lustre/osd-ldiskfs/osd_compat.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/osd/osd_compat.c
  39  *
  40  * on-disk compatibility stuff for OST
  41  *
  42  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  43  */
  44
  45 /* LUSTRE_VERSION_CODE */
  46 #include <lustre_ver.h>
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/types.h>
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/fs.h>
  51
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  * OBD_FAIL_CHECK
  55  */
  56 #include <obd_support.h>
  57 #include <lvfs.h>
  58
  59 #include "osd_internal.h"
  60
  61 struct osd_compat_objid_seq {
  62         /* protects on-fly initialization */
  63         cfs_semaphore_t        dir_init_sem;
  64         /* file storing last created objid */
  65         struct osd_inode_id    last_id;
  66         struct dentry         *groot; /* O/<seq> */
  67         struct dentry        **dirs;  /* O/<seq>/d0-dXX */
  68 };
  69
  70 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
  71
  72 struct osd_compat_objid {
  73         int                          subdir_count;
  74         struct dentry               *root;
  75         struct osd_inode_id          last_rcvd_id;
  76         struct osd_inode_id          last_seq_id;
  77         struct osd_compat_objid_seq  groups[MAX_OBJID_GROUP];
  78 };
  79
  80 static void osd_push_ctxt(const struct osd_device *dev,
  81                           struct lvfs_run_ctxt *newctxt,
  82                           struct lvfs_run_ctxt *save)
  83 {
  84         OBD_SET_CTXT_MAGIC(newctxt);
  85         newctxt->pwdmnt = dev->od_mnt;
  86         newctxt->pwd = dev->od_mnt->mnt_root;
  87         newctxt->fs = get_ds();
  88
  89         push_ctxt(save, newctxt, NULL);
  90 }
  91
  92 void osd_compat_seq_fini(struct osd_device *osd, int seq)
  93 {
  94         struct osd_compat_objid_seq *grp;
  95         struct osd_compat_objid     *map = osd->od_ost_map;
  96         int                          i;
  97
  98         ENTRY;
  99
 100         grp = &map->groups[seq];
 101         if (grp->groot ==NULL)
 102                 RETURN_EXIT;
 103         LASSERT(grp->dirs);
 104
 105         for (i = 0; i < map->subdir_count; i++) {
 106                 if (grp->dirs[i] == NULL)
 107                         break;
 108                 dput(grp->dirs[i]);
 109         }
 110
 111         OBD_FREE(grp->dirs, sizeof(struct dentry *) * map->subdir_count);
 112         dput(grp->groot);
 113         EXIT;
 114 }
 115
 116 int osd_compat_seq_init(struct osd_device *osd, int seq)
 117 {
 118         struct osd_compat_objid_seq *grp;
 119         struct osd_compat_objid     *map;
 120         struct dentry               *d;
 121         int                          rc = 0;
 122         char                         name[32];
 123         int                          i;
 124         ENTRY;
 125
 126         map = osd->od_ost_map;
 127         LASSERT(map);
 128         LASSERT(map->root);
 129         grp = &map->groups[seq];
 130
 131         if (grp->groot != NULL)
 132                 RETURN(0);
 133
 134         cfs_down(&grp->dir_init_sem);
 135
 136         sprintf(name, "%d", seq);
 137         d = simple_mkdir(map->root, osd->od_mnt, name, 0755, 1);
 138         if (IS_ERR(d)) {
 139                 rc = PTR_ERR(d);
 140                 GOTO(out, rc);
 141         } else if (d->d_inode == NULL) {
 142                 rc = -EFAULT;
 143                 dput(d);
 144                 GOTO(out, rc);
 145         }
 146
 147         LASSERT(grp->dirs == NULL);
 148         OBD_ALLOC(grp->dirs, sizeof(d) * map->subdir_count);
 149         if (grp->dirs == NULL) {
 150                 dput(d);
 151                 GOTO(out, rc = -ENOMEM);
 152         }
 153
 154         grp->groot = d;
 155         for (i = 0; i < map->subdir_count; i++) {
 156                 sprintf(name, "d%d", i);
 157                 d = simple_mkdir(grp->groot, osd->od_mnt, name, 0755, 1);
 158                 if (IS_ERR(d)) {
 159                         rc = PTR_ERR(d);
 160                         break;
 161                 } else if (d->d_inode == NULL) {
 162                         rc = -EFAULT;
 163                         dput(d);
 164                         break;
 165                 }
 166
 167                 grp->dirs[i] = d;
 168         }
 169
 170         if (rc)
 171                 osd_compat_seq_fini(osd, seq);
 172 out:
 173         cfs_up(&grp->dir_init_sem);
 174         RETURN(rc);
 175 }
 176
 177 int osd_last_rcvd_subdir_count(struct osd_device *osd)
 178 {
 179         struct lr_server_data lsd;
 180         struct dentry        *dlast;
 181         loff_t                off;
 182         int                   rc = 0;
 183         int                   count = 0;
 184
 185         ENTRY;
 186
 187         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 188                                   strlen(LAST_RCVD));
 189         if (IS_ERR(dlast))
 190                 return PTR_ERR(dlast);
 191         else if (dlast->d_inode == NULL)
 192                 goto out;
 193
 194         off = 0;
 195         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 196         if (rc == sizeof(lsd)) {
 197                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 198                        "subdir count = %d\n", lsd.lsd_uuid,
 199                        lsd.lsd_subdir_count);
 200                 count = le16_to_cpu(lsd.lsd_subdir_count);
 201         } else if (rc != 0) {
 202                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 203                 if (rc > 0)
 204                         rc = -EFAULT;
 205                 goto out;
 206         } else {
 207                 count = FILTER_SUBDIR_COUNT;
 208         }
 209
 210         rc = count;
 211 out:
 212         dput(dlast);
 213         return rc;
 214 }
 215
 216 void osd_compat_fini(struct osd_device *dev)
 217 {
 218         int i;
 219
 220         ENTRY;
 221
 222         if (dev->od_ost_map == NULL)
 223                 RETURN_EXIT;
 224
 225         for (i = 0; i < MAX_OBJID_GROUP; i++)
 226                 osd_compat_seq_fini(dev, i);
 227
 228         dput(dev->od_ost_map->root);
 229         OBD_FREE_PTR(dev->od_ost_map);
 230         dev->od_ost_map = NULL;
 231
 232         EXIT;
 233 }
 234
 235 /*
 236  * directory structure on legacy OST:
 237  *
 238  * O/<seq>/d0-31/<objid>
 239  * O/<seq>/LAST_ID
 240  * last_rcvd
 241  * LAST_GROUP
 242  * CONFIGS
 243  *
 244  */
 245 int osd_compat_init(struct osd_device *dev)
 246 {
 247         struct lvfs_run_ctxt  new;
 248         struct lvfs_run_ctxt  save;
 249         struct dentry        *rootd = osd_sb(dev)->s_root;
 250         struct dentry        *d;
 251         int                   rc;
 252         int                   i;
 253
 254         ENTRY;
 255
 256         /* to get subdir count from last_rcvd */
 257         rc = osd_last_rcvd_subdir_count(dev);
 258         if (rc <= 0)
 259                 RETURN(rc);
 260
 261         dev->od_ost_map->subdir_count = rc;
 262         rc = 0;
 263
 264         OBD_ALLOC_PTR(dev->od_ost_map);
 265         if (dev->od_ost_map == NULL)
 266                 RETURN(-ENOMEM);
 267
 268         LASSERT(dev->od_fsops);
 269         osd_push_ctxt(dev, &new, &save);
 270
 271         d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 272         pop_ctxt(&save, &new, NULL);
 273         if (IS_ERR(d)) {
 274                 OBD_FREE_PTR(dev->od_ost_map);
 275                 RETURN(PTR_ERR(d));
 276         }
 277
 278         dev->od_ost_map->root = d;
 279
 280         /* Initialize all groups */
 281         for (i = 0; i < MAX_OBJID_GROUP; i++) {
 282                 cfs_sema_init(&dev->od_ost_map->groups[i].dir_init_sem, 1);
 283                 rc = osd_compat_seq_init(dev, i);
 284                 if (rc) {
 285                         osd_compat_fini(dev);
 286                         break;
 287                 }
 288         }
 289
 290         RETURN(rc);
 291 }
 292
 293 int osd_compat_del_entry(struct osd_thread_info *info, struct osd_device *osd,
 294                          struct dentry *dird, char *name, struct thandle *th)
 295 {
 296         struct ldiskfs_dir_entry_2 *de;
 297         struct buffer_head         *bh;
 298         struct osd_thandle         *oh;
 299         struct dentry              *child;
 300         struct inode               *dir = dird->d_inode;
 301         int                         rc;
 302
 303         ENTRY;
 304
 305         oh = container_of(th, struct osd_thandle, ot_super);
 306         LASSERT(oh->ot_handle != NULL);
 307         LASSERT(oh->ot_handle->h_transaction != NULL);
 308
 309
 310         child = &info->oti_child_dentry;
 311         child->d_name.hash = 0;
 312         child->d_name.name = name;
 313         child->d_name.len = strlen(name);
 314         child->d_parent = dird;
 315         child->d_inode = NULL;
 316
 317         LOCK_INODE_MUTEX(dir);
 318         rc = -ENOENT;
 319         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 320         if (bh) {
 321                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
 322                 brelse(bh);
 323         }
 324         UNLOCK_INODE_MUTEX(dir);
 325
 326         RETURN(rc);
 327 }
 328
 329 int osd_compat_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 330                          struct dentry *dir, char *name,
 331                          const struct osd_inode_id *id, struct thandle *th)
 332 {
 333         struct osd_thandle *oh;
 334         struct dentry *child;
 335         struct inode *inode;
 336         int rc;
 337
 338         ENTRY;
 339
 340         oh = container_of(th, struct osd_thandle, ot_super);
 341         LASSERT(oh->ot_handle != NULL);
 342         LASSERT(oh->ot_handle->h_transaction != NULL);
 343
 344         inode = &info->oti_inode;
 345         inode->i_sb = osd_sb(osd);
 346         inode->i_ino = id->oii_ino;
 347         inode->i_generation = id->oii_gen;
 348
 349         child = &info->oti_child_dentry;
 350         child->d_name.hash = 0;
 351         child->d_name.name = name;
 352         child->d_name.len = strlen(name);
 353         child->d_parent = dir;
 354         child->d_inode = inode;
 355
 356         LOCK_INODE_MUTEX(dir->d_inode);
 357         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 358         UNLOCK_INODE_MUTEX(dir->d_inode);
 359
 360         RETURN(rc);
 361 }
 362
 363 int osd_compat_objid_lookup(struct osd_thread_info *info,
 364                             struct osd_device *dev, const struct lu_fid *fid,
 365                             struct osd_inode_id *id)
 366 {
 367         struct osd_compat_objid    *map;
 368         struct dentry              *d;
 369         struct dentry              *d_seq;
 370         struct ost_id              *ostid = &info->oti_ostid;
 371         int                         rc = 0;
 372         int                         dirn;
 373         char                        name[32];
 374         struct ldiskfs_dir_entry_2 *de;
 375         struct buffer_head         *bh;
 376         struct inode               *dir;
 377         ENTRY;
 378
 379         /* on the very first lookup we find and open directories */
 380
 381         map = dev->od_ost_map;
 382         LASSERT(map);
 383         LASSERT(map->root);
 384
 385         fid_ostid_pack(fid, ostid);
 386         LASSERT(ostid->oi_seq < MAX_OBJID_GROUP);
 387         LASSERT(map->subdir_count > 0);
 388         LASSERT(map->groups[ostid->oi_seq].groot);
 389
 390         dirn = ostid->oi_id & (map->subdir_count - 1);
 391         d = map->groups[ostid->oi_seq].dirs[dirn];
 392         LASSERT(d);
 393
 394         sprintf(name, "%llu", ostid->oi_id);
 395         d_seq = &info->oti_child_dentry;
 396         d_seq->d_parent = d;
 397         d_seq->d_name.hash = 0;
 398         d_seq->d_name.name = name;
 399         /* XXX: we can use rc from sprintf() instead of strlen() */
 400         d_seq->d_name.len = strlen(name);
 401
 402         dir = d->d_inode;
 403         LOCK_INODE_MUTEX(dir);
 404         bh = osd_ldiskfs_find_entry(dir, d_seq, &de, NULL);
 405         UNLOCK_INODE_MUTEX(dir);
 406
 407         rc = -ENOENT;
 408         if (bh) {
 409                 struct inode *inode;
 410
 411                 id->oii_ino = le32_to_cpu(de->inode);
 412                 brelse(bh);
 413
 414                 id->oii_gen = OSD_OII_NOGEN;
 415                 inode = osd_iget(info, dev, id);
 416
 417                 if (IS_ERR(inode))
 418                         GOTO(cleanup, rc = PTR_ERR(inode));
 419                 rc = 0;
 420                 id->oii_gen = inode->i_generation;
 421                 iput(inode);
 422         }
 423
 424 cleanup:
 425         RETURN(rc);
 426 }
 427
 428 int osd_compat_objid_insert(struct osd_thread_info *info,
 429                             struct osd_device *osd,
 430                             const struct lu_fid *fid,
 431                             const struct osd_inode_id *id,
 432                             struct thandle *th)
 433 {
 434         struct osd_compat_objid *map;
 435         struct dentry           *d;
 436         struct ost_id           *ostid = &info->oti_ostid;
 437         int                      dirn, rc = 0;
 438         char                     name[32];
 439         ENTRY;
 440
 441         map = osd->od_ost_map;
 442         LASSERT(map);
 443         LASSERT(map->root);
 444         LASSERT(map->subdir_count > 0);
 445         LASSERT(map->groups[ostid->oi_seq].groot);
 446
 447         /* map fid to group:objid */
 448         fid_ostid_pack(fid, ostid);
 449         dirn = ostid->oi_id & (map->subdir_count - 1);
 450         d = map->groups[ostid->oi_seq].dirs[dirn];
 451         LASSERT(d);
 452
 453         sprintf(name, "%llu", ostid->oi_id);
 454         rc = osd_compat_add_entry(info, osd, d, name, id, th);
 455
 456         RETURN(rc);
 457 }
 458
 459 int osd_compat_objid_delete(struct osd_thread_info *info,
 460                             struct osd_device *osd,
 461                             const struct lu_fid *fid, struct thandle *th)
 462 {
 463         struct osd_compat_objid *map;
 464         struct dentry           *d;
 465         struct ost_id           *ostid = &info->oti_ostid;
 466         int                      dirn, rc = 0;
 467         char                     name[32];
 468         ENTRY;
 469
 470         map = osd->od_ost_map;
 471         LASSERT(map);
 472         LASSERT(map->root);
 473         LASSERT(map->subdir_count > 0);
 474         LASSERT(map->groups[ostid->oi_seq].groot);
 475
 476         /* map fid to group:objid */
 477         fid_ostid_pack(fid, ostid);
 478         dirn = ostid->oi_id & (map->subdir_count - 1);
 479         d = map->groups[ostid->oi_seq].dirs[dirn];
 480         LASSERT(d);
 481
 482         sprintf(name, "%llu", ostid->oi_id);
 483         rc = osd_compat_del_entry(info, osd, d, name, th);
 484
 485         RETURN(rc);
 486 }
 487
 488 struct named_oid {
 489         unsigned long  oid;
 490         char          *name;
 491 };
 492
 493 static const struct named_oid oids[] = {
 494         { FLD_INDEX_OID,        "" /* "fld" */ },
 495         { FID_SEQ_CTL_OID,      "" /* "seq_ctl" */ },
 496         { FID_SEQ_SRV_OID,      "" /* "seq_srv" */ },
 497         { MDD_ROOT_INDEX_OID,   "" /* "ROOT" */ },
 498         { MDD_ORPHAN_OID,       "" /* "PENDING" */ },
 499         { MDD_LOV_OBJ_OID,      "" /* LOV_OBJID */ },
 500         { MDD_CAPA_KEYS_OID,    "" /* CAPA_KEYS */ },
 501         { MDT_LAST_RECV_OID,    "" /* LAST_RCVD */ },
 502         { OFD_LAST_RECV_OID,    "" /* LAST_RCVD */ },
 503         { OFD_LAST_GROUP_OID,   "" /* "LAST_GROUP" */ },
 504         { LLOG_CATALOGS_OID,    "" /* "CATALOGS" */ },
 505         { MGS_CONFIGS_OID,      "" /* MOUNT_CONFIGS_DIR */ },
 506         { OFD_HEALTH_CHECK_OID, "" /* HEALTH_CHECK */ },
 507         { 0,                    NULL }
 508 };
 509
 510 static char *oid2name(const unsigned long oid)
 511 {
 512         int i = 0;
 513
 514         while (oids[i].oid) {
 515                 if (oids[i].oid == oid)
 516                         return oids[i].name;
 517                 i++;
 518         }
 519         return NULL;
 520 }
 521
 522 int osd_compat_spec_insert(struct osd_thread_info *info,
 523                            struct osd_device *osd, const struct lu_fid *fid,
 524                            const struct osd_inode_id *id, struct thandle *th)
 525 {
 526         struct osd_compat_objid *map = osd->od_ost_map;
 527         struct dentry           *root = osd_sb(osd)->s_root;
 528         char                    *name;
 529         int                      rc = 0;
 530         int                      seq;
 531         ENTRY;
 532
 533         if (fid_oid(fid) >= OFD_GROUP0_LAST_OID &&
 534             fid_oid(fid) < OFD_GROUP4K_LAST_OID) {
 535                 /* on creation of LAST_ID we create O/<group> hierarchy */
 536                 LASSERT(map);
 537                 seq = fid_oid(fid) - OFD_GROUP0_LAST_OID;
 538                 LASSERT(seq < MAX_OBJID_GROUP);
 539                 LASSERT(map->groups[seq].groot);
 540         } else {
 541                 name = oid2name(fid_oid(fid));
 542                 if (name == NULL)
 543                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
 544                 else if (name[0])
 545                         rc = osd_compat_add_entry(info, osd, root, name, id,
 546                                                   th);
 547         }
 548
 549         RETURN(rc);
 550 }
 551
 552 int osd_compat_spec_lookup(struct osd_thread_info *info,
 553                            struct osd_device *osd, const struct lu_fid *fid,
 554                            struct osd_inode_id *id)
 555 {
 556         struct dentry *dentry;
 557         char          *name;
 558         int            rc = -ERESTART;
 559
 560         ENTRY;
 561
 562         name = oid2name(fid_oid(fid));
 563         if (name == NULL || strlen(name) == 0)
 564                 return -ERESTART;
 565
 566         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
 567         if (!IS_ERR(dentry)) {
 568                 if (dentry->d_inode) {
 569                         if (is_bad_inode(dentry->d_inode)) {
 570                                 rc = -EIO;
 571                         } else {
 572                                 id->oii_ino = dentry->d_inode->i_ino;
 573                                 id->oii_gen = dentry->d_inode->i_generation;
 574                                 rc = 0;
 575                         }
 576                 }
 577                 dput(dentry);
 578         }
 579
 580         RETURN(rc);
 581 }
 582