| 1 | /* |
| 2 | * GPL HEADER START |
| 3 | * |
| 4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License version 2 only, |
| 8 | * as published by the Free Software Foundation. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, but |
| 11 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | * General Public License version 2 for more details (a copy is included |
| 14 | * in the LICENSE file that accompanied this code). |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * version 2 along with this program; If not, see |
| 18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf |
| 19 | * |
| 20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
| 21 | * CA 95054 USA or visit www.sun.com if you need additional information or |
| 22 | * have any questions. |
| 23 | * |
| 24 | * GPL HEADER END |
| 25 | */ |
| 26 | /* |
| 27 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. |
| 28 | * Use is subject to license terms. |
| 29 | * |
| 30 | * Copyright (c) 2011, 2012, Intel Corporation. |
| 31 | */ |
| 32 | /* |
| 33 | * This file is part of Lustre, http://www.lustre.org/ |
| 34 | * Lustre is a trademark of Sun Microsystems, Inc. |
| 35 | * |
| 36 | * lustre/lov/lov_obd.c |
| 37 | * |
| 38 | * Author: Phil Schwan <phil@clusterfs.com> |
| 39 | * Author: Peter Braam <braam@clusterfs.com> |
| 40 | * Author: Mike Shaver <shaver@clusterfs.com> |
| 41 | * Author: Nathan Rutman <nathan@clusterfs.com> |
| 42 | */ |
| 43 | |
| 44 | #define DEBUG_SUBSYSTEM S_LOV |
| 45 | #include "../../include/linux/libcfs/libcfs.h" |
| 46 | |
| 47 | #include "../include/obd_support.h" |
| 48 | #include "../include/lustre_lib.h" |
| 49 | #include "../include/lustre_net.h" |
| 50 | #include "../include/lustre/lustre_idl.h" |
| 51 | #include "../include/lustre_dlm.h" |
| 52 | #include "../include/lustre_mds.h" |
| 53 | #include "../include/obd_class.h" |
| 54 | #include "../include/lprocfs_status.h" |
| 55 | #include "../include/lustre_param.h" |
| 56 | #include "../include/cl_object.h" |
| 57 | #include "../include/lclient.h" /* for cl_client_lru */ |
| 58 | #include "../include/lustre/ll_fiemap.h" |
| 59 | #include "../include/lustre_fid.h" |
| 60 | |
| 61 | #include "lov_internal.h" |
| 62 | |
| 63 | /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion. |
| 64 | Any function that expects lov_tgts to remain stationary must take a ref. */ |
| 65 | static void lov_getref(struct obd_device *obd) |
| 66 | { |
| 67 | struct lov_obd *lov = &obd->u.lov; |
| 68 | |
| 69 | /* nobody gets through here until lov_putref is done */ |
| 70 | mutex_lock(&lov->lov_lock); |
| 71 | atomic_inc(&lov->lov_refcount); |
| 72 | mutex_unlock(&lov->lov_lock); |
| 73 | return; |
| 74 | } |
| 75 | |
| 76 | static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt); |
| 77 | |
| 78 | static void lov_putref(struct obd_device *obd) |
| 79 | { |
| 80 | struct lov_obd *lov = &obd->u.lov; |
| 81 | |
| 82 | mutex_lock(&lov->lov_lock); |
| 83 | /* ok to dec to 0 more than once -- ltd_exp's will be null */ |
| 84 | if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) { |
| 85 | LIST_HEAD(kill); |
| 86 | int i; |
| 87 | struct lov_tgt_desc *tgt, *n; |
| 88 | CDEBUG(D_CONFIG, "destroying %d lov targets\n", |
| 89 | lov->lov_death_row); |
| 90 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 91 | tgt = lov->lov_tgts[i]; |
| 92 | |
| 93 | if (!tgt || !tgt->ltd_reap) |
| 94 | continue; |
| 95 | list_add(&tgt->ltd_kill, &kill); |
| 96 | /* XXX - right now there is a dependency on ld_tgt_count |
| 97 | * being the maximum tgt index for computing the |
| 98 | * mds_max_easize. So we can't shrink it. */ |
| 99 | lov_ost_pool_remove(&lov->lov_packed, i); |
| 100 | lov->lov_tgts[i] = NULL; |
| 101 | lov->lov_death_row--; |
| 102 | } |
| 103 | mutex_unlock(&lov->lov_lock); |
| 104 | |
| 105 | list_for_each_entry_safe(tgt, n, &kill, ltd_kill) { |
| 106 | list_del(&tgt->ltd_kill); |
| 107 | /* Disconnect */ |
| 108 | __lov_del_obd(obd, tgt); |
| 109 | } |
| 110 | |
| 111 | if (lov->lov_tgts_kobj) |
| 112 | kobject_put(lov->lov_tgts_kobj); |
| 113 | |
| 114 | } else { |
| 115 | mutex_unlock(&lov->lov_lock); |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid, |
| 120 | enum obd_notify_event ev); |
| 121 | static int lov_notify(struct obd_device *obd, struct obd_device *watched, |
| 122 | enum obd_notify_event ev, void *data); |
| 123 | |
| 124 | |
| 125 | #define MAX_STRING_SIZE 128 |
| 126 | int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, |
| 127 | struct obd_connect_data *data) |
| 128 | { |
| 129 | struct lov_obd *lov = &obd->u.lov; |
| 130 | struct obd_uuid *tgt_uuid; |
| 131 | struct obd_device *tgt_obd; |
| 132 | static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" }; |
| 133 | struct obd_import *imp; |
| 134 | int rc; |
| 135 | |
| 136 | if (!lov->lov_tgts[index]) |
| 137 | return -EINVAL; |
| 138 | |
| 139 | tgt_uuid = &lov->lov_tgts[index]->ltd_uuid; |
| 140 | tgt_obd = lov->lov_tgts[index]->ltd_obd; |
| 141 | |
| 142 | if (!tgt_obd->obd_set_up) { |
| 143 | CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid)); |
| 144 | return -EINVAL; |
| 145 | } |
| 146 | |
| 147 | /* override the sp_me from lov */ |
| 148 | tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me; |
| 149 | |
| 150 | if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX)) |
| 151 | data->ocd_index = index; |
| 152 | |
| 153 | /* |
| 154 | * Divine LOV knows that OBDs under it are OSCs. |
| 155 | */ |
| 156 | imp = tgt_obd->u.cli.cl_import; |
| 157 | |
| 158 | if (activate) { |
| 159 | tgt_obd->obd_no_recov = 0; |
| 160 | /* FIXME this is probably supposed to be |
| 161 | ptlrpc_set_import_active. Horrible naming. */ |
| 162 | ptlrpc_activate_import(imp); |
| 163 | } |
| 164 | |
| 165 | rc = obd_register_observer(tgt_obd, obd); |
| 166 | if (rc) { |
| 167 | CERROR("Target %s register_observer error %d\n", |
| 168 | obd_uuid2str(tgt_uuid), rc); |
| 169 | return rc; |
| 170 | } |
| 171 | |
| 172 | |
| 173 | if (imp->imp_invalid) { |
| 174 | CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n", |
| 175 | obd_uuid2str(tgt_uuid)); |
| 176 | return 0; |
| 177 | } |
| 178 | |
| 179 | rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd, |
| 180 | &lov_osc_uuid, data, NULL); |
| 181 | if (rc || !lov->lov_tgts[index]->ltd_exp) { |
| 182 | CERROR("Target %s connect error %d\n", |
| 183 | obd_uuid2str(tgt_uuid), rc); |
| 184 | return -ENODEV; |
| 185 | } |
| 186 | |
| 187 | lov->lov_tgts[index]->ltd_reap = 0; |
| 188 | |
| 189 | CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index, |
| 190 | obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in"); |
| 191 | |
| 192 | if (lov->lov_tgts_kobj) |
| 193 | /* Even if we failed, that's ok */ |
| 194 | rc = sysfs_create_link(lov->lov_tgts_kobj, &tgt_obd->obd_kobj, |
| 195 | tgt_obd->obd_name); |
| 196 | |
| 197 | return 0; |
| 198 | } |
| 199 | |
| 200 | static int lov_connect(const struct lu_env *env, |
| 201 | struct obd_export **exp, struct obd_device *obd, |
| 202 | struct obd_uuid *cluuid, struct obd_connect_data *data, |
| 203 | void *localdata) |
| 204 | { |
| 205 | struct lov_obd *lov = &obd->u.lov; |
| 206 | struct lov_tgt_desc *tgt; |
| 207 | struct lustre_handle conn; |
| 208 | int i, rc; |
| 209 | |
| 210 | CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects); |
| 211 | |
| 212 | rc = class_connect(&conn, obd, cluuid); |
| 213 | if (rc) |
| 214 | return rc; |
| 215 | |
| 216 | *exp = class_conn2export(&conn); |
| 217 | |
| 218 | /* Why should there ever be more than 1 connect? */ |
| 219 | lov->lov_connects++; |
| 220 | LASSERT(lov->lov_connects == 1); |
| 221 | |
| 222 | memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd)); |
| 223 | if (data) |
| 224 | lov->lov_ocd = *data; |
| 225 | |
| 226 | obd_getref(obd); |
| 227 | |
| 228 | lov->lov_tgts_kobj = kobject_create_and_add("target_obds", |
| 229 | &obd->obd_kobj); |
| 230 | |
| 231 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 232 | tgt = lov->lov_tgts[i]; |
| 233 | if (!tgt || obd_uuid_empty(&tgt->ltd_uuid)) |
| 234 | continue; |
| 235 | /* Flags will be lowest common denominator */ |
| 236 | rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd); |
| 237 | if (rc) { |
| 238 | CERROR("%s: lov connect tgt %d failed: %d\n", |
| 239 | obd->obd_name, i, rc); |
| 240 | continue; |
| 241 | } |
| 242 | /* connect to administrative disabled ost */ |
| 243 | if (!lov->lov_tgts[i]->ltd_exp) |
| 244 | continue; |
| 245 | |
| 246 | rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd, |
| 247 | OBD_NOTIFY_CONNECT, (void *)&i); |
| 248 | if (rc) { |
| 249 | CERROR("%s error sending notify %d\n", |
| 250 | obd->obd_name, rc); |
| 251 | } |
| 252 | } |
| 253 | obd_putref(obd); |
| 254 | |
| 255 | return 0; |
| 256 | } |
| 257 | |
| 258 | static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt) |
| 259 | { |
| 260 | struct lov_obd *lov = &obd->u.lov; |
| 261 | struct obd_device *osc_obd; |
| 262 | int rc; |
| 263 | |
| 264 | osc_obd = class_exp2obd(tgt->ltd_exp); |
| 265 | CDEBUG(D_CONFIG, "%s: disconnecting target %s\n", |
| 266 | obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL"); |
| 267 | |
| 268 | if (tgt->ltd_active) { |
| 269 | tgt->ltd_active = 0; |
| 270 | lov->desc.ld_active_tgt_count--; |
| 271 | tgt->ltd_exp->exp_obd->obd_inactive = 1; |
| 272 | } |
| 273 | |
| 274 | if (osc_obd) { |
| 275 | if (lov->lov_tgts_kobj) |
| 276 | sysfs_remove_link(lov->lov_tgts_kobj, |
| 277 | osc_obd->obd_name); |
| 278 | |
| 279 | /* Pass it on to our clients. |
| 280 | * XXX This should be an argument to disconnect, |
| 281 | * XXX not a back-door flag on the OBD. Ah well. |
| 282 | */ |
| 283 | osc_obd->obd_force = obd->obd_force; |
| 284 | osc_obd->obd_fail = obd->obd_fail; |
| 285 | osc_obd->obd_no_recov = obd->obd_no_recov; |
| 286 | } |
| 287 | |
| 288 | obd_register_observer(osc_obd, NULL); |
| 289 | |
| 290 | rc = obd_disconnect(tgt->ltd_exp); |
| 291 | if (rc) { |
| 292 | CERROR("Target %s disconnect error %d\n", |
| 293 | tgt->ltd_uuid.uuid, rc); |
| 294 | rc = 0; |
| 295 | } |
| 296 | |
| 297 | tgt->ltd_exp = NULL; |
| 298 | return 0; |
| 299 | } |
| 300 | |
| 301 | static int lov_disconnect(struct obd_export *exp) |
| 302 | { |
| 303 | struct obd_device *obd = class_exp2obd(exp); |
| 304 | struct lov_obd *lov = &obd->u.lov; |
| 305 | int i, rc; |
| 306 | |
| 307 | if (!lov->lov_tgts) |
| 308 | goto out; |
| 309 | |
| 310 | /* Only disconnect the underlying layers on the final disconnect. */ |
| 311 | lov->lov_connects--; |
| 312 | if (lov->lov_connects != 0) { |
| 313 | /* why should there be more than 1 connect? */ |
| 314 | CERROR("disconnect #%d\n", lov->lov_connects); |
| 315 | goto out; |
| 316 | } |
| 317 | |
| 318 | /* Let's hold another reference so lov_del_obd doesn't spin through |
| 319 | putref every time */ |
| 320 | obd_getref(obd); |
| 321 | |
| 322 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 323 | if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) { |
| 324 | /* Disconnection is the last we know about an obd */ |
| 325 | lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen); |
| 326 | } |
| 327 | } |
| 328 | |
| 329 | obd_putref(obd); |
| 330 | |
| 331 | out: |
| 332 | rc = class_disconnect(exp); /* bz 9811 */ |
| 333 | return rc; |
| 334 | } |
| 335 | |
| 336 | /* Error codes: |
| 337 | * |
| 338 | * -EINVAL : UUID can't be found in the LOV's target list |
| 339 | * -ENOTCONN: The UUID is found, but the target connection is bad (!) |
| 340 | * -EBADF : The UUID is found, but the OBD is the wrong type (!) |
| 341 | * any >= 0 : is log target index |
| 342 | */ |
| 343 | static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid, |
| 344 | enum obd_notify_event ev) |
| 345 | { |
| 346 | struct lov_obd *lov = &obd->u.lov; |
| 347 | struct lov_tgt_desc *tgt; |
| 348 | int index, activate, active; |
| 349 | |
| 350 | CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n", |
| 351 | lov, uuid->uuid, ev); |
| 352 | |
| 353 | obd_getref(obd); |
| 354 | for (index = 0; index < lov->desc.ld_tgt_count; index++) { |
| 355 | tgt = lov->lov_tgts[index]; |
| 356 | if (!tgt) |
| 357 | continue; |
| 358 | /* |
| 359 | * LU-642, initially inactive OSC could miss the obd_connect, |
| 360 | * we make up for it here. |
| 361 | */ |
| 362 | if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL && |
| 363 | obd_uuid_equals(uuid, &tgt->ltd_uuid)) { |
| 364 | struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"}; |
| 365 | |
| 366 | obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd, |
| 367 | &lov_osc_uuid, &lov->lov_ocd, NULL); |
| 368 | } |
| 369 | if (!tgt->ltd_exp) |
| 370 | continue; |
| 371 | |
| 372 | CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n", |
| 373 | index, obd_uuid2str(&tgt->ltd_uuid), |
| 374 | tgt->ltd_exp->exp_handle.h_cookie); |
| 375 | if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) |
| 376 | break; |
| 377 | } |
| 378 | |
| 379 | if (index == lov->desc.ld_tgt_count) { |
| 380 | index = -EINVAL; |
| 381 | goto out; |
| 382 | } |
| 383 | |
| 384 | if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) { |
| 385 | activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0; |
| 386 | |
| 387 | if (lov->lov_tgts[index]->ltd_activate == activate) { |
| 388 | CDEBUG(D_INFO, "OSC %s already %sactivate!\n", |
| 389 | uuid->uuid, activate ? "" : "de"); |
| 390 | } else { |
| 391 | lov->lov_tgts[index]->ltd_activate = activate; |
| 392 | CDEBUG(D_CONFIG, "%sactivate OSC %s\n", |
| 393 | activate ? "" : "de", obd_uuid2str(uuid)); |
| 394 | } |
| 395 | |
| 396 | } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) { |
| 397 | active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0; |
| 398 | |
| 399 | if (lov->lov_tgts[index]->ltd_active == active) { |
| 400 | CDEBUG(D_INFO, "OSC %s already %sactive!\n", |
| 401 | uuid->uuid, active ? "" : "in"); |
| 402 | goto out; |
| 403 | } else { |
| 404 | CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", |
| 405 | obd_uuid2str(uuid), active ? "" : "in"); |
| 406 | } |
| 407 | |
| 408 | lov->lov_tgts[index]->ltd_active = active; |
| 409 | if (active) { |
| 410 | lov->desc.ld_active_tgt_count++; |
| 411 | lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0; |
| 412 | } else { |
| 413 | lov->desc.ld_active_tgt_count--; |
| 414 | lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1; |
| 415 | } |
| 416 | } else { |
| 417 | CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid); |
| 418 | } |
| 419 | |
| 420 | out: |
| 421 | obd_putref(obd); |
| 422 | return index; |
| 423 | } |
| 424 | |
| 425 | static int lov_notify(struct obd_device *obd, struct obd_device *watched, |
| 426 | enum obd_notify_event ev, void *data) |
| 427 | { |
| 428 | int rc = 0; |
| 429 | struct lov_obd *lov = &obd->u.lov; |
| 430 | |
| 431 | down_read(&lov->lov_notify_lock); |
| 432 | if (!lov->lov_connects) { |
| 433 | up_read(&lov->lov_notify_lock); |
| 434 | return rc; |
| 435 | } |
| 436 | |
| 437 | if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE || |
| 438 | ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) { |
| 439 | struct obd_uuid *uuid; |
| 440 | |
| 441 | LASSERT(watched); |
| 442 | |
| 443 | if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) { |
| 444 | up_read(&lov->lov_notify_lock); |
| 445 | CERROR("unexpected notification of %s %s!\n", |
| 446 | watched->obd_type->typ_name, |
| 447 | watched->obd_name); |
| 448 | return -EINVAL; |
| 449 | } |
| 450 | uuid = &watched->u.cli.cl_target_uuid; |
| 451 | |
| 452 | /* Set OSC as active before notifying the observer, so the |
| 453 | * observer can use the OSC normally. |
| 454 | */ |
| 455 | rc = lov_set_osc_active(obd, uuid, ev); |
| 456 | if (rc < 0) { |
| 457 | up_read(&lov->lov_notify_lock); |
| 458 | CERROR("event(%d) of %s failed: %d\n", ev, |
| 459 | obd_uuid2str(uuid), rc); |
| 460 | return rc; |
| 461 | } |
| 462 | /* active event should be pass lov target index as data */ |
| 463 | data = &rc; |
| 464 | } |
| 465 | |
| 466 | /* Pass the notification up the chain. */ |
| 467 | if (watched) { |
| 468 | rc = obd_notify_observer(obd, watched, ev, data); |
| 469 | } else { |
| 470 | /* NULL watched means all osc's in the lov (only for syncs) */ |
| 471 | /* sync event should be send lov idx as data */ |
| 472 | struct lov_obd *lov = &obd->u.lov; |
| 473 | int i, is_sync; |
| 474 | |
| 475 | data = &i; |
| 476 | is_sync = (ev == OBD_NOTIFY_SYNC) || |
| 477 | (ev == OBD_NOTIFY_SYNC_NONBLOCK); |
| 478 | |
| 479 | obd_getref(obd); |
| 480 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 481 | if (!lov->lov_tgts[i]) |
| 482 | continue; |
| 483 | |
| 484 | /* don't send sync event if target not |
| 485 | * connected/activated */ |
| 486 | if (is_sync && !lov->lov_tgts[i]->ltd_active) |
| 487 | continue; |
| 488 | |
| 489 | rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd, |
| 490 | ev, data); |
| 491 | if (rc) { |
| 492 | CERROR("%s: notify %s of %s failed %d\n", |
| 493 | obd->obd_name, |
| 494 | obd->obd_observer->obd_name, |
| 495 | lov->lov_tgts[i]->ltd_obd->obd_name, |
| 496 | rc); |
| 497 | } |
| 498 | } |
| 499 | obd_putref(obd); |
| 500 | } |
| 501 | |
| 502 | up_read(&lov->lov_notify_lock); |
| 503 | return rc; |
| 504 | } |
| 505 | |
| 506 | static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, |
| 507 | __u32 index, int gen, int active) |
| 508 | { |
| 509 | struct lov_obd *lov = &obd->u.lov; |
| 510 | struct lov_tgt_desc *tgt; |
| 511 | struct obd_device *tgt_obd; |
| 512 | int rc; |
| 513 | |
| 514 | CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n", |
| 515 | uuidp->uuid, index, gen, active); |
| 516 | |
| 517 | if (gen <= 0) { |
| 518 | CERROR("request to add OBD %s with invalid generation: %d\n", |
| 519 | uuidp->uuid, gen); |
| 520 | return -EINVAL; |
| 521 | } |
| 522 | |
| 523 | tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, |
| 524 | &obd->obd_uuid); |
| 525 | if (tgt_obd == NULL) |
| 526 | return -EINVAL; |
| 527 | |
| 528 | mutex_lock(&lov->lov_lock); |
| 529 | |
| 530 | if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) { |
| 531 | tgt = lov->lov_tgts[index]; |
| 532 | CERROR("UUID %s already assigned at LOV target index %d\n", |
| 533 | obd_uuid2str(&tgt->ltd_uuid), index); |
| 534 | mutex_unlock(&lov->lov_lock); |
| 535 | return -EEXIST; |
| 536 | } |
| 537 | |
| 538 | if (index >= lov->lov_tgt_size) { |
| 539 | /* We need to reallocate the lov target array. */ |
| 540 | struct lov_tgt_desc **newtgts, **old = NULL; |
| 541 | __u32 newsize, oldsize = 0; |
| 542 | |
| 543 | newsize = max_t(__u32, lov->lov_tgt_size, 2); |
| 544 | while (newsize < index + 1) |
| 545 | newsize <<= 1; |
| 546 | newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS); |
| 547 | if (newtgts == NULL) { |
| 548 | mutex_unlock(&lov->lov_lock); |
| 549 | return -ENOMEM; |
| 550 | } |
| 551 | |
| 552 | if (lov->lov_tgt_size) { |
| 553 | memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) * |
| 554 | lov->lov_tgt_size); |
| 555 | old = lov->lov_tgts; |
| 556 | oldsize = lov->lov_tgt_size; |
| 557 | } |
| 558 | |
| 559 | lov->lov_tgts = newtgts; |
| 560 | lov->lov_tgt_size = newsize; |
| 561 | smp_rmb(); |
| 562 | kfree(old); |
| 563 | |
| 564 | CDEBUG(D_CONFIG, "tgts: %p size: %d\n", |
| 565 | lov->lov_tgts, lov->lov_tgt_size); |
| 566 | } |
| 567 | |
| 568 | tgt = kzalloc(sizeof(*tgt), GFP_NOFS); |
| 569 | if (!tgt) { |
| 570 | mutex_unlock(&lov->lov_lock); |
| 571 | return -ENOMEM; |
| 572 | } |
| 573 | |
| 574 | rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size); |
| 575 | if (rc) { |
| 576 | mutex_unlock(&lov->lov_lock); |
| 577 | kfree(tgt); |
| 578 | return rc; |
| 579 | } |
| 580 | |
| 581 | tgt->ltd_uuid = *uuidp; |
| 582 | tgt->ltd_obd = tgt_obd; |
| 583 | /* XXX - add a sanity check on the generation number. */ |
| 584 | tgt->ltd_gen = gen; |
| 585 | tgt->ltd_index = index; |
| 586 | tgt->ltd_activate = active; |
| 587 | lov->lov_tgts[index] = tgt; |
| 588 | if (index >= lov->desc.ld_tgt_count) |
| 589 | lov->desc.ld_tgt_count = index + 1; |
| 590 | |
| 591 | mutex_unlock(&lov->lov_lock); |
| 592 | |
| 593 | CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n", |
| 594 | index, tgt->ltd_gen, lov->desc.ld_tgt_count); |
| 595 | |
| 596 | rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index); |
| 597 | |
| 598 | if (lov->lov_connects == 0) { |
| 599 | /* lov_connect hasn't been called yet. We'll do the |
| 600 | lov_connect_obd on this target when that fn first runs, |
| 601 | because we don't know the connect flags yet. */ |
| 602 | return 0; |
| 603 | } |
| 604 | |
| 605 | obd_getref(obd); |
| 606 | |
| 607 | rc = lov_connect_obd(obd, index, active, &lov->lov_ocd); |
| 608 | if (rc) |
| 609 | goto out; |
| 610 | |
| 611 | /* connect to administrative disabled ost */ |
| 612 | if (!tgt->ltd_exp) { |
| 613 | rc = 0; |
| 614 | goto out; |
| 615 | } |
| 616 | |
| 617 | if (lov->lov_cache != NULL) { |
| 618 | rc = obd_set_info_async(NULL, tgt->ltd_exp, |
| 619 | sizeof(KEY_CACHE_SET), KEY_CACHE_SET, |
| 620 | sizeof(struct cl_client_cache), lov->lov_cache, |
| 621 | NULL); |
| 622 | if (rc < 0) |
| 623 | goto out; |
| 624 | } |
| 625 | |
| 626 | rc = lov_notify(obd, tgt->ltd_exp->exp_obd, |
| 627 | active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE, |
| 628 | (void *)&index); |
| 629 | |
| 630 | out: |
| 631 | if (rc) { |
| 632 | CERROR("add failed (%d), deleting %s\n", rc, |
| 633 | obd_uuid2str(&tgt->ltd_uuid)); |
| 634 | lov_del_target(obd, index, NULL, 0); |
| 635 | } |
| 636 | obd_putref(obd); |
| 637 | return rc; |
| 638 | } |
| 639 | |
| 640 | /* Schedule a target for deletion */ |
| 641 | int lov_del_target(struct obd_device *obd, __u32 index, |
| 642 | struct obd_uuid *uuidp, int gen) |
| 643 | { |
| 644 | struct lov_obd *lov = &obd->u.lov; |
| 645 | int count = lov->desc.ld_tgt_count; |
| 646 | int rc = 0; |
| 647 | |
| 648 | if (index >= count) { |
| 649 | CERROR("LOV target index %d >= number of LOV OBDs %d.\n", |
| 650 | index, count); |
| 651 | return -EINVAL; |
| 652 | } |
| 653 | |
| 654 | /* to make sure there's no ongoing lov_notify() now */ |
| 655 | down_write(&lov->lov_notify_lock); |
| 656 | obd_getref(obd); |
| 657 | |
| 658 | if (!lov->lov_tgts[index]) { |
| 659 | CERROR("LOV target at index %d is not setup.\n", index); |
| 660 | rc = -EINVAL; |
| 661 | goto out; |
| 662 | } |
| 663 | |
| 664 | if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) { |
| 665 | CERROR("LOV target UUID %s at index %d doesn't match %s.\n", |
| 666 | lov_uuid2str(lov, index), index, |
| 667 | obd_uuid2str(uuidp)); |
| 668 | rc = -EINVAL; |
| 669 | goto out; |
| 670 | } |
| 671 | |
| 672 | CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n", |
| 673 | lov_uuid2str(lov, index), index, |
| 674 | lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp, |
| 675 | lov->lov_tgts[index]->ltd_active); |
| 676 | |
| 677 | lov->lov_tgts[index]->ltd_reap = 1; |
| 678 | lov->lov_death_row++; |
| 679 | /* we really delete it from obd_putref */ |
| 680 | out: |
| 681 | obd_putref(obd); |
| 682 | up_write(&lov->lov_notify_lock); |
| 683 | |
| 684 | return rc; |
| 685 | } |
| 686 | |
| 687 | static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt) |
| 688 | { |
| 689 | struct obd_device *osc_obd; |
| 690 | |
| 691 | LASSERT(tgt); |
| 692 | LASSERT(tgt->ltd_reap); |
| 693 | |
| 694 | osc_obd = class_exp2obd(tgt->ltd_exp); |
| 695 | |
| 696 | CDEBUG(D_CONFIG, "Removing tgt %s : %s\n", |
| 697 | tgt->ltd_uuid.uuid, |
| 698 | osc_obd ? osc_obd->obd_name : "<no obd>"); |
| 699 | |
| 700 | if (tgt->ltd_exp) |
| 701 | lov_disconnect_obd(obd, tgt); |
| 702 | |
| 703 | kfree(tgt); |
| 704 | |
| 705 | /* Manual cleanup - no cleanup logs to clean up the osc's. We must |
| 706 | do it ourselves. And we can't do it from lov_cleanup, |
| 707 | because we just lost our only reference to it. */ |
| 708 | if (osc_obd) |
| 709 | class_manual_cleanup(osc_obd); |
| 710 | } |
| 711 | |
| 712 | void lov_fix_desc_stripe_size(__u64 *val) |
| 713 | { |
| 714 | if (*val < LOV_MIN_STRIPE_SIZE) { |
| 715 | if (*val != 0) |
| 716 | LCONSOLE_INFO("Increasing default stripe size to minimum %u\n", |
| 717 | LOV_DESC_STRIPE_SIZE_DEFAULT); |
| 718 | *val = LOV_DESC_STRIPE_SIZE_DEFAULT; |
| 719 | } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) { |
| 720 | *val &= ~(LOV_MIN_STRIPE_SIZE - 1); |
| 721 | LCONSOLE_WARN("Changing default stripe size to %llu (a multiple of %u)\n", |
| 722 | *val, LOV_MIN_STRIPE_SIZE); |
| 723 | } |
| 724 | } |
| 725 | |
| 726 | void lov_fix_desc_stripe_count(__u32 *val) |
| 727 | { |
| 728 | if (*val == 0) |
| 729 | *val = 1; |
| 730 | } |
| 731 | |
| 732 | void lov_fix_desc_pattern(__u32 *val) |
| 733 | { |
| 734 | /* from lov_setstripe */ |
| 735 | if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) { |
| 736 | LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val); |
| 737 | *val = 0; |
| 738 | } |
| 739 | } |
| 740 | |
| 741 | void lov_fix_desc_qos_maxage(__u32 *val) |
| 742 | { |
| 743 | if (*val == 0) |
| 744 | *val = LOV_DESC_QOS_MAXAGE_DEFAULT; |
| 745 | } |
| 746 | |
| 747 | void lov_fix_desc(struct lov_desc *desc) |
| 748 | { |
| 749 | lov_fix_desc_stripe_size(&desc->ld_default_stripe_size); |
| 750 | lov_fix_desc_stripe_count(&desc->ld_default_stripe_count); |
| 751 | lov_fix_desc_pattern(&desc->ld_pattern); |
| 752 | lov_fix_desc_qos_maxage(&desc->ld_qos_maxage); |
| 753 | } |
| 754 | |
| 755 | int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg) |
| 756 | { |
| 757 | struct lprocfs_static_vars lvars = { NULL }; |
| 758 | struct lov_desc *desc; |
| 759 | struct lov_obd *lov = &obd->u.lov; |
| 760 | int rc; |
| 761 | |
| 762 | if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { |
| 763 | CERROR("LOV setup requires a descriptor\n"); |
| 764 | return -EINVAL; |
| 765 | } |
| 766 | |
| 767 | desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1); |
| 768 | |
| 769 | if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) { |
| 770 | CERROR("descriptor size wrong: %d > %d\n", |
| 771 | (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1)); |
| 772 | return -EINVAL; |
| 773 | } |
| 774 | |
| 775 | if (desc->ld_magic != LOV_DESC_MAGIC) { |
| 776 | if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) { |
| 777 | CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n", |
| 778 | obd->obd_name, desc); |
| 779 | lustre_swab_lov_desc(desc); |
| 780 | } else { |
| 781 | CERROR("%s: Bad lov desc magic: %#x\n", |
| 782 | obd->obd_name, desc->ld_magic); |
| 783 | return -EINVAL; |
| 784 | } |
| 785 | } |
| 786 | |
| 787 | lov_fix_desc(desc); |
| 788 | |
| 789 | desc->ld_active_tgt_count = 0; |
| 790 | lov->desc = *desc; |
| 791 | lov->lov_tgt_size = 0; |
| 792 | |
| 793 | mutex_init(&lov->lov_lock); |
| 794 | atomic_set(&lov->lov_refcount, 0); |
| 795 | lov->lov_sp_me = LUSTRE_SP_CLI; |
| 796 | |
| 797 | init_rwsem(&lov->lov_notify_lock); |
| 798 | |
| 799 | lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS, |
| 800 | HASH_POOLS_MAX_BITS, |
| 801 | HASH_POOLS_BKT_BITS, 0, |
| 802 | CFS_HASH_MIN_THETA, |
| 803 | CFS_HASH_MAX_THETA, |
| 804 | &pool_hash_operations, |
| 805 | CFS_HASH_DEFAULT); |
| 806 | INIT_LIST_HEAD(&lov->lov_pool_list); |
| 807 | lov->lov_pool_count = 0; |
| 808 | rc = lov_ost_pool_init(&lov->lov_packed, 0); |
| 809 | if (rc) |
| 810 | goto out; |
| 811 | |
| 812 | lprocfs_lov_init_vars(&lvars); |
| 813 | lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars); |
| 814 | |
| 815 | rc = ldebugfs_seq_create(obd->obd_debugfs_entry, "target_obd", |
| 816 | 0444, &lov_proc_target_fops, obd); |
| 817 | if (rc) |
| 818 | CWARN("Error adding the target_obd file\n"); |
| 819 | |
| 820 | lov->lov_pool_debugfs_entry = ldebugfs_register("pools", |
| 821 | obd->obd_debugfs_entry, |
| 822 | NULL, NULL); |
| 823 | return 0; |
| 824 | |
| 825 | out: |
| 826 | return rc; |
| 827 | } |
| 828 | |
| 829 | static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) |
| 830 | { |
| 831 | struct lov_obd *lov = &obd->u.lov; |
| 832 | |
| 833 | switch (stage) { |
| 834 | case OBD_CLEANUP_EARLY: { |
| 835 | int i; |
| 836 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 837 | if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) |
| 838 | continue; |
| 839 | obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp), |
| 840 | OBD_CLEANUP_EARLY); |
| 841 | } |
| 842 | break; |
| 843 | } |
| 844 | default: |
| 845 | break; |
| 846 | } |
| 847 | |
| 848 | return 0; |
| 849 | } |
| 850 | |
| 851 | static int lov_cleanup(struct obd_device *obd) |
| 852 | { |
| 853 | struct lov_obd *lov = &obd->u.lov; |
| 854 | struct list_head *pos, *tmp; |
| 855 | struct pool_desc *pool; |
| 856 | |
| 857 | list_for_each_safe(pos, tmp, &lov->lov_pool_list) { |
| 858 | pool = list_entry(pos, struct pool_desc, pool_list); |
| 859 | /* free pool structs */ |
| 860 | CDEBUG(D_INFO, "delete pool %p\n", pool); |
| 861 | /* In the function below, .hs_keycmp resolves to |
| 862 | * pool_hashkey_keycmp() */ |
| 863 | /* coverity[overrun-buffer-val] */ |
| 864 | lov_pool_del(obd, pool->pool_name); |
| 865 | } |
| 866 | cfs_hash_putref(lov->lov_pools_hash_body); |
| 867 | lov_ost_pool_free(&lov->lov_packed); |
| 868 | |
| 869 | lprocfs_obd_cleanup(obd); |
| 870 | if (lov->lov_tgts) { |
| 871 | int i; |
| 872 | obd_getref(obd); |
| 873 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 874 | if (!lov->lov_tgts[i]) |
| 875 | continue; |
| 876 | |
| 877 | /* Inactive targets may never have connected */ |
| 878 | if (lov->lov_tgts[i]->ltd_active || |
| 879 | atomic_read(&lov->lov_refcount)) |
| 880 | /* We should never get here - these |
| 881 | should have been removed in the |
| 882 | disconnect. */ |
| 883 | CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n", |
| 884 | i, lov->lov_death_row, |
| 885 | atomic_read(&lov->lov_refcount)); |
| 886 | lov_del_target(obd, i, NULL, 0); |
| 887 | } |
| 888 | obd_putref(obd); |
| 889 | kfree(lov->lov_tgts); |
| 890 | lov->lov_tgt_size = 0; |
| 891 | } |
| 892 | return 0; |
| 893 | } |
| 894 | |
| 895 | int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg, |
| 896 | __u32 *indexp, int *genp) |
| 897 | { |
| 898 | struct obd_uuid obd_uuid; |
| 899 | int cmd; |
| 900 | int rc = 0; |
| 901 | |
| 902 | switch (cmd = lcfg->lcfg_command) { |
| 903 | case LCFG_LOV_ADD_OBD: |
| 904 | case LCFG_LOV_ADD_INA: |
| 905 | case LCFG_LOV_DEL_OBD: { |
| 906 | __u32 index; |
| 907 | int gen; |
| 908 | /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */ |
| 909 | if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) { |
| 910 | rc = -EINVAL; |
| 911 | goto out; |
| 912 | } |
| 913 | |
| 914 | obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1)); |
| 915 | |
| 916 | if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1) { |
| 917 | rc = -EINVAL; |
| 918 | goto out; |
| 919 | } |
| 920 | if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1) { |
| 921 | rc = -EINVAL; |
| 922 | goto out; |
| 923 | } |
| 924 | index = *indexp; |
| 925 | gen = *genp; |
| 926 | if (cmd == LCFG_LOV_ADD_OBD) |
| 927 | rc = lov_add_target(obd, &obd_uuid, index, gen, 1); |
| 928 | else if (cmd == LCFG_LOV_ADD_INA) |
| 929 | rc = lov_add_target(obd, &obd_uuid, index, gen, 0); |
| 930 | else |
| 931 | rc = lov_del_target(obd, index, &obd_uuid, gen); |
| 932 | goto out; |
| 933 | } |
| 934 | case LCFG_PARAM: { |
| 935 | struct lprocfs_static_vars lvars = { NULL }; |
| 936 | struct lov_desc *desc = &(obd->u.lov.desc); |
| 937 | |
| 938 | if (!desc) { |
| 939 | rc = -EINVAL; |
| 940 | goto out; |
| 941 | } |
| 942 | |
| 943 | lprocfs_lov_init_vars(&lvars); |
| 944 | |
| 945 | rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars, |
| 946 | lcfg, obd); |
| 947 | if (rc > 0) |
| 948 | rc = 0; |
| 949 | goto out; |
| 950 | } |
| 951 | case LCFG_POOL_NEW: |
| 952 | case LCFG_POOL_ADD: |
| 953 | case LCFG_POOL_DEL: |
| 954 | case LCFG_POOL_REM: |
| 955 | goto out; |
| 956 | |
| 957 | default: { |
| 958 | CERROR("Unknown command: %d\n", lcfg->lcfg_command); |
| 959 | rc = -EINVAL; |
| 960 | goto out; |
| 961 | |
| 962 | } |
| 963 | } |
| 964 | out: |
| 965 | return rc; |
| 966 | } |
| 967 | |
| 968 | static int lov_recreate(struct obd_export *exp, struct obdo *src_oa, |
| 969 | struct lov_stripe_md **ea, struct obd_trans_info *oti) |
| 970 | { |
| 971 | struct lov_stripe_md *obj_mdp, *lsm; |
| 972 | struct lov_obd *lov = &exp->exp_obd->u.lov; |
| 973 | unsigned ost_idx; |
| 974 | int rc, i; |
| 975 | |
| 976 | LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS && |
| 977 | src_oa->o_flags & OBD_FL_RECREATE_OBJS); |
| 978 | |
| 979 | obj_mdp = kzalloc(sizeof(*obj_mdp), GFP_NOFS); |
| 980 | if (!obj_mdp) |
| 981 | return -ENOMEM; |
| 982 | |
| 983 | ost_idx = src_oa->o_nlink; |
| 984 | lsm = *ea; |
| 985 | if (lsm == NULL) { |
| 986 | rc = -EINVAL; |
| 987 | goto out; |
| 988 | } |
| 989 | if (ost_idx >= lov->desc.ld_tgt_count || |
| 990 | !lov->lov_tgts[ost_idx]) { |
| 991 | rc = -EINVAL; |
| 992 | goto out; |
| 993 | } |
| 994 | |
| 995 | for (i = 0; i < lsm->lsm_stripe_count; i++) { |
| 996 | struct lov_oinfo *loi = lsm->lsm_oinfo[i]; |
| 997 | |
| 998 | if (lov_oinfo_is_dummy(loi)) |
| 999 | continue; |
| 1000 | |
| 1001 | if (loi->loi_ost_idx == ost_idx) { |
| 1002 | if (ostid_id(&loi->loi_oi) != ostid_id(&src_oa->o_oi)) { |
| 1003 | rc = -EINVAL; |
| 1004 | goto out; |
| 1005 | } |
| 1006 | break; |
| 1007 | } |
| 1008 | } |
| 1009 | if (i == lsm->lsm_stripe_count) { |
| 1010 | rc = -EINVAL; |
| 1011 | goto out; |
| 1012 | } |
| 1013 | |
| 1014 | rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp, |
| 1015 | src_oa, &obj_mdp, oti); |
| 1016 | out: |
| 1017 | kfree(obj_mdp); |
| 1018 | return rc; |
| 1019 | } |
| 1020 | |
| 1021 | /* the LOV expects oa->o_id to be set to the LOV object id */ |
| 1022 | static int lov_create(const struct lu_env *env, struct obd_export *exp, |
| 1023 | struct obdo *src_oa, struct lov_stripe_md **ea, |
| 1024 | struct obd_trans_info *oti) |
| 1025 | { |
| 1026 | struct lov_obd *lov; |
| 1027 | int rc = 0; |
| 1028 | |
| 1029 | LASSERT(ea != NULL); |
| 1030 | if (exp == NULL) |
| 1031 | return -EINVAL; |
| 1032 | |
| 1033 | if ((src_oa->o_valid & OBD_MD_FLFLAGS) && |
| 1034 | src_oa->o_flags == OBD_FL_DELORPHAN) { |
| 1035 | /* should be used with LOV anymore */ |
| 1036 | LBUG(); |
| 1037 | } |
| 1038 | |
| 1039 | lov = &exp->exp_obd->u.lov; |
| 1040 | if (!lov->desc.ld_active_tgt_count) |
| 1041 | return -EIO; |
| 1042 | |
| 1043 | obd_getref(exp->exp_obd); |
| 1044 | /* Recreate a specific object id at the given OST index */ |
| 1045 | if ((src_oa->o_valid & OBD_MD_FLFLAGS) && |
| 1046 | (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) { |
| 1047 | rc = lov_recreate(exp, src_oa, ea, oti); |
| 1048 | } |
| 1049 | |
| 1050 | obd_putref(exp->exp_obd); |
| 1051 | return rc; |
| 1052 | } |
| 1053 | |
| 1054 | #define ASSERT_LSM_MAGIC(lsmp) \ |
| 1055 | do { \ |
| 1056 | LASSERT((lsmp) != NULL); \ |
| 1057 | LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \ |
| 1058 | (lsmp)->lsm_magic == LOV_MAGIC_V3), \ |
| 1059 | "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \ |
| 1060 | } while (0) |
| 1061 | |
| 1062 | static int lov_destroy(const struct lu_env *env, struct obd_export *exp, |
| 1063 | struct obdo *oa, struct lov_stripe_md *lsm, |
| 1064 | struct obd_trans_info *oti, struct obd_export *md_exp, |
| 1065 | void *capa) |
| 1066 | { |
| 1067 | struct lov_request_set *set; |
| 1068 | struct obd_info oinfo; |
| 1069 | struct lov_request *req; |
| 1070 | struct list_head *pos; |
| 1071 | struct lov_obd *lov; |
| 1072 | int rc = 0, err = 0; |
| 1073 | |
| 1074 | ASSERT_LSM_MAGIC(lsm); |
| 1075 | |
| 1076 | if (!exp || !exp->exp_obd) |
| 1077 | return -ENODEV; |
| 1078 | |
| 1079 | if (oa->o_valid & OBD_MD_FLCOOKIE) { |
| 1080 | LASSERT(oti); |
| 1081 | LASSERT(oti->oti_logcookies); |
| 1082 | } |
| 1083 | |
| 1084 | lov = &exp->exp_obd->u.lov; |
| 1085 | obd_getref(exp->exp_obd); |
| 1086 | rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set); |
| 1087 | if (rc) |
| 1088 | goto out; |
| 1089 | |
| 1090 | list_for_each(pos, &set->set_list) { |
| 1091 | req = list_entry(pos, struct lov_request, rq_link); |
| 1092 | |
| 1093 | if (oa->o_valid & OBD_MD_FLCOOKIE) |
| 1094 | oti->oti_logcookies = set->set_cookies + req->rq_stripe; |
| 1095 | |
| 1096 | err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp, |
| 1097 | req->rq_oi.oi_oa, NULL, oti, NULL, capa); |
| 1098 | err = lov_update_common_set(set, req, err); |
| 1099 | if (err) { |
| 1100 | CERROR("%s: destroying objid "DOSTID" subobj " |
| 1101 | DOSTID" on OST idx %d: rc = %d\n", |
| 1102 | exp->exp_obd->obd_name, POSTID(&oa->o_oi), |
| 1103 | POSTID(&req->rq_oi.oi_oa->o_oi), |
| 1104 | req->rq_idx, err); |
| 1105 | if (!rc) |
| 1106 | rc = err; |
| 1107 | } |
| 1108 | } |
| 1109 | |
| 1110 | if (rc == 0) { |
| 1111 | LASSERT(lsm_op_find(lsm->lsm_magic) != NULL); |
| 1112 | rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp); |
| 1113 | } |
| 1114 | err = lov_fini_destroy_set(set); |
| 1115 | out: |
| 1116 | obd_putref(exp->exp_obd); |
| 1117 | return rc ? rc : err; |
| 1118 | } |
| 1119 | |
| 1120 | static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, |
| 1121 | void *data, int rc) |
| 1122 | { |
| 1123 | struct lov_request_set *lovset = (struct lov_request_set *)data; |
| 1124 | int err; |
| 1125 | |
| 1126 | /* don't do attribute merge if this async op failed */ |
| 1127 | if (rc) |
| 1128 | atomic_set(&lovset->set_completes, 0); |
| 1129 | err = lov_fini_getattr_set(lovset); |
| 1130 | return rc ? rc : err; |
| 1131 | } |
| 1132 | |
| 1133 | static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo, |
| 1134 | struct ptlrpc_request_set *rqset) |
| 1135 | { |
| 1136 | struct lov_request_set *lovset; |
| 1137 | struct lov_obd *lov; |
| 1138 | struct list_head *pos; |
| 1139 | struct lov_request *req; |
| 1140 | int rc = 0, err; |
| 1141 | |
| 1142 | LASSERT(oinfo); |
| 1143 | ASSERT_LSM_MAGIC(oinfo->oi_md); |
| 1144 | |
| 1145 | if (!exp || !exp->exp_obd) |
| 1146 | return -ENODEV; |
| 1147 | |
| 1148 | lov = &exp->exp_obd->u.lov; |
| 1149 | |
| 1150 | rc = lov_prep_getattr_set(exp, oinfo, &lovset); |
| 1151 | if (rc) |
| 1152 | return rc; |
| 1153 | |
| 1154 | CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n", |
| 1155 | POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count, |
| 1156 | oinfo->oi_md->lsm_stripe_size); |
| 1157 | |
| 1158 | list_for_each(pos, &lovset->set_list) { |
| 1159 | req = list_entry(pos, struct lov_request, rq_link); |
| 1160 | |
| 1161 | CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n", |
| 1162 | POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe, |
| 1163 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx); |
| 1164 | rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp, |
| 1165 | &req->rq_oi, rqset); |
| 1166 | if (rc) { |
| 1167 | CERROR("%s: getattr objid "DOSTID" subobj" |
| 1168 | DOSTID" on OST idx %d: rc = %d\n", |
| 1169 | exp->exp_obd->obd_name, |
| 1170 | POSTID(&oinfo->oi_oa->o_oi), |
| 1171 | POSTID(&req->rq_oi.oi_oa->o_oi), |
| 1172 | req->rq_idx, rc); |
| 1173 | goto out; |
| 1174 | } |
| 1175 | } |
| 1176 | |
| 1177 | if (!list_empty(&rqset->set_requests)) { |
| 1178 | LASSERT(rc == 0); |
| 1179 | LASSERT(rqset->set_interpret == NULL); |
| 1180 | rqset->set_interpret = lov_getattr_interpret; |
| 1181 | rqset->set_arg = (void *)lovset; |
| 1182 | return rc; |
| 1183 | } |
| 1184 | out: |
| 1185 | if (rc) |
| 1186 | atomic_set(&lovset->set_completes, 0); |
| 1187 | err = lov_fini_getattr_set(lovset); |
| 1188 | return rc ? rc : err; |
| 1189 | } |
| 1190 | |
| 1191 | static int lov_setattr_interpret(struct ptlrpc_request_set *rqset, |
| 1192 | void *data, int rc) |
| 1193 | { |
| 1194 | struct lov_request_set *lovset = (struct lov_request_set *)data; |
| 1195 | int err; |
| 1196 | |
| 1197 | if (rc) |
| 1198 | atomic_set(&lovset->set_completes, 0); |
| 1199 | err = lov_fini_setattr_set(lovset); |
| 1200 | return rc ? rc : err; |
| 1201 | } |
| 1202 | |
| 1203 | /* If @oti is given, the request goes from MDS and responses from OSTs are not |
| 1204 | needed. Otherwise, a client is waiting for responses. */ |
| 1205 | static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo, |
| 1206 | struct obd_trans_info *oti, |
| 1207 | struct ptlrpc_request_set *rqset) |
| 1208 | { |
| 1209 | struct lov_request_set *set; |
| 1210 | struct lov_request *req; |
| 1211 | struct list_head *pos; |
| 1212 | struct lov_obd *lov; |
| 1213 | int rc = 0; |
| 1214 | |
| 1215 | LASSERT(oinfo); |
| 1216 | ASSERT_LSM_MAGIC(oinfo->oi_md); |
| 1217 | if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) { |
| 1218 | LASSERT(oti); |
| 1219 | LASSERT(oti->oti_logcookies); |
| 1220 | } |
| 1221 | |
| 1222 | if (!exp || !exp->exp_obd) |
| 1223 | return -ENODEV; |
| 1224 | |
| 1225 | lov = &exp->exp_obd->u.lov; |
| 1226 | rc = lov_prep_setattr_set(exp, oinfo, oti, &set); |
| 1227 | if (rc) |
| 1228 | return rc; |
| 1229 | |
| 1230 | CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n", |
| 1231 | POSTID(&oinfo->oi_md->lsm_oi), |
| 1232 | oinfo->oi_md->lsm_stripe_count, |
| 1233 | oinfo->oi_md->lsm_stripe_size); |
| 1234 | |
| 1235 | list_for_each(pos, &set->set_list) { |
| 1236 | req = list_entry(pos, struct lov_request, rq_link); |
| 1237 | |
| 1238 | if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) |
| 1239 | oti->oti_logcookies = set->set_cookies + req->rq_stripe; |
| 1240 | |
| 1241 | CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n", |
| 1242 | POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe, |
| 1243 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx); |
| 1244 | |
| 1245 | rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp, |
| 1246 | &req->rq_oi, oti, rqset); |
| 1247 | if (rc) { |
| 1248 | CERROR("error: setattr objid "DOSTID" subobj" |
| 1249 | DOSTID" on OST idx %d: rc = %d\n", |
| 1250 | POSTID(&set->set_oi->oi_oa->o_oi), |
| 1251 | POSTID(&req->rq_oi.oi_oa->o_oi), |
| 1252 | req->rq_idx, rc); |
| 1253 | break; |
| 1254 | } |
| 1255 | } |
| 1256 | |
| 1257 | /* If we are not waiting for responses on async requests, return. */ |
| 1258 | if (rc || !rqset || list_empty(&rqset->set_requests)) { |
| 1259 | int err; |
| 1260 | if (rc) |
| 1261 | atomic_set(&set->set_completes, 0); |
| 1262 | err = lov_fini_setattr_set(set); |
| 1263 | return rc ? rc : err; |
| 1264 | } |
| 1265 | |
| 1266 | LASSERT(rqset->set_interpret == NULL); |
| 1267 | rqset->set_interpret = lov_setattr_interpret; |
| 1268 | rqset->set_arg = (void *)set; |
| 1269 | |
| 1270 | return 0; |
| 1271 | } |
| 1272 | |
| 1273 | /* find any ldlm lock of the inode in lov |
| 1274 | * return 0 not find |
| 1275 | * 1 find one |
| 1276 | * < 0 error */ |
| 1277 | static int lov_find_cbdata(struct obd_export *exp, |
| 1278 | struct lov_stripe_md *lsm, ldlm_iterator_t it, |
| 1279 | void *data) |
| 1280 | { |
| 1281 | struct lov_obd *lov; |
| 1282 | int rc = 0, i; |
| 1283 | |
| 1284 | ASSERT_LSM_MAGIC(lsm); |
| 1285 | |
| 1286 | if (!exp || !exp->exp_obd) |
| 1287 | return -ENODEV; |
| 1288 | |
| 1289 | lov = &exp->exp_obd->u.lov; |
| 1290 | for (i = 0; i < lsm->lsm_stripe_count; i++) { |
| 1291 | struct lov_stripe_md submd; |
| 1292 | struct lov_oinfo *loi = lsm->lsm_oinfo[i]; |
| 1293 | |
| 1294 | if (lov_oinfo_is_dummy(loi)) |
| 1295 | continue; |
| 1296 | |
| 1297 | if (!lov->lov_tgts[loi->loi_ost_idx]) { |
| 1298 | CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx); |
| 1299 | continue; |
| 1300 | } |
| 1301 | |
| 1302 | submd.lsm_oi = loi->loi_oi; |
| 1303 | submd.lsm_stripe_count = 0; |
| 1304 | rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, |
| 1305 | &submd, it, data); |
| 1306 | if (rc != 0) |
| 1307 | return rc; |
| 1308 | } |
| 1309 | return rc; |
| 1310 | } |
| 1311 | |
| 1312 | int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc) |
| 1313 | { |
| 1314 | struct lov_request_set *lovset = (struct lov_request_set *)data; |
| 1315 | int err; |
| 1316 | |
| 1317 | if (rc) |
| 1318 | atomic_set(&lovset->set_completes, 0); |
| 1319 | |
| 1320 | err = lov_fini_statfs_set(lovset); |
| 1321 | return rc ? rc : err; |
| 1322 | } |
| 1323 | |
| 1324 | static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo, |
| 1325 | __u64 max_age, struct ptlrpc_request_set *rqset) |
| 1326 | { |
| 1327 | struct obd_device *obd = class_exp2obd(exp); |
| 1328 | struct lov_request_set *set; |
| 1329 | struct lov_request *req; |
| 1330 | struct list_head *pos; |
| 1331 | struct lov_obd *lov; |
| 1332 | int rc = 0; |
| 1333 | |
| 1334 | LASSERT(oinfo != NULL); |
| 1335 | LASSERT(oinfo->oi_osfs != NULL); |
| 1336 | |
| 1337 | lov = &obd->u.lov; |
| 1338 | rc = lov_prep_statfs_set(obd, oinfo, &set); |
| 1339 | if (rc) |
| 1340 | return rc; |
| 1341 | |
| 1342 | list_for_each(pos, &set->set_list) { |
| 1343 | req = list_entry(pos, struct lov_request, rq_link); |
| 1344 | rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp, |
| 1345 | &req->rq_oi, max_age, rqset); |
| 1346 | if (rc) |
| 1347 | break; |
| 1348 | } |
| 1349 | |
| 1350 | if (rc || list_empty(&rqset->set_requests)) { |
| 1351 | int err; |
| 1352 | if (rc) |
| 1353 | atomic_set(&set->set_completes, 0); |
| 1354 | err = lov_fini_statfs_set(set); |
| 1355 | return rc ? rc : err; |
| 1356 | } |
| 1357 | |
| 1358 | LASSERT(rqset->set_interpret == NULL); |
| 1359 | rqset->set_interpret = lov_statfs_interpret; |
| 1360 | rqset->set_arg = (void *)set; |
| 1361 | return 0; |
| 1362 | } |
| 1363 | |
| 1364 | static int lov_statfs(const struct lu_env *env, struct obd_export *exp, |
| 1365 | struct obd_statfs *osfs, __u64 max_age, __u32 flags) |
| 1366 | { |
| 1367 | struct ptlrpc_request_set *set = NULL; |
| 1368 | struct obd_info oinfo = { { { 0 } } }; |
| 1369 | int rc = 0; |
| 1370 | |
| 1371 | /* for obdclass we forbid using obd_statfs_rqset, but prefer using async |
| 1372 | * statfs requests */ |
| 1373 | set = ptlrpc_prep_set(); |
| 1374 | if (set == NULL) |
| 1375 | return -ENOMEM; |
| 1376 | |
| 1377 | oinfo.oi_osfs = osfs; |
| 1378 | oinfo.oi_flags = flags; |
| 1379 | rc = lov_statfs_async(exp, &oinfo, max_age, set); |
| 1380 | if (rc == 0) |
| 1381 | rc = ptlrpc_set_wait(set); |
| 1382 | ptlrpc_set_destroy(set); |
| 1383 | |
| 1384 | return rc; |
| 1385 | } |
| 1386 | |
| 1387 | static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, |
| 1388 | void *karg, void *uarg) |
| 1389 | { |
| 1390 | struct obd_device *obddev = class_exp2obd(exp); |
| 1391 | struct lov_obd *lov = &obddev->u.lov; |
| 1392 | int i = 0, rc = 0, count = lov->desc.ld_tgt_count; |
| 1393 | struct obd_uuid *uuidp; |
| 1394 | |
| 1395 | switch (cmd) { |
| 1396 | case IOC_OBD_STATFS: { |
| 1397 | struct obd_ioctl_data *data = karg; |
| 1398 | struct obd_device *osc_obd; |
| 1399 | struct obd_statfs stat_buf = {0}; |
| 1400 | __u32 index; |
| 1401 | __u32 flags; |
| 1402 | |
| 1403 | memcpy(&index, data->ioc_inlbuf2, sizeof(__u32)); |
| 1404 | if (index >= count) |
| 1405 | return -ENODEV; |
| 1406 | |
| 1407 | if (!lov->lov_tgts[index]) |
| 1408 | /* Try again with the next index */ |
| 1409 | return -EAGAIN; |
| 1410 | if (!lov->lov_tgts[index]->ltd_active) |
| 1411 | return -ENODATA; |
| 1412 | |
| 1413 | osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp); |
| 1414 | if (!osc_obd) |
| 1415 | return -EINVAL; |
| 1416 | |
| 1417 | /* copy UUID */ |
| 1418 | if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd), |
| 1419 | min((int) data->ioc_plen2, |
| 1420 | (int) sizeof(struct obd_uuid)))) |
| 1421 | return -EFAULT; |
| 1422 | |
| 1423 | flags = uarg ? *(__u32 *)uarg : 0; |
| 1424 | /* got statfs data */ |
| 1425 | rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf, |
| 1426 | cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), |
| 1427 | flags); |
| 1428 | if (rc) |
| 1429 | return rc; |
| 1430 | if (copy_to_user(data->ioc_pbuf1, &stat_buf, |
| 1431 | min((int) data->ioc_plen1, |
| 1432 | (int) sizeof(stat_buf)))) |
| 1433 | return -EFAULT; |
| 1434 | break; |
| 1435 | } |
| 1436 | case OBD_IOC_LOV_GET_CONFIG: { |
| 1437 | struct obd_ioctl_data *data; |
| 1438 | struct lov_desc *desc; |
| 1439 | char *buf = NULL; |
| 1440 | __u32 *genp; |
| 1441 | |
| 1442 | len = 0; |
| 1443 | if (obd_ioctl_getdata(&buf, &len, uarg)) |
| 1444 | return -EINVAL; |
| 1445 | |
| 1446 | data = (struct obd_ioctl_data *)buf; |
| 1447 | |
| 1448 | if (sizeof(*desc) > data->ioc_inllen1) { |
| 1449 | obd_ioctl_freedata(buf, len); |
| 1450 | return -EINVAL; |
| 1451 | } |
| 1452 | |
| 1453 | if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) { |
| 1454 | obd_ioctl_freedata(buf, len); |
| 1455 | return -EINVAL; |
| 1456 | } |
| 1457 | |
| 1458 | if (sizeof(__u32) * count > data->ioc_inllen3) { |
| 1459 | obd_ioctl_freedata(buf, len); |
| 1460 | return -EINVAL; |
| 1461 | } |
| 1462 | |
| 1463 | desc = (struct lov_desc *)data->ioc_inlbuf1; |
| 1464 | memcpy(desc, &(lov->desc), sizeof(*desc)); |
| 1465 | |
| 1466 | uuidp = (struct obd_uuid *)data->ioc_inlbuf2; |
| 1467 | genp = (__u32 *)data->ioc_inlbuf3; |
| 1468 | /* the uuid will be empty for deleted OSTs */ |
| 1469 | for (i = 0; i < count; i++, uuidp++, genp++) { |
| 1470 | if (!lov->lov_tgts[i]) |
| 1471 | continue; |
| 1472 | *uuidp = lov->lov_tgts[i]->ltd_uuid; |
| 1473 | *genp = lov->lov_tgts[i]->ltd_gen; |
| 1474 | } |
| 1475 | |
| 1476 | if (copy_to_user(uarg, buf, len)) |
| 1477 | rc = -EFAULT; |
| 1478 | obd_ioctl_freedata(buf, len); |
| 1479 | break; |
| 1480 | } |
| 1481 | case LL_IOC_LOV_GETSTRIPE: |
| 1482 | rc = lov_getstripe(exp, karg, uarg); |
| 1483 | break; |
| 1484 | case OBD_IOC_QUOTACTL: { |
| 1485 | struct if_quotactl *qctl = karg; |
| 1486 | struct lov_tgt_desc *tgt = NULL; |
| 1487 | struct obd_quotactl *oqctl; |
| 1488 | |
| 1489 | if (qctl->qc_valid == QC_OSTIDX) { |
| 1490 | if (qctl->qc_idx < 0 || count <= qctl->qc_idx) |
| 1491 | return -EINVAL; |
| 1492 | |
| 1493 | tgt = lov->lov_tgts[qctl->qc_idx]; |
| 1494 | if (!tgt || !tgt->ltd_exp) |
| 1495 | return -EINVAL; |
| 1496 | } else if (qctl->qc_valid == QC_UUID) { |
| 1497 | for (i = 0; i < count; i++) { |
| 1498 | tgt = lov->lov_tgts[i]; |
| 1499 | if (!tgt || |
| 1500 | !obd_uuid_equals(&tgt->ltd_uuid, |
| 1501 | &qctl->obd_uuid)) |
| 1502 | continue; |
| 1503 | |
| 1504 | if (tgt->ltd_exp == NULL) |
| 1505 | return -EINVAL; |
| 1506 | |
| 1507 | break; |
| 1508 | } |
| 1509 | } else { |
| 1510 | return -EINVAL; |
| 1511 | } |
| 1512 | |
| 1513 | if (i >= count) |
| 1514 | return -EAGAIN; |
| 1515 | |
| 1516 | LASSERT(tgt && tgt->ltd_exp); |
| 1517 | oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS); |
| 1518 | if (!oqctl) |
| 1519 | return -ENOMEM; |
| 1520 | |
| 1521 | QCTL_COPY(oqctl, qctl); |
| 1522 | rc = obd_quotactl(tgt->ltd_exp, oqctl); |
| 1523 | if (rc == 0) { |
| 1524 | QCTL_COPY(qctl, oqctl); |
| 1525 | qctl->qc_valid = QC_OSTIDX; |
| 1526 | qctl->obd_uuid = tgt->ltd_uuid; |
| 1527 | } |
| 1528 | kfree(oqctl); |
| 1529 | break; |
| 1530 | } |
| 1531 | default: { |
| 1532 | int set = 0; |
| 1533 | |
| 1534 | if (count == 0) |
| 1535 | return -ENOTTY; |
| 1536 | |
| 1537 | for (i = 0; i < count; i++) { |
| 1538 | int err; |
| 1539 | struct obd_device *osc_obd; |
| 1540 | |
| 1541 | /* OST was disconnected */ |
| 1542 | if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp) |
| 1543 | continue; |
| 1544 | |
| 1545 | /* ll_umount_begin() sets force flag but for lov, not |
| 1546 | * osc. Let's pass it through */ |
| 1547 | osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp); |
| 1548 | osc_obd->obd_force = obddev->obd_force; |
| 1549 | err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp, |
| 1550 | len, karg, uarg); |
| 1551 | if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) { |
| 1552 | return err; |
| 1553 | } else if (err) { |
| 1554 | if (lov->lov_tgts[i]->ltd_active) { |
| 1555 | CDEBUG(err == -ENOTTY ? |
| 1556 | D_IOCTL : D_WARNING, |
| 1557 | "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n", |
| 1558 | lov_uuid2str(lov, i), |
| 1559 | i, cmd, err); |
| 1560 | if (!rc) |
| 1561 | rc = err; |
| 1562 | } |
| 1563 | } else { |
| 1564 | set = 1; |
| 1565 | } |
| 1566 | } |
| 1567 | if (!set && !rc) |
| 1568 | rc = -EIO; |
| 1569 | } |
| 1570 | } |
| 1571 | |
| 1572 | return rc; |
| 1573 | } |
| 1574 | |
| 1575 | #define FIEMAP_BUFFER_SIZE 4096 |
| 1576 | |
| 1577 | /** |
| 1578 | * Non-zero fe_logical indicates that this is a continuation FIEMAP |
| 1579 | * call. The local end offset and the device are sent in the first |
| 1580 | * fm_extent. This function calculates the stripe number from the index. |
| 1581 | * This function returns a stripe_no on which mapping is to be restarted. |
| 1582 | * |
| 1583 | * This function returns fm_end_offset which is the in-OST offset at which |
| 1584 | * mapping should be restarted. If fm_end_offset=0 is returned then caller |
| 1585 | * will re-calculate proper offset in next stripe. |
| 1586 | * Note that the first extent is passed to lov_get_info via the value field. |
| 1587 | * |
| 1588 | * \param fiemap fiemap request header |
| 1589 | * \param lsm striping information for the file |
| 1590 | * \param fm_start logical start of mapping |
| 1591 | * \param fm_end logical end of mapping |
| 1592 | * \param start_stripe starting stripe will be returned in this |
| 1593 | */ |
| 1594 | static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap, |
| 1595 | struct lov_stripe_md *lsm, u64 fm_start, |
| 1596 | u64 fm_end, int *start_stripe) |
| 1597 | { |
| 1598 | u64 local_end = fiemap->fm_extents[0].fe_logical; |
| 1599 | u64 lun_start, lun_end; |
| 1600 | u64 fm_end_offset; |
| 1601 | int stripe_no = -1, i; |
| 1602 | |
| 1603 | if (fiemap->fm_extent_count == 0 || |
| 1604 | fiemap->fm_extents[0].fe_logical == 0) |
| 1605 | return 0; |
| 1606 | |
| 1607 | /* Find out stripe_no from ost_index saved in the fe_device */ |
| 1608 | for (i = 0; i < lsm->lsm_stripe_count; i++) { |
| 1609 | struct lov_oinfo *oinfo = lsm->lsm_oinfo[i]; |
| 1610 | |
| 1611 | if (lov_oinfo_is_dummy(oinfo)) |
| 1612 | continue; |
| 1613 | |
| 1614 | if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) { |
| 1615 | stripe_no = i; |
| 1616 | break; |
| 1617 | } |
| 1618 | } |
| 1619 | if (stripe_no == -1) |
| 1620 | return -EINVAL; |
| 1621 | |
| 1622 | /* If we have finished mapping on previous device, shift logical |
| 1623 | * offset to start of next device */ |
| 1624 | if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end, |
| 1625 | &lun_start, &lun_end)) != 0 && |
| 1626 | local_end < lun_end) { |
| 1627 | fm_end_offset = local_end; |
| 1628 | *start_stripe = stripe_no; |
| 1629 | } else { |
| 1630 | /* This is a special value to indicate that caller should |
| 1631 | * calculate offset in next stripe. */ |
| 1632 | fm_end_offset = 0; |
| 1633 | *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count; |
| 1634 | } |
| 1635 | |
| 1636 | return fm_end_offset; |
| 1637 | } |
| 1638 | |
| 1639 | /** |
| 1640 | * We calculate on which OST the mapping will end. If the length of mapping |
| 1641 | * is greater than (stripe_size * stripe_count) then the last_stripe will |
| 1642 | * will be one just before start_stripe. Else we check if the mapping |
| 1643 | * intersects each OST and find last_stripe. |
| 1644 | * This function returns the last_stripe and also sets the stripe_count |
| 1645 | * over which the mapping is spread |
| 1646 | * |
| 1647 | * \param lsm striping information for the file |
| 1648 | * \param fm_start logical start of mapping |
| 1649 | * \param fm_end logical end of mapping |
| 1650 | * \param start_stripe starting stripe of the mapping |
| 1651 | * \param stripe_count the number of stripes across which to map is returned |
| 1652 | * |
| 1653 | * \retval last_stripe return the last stripe of the mapping |
| 1654 | */ |
| 1655 | static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start, |
| 1656 | u64 fm_end, int start_stripe, |
| 1657 | int *stripe_count) |
| 1658 | { |
| 1659 | int last_stripe; |
| 1660 | u64 obd_start, obd_end; |
| 1661 | int i, j; |
| 1662 | |
| 1663 | if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) { |
| 1664 | last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 : |
| 1665 | start_stripe - 1; |
| 1666 | *stripe_count = lsm->lsm_stripe_count; |
| 1667 | } else { |
| 1668 | for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count; |
| 1669 | i = (i + 1) % lsm->lsm_stripe_count, j++) { |
| 1670 | if ((lov_stripe_intersects(lsm, i, fm_start, fm_end, |
| 1671 | &obd_start, &obd_end)) == 0) |
| 1672 | break; |
| 1673 | } |
| 1674 | *stripe_count = j; |
| 1675 | last_stripe = (start_stripe + j - 1) % lsm->lsm_stripe_count; |
| 1676 | } |
| 1677 | |
| 1678 | return last_stripe; |
| 1679 | } |
| 1680 | |
| 1681 | /** |
| 1682 | * Set fe_device and copy extents from local buffer into main return buffer. |
| 1683 | * |
| 1684 | * \param fiemap fiemap request header |
| 1685 | * \param lcl_fm_ext array of local fiemap extents to be copied |
| 1686 | * \param ost_index OST index to be written into the fm_device field for each |
| 1687 | extent |
| 1688 | * \param ext_count number of extents to be copied |
| 1689 | * \param current_extent where to start copying in main extent array |
| 1690 | */ |
| 1691 | static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap, |
| 1692 | struct ll_fiemap_extent *lcl_fm_ext, |
| 1693 | int ost_index, unsigned int ext_count, |
| 1694 | int current_extent) |
| 1695 | { |
| 1696 | char *to; |
| 1697 | int ext; |
| 1698 | |
| 1699 | for (ext = 0; ext < ext_count; ext++) { |
| 1700 | lcl_fm_ext[ext].fe_device = ost_index; |
| 1701 | lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET; |
| 1702 | } |
| 1703 | |
| 1704 | /* Copy fm_extent's from fm_local to return buffer */ |
| 1705 | to = (char *)fiemap + fiemap_count_to_size(current_extent); |
| 1706 | memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent)); |
| 1707 | } |
| 1708 | |
| 1709 | /** |
| 1710 | * Break down the FIEMAP request and send appropriate calls to individual OSTs. |
| 1711 | * This also handles the restarting of FIEMAP calls in case mapping overflows |
| 1712 | * the available number of extents in single call. |
| 1713 | */ |
| 1714 | static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key, |
| 1715 | __u32 *vallen, void *val, struct lov_stripe_md *lsm) |
| 1716 | { |
| 1717 | struct ll_fiemap_info_key *fm_key = key; |
| 1718 | struct ll_user_fiemap *fiemap = val; |
| 1719 | struct ll_user_fiemap *fm_local = NULL; |
| 1720 | struct ll_fiemap_extent *lcl_fm_ext; |
| 1721 | int count_local; |
| 1722 | unsigned int get_num_extents = 0; |
| 1723 | int ost_index = 0, actual_start_stripe, start_stripe; |
| 1724 | u64 fm_start, fm_end, fm_length, fm_end_offset; |
| 1725 | u64 curr_loc; |
| 1726 | int current_extent = 0, rc = 0, i; |
| 1727 | int ost_eof = 0; /* EOF for object */ |
| 1728 | int ost_done = 0; /* done with required mapping for this OST? */ |
| 1729 | int last_stripe; |
| 1730 | int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count; |
| 1731 | unsigned int buffer_size = FIEMAP_BUFFER_SIZE; |
| 1732 | |
| 1733 | if (!lsm_has_objects(lsm)) { |
| 1734 | rc = 0; |
| 1735 | goto out; |
| 1736 | } |
| 1737 | |
| 1738 | if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size) |
| 1739 | buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count); |
| 1740 | |
| 1741 | fm_local = libcfs_kvzalloc(buffer_size, GFP_NOFS); |
| 1742 | if (fm_local == NULL) { |
| 1743 | rc = -ENOMEM; |
| 1744 | goto out; |
| 1745 | } |
| 1746 | lcl_fm_ext = &fm_local->fm_extents[0]; |
| 1747 | |
| 1748 | count_local = fiemap_size_to_count(buffer_size); |
| 1749 | |
| 1750 | memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap)); |
| 1751 | fm_start = fiemap->fm_start; |
| 1752 | fm_length = fiemap->fm_length; |
| 1753 | /* Calculate start stripe, last stripe and length of mapping */ |
| 1754 | actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start); |
| 1755 | fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size : |
| 1756 | fm_start + fm_length - 1); |
| 1757 | /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */ |
| 1758 | if (fm_end > fm_key->oa.o_size) |
| 1759 | fm_end = fm_key->oa.o_size; |
| 1760 | |
| 1761 | last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end, |
| 1762 | actual_start_stripe, &stripe_count); |
| 1763 | |
| 1764 | fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, |
| 1765 | fm_end, &start_stripe); |
| 1766 | if (fm_end_offset == -EINVAL) { |
| 1767 | rc = -EINVAL; |
| 1768 | goto out; |
| 1769 | } |
| 1770 | |
| 1771 | if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen) |
| 1772 | fiemap->fm_extent_count = fiemap_size_to_count(*vallen); |
| 1773 | if (fiemap->fm_extent_count == 0) { |
| 1774 | get_num_extents = 1; |
| 1775 | count_local = 0; |
| 1776 | } |
| 1777 | /* Check each stripe */ |
| 1778 | for (cur_stripe = start_stripe, i = 0; i < stripe_count; |
| 1779 | i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) { |
| 1780 | u64 req_fm_len; /* Stores length of required mapping */ |
| 1781 | u64 len_mapped_single_call; |
| 1782 | u64 lun_start, lun_end, obd_object_end; |
| 1783 | unsigned int ext_count; |
| 1784 | |
| 1785 | cur_stripe_wrap = cur_stripe; |
| 1786 | |
| 1787 | /* Find out range of mapping on this stripe */ |
| 1788 | if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end, |
| 1789 | &lun_start, &obd_object_end)) == 0) |
| 1790 | continue; |
| 1791 | |
| 1792 | if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) { |
| 1793 | rc = -EIO; |
| 1794 | goto out; |
| 1795 | } |
| 1796 | |
| 1797 | /* If this is a continuation FIEMAP call and we are on |
| 1798 | * starting stripe then lun_start needs to be set to |
| 1799 | * fm_end_offset */ |
| 1800 | if (fm_end_offset != 0 && cur_stripe == start_stripe) |
| 1801 | lun_start = fm_end_offset; |
| 1802 | |
| 1803 | if (fm_length != ~0ULL) { |
| 1804 | /* Handle fm_start + fm_length overflow */ |
| 1805 | if (fm_start + fm_length < fm_start) |
| 1806 | fm_length = ~0ULL - fm_start; |
| 1807 | lun_end = lov_size_to_stripe(lsm, fm_start + fm_length, |
| 1808 | cur_stripe); |
| 1809 | } else { |
| 1810 | lun_end = ~0ULL; |
| 1811 | } |
| 1812 | |
| 1813 | if (lun_start == lun_end) |
| 1814 | continue; |
| 1815 | |
| 1816 | req_fm_len = obd_object_end - lun_start; |
| 1817 | fm_local->fm_length = 0; |
| 1818 | len_mapped_single_call = 0; |
| 1819 | |
| 1820 | /* If the output buffer is very large and the objects have many |
| 1821 | * extents we may need to loop on a single OST repeatedly */ |
| 1822 | ost_eof = 0; |
| 1823 | ost_done = 0; |
| 1824 | do { |
| 1825 | if (get_num_extents == 0) { |
| 1826 | /* Don't get too many extents. */ |
| 1827 | if (current_extent + count_local > |
| 1828 | fiemap->fm_extent_count) |
| 1829 | count_local = fiemap->fm_extent_count - |
| 1830 | current_extent; |
| 1831 | } |
| 1832 | |
| 1833 | lun_start += len_mapped_single_call; |
| 1834 | fm_local->fm_length = req_fm_len - len_mapped_single_call; |
| 1835 | req_fm_len = fm_local->fm_length; |
| 1836 | fm_local->fm_extent_count = count_local; |
| 1837 | fm_local->fm_mapped_extents = 0; |
| 1838 | fm_local->fm_flags = fiemap->fm_flags; |
| 1839 | |
| 1840 | fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi; |
| 1841 | ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx; |
| 1842 | |
| 1843 | if (ost_index < 0 || |
| 1844 | ost_index >= lov->desc.ld_tgt_count) { |
| 1845 | rc = -EINVAL; |
| 1846 | goto out; |
| 1847 | } |
| 1848 | |
| 1849 | /* If OST is inactive, return extent with UNKNOWN flag */ |
| 1850 | if (!lov->lov_tgts[ost_index]->ltd_active) { |
| 1851 | fm_local->fm_flags |= FIEMAP_EXTENT_LAST; |
| 1852 | fm_local->fm_mapped_extents = 1; |
| 1853 | |
| 1854 | lcl_fm_ext[0].fe_logical = lun_start; |
| 1855 | lcl_fm_ext[0].fe_length = obd_object_end - |
| 1856 | lun_start; |
| 1857 | lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN; |
| 1858 | |
| 1859 | goto inactive_tgt; |
| 1860 | } |
| 1861 | |
| 1862 | fm_local->fm_start = lun_start; |
| 1863 | fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER; |
| 1864 | memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local)); |
| 1865 | *vallen = fiemap_count_to_size(fm_local->fm_extent_count); |
| 1866 | rc = obd_get_info(NULL, |
| 1867 | lov->lov_tgts[ost_index]->ltd_exp, |
| 1868 | keylen, key, vallen, fm_local, lsm); |
| 1869 | if (rc != 0) |
| 1870 | goto out; |
| 1871 | |
| 1872 | inactive_tgt: |
| 1873 | ext_count = fm_local->fm_mapped_extents; |
| 1874 | if (ext_count == 0) { |
| 1875 | ost_done = 1; |
| 1876 | /* If last stripe has hole at the end, |
| 1877 | * then we need to return */ |
| 1878 | if (cur_stripe_wrap == last_stripe) { |
| 1879 | fiemap->fm_mapped_extents = 0; |
| 1880 | goto finish; |
| 1881 | } |
| 1882 | break; |
| 1883 | } |
| 1884 | |
| 1885 | /* If we just need num of extents then go to next device */ |
| 1886 | if (get_num_extents) { |
| 1887 | current_extent += ext_count; |
| 1888 | break; |
| 1889 | } |
| 1890 | |
| 1891 | len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical - |
| 1892 | lun_start + lcl_fm_ext[ext_count - 1].fe_length; |
| 1893 | |
| 1894 | /* Have we finished mapping on this device? */ |
| 1895 | if (req_fm_len <= len_mapped_single_call) |
| 1896 | ost_done = 1; |
| 1897 | |
| 1898 | /* Clear the EXTENT_LAST flag which can be present on |
| 1899 | * last extent */ |
| 1900 | if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST) |
| 1901 | lcl_fm_ext[ext_count - 1].fe_flags &= |
| 1902 | ~FIEMAP_EXTENT_LAST; |
| 1903 | |
| 1904 | curr_loc = lov_stripe_size(lsm, |
| 1905 | lcl_fm_ext[ext_count - 1].fe_logical+ |
| 1906 | lcl_fm_ext[ext_count - 1].fe_length, |
| 1907 | cur_stripe); |
| 1908 | if (curr_loc >= fm_key->oa.o_size) |
| 1909 | ost_eof = 1; |
| 1910 | |
| 1911 | fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext, |
| 1912 | ost_index, ext_count, |
| 1913 | current_extent); |
| 1914 | |
| 1915 | current_extent += ext_count; |
| 1916 | |
| 1917 | /* Ran out of available extents? */ |
| 1918 | if (current_extent >= fiemap->fm_extent_count) |
| 1919 | goto finish; |
| 1920 | } while (ost_done == 0 && ost_eof == 0); |
| 1921 | |
| 1922 | if (cur_stripe_wrap == last_stripe) |
| 1923 | goto finish; |
| 1924 | } |
| 1925 | |
| 1926 | finish: |
| 1927 | /* Indicate that we are returning device offsets unless file just has |
| 1928 | * single stripe */ |
| 1929 | if (lsm->lsm_stripe_count > 1) |
| 1930 | fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER; |
| 1931 | |
| 1932 | if (get_num_extents) |
| 1933 | goto skip_last_device_calc; |
| 1934 | |
| 1935 | /* Check if we have reached the last stripe and whether mapping for that |
| 1936 | * stripe is done. */ |
| 1937 | if (cur_stripe_wrap == last_stripe) { |
| 1938 | if (ost_done || ost_eof) |
| 1939 | fiemap->fm_extents[current_extent - 1].fe_flags |= |
| 1940 | FIEMAP_EXTENT_LAST; |
| 1941 | } |
| 1942 | |
| 1943 | skip_last_device_calc: |
| 1944 | fiemap->fm_mapped_extents = current_extent; |
| 1945 | |
| 1946 | out: |
| 1947 | kvfree(fm_local); |
| 1948 | return rc; |
| 1949 | } |
| 1950 | |
| 1951 | static int lov_get_info(const struct lu_env *env, struct obd_export *exp, |
| 1952 | __u32 keylen, void *key, __u32 *vallen, void *val, |
| 1953 | struct lov_stripe_md *lsm) |
| 1954 | { |
| 1955 | struct obd_device *obddev = class_exp2obd(exp); |
| 1956 | struct lov_obd *lov = &obddev->u.lov; |
| 1957 | int i, rc; |
| 1958 | |
| 1959 | if (!vallen || !val) |
| 1960 | return -EFAULT; |
| 1961 | |
| 1962 | obd_getref(obddev); |
| 1963 | |
| 1964 | if (KEY_IS(KEY_LOCK_TO_STRIPE)) { |
| 1965 | struct { |
| 1966 | char name[16]; |
| 1967 | struct ldlm_lock *lock; |
| 1968 | } *data = key; |
| 1969 | struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name; |
| 1970 | struct lov_oinfo *loi; |
| 1971 | __u32 *stripe = val; |
| 1972 | |
| 1973 | if (*vallen < sizeof(*stripe)) { |
| 1974 | rc = -EFAULT; |
| 1975 | goto out; |
| 1976 | } |
| 1977 | *vallen = sizeof(*stripe); |
| 1978 | |
| 1979 | /* XXX This is another one of those bits that will need to |
| 1980 | * change if we ever actually support nested LOVs. It uses |
| 1981 | * the lock's export to find out which stripe it is. */ |
| 1982 | /* XXX - it's assumed all the locks for deleted OSTs have |
| 1983 | * been cancelled. Also, the export for deleted OSTs will |
| 1984 | * be NULL and won't match the lock's export. */ |
| 1985 | for (i = 0; i < lsm->lsm_stripe_count; i++) { |
| 1986 | loi = lsm->lsm_oinfo[i]; |
| 1987 | if (lov_oinfo_is_dummy(loi)) |
| 1988 | continue; |
| 1989 | |
| 1990 | if (!lov->lov_tgts[loi->loi_ost_idx]) |
| 1991 | continue; |
| 1992 | if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp == |
| 1993 | data->lock->l_conn_export && |
| 1994 | ostid_res_name_eq(&loi->loi_oi, res_id)) { |
| 1995 | *stripe = i; |
| 1996 | rc = 0; |
| 1997 | goto out; |
| 1998 | } |
| 1999 | } |
| 2000 | LDLM_ERROR(data->lock, "lock on inode without such object"); |
| 2001 | dump_lsm(D_ERROR, lsm); |
| 2002 | rc = -ENXIO; |
| 2003 | goto out; |
| 2004 | } else if (KEY_IS(KEY_LAST_ID)) { |
| 2005 | struct obd_id_info *info = val; |
| 2006 | __u32 size = sizeof(u64); |
| 2007 | struct lov_tgt_desc *tgt; |
| 2008 | |
| 2009 | LASSERT(*vallen == sizeof(struct obd_id_info)); |
| 2010 | tgt = lov->lov_tgts[info->idx]; |
| 2011 | |
| 2012 | if (!tgt || !tgt->ltd_active) { |
| 2013 | rc = -ESRCH; |
| 2014 | goto out; |
| 2015 | } |
| 2016 | |
| 2017 | rc = obd_get_info(env, tgt->ltd_exp, keylen, key, |
| 2018 | &size, info->data, NULL); |
| 2019 | rc = 0; |
| 2020 | goto out; |
| 2021 | } else if (KEY_IS(KEY_LOVDESC)) { |
| 2022 | struct lov_desc *desc_ret = val; |
| 2023 | *desc_ret = lov->desc; |
| 2024 | |
| 2025 | rc = 0; |
| 2026 | goto out; |
| 2027 | } else if (KEY_IS(KEY_FIEMAP)) { |
| 2028 | rc = lov_fiemap(lov, keylen, key, vallen, val, lsm); |
| 2029 | goto out; |
| 2030 | } else if (KEY_IS(KEY_CONNECT_FLAG)) { |
| 2031 | struct lov_tgt_desc *tgt; |
| 2032 | __u64 ost_idx = *((__u64 *)val); |
| 2033 | |
| 2034 | LASSERT(*vallen == sizeof(__u64)); |
| 2035 | LASSERT(ost_idx < lov->desc.ld_tgt_count); |
| 2036 | tgt = lov->lov_tgts[ost_idx]; |
| 2037 | |
| 2038 | if (!tgt || !tgt->ltd_exp) { |
| 2039 | rc = -ESRCH; |
| 2040 | goto out; |
| 2041 | } |
| 2042 | |
| 2043 | *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp); |
| 2044 | rc = 0; |
| 2045 | goto out; |
| 2046 | } else if (KEY_IS(KEY_TGT_COUNT)) { |
| 2047 | *((int *)val) = lov->desc.ld_tgt_count; |
| 2048 | rc = 0; |
| 2049 | goto out; |
| 2050 | } |
| 2051 | |
| 2052 | rc = -EINVAL; |
| 2053 | |
| 2054 | out: |
| 2055 | obd_putref(obddev); |
| 2056 | return rc; |
| 2057 | } |
| 2058 | |
| 2059 | static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp, |
| 2060 | u32 keylen, void *key, u32 vallen, |
| 2061 | void *val, struct ptlrpc_request_set *set) |
| 2062 | { |
| 2063 | struct obd_device *obddev = class_exp2obd(exp); |
| 2064 | struct lov_obd *lov = &obddev->u.lov; |
| 2065 | u32 count; |
| 2066 | int i, rc = 0, err; |
| 2067 | struct lov_tgt_desc *tgt; |
| 2068 | unsigned incr, check_uuid, |
| 2069 | do_inactive, no_set; |
| 2070 | unsigned next_id = 0, mds_con = 0, capa = 0; |
| 2071 | |
| 2072 | incr = check_uuid = do_inactive = no_set = 0; |
| 2073 | if (set == NULL) { |
| 2074 | no_set = 1; |
| 2075 | set = ptlrpc_prep_set(); |
| 2076 | if (!set) |
| 2077 | return -ENOMEM; |
| 2078 | } |
| 2079 | |
| 2080 | obd_getref(obddev); |
| 2081 | count = lov->desc.ld_tgt_count; |
| 2082 | |
| 2083 | if (KEY_IS(KEY_NEXT_ID)) { |
| 2084 | count = vallen / sizeof(struct obd_id_info); |
| 2085 | vallen = sizeof(u64); |
| 2086 | incr = sizeof(struct obd_id_info); |
| 2087 | do_inactive = 1; |
| 2088 | next_id = 1; |
| 2089 | } else if (KEY_IS(KEY_CHECKSUM)) { |
| 2090 | do_inactive = 1; |
| 2091 | } else if (KEY_IS(KEY_EVICT_BY_NID)) { |
| 2092 | /* use defaults: do_inactive = incr = 0; */ |
| 2093 | } else if (KEY_IS(KEY_MDS_CONN)) { |
| 2094 | mds_con = 1; |
| 2095 | } else if (KEY_IS(KEY_CAPA_KEY)) { |
| 2096 | capa = 1; |
| 2097 | } else if (KEY_IS(KEY_CACHE_SET)) { |
| 2098 | LASSERT(lov->lov_cache == NULL); |
| 2099 | lov->lov_cache = val; |
| 2100 | do_inactive = 1; |
| 2101 | } |
| 2102 | |
| 2103 | for (i = 0; i < count; i++, val = (char *)val + incr) { |
| 2104 | if (next_id) { |
| 2105 | tgt = lov->lov_tgts[((struct obd_id_info *)val)->idx]; |
| 2106 | } else { |
| 2107 | tgt = lov->lov_tgts[i]; |
| 2108 | } |
| 2109 | /* OST was disconnected */ |
| 2110 | if (!tgt || !tgt->ltd_exp) |
| 2111 | continue; |
| 2112 | |
| 2113 | /* OST is inactive and we don't want inactive OSCs */ |
| 2114 | if (!tgt->ltd_active && !do_inactive) |
| 2115 | continue; |
| 2116 | |
| 2117 | if (mds_con) { |
| 2118 | struct mds_group_info *mgi; |
| 2119 | |
| 2120 | LASSERT(vallen == sizeof(*mgi)); |
| 2121 | mgi = (struct mds_group_info *)val; |
| 2122 | |
| 2123 | /* Only want a specific OSC */ |
| 2124 | if (mgi->uuid && !obd_uuid_equals(mgi->uuid, |
| 2125 | &tgt->ltd_uuid)) |
| 2126 | continue; |
| 2127 | |
| 2128 | err = obd_set_info_async(env, tgt->ltd_exp, |
| 2129 | keylen, key, sizeof(int), |
| 2130 | &mgi->group, set); |
| 2131 | } else if (next_id) { |
| 2132 | err = obd_set_info_async(env, tgt->ltd_exp, |
| 2133 | keylen, key, vallen, |
| 2134 | ((struct obd_id_info *)val)->data, set); |
| 2135 | } else if (capa) { |
| 2136 | struct mds_capa_info *info = (struct mds_capa_info *)val; |
| 2137 | |
| 2138 | LASSERT(vallen == sizeof(*info)); |
| 2139 | |
| 2140 | /* Only want a specific OSC */ |
| 2141 | if (info->uuid && |
| 2142 | !obd_uuid_equals(info->uuid, &tgt->ltd_uuid)) |
| 2143 | continue; |
| 2144 | |
| 2145 | err = obd_set_info_async(env, tgt->ltd_exp, keylen, |
| 2146 | key, sizeof(*info->capa), |
| 2147 | info->capa, set); |
| 2148 | } else { |
| 2149 | /* Only want a specific OSC */ |
| 2150 | if (check_uuid && |
| 2151 | !obd_uuid_equals(val, &tgt->ltd_uuid)) |
| 2152 | continue; |
| 2153 | |
| 2154 | err = obd_set_info_async(env, tgt->ltd_exp, |
| 2155 | keylen, key, vallen, val, set); |
| 2156 | } |
| 2157 | |
| 2158 | if (!rc) |
| 2159 | rc = err; |
| 2160 | } |
| 2161 | |
| 2162 | obd_putref(obddev); |
| 2163 | if (no_set) { |
| 2164 | err = ptlrpc_set_wait(set); |
| 2165 | if (!rc) |
| 2166 | rc = err; |
| 2167 | ptlrpc_set_destroy(set); |
| 2168 | } |
| 2169 | return rc; |
| 2170 | } |
| 2171 | |
| 2172 | void lov_stripe_lock(struct lov_stripe_md *md) |
| 2173 | __acquires(&md->lsm_lock) |
| 2174 | { |
| 2175 | LASSERT(md->lsm_lock_owner != current_pid()); |
| 2176 | spin_lock(&md->lsm_lock); |
| 2177 | LASSERT(md->lsm_lock_owner == 0); |
| 2178 | md->lsm_lock_owner = current_pid(); |
| 2179 | } |
| 2180 | EXPORT_SYMBOL(lov_stripe_lock); |
| 2181 | |
| 2182 | void lov_stripe_unlock(struct lov_stripe_md *md) |
| 2183 | __releases(&md->lsm_lock) |
| 2184 | { |
| 2185 | LASSERT(md->lsm_lock_owner == current_pid()); |
| 2186 | md->lsm_lock_owner = 0; |
| 2187 | spin_unlock(&md->lsm_lock); |
| 2188 | } |
| 2189 | EXPORT_SYMBOL(lov_stripe_unlock); |
| 2190 | |
| 2191 | static int lov_quotactl(struct obd_device *obd, struct obd_export *exp, |
| 2192 | struct obd_quotactl *oqctl) |
| 2193 | { |
| 2194 | struct lov_obd *lov = &obd->u.lov; |
| 2195 | struct lov_tgt_desc *tgt; |
| 2196 | __u64 curspace = 0; |
| 2197 | __u64 bhardlimit = 0; |
| 2198 | int i, rc = 0; |
| 2199 | |
| 2200 | if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON && |
| 2201 | oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF && |
| 2202 | oqctl->qc_cmd != Q_GETOQUOTA && |
| 2203 | oqctl->qc_cmd != Q_INITQUOTA && |
| 2204 | oqctl->qc_cmd != LUSTRE_Q_SETQUOTA && |
| 2205 | oqctl->qc_cmd != Q_FINVALIDATE) { |
| 2206 | CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd); |
| 2207 | return -EFAULT; |
| 2208 | } |
| 2209 | |
| 2210 | /* for lov tgt */ |
| 2211 | obd_getref(obd); |
| 2212 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 2213 | int err; |
| 2214 | |
| 2215 | tgt = lov->lov_tgts[i]; |
| 2216 | |
| 2217 | if (!tgt) |
| 2218 | continue; |
| 2219 | |
| 2220 | if (!tgt->ltd_active || tgt->ltd_reap) { |
| 2221 | if (oqctl->qc_cmd == Q_GETOQUOTA && |
| 2222 | lov->lov_tgts[i]->ltd_activate) { |
| 2223 | rc = -EREMOTEIO; |
| 2224 | CERROR("ost %d is inactive\n", i); |
| 2225 | } else { |
| 2226 | CDEBUG(D_HA, "ost %d is inactive\n", i); |
| 2227 | } |
| 2228 | continue; |
| 2229 | } |
| 2230 | |
| 2231 | err = obd_quotactl(tgt->ltd_exp, oqctl); |
| 2232 | if (err) { |
| 2233 | if (tgt->ltd_active && !rc) |
| 2234 | rc = err; |
| 2235 | continue; |
| 2236 | } |
| 2237 | |
| 2238 | if (oqctl->qc_cmd == Q_GETOQUOTA) { |
| 2239 | curspace += oqctl->qc_dqblk.dqb_curspace; |
| 2240 | bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit; |
| 2241 | } |
| 2242 | } |
| 2243 | obd_putref(obd); |
| 2244 | |
| 2245 | if (oqctl->qc_cmd == Q_GETOQUOTA) { |
| 2246 | oqctl->qc_dqblk.dqb_curspace = curspace; |
| 2247 | oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit; |
| 2248 | } |
| 2249 | return rc; |
| 2250 | } |
| 2251 | |
| 2252 | static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp, |
| 2253 | struct obd_quotactl *oqctl) |
| 2254 | { |
| 2255 | struct lov_obd *lov = &obd->u.lov; |
| 2256 | int i, rc = 0; |
| 2257 | |
| 2258 | obd_getref(obd); |
| 2259 | |
| 2260 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 2261 | if (!lov->lov_tgts[i]) |
| 2262 | continue; |
| 2263 | |
| 2264 | /* Skip quota check on the administratively disabled OSTs. */ |
| 2265 | if (!lov->lov_tgts[i]->ltd_activate) { |
| 2266 | CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n", |
| 2267 | i); |
| 2268 | continue; |
| 2269 | } |
| 2270 | |
| 2271 | if (!lov->lov_tgts[i]->ltd_active) { |
| 2272 | CERROR("lov idx %d inactive\n", i); |
| 2273 | rc = -EIO; |
| 2274 | goto out; |
| 2275 | } |
| 2276 | } |
| 2277 | |
| 2278 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { |
| 2279 | int err; |
| 2280 | |
| 2281 | if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate) |
| 2282 | continue; |
| 2283 | |
| 2284 | err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl); |
| 2285 | if (err && !rc) |
| 2286 | rc = err; |
| 2287 | } |
| 2288 | |
| 2289 | out: |
| 2290 | obd_putref(obd); |
| 2291 | |
| 2292 | return rc; |
| 2293 | } |
| 2294 | |
| 2295 | static struct obd_ops lov_obd_ops = { |
| 2296 | .o_owner = THIS_MODULE, |
| 2297 | .o_setup = lov_setup, |
| 2298 | .o_precleanup = lov_precleanup, |
| 2299 | .o_cleanup = lov_cleanup, |
| 2300 | /*.o_process_config = lov_process_config,*/ |
| 2301 | .o_connect = lov_connect, |
| 2302 | .o_disconnect = lov_disconnect, |
| 2303 | .o_statfs = lov_statfs, |
| 2304 | .o_statfs_async = lov_statfs_async, |
| 2305 | .o_packmd = lov_packmd, |
| 2306 | .o_unpackmd = lov_unpackmd, |
| 2307 | .o_create = lov_create, |
| 2308 | .o_destroy = lov_destroy, |
| 2309 | .o_getattr_async = lov_getattr_async, |
| 2310 | .o_setattr_async = lov_setattr_async, |
| 2311 | .o_adjust_kms = lov_adjust_kms, |
| 2312 | .o_find_cbdata = lov_find_cbdata, |
| 2313 | .o_iocontrol = lov_iocontrol, |
| 2314 | .o_get_info = lov_get_info, |
| 2315 | .o_set_info_async = lov_set_info_async, |
| 2316 | .o_notify = lov_notify, |
| 2317 | .o_pool_new = lov_pool_new, |
| 2318 | .o_pool_rem = lov_pool_remove, |
| 2319 | .o_pool_add = lov_pool_add, |
| 2320 | .o_pool_del = lov_pool_del, |
| 2321 | .o_getref = lov_getref, |
| 2322 | .o_putref = lov_putref, |
| 2323 | .o_quotactl = lov_quotactl, |
| 2324 | .o_quotacheck = lov_quotacheck, |
| 2325 | }; |
| 2326 | |
| 2327 | struct kmem_cache *lov_oinfo_slab; |
| 2328 | |
| 2329 | static int __init lov_init(void) |
| 2330 | { |
| 2331 | struct lprocfs_static_vars lvars = { NULL }; |
| 2332 | int rc; |
| 2333 | |
| 2334 | /* print an address of _any_ initialized kernel symbol from this |
| 2335 | * module, to allow debugging with gdb that doesn't support data |
| 2336 | * symbols from modules.*/ |
| 2337 | CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches); |
| 2338 | |
| 2339 | rc = lu_kmem_init(lov_caches); |
| 2340 | if (rc) |
| 2341 | return rc; |
| 2342 | |
| 2343 | lov_oinfo_slab = kmem_cache_create("lov_oinfo", |
| 2344 | sizeof(struct lov_oinfo), |
| 2345 | 0, SLAB_HWCACHE_ALIGN, NULL); |
| 2346 | if (lov_oinfo_slab == NULL) { |
| 2347 | lu_kmem_fini(lov_caches); |
| 2348 | return -ENOMEM; |
| 2349 | } |
| 2350 | lprocfs_lov_init_vars(&lvars); |
| 2351 | |
| 2352 | rc = class_register_type(&lov_obd_ops, NULL, |
| 2353 | LUSTRE_LOV_NAME, &lov_device_type); |
| 2354 | |
| 2355 | if (rc) { |
| 2356 | kmem_cache_destroy(lov_oinfo_slab); |
| 2357 | lu_kmem_fini(lov_caches); |
| 2358 | } |
| 2359 | |
| 2360 | return rc; |
| 2361 | } |
| 2362 | |
| 2363 | static void /*__exit*/ lov_exit(void) |
| 2364 | { |
| 2365 | class_unregister_type(LUSTRE_LOV_NAME); |
| 2366 | kmem_cache_destroy(lov_oinfo_slab); |
| 2367 | |
| 2368 | lu_kmem_fini(lov_caches); |
| 2369 | } |
| 2370 | |
| 2371 | MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>"); |
| 2372 | MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver"); |
| 2373 | MODULE_LICENSE("GPL"); |
| 2374 | MODULE_VERSION(LUSTRE_VERSION_STRING); |
| 2375 | |
| 2376 | module_init(lov_init); |
| 2377 | module_exit(lov_exit); |