + DECLARE_WAITQUEUE(myself, current); + + _enter("%p,{fid={%u,%u,%u},sz=%Zu,of=%lu}", + server, + desc->fid.vid, + desc->fid.vnode, + desc->fid.unique, + desc->size, + desc->offset); + + /* get hold of the fileserver connection */ + ret = afs_server_request_callslot(server, &callslot); + if (ret < 0) + goto out; + + /* create a call through that connection */ + ret = rxrpc_create_call(callslot.conn, NULL, NULL, afs_rxfs_aemap, &call); + if (ret < 0) { + printk("kAFS: Unable to create call: %d\n", ret); + goto out_put_conn; + } + call->app_opcode = FSFETCHDATA; + + /* we want to get event notifications from the call */ + add_wait_queue(&call->waitq, &myself); + + /* marshall the parameters */ + bp = rxrpc_call_alloc_scratch(call, 24); + bp[0] = htonl(FSFETCHDATA); + bp[1] = htonl(desc->fid.vid); + bp[2] = htonl(desc->fid.vnode); + bp[3] = htonl(desc->fid.unique); + bp[4] = htonl(desc->offset); + bp[5] = htonl(desc->size); + + piov[0].iov_len = 24; + piov[0].iov_base = bp; + + /* send the parameters to the server */ + ret = rxrpc_call_write_data(call, 1, piov, RXRPC_LAST_PACKET, GFP_NOFS, + 0, &sent); + if (ret < 0) + goto abort; + + /* wait for the data count to arrive */ + ret = rxrpc_call_read_data(call, bp, 4, RXRPC_CALL_READ_BLOCK); + if (ret < 0) + goto read_failed; + + desc->actual = ntohl(bp[0]); + if (desc->actual != desc->size) { + ret = -EBADMSG; + goto abort; + } + + /* call the app to read the actual data */ + rxrpc_call_reset_scratch(call); + + ret = rxrpc_call_read_data(call, desc->buffer, desc->actual, + RXRPC_CALL_READ_BLOCK); + if (ret < 0) + goto read_failed; + + /* wait for the rest of the reply to completely arrive */ + rxrpc_call_reset_scratch(call); + bp = rxrpc_call_alloc_scratch(call, 120); + + ret = rxrpc_call_read_data(call, bp, 120, + RXRPC_CALL_READ_BLOCK | + RXRPC_CALL_READ_ALL); + if (ret < 0) + goto read_failed; + + /* unmarshall the reply */ + vnode->status.if_version = ntohl(*bp++); + vnode->status.type = ntohl(*bp++); + vnode->status.nlink = ntohl(*bp++); + vnode->status.size = ntohl(*bp++); + vnode->status.version = ntohl(*bp++); + vnode->status.author = ntohl(*bp++); + vnode->status.owner = ntohl(*bp++); + vnode->status.caller_access = ntohl(*bp++); + vnode->status.anon_access = ntohl(*bp++); + vnode->status.mode = ntohl(*bp++); + vnode->status.parent.vid = desc->fid.vid; + vnode->status.parent.vnode = ntohl(*bp++); + vnode->status.parent.unique = ntohl(*bp++); + bp++; /* seg size */ + vnode->status.mtime_client = ntohl(*bp++); + vnode->status.mtime_server = ntohl(*bp++); + bp++; /* group */ + bp++; /* sync counter */ + vnode->status.version |= ((unsigned long long) ntohl(*bp++)) << 32; + bp++; /* spare2 */ + bp++; /* spare3 */ + bp++; /* spare4 */ + + vnode->cb_version = ntohl(*bp++); + vnode->cb_expiry = ntohl(*bp++); + vnode->cb_type = ntohl(*bp++); + + if (volsync) { + volsync->creation = ntohl(*bp++); + bp++; /* spare2 */ + bp++; /* spare3 */ + bp++; /* spare4 */ + bp++; /* spare5 */ + bp++; /* spare6 */ + } + + /* success */ + ret = 0; + + out_unwait: + set_current_state(TASK_RUNNING); + remove_wait_queue(&call->waitq,&myself); + rxrpc_put_call(call); + out_put_conn: + afs_server_release_callslot(server, &callslot); + out: + _leave(" = %d", ret); + return ret; + + read_failed: + if (ret == -ECONNABORTED) { + ret = call->app_errno; + goto out_unwait; + } + + abort: + set_current_state(TASK_UNINTERRUPTIBLE); + rxrpc_call_abort(call, ret); + schedule(); + goto out_unwait; + +} /* end afs_rxfs_fetch_file_data() */ + +/*****************************************************************************/ +/* + * ask the AFS fileserver to discard a callback request on a file + */ +int afs_rxfs_give_up_callback(struct afs_server *server, + struct afs_vnode *vnode) +{ + struct afs_server_callslot callslot; + struct rxrpc_call *call; + struct kvec piov[1]; + size_t sent; + int ret; + __be32 *bp; + + DECLARE_WAITQUEUE(myself, current); + + _enter("%p,{%u,%u,%u}", + server, vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique); + + /* get hold of the fileserver connection */ + ret = afs_server_request_callslot(server, &callslot); + if (ret < 0) + goto out; + + /* create a call through that connection */ + ret = rxrpc_create_call(callslot.conn, NULL, NULL, afs_rxfs_aemap, &call); + if (ret < 0) { + printk("kAFS: Unable to create call: %d\n", ret); + goto out_put_conn; + } + call->app_opcode = FSGIVEUPCALLBACKS; + + /* we want to get event notifications from the call */ + add_wait_queue(&call->waitq, &myself); + + /* marshall the parameters */ + bp = rxrpc_call_alloc_scratch(call, (1 + 4 + 4) * 4); + + piov[0].iov_len = (1 + 4 + 4) * 4; + piov[0].iov_base = bp; + + *bp++ = htonl(FSGIVEUPCALLBACKS); + *bp++ = htonl(1); + *bp++ = htonl(vnode->fid.vid); + *bp++ = htonl(vnode->fid.vnode); + *bp++ = htonl(vnode->fid.unique); + *bp++ = htonl(1); + *bp++ = htonl(vnode->cb_version); + *bp++ = htonl(vnode->cb_expiry); + *bp++ = htonl(vnode->cb_type); + + /* send the parameters to the server */ + ret = rxrpc_call_write_data(call, 1, piov, RXRPC_LAST_PACKET, GFP_NOFS, + 0, &sent); + if (ret < 0) + goto abort; + + /* wait for the reply to completely arrive */ + for (;;) { + set_current_state(TASK_INTERRUPTIBLE); + if (call->app_call_state != RXRPC_CSTATE_CLNT_RCV_REPLY || + signal_pending(current)) + break; + schedule(); + } + set_current_state(TASK_RUNNING); + + ret = -EINTR; + if (signal_pending(current)) + goto abort; + + switch (call->app_call_state) { + case RXRPC_CSTATE_ERROR: + ret = call->app_errno; + goto out_unwait; + + case RXRPC_CSTATE_CLNT_GOT_REPLY: + ret = 0; + goto out_unwait; + + default: + BUG(); + } + + out_unwait: + set_current_state(TASK_RUNNING); + remove_wait_queue(&call->waitq, &myself); + rxrpc_put_call(call); + out_put_conn: + afs_server_release_callslot(server, &callslot); + out: + _leave(""); + return ret; + + abort: + set_current_state(TASK_UNINTERRUPTIBLE); + rxrpc_call_abort(call, ret); + schedule(); + goto out_unwait; +} /* end afs_rxfs_give_up_callback() */ + +/*****************************************************************************/ +/* + * look a filename up in a directory + * - this operation doesn't seem to work correctly in OpenAFS server 1.2.2 + */ +#if 0 +int afs_rxfs_lookup(struct afs_server *server, + struct afs_vnode *dir, + const char *filename, + struct afs_vnode *vnode, + struct afs_volsync *volsync) +{ + struct rxrpc_connection *conn; + struct rxrpc_call *call; + struct kvec piov[3]; + size_t sent; + int ret; + u32 *bp, zero; + + DECLARE_WAITQUEUE(myself, current); + + kenter("%p,{%u,%u,%u},%s", + server, fid->vid, fid->vnode, fid->unique, filename); + + /* get hold of the fileserver connection */ + ret = afs_server_get_fsconn(server, &conn); + if (ret < 0) + goto out; + + /* create a call through that connection */ + ret = rxrpc_create_call(conn, NULL, NULL, afs_rxfs_aemap, &call); + if (ret < 0) { + printk("kAFS: Unable to create call: %d\n", ret); + goto out_put_conn; + } + call->app_opcode = FSLOOKUP; + + /* we want to get event notifications from the call */ + add_wait_queue(&call->waitq,&myself); + + /* marshall the parameters */ + bp = rxrpc_call_alloc_scratch(call, 20); + + zero = 0; + + piov[0].iov_len = 20; + piov[0].iov_base = bp; + piov[1].iov_len = strlen(filename); + piov[1].iov_base = (char *) filename; + piov[2].iov_len = (4 - (piov[1].iov_len & 3)) & 3; + piov[2].iov_base = &zero; + + *bp++ = htonl(FSLOOKUP); + *bp++ = htonl(dirfid->vid); + *bp++ = htonl(dirfid->vnode); + *bp++ = htonl(dirfid->unique); + *bp++ = htonl(piov[1].iov_len); + + /* send the parameters to the server */ + ret = rxrpc_call_write_data(call, 3, piov, RXRPC_LAST_PACKET, GFP_NOFS, + 0, &sent); + if (ret < 0) + goto abort; + + /* wait for the reply to completely arrive */ + bp = rxrpc_call_alloc_scratch(call, 220); + + ret = rxrpc_call_read_data(call, bp, 220, + RXRPC_CALL_READ_BLOCK | + RXRPC_CALL_READ_ALL); + if (ret < 0) { + if (ret == -ECONNABORTED) { + ret = call->app_errno; + goto out_unwait; + } + goto abort; + } + + /* unmarshall the reply */ + fid->vid = ntohl(*bp++); + fid->vnode = ntohl(*bp++); + fid->unique = ntohl(*bp++); + + vnode->status.if_version = ntohl(*bp++); + vnode->status.type = ntohl(*bp++); + vnode->status.nlink = ntohl(*bp++); + vnode->status.size = ntohl(*bp++); + vnode->status.version = ntohl(*bp++); + vnode->status.author = ntohl(*bp++); + vnode->status.owner = ntohl(*bp++); + vnode->status.caller_access = ntohl(*bp++); + vnode->status.anon_access = ntohl(*bp++); + vnode->status.mode = ntohl(*bp++); + vnode->status.parent.vid = dirfid->vid; + vnode->status.parent.vnode = ntohl(*bp++); + vnode->status.parent.unique = ntohl(*bp++); + bp++; /* seg size */ + vnode->status.mtime_client = ntohl(*bp++); + vnode->status.mtime_server = ntohl(*bp++); + bp++; /* group */ + bp++; /* sync counter */ + vnode->status.version |= ((unsigned long long) ntohl(*bp++)) << 32; + bp++; /* spare2 */ + bp++; /* spare3 */ + bp++; /* spare4 */ + + dir->status.if_version = ntohl(*bp++); + dir->status.type = ntohl(*bp++); + dir->status.nlink = ntohl(*bp++); + dir->status.size = ntohl(*bp++); + dir->status.version = ntohl(*bp++); + dir->status.author = ntohl(*bp++); + dir->status.owner = ntohl(*bp++); + dir->status.caller_access = ntohl(*bp++); + dir->status.anon_access = ntohl(*bp++); + dir->status.mode = ntohl(*bp++); + dir->status.parent.vid = dirfid->vid; + dir->status.parent.vnode = ntohl(*bp++); + dir->status.parent.unique = ntohl(*bp++); + bp++; /* seg size */ + dir->status.mtime_client = ntohl(*bp++); + dir->status.mtime_server = ntohl(*bp++); + bp++; /* group */ + bp++; /* sync counter */ + dir->status.version |= ((unsigned long long) ntohl(*bp++)) << 32; + bp++; /* spare2 */ + bp++; /* spare3 */ + bp++; /* spare4 */ + + callback->fid = *fid; + callback->version = ntohl(*bp++); + callback->expiry = ntohl(*bp++); + callback->type = ntohl(*bp++); + + if (volsync) { + volsync->creation = ntohl(*bp++); + bp++; /* spare2 */ + bp++; /* spare3 */ + bp++; /* spare4 */ + bp++; /* spare5 */ + bp++; /* spare6 */ + } + + /* success */ + ret = 0; + + out_unwait: + set_current_state(TASK_RUNNING); + remove_wait_queue(&call->waitq, &myself); + rxrpc_put_call(call); + out_put_conn: + afs_server_release_fsconn(server, conn); + out: + kleave(""); + return ret; + + abort: + set_current_state(TASK_UNINTERRUPTIBLE); + rxrpc_call_abort(call, ret); + schedule(); + goto out_unwait; +} /* end afs_rxfs_lookup() */ +#endif |