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 = FILTER_SUBDIR_COUNT;
 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                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 199                         count = le16_to_cpu(lsd.lsd_subdir_count);
 200         } else if (rc != 0) {
 201                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 202                 if (rc > 0)
 203                         rc = -EFAULT;
 204                 dput(dlast);
 205                 return rc;
 206         } else {
 207                 count = FILTER_SUBDIR_COUNT;
 208         }
 209 out:
 210         dput(dlast);
 211         LASSERT(count > 0);
 212         return rc;
 213 }
 214
 215 void osd_compat_fini(struct osd_device *dev)
 216 {
 217         int i;
 218
 219         ENTRY;
 220
 221         if (dev->od_ost_map == NULL)
 222                 RETURN_EXIT;
 223
 224         for (i = 0; i < MAX_OBJID_GROUP; i++)
 225                 osd_compat_seq_fini(dev, i);
 226
 227         dput(dev->od_ost_map->root);
 228         OBD_FREE_PTR(dev->od_ost_map);
 229         dev->od_ost_map = NULL;
 230
 231         EXIT;
 232 }
 233
 234 /*
 235  * directory structure on legacy OST:
 236  *
 237  * O/<seq>/d0-31/<objid>
 238  * O/<seq>/LAST_ID
 239  * last_rcvd
 240  * LAST_GROUP
 241  * CONFIGS
 242  *
 243  */
 244 int osd_compat_init(struct osd_device *dev)
 245 {
 246         struct lvfs_run_ctxt  new;
 247         struct lvfs_run_ctxt  save;
 248         struct dentry        *rootd = osd_sb(dev)->s_root;
 249         struct dentry        *d;
 250         int                   rc;
 251         int                   i;
 252
 253         ENTRY;
 254
 255         OBD_ALLOC_PTR(dev->od_ost_map);
 256         if (dev->od_ost_map == NULL)
 257                 RETURN(-ENOMEM);
 258
 259         /* to get subdir count from last_rcvd */
 260         rc = osd_last_rcvd_subdir_count(dev);
 261         if (rc < 0) {
 262                 OBD_FREE_PTR(dev->od_ost_map);
 263                 RETURN(rc);
 264         }
 265
 266         dev->od_ost_map->subdir_count = rc;
 267         rc = 0;
 268
 269         LASSERT(dev->od_fsops);
 270         osd_push_ctxt(dev, &new, &save);
 271
 272         d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 273         pop_ctxt(&save, &new, NULL);
 274         if (IS_ERR(d)) {
 275                 OBD_FREE_PTR(dev->od_ost_map);
 276                 RETURN(PTR_ERR(d));
 277         }
 278
 279         dev->od_ost_map->root = d;
 280
 281         /* Initialize all groups */
 282         for (i = 0; i < MAX_OBJID_GROUP; i++) {
 283                 cfs_sema_init(&dev->od_ost_map->groups[i].dir_init_sem, 1);
 284                 rc = osd_compat_seq_init(dev, i);
 285                 if (rc) {
 286                         osd_compat_fini(dev);
 287                         break;
 288                 }
 289         }
 290
 291         RETURN(rc);
 292 }
 293
 294 int osd_compat_del_entry(struct osd_thread_info *info, struct osd_device *osd,
 295                          struct dentry *dird, char *name, struct thandle *th)
 296 {
 297         struct ldiskfs_dir_entry_2 *de;
 298         struct buffer_head         *bh;
 299         struct osd_thandle         *oh;
 300         struct dentry              *child;
 301         struct inode               *dir = dird->d_inode;
 302         int                         rc;
 303
 304         ENTRY;
 305
 306         oh = container_of(th, struct osd_thandle, ot_super);
 307         LASSERT(oh->ot_handle != NULL);
 308         LASSERT(oh->ot_handle->h_transaction != NULL);
 309
 310
 311         child = &info->oti_child_dentry;
 312         child->d_name.hash = 0;
 313         child->d_name.name = name;
 314         child->d_name.len = strlen(name);
 315         child->d_parent = dird;
 316         child->d_inode = NULL;
 317
 318         LOCK_INODE_MUTEX(dir);
 319         rc = -ENOENT;
 320         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 321         if (bh) {
 322                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
 323                 brelse(bh);
 324         }
 325         UNLOCK_INODE_MUTEX(dir);
 326
 327         RETURN(rc);
 328 }
 329
 330 int osd_compat_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 331                          struct dentry *dir, char *name,
 332                          const struct osd_inode_id *id, struct thandle *th)
 333 {
 334         struct osd_thandle *oh;
 335         struct dentry *child;
 336         struct inode *inode;
 337         int rc;
 338
 339         ENTRY;
 340
 341         oh = container_of(th, struct osd_thandle, ot_super);
 342         LASSERT(oh->ot_handle != NULL);
 343         LASSERT(oh->ot_handle->h_transaction != NULL);
 344
 345         inode = &info->oti_inode;
 346         inode->i_sb = osd_sb(osd);
 347         inode->i_ino = id->oii_ino;
 348         inode->i_generation = id->oii_gen;
 349
 350         child = &info->oti_child_dentry;
 351         child->d_name.hash = 0;
 352         child->d_name.name = name;
 353         child->d_name.len = strlen(name);
 354         child->d_parent = dir;
 355         child->d_inode = inode;
 356
 357         LOCK_INODE_MUTEX(dir->d_inode);
 358         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 359         UNLOCK_INODE_MUTEX(dir->d_inode);
 360
 361         RETURN(rc);
 362 }
 363
 364 int osd_compat_objid_lookup(struct osd_thread_info *info,
 365                             struct osd_device *dev, const struct lu_fid *fid,
 366                             struct osd_inode_id *id)
 367 {
 368         struct osd_compat_objid    *map;
 369         struct dentry              *d;
 370         struct dentry              *d_seq;
 371         struct ost_id              *ostid = &info->oti_ostid;
 372         int                         rc = 0;
 373         int                         dirn;
 374         char                        name[32];
 375         struct ldiskfs_dir_entry_2 *de;
 376         struct buffer_head         *bh;
 377         struct inode               *dir;
 378         ENTRY;
 379
 380         /* on the very first lookup we find and open directories */
 381
 382         map = dev->od_ost_map;
 383         LASSERT(map);
 384         LASSERT(map->root);
 385
 386         fid_ostid_pack(fid, ostid);
 387         LASSERT(ostid->oi_seq < MAX_OBJID_GROUP);
 388         LASSERT(map->subdir_count > 0);
 389         LASSERT(map->groups[ostid->oi_seq].groot);
 390
 391         dirn = ostid->oi_id & (map->subdir_count - 1);
 392         d = map->groups[ostid->oi_seq].dirs[dirn];
 393         LASSERT(d);
 394
 395         sprintf(name, "%llu", ostid->oi_id);
 396         d_seq = &info->oti_child_dentry;
 397         d_seq->d_parent = d;
 398         d_seq->d_name.hash = 0;
 399         d_seq->d_name.name = name;
 400         /* XXX: we can use rc from sprintf() instead of strlen() */
 401         d_seq->d_name.len = strlen(name);
 402
 403         dir = d->d_inode;
 404         LOCK_INODE_MUTEX(dir);
 405         bh = osd_ldiskfs_find_entry(dir, d_seq, &de, NULL);
 406         UNLOCK_INODE_MUTEX(dir);
 407
 408         rc = -ENOENT;
 409         if (bh) {
 410                 struct inode *inode;
 411
 412                 id->oii_ino = le32_to_cpu(de->inode);
 413                 brelse(bh);
 414
 415                 id->oii_gen = OSD_OII_NOGEN;
 416                 inode = osd_iget(info, dev, id);
 417
 418                 if (IS_ERR(inode))
 419                         GOTO(cleanup, rc = PTR_ERR(inode));
 420                 rc = 0;
 421                 id->oii_gen = inode->i_generation;
 422                 iput(inode);
 423         }
 424
 425 cleanup:
 426         RETURN(rc);
 427 }
 428
 429 int osd_compat_objid_insert(struct osd_thread_info *info,
 430                             struct osd_device *osd,
 431                             const struct lu_fid *fid,
 432                             const struct osd_inode_id *id,
 433                             struct thandle *th)
 434 {
 435         struct osd_compat_objid *map;
 436         struct dentry           *d;
 437         struct ost_id           *ostid = &info->oti_ostid;
 438         int                      dirn, rc = 0;
 439         char                     name[32];
 440         ENTRY;
 441
 442         map = osd->od_ost_map;
 443         LASSERT(map);
 444         LASSERT(map->root);
 445         LASSERT(map->subdir_count > 0);
 446         LASSERT(map->groups[ostid->oi_seq].groot);
 447
 448         /* map fid to group:objid */
 449         fid_ostid_pack(fid, ostid);
 450         dirn = ostid->oi_id & (map->subdir_count - 1);
 451         d = map->groups[ostid->oi_seq].dirs[dirn];
 452         LASSERT(d);
 453
 454         sprintf(name, "%llu", ostid->oi_id);
 455         rc = osd_compat_add_entry(info, osd, d, name, id, th);
 456
 457         RETURN(rc);
 458 }
 459
 460 int osd_compat_objid_delete(struct osd_thread_info *info,
 461                             struct osd_device *osd,
 462                             const struct lu_fid *fid, struct thandle *th)
 463 {
 464         struct osd_compat_objid *map;
 465         struct dentry           *d;
 466         struct ost_id           *ostid = &info->oti_ostid;
 467         int                      dirn, rc = 0;
 468         char                     name[32];
 469         ENTRY;
 470
 471         map = osd->od_ost_map;
 472         LASSERT(map);
 473         LASSERT(map->root);
 474         LASSERT(map->subdir_count > 0);
 475         LASSERT(map->groups[ostid->oi_seq].groot);
 476
 477         /* map fid to group:objid */
 478         fid_ostid_pack(fid, ostid);
 479         dirn = ostid->oi_id & (map->subdir_count - 1);
 480         d = map->groups[ostid->oi_seq].dirs[dirn];
 481         LASSERT(d);
 482
 483         sprintf(name, "%llu", ostid->oi_id);
 484         rc = osd_compat_del_entry(info, osd, d, name, th);
 485
 486         RETURN(rc);
 487 }
 488
 489 struct named_oid {
 490         unsigned long  oid;
 491         char          *name;
 492 };
 493
 494 static const struct named_oid oids[] = {
 495         { FLD_INDEX_OID,        "" /* "fld" */ },
 496         { FID_SEQ_CTL_OID,      "" /* "seq_ctl" */ },
 497         { FID_SEQ_SRV_OID,      "" /* "seq_srv" */ },
 498         { MDD_ROOT_INDEX_OID,   "" /* "ROOT" */ },
 499         { MDD_ORPHAN_OID,       "" /* "PENDING" */ },
 500         { MDD_LOV_OBJ_OID,      "" /* LOV_OBJID */ },
 501         { MDD_CAPA_KEYS_OID,    "" /* CAPA_KEYS */ },
 502         { MDT_LAST_RECV_OID,    LAST_RCVD },
 503         { OFD_LAST_RECV_OID,    "" /* LAST_RCVD */ },
 504         { OFD_LAST_GROUP_OID,   "" /* "LAST_GROUP" */ },
 505         { LLOG_CATALOGS_OID,    "" /* "CATALOGS" */ },
 506         { MGS_CONFIGS_OID,      "" /* MOUNT_CONFIGS_DIR */ },
 507         { OFD_HEALTH_CHECK_OID, "" /* HEALTH_CHECK */ },
 508         { 0,                    NULL }
 509 };
 510
 511 static char *oid2name(const unsigned long oid)
 512 {
 513         int i = 0;
 514
 515         while (oids[i].oid) {
 516                 if (oids[i].oid == oid)
 517                         return oids[i].name;
 518                 i++;
 519         }
 520         return NULL;
 521 }
 522
 523 int osd_compat_spec_insert(struct osd_thread_info *info,
 524                            struct osd_device *osd, const struct lu_fid *fid,
 525                            const struct osd_inode_id *id, struct thandle *th)
 526 {
 527         struct osd_compat_objid *map = osd->od_ost_map;
 528         struct dentry           *root = osd_sb(osd)->s_root;
 529         char                    *name;
 530         int                      rc = 0;
 531         int                      seq;
 532         ENTRY;
 533
 534         if (fid_oid(fid) >= OFD_GROUP0_LAST_OID &&
 535             fid_oid(fid) < OFD_GROUP4K_LAST_OID) {
 536                 /* on creation of LAST_ID we create O/<group> hierarchy */
 537                 LASSERT(map);
 538                 seq = fid_oid(fid) - OFD_GROUP0_LAST_OID;
 539                 LASSERT(seq < MAX_OBJID_GROUP);
 540                 LASSERT(map->groups[seq].groot);
 541         } else {
 542                 name = oid2name(fid_oid(fid));
 543                 if (name == NULL)
 544                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
 545                 else if (name[0])
 546                         rc = osd_compat_add_entry(info, osd, root, name, id,
 547                                                   th);
 548         }
 549
 550         RETURN(rc);
 551 }
 552
 553 int osd_compat_spec_lookup(struct osd_thread_info *info,
 554                            struct osd_device *osd, const struct lu_fid *fid,
 555                            struct osd_inode_id *id)
 556 {
 557         struct dentry *dentry;
 558         char          *name;
 559         int            rc = -ERESTART;
 560
 561         ENTRY;
 562
 563         name = oid2name(fid_oid(fid));
 564         if (name == NULL || strlen(name) == 0)
 565                 return -ERESTART;
 566
 567         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 568         if (!IS_ERR(dentry)) {
 569                 if (dentry->d_inode) {
 570                         if (is_bad_inode(dentry->d_inode)) {
 571                                 rc = -EIO;
 572                         } else {
 573                                 id->oii_ino = dentry->d_inode->i_ino;
 574                                 id->oii_gen = dentry->d_inode->i_generation;
 575                                 rc = 0;
 576                         }
 577                 }
 578                 dput(dentry);
 579         }
 580
 581         RETURN(rc);
 582 }
 583