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 #include "osd_oi.h"
  59
  60 struct osd_compat_objid_seq {
  61         /* protects on-fly initialization */
  62         struct semaphore        dir_init_sem;
  63         /* file storing last created objid */
  64         struct osd_inode_id    last_id;
  65         struct dentry         *groot; /* O/<seq> */
  66         struct dentry        **dirs;  /* O/<seq>/d0-dXX */
  67 };
  68
  69 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
  70
  71 struct osd_compat_objid {
  72         int                          subdir_count;
  73         struct dentry               *root;
  74         struct osd_inode_id          last_rcvd_id;
  75         struct osd_inode_id          last_seq_id;
  76         struct osd_compat_objid_seq  groups[MAX_OBJID_GROUP];
  77 };
  78
  79 static void osd_push_ctxt(const struct osd_device *dev,
  80                           struct lvfs_run_ctxt *newctxt,
  81                           struct lvfs_run_ctxt *save)
  82 {
  83         OBD_SET_CTXT_MAGIC(newctxt);
  84         newctxt->pwdmnt = dev->od_mnt;
  85         newctxt->pwd = dev->od_mnt->mnt_root;
  86         newctxt->fs = get_ds();
  87
  88         push_ctxt(save, newctxt, NULL);
  89 }
  90
  91 void osd_compat_seq_fini(struct osd_device *osd, int seq)
  92 {
  93         struct osd_compat_objid_seq *grp;
  94         struct osd_compat_objid     *map = osd->od_ost_map;
  95         int                          i;
  96
  97         ENTRY;
  98
  99         grp = &map->groups[seq];
 100         if (grp->groot ==NULL)
 101                 RETURN_EXIT;
 102         LASSERT(grp->dirs);
 103
 104         for (i = 0; i < map->subdir_count; i++) {
 105                 if (grp->dirs[i] == NULL)
 106                         break;
 107                 dput(grp->dirs[i]);
 108         }
 109
 110         OBD_FREE(grp->dirs, sizeof(struct dentry *) * map->subdir_count);
 111         dput(grp->groot);
 112         EXIT;
 113 }
 114
 115 int osd_compat_seq_init(struct osd_device *osd, int seq)
 116 {
 117         struct osd_compat_objid_seq *grp;
 118         struct osd_compat_objid     *map;
 119         struct dentry               *d;
 120         int                          rc = 0;
 121         char                         name[32];
 122         int                          i;
 123         ENTRY;
 124
 125         map = osd->od_ost_map;
 126         LASSERT(map);
 127         LASSERT(map->root);
 128         grp = &map->groups[seq];
 129
 130         if (grp->groot != NULL)
 131                 RETURN(0);
 132
 133         down(&grp->dir_init_sem);
 134
 135         sprintf(name, "%d", seq);
 136         d = simple_mkdir(map->root, osd->od_mnt, name, 0755, 1);
 137         if (IS_ERR(d)) {
 138                 rc = PTR_ERR(d);
 139                 GOTO(out, rc);
 140         } else if (d->d_inode == NULL) {
 141                 rc = -EFAULT;
 142                 dput(d);
 143                 GOTO(out, rc);
 144         }
 145
 146         LASSERT(grp->dirs == NULL);
 147         OBD_ALLOC(grp->dirs, sizeof(d) * map->subdir_count);
 148         if (grp->dirs == NULL) {
 149                 dput(d);
 150                 GOTO(out, rc = -ENOMEM);
 151         }
 152
 153         grp->groot = d;
 154         for (i = 0; i < map->subdir_count; i++) {
 155                 sprintf(name, "d%d", i);
 156                 d = simple_mkdir(grp->groot, osd->od_mnt, name, 0755, 1);
 157                 if (IS_ERR(d)) {
 158                         rc = PTR_ERR(d);
 159                         break;
 160                 } else if (d->d_inode == NULL) {
 161                         rc = -EFAULT;
 162                         dput(d);
 163                         break;
 164                 }
 165
 166                 grp->dirs[i] = d;
 167         }
 168
 169         if (rc)
 170                 osd_compat_seq_fini(osd, seq);
 171 out:
 172         up(&grp->dir_init_sem);
 173         RETURN(rc);
 174 }
 175
 176 int osd_last_rcvd_subdir_count(struct osd_device *osd)
 177 {
 178         struct lr_server_data lsd;
 179         struct dentry        *dlast;
 180         loff_t                off;
 181         int                   rc = 0;
 182         int                   count = FILTER_SUBDIR_COUNT;
 183
 184         ENTRY;
 185
 186         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 187                                   strlen(LAST_RCVD));
 188         if (IS_ERR(dlast))
 189                 return PTR_ERR(dlast);
 190         else if (dlast->d_inode == NULL)
 191                 goto out;
 192
 193         off = 0;
 194         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 195         if (rc == sizeof(lsd)) {
 196                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 197                        "subdir count = %d\n", lsd.lsd_uuid,
 198                        lsd.lsd_subdir_count);
 199                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 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                 dput(dlast);
 206                 return rc;
 207         }
 208 out:
 209         dput(dlast);
 210         LASSERT(count > 0);
 211         return count;
 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         OBD_ALLOC_PTR(dev->od_ost_map);
 255         if (dev->od_ost_map == NULL)
 256                 RETURN(-ENOMEM);
 257
 258         /* to get subdir count from last_rcvd */
 259         rc = osd_last_rcvd_subdir_count(dev);
 260         if (rc < 0) {
 261                 OBD_FREE_PTR(dev->od_ost_map);
 262                 RETURN(rc);
 263         }
 264
 265         dev->od_ost_map->subdir_count = rc;
 266         rc = 0;
 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                 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         ll_vfs_dq_init(dir);
 318         mutex_lock(&dir->i_mutex);
 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         mutex_unlock(&dir->i_mutex);
 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         osd_id_to_inode(inode, id);
 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         ll_vfs_dq_init(dir->d_inode);
 357         mutex_lock(&dir->d_inode->i_mutex);
 358         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 359         mutex_unlock(&dir->d_inode->i_mutex);
 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                         dirn;
 373         char                        name[32];
 374         struct ldiskfs_dir_entry_2 *de;
 375         struct buffer_head         *bh;
 376         struct inode               *dir;
 377         struct inode               *inode;
 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         mutex_lock(&dir->i_mutex);
 405         bh = osd_ldiskfs_find_entry(dir, d_seq, &de, NULL);
 406         mutex_unlock(&dir->i_mutex);
 407
 408         if (bh == NULL)
 409                 RETURN(-ENOENT);
 410
 411         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 412         brelse(bh);
 413
 414         inode = osd_iget(info, dev, id);
 415         if (IS_ERR(inode))
 416                 RETURN(PTR_ERR(inode));
 417
 418         iput(inode);
 419         RETURN(0);
 420 }
 421
 422 int osd_compat_objid_insert(struct osd_thread_info *info,
 423                             struct osd_device *osd,
 424                             const struct lu_fid *fid,
 425                             const struct osd_inode_id *id,
 426                             struct thandle *th)
 427 {
 428         struct osd_compat_objid *map;
 429         struct dentry           *d;
 430         struct ost_id           *ostid = &info->oti_ostid;
 431         int                      dirn, rc = 0;
 432         char                     name[32];
 433         ENTRY;
 434
 435         map = osd->od_ost_map;
 436         LASSERT(map);
 437         LASSERT(map->root);
 438         LASSERT(map->subdir_count > 0);
 439         LASSERT(map->groups[ostid->oi_seq].groot);
 440
 441         /* map fid to group:objid */
 442         fid_ostid_pack(fid, ostid);
 443         dirn = ostid->oi_id & (map->subdir_count - 1);
 444         d = map->groups[ostid->oi_seq].dirs[dirn];
 445         LASSERT(d);
 446
 447         sprintf(name, "%llu", ostid->oi_id);
 448         rc = osd_compat_add_entry(info, osd, d, name, id, th);
 449
 450         RETURN(rc);
 451 }
 452
 453 int osd_compat_objid_delete(struct osd_thread_info *info,
 454                             struct osd_device *osd,
 455                             const struct lu_fid *fid, struct thandle *th)
 456 {
 457         struct osd_compat_objid *map;
 458         struct dentry           *d;
 459         struct ost_id           *ostid = &info->oti_ostid;
 460         int                      dirn, rc = 0;
 461         char                     name[32];
 462         ENTRY;
 463
 464         map = osd->od_ost_map;
 465         LASSERT(map);
 466         LASSERT(map->root);
 467         LASSERT(map->subdir_count > 0);
 468         LASSERT(map->groups[ostid->oi_seq].groot);
 469
 470         /* map fid to group:objid */
 471         fid_ostid_pack(fid, ostid);
 472         dirn = ostid->oi_id & (map->subdir_count - 1);
 473         d = map->groups[ostid->oi_seq].dirs[dirn];
 474         LASSERT(d);
 475
 476         sprintf(name, "%llu", ostid->oi_id);
 477         rc = osd_compat_del_entry(info, osd, d, name, th);
 478
 479         RETURN(rc);
 480 }
 481
 482 struct named_oid {
 483         unsigned long  oid;
 484         char          *name;
 485 };
 486
 487 static const struct named_oid oids[] = {
 488         { FLD_INDEX_OID,        "fld" },
 489         { FID_SEQ_CTL_OID,      "seq_ctl" },
 490         { FID_SEQ_SRV_OID,      "seq_srv" },
 491         { MDD_ROOT_INDEX_OID,   "" /* "ROOT" */ },
 492         { MDD_ORPHAN_OID,       "" /* "PENDING" */ },
 493         { MDD_LOV_OBJ_OID,      LOV_OBJID },
 494         { MDD_CAPA_KEYS_OID,    "" /* CAPA_KEYS */ },
 495         { MDT_LAST_RECV_OID,    LAST_RCVD },
 496         { LFSCK_BOOKMARK_OID,   "" /* "lfsck_bookmark" */ },
 497         { OTABLE_IT_OID,        "" /* "otable iterator" */},
 498         { OFD_LAST_RECV_OID,    LAST_RCVD },
 499         { OFD_LAST_GROUP_OID,   "LAST_GROUP" },
 500         { LLOG_CATALOGS_OID,    "CATALOGS" },
 501         { MGS_CONFIGS_OID,      "" /* MOUNT_CONFIGS_DIR */ },
 502         { OFD_HEALTH_CHECK_OID, HEALTH_CHECK },
 503         { 0,                    NULL }
 504 };
 505
 506 static char *oid2name(const unsigned long oid)
 507 {
 508         int i = 0;
 509
 510         while (oids[i].oid) {
 511                 if (oids[i].oid == oid)
 512                         return oids[i].name;
 513                 i++;
 514         }
 515         return NULL;
 516 }
 517
 518 int osd_compat_spec_insert(struct osd_thread_info *info,
 519                            struct osd_device *osd, const struct lu_fid *fid,
 520                            const struct osd_inode_id *id, struct thandle *th)
 521 {
 522         struct osd_compat_objid *map = osd->od_ost_map;
 523         struct dentry           *root = osd_sb(osd)->s_root;
 524         char                    *name;
 525         int                      rc = 0;
 526         int                      seq;
 527         ENTRY;
 528
 529         if (fid_oid(fid) >= OFD_GROUP0_LAST_OID &&
 530             fid_oid(fid) < OFD_GROUP4K_LAST_OID) {
 531                 /* on creation of LAST_ID we create O/<group> hierarchy */
 532                 LASSERT(map);
 533                 seq = fid_oid(fid) - OFD_GROUP0_LAST_OID;
 534                 LASSERT(seq < MAX_OBJID_GROUP);
 535                 LASSERT(map->groups[seq].groot);
 536                 rc = osd_compat_add_entry(info, osd, map->groups[seq].groot,
 537                                           "LAST_ID", id, th);
 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   *root;
 555         struct dentry *dentry;
 556         struct inode  *inode;
 557         char          *name;
 558         int            rc = -ENOENT;
 559         ENTRY;
 560
 561         if (fid_oid(fid) >= OFD_GROUP0_LAST_OID &&
 562             fid_oid(fid) < OFD_GROUP4K_LAST_OID) {
 563                 struct osd_compat_objid *map = osd->od_ost_map;
 564                 int                      seq;
 565
 566                 LASSERT(map);
 567                 seq = fid_oid(fid) - OFD_GROUP0_LAST_OID;
 568                 LASSERT(seq < MAX_OBJID_GROUP);
 569                 LASSERT(map->groups[seq].groot);
 570                 root = map->groups[seq].groot;
 571                 name = "LAST_ID";
 572         } else {
 573                 root = osd_sb(osd)->s_root;
 574                 name = oid2name(fid_oid(fid));
 575                 if (name == NULL || strlen(name) == 0)
 576                         RETURN(-ENOENT);
 577         }
 578
 579         dentry = ll_lookup_one_len(name, root, strlen(name));
 580         if (!IS_ERR(dentry)) {
 581                 inode = dentry->d_inode;
 582                 if (inode) {
 583                         if (is_bad_inode(inode)) {
 584                                 rc = -EIO;
 585                         } else {
 586                                 osd_id_gen(id, inode->i_ino,
 587                                            inode->i_generation);
 588                                 rc = 0;
 589                         }
 590                 }
 591                 /* if dentry is accessible after osd_compat_spec_insert it
 592                  * will still contain NULL inode, so don't keep it in cache */
 593                 d_invalidate(dentry);
 594                 dput(dentry);
 595         }
 596
 597         RETURN(rc);
 598 }