lustre/osd-ldiskfs/osd_compat.c

   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) 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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/osd/osd_compat.c
  37  *
  38  * on-disk compatibility stuff for OST
  39  *
  40  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  41  */
  42
  43 /* LUSTRE_VERSION_CODE */
  44 #include <lustre_ver.h>
  45 /* prerequisite for linux/xattr.h */
  46 #include <linux/types.h>
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/fs.h>
  49
  50 /*
  51  * struct OBD_{ALLOC,FREE}*()
  52  * OBD_FAIL_CHECK
  53  */
  54 #include <obd_support.h>
  55 #include <lvfs.h>
  56
  57 #include "osd_internal.h"
  58
  59 struct osd_compat_objid_seq {
  60         /* protects on-fly initialization */
  61         cfs_semaphore_t        dir_init_sem;
  62         /* file storing last created objid */
  63         struct osd_inode_id    last_id;
  64         struct dentry         *groot; /* O/<seq> */
  65         struct dentry        **dirs;  /* O/<seq>/d0-dXX */
  66 };
  67
  68 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
  69
  70 struct osd_compat_objid {
  71         int                          subdir_count;
  72         struct dentry               *root;
  73         struct osd_inode_id          last_rcvd_id;
  74         struct osd_inode_id          last_seq_id;
  75         struct osd_compat_objid_seq  groups[MAX_OBJID_GROUP];
  76 };
  77
  78 static void osd_push_ctxt(const struct osd_device *dev,
  79                           struct lvfs_run_ctxt *newctxt,
  80                           struct lvfs_run_ctxt *save)
  81 {
  82         OBD_SET_CTXT_MAGIC(newctxt);
  83         newctxt->pwdmnt = dev->od_mnt;
  84         newctxt->pwd = dev->od_mnt->mnt_root;
  85         newctxt->fs = get_ds();
  86
  87         push_ctxt(save, newctxt, NULL);
  88 }
  89
  90 void osd_compat_seq_fini(struct osd_device *osd, int seq)
  91 {
  92         struct osd_compat_objid_seq *grp;
  93         struct osd_compat_objid     *map = osd->od_ost_map;
  94         int                          i;
  95
  96         ENTRY;
  97
  98         grp = &map->groups[seq];
  99         if (grp->groot ==NULL)
 100                 RETURN_EXIT;
 101         LASSERT(grp->dirs);
 102
 103         for (i = 0; i < map->subdir_count; i++) {
 104                 if (grp->dirs[i] == NULL)
 105                         break;
 106                 dput(grp->dirs[i]);
 107         }
 108
 109         OBD_FREE(grp->dirs, sizeof(struct dentry *) * map->subdir_count);
 110         dput(grp->groot);
 111         EXIT;
 112 }
 113
 114 int osd_compat_seq_init(struct osd_device *osd, int seq)
 115 {
 116         struct osd_compat_objid_seq *grp;
 117         struct osd_compat_objid     *map;
 118         struct dentry               *d;
 119         int                          rc = 0;
 120         char                         name[32];
 121         int                          i;
 122         ENTRY;
 123
 124         map = osd->od_ost_map;
 125         LASSERT(map);
 126         LASSERT(map->root);
 127         grp = &map->groups[seq];
 128
 129         if (grp->groot != NULL)
 130                 RETURN(0);
 131
 132         cfs_down(&grp->dir_init_sem);
 133
 134         sprintf(name, "%d", seq);
 135         d = simple_mkdir(map->root, osd->od_mnt, name, 0755, 1);
 136         if (IS_ERR(d)) {
 137                 rc = PTR_ERR(d);
 138                 GOTO(out, rc);
 139         } else if (d->d_inode == NULL) {
 140                 rc = -EFAULT;
 141                 dput(d);
 142                 GOTO(out, rc);
 143         }
 144
 145         LASSERT(grp->dirs == NULL);
 146         OBD_ALLOC(grp->dirs, sizeof(d) * map->subdir_count);
 147         if (grp->dirs == NULL) {
 148                 dput(d);
 149                 GOTO(out, rc = -ENOMEM);
 150         }
 151
 152         grp->groot = d;
 153         for (i = 0; i < map->subdir_count; i++) {
 154                 sprintf(name, "d%d", i);
 155                 d = simple_mkdir(grp->groot, osd->od_mnt, name, 0755, 1);
 156                 if (IS_ERR(d)) {
 157                         rc = PTR_ERR(d);
 158                         break;
 159                 } else if (d->d_inode == NULL) {
 160                         rc = -EFAULT;
 161                         dput(d);
 162                         break;
 163                 }
 164
 165                 grp->dirs[i] = d;
 166         }
 167
 168         if (rc)
 169                 osd_compat_seq_fini(osd, seq);
 170 out:
 171         cfs_up(&grp->dir_init_sem);
 172         RETURN(rc);
 173 }
 174
 175 int osd_last_rcvd_subdir_count(struct osd_device *osd)
 176 {
 177         struct lr_server_data lsd;
 178         struct dentry        *dlast;
 179         loff_t                off;
 180         int                   rc = 0;
 181         int                   count = 0;
 182
 183         ENTRY;
 184
 185         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 186                                   strlen(LAST_RCVD));
 187         if (IS_ERR(dlast))
 188                 return PTR_ERR(dlast);
 189         else if (dlast->d_inode == NULL)
 190                 goto out;
 191
 192         off = 0;
 193         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 194         if (rc == sizeof(lsd)) {
 195                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 196                        "subdir count = %d\n", lsd.lsd_uuid,
 197                        lsd.lsd_subdir_count);
 198                 count = le16_to_cpu(lsd.lsd_subdir_count);
 199         } else if (rc != 0) {
 200                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 201                 if (rc > 0)
 202                         rc = -EFAULT;
 203                 goto out;
 204         } else {
 205                 count = FILTER_SUBDIR_COUNT;
 206         }
 207
 208         rc = count;
 209 out:
 210         dput(dlast);
 211         return rc;
 212 }
 213
 214 void osd_compat_fini(struct osd_device *dev)
 215 {
 216         int i;
 217
 218         ENTRY;
 219
 220         if (dev->od_ost_map == NULL)
 221                 RETURN_EXIT;
 222
 223         for (i = 0; i < MAX_OBJID_GROUP; i++)
 224                 osd_compat_seq_fini(dev, i);
 225
 226         dput(dev->od_ost_map->root);
 227         OBD_FREE_PTR(dev->od_ost_map);
 228         dev->od_ost_map = NULL;
 229
 230         EXIT;
 231 }
 232
 233 /*
 234  * directory structure on legacy OST:
 235  *
 236  * O/<seq>/d0-31/<objid>
 237  * O/<seq>/LAST_ID
 238  * last_rcvd
 239  * LAST_GROUP
 240  * CONFIGS
 241  *
 242  */
 243 int osd_compat_init(struct osd_device *dev)
 244 {
 245         struct lvfs_run_ctxt  new;
 246         struct lvfs_run_ctxt  save;
 247         struct dentry        *rootd = osd_sb(dev)->s_root;
 248         struct dentry        *d;
 249         int                   rc;
 250         int                   i;
 251
 252         ENTRY;
 253
 254         /* to get subdir count from last_rcvd */
 255         rc = osd_last_rcvd_subdir_count(dev);
 256         if (rc <= 0)
 257                 RETURN(rc);
 258
 259         dev->od_ost_map->subdir_count = rc;
 260         rc = 0;
 261
 262         OBD_ALLOC_PTR(dev->od_ost_map);
 263         if (dev->od_ost_map == NULL)
 264                 RETURN(-ENOMEM);
 265
 266         LASSERT(dev->od_fsops);
 267         osd_push_ctxt(dev, &new, &save);
 268
 269         d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 270         pop_ctxt(&save, &new, NULL);
 271         if (IS_ERR(d)) {
 272                 OBD_FREE_PTR(dev->od_ost_map);
 273                 RETURN(PTR_ERR(d));
 274         }
 275
 276         dev->od_ost_map->root = d;
 277
 278         /* Initialize all groups */
 279         for (i = 0; i < MAX_OBJID_GROUP; i++) {
 280                 cfs_sema_init(&dev->od_ost_map->groups[i].dir_init_sem, 1);
 281                 rc = osd_compat_seq_init(dev, i);
 282                 if (rc) {
 283                         osd_compat_fini(dev);
 284                         break;
 285                 }
 286         }
 287
 288         RETURN(rc);
 289 }
 290
 291 int osd_compat_del_entry(struct osd_thread_info *info, struct osd_device *osd,
 292                          struct dentry *dird, char *name, struct thandle *th)
 293 {
 294         struct ldiskfs_dir_entry_2 *de;
 295         struct buffer_head         *bh;
 296         struct osd_thandle         *oh;
 297         struct dentry              *child;
 298         struct inode               *dir = dird->d_inode;
 299         int                         rc;
 300
 301         ENTRY;
 302
 303         oh = container_of(th, struct osd_thandle, ot_super);
 304         LASSERT(oh->ot_handle != NULL);
 305         LASSERT(oh->ot_handle->h_transaction != NULL);
 306
 307
 308         child = &info->oti_child_dentry;
 309         child->d_name.hash = 0;
 310         child->d_name.name = name;
 311         child->d_name.len = strlen(name);
 312         child->d_parent = dird;
 313         child->d_inode = NULL;
 314
 315         LOCK_INODE_MUTEX(dir);
 316         rc = -ENOENT;
 317         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 318         if (bh) {
 319                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
 320                 brelse(bh);
 321         }
 322         UNLOCK_INODE_MUTEX(dir);
 323
 324         RETURN(rc);
 325 }
 326
 327 int osd_compat_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 328                          struct dentry *dir, char *name,
 329                          const struct osd_inode_id *id, struct thandle *th)
 330 {
 331         struct osd_thandle *oh;
 332         struct dentry *child;
 333         struct inode *inode;
 334         int rc;
 335
 336         ENTRY;
 337
 338         oh = container_of(th, struct osd_thandle, ot_super);
 339         LASSERT(oh->ot_handle != NULL);
 340         LASSERT(oh->ot_handle->h_transaction != NULL);
 341
 342         inode = &info->oti_inode;
 343         inode->i_sb = osd_sb(osd);
 344         inode->i_ino = id->oii_ino;
 345         inode->i_generation = id->oii_gen;
 346
 347         child = &info->oti_child_dentry;
 348         child->d_name.hash = 0;
 349         child->d_name.name = name;
 350         child->d_name.len = strlen(name);
 351         child->d_parent = dir;
 352         child->d_inode = inode;
 353
 354         LOCK_INODE_MUTEX(dir->d_inode);
 355         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 356         UNLOCK_INODE_MUTEX(dir->d_inode);
 357
 358         RETURN(rc);
 359 }
 360
 361 int osd_compat_objid_lookup(struct osd_thread_info *info,
 362                             struct osd_device *dev, const struct lu_fid *fid,
 363                             struct osd_inode_id *id)
 364 {
 365         struct osd_compat_objid    *map;
 366         struct dentry              *d;
 367         struct dentry              *d_seq;
 368         struct ost_id              *ostid = &info->oti_ostid;
 369         int                         rc = 0;
 370         int                         dirn;
 371         char                        name[32];
 372         struct ldiskfs_dir_entry_2 *de;
 373         struct buffer_head         *bh;
 374         struct inode               *dir;
 375         ENTRY;
 376
 377         /* on the very first lookup we find and open directories */
 378
 379         map = dev->od_ost_map;
 380         LASSERT(map);
 381         LASSERT(map->root);
 382
 383         fid_ostid_pack(fid, ostid);
 384         LASSERT(ostid->oi_seq < MAX_OBJID_GROUP);
 385         LASSERT(map->subdir_count > 0);
 386         LASSERT(map->groups[ostid->oi_seq].groot);
 387
 388         dirn = ostid->oi_id & (map->subdir_count - 1);
 389         d = map->groups[ostid->oi_seq].dirs[dirn];
 390         LASSERT(d);
 391
 392         sprintf(name, "%llu", ostid->oi_id);
 393         d_seq = &info->oti_child_dentry;
 394         d_seq->d_parent = d;
 395         d_seq->d_name.hash = 0;
 396         d_seq->d_name.name = name;
 397         /* XXX: we can use rc from sprintf() instead of strlen() */
 398         d_seq->d_name.len = strlen(name);
 399
 400         dir = d->d_inode;
 401         LOCK_INODE_MUTEX(dir);
 402         bh = osd_ldiskfs_find_entry(dir, d_seq, &de, NULL);
 403         UNLOCK_INODE_MUTEX(dir);
 404
 405         rc = -ENOENT;
 406         if (bh) {
 407                 struct inode *inode;
 408
 409                 id->oii_ino = le32_to_cpu(de->inode);
 410                 brelse(bh);
 411
 412                 id->oii_gen = OSD_OII_NOGEN;
 413                 inode = osd_iget(info, dev, id);
 414
 415                 if (IS_ERR(inode))
 416                         GOTO(cleanup, rc = PTR_ERR(inode));
 417                 rc = 0;
 418                 id->oii_gen = inode->i_generation;
 419                 iput(inode);
 420         }
 421
 422 cleanup:
 423         RETURN(rc);
 424 }
 425
 426 int osd_compat_objid_insert(struct osd_thread_info *info,
 427                             struct osd_device *osd,
 428                             const struct lu_fid *fid,
 429                             const struct osd_inode_id *id,
 430                             struct thandle *th)
 431 {
 432         struct osd_compat_objid *map;
 433         struct dentry           *d;
 434         struct ost_id           *ostid = &info->oti_ostid;
 435         int                      dirn, rc = 0;
 436         char                     name[32];
 437         ENTRY;
 438
 439         map = osd->od_ost_map;
 440         LASSERT(map);
 441         LASSERT(map->root);
 442         LASSERT(map->subdir_count > 0);
 443         LASSERT(map->groups[ostid->oi_seq].groot);
 444
 445         /* map fid to group:objid */
 446         fid_ostid_pack(fid, ostid);
 447         dirn = ostid->oi_id & (map->subdir_count - 1);
 448         d = map->groups[ostid->oi_seq].dirs[dirn];
 449         LASSERT(d);
 450
 451         sprintf(name, "%llu", ostid->oi_id);
 452         rc = osd_compat_add_entry(info, osd, d, name, id, th);
 453
 454         RETURN(rc);
 455 }
 456
 457 int osd_compat_objid_delete(struct osd_thread_info *info,
 458                             struct osd_device *osd,
 459                             const struct lu_fid *fid, struct thandle *th)
 460 {
 461         struct osd_compat_objid *map;
 462         struct dentry           *d;
 463         struct ost_id           *ostid = &info->oti_ostid;
 464         int                      dirn, rc = 0;
 465         char                     name[32];
 466         ENTRY;
 467
 468         map = osd->od_ost_map;
 469         LASSERT(map);
 470         LASSERT(map->root);
 471         LASSERT(map->subdir_count > 0);
 472         LASSERT(map->groups[ostid->oi_seq].groot);
 473
 474         /* map fid to group:objid */
 475         fid_ostid_pack(fid, ostid);
 476         dirn = ostid->oi_id & (map->subdir_count - 1);
 477         d = map->groups[ostid->oi_seq].dirs[dirn];
 478         LASSERT(d);
 479
 480         sprintf(name, "%llu", ostid->oi_id);
 481         rc = osd_compat_del_entry(info, osd, d, name, th);
 482
 483         RETURN(rc);
 484 }
 485
 486 struct named_oid {
 487         unsigned long  oid;
 488         char          *name;
 489 };
 490
 491 static const struct named_oid oids[] = {
 492         { FLD_INDEX_OID,        "" /* "fld" */ },
 493         { FID_SEQ_CTL_OID,      "" /* "seq_ctl" */ },
 494         { FID_SEQ_SRV_OID,      "" /* "seq_srv" */ },
 495         { MDD_ROOT_INDEX_OID,   "" /* "ROOT" */ },
 496         { MDD_ORPHAN_OID,       "" /* "PENDING" */ },
 497         { MDD_LOV_OBJ_OID,      "" /* LOV_OBJID */ },
 498         { MDD_CAPA_KEYS_OID,    "" /* CAPA_KEYS */ },
 499         { MDT_LAST_RECV_OID,    "" /* LAST_RCVD */ },
 500         { OFD_LAST_RECV_OID,    "" /* LAST_RCVD */ },
 501         { OFD_LAST_GROUP_OID,   "" /* "LAST_GROUP" */ },
 502         { LLOG_CATALOGS_OID,    "" /* "CATALOGS" */ },
 503         { MGS_CONFIGS_OID,      "" /* MOUNT_CONFIGS_DIR */ },
 504         { OFD_HEALTH_CHECK_OID, "" /* HEALTH_CHECK */ },
 505         { 0,                    NULL }
 506 };
 507
 508 static char *oid2name(const unsigned long oid)
 509 {
 510         int i = 0;
 511
 512         while (oids[i].oid) {
 513                 if (oids[i].oid == oid)
 514                         return oids[i].name;
 515                 i++;
 516         }
 517         return NULL;
 518 }
 519
 520 int osd_compat_spec_insert(struct osd_thread_info *info,
 521                            struct osd_device *osd, const struct lu_fid *fid,
 522                            const struct osd_inode_id *id, struct thandle *th)
 523 {
 524         struct osd_compat_objid *map = osd->od_ost_map;
 525         struct dentry           *root = osd_sb(osd)->s_root;
 526         char                    *name;
 527         int                      rc = 0;
 528         int                      seq;
 529         ENTRY;
 530
 531         if (fid_oid(fid) >= OFD_GROUP0_LAST_OID &&
 532             fid_oid(fid) < OFD_GROUP4K_LAST_OID) {
 533                 /* on creation of LAST_ID we create O/<group> hierarchy */
 534                 LASSERT(map);
 535                 seq = fid_oid(fid) - OFD_GROUP0_LAST_OID;
 536                 LASSERT(seq < MAX_OBJID_GROUP);
 537                 LASSERT(map->groups[seq].groot);
 538         } else {
 539                 name = oid2name(fid_oid(fid));
 540                 if (name == NULL)
 541                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
 542                 else if (name[0])
 543                         rc = osd_compat_add_entry(info, osd, root, name, id,
 544                                                   th);
 545         }
 546
 547         RETURN(rc);
 548 }
 549
 550 int osd_compat_spec_lookup(struct osd_thread_info *info,
 551                            struct osd_device *osd, const struct lu_fid *fid,
 552                            struct osd_inode_id *id)
 553 {
 554         struct dentry *dentry;
 555         char          *name;
 556         int            rc = -ERESTART;
 557
 558         ENTRY;
 559
 560         name = oid2name(fid_oid(fid));
 561         if (name == NULL || strlen(name) == 0)
 562                 return -ERESTART;
 563
 564         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 565         if (!IS_ERR(dentry)) {
 566                 if (dentry->d_inode) {
 567                         if (is_bad_inode(dentry->d_inode)) {
 568                                 rc = -EIO;
 569                         } else {
 570                                 id->oii_ino = dentry->d_inode->i_ino;
 571                                 id->oii_gen = dentry->d_inode->i_generation;
 572                                 rc = 0;
 573                         }
 574                 }
 575                 dput(dentry);
 576         }
 577
 578         RETURN(rc);
 579 }
 580