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 #include <linux/btrfs.h>
  27
  28 #include "ctree.h"
  29 #include "transaction.h"
  30 #include "disk-io.h"
  31 #include "locking.h"
  32 #include "ulist.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_ioctl_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                 while (!list_empty(&qgroup->groups)) {
 424                         list = list_first_entry(&qgroup->groups,
 425                                                 struct btrfs_qgroup_list,
 426                                                 next_group);
 427                         list_del(&list->next_group);
 428                         list_del(&list->next_member);
 429                         kfree(list);
 430                 }
 431
 432                 while (!list_empty(&qgroup->members)) {
 433                         list = list_first_entry(&qgroup->members,
 434                                                 struct btrfs_qgroup_list,
 435                                                 next_member);
 436                         list_del(&list->next_group);
 437                         list_del(&list->next_member);
 438                         kfree(list);
 439                 }
 440                 kfree(qgroup);
 441         }
 442 }
 443
 444 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
 445                                     struct btrfs_root *quota_root,
 446                                     u64 src, u64 dst)
 447 {
 448         int ret;
 449         struct btrfs_path *path;
 450         struct btrfs_key key;
 451
 452         path = btrfs_alloc_path();
 453         if (!path)
 454                 return -ENOMEM;
 455
 456         key.objectid = src;
 457         key.type = BTRFS_QGROUP_RELATION_KEY;
 458         key.offset = dst;
 459
 460         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 461
 462         btrfs_mark_buffer_dirty(path->nodes[0]);
 463
 464         btrfs_free_path(path);
 465         return ret;
 466 }
 467
 468 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
 469                                     struct btrfs_root *quota_root,
 470                                     u64 src, u64 dst)
 471 {
 472         int ret;
 473         struct btrfs_path *path;
 474         struct btrfs_key key;
 475
 476         path = btrfs_alloc_path();
 477         if (!path)
 478                 return -ENOMEM;
 479
 480         key.objectid = src;
 481         key.type = BTRFS_QGROUP_RELATION_KEY;
 482         key.offset = dst;
 483
 484         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 485         if (ret < 0)
 486                 goto out;
 487
 488         if (ret > 0) {
 489                 ret = -ENOENT;
 490                 goto out;
 491         }
 492
 493         ret = btrfs_del_item(trans, quota_root, path);
 494 out:
 495         btrfs_free_path(path);
 496         return ret;
 497 }
 498
 499 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 500                            struct btrfs_root *quota_root, u64 qgroupid)
 501 {
 502         int ret;
 503         struct btrfs_path *path;
 504         struct btrfs_qgroup_info_item *qgroup_info;
 505         struct btrfs_qgroup_limit_item *qgroup_limit;
 506         struct extent_buffer *leaf;
 507         struct btrfs_key key;
 508
 509         path = btrfs_alloc_path();
 510         if (!path)
 511                 return -ENOMEM;
 512
 513         key.objectid = 0;
 514         key.type = BTRFS_QGROUP_INFO_KEY;
 515         key.offset = qgroupid;
 516
 517         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 518                                       sizeof(*qgroup_info));
 519         if (ret)
 520                 goto out;
 521
 522         leaf = path->nodes[0];
 523         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 524                                  struct btrfs_qgroup_info_item);
 525         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 526         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 527         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 528         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 529         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 530
 531         btrfs_mark_buffer_dirty(leaf);
 532
 533         btrfs_release_path(path);
 534
 535         key.type = BTRFS_QGROUP_LIMIT_KEY;
 536         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 537                                       sizeof(*qgroup_limit));
 538         if (ret)
 539                 goto out;
 540
 541         leaf = path->nodes[0];
 542         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 543                                   struct btrfs_qgroup_limit_item);
 544         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 545         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 546         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 547         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 548         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 549
 550         btrfs_mark_buffer_dirty(leaf);
 551
 552         ret = 0;
 553 out:
 554         btrfs_free_path(path);
 555         return ret;
 556 }
 557
 558 static int del_qgroup_item(struct btrfs_trans_handle *trans,
 559                            struct btrfs_root *quota_root, u64 qgroupid)
 560 {
 561         int ret;
 562         struct btrfs_path *path;
 563         struct btrfs_key key;
 564
 565         path = btrfs_alloc_path();
 566         if (!path)
 567                 return -ENOMEM;
 568
 569         key.objectid = 0;
 570         key.type = BTRFS_QGROUP_INFO_KEY;
 571         key.offset = qgroupid;
 572         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 573         if (ret < 0)
 574                 goto out;
 575
 576         if (ret > 0) {
 577                 ret = -ENOENT;
 578                 goto out;
 579         }
 580
 581         ret = btrfs_del_item(trans, quota_root, path);
 582         if (ret)
 583                 goto out;
 584
 585         btrfs_release_path(path);
 586
 587         key.type = BTRFS_QGROUP_LIMIT_KEY;
 588         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 589         if (ret < 0)
 590                 goto out;
 591
 592         if (ret > 0) {
 593                 ret = -ENOENT;
 594                 goto out;
 595         }
 596
 597         ret = btrfs_del_item(trans, quota_root, path);
 598
 599 out:
 600         btrfs_free_path(path);
 601         return ret;
 602 }
 603
 604 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 605                                     struct btrfs_root *root, u64 qgroupid,
 606                                     u64 flags, u64 max_rfer, u64 max_excl,
 607                                     u64 rsv_rfer, u64 rsv_excl)
 608 {
 609         struct btrfs_path *path;
 610         struct btrfs_key key;
 611         struct extent_buffer *l;
 612         struct btrfs_qgroup_limit_item *qgroup_limit;
 613         int ret;
 614         int slot;
 615
 616         key.objectid = 0;
 617         key.type = BTRFS_QGROUP_LIMIT_KEY;
 618         key.offset = qgroupid;
 619
 620         path = btrfs_alloc_path();
 621         if (!path)
 622                 return -ENOMEM;
 623
 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         if (!path)
 665                 return -ENOMEM;
 666
 667         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 668         if (ret > 0)
 669                 ret = -ENOENT;
 670
 671         if (ret)
 672                 goto out;
 673
 674         l = path->nodes[0];
 675         slot = path->slots[0];
 676         qgroup_info = btrfs_item_ptr(l, path->slots[0],
 677                                  struct btrfs_qgroup_info_item);
 678         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 679         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 680         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 681         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 682         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 683
 684         btrfs_mark_buffer_dirty(l);
 685
 686 out:
 687         btrfs_free_path(path);
 688         return ret;
 689 }
 690
 691 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 692                                      struct btrfs_fs_info *fs_info,
 693                                     struct btrfs_root *root)
 694 {
 695         struct btrfs_path *path;
 696         struct btrfs_key key;
 697         struct extent_buffer *l;
 698         struct btrfs_qgroup_status_item *ptr;
 699         int ret;
 700         int slot;
 701
 702         key.objectid = 0;
 703         key.type = BTRFS_QGROUP_STATUS_KEY;
 704         key.offset = 0;
 705
 706         path = btrfs_alloc_path();
 707         if (!path)
 708                 return -ENOMEM;
 709
 710         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 711         if (ret > 0)
 712                 ret = -ENOENT;
 713
 714         if (ret)
 715                 goto out;
 716
 717         l = path->nodes[0];
 718         slot = path->slots[0];
 719         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 720         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 721         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 722         /* XXX scan */
 723
 724         btrfs_mark_buffer_dirty(l);
 725
 726 out:
 727         btrfs_free_path(path);
 728         return ret;
 729 }
 730
 731 /*
 732  * called with qgroup_lock held
 733  */
 734 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 735                                   struct btrfs_root *root)
 736 {
 737         struct btrfs_path *path;
 738         struct btrfs_key key;
 739         struct extent_buffer *leaf = NULL;
 740         int ret;
 741         int nr = 0;
 742
 743         path = btrfs_alloc_path();
 744         if (!path)
 745                 return -ENOMEM;
 746
 747         path->leave_spinning = 1;
 748
 749         key.objectid = 0;
 750         key.offset = 0;
 751         key.type = 0;
 752
 753         while (1) {
 754                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 755                 if (ret < 0)
 756                         goto out;
 757                 leaf = path->nodes[0];
 758                 nr = btrfs_header_nritems(leaf);
 759                 if (!nr)
 760                         break;
 761                 /*
 762                  * delete the leaf one by one
 763                  * since the whole tree is going
 764                  * to be deleted.
 765                  */
 766                 path->slots[0] = 0;
 767                 ret = btrfs_del_items(trans, root, path, 0, nr);
 768                 if (ret)
 769                         goto out;
 770
 771                 btrfs_release_path(path);
 772         }
 773         ret = 0;
 774 out:
 775         root->fs_info->pending_quota_state = 0;
 776         btrfs_free_path(path);
 777         return ret;
 778 }
 779
 780 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 781                        struct btrfs_fs_info *fs_info)
 782 {
 783         struct btrfs_root *quota_root;
 784         struct btrfs_root *tree_root = fs_info->tree_root;
 785         struct btrfs_path *path = NULL;
 786         struct btrfs_qgroup_status_item *ptr;
 787         struct extent_buffer *leaf;
 788         struct btrfs_key key;
 789         struct btrfs_key found_key;
 790         struct btrfs_qgroup *qgroup = NULL;
 791         int ret = 0;
 792         int slot;
 793
 794         mutex_lock(&fs_info->qgroup_ioctl_lock);
 795         if (fs_info->quota_root) {
 796                 fs_info->pending_quota_state = 1;
 797                 goto out;
 798         }
 799
 800         /*
 801          * initially create the quota tree
 802          */
 803         quota_root = btrfs_create_tree(trans, fs_info,
 804                                        BTRFS_QUOTA_TREE_OBJECTID);
 805         if (IS_ERR(quota_root)) {
 806                 ret =  PTR_ERR(quota_root);
 807                 goto out;
 808         }
 809
 810         path = btrfs_alloc_path();
 811         if (!path) {
 812                 ret = -ENOMEM;
 813                 goto out_free_root;
 814         }
 815
 816         key.objectid = 0;
 817         key.type = BTRFS_QGROUP_STATUS_KEY;
 818         key.offset = 0;
 819
 820         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 821                                       sizeof(*ptr));
 822         if (ret)
 823                 goto out_free_path;
 824
 825         leaf = path->nodes[0];
 826         ptr = btrfs_item_ptr(leaf, path->slots[0],
 827                                  struct btrfs_qgroup_status_item);
 828         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 829         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 830         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 831                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 832         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 833         btrfs_set_qgroup_status_scan(leaf, ptr, 0);
 834
 835         btrfs_mark_buffer_dirty(leaf);
 836
 837         key.objectid = 0;
 838         key.type = BTRFS_ROOT_REF_KEY;
 839         key.offset = 0;
 840
 841         btrfs_release_path(path);
 842         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 843         if (ret > 0)
 844                 goto out_add_root;
 845         if (ret < 0)
 846                 goto out_free_path;
 847
 848
 849         while (1) {
 850                 slot = path->slots[0];
 851                 leaf = path->nodes[0];
 852                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 853
 854                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 855                         ret = add_qgroup_item(trans, quota_root,
 856                                               found_key.offset);
 857                         if (ret)
 858                                 goto out_free_path;
 859
 860                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 861                         if (IS_ERR(qgroup)) {
 862                                 ret = PTR_ERR(qgroup);
 863                                 goto out_free_path;
 864                         }
 865                 }
 866                 ret = btrfs_next_item(tree_root, path);
 867                 if (ret < 0)
 868                         goto out_free_path;
 869                 if (ret)
 870                         break;
 871         }
 872
 873 out_add_root:
 874         btrfs_release_path(path);
 875         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
 876         if (ret)
 877                 goto out_free_path;
 878
 879         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
 880         if (IS_ERR(qgroup)) {
 881                 ret = PTR_ERR(qgroup);
 882                 goto out_free_path;
 883         }
 884         spin_lock(&fs_info->qgroup_lock);
 885         fs_info->quota_root = quota_root;
 886         fs_info->pending_quota_state = 1;
 887         spin_unlock(&fs_info->qgroup_lock);
 888 out_free_path:
 889         btrfs_free_path(path);
 890 out_free_root:
 891         if (ret) {
 892                 free_extent_buffer(quota_root->node);
 893                 free_extent_buffer(quota_root->commit_root);
 894                 kfree(quota_root);
 895         }
 896 out:
 897         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 898         return ret;
 899 }
 900
 901 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
 902                         struct btrfs_fs_info *fs_info)
 903 {
 904         struct btrfs_root *tree_root = fs_info->tree_root;
 905         struct btrfs_root *quota_root;
 906         int ret = 0;
 907
 908         mutex_lock(&fs_info->qgroup_ioctl_lock);
 909         if (!fs_info->quota_root)
 910                 goto out;
 911         spin_lock(&fs_info->qgroup_lock);
 912         fs_info->quota_enabled = 0;
 913         fs_info->pending_quota_state = 0;
 914         quota_root = fs_info->quota_root;
 915         fs_info->quota_root = NULL;
 916         btrfs_free_qgroup_config(fs_info);
 917         spin_unlock(&fs_info->qgroup_lock);
 918
 919         if (!quota_root) {
 920                 ret = -EINVAL;
 921                 goto out;
 922         }
 923
 924         ret = btrfs_clean_quota_tree(trans, quota_root);
 925         if (ret)
 926                 goto out;
 927
 928         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
 929         if (ret)
 930                 goto out;
 931
 932         list_del(&quota_root->dirty_list);
 933
 934         btrfs_tree_lock(quota_root->node);
 935         clean_tree_block(trans, tree_root, quota_root->node);
 936         btrfs_tree_unlock(quota_root->node);
 937         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
 938
 939         free_extent_buffer(quota_root->node);
 940         free_extent_buffer(quota_root->commit_root);
 941         kfree(quota_root);
 942 out:
 943         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 944         return ret;
 945 }
 946
 947 int btrfs_quota_rescan(struct btrfs_fs_info *fs_info)
 948 {
 949         /* FIXME */
 950         return 0;
 951 }
 952
 953 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
 954                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
 955 {
 956         struct btrfs_root *quota_root;
 957         struct btrfs_qgroup *parent;
 958         struct btrfs_qgroup *member;
 959         int ret = 0;
 960
 961         mutex_lock(&fs_info->qgroup_ioctl_lock);
 962         quota_root = fs_info->quota_root;
 963         if (!quota_root) {
 964                 ret = -EINVAL;
 965                 goto out;
 966         }
 967         member = find_qgroup_rb(fs_info, src);
 968         parent = find_qgroup_rb(fs_info, dst);
 969         if (!member || !parent) {
 970                 ret = -EINVAL;
 971                 goto out;
 972         }
 973
 974         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
 975         if (ret)
 976                 goto out;
 977
 978         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
 979         if (ret) {
 980                 del_qgroup_relation_item(trans, quota_root, src, dst);
 981                 goto out;
 982         }
 983
 984         spin_lock(&fs_info->qgroup_lock);
 985         ret = add_relation_rb(quota_root->fs_info, src, dst);
 986         spin_unlock(&fs_info->qgroup_lock);
 987 out:
 988         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 989         return ret;
 990 }
 991
 992 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
 993                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
 994 {
 995         struct btrfs_root *quota_root;
 996         int ret = 0;
 997         int err;
 998
 999         mutex_lock(&fs_info->qgroup_ioctl_lock);
1000         quota_root = fs_info->quota_root;
1001         if (!quota_root) {
1002                 ret = -EINVAL;
1003                 goto out;
1004         }
1005
1006         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1007         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1008         if (err && !ret)
1009                 ret = err;
1010
1011         spin_lock(&fs_info->qgroup_lock);
1012         del_relation_rb(fs_info, src, dst);
1013
1014         spin_unlock(&fs_info->qgroup_lock);
1015 out:
1016         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1017         return ret;
1018 }
1019
1020 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1021                         struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
1022 {
1023         struct btrfs_root *quota_root;
1024         struct btrfs_qgroup *qgroup;
1025         int ret = 0;
1026
1027         mutex_lock(&fs_info->qgroup_ioctl_lock);
1028         quota_root = fs_info->quota_root;
1029         if (!quota_root) {
1030                 ret = -EINVAL;
1031                 goto out;
1032         }
1033
1034         ret = add_qgroup_item(trans, quota_root, qgroupid);
1035
1036         spin_lock(&fs_info->qgroup_lock);
1037         qgroup = add_qgroup_rb(fs_info, qgroupid);
1038         spin_unlock(&fs_info->qgroup_lock);
1039
1040         if (IS_ERR(qgroup))
1041                 ret = PTR_ERR(qgroup);
1042 out:
1043         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1044         return ret;
1045 }
1046
1047 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1048                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1049 {
1050         struct btrfs_root *quota_root;
1051         struct btrfs_qgroup *qgroup;
1052         int ret = 0;
1053
1054         mutex_lock(&fs_info->qgroup_ioctl_lock);
1055         quota_root = fs_info->quota_root;
1056         if (!quota_root) {
1057                 ret = -EINVAL;
1058                 goto out;
1059         }
1060
1061         /* check if there are no relations to this qgroup */
1062         qgroup = find_qgroup_rb(fs_info, qgroupid);
1063         if (qgroup) {
1064                 if (!list_empty(&qgroup->groups) || !list_empty(&qgroup->members)) {
1065                         ret = -EBUSY;
1066                         goto out;
1067                 }
1068         }
1069
1070         ret = del_qgroup_item(trans, quota_root, qgroupid);
1071
1072         spin_lock(&fs_info->qgroup_lock);
1073         del_qgroup_rb(quota_root->fs_info, qgroupid);
1074         spin_unlock(&fs_info->qgroup_lock);
1075 out:
1076         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1077         return ret;
1078 }
1079
1080 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1081                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1082                        struct btrfs_qgroup_limit *limit)
1083 {
1084         struct btrfs_root *quota_root;
1085         struct btrfs_qgroup *qgroup;
1086         int ret = 0;
1087
1088         mutex_lock(&fs_info->qgroup_ioctl_lock);
1089         quota_root = fs_info->quota_root;
1090         if (!quota_root) {
1091                 ret = -EINVAL;
1092                 goto out;
1093         }
1094
1095         qgroup = find_qgroup_rb(fs_info, qgroupid);
1096         if (!qgroup) {
1097                 ret = -ENOENT;
1098                 goto out;
1099         }
1100         ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1101                                        limit->flags, limit->max_rfer,
1102                                        limit->max_excl, limit->rsv_rfer,
1103                                        limit->rsv_excl);
1104         if (ret) {
1105                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1106                 printk(KERN_INFO "unable to update quota limit for %llu\n",
1107                        (unsigned long long)qgroupid);
1108         }
1109
1110         spin_lock(&fs_info->qgroup_lock);
1111         qgroup->lim_flags = limit->flags;
1112         qgroup->max_rfer = limit->max_rfer;
1113         qgroup->max_excl = limit->max_excl;
1114         qgroup->rsv_rfer = limit->rsv_rfer;
1115         qgroup->rsv_excl = limit->rsv_excl;
1116         spin_unlock(&fs_info->qgroup_lock);
1117 out:
1118         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1119         return ret;
1120 }
1121
1122 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1123                          struct btrfs_qgroup *qgroup)
1124 {
1125         if (list_empty(&qgroup->dirty))
1126                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1127 }
1128
1129 /*
1130  * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1131  * the modification into a list that's later used by btrfs_end_transaction to
1132  * pass the recorded modifications on to btrfs_qgroup_account_ref.
1133  */
1134 int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1135                             struct btrfs_delayed_ref_node *node,
1136                             struct btrfs_delayed_extent_op *extent_op)
1137 {
1138         struct qgroup_update *u;
1139
1140         BUG_ON(!trans->delayed_ref_elem.seq);
1141         u = kmalloc(sizeof(*u), GFP_NOFS);
1142         if (!u)
1143                 return -ENOMEM;
1144
1145         u->node = node;
1146         u->extent_op = extent_op;
1147         list_add_tail(&u->list, &trans->qgroup_ref_list);
1148
1149         return 0;
1150 }
1151
1152 /*
1153  * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1154  * from the fs. First, all roots referencing the extent are searched, and
1155  * then the space is accounted accordingly to the different roots. The
1156  * accounting algorithm works in 3 steps documented inline.
1157  */
1158 int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1159                              struct btrfs_fs_info *fs_info,
1160                              struct btrfs_delayed_ref_node *node,
1161                              struct btrfs_delayed_extent_op *extent_op)
1162 {
1163         struct btrfs_key ins;
1164         struct btrfs_root *quota_root;
1165         u64 ref_root;
1166         struct btrfs_qgroup *qgroup;
1167         struct ulist_node *unode;
1168         struct ulist *roots = NULL;
1169         struct ulist *tmp = NULL;
1170         struct ulist_iterator uiter;
1171         u64 seq;
1172         int ret = 0;
1173         int sgn;
1174
1175         if (!fs_info->quota_enabled)
1176                 return 0;
1177
1178         BUG_ON(!fs_info->quota_root);
1179
1180         ins.objectid = node->bytenr;
1181         ins.offset = node->num_bytes;
1182         ins.type = BTRFS_EXTENT_ITEM_KEY;
1183
1184         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1185             node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1186                 struct btrfs_delayed_tree_ref *ref;
1187                 ref = btrfs_delayed_node_to_tree_ref(node);
1188                 ref_root = ref->root;
1189         } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1190                    node->type == BTRFS_SHARED_DATA_REF_KEY) {
1191                 struct btrfs_delayed_data_ref *ref;
1192                 ref = btrfs_delayed_node_to_data_ref(node);
1193                 ref_root = ref->root;
1194         } else {
1195                 BUG();
1196         }
1197
1198         if (!is_fstree(ref_root)) {
1199                 /*
1200                  * non-fs-trees are not being accounted
1201                  */
1202                 return 0;
1203         }
1204
1205         switch (node->action) {
1206         case BTRFS_ADD_DELAYED_REF:
1207         case BTRFS_ADD_DELAYED_EXTENT:
1208                 sgn = 1;
1209                 break;
1210         case BTRFS_DROP_DELAYED_REF:
1211                 sgn = -1;
1212                 break;
1213         case BTRFS_UPDATE_DELAYED_HEAD:
1214                 return 0;
1215         default:
1216                 BUG();
1217         }
1218
1219         /*
1220          * the delayed ref sequence number we pass depends on the direction of
1221          * the operation. for add operations, we pass (node->seq - 1) to skip
1222          * the delayed ref's current sequence number, because we need the state
1223          * of the tree before the add operation. for delete operations, we pass
1224          * (node->seq) to include the delayed ref's current sequence number,
1225          * because we need the state of the tree after the delete operation.
1226          */
1227         ret = btrfs_find_all_roots(trans, fs_info, node->bytenr,
1228                                    sgn > 0 ? node->seq - 1 : node->seq, &roots);
1229         if (ret < 0)
1230                 return ret;
1231
1232         spin_lock(&fs_info->qgroup_lock);
1233         quota_root = fs_info->quota_root;
1234         if (!quota_root)
1235                 goto unlock;
1236
1237         qgroup = find_qgroup_rb(fs_info, ref_root);
1238         if (!qgroup)
1239                 goto unlock;
1240
1241         /*
1242          * step 1: for each old ref, visit all nodes once and inc refcnt
1243          */
1244         tmp = ulist_alloc(GFP_ATOMIC);
1245         if (!tmp) {
1246                 ret = -ENOMEM;
1247                 goto unlock;
1248         }
1249         seq = fs_info->qgroup_seq;
1250         fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1251
1252         ULIST_ITER_INIT(&uiter);
1253         while ((unode = ulist_next(roots, &uiter))) {
1254                 struct ulist_node *tmp_unode;
1255                 struct ulist_iterator tmp_uiter;
1256                 struct btrfs_qgroup *qg;
1257
1258                 qg = find_qgroup_rb(fs_info, unode->val);
1259                 if (!qg)
1260                         continue;
1261
1262                 ulist_reinit(tmp);
1263                                                 /* XXX id not needed */
1264                 ret = ulist_add(tmp, qg->qgroupid,
1265                                 (u64)(uintptr_t)qg, GFP_ATOMIC);
1266                 if (ret < 0)
1267                         goto unlock;
1268                 ULIST_ITER_INIT(&tmp_uiter);
1269                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1270                         struct btrfs_qgroup_list *glist;
1271
1272                         qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1273                         if (qg->refcnt < seq)
1274                                 qg->refcnt = seq + 1;
1275                         else
1276                                 ++qg->refcnt;
1277
1278                         list_for_each_entry(glist, &qg->groups, next_group) {
1279                                 ret = ulist_add(tmp, glist->group->qgroupid,
1280                                                 (u64)(uintptr_t)glist->group,
1281                                                 GFP_ATOMIC);
1282                                 if (ret < 0)
1283                                         goto unlock;
1284                         }
1285                 }
1286         }
1287
1288         /*
1289          * step 2: walk from the new root
1290          */
1291         ulist_reinit(tmp);
1292         ret = ulist_add(tmp, qgroup->qgroupid,
1293                         (uintptr_t)qgroup, GFP_ATOMIC);
1294         if (ret < 0)
1295                 goto unlock;
1296         ULIST_ITER_INIT(&uiter);
1297         while ((unode = ulist_next(tmp, &uiter))) {
1298                 struct btrfs_qgroup *qg;
1299                 struct btrfs_qgroup_list *glist;
1300
1301                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1302                 if (qg->refcnt < seq) {
1303                         /* not visited by step 1 */
1304                         qg->rfer += sgn * node->num_bytes;
1305                         qg->rfer_cmpr += sgn * node->num_bytes;
1306                         if (roots->nnodes == 0) {
1307                                 qg->excl += sgn * node->num_bytes;
1308                                 qg->excl_cmpr += sgn * node->num_bytes;
1309                         }
1310                         qgroup_dirty(fs_info, qg);
1311                 }
1312                 WARN_ON(qg->tag >= seq);
1313                 qg->tag = seq;
1314
1315                 list_for_each_entry(glist, &qg->groups, next_group) {
1316                         ret = ulist_add(tmp, glist->group->qgroupid,
1317                                         (uintptr_t)glist->group, GFP_ATOMIC);
1318                         if (ret < 0)
1319                                 goto unlock;
1320                 }
1321         }
1322
1323         /*
1324          * step 3: walk again from old refs
1325          */
1326         ULIST_ITER_INIT(&uiter);
1327         while ((unode = ulist_next(roots, &uiter))) {
1328                 struct btrfs_qgroup *qg;
1329                 struct ulist_node *tmp_unode;
1330                 struct ulist_iterator tmp_uiter;
1331
1332                 qg = find_qgroup_rb(fs_info, unode->val);
1333                 if (!qg)
1334                         continue;
1335
1336                 ulist_reinit(tmp);
1337                 ret = ulist_add(tmp, qg->qgroupid,
1338                                 (uintptr_t)qg, GFP_ATOMIC);
1339                 if (ret < 0)
1340                         goto unlock;
1341                 ULIST_ITER_INIT(&tmp_uiter);
1342                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1343                         struct btrfs_qgroup_list *glist;
1344
1345                         qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1346                         if (qg->tag == seq)
1347                                 continue;
1348
1349                         if (qg->refcnt - seq == roots->nnodes) {
1350                                 qg->excl -= sgn * node->num_bytes;
1351                                 qg->excl_cmpr -= sgn * node->num_bytes;
1352                                 qgroup_dirty(fs_info, qg);
1353                         }
1354
1355                         list_for_each_entry(glist, &qg->groups, next_group) {
1356                                 ret = ulist_add(tmp, glist->group->qgroupid,
1357                                                 (uintptr_t)glist->group,
1358                                                 GFP_ATOMIC);
1359                                 if (ret < 0)
1360                                         goto unlock;
1361                         }
1362                 }
1363         }
1364         ret = 0;
1365 unlock:
1366         spin_unlock(&fs_info->qgroup_lock);
1367         ulist_free(roots);
1368         ulist_free(tmp);
1369
1370         return ret;
1371 }
1372
1373 /*
1374  * called from commit_transaction. Writes all changed qgroups to disk.
1375  */
1376 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1377                       struct btrfs_fs_info *fs_info)
1378 {
1379         struct btrfs_root *quota_root = fs_info->quota_root;
1380         int ret = 0;
1381
1382         if (!quota_root)
1383                 goto out;
1384
1385         fs_info->quota_enabled = fs_info->pending_quota_state;
1386
1387         spin_lock(&fs_info->qgroup_lock);
1388         while (!list_empty(&fs_info->dirty_qgroups)) {
1389                 struct btrfs_qgroup *qgroup;
1390                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1391                                           struct btrfs_qgroup, dirty);
1392                 list_del_init(&qgroup->dirty);
1393                 spin_unlock(&fs_info->qgroup_lock);
1394                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1395                 if (ret)
1396                         fs_info->qgroup_flags |=
1397                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1398                 spin_lock(&fs_info->qgroup_lock);
1399         }
1400         if (fs_info->quota_enabled)
1401                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1402         else
1403                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1404         spin_unlock(&fs_info->qgroup_lock);
1405
1406         ret = update_qgroup_status_item(trans, fs_info, quota_root);
1407         if (ret)
1408                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1409
1410 out:
1411
1412         return ret;
1413 }
1414
1415 /*
1416  * copy the acounting information between qgroups. This is necessary when a
1417  * snapshot or a subvolume is created
1418  */
1419 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1420                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1421                          struct btrfs_qgroup_inherit *inherit)
1422 {
1423         int ret = 0;
1424         int i;
1425         u64 *i_qgroups;
1426         struct btrfs_root *quota_root = fs_info->quota_root;
1427         struct btrfs_qgroup *srcgroup;
1428         struct btrfs_qgroup *dstgroup;
1429         u32 level_size = 0;
1430         u64 nums;
1431
1432         mutex_lock(&fs_info->qgroup_ioctl_lock);
1433         if (!fs_info->quota_enabled)
1434                 goto out;
1435
1436         if (!quota_root) {
1437                 ret = -EINVAL;
1438                 goto out;
1439         }
1440
1441         if (inherit) {
1442                 i_qgroups = (u64 *)(inherit + 1);
1443                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1444                        2 * inherit->num_excl_copies;
1445                 for (i = 0; i < nums; ++i) {
1446                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1447                         if (!srcgroup) {
1448                                 ret = -EINVAL;
1449                                 goto out;
1450                         }
1451                         ++i_qgroups;
1452                 }
1453         }
1454
1455         /*
1456          * create a tracking group for the subvol itself
1457          */
1458         ret = add_qgroup_item(trans, quota_root, objectid);
1459         if (ret)
1460                 goto out;
1461
1462         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1463                 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1464                                                inherit->lim.flags,
1465                                                inherit->lim.max_rfer,
1466                                                inherit->lim.max_excl,
1467                                                inherit->lim.rsv_rfer,
1468                                                inherit->lim.rsv_excl);
1469                 if (ret)
1470                         goto out;
1471         }
1472
1473         if (srcid) {
1474                 struct btrfs_root *srcroot;
1475                 struct btrfs_key srckey;
1476                 int srcroot_level;
1477
1478                 srckey.objectid = srcid;
1479                 srckey.type = BTRFS_ROOT_ITEM_KEY;
1480                 srckey.offset = (u64)-1;
1481                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1482                 if (IS_ERR(srcroot)) {
1483                         ret = PTR_ERR(srcroot);
1484                         goto out;
1485                 }
1486
1487                 rcu_read_lock();
1488                 srcroot_level = btrfs_header_level(srcroot->node);
1489                 level_size = btrfs_level_size(srcroot, srcroot_level);
1490                 rcu_read_unlock();
1491         }
1492
1493         /*
1494          * add qgroup to all inherited groups
1495          */
1496         if (inherit) {
1497                 i_qgroups = (u64 *)(inherit + 1);
1498                 for (i = 0; i < inherit->num_qgroups; ++i) {
1499                         ret = add_qgroup_relation_item(trans, quota_root,
1500                                                        objectid, *i_qgroups);
1501                         if (ret)
1502                                 goto out;
1503                         ret = add_qgroup_relation_item(trans, quota_root,
1504                                                        *i_qgroups, objectid);
1505                         if (ret)
1506                                 goto out;
1507                         ++i_qgroups;
1508                 }
1509         }
1510
1511
1512         spin_lock(&fs_info->qgroup_lock);
1513
1514         dstgroup = add_qgroup_rb(fs_info, objectid);
1515         if (IS_ERR(dstgroup)) {
1516                 ret = PTR_ERR(dstgroup);
1517                 goto unlock;
1518         }
1519
1520         if (srcid) {
1521                 srcgroup = find_qgroup_rb(fs_info, srcid);
1522                 if (!srcgroup)
1523                         goto unlock;
1524                 dstgroup->rfer = srcgroup->rfer - level_size;
1525                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1526                 srcgroup->excl = level_size;
1527                 srcgroup->excl_cmpr = level_size;
1528                 qgroup_dirty(fs_info, dstgroup);
1529                 qgroup_dirty(fs_info, srcgroup);
1530         }
1531
1532         if (!inherit)
1533                 goto unlock;
1534
1535         i_qgroups = (u64 *)(inherit + 1);
1536         for (i = 0; i < inherit->num_qgroups; ++i) {
1537                 ret = add_relation_rb(quota_root->fs_info, objectid,
1538                                       *i_qgroups);
1539                 if (ret)
1540                         goto unlock;
1541                 ++i_qgroups;
1542         }
1543
1544         for (i = 0; i <  inherit->num_ref_copies; ++i) {
1545                 struct btrfs_qgroup *src;
1546                 struct btrfs_qgroup *dst;
1547
1548                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1549                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1550
1551                 if (!src || !dst) {
1552                         ret = -EINVAL;
1553                         goto unlock;
1554                 }
1555
1556                 dst->rfer = src->rfer - level_size;
1557                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1558                 i_qgroups += 2;
1559         }
1560         for (i = 0; i <  inherit->num_excl_copies; ++i) {
1561                 struct btrfs_qgroup *src;
1562                 struct btrfs_qgroup *dst;
1563
1564                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1565                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1566
1567                 if (!src || !dst) {
1568                         ret = -EINVAL;
1569                         goto unlock;
1570                 }
1571
1572                 dst->excl = src->excl + level_size;
1573                 dst->excl_cmpr = src->excl_cmpr + level_size;
1574                 i_qgroups += 2;
1575         }
1576
1577 unlock:
1578         spin_unlock(&fs_info->qgroup_lock);
1579 out:
1580         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1581         return ret;
1582 }
1583
1584 /*
1585  * reserve some space for a qgroup and all its parents. The reservation takes
1586  * place with start_transaction or dealloc_reserve, similar to ENOSPC
1587  * accounting. If not enough space is available, EDQUOT is returned.
1588  * We assume that the requested space is new for all qgroups.
1589  */
1590 int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1591 {
1592         struct btrfs_root *quota_root;
1593         struct btrfs_qgroup *qgroup;
1594         struct btrfs_fs_info *fs_info = root->fs_info;
1595         u64 ref_root = root->root_key.objectid;
1596         int ret = 0;
1597         struct ulist *ulist = NULL;
1598         struct ulist_node *unode;
1599         struct ulist_iterator uiter;
1600
1601         if (!is_fstree(ref_root))
1602                 return 0;
1603
1604         if (num_bytes == 0)
1605                 return 0;
1606
1607         spin_lock(&fs_info->qgroup_lock);
1608         quota_root = fs_info->quota_root;
1609         if (!quota_root)
1610                 goto out;
1611
1612         qgroup = find_qgroup_rb(fs_info, ref_root);
1613         if (!qgroup)
1614                 goto out;
1615
1616         /*
1617          * in a first step, we check all affected qgroups if any limits would
1618          * be exceeded
1619          */
1620         ulist = ulist_alloc(GFP_ATOMIC);
1621         if (!ulist) {
1622                 ret = -ENOMEM;
1623                 goto out;
1624         }
1625         ret = ulist_add(ulist, qgroup->qgroupid,
1626                         (uintptr_t)qgroup, GFP_ATOMIC);
1627         if (ret < 0)
1628                 goto out;
1629         ULIST_ITER_INIT(&uiter);
1630         while ((unode = ulist_next(ulist, &uiter))) {
1631                 struct btrfs_qgroup *qg;
1632                 struct btrfs_qgroup_list *glist;
1633
1634                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1635
1636                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1637                     qg->reserved + (s64)qg->rfer + num_bytes >
1638                     qg->max_rfer) {
1639                         ret = -EDQUOT;
1640                         goto out;
1641                 }
1642
1643                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1644                     qg->reserved + (s64)qg->excl + num_bytes >
1645                     qg->max_excl) {
1646                         ret = -EDQUOT;
1647                         goto out;
1648                 }
1649
1650                 list_for_each_entry(glist, &qg->groups, next_group) {
1651                         ret = ulist_add(ulist, glist->group->qgroupid,
1652                                         (uintptr_t)glist->group, GFP_ATOMIC);
1653                         if (ret < 0)
1654                                 goto out;
1655                 }
1656         }
1657         ret = 0;
1658         /*
1659          * no limits exceeded, now record the reservation into all qgroups
1660          */
1661         ULIST_ITER_INIT(&uiter);
1662         while ((unode = ulist_next(ulist, &uiter))) {
1663                 struct btrfs_qgroup *qg;
1664
1665                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1666
1667                 qg->reserved += num_bytes;
1668         }
1669
1670 out:
1671         spin_unlock(&fs_info->qgroup_lock);
1672         ulist_free(ulist);
1673
1674         return ret;
1675 }
1676
1677 void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1678 {
1679         struct btrfs_root *quota_root;
1680         struct btrfs_qgroup *qgroup;
1681         struct btrfs_fs_info *fs_info = root->fs_info;
1682         struct ulist *ulist = NULL;
1683         struct ulist_node *unode;
1684         struct ulist_iterator uiter;
1685         u64 ref_root = root->root_key.objectid;
1686         int ret = 0;
1687
1688         if (!is_fstree(ref_root))
1689                 return;
1690
1691         if (num_bytes == 0)
1692                 return;
1693
1694         spin_lock(&fs_info->qgroup_lock);
1695
1696         quota_root = fs_info->quota_root;
1697         if (!quota_root)
1698                 goto out;
1699
1700         qgroup = find_qgroup_rb(fs_info, ref_root);
1701         if (!qgroup)
1702                 goto out;
1703
1704         ulist = ulist_alloc(GFP_ATOMIC);
1705         if (!ulist) {
1706                 btrfs_std_error(fs_info, -ENOMEM);
1707                 goto out;
1708         }
1709         ret = ulist_add(ulist, qgroup->qgroupid,
1710                         (uintptr_t)qgroup, GFP_ATOMIC);
1711         if (ret < 0)
1712                 goto out;
1713         ULIST_ITER_INIT(&uiter);
1714         while ((unode = ulist_next(ulist, &uiter))) {
1715                 struct btrfs_qgroup *qg;
1716                 struct btrfs_qgroup_list *glist;
1717
1718                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1719
1720                 qg->reserved -= num_bytes;
1721
1722                 list_for_each_entry(glist, &qg->groups, next_group) {
1723                         ret = ulist_add(ulist, glist->group->qgroupid,
1724                                         (uintptr_t)glist->group, GFP_ATOMIC);
1725                         if (ret < 0)
1726                                 goto out;
1727                 }
1728         }
1729
1730 out:
1731         spin_unlock(&fs_info->qgroup_lock);
1732         ulist_free(ulist);
1733 }
1734
1735 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1736 {
1737         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1738                 return;
1739         printk(KERN_ERR "btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1740                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1741                 trans->delayed_ref_elem.seq);
1742         BUG();
1743 }