fs/btrfs/qgroup.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/sched.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/writeback.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/rbtree.h>
  24 #include <linux/slab.h>
  25 #include <linux/workqueue.h>
  26
  27 #include "ctree.h"
  28 #include "transaction.h"
  29 #include "disk-io.h"
  30 #include "locking.h"
  31 #include "ulist.h"
  32 #include "ioctl.h"
  33 #include "backref.h"
  34
  35 /* TODO XXX FIXME
  36  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  37  *  - reorganize keys
  38  *  - compressed
  39  *  - sync
  40  *  - rescan
  41  *  - copy also limits on subvol creation
  42  *  - limit
  43  *  - caches fuer ulists
  44  *  - performance benchmarks
  45  *  - check all ioctl parameters
  46  */
  47
  48 /*
  49  * one struct for each qgroup, organized in fs_info->qgroup_tree.
  50  */
  51 struct btrfs_qgroup {
  52         u64 qgroupid;
  53
  54         /*
  55          * state
  56          */
  57         u64 rfer;       /* referenced */
  58         u64 rfer_cmpr;  /* referenced compressed */
  59         u64 excl;       /* exclusive */
  60         u64 excl_cmpr;  /* exclusive compressed */
  61
  62         /*
  63          * limits
  64          */
  65         u64 lim_flags;  /* which limits are set */
  66         u64 max_rfer;
  67         u64 max_excl;
  68         u64 rsv_rfer;
  69         u64 rsv_excl;
  70
  71         /*
  72          * reservation tracking
  73          */
  74         u64 reserved;
  75
  76         /*
  77          * lists
  78          */
  79         struct list_head groups;  /* groups this group is member of */
  80         struct list_head members; /* groups that are members of this group */
  81         struct list_head dirty;   /* dirty groups */
  82         struct rb_node node;      /* tree of qgroups */
  83
  84         /*
  85          * temp variables for accounting operations
  86          */
  87         u64 tag;
  88         u64 refcnt;
  89 };
  90
  91 /*
  92  * glue structure to represent the relations between qgroups.
  93  */
  94 struct btrfs_qgroup_list {
  95         struct list_head next_group;
  96         struct list_head next_member;
  97         struct btrfs_qgroup *group;
  98         struct btrfs_qgroup *member;
  99 };
 100
 101 /* must be called with qgroup_lock held */
 102 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 103                                            u64 qgroupid)
 104 {
 105         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 106         struct btrfs_qgroup *qgroup;
 107
 108         while (n) {
 109                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 110                 if (qgroup->qgroupid < qgroupid)
 111                         n = n->rb_left;
 112                 else if (qgroup->qgroupid > qgroupid)
 113                         n = n->rb_right;
 114                 else
 115                         return qgroup;
 116         }
 117         return NULL;
 118 }
 119
 120 /* must be called with qgroup_lock held */
 121 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 122                                           u64 qgroupid)
 123 {
 124         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 125         struct rb_node *parent = NULL;
 126         struct btrfs_qgroup *qgroup;
 127
 128         while (*p) {
 129                 parent = *p;
 130                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 131
 132                 if (qgroup->qgroupid < qgroupid)
 133                         p = &(*p)->rb_left;
 134                 else if (qgroup->qgroupid > qgroupid)
 135                         p = &(*p)->rb_right;
 136                 else
 137                         return qgroup;
 138         }
 139
 140         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 141         if (!qgroup)
 142                 return ERR_PTR(-ENOMEM);
 143
 144         qgroup->qgroupid = qgroupid;
 145         INIT_LIST_HEAD(&qgroup->groups);
 146         INIT_LIST_HEAD(&qgroup->members);
 147         INIT_LIST_HEAD(&qgroup->dirty);
 148
 149         rb_link_node(&qgroup->node, parent, p);
 150         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 151
 152         return qgroup;
 153 }
 154
 155 /* must be called with qgroup_lock held */
 156 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 157 {
 158         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 159         struct btrfs_qgroup_list *list;
 160
 161         if (!qgroup)
 162                 return -ENOENT;
 163
 164         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 165         list_del(&qgroup->dirty);
 166
 167         while (!list_empty(&qgroup->groups)) {
 168                 list = list_first_entry(&qgroup->groups,
 169                                         struct btrfs_qgroup_list, next_group);
 170                 list_del(&list->next_group);
 171                 list_del(&list->next_member);
 172                 kfree(list);
 173         }
 174
 175         while (!list_empty(&qgroup->members)) {
 176                 list = list_first_entry(&qgroup->members,
 177                                         struct btrfs_qgroup_list, next_member);
 178                 list_del(&list->next_group);
 179                 list_del(&list->next_member);
 180                 kfree(list);
 181         }
 182         kfree(qgroup);
 183
 184         return 0;
 185 }
 186
 187 /* must be called with qgroup_lock held */
 188 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 189                            u64 memberid, u64 parentid)
 190 {
 191         struct btrfs_qgroup *member;
 192         struct btrfs_qgroup *parent;
 193         struct btrfs_qgroup_list *list;
 194
 195         member = find_qgroup_rb(fs_info, memberid);
 196         parent = find_qgroup_rb(fs_info, parentid);
 197         if (!member || !parent)
 198                 return -ENOENT;
 199
 200         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 201         if (!list)
 202                 return -ENOMEM;
 203
 204         list->group = parent;
 205         list->member = member;
 206         list_add_tail(&list->next_group, &member->groups);
 207         list_add_tail(&list->next_member, &parent->members);
 208
 209         return 0;
 210 }
 211
 212 /* must be called with qgroup_lock held */
 213 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 214                            u64 memberid, u64 parentid)
 215 {
 216         struct btrfs_qgroup *member;
 217         struct btrfs_qgroup *parent;
 218         struct btrfs_qgroup_list *list;
 219
 220         member = find_qgroup_rb(fs_info, memberid);
 221         parent = find_qgroup_rb(fs_info, parentid);
 222         if (!member || !parent)
 223                 return -ENOENT;
 224
 225         list_for_each_entry(list, &member->groups, next_group) {
 226                 if (list->group == parent) {
 227                         list_del(&list->next_group);
 228                         list_del(&list->next_member);
 229                         kfree(list);
 230                         return 0;
 231                 }
 232         }
 233         return -ENOENT;
 234 }
 235
 236 /*
 237  * The full config is read in one go, only called from open_ctree()
 238  * It doesn't use any locking, as at this point we're still single-threaded
 239  */
 240 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 241 {
 242         struct btrfs_key key;
 243         struct btrfs_key found_key;
 244         struct btrfs_root *quota_root = fs_info->quota_root;
 245         struct btrfs_path *path = NULL;
 246         struct extent_buffer *l;
 247         int slot;
 248         int ret = 0;
 249         u64 flags = 0;
 250
 251         if (!fs_info->quota_enabled)
 252                 return 0;
 253
 254         path = btrfs_alloc_path();
 255         if (!path) {
 256                 ret = -ENOMEM;
 257                 goto out;
 258         }
 259
 260         /* default this to quota off, in case no status key is found */
 261         fs_info->qgroup_flags = 0;
 262
 263         /*
 264          * pass 1: read status, all qgroup infos and limits
 265          */
 266         key.objectid = 0;
 267         key.type = 0;
 268         key.offset = 0;
 269         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 270         if (ret)
 271                 goto out;
 272
 273         while (1) {
 274                 struct btrfs_qgroup *qgroup;
 275
 276                 slot = path->slots[0];
 277                 l = path->nodes[0];
 278                 btrfs_item_key_to_cpu(l, &found_key, slot);
 279
 280                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 281                         struct btrfs_qgroup_status_item *ptr;
 282
 283                         ptr = btrfs_item_ptr(l, slot,
 284                                              struct btrfs_qgroup_status_item);
 285
 286                         if (btrfs_qgroup_status_version(l, ptr) !=
 287                             BTRFS_QGROUP_STATUS_VERSION) {
 288                                 printk(KERN_ERR
 289                                  "btrfs: old qgroup version, quota disabled\n");
 290                                 goto out;
 291                         }
 292                         if (btrfs_qgroup_status_generation(l, ptr) !=
 293                             fs_info->generation) {
 294                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 295                                 printk(KERN_ERR
 296                                         "btrfs: qgroup generation mismatch, "
 297                                         "marked as inconsistent\n");
 298                         }
 299                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 300                                                                           ptr);
 301                         /* FIXME read scan element */
 302                         goto next1;
 303                 }
 304
 305                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 306                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 307                         goto next1;
 308
 309                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 310                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 311                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 312                         printk(KERN_ERR "btrfs: inconsitent qgroup config\n");
 313                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 314                 }
 315                 if (!qgroup) {
 316                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 317                         if (IS_ERR(qgroup)) {
 318                                 ret = PTR_ERR(qgroup);
 319                                 goto out;
 320                         }
 321                 }
 322                 switch (found_key.type) {
 323                 case BTRFS_QGROUP_INFO_KEY: {
 324                         struct btrfs_qgroup_info_item *ptr;
 325
 326                         ptr = btrfs_item_ptr(l, slot,
 327                                              struct btrfs_qgroup_info_item);
 328                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 329                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 330                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 331                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 332                         /* generation currently unused */
 333                         break;
 334                 }
 335                 case BTRFS_QGROUP_LIMIT_KEY: {
 336                         struct btrfs_qgroup_limit_item *ptr;
 337
 338                         ptr = btrfs_item_ptr(l, slot,
 339                                              struct btrfs_qgroup_limit_item);
 340                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 341                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 342                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 343                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 344                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 345                         break;
 346                 }
 347                 }
 348 next1:
 349                 ret = btrfs_next_item(quota_root, path);
 350                 if (ret < 0)
 351                         goto out;
 352                 if (ret)
 353                         break;
 354         }
 355         btrfs_release_path(path);
 356
 357         /*
 358          * pass 2: read all qgroup relations
 359          */
 360         key.objectid = 0;
 361         key.type = BTRFS_QGROUP_RELATION_KEY;
 362         key.offset = 0;
 363         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 364         if (ret)
 365                 goto out;
 366         while (1) {
 367                 slot = path->slots[0];
 368                 l = path->nodes[0];
 369                 btrfs_item_key_to_cpu(l, &found_key, slot);
 370
 371                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 372                         goto next2;
 373
 374                 if (found_key.objectid > found_key.offset) {
 375                         /* parent <- member, not needed to build config */
 376                         /* FIXME should we omit the key completely? */
 377                         goto next2;
 378                 }
 379
 380                 ret = add_relation_rb(fs_info, found_key.objectid,
 381                                       found_key.offset);
 382                 if (ret)
 383                         goto out;
 384 next2:
 385                 ret = btrfs_next_item(quota_root, path);
 386                 if (ret < 0)
 387                         goto out;
 388                 if (ret)
 389                         break;
 390         }
 391 out:
 392         fs_info->qgroup_flags |= flags;
 393         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
 394                 fs_info->quota_enabled = 0;
 395                 fs_info->pending_quota_state = 0;
 396         }
 397         btrfs_free_path(path);
 398
 399         return ret < 0 ? ret : 0;
 400 }
 401
 402 /*
 403  * This is only called from close_ctree() or open_ctree(), both in single-
 404  * treaded paths. Clean up the in-memory structures. No locking needed.
 405  */
 406 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 407 {
 408         struct rb_node *n;
 409         struct btrfs_qgroup *qgroup;
 410         struct btrfs_qgroup_list *list;
 411
 412         while ((n = rb_first(&fs_info->qgroup_tree))) {
 413                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 414                 rb_erase(n, &fs_info->qgroup_tree);
 415
 416                 WARN_ON(!list_empty(&qgroup->dirty));
 417
 418                 while (!list_empty(&qgroup->groups)) {
 419                         list = list_first_entry(&qgroup->groups,
 420                                                 struct btrfs_qgroup_list,
 421                                                 next_group);
 422                         list_del(&list->next_group);
 423                         list_del(&list->next_member);
 424                         kfree(list);
 425                 }
 426
 427                 while (!list_empty(&qgroup->members)) {
 428                         list = list_first_entry(&qgroup->members,
 429                                                 struct btrfs_qgroup_list,
 430                                                 next_member);
 431                         list_del(&list->next_group);
 432                         list_del(&list->next_member);
 433                         kfree(list);
 434                 }
 435                 kfree(qgroup);
 436         }
 437 }
 438
 439 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
 440                                     struct btrfs_root *quota_root,
 441                                     u64 src, u64 dst)
 442 {
 443         int ret;
 444         struct btrfs_path *path;
 445         struct btrfs_key key;
 446
 447         path = btrfs_alloc_path();
 448         if (!path)
 449                 return -ENOMEM;
 450
 451         key.objectid = src;
 452         key.type = BTRFS_QGROUP_RELATION_KEY;
 453         key.offset = dst;
 454
 455         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 456
 457         btrfs_mark_buffer_dirty(path->nodes[0]);
 458
 459         btrfs_free_path(path);
 460         return ret;
 461 }
 462
 463 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
 464                                     struct btrfs_root *quota_root,
 465                                     u64 src, u64 dst)
 466 {
 467         int ret;
 468         struct btrfs_path *path;
 469         struct btrfs_key key;
 470
 471         path = btrfs_alloc_path();
 472         if (!path)
 473                 return -ENOMEM;
 474
 475         key.objectid = src;
 476         key.type = BTRFS_QGROUP_RELATION_KEY;
 477         key.offset = dst;
 478
 479         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 480         if (ret < 0)
 481                 goto out;
 482
 483         if (ret > 0) {
 484                 ret = -ENOENT;
 485                 goto out;
 486         }
 487
 488         ret = btrfs_del_item(trans, quota_root, path);
 489 out:
 490         btrfs_free_path(path);
 491         return ret;
 492 }
 493
 494 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 495                            struct btrfs_root *quota_root, u64 qgroupid)
 496 {
 497         int ret;
 498         struct btrfs_path *path;
 499         struct btrfs_qgroup_info_item *qgroup_info;
 500         struct btrfs_qgroup_limit_item *qgroup_limit;
 501         struct extent_buffer *leaf;
 502         struct btrfs_key key;
 503
 504         path = btrfs_alloc_path();
 505         if (!path)
 506                 return -ENOMEM;
 507
 508         key.objectid = 0;
 509         key.type = BTRFS_QGROUP_INFO_KEY;
 510         key.offset = qgroupid;
 511
 512         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 513                                       sizeof(*qgroup_info));
 514         if (ret)
 515                 goto out;
 516
 517         leaf = path->nodes[0];
 518         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 519                                  struct btrfs_qgroup_info_item);
 520         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 521         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 522         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 523         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 524         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 525
 526         btrfs_mark_buffer_dirty(leaf);
 527
 528         btrfs_release_path(path);
 529
 530         key.type = BTRFS_QGROUP_LIMIT_KEY;
 531         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 532                                       sizeof(*qgroup_limit));
 533         if (ret)
 534                 goto out;
 535
 536         leaf = path->nodes[0];
 537         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 538                                   struct btrfs_qgroup_limit_item);
 539         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 540         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 541         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 542         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 543         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 544
 545         btrfs_mark_buffer_dirty(leaf);
 546
 547         ret = 0;
 548 out:
 549         btrfs_free_path(path);
 550         return ret;
 551 }
 552
 553 static int del_qgroup_item(struct btrfs_trans_handle *trans,
 554                            struct btrfs_root *quota_root, u64 qgroupid)
 555 {
 556         int ret;
 557         struct btrfs_path *path;
 558         struct btrfs_key key;
 559
 560         path = btrfs_alloc_path();
 561         if (!path)
 562                 return -ENOMEM;
 563
 564         key.objectid = 0;
 565         key.type = BTRFS_QGROUP_INFO_KEY;
 566         key.offset = qgroupid;
 567         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 568         if (ret < 0)
 569                 goto out;
 570
 571         if (ret > 0) {
 572                 ret = -ENOENT;
 573                 goto out;
 574         }
 575
 576         ret = btrfs_del_item(trans, quota_root, path);
 577         if (ret)
 578                 goto out;
 579
 580         btrfs_release_path(path);
 581
 582         key.type = BTRFS_QGROUP_LIMIT_KEY;
 583         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 584         if (ret < 0)
 585                 goto out;
 586
 587         if (ret > 0) {
 588                 ret = -ENOENT;
 589                 goto out;
 590         }
 591
 592         ret = btrfs_del_item(trans, quota_root, path);
 593
 594 out:
 595         btrfs_free_path(path);
 596         return ret;
 597 }
 598
 599 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 600                                     struct btrfs_root *root, u64 qgroupid,
 601                                     u64 flags, u64 max_rfer, u64 max_excl,
 602                                     u64 rsv_rfer, u64 rsv_excl)
 603 {
 604         struct btrfs_path *path;
 605         struct btrfs_key key;
 606         struct extent_buffer *l;
 607         struct btrfs_qgroup_limit_item *qgroup_limit;
 608         int ret;
 609         int slot;
 610
 611         key.objectid = 0;
 612         key.type = BTRFS_QGROUP_LIMIT_KEY;
 613         key.offset = qgroupid;
 614
 615         path = btrfs_alloc_path();
 616         BUG_ON(!path);
 617         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 618         if (ret > 0)
 619                 ret = -ENOENT;
 620
 621         if (ret)
 622                 goto out;
 623
 624         l = path->nodes[0];
 625         slot = path->slots[0];
 626         qgroup_limit = btrfs_item_ptr(l, path->slots[0],
 627                                       struct btrfs_qgroup_limit_item);
 628         btrfs_set_qgroup_limit_flags(l, qgroup_limit, flags);
 629         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, max_rfer);
 630         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, max_excl);
 631         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, rsv_rfer);
 632         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, rsv_excl);
 633
 634         btrfs_mark_buffer_dirty(l);
 635
 636 out:
 637         btrfs_free_path(path);
 638         return ret;
 639 }
 640
 641 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 642                                    struct btrfs_root *root,
 643                                    struct btrfs_qgroup *qgroup)
 644 {
 645         struct btrfs_path *path;
 646         struct btrfs_key key;
 647         struct extent_buffer *l;
 648         struct btrfs_qgroup_info_item *qgroup_info;
 649         int ret;
 650         int slot;
 651
 652         key.objectid = 0;
 653         key.type = BTRFS_QGROUP_INFO_KEY;
 654         key.offset = qgroup->qgroupid;
 655
 656         path = btrfs_alloc_path();
 657         BUG_ON(!path);
 658         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 659         if (ret > 0)
 660                 ret = -ENOENT;
 661
 662         if (ret)
 663                 goto out;
 664
 665         l = path->nodes[0];
 666         slot = path->slots[0];
 667         qgroup_info = btrfs_item_ptr(l, path->slots[0],
 668                                  struct btrfs_qgroup_info_item);
 669         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 670         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 671         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 672         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 673         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 674
 675         btrfs_mark_buffer_dirty(l);
 676
 677 out:
 678         btrfs_free_path(path);
 679         return ret;
 680 }
 681
 682 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 683                                      struct btrfs_fs_info *fs_info,
 684                                     struct btrfs_root *root)
 685 {
 686         struct btrfs_path *path;
 687         struct btrfs_key key;
 688         struct extent_buffer *l;
 689         struct btrfs_qgroup_status_item *ptr;
 690         int ret;
 691         int slot;
 692
 693         key.objectid = 0;
 694         key.type = BTRFS_QGROUP_STATUS_KEY;
 695         key.offset = 0;
 696
 697         path = btrfs_alloc_path();
 698         BUG_ON(!path);
 699         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 700         if (ret > 0)
 701                 ret = -ENOENT;
 702
 703         if (ret)
 704                 goto out;
 705
 706         l = path->nodes[0];
 707         slot = path->slots[0];
 708         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 709         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 710         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 711         /* XXX scan */
 712
 713         btrfs_mark_buffer_dirty(l);
 714
 715 out:
 716         btrfs_free_path(path);
 717         return ret;
 718 }
 719
 720 /*
 721  * called with qgroup_lock held
 722  */
 723 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 724                                   struct btrfs_root *root)
 725 {
 726         struct btrfs_path *path;
 727         struct btrfs_key key;
 728         int ret;
 729
 730         if (!root)
 731                 return -EINVAL;
 732
 733         path = btrfs_alloc_path();
 734         if (!path)
 735                 return -ENOMEM;
 736
 737         while (1) {
 738                 key.objectid = 0;
 739                 key.offset = 0;
 740                 key.type = 0;
 741
 742                 path->leave_spinning = 1;
 743                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 744                 if (ret > 0) {
 745                         if (path->slots[0] == 0)
 746                                 break;
 747                         path->slots[0]--;
 748                 } else if (ret < 0) {
 749                         break;
 750                 }
 751
 752                 ret = btrfs_del_item(trans, root, path);
 753                 if (ret)
 754                         goto out;
 755                 btrfs_release_path(path);
 756         }
 757         ret = 0;
 758 out:
 759         root->fs_info->pending_quota_state = 0;
 760         btrfs_free_path(path);
 761         return ret;
 762 }
 763
 764 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 765                        struct btrfs_fs_info *fs_info)
 766 {
 767         struct btrfs_root *quota_root;
 768         struct btrfs_path *path = NULL;
 769         struct btrfs_qgroup_status_item *ptr;
 770         struct extent_buffer *leaf;
 771         struct btrfs_key key;
 772         int ret = 0;
 773
 774         spin_lock(&fs_info->qgroup_lock);
 775         if (fs_info->quota_root) {
 776                 fs_info->pending_quota_state = 1;
 777                 spin_unlock(&fs_info->qgroup_lock);
 778                 goto out;
 779         }
 780         spin_unlock(&fs_info->qgroup_lock);
 781
 782         /*
 783          * initially create the quota tree
 784          */
 785         quota_root = btrfs_create_tree(trans, fs_info,
 786                                        BTRFS_QUOTA_TREE_OBJECTID);
 787         if (IS_ERR(quota_root)) {
 788                 ret =  PTR_ERR(quota_root);
 789                 goto out;
 790         }
 791
 792         path = btrfs_alloc_path();
 793         if (!path)
 794                 return -ENOMEM;
 795
 796         key.objectid = 0;
 797         key.type = BTRFS_QGROUP_STATUS_KEY;
 798         key.offset = 0;
 799
 800         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 801                                       sizeof(*ptr));
 802         if (ret)
 803                 goto out;
 804
 805         leaf = path->nodes[0];
 806         ptr = btrfs_item_ptr(leaf, path->slots[0],
 807                                  struct btrfs_qgroup_status_item);
 808         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 809         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 810         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 811                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 812         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 813         btrfs_set_qgroup_status_scan(leaf, ptr, 0);
 814
 815         btrfs_mark_buffer_dirty(leaf);
 816
 817         spin_lock(&fs_info->qgroup_lock);
 818         fs_info->quota_root = quota_root;
 819         fs_info->pending_quota_state = 1;
 820         spin_unlock(&fs_info->qgroup_lock);
 821 out:
 822         btrfs_free_path(path);
 823         return ret;
 824 }
 825
 826 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
 827                         struct btrfs_fs_info *fs_info)
 828 {
 829         struct btrfs_root *tree_root = fs_info->tree_root;
 830         struct btrfs_root *quota_root;
 831         int ret = 0;
 832
 833         spin_lock(&fs_info->qgroup_lock);
 834         fs_info->quota_enabled = 0;
 835         fs_info->pending_quota_state = 0;
 836         quota_root = fs_info->quota_root;
 837         fs_info->quota_root = NULL;
 838         btrfs_free_qgroup_config(fs_info);
 839         spin_unlock(&fs_info->qgroup_lock);
 840
 841         if (!quota_root)
 842                 return -EINVAL;
 843
 844         ret = btrfs_clean_quota_tree(trans, quota_root);
 845         if (ret)
 846                 goto out;
 847
 848         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
 849         if (ret)
 850                 goto out;
 851
 852         list_del(&quota_root->dirty_list);
 853
 854         btrfs_tree_lock(quota_root->node);
 855         clean_tree_block(trans, tree_root, quota_root->node);
 856         btrfs_tree_unlock(quota_root->node);
 857         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
 858
 859         free_extent_buffer(quota_root->node);
 860         free_extent_buffer(quota_root->commit_root);
 861         kfree(quota_root);
 862 out:
 863         return ret;
 864 }
 865
 866 int btrfs_quota_rescan(struct btrfs_fs_info *fs_info)
 867 {
 868         /* FIXME */
 869         return 0;
 870 }
 871
 872 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
 873                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
 874 {
 875         struct btrfs_root *quota_root;
 876         int ret = 0;
 877
 878         quota_root = fs_info->quota_root;
 879         if (!quota_root)
 880                 return -EINVAL;
 881
 882         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
 883         if (ret)
 884                 return ret;
 885
 886         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
 887         if (ret) {
 888                 del_qgroup_relation_item(trans, quota_root, src, dst);
 889                 return ret;
 890         }
 891
 892         spin_lock(&fs_info->qgroup_lock);
 893         ret = add_relation_rb(quota_root->fs_info, src, dst);
 894         spin_unlock(&fs_info->qgroup_lock);
 895
 896         return ret;
 897 }
 898
 899 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
 900                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
 901 {
 902         struct btrfs_root *quota_root;
 903         int ret = 0;
 904         int err;
 905
 906         quota_root = fs_info->quota_root;
 907         if (!quota_root)
 908                 return -EINVAL;
 909
 910         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
 911         err = del_qgroup_relation_item(trans, quota_root, dst, src);
 912         if (err && !ret)
 913                 ret = err;
 914
 915         spin_lock(&fs_info->qgroup_lock);
 916         del_relation_rb(fs_info, src, dst);
 917
 918         spin_unlock(&fs_info->qgroup_lock);
 919
 920         return ret;
 921 }
 922
 923 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
 924                         struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
 925 {
 926         struct btrfs_root *quota_root;
 927         struct btrfs_qgroup *qgroup;
 928         int ret = 0;
 929
 930         quota_root = fs_info->quota_root;
 931         if (!quota_root)
 932                 return -EINVAL;
 933
 934         ret = add_qgroup_item(trans, quota_root, qgroupid);
 935
 936         spin_lock(&fs_info->qgroup_lock);
 937         qgroup = add_qgroup_rb(fs_info, qgroupid);
 938         spin_unlock(&fs_info->qgroup_lock);
 939
 940         if (IS_ERR(qgroup))
 941                 ret = PTR_ERR(qgroup);
 942
 943         return ret;
 944 }
 945
 946 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
 947                         struct btrfs_fs_info *fs_info, u64 qgroupid)
 948 {
 949         struct btrfs_root *quota_root;
 950         int ret = 0;
 951
 952         quota_root = fs_info->quota_root;
 953         if (!quota_root)
 954                 return -EINVAL;
 955
 956         ret = del_qgroup_item(trans, quota_root, qgroupid);
 957
 958         spin_lock(&fs_info->qgroup_lock);
 959         del_qgroup_rb(quota_root->fs_info, qgroupid);
 960
 961         spin_unlock(&fs_info->qgroup_lock);
 962
 963         return ret;
 964 }
 965
 966 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
 967                        struct btrfs_fs_info *fs_info, u64 qgroupid,
 968                        struct btrfs_qgroup_limit *limit)
 969 {
 970         struct btrfs_root *quota_root = fs_info->quota_root;
 971         struct btrfs_qgroup *qgroup;
 972         int ret = 0;
 973
 974         if (!quota_root)
 975                 return -EINVAL;
 976
 977         ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
 978                                        limit->flags, limit->max_rfer,
 979                                        limit->max_excl, limit->rsv_rfer,
 980                                        limit->rsv_excl);
 981         if (ret) {
 982                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 983                 printk(KERN_INFO "unable to update quota limit for %llu\n",
 984                        (unsigned long long)qgroupid);
 985         }
 986
 987         spin_lock(&fs_info->qgroup_lock);
 988
 989         qgroup = find_qgroup_rb(fs_info, qgroupid);
 990         if (!qgroup) {
 991                 ret = -ENOENT;
 992                 goto unlock;
 993         }
 994         qgroup->lim_flags = limit->flags;
 995         qgroup->max_rfer = limit->max_rfer;
 996         qgroup->max_excl = limit->max_excl;
 997         qgroup->rsv_rfer = limit->rsv_rfer;
 998         qgroup->rsv_excl = limit->rsv_excl;
 999
1000 unlock:
1001         spin_unlock(&fs_info->qgroup_lock);
1002
1003         return ret;
1004 }
1005
1006 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1007                          struct btrfs_qgroup *qgroup)
1008 {
1009         if (list_empty(&qgroup->dirty))
1010                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1011 }
1012
1013 /*
1014  * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1015  * the modification into a list that's later used by btrfs_end_transaction to
1016  * pass the recorded modifications on to btrfs_qgroup_account_ref.
1017  */
1018 int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1019                             struct btrfs_delayed_ref_node *node,
1020                             struct btrfs_delayed_extent_op *extent_op)
1021 {
1022         struct qgroup_update *u;
1023
1024         BUG_ON(!trans->delayed_ref_elem.seq);
1025         u = kmalloc(sizeof(*u), GFP_NOFS);
1026         if (!u)
1027                 return -ENOMEM;
1028
1029         u->node = node;
1030         u->extent_op = extent_op;
1031         list_add_tail(&u->list, &trans->qgroup_ref_list);
1032
1033         return 0;
1034 }
1035
1036 /*
1037  * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1038  * from the fs. First, all roots referencing the extent are searched, and
1039  * then the space is accounted accordingly to the different roots. The
1040  * accounting algorithm works in 3 steps documented inline.
1041  */
1042 int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1043                              struct btrfs_fs_info *fs_info,
1044                              struct btrfs_delayed_ref_node *node,
1045                              struct btrfs_delayed_extent_op *extent_op)
1046 {
1047         struct btrfs_key ins;
1048         struct btrfs_root *quota_root;
1049         u64 ref_root;
1050         struct btrfs_qgroup *qgroup;
1051         struct ulist_node *unode;
1052         struct ulist *roots = NULL;
1053         struct ulist *tmp = NULL;
1054         struct ulist_iterator uiter;
1055         u64 seq;
1056         int ret = 0;
1057         int sgn;
1058
1059         if (!fs_info->quota_enabled)
1060                 return 0;
1061
1062         BUG_ON(!fs_info->quota_root);
1063
1064         ins.objectid = node->bytenr;
1065         ins.offset = node->num_bytes;
1066         ins.type = BTRFS_EXTENT_ITEM_KEY;
1067
1068         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1069             node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1070                 struct btrfs_delayed_tree_ref *ref;
1071                 ref = btrfs_delayed_node_to_tree_ref(node);
1072                 ref_root = ref->root;
1073         } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1074                    node->type == BTRFS_SHARED_DATA_REF_KEY) {
1075                 struct btrfs_delayed_data_ref *ref;
1076                 ref = btrfs_delayed_node_to_data_ref(node);
1077                 ref_root = ref->root;
1078         } else {
1079                 BUG();
1080         }
1081
1082         if (!is_fstree(ref_root)) {
1083                 /*
1084                  * non-fs-trees are not being accounted
1085                  */
1086                 return 0;
1087         }
1088
1089         switch (node->action) {
1090         case BTRFS_ADD_DELAYED_REF:
1091         case BTRFS_ADD_DELAYED_EXTENT:
1092                 sgn = 1;
1093                 break;
1094         case BTRFS_DROP_DELAYED_REF:
1095                 sgn = -1;
1096                 break;
1097         case BTRFS_UPDATE_DELAYED_HEAD:
1098                 return 0;
1099         default:
1100                 BUG();
1101         }
1102
1103         /*
1104          * the delayed ref sequence number we pass depends on the direction of
1105          * the operation. for add operations, we pass (node->seq - 1) to skip
1106          * the delayed ref's current sequence number, because we need the state
1107          * of the tree before the add operation. for delete operations, we pass
1108          * (node->seq) to include the delayed ref's current sequence number,
1109          * because we need the state of the tree after the delete operation.
1110          */
1111         ret = btrfs_find_all_roots(trans, fs_info, node->bytenr,
1112                                    sgn > 0 ? node->seq - 1 : node->seq, &roots);
1113         if (ret < 0)
1114                 goto out;
1115
1116         spin_lock(&fs_info->qgroup_lock);
1117         quota_root = fs_info->quota_root;
1118         if (!quota_root)
1119                 goto unlock;
1120
1121         qgroup = find_qgroup_rb(fs_info, ref_root);
1122         if (!qgroup)
1123                 goto unlock;
1124
1125         /*
1126          * step 1: for each old ref, visit all nodes once and inc refcnt
1127          */
1128         tmp = ulist_alloc(GFP_ATOMIC);
1129         if (!tmp) {
1130                 ret = -ENOMEM;
1131                 goto unlock;
1132         }
1133         seq = fs_info->qgroup_seq;
1134         fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1135
1136         ULIST_ITER_INIT(&uiter);
1137         while ((unode = ulist_next(roots, &uiter))) {
1138                 struct ulist_node *tmp_unode;
1139                 struct ulist_iterator tmp_uiter;
1140                 struct btrfs_qgroup *qg;
1141
1142                 qg = find_qgroup_rb(fs_info, unode->val);
1143                 if (!qg)
1144                         continue;
1145
1146                 ulist_reinit(tmp);
1147                                                 /* XXX id not needed */
1148                 ulist_add(tmp, qg->qgroupid, (unsigned long)qg, GFP_ATOMIC);
1149                 ULIST_ITER_INIT(&tmp_uiter);
1150                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1151                         struct btrfs_qgroup_list *glist;
1152
1153                         qg = (struct btrfs_qgroup *)tmp_unode->aux;
1154                         if (qg->refcnt < seq)
1155                                 qg->refcnt = seq + 1;
1156                         else
1157                                 ++qg->refcnt;
1158
1159                         list_for_each_entry(glist, &qg->groups, next_group) {
1160                                 ulist_add(tmp, glist->group->qgroupid,
1161                                           (unsigned long)glist->group,
1162                                           GFP_ATOMIC);
1163                         }
1164                 }
1165         }
1166
1167         /*
1168          * step 2: walk from the new root
1169          */
1170         ulist_reinit(tmp);
1171         ulist_add(tmp, qgroup->qgroupid, (unsigned long)qgroup, GFP_ATOMIC);
1172         ULIST_ITER_INIT(&uiter);
1173         while ((unode = ulist_next(tmp, &uiter))) {
1174                 struct btrfs_qgroup *qg;
1175                 struct btrfs_qgroup_list *glist;
1176
1177                 qg = (struct btrfs_qgroup *)unode->aux;
1178                 if (qg->refcnt < seq) {
1179                         /* not visited by step 1 */
1180                         qg->rfer += sgn * node->num_bytes;
1181                         qg->rfer_cmpr += sgn * node->num_bytes;
1182                         if (roots->nnodes == 0) {
1183                                 qg->excl += sgn * node->num_bytes;
1184                                 qg->excl_cmpr += sgn * node->num_bytes;
1185                         }
1186                         qgroup_dirty(fs_info, qg);
1187                 }
1188                 WARN_ON(qg->tag >= seq);
1189                 qg->tag = seq;
1190
1191                 list_for_each_entry(glist, &qg->groups, next_group) {
1192                         ulist_add(tmp, glist->group->qgroupid,
1193                                   (unsigned long)glist->group, GFP_ATOMIC);
1194                 }
1195         }
1196
1197         /*
1198          * step 3: walk again from old refs
1199          */
1200         ULIST_ITER_INIT(&uiter);
1201         while ((unode = ulist_next(roots, &uiter))) {
1202                 struct btrfs_qgroup *qg;
1203                 struct ulist_node *tmp_unode;
1204                 struct ulist_iterator tmp_uiter;
1205
1206                 qg = find_qgroup_rb(fs_info, unode->val);
1207                 if (!qg)
1208                         continue;
1209
1210                 ulist_reinit(tmp);
1211                 ulist_add(tmp, qg->qgroupid, (unsigned long)qg, GFP_ATOMIC);
1212                 ULIST_ITER_INIT(&tmp_uiter);
1213                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1214                         struct btrfs_qgroup_list *glist;
1215
1216                         qg = (struct btrfs_qgroup *)tmp_unode->aux;
1217                         if (qg->tag == seq)
1218                                 continue;
1219
1220                         if (qg->refcnt - seq == roots->nnodes) {
1221                                 qg->excl -= sgn * node->num_bytes;
1222                                 qg->excl_cmpr -= sgn * node->num_bytes;
1223                                 qgroup_dirty(fs_info, qg);
1224                         }
1225
1226                         list_for_each_entry(glist, &qg->groups, next_group) {
1227                                 ulist_add(tmp, glist->group->qgroupid,
1228                                           (unsigned long)glist->group,
1229                                           GFP_ATOMIC);
1230                         }
1231                 }
1232         }
1233         ret = 0;
1234 unlock:
1235         spin_unlock(&fs_info->qgroup_lock);
1236 out:
1237         ulist_free(roots);
1238         ulist_free(tmp);
1239
1240         return ret;
1241 }
1242
1243 /*
1244  * called from commit_transaction. Writes all changed qgroups to disk.
1245  */
1246 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1247                       struct btrfs_fs_info *fs_info)
1248 {
1249         struct btrfs_root *quota_root = fs_info->quota_root;
1250         int ret = 0;
1251
1252         if (!quota_root)
1253                 goto out;
1254
1255         fs_info->quota_enabled = fs_info->pending_quota_state;
1256
1257         spin_lock(&fs_info->qgroup_lock);
1258         while (!list_empty(&fs_info->dirty_qgroups)) {
1259                 struct btrfs_qgroup *qgroup;
1260                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1261                                           struct btrfs_qgroup, dirty);
1262                 list_del_init(&qgroup->dirty);
1263                 spin_unlock(&fs_info->qgroup_lock);
1264                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1265                 if (ret)
1266                         fs_info->qgroup_flags |=
1267                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1268                 spin_lock(&fs_info->qgroup_lock);
1269         }
1270         if (fs_info->quota_enabled)
1271                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1272         else
1273                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1274         spin_unlock(&fs_info->qgroup_lock);
1275
1276         ret = update_qgroup_status_item(trans, fs_info, quota_root);
1277         if (ret)
1278                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1279
1280 out:
1281
1282         return ret;
1283 }
1284
1285 /*
1286  * copy the acounting information between qgroups. This is necessary when a
1287  * snapshot or a subvolume is created
1288  */
1289 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1290                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1291                          struct btrfs_qgroup_inherit *inherit)
1292 {
1293         int ret = 0;
1294         int i;
1295         u64 *i_qgroups;
1296         struct btrfs_root *quota_root = fs_info->quota_root;
1297         struct btrfs_qgroup *srcgroup;
1298         struct btrfs_qgroup *dstgroup;
1299         u32 level_size = 0;
1300
1301         if (!fs_info->quota_enabled)
1302                 return 0;
1303
1304         if (!quota_root)
1305                 return -EINVAL;
1306
1307         /*
1308          * create a tracking group for the subvol itself
1309          */
1310         ret = add_qgroup_item(trans, quota_root, objectid);
1311         if (ret)
1312                 goto out;
1313
1314         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1315                 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1316                                                inherit->lim.flags,
1317                                                inherit->lim.max_rfer,
1318                                                inherit->lim.max_excl,
1319                                                inherit->lim.rsv_rfer,
1320                                                inherit->lim.rsv_excl);
1321                 if (ret)
1322                         goto out;
1323         }
1324
1325         if (srcid) {
1326                 struct btrfs_root *srcroot;
1327                 struct btrfs_key srckey;
1328                 int srcroot_level;
1329
1330                 srckey.objectid = srcid;
1331                 srckey.type = BTRFS_ROOT_ITEM_KEY;
1332                 srckey.offset = (u64)-1;
1333                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1334                 if (IS_ERR(srcroot)) {
1335                         ret = PTR_ERR(srcroot);
1336                         goto out;
1337                 }
1338
1339                 rcu_read_lock();
1340                 srcroot_level = btrfs_header_level(srcroot->node);
1341                 level_size = btrfs_level_size(srcroot, srcroot_level);
1342                 rcu_read_unlock();
1343         }
1344
1345         /*
1346          * add qgroup to all inherited groups
1347          */
1348         if (inherit) {
1349                 i_qgroups = (u64 *)(inherit + 1);
1350                 for (i = 0; i < inherit->num_qgroups; ++i) {
1351                         ret = add_qgroup_relation_item(trans, quota_root,
1352                                                        objectid, *i_qgroups);
1353                         if (ret)
1354                                 goto out;
1355                         ret = add_qgroup_relation_item(trans, quota_root,
1356                                                        *i_qgroups, objectid);
1357                         if (ret)
1358                                 goto out;
1359                         ++i_qgroups;
1360                 }
1361         }
1362
1363
1364         spin_lock(&fs_info->qgroup_lock);
1365
1366         dstgroup = add_qgroup_rb(fs_info, objectid);
1367         if (!dstgroup)
1368                 goto unlock;
1369
1370         if (srcid) {
1371                 srcgroup = find_qgroup_rb(fs_info, srcid);
1372                 if (!srcgroup)
1373                         goto unlock;
1374                 dstgroup->rfer = srcgroup->rfer - level_size;
1375                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1376                 srcgroup->excl = level_size;
1377                 srcgroup->excl_cmpr = level_size;
1378                 qgroup_dirty(fs_info, dstgroup);
1379                 qgroup_dirty(fs_info, srcgroup);
1380         }
1381
1382         if (!inherit)
1383                 goto unlock;
1384
1385         i_qgroups = (u64 *)(inherit + 1);
1386         for (i = 0; i < inherit->num_qgroups; ++i) {
1387                 ret = add_relation_rb(quota_root->fs_info, objectid,
1388                                       *i_qgroups);
1389                 if (ret)
1390                         goto unlock;
1391                 ++i_qgroups;
1392         }
1393
1394         for (i = 0; i <  inherit->num_ref_copies; ++i) {
1395                 struct btrfs_qgroup *src;
1396                 struct btrfs_qgroup *dst;
1397
1398                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1399                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1400
1401                 if (!src || !dst) {
1402                         ret = -EINVAL;
1403                         goto unlock;
1404                 }
1405
1406                 dst->rfer = src->rfer - level_size;
1407                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1408                 i_qgroups += 2;
1409         }
1410         for (i = 0; i <  inherit->num_excl_copies; ++i) {
1411                 struct btrfs_qgroup *src;
1412                 struct btrfs_qgroup *dst;
1413
1414                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1415                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1416
1417                 if (!src || !dst) {
1418                         ret = -EINVAL;
1419                         goto unlock;
1420                 }
1421
1422                 dst->excl = src->excl + level_size;
1423                 dst->excl_cmpr = src->excl_cmpr + level_size;
1424                 i_qgroups += 2;
1425         }
1426
1427 unlock:
1428         spin_unlock(&fs_info->qgroup_lock);
1429 out:
1430         return ret;
1431 }
1432
1433 /*
1434  * reserve some space for a qgroup and all its parents. The reservation takes
1435  * place with start_transaction or dealloc_reserve, similar to ENOSPC
1436  * accounting. If not enough space is available, EDQUOT is returned.
1437  * We assume that the requested space is new for all qgroups.
1438  */
1439 int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1440 {
1441         struct btrfs_root *quota_root;
1442         struct btrfs_qgroup *qgroup;
1443         struct btrfs_fs_info *fs_info = root->fs_info;
1444         u64 ref_root = root->root_key.objectid;
1445         int ret = 0;
1446         struct ulist *ulist = NULL;
1447         struct ulist_node *unode;
1448         struct ulist_iterator uiter;
1449
1450         if (!is_fstree(ref_root))
1451                 return 0;
1452
1453         if (num_bytes == 0)
1454                 return 0;
1455
1456         spin_lock(&fs_info->qgroup_lock);
1457         quota_root = fs_info->quota_root;
1458         if (!quota_root)
1459                 goto out;
1460
1461         qgroup = find_qgroup_rb(fs_info, ref_root);
1462         if (!qgroup)
1463                 goto out;
1464
1465         /*
1466          * in a first step, we check all affected qgroups if any limits would
1467          * be exceeded
1468          */
1469         ulist = ulist_alloc(GFP_ATOMIC);
1470         ulist_add(ulist, qgroup->qgroupid, (unsigned long)qgroup, GFP_ATOMIC);
1471         ULIST_ITER_INIT(&uiter);
1472         while ((unode = ulist_next(ulist, &uiter))) {
1473                 struct btrfs_qgroup *qg;
1474                 struct btrfs_qgroup_list *glist;
1475
1476                 qg = (struct btrfs_qgroup *)unode->aux;
1477
1478                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1479                     qg->reserved + qg->rfer + num_bytes >
1480                     qg->max_rfer)
1481                         ret = -EDQUOT;
1482
1483                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1484                     qg->reserved + qg->excl + num_bytes >
1485                     qg->max_excl)
1486                         ret = -EDQUOT;
1487
1488                 list_for_each_entry(glist, &qg->groups, next_group) {
1489                         ulist_add(ulist, glist->group->qgroupid,
1490                                   (unsigned long)glist->group, GFP_ATOMIC);
1491                 }
1492         }
1493         if (ret)
1494                 goto out;
1495
1496         /*
1497          * no limits exceeded, now record the reservation into all qgroups
1498          */
1499         ULIST_ITER_INIT(&uiter);
1500         while ((unode = ulist_next(ulist, &uiter))) {
1501                 struct btrfs_qgroup *qg;
1502
1503                 qg = (struct btrfs_qgroup *)unode->aux;
1504
1505                 qg->reserved += num_bytes;
1506         }
1507
1508 out:
1509         spin_unlock(&fs_info->qgroup_lock);
1510         ulist_free(ulist);
1511
1512         return ret;
1513 }
1514
1515 void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1516 {
1517         struct btrfs_root *quota_root;
1518         struct btrfs_qgroup *qgroup;
1519         struct btrfs_fs_info *fs_info = root->fs_info;
1520         struct ulist *ulist = NULL;
1521         struct ulist_node *unode;
1522         struct ulist_iterator uiter;
1523         u64 ref_root = root->root_key.objectid;
1524
1525         if (!is_fstree(ref_root))
1526                 return;
1527
1528         if (num_bytes == 0)
1529                 return;
1530
1531         spin_lock(&fs_info->qgroup_lock);
1532
1533         quota_root = fs_info->quota_root;
1534         if (!quota_root)
1535                 goto out;
1536
1537         qgroup = find_qgroup_rb(fs_info, ref_root);
1538         if (!qgroup)
1539                 goto out;
1540
1541         ulist = ulist_alloc(GFP_ATOMIC);
1542         ulist_add(ulist, qgroup->qgroupid, (unsigned long)qgroup, GFP_ATOMIC);
1543         ULIST_ITER_INIT(&uiter);
1544         while ((unode = ulist_next(ulist, &uiter))) {
1545                 struct btrfs_qgroup *qg;
1546                 struct btrfs_qgroup_list *glist;
1547
1548                 qg = (struct btrfs_qgroup *)unode->aux;
1549
1550                 qg->reserved -= num_bytes;
1551
1552                 list_for_each_entry(glist, &qg->groups, next_group) {
1553                         ulist_add(ulist, glist->group->qgroupid,
1554                                   (unsigned long)glist->group, GFP_ATOMIC);
1555                 }
1556         }
1557
1558 out:
1559         spin_unlock(&fs_info->qgroup_lock);
1560         ulist_free(ulist);
1561 }
1562
1563 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1564 {
1565         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1566                 return;
1567         printk(KERN_ERR "btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1568                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1569                 trans->delayed_ref_elem.seq);
1570         BUG();
1571 }