David Howells | 9cc6fc5 | 2017-11-02 15:27:50 +0000 | [diff] [blame^] | 1 | /* Handle fileserver selection and rotation. |
| 2 | * |
| 3 | * Copyright (C) 2017 Red Hat, Inc. All Rights Reserved. |
| 4 | * Written by David Howells (dhowells@redhat.com) |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public Licence |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the Licence, or (at your option) any later version. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/kernel.h> |
| 13 | #include <linux/slab.h> |
| 14 | #include "internal.h" |
| 15 | |
| 16 | /* |
| 17 | * Initialise a filesystem server cursor for iterating over FS servers. |
| 18 | */ |
| 19 | void afs_init_fs_cursor(struct afs_fs_cursor *fc, struct afs_vnode *vnode) |
| 20 | { |
| 21 | memset(fc, 0, sizeof(*fc)); |
| 22 | } |
| 23 | |
| 24 | /* |
| 25 | * Set a filesystem server cursor for using a specific FS server. |
| 26 | */ |
| 27 | int afs_set_fs_cursor(struct afs_fs_cursor *fc, struct afs_vnode *vnode) |
| 28 | { |
| 29 | afs_init_fs_cursor(fc, vnode); |
| 30 | |
| 31 | read_seqlock_excl(&vnode->cb_lock); |
| 32 | if (vnode->cb_interest) { |
| 33 | if (vnode->cb_interest->server->fs_state == 0) |
| 34 | fc->server = afs_get_server(vnode->cb_interest->server); |
| 35 | else |
| 36 | fc->ac.error = vnode->cb_interest->server->fs_state; |
| 37 | } else { |
| 38 | fc->ac.error = -ESTALE; |
| 39 | } |
| 40 | read_sequnlock_excl(&vnode->cb_lock); |
| 41 | |
| 42 | return fc->ac.error; |
| 43 | } |
| 44 | |
| 45 | /* |
| 46 | * pick a server to use to try accessing this volume |
| 47 | * - returns with an elevated usage count on the server chosen |
| 48 | */ |
| 49 | bool afs_volume_pick_fileserver(struct afs_fs_cursor *fc, struct afs_vnode *vnode) |
| 50 | { |
| 51 | struct afs_volume *volume = vnode->volume; |
| 52 | struct afs_server *server; |
| 53 | int ret, state, loop; |
| 54 | |
| 55 | _enter("%s", volume->vlocation->vldb.name); |
| 56 | |
| 57 | /* stick with the server we're already using if we can */ |
| 58 | if (vnode->cb_interest && vnode->cb_interest->server->fs_state == 0) { |
| 59 | fc->server = afs_get_server(vnode->cb_interest->server); |
| 60 | goto set_server; |
| 61 | } |
| 62 | |
| 63 | down_read(&volume->server_sem); |
| 64 | |
| 65 | /* handle the no-server case */ |
| 66 | if (volume->nservers == 0) { |
| 67 | fc->ac.error = volume->rjservers ? -ENOMEDIUM : -ESTALE; |
| 68 | up_read(&volume->server_sem); |
| 69 | _leave(" = f [no servers %d]", fc->ac.error); |
| 70 | return false; |
| 71 | } |
| 72 | |
| 73 | /* basically, just search the list for the first live server and use |
| 74 | * that */ |
| 75 | ret = 0; |
| 76 | for (loop = 0; loop < volume->nservers; loop++) { |
| 77 | server = volume->servers[loop]; |
| 78 | state = server->fs_state; |
| 79 | |
| 80 | _debug("consider %d [%d]", loop, state); |
| 81 | |
| 82 | switch (state) { |
| 83 | case 0: |
| 84 | goto picked_server; |
| 85 | |
| 86 | case -ENETUNREACH: |
| 87 | if (ret == 0) |
| 88 | ret = state; |
| 89 | break; |
| 90 | |
| 91 | case -EHOSTUNREACH: |
| 92 | if (ret == 0 || |
| 93 | ret == -ENETUNREACH) |
| 94 | ret = state; |
| 95 | break; |
| 96 | |
| 97 | case -ECONNREFUSED: |
| 98 | if (ret == 0 || |
| 99 | ret == -ENETUNREACH || |
| 100 | ret == -EHOSTUNREACH) |
| 101 | ret = state; |
| 102 | break; |
| 103 | |
| 104 | default: |
| 105 | case -EREMOTEIO: |
| 106 | if (ret == 0 || |
| 107 | ret == -ENETUNREACH || |
| 108 | ret == -EHOSTUNREACH || |
| 109 | ret == -ECONNREFUSED) |
| 110 | ret = state; |
| 111 | break; |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | error: |
| 116 | fc->ac.error = ret; |
| 117 | |
| 118 | /* no available servers |
| 119 | * - TODO: handle the no active servers case better |
| 120 | */ |
| 121 | up_read(&volume->server_sem); |
| 122 | _leave(" = f [%d]", fc->ac.error); |
| 123 | return false; |
| 124 | |
| 125 | picked_server: |
| 126 | /* Found an apparently healthy server. We need to register an interest |
| 127 | * in receiving callbacks before we talk to it. |
| 128 | */ |
| 129 | ret = afs_register_server_cb_interest(vnode, |
| 130 | &volume->cb_interests[loop], server); |
| 131 | if (ret < 0) |
| 132 | goto error; |
| 133 | |
| 134 | fc->server = afs_get_server(server); |
| 135 | up_read(&volume->server_sem); |
| 136 | set_server: |
| 137 | fc->ac.alist = afs_get_addrlist(fc->server->addrs); |
| 138 | fc->ac.addr = &fc->ac.alist->addrs[0]; |
| 139 | _debug("USING SERVER: %pIS\n", &fc->ac.addr->transport); |
| 140 | _leave(" = t (picked %pIS)", &fc->ac.addr->transport); |
| 141 | return true; |
| 142 | } |
| 143 | |
| 144 | /* |
| 145 | * release a server after use |
| 146 | * - releases the ref on the server struct that was acquired by picking |
| 147 | * - records result of using a particular server to access a volume |
| 148 | * - return true to try again, false if okay or to issue error |
| 149 | * - the caller must release the server struct if result was false |
| 150 | */ |
| 151 | bool afs_iterate_fs_cursor(struct afs_fs_cursor *fc, |
| 152 | struct afs_vnode *vnode) |
| 153 | { |
| 154 | struct afs_volume *volume = vnode->volume; |
| 155 | struct afs_server *server = fc->server; |
| 156 | unsigned loop; |
| 157 | |
| 158 | _enter("%s,%pIS,%d", |
| 159 | volume->vlocation->vldb.name, &fc->ac.addr->transport, |
| 160 | fc->ac.error); |
| 161 | |
| 162 | switch (fc->ac.error) { |
| 163 | /* success */ |
| 164 | case 0: |
| 165 | server->fs_state = 0; |
| 166 | _leave(" = f"); |
| 167 | return false; |
| 168 | |
| 169 | /* the fileserver denied all knowledge of the volume */ |
| 170 | case -ENOMEDIUM: |
| 171 | down_write(&volume->server_sem); |
| 172 | |
| 173 | /* firstly, find where the server is in the active list (if it |
| 174 | * is) */ |
| 175 | for (loop = 0; loop < volume->nservers; loop++) |
| 176 | if (volume->servers[loop] == server) |
| 177 | goto present; |
| 178 | |
| 179 | /* no longer there - may have been discarded by another op */ |
| 180 | goto try_next_server_upw; |
| 181 | |
| 182 | present: |
| 183 | volume->nservers--; |
| 184 | memmove(&volume->servers[loop], |
| 185 | &volume->servers[loop + 1], |
| 186 | sizeof(volume->servers[loop]) * |
| 187 | (volume->nservers - loop)); |
| 188 | volume->servers[volume->nservers] = NULL; |
| 189 | afs_put_server(afs_v2net(vnode), server); |
| 190 | volume->rjservers++; |
| 191 | |
| 192 | if (volume->nservers > 0) |
| 193 | /* another server might acknowledge its existence */ |
| 194 | goto try_next_server_upw; |
| 195 | |
| 196 | /* handle the case where all the fileservers have rejected the |
| 197 | * volume |
| 198 | * - TODO: try asking the fileservers for volume information |
| 199 | * - TODO: contact the VL server again to see if the volume is |
| 200 | * no longer registered |
| 201 | */ |
| 202 | up_write(&volume->server_sem); |
| 203 | afs_put_server(afs_v2net(vnode), server); |
| 204 | fc->server = NULL; |
| 205 | _leave(" = f [completely rejected]"); |
| 206 | return false; |
| 207 | |
| 208 | /* problem reaching the server */ |
| 209 | case -ENETUNREACH: |
| 210 | case -EHOSTUNREACH: |
| 211 | case -ECONNREFUSED: |
| 212 | case -ETIME: |
| 213 | case -ETIMEDOUT: |
| 214 | case -EREMOTEIO: |
| 215 | /* mark the server as dead |
| 216 | * TODO: vary dead timeout depending on error |
| 217 | */ |
| 218 | spin_lock(&server->fs_lock); |
| 219 | if (!server->fs_state) { |
| 220 | server->fs_state = fc->ac.error; |
| 221 | printk("kAFS: SERVER DEAD state=%d\n", fc->ac.error); |
| 222 | } |
| 223 | spin_unlock(&server->fs_lock); |
| 224 | goto try_next_server; |
| 225 | |
| 226 | /* miscellaneous error */ |
| 227 | default: |
| 228 | case -ENOMEM: |
| 229 | case -ENONET: |
| 230 | /* tell the caller to accept the result */ |
| 231 | afs_put_server(afs_v2net(vnode), server); |
| 232 | fc->server = NULL; |
| 233 | _leave(" = f [local failure]"); |
| 234 | return false; |
| 235 | } |
| 236 | |
| 237 | /* tell the caller to loop around and try the next server */ |
| 238 | try_next_server_upw: |
| 239 | up_write(&volume->server_sem); |
| 240 | try_next_server: |
| 241 | afs_put_server(afs_v2net(vnode), server); |
| 242 | _leave(" = t [try next server]"); |
| 243 | return true; |
| 244 | } |
| 245 | |
| 246 | /* |
| 247 | * Clean up a fileserver cursor. |
| 248 | */ |
| 249 | int afs_end_fs_cursor(struct afs_fs_cursor *fc, struct afs_net *net) |
| 250 | { |
| 251 | afs_end_cursor(&fc->ac); |
| 252 | afs_put_server(net, fc->server); |
| 253 | return fc->ac.error; |
| 254 | } |