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 == -ENOENT) {
 383                         printk(KERN_WARNING
 384                                 "btrfs: orphan qgroup relation 0x%llx->0x%llx\n",
 385                                 (unsigned long long)found_key.objectid,
 386                                 (unsigned long long)found_key.offset);
 387                         ret = 0;        /* ignore the error */
 388                 }
 389                 if (ret)
 390                         goto out;
 391 next2:
 392                 ret = btrfs_next_item(quota_root, path);
 393                 if (ret < 0)
 394                         goto out;
 395                 if (ret)
 396                         break;
 397         }
 398 out:
 399         fs_info->qgroup_flags |= flags;
 400         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
 401                 fs_info->quota_enabled = 0;
 402                 fs_info->pending_quota_state = 0;
 403         }
 404         btrfs_free_path(path);
 405
 406         return ret < 0 ? ret : 0;
 407 }
 408
 409 /*
 410  * This is only called from close_ctree() or open_ctree(), both in single-
 411  * treaded paths. Clean up the in-memory structures. No locking needed.
 412  */
 413 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 414 {
 415         struct rb_node *n;
 416         struct btrfs_qgroup *qgroup;
 417         struct btrfs_qgroup_list *list;
 418
 419         while ((n = rb_first(&fs_info->qgroup_tree))) {
 420                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 421                 rb_erase(n, &fs_info->qgroup_tree);
 422
 423                 WARN_ON(!list_empty(&qgroup->dirty));
 424
 425                 while (!list_empty(&qgroup->groups)) {
 426                         list = list_first_entry(&qgroup->groups,
 427                                                 struct btrfs_qgroup_list,
 428                                                 next_group);
 429                         list_del(&list->next_group);
 430                         list_del(&list->next_member);
 431                         kfree(list);
 432                 }
 433
 434                 while (!list_empty(&qgroup->members)) {
 435                         list = list_first_entry(&qgroup->members,
 436                                                 struct btrfs_qgroup_list,
 437                                                 next_member);
 438                         list_del(&list->next_group);
 439                         list_del(&list->next_member);
 440                         kfree(list);
 441                 }
 442                 kfree(qgroup);
 443         }
 444 }
 445
 446 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
 447                                     struct btrfs_root *quota_root,
 448                                     u64 src, u64 dst)
 449 {
 450         int ret;
 451         struct btrfs_path *path;
 452         struct btrfs_key key;
 453
 454         path = btrfs_alloc_path();
 455         if (!path)
 456                 return -ENOMEM;
 457
 458         key.objectid = src;
 459         key.type = BTRFS_QGROUP_RELATION_KEY;
 460         key.offset = dst;
 461
 462         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 463
 464         btrfs_mark_buffer_dirty(path->nodes[0]);
 465
 466         btrfs_free_path(path);
 467         return ret;
 468 }
 469
 470 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
 471                                     struct btrfs_root *quota_root,
 472                                     u64 src, u64 dst)
 473 {
 474         int ret;
 475         struct btrfs_path *path;
 476         struct btrfs_key key;
 477
 478         path = btrfs_alloc_path();
 479         if (!path)
 480                 return -ENOMEM;
 481
 482         key.objectid = src;
 483         key.type = BTRFS_QGROUP_RELATION_KEY;
 484         key.offset = dst;
 485
 486         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 487         if (ret < 0)
 488                 goto out;
 489
 490         if (ret > 0) {
 491                 ret = -ENOENT;
 492                 goto out;
 493         }
 494
 495         ret = btrfs_del_item(trans, quota_root, path);
 496 out:
 497         btrfs_free_path(path);
 498         return ret;
 499 }
 500
 501 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 502                            struct btrfs_root *quota_root, u64 qgroupid)
 503 {
 504         int ret;
 505         struct btrfs_path *path;
 506         struct btrfs_qgroup_info_item *qgroup_info;
 507         struct btrfs_qgroup_limit_item *qgroup_limit;
 508         struct extent_buffer *leaf;
 509         struct btrfs_key key;
 510
 511         path = btrfs_alloc_path();
 512         if (!path)
 513                 return -ENOMEM;
 514
 515         key.objectid = 0;
 516         key.type = BTRFS_QGROUP_INFO_KEY;
 517         key.offset = qgroupid;
 518
 519         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 520                                       sizeof(*qgroup_info));
 521         if (ret)
 522                 goto out;
 523
 524         leaf = path->nodes[0];
 525         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 526                                  struct btrfs_qgroup_info_item);
 527         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 528         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 529         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 530         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 531         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 532
 533         btrfs_mark_buffer_dirty(leaf);
 534
 535         btrfs_release_path(path);
 536
 537         key.type = BTRFS_QGROUP_LIMIT_KEY;
 538         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 539                                       sizeof(*qgroup_limit));
 540         if (ret)
 541                 goto out;
 542
 543         leaf = path->nodes[0];
 544         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 545                                   struct btrfs_qgroup_limit_item);
 546         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 547         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 548         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 549         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 550         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 551
 552         btrfs_mark_buffer_dirty(leaf);
 553
 554         ret = 0;
 555 out:
 556         btrfs_free_path(path);
 557         return ret;
 558 }
 559
 560 static int del_qgroup_item(struct btrfs_trans_handle *trans,
 561                            struct btrfs_root *quota_root, u64 qgroupid)
 562 {
 563         int ret;
 564         struct btrfs_path *path;
 565         struct btrfs_key key;
 566
 567         path = btrfs_alloc_path();
 568         if (!path)
 569                 return -ENOMEM;
 570
 571         key.objectid = 0;
 572         key.type = BTRFS_QGROUP_INFO_KEY;
 573         key.offset = qgroupid;
 574         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 575         if (ret < 0)
 576                 goto out;
 577
 578         if (ret > 0) {
 579                 ret = -ENOENT;
 580                 goto out;
 581         }
 582
 583         ret = btrfs_del_item(trans, quota_root, path);
 584         if (ret)
 585                 goto out;
 586
 587         btrfs_release_path(path);
 588
 589         key.type = BTRFS_QGROUP_LIMIT_KEY;
 590         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 591         if (ret < 0)
 592                 goto out;
 593
 594         if (ret > 0) {
 595                 ret = -ENOENT;
 596                 goto out;
 597         }
 598
 599         ret = btrfs_del_item(trans, quota_root, path);
 600
 601 out:
 602         btrfs_free_path(path);
 603         return ret;
 604 }
 605
 606 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 607                                     struct btrfs_root *root, u64 qgroupid,
 608                                     u64 flags, u64 max_rfer, u64 max_excl,
 609                                     u64 rsv_rfer, u64 rsv_excl)
 610 {
 611         struct btrfs_path *path;
 612         struct btrfs_key key;
 613         struct extent_buffer *l;
 614         struct btrfs_qgroup_limit_item *qgroup_limit;
 615         int ret;
 616         int slot;
 617
 618         key.objectid = 0;
 619         key.type = BTRFS_QGROUP_LIMIT_KEY;
 620         key.offset = qgroupid;
 621
 622         path = btrfs_alloc_path();
 623         BUG_ON(!path);
 624         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 625         if (ret > 0)
 626                 ret = -ENOENT;
 627
 628         if (ret)
 629                 goto out;
 630
 631         l = path->nodes[0];
 632         slot = path->slots[0];
 633         qgroup_limit = btrfs_item_ptr(l, path->slots[0],
 634                                       struct btrfs_qgroup_limit_item);
 635         btrfs_set_qgroup_limit_flags(l, qgroup_limit, flags);
 636         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, max_rfer);
 637         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, max_excl);
 638         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, rsv_rfer);
 639         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, rsv_excl);
 640
 641         btrfs_mark_buffer_dirty(l);
 642
 643 out:
 644         btrfs_free_path(path);
 645         return ret;
 646 }
 647
 648 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 649                                    struct btrfs_root *root,
 650                                    struct btrfs_qgroup *qgroup)
 651 {
 652         struct btrfs_path *path;
 653         struct btrfs_key key;
 654         struct extent_buffer *l;
 655         struct btrfs_qgroup_info_item *qgroup_info;
 656         int ret;
 657         int slot;
 658
 659         key.objectid = 0;
 660         key.type = BTRFS_QGROUP_INFO_KEY;
 661         key.offset = qgroup->qgroupid;
 662
 663         path = btrfs_alloc_path();
 664         BUG_ON(!path);
 665         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 666         if (ret > 0)
 667                 ret = -ENOENT;
 668
 669         if (ret)
 670                 goto out;
 671
 672         l = path->nodes[0];
 673         slot = path->slots[0];
 674         qgroup_info = btrfs_item_ptr(l, path->slots[0],
 675                                  struct btrfs_qgroup_info_item);
 676         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 677         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 678         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 679         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 680         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 681
 682         btrfs_mark_buffer_dirty(l);
 683
 684 out:
 685         btrfs_free_path(path);
 686         return ret;
 687 }
 688
 689 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 690                                      struct btrfs_fs_info *fs_info,
 691                                     struct btrfs_root *root)
 692 {
 693         struct btrfs_path *path;
 694         struct btrfs_key key;
 695         struct extent_buffer *l;
 696         struct btrfs_qgroup_status_item *ptr;
 697         int ret;
 698         int slot;
 699
 700         key.objectid = 0;
 701         key.type = BTRFS_QGROUP_STATUS_KEY;
 702         key.offset = 0;
 703
 704         path = btrfs_alloc_path();
 705         BUG_ON(!path);
 706         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 707         if (ret > 0)
 708                 ret = -ENOENT;
 709
 710         if (ret)
 711                 goto out;
 712
 713         l = path->nodes[0];
 714         slot = path->slots[0];
 715         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 716         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 717         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 718         /* XXX scan */
 719
 720         btrfs_mark_buffer_dirty(l);
 721
 722 out:
 723         btrfs_free_path(path);
 724         return ret;
 725 }
 726
 727 /*
 728  * called with qgroup_lock held
 729  */
 730 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 731                                   struct btrfs_root *root)
 732 {
 733         struct btrfs_path *path;
 734         struct btrfs_key key;
 735         int ret;
 736
 737         if (!root)
 738                 return -EINVAL;
 739
 740         path = btrfs_alloc_path();
 741         if (!path)
 742                 return -ENOMEM;
 743
 744         while (1) {
 745                 key.objectid = 0;
 746                 key.offset = 0;
 747                 key.type = 0;
 748
 749                 path->leave_spinning = 1;
 750                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 751                 if (ret > 0) {
 752                         if (path->slots[0] == 0)
 753                                 break;
 754                         path->slots[0]--;
 755                 } else if (ret < 0) {
 756                         break;
 757                 }
 758
 759                 ret = btrfs_del_item(trans, root, path);
 760                 if (ret)
 761                         goto out;
 762                 btrfs_release_path(path);
 763         }
 764         ret = 0;
 765 out:
 766         root->fs_info->pending_quota_state = 0;
 767         btrfs_free_path(path);
 768         return ret;
 769 }
 770
 771 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 772                        struct btrfs_fs_info *fs_info)
 773 {
 774         struct btrfs_root *quota_root;
 775         struct btrfs_path *path = NULL;
 776         struct btrfs_qgroup_status_item *ptr;
 777         struct extent_buffer *leaf;
 778         struct btrfs_key key;
 779         int ret = 0;
 780
 781         spin_lock(&fs_info->qgroup_lock);
 782         if (fs_info->quota_root) {
 783                 fs_info->pending_quota_state = 1;
 784                 spin_unlock(&fs_info->qgroup_lock);
 785                 goto out;
 786         }
 787         spin_unlock(&fs_info->qgroup_lock);
 788
 789         /*
 790          * initially create the quota tree
 791          */
 792         quota_root = btrfs_create_tree(trans, fs_info,
 793                                        BTRFS_QUOTA_TREE_OBJECTID);
 794         if (IS_ERR(quota_root)) {
 795                 ret =  PTR_ERR(quota_root);
 796                 goto out;
 797         }
 798
 799         path = btrfs_alloc_path();
 800         if (!path) {
 801                 ret = -ENOMEM;
 802                 goto out_free_root;
 803         }
 804
 805         key.objectid = 0;
 806         key.type = BTRFS_QGROUP_STATUS_KEY;
 807         key.offset = 0;
 808
 809         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 810                                       sizeof(*ptr));
 811         if (ret)
 812                 goto out_free_path;
 813
 814         leaf = path->nodes[0];
 815         ptr = btrfs_item_ptr(leaf, path->slots[0],
 816                                  struct btrfs_qgroup_status_item);
 817         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 818         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 819         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 820                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 821         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 822         btrfs_set_qgroup_status_scan(leaf, ptr, 0);
 823
 824         btrfs_mark_buffer_dirty(leaf);
 825
 826         spin_lock(&fs_info->qgroup_lock);
 827         fs_info->quota_root = quota_root;
 828         fs_info->pending_quota_state = 1;
 829         spin_unlock(&fs_info->qgroup_lock);
 830 out_free_path:
 831         btrfs_free_path(path);
 832 out_free_root:
 833         if (ret) {
 834                 free_extent_buffer(quota_root->node);
 835                 free_extent_buffer(quota_root->commit_root);
 836                 kfree(quota_root);
 837         }
 838 out:
 839         return ret;
 840 }
 841
 842 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
 843                         struct btrfs_fs_info *fs_info)
 844 {
 845         struct btrfs_root *tree_root = fs_info->tree_root;
 846         struct btrfs_root *quota_root;
 847         int ret = 0;
 848
 849         spin_lock(&fs_info->qgroup_lock);
 850         fs_info->quota_enabled = 0;
 851         fs_info->pending_quota_state = 0;
 852         quota_root = fs_info->quota_root;
 853         fs_info->quota_root = NULL;
 854         btrfs_free_qgroup_config(fs_info);
 855         spin_unlock(&fs_info->qgroup_lock);
 856
 857         if (!quota_root)
 858                 return -EINVAL;
 859
 860         ret = btrfs_clean_quota_tree(trans, quota_root);
 861         if (ret)
 862                 goto out;
 863
 864         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
 865         if (ret)
 866                 goto out;
 867
 868         list_del(&quota_root->dirty_list);
 869
 870         btrfs_tree_lock(quota_root->node);
 871         clean_tree_block(trans, tree_root, quota_root->node);
 872         btrfs_tree_unlock(quota_root->node);
 873         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
 874
 875         free_extent_buffer(quota_root->node);
 876         free_extent_buffer(quota_root->commit_root);
 877         kfree(quota_root);
 878 out:
 879         return ret;
 880 }
 881
 882 int btrfs_quota_rescan(struct btrfs_fs_info *fs_info)
 883 {
 884         /* FIXME */
 885         return 0;
 886 }
 887
 888 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
 889                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
 890 {
 891         struct btrfs_root *quota_root;
 892         int ret = 0;
 893
 894         quota_root = fs_info->quota_root;
 895         if (!quota_root)
 896                 return -EINVAL;
 897
 898         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
 899         if (ret)
 900                 return ret;
 901
 902         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
 903         if (ret) {
 904                 del_qgroup_relation_item(trans, quota_root, src, dst);
 905                 return ret;
 906         }
 907
 908         spin_lock(&fs_info->qgroup_lock);
 909         ret = add_relation_rb(quota_root->fs_info, src, dst);
 910         spin_unlock(&fs_info->qgroup_lock);
 911
 912         return ret;
 913 }
 914
 915 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
 916                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
 917 {
 918         struct btrfs_root *quota_root;
 919         int ret = 0;
 920         int err;
 921
 922         quota_root = fs_info->quota_root;
 923         if (!quota_root)
 924                 return -EINVAL;
 925
 926         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
 927         err = del_qgroup_relation_item(trans, quota_root, dst, src);
 928         if (err && !ret)
 929                 ret = err;
 930
 931         spin_lock(&fs_info->qgroup_lock);
 932         del_relation_rb(fs_info, src, dst);
 933
 934         spin_unlock(&fs_info->qgroup_lock);
 935
 936         return ret;
 937 }
 938
 939 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
 940                         struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
 941 {
 942         struct btrfs_root *quota_root;
 943         struct btrfs_qgroup *qgroup;
 944         int ret = 0;
 945
 946         quota_root = fs_info->quota_root;
 947         if (!quota_root)
 948                 return -EINVAL;
 949
 950         ret = add_qgroup_item(trans, quota_root, qgroupid);
 951
 952         spin_lock(&fs_info->qgroup_lock);
 953         qgroup = add_qgroup_rb(fs_info, qgroupid);
 954         spin_unlock(&fs_info->qgroup_lock);
 955
 956         if (IS_ERR(qgroup))
 957                 ret = PTR_ERR(qgroup);
 958
 959         return ret;
 960 }
 961
 962 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
 963                         struct btrfs_fs_info *fs_info, u64 qgroupid)
 964 {
 965         struct btrfs_root *quota_root;
 966         struct btrfs_qgroup *qgroup;
 967         int ret = 0;
 968
 969         quota_root = fs_info->quota_root;
 970         if (!quota_root)
 971                 return -EINVAL;
 972
 973         /* check if there are no relations to this qgroup */
 974         spin_lock(&fs_info->qgroup_lock);
 975         qgroup = find_qgroup_rb(fs_info, qgroupid);
 976         if (qgroup) {
 977                 if (!list_empty(&qgroup->groups) || !list_empty(&qgroup->members)) {
 978                         spin_unlock(&fs_info->qgroup_lock);
 979                         return -EBUSY;
 980                 }
 981         }
 982         spin_unlock(&fs_info->qgroup_lock);
 983
 984         ret = del_qgroup_item(trans, quota_root, qgroupid);
 985
 986         spin_lock(&fs_info->qgroup_lock);
 987         del_qgroup_rb(quota_root->fs_info, qgroupid);
 988         spin_unlock(&fs_info->qgroup_lock);
 989
 990         return ret;
 991 }
 992
 993 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
 994                        struct btrfs_fs_info *fs_info, u64 qgroupid,
 995                        struct btrfs_qgroup_limit *limit)
 996 {
 997         struct btrfs_root *quota_root = fs_info->quota_root;
 998         struct btrfs_qgroup *qgroup;
 999         int ret = 0;
1000
1001         if (!quota_root)
1002                 return -EINVAL;
1003
1004         ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1005                                        limit->flags, limit->max_rfer,
1006                                        limit->max_excl, limit->rsv_rfer,
1007                                        limit->rsv_excl);
1008         if (ret) {
1009                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1010                 printk(KERN_INFO "unable to update quota limit for %llu\n",
1011                        (unsigned long long)qgroupid);
1012         }
1013
1014         spin_lock(&fs_info->qgroup_lock);
1015
1016         qgroup = find_qgroup_rb(fs_info, qgroupid);
1017         if (!qgroup) {
1018                 ret = -ENOENT;
1019                 goto unlock;
1020         }
1021         qgroup->lim_flags = limit->flags;
1022         qgroup->max_rfer = limit->max_rfer;
1023         qgroup->max_excl = limit->max_excl;
1024         qgroup->rsv_rfer = limit->rsv_rfer;
1025         qgroup->rsv_excl = limit->rsv_excl;
1026
1027 unlock:
1028         spin_unlock(&fs_info->qgroup_lock);
1029
1030         return ret;
1031 }
1032
1033 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1034                          struct btrfs_qgroup *qgroup)
1035 {
1036         if (list_empty(&qgroup->dirty))
1037                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1038 }
1039
1040 /*
1041  * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1042  * the modification into a list that's later used by btrfs_end_transaction to
1043  * pass the recorded modifications on to btrfs_qgroup_account_ref.
1044  */
1045 int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1046                             struct btrfs_delayed_ref_node *node,
1047                             struct btrfs_delayed_extent_op *extent_op)
1048 {
1049         struct qgroup_update *u;
1050
1051         BUG_ON(!trans->delayed_ref_elem.seq);
1052         u = kmalloc(sizeof(*u), GFP_NOFS);
1053         if (!u)
1054                 return -ENOMEM;
1055
1056         u->node = node;
1057         u->extent_op = extent_op;
1058         list_add_tail(&u->list, &trans->qgroup_ref_list);
1059
1060         return 0;
1061 }
1062
1063 /*
1064  * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1065  * from the fs. First, all roots referencing the extent are searched, and
1066  * then the space is accounted accordingly to the different roots. The
1067  * accounting algorithm works in 3 steps documented inline.
1068  */
1069 int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1070                              struct btrfs_fs_info *fs_info,
1071                              struct btrfs_delayed_ref_node *node,
1072                              struct btrfs_delayed_extent_op *extent_op)
1073 {
1074         struct btrfs_key ins;
1075         struct btrfs_root *quota_root;
1076         u64 ref_root;
1077         struct btrfs_qgroup *qgroup;
1078         struct ulist_node *unode;
1079         struct ulist *roots = NULL;
1080         struct ulist *tmp = NULL;
1081         struct ulist_iterator uiter;
1082         u64 seq;
1083         int ret = 0;
1084         int sgn;
1085
1086         if (!fs_info->quota_enabled)
1087                 return 0;
1088
1089         BUG_ON(!fs_info->quota_root);
1090
1091         ins.objectid = node->bytenr;
1092         ins.offset = node->num_bytes;
1093         ins.type = BTRFS_EXTENT_ITEM_KEY;
1094
1095         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1096             node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1097                 struct btrfs_delayed_tree_ref *ref;
1098                 ref = btrfs_delayed_node_to_tree_ref(node);
1099                 ref_root = ref->root;
1100         } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1101                    node->type == BTRFS_SHARED_DATA_REF_KEY) {
1102                 struct btrfs_delayed_data_ref *ref;
1103                 ref = btrfs_delayed_node_to_data_ref(node);
1104                 ref_root = ref->root;
1105         } else {
1106                 BUG();
1107         }
1108
1109         if (!is_fstree(ref_root)) {
1110                 /*
1111                  * non-fs-trees are not being accounted
1112                  */
1113                 return 0;
1114         }
1115
1116         switch (node->action) {
1117         case BTRFS_ADD_DELAYED_REF:
1118         case BTRFS_ADD_DELAYED_EXTENT:
1119                 sgn = 1;
1120                 break;
1121         case BTRFS_DROP_DELAYED_REF:
1122                 sgn = -1;
1123                 break;
1124         case BTRFS_UPDATE_DELAYED_HEAD:
1125                 return 0;
1126         default:
1127                 BUG();
1128         }
1129
1130         /*
1131          * the delayed ref sequence number we pass depends on the direction of
1132          * the operation. for add operations, we pass (node->seq - 1) to skip
1133          * the delayed ref's current sequence number, because we need the state
1134          * of the tree before the add operation. for delete operations, we pass
1135          * (node->seq) to include the delayed ref's current sequence number,
1136          * because we need the state of the tree after the delete operation.
1137          */
1138         ret = btrfs_find_all_roots(trans, fs_info, node->bytenr,
1139                                    sgn > 0 ? node->seq - 1 : node->seq, &roots);
1140         if (ret < 0)
1141                 goto out;
1142
1143         spin_lock(&fs_info->qgroup_lock);
1144         quota_root = fs_info->quota_root;
1145         if (!quota_root)
1146                 goto unlock;
1147
1148         qgroup = find_qgroup_rb(fs_info, ref_root);
1149         if (!qgroup)
1150                 goto unlock;
1151
1152         /*
1153          * step 1: for each old ref, visit all nodes once and inc refcnt
1154          */
1155         tmp = ulist_alloc(GFP_ATOMIC);
1156         if (!tmp) {
1157                 ret = -ENOMEM;
1158                 goto unlock;
1159         }
1160         seq = fs_info->qgroup_seq;
1161         fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1162
1163         ULIST_ITER_INIT(&uiter);
1164         while ((unode = ulist_next(roots, &uiter))) {
1165                 struct ulist_node *tmp_unode;
1166                 struct ulist_iterator tmp_uiter;
1167                 struct btrfs_qgroup *qg;
1168
1169                 qg = find_qgroup_rb(fs_info, unode->val);
1170                 if (!qg)
1171                         continue;
1172
1173                 ulist_reinit(tmp);
1174                                                 /* XXX id not needed */
1175                 ulist_add(tmp, qg->qgroupid, (u64)(uintptr_t)qg, GFP_ATOMIC);
1176                 ULIST_ITER_INIT(&tmp_uiter);
1177                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1178                         struct btrfs_qgroup_list *glist;
1179
1180                         qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1181                         if (qg->refcnt < seq)
1182                                 qg->refcnt = seq + 1;
1183                         else
1184                                 ++qg->refcnt;
1185
1186                         list_for_each_entry(glist, &qg->groups, next_group) {
1187                                 ulist_add(tmp, glist->group->qgroupid,
1188                                           (u64)(uintptr_t)glist->group,
1189                                           GFP_ATOMIC);
1190                         }
1191                 }
1192         }
1193
1194         /*
1195          * step 2: walk from the new root
1196          */
1197         ulist_reinit(tmp);
1198         ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
1199         ULIST_ITER_INIT(&uiter);
1200         while ((unode = ulist_next(tmp, &uiter))) {
1201                 struct btrfs_qgroup *qg;
1202                 struct btrfs_qgroup_list *glist;
1203
1204                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1205                 if (qg->refcnt < seq) {
1206                         /* not visited by step 1 */
1207                         qg->rfer += sgn * node->num_bytes;
1208                         qg->rfer_cmpr += sgn * node->num_bytes;
1209                         if (roots->nnodes == 0) {
1210                                 qg->excl += sgn * node->num_bytes;
1211                                 qg->excl_cmpr += sgn * node->num_bytes;
1212                         }
1213                         qgroup_dirty(fs_info, qg);
1214                 }
1215                 WARN_ON(qg->tag >= seq);
1216                 qg->tag = seq;
1217
1218                 list_for_each_entry(glist, &qg->groups, next_group) {
1219                         ulist_add(tmp, glist->group->qgroupid,
1220                                   (uintptr_t)glist->group, GFP_ATOMIC);
1221                 }
1222         }
1223
1224         /*
1225          * step 3: walk again from old refs
1226          */
1227         ULIST_ITER_INIT(&uiter);
1228         while ((unode = ulist_next(roots, &uiter))) {
1229                 struct btrfs_qgroup *qg;
1230                 struct ulist_node *tmp_unode;
1231                 struct ulist_iterator tmp_uiter;
1232
1233                 qg = find_qgroup_rb(fs_info, unode->val);
1234                 if (!qg)
1235                         continue;
1236
1237                 ulist_reinit(tmp);
1238                 ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
1239                 ULIST_ITER_INIT(&tmp_uiter);
1240                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1241                         struct btrfs_qgroup_list *glist;
1242
1243                         qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1244                         if (qg->tag == seq)
1245                                 continue;
1246
1247                         if (qg->refcnt - seq == roots->nnodes) {
1248                                 qg->excl -= sgn * node->num_bytes;
1249                                 qg->excl_cmpr -= sgn * node->num_bytes;
1250                                 qgroup_dirty(fs_info, qg);
1251                         }
1252
1253                         list_for_each_entry(glist, &qg->groups, next_group) {
1254                                 ulist_add(tmp, glist->group->qgroupid,
1255                                           (uintptr_t)glist->group,
1256                                           GFP_ATOMIC);
1257                         }
1258                 }
1259         }
1260         ret = 0;
1261 unlock:
1262         spin_unlock(&fs_info->qgroup_lock);
1263 out:
1264         ulist_free(roots);
1265         ulist_free(tmp);
1266
1267         return ret;
1268 }
1269
1270 /*
1271  * called from commit_transaction. Writes all changed qgroups to disk.
1272  */
1273 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1274                       struct btrfs_fs_info *fs_info)
1275 {
1276         struct btrfs_root *quota_root = fs_info->quota_root;
1277         int ret = 0;
1278
1279         if (!quota_root)
1280                 goto out;
1281
1282         fs_info->quota_enabled = fs_info->pending_quota_state;
1283
1284         spin_lock(&fs_info->qgroup_lock);
1285         while (!list_empty(&fs_info->dirty_qgroups)) {
1286                 struct btrfs_qgroup *qgroup;
1287                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1288                                           struct btrfs_qgroup, dirty);
1289                 list_del_init(&qgroup->dirty);
1290                 spin_unlock(&fs_info->qgroup_lock);
1291                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1292                 if (ret)
1293                         fs_info->qgroup_flags |=
1294                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1295                 spin_lock(&fs_info->qgroup_lock);
1296         }
1297         if (fs_info->quota_enabled)
1298                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1299         else
1300                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1301         spin_unlock(&fs_info->qgroup_lock);
1302
1303         ret = update_qgroup_status_item(trans, fs_info, quota_root);
1304         if (ret)
1305                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1306
1307 out:
1308
1309         return ret;
1310 }
1311
1312 /*
1313  * copy the acounting information between qgroups. This is necessary when a
1314  * snapshot or a subvolume is created
1315  */
1316 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1317                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1318                          struct btrfs_qgroup_inherit *inherit)
1319 {
1320         int ret = 0;
1321         int i;
1322         u64 *i_qgroups;
1323         struct btrfs_root *quota_root = fs_info->quota_root;
1324         struct btrfs_qgroup *srcgroup;
1325         struct btrfs_qgroup *dstgroup;
1326         u32 level_size = 0;
1327
1328         if (!fs_info->quota_enabled)
1329                 return 0;
1330
1331         if (!quota_root)
1332                 return -EINVAL;
1333
1334         /*
1335          * create a tracking group for the subvol itself
1336          */
1337         ret = add_qgroup_item(trans, quota_root, objectid);
1338         if (ret)
1339                 goto out;
1340
1341         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1342                 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1343                                                inherit->lim.flags,
1344                                                inherit->lim.max_rfer,
1345                                                inherit->lim.max_excl,
1346                                                inherit->lim.rsv_rfer,
1347                                                inherit->lim.rsv_excl);
1348                 if (ret)
1349                         goto out;
1350         }
1351
1352         if (srcid) {
1353                 struct btrfs_root *srcroot;
1354                 struct btrfs_key srckey;
1355                 int srcroot_level;
1356
1357                 srckey.objectid = srcid;
1358                 srckey.type = BTRFS_ROOT_ITEM_KEY;
1359                 srckey.offset = (u64)-1;
1360                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1361                 if (IS_ERR(srcroot)) {
1362                         ret = PTR_ERR(srcroot);
1363                         goto out;
1364                 }
1365
1366                 rcu_read_lock();
1367                 srcroot_level = btrfs_header_level(srcroot->node);
1368                 level_size = btrfs_level_size(srcroot, srcroot_level);
1369                 rcu_read_unlock();
1370         }
1371
1372         /*
1373          * add qgroup to all inherited groups
1374          */
1375         if (inherit) {
1376                 i_qgroups = (u64 *)(inherit + 1);
1377                 for (i = 0; i < inherit->num_qgroups; ++i) {
1378                         ret = add_qgroup_relation_item(trans, quota_root,
1379                                                        objectid, *i_qgroups);
1380                         if (ret)
1381                                 goto out;
1382                         ret = add_qgroup_relation_item(trans, quota_root,
1383                                                        *i_qgroups, objectid);
1384                         if (ret)
1385                                 goto out;
1386                         ++i_qgroups;
1387                 }
1388         }
1389
1390
1391         spin_lock(&fs_info->qgroup_lock);
1392
1393         dstgroup = add_qgroup_rb(fs_info, objectid);
1394         if (IS_ERR(dstgroup)) {
1395                 ret = PTR_ERR(dstgroup);
1396                 goto unlock;
1397         }
1398
1399         if (srcid) {
1400                 srcgroup = find_qgroup_rb(fs_info, srcid);
1401                 if (!srcgroup)
1402                         goto unlock;
1403                 dstgroup->rfer = srcgroup->rfer - level_size;
1404                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1405                 srcgroup->excl = level_size;
1406                 srcgroup->excl_cmpr = level_size;
1407                 qgroup_dirty(fs_info, dstgroup);
1408                 qgroup_dirty(fs_info, srcgroup);
1409         }
1410
1411         if (!inherit)
1412                 goto unlock;
1413
1414         i_qgroups = (u64 *)(inherit + 1);
1415         for (i = 0; i < inherit->num_qgroups; ++i) {
1416                 ret = add_relation_rb(quota_root->fs_info, objectid,
1417                                       *i_qgroups);
1418                 if (ret)
1419                         goto unlock;
1420                 ++i_qgroups;
1421         }
1422
1423         for (i = 0; i <  inherit->num_ref_copies; ++i) {
1424                 struct btrfs_qgroup *src;
1425                 struct btrfs_qgroup *dst;
1426
1427                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1428                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1429
1430                 if (!src || !dst) {
1431                         ret = -EINVAL;
1432                         goto unlock;
1433                 }
1434
1435                 dst->rfer = src->rfer - level_size;
1436                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1437                 i_qgroups += 2;
1438         }
1439         for (i = 0; i <  inherit->num_excl_copies; ++i) {
1440                 struct btrfs_qgroup *src;
1441                 struct btrfs_qgroup *dst;
1442
1443                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1444                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1445
1446                 if (!src || !dst) {
1447                         ret = -EINVAL;
1448                         goto unlock;
1449                 }
1450
1451                 dst->excl = src->excl + level_size;
1452                 dst->excl_cmpr = src->excl_cmpr + level_size;
1453                 i_qgroups += 2;
1454         }
1455
1456 unlock:
1457         spin_unlock(&fs_info->qgroup_lock);
1458 out:
1459         return ret;
1460 }
1461
1462 /*
1463  * reserve some space for a qgroup and all its parents. The reservation takes
1464  * place with start_transaction or dealloc_reserve, similar to ENOSPC
1465  * accounting. If not enough space is available, EDQUOT is returned.
1466  * We assume that the requested space is new for all qgroups.
1467  */
1468 int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1469 {
1470         struct btrfs_root *quota_root;
1471         struct btrfs_qgroup *qgroup;
1472         struct btrfs_fs_info *fs_info = root->fs_info;
1473         u64 ref_root = root->root_key.objectid;
1474         int ret = 0;
1475         struct ulist *ulist = NULL;
1476         struct ulist_node *unode;
1477         struct ulist_iterator uiter;
1478
1479         if (!is_fstree(ref_root))
1480                 return 0;
1481
1482         if (num_bytes == 0)
1483                 return 0;
1484
1485         spin_lock(&fs_info->qgroup_lock);
1486         quota_root = fs_info->quota_root;
1487         if (!quota_root)
1488                 goto out;
1489
1490         qgroup = find_qgroup_rb(fs_info, ref_root);
1491         if (!qgroup)
1492                 goto out;
1493
1494         /*
1495          * in a first step, we check all affected qgroups if any limits would
1496          * be exceeded
1497          */
1498         ulist = ulist_alloc(GFP_ATOMIC);
1499         if (!ulist) {
1500                 ret = -ENOMEM;
1501                 goto out;
1502         }
1503         ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
1504         ULIST_ITER_INIT(&uiter);
1505         while ((unode = ulist_next(ulist, &uiter))) {
1506                 struct btrfs_qgroup *qg;
1507                 struct btrfs_qgroup_list *glist;
1508
1509                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1510
1511                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1512                     qg->reserved + qg->rfer + num_bytes >
1513                     qg->max_rfer)
1514                         ret = -EDQUOT;
1515
1516                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1517                     qg->reserved + qg->excl + num_bytes >
1518                     qg->max_excl)
1519                         ret = -EDQUOT;
1520
1521                 list_for_each_entry(glist, &qg->groups, next_group) {
1522                         ulist_add(ulist, glist->group->qgroupid,
1523                                   (uintptr_t)glist->group, GFP_ATOMIC);
1524                 }
1525         }
1526         if (ret)
1527                 goto out;
1528
1529         /*
1530          * no limits exceeded, now record the reservation into all qgroups
1531          */
1532         ULIST_ITER_INIT(&uiter);
1533         while ((unode = ulist_next(ulist, &uiter))) {
1534                 struct btrfs_qgroup *qg;
1535
1536                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1537
1538                 qg->reserved += num_bytes;
1539         }
1540
1541 out:
1542         spin_unlock(&fs_info->qgroup_lock);
1543         ulist_free(ulist);
1544
1545         return ret;
1546 }
1547
1548 void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1549 {
1550         struct btrfs_root *quota_root;
1551         struct btrfs_qgroup *qgroup;
1552         struct btrfs_fs_info *fs_info = root->fs_info;
1553         struct ulist *ulist = NULL;
1554         struct ulist_node *unode;
1555         struct ulist_iterator uiter;
1556         u64 ref_root = root->root_key.objectid;
1557
1558         if (!is_fstree(ref_root))
1559                 return;
1560
1561         if (num_bytes == 0)
1562                 return;
1563
1564         spin_lock(&fs_info->qgroup_lock);
1565
1566         quota_root = fs_info->quota_root;
1567         if (!quota_root)
1568                 goto out;
1569
1570         qgroup = find_qgroup_rb(fs_info, ref_root);
1571         if (!qgroup)
1572                 goto out;
1573
1574         ulist = ulist_alloc(GFP_ATOMIC);
1575         if (!ulist) {
1576                 btrfs_std_error(fs_info, -ENOMEM);
1577                 goto out;
1578         }
1579         ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
1580         ULIST_ITER_INIT(&uiter);
1581         while ((unode = ulist_next(ulist, &uiter))) {
1582                 struct btrfs_qgroup *qg;
1583                 struct btrfs_qgroup_list *glist;
1584
1585                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1586
1587                 qg->reserved -= num_bytes;
1588
1589                 list_for_each_entry(glist, &qg->groups, next_group) {
1590                         ulist_add(ulist, glist->group->qgroupid,
1591                                   (uintptr_t)glist->group, GFP_ATOMIC);
1592                 }
1593         }
1594
1595 out:
1596         spin_unlock(&fs_info->qgroup_lock);
1597         ulist_free(ulist);
1598 }
1599
1600 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1601 {
1602         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1603                 return;
1604         printk(KERN_ERR "btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1605                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1606                 trans->delayed_ref_elem.seq);
1607         BUG();
1608 }