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 #include "extent_io.h"
  35 #include "qgroup.h"
  36
  37
  38 /* TODO XXX FIXME
  39  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  40  *  - reorganize keys
  41  *  - compressed
  42  *  - sync
  43  *  - copy also limits on subvol creation
  44  *  - limit
  45  *  - caches fuer ulists
  46  *  - performance benchmarks
  47  *  - check all ioctl parameters
  48  */
  49
  50 /*
  51  * one struct for each qgroup, organized in fs_info->qgroup_tree.
  52  */
  53 struct btrfs_qgroup {
  54         u64 qgroupid;
  55
  56         /*
  57          * state
  58          */
  59         u64 rfer;       /* referenced */
  60         u64 rfer_cmpr;  /* referenced compressed */
  61         u64 excl;       /* exclusive */
  62         u64 excl_cmpr;  /* exclusive compressed */
  63
  64         /*
  65          * limits
  66          */
  67         u64 lim_flags;  /* which limits are set */
  68         u64 max_rfer;
  69         u64 max_excl;
  70         u64 rsv_rfer;
  71         u64 rsv_excl;
  72
  73         /*
  74          * reservation tracking
  75          */
  76         u64 reserved;
  77
  78         /*
  79          * lists
  80          */
  81         struct list_head groups;  /* groups this group is member of */
  82         struct list_head members; /* groups that are members of this group */
  83         struct list_head dirty;   /* dirty groups */
  84         struct rb_node node;      /* tree of qgroups */
  85
  86         /*
  87          * temp variables for accounting operations
  88          * Refer to qgroup_shared_accounting() for details.
  89          */
  90         u64 old_refcnt;
  91         u64 new_refcnt;
  92 };
  93
  94 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
  95                                            int mod)
  96 {
  97         if (qg->old_refcnt < seq)
  98                 qg->old_refcnt = seq;
  99         qg->old_refcnt += mod;
 100 }
 101
 102 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
 103                                            int mod)
 104 {
 105         if (qg->new_refcnt < seq)
 106                 qg->new_refcnt = seq;
 107         qg->new_refcnt += mod;
 108 }
 109
 110 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
 111 {
 112         if (qg->old_refcnt < seq)
 113                 return 0;
 114         return qg->old_refcnt - seq;
 115 }
 116
 117 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
 118 {
 119         if (qg->new_refcnt < seq)
 120                 return 0;
 121         return qg->new_refcnt - seq;
 122 }
 123
 124 /*
 125  * glue structure to represent the relations between qgroups.
 126  */
 127 struct btrfs_qgroup_list {
 128         struct list_head next_group;
 129         struct list_head next_member;
 130         struct btrfs_qgroup *group;
 131         struct btrfs_qgroup *member;
 132 };
 133
 134 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
 135 {
 136         return (u64)(uintptr_t)qg;
 137 }
 138
 139 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
 140 {
 141         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
 142 }
 143
 144 static int
 145 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 146                    int init_flags);
 147 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 148
 149 /* must be called with qgroup_ioctl_lock held */
 150 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 151                                            u64 qgroupid)
 152 {
 153         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 154         struct btrfs_qgroup *qgroup;
 155
 156         while (n) {
 157                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 158                 if (qgroup->qgroupid < qgroupid)
 159                         n = n->rb_left;
 160                 else if (qgroup->qgroupid > qgroupid)
 161                         n = n->rb_right;
 162                 else
 163                         return qgroup;
 164         }
 165         return NULL;
 166 }
 167
 168 /* must be called with qgroup_lock held */
 169 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 170                                           u64 qgroupid)
 171 {
 172         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 173         struct rb_node *parent = NULL;
 174         struct btrfs_qgroup *qgroup;
 175
 176         while (*p) {
 177                 parent = *p;
 178                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 179
 180                 if (qgroup->qgroupid < qgroupid)
 181                         p = &(*p)->rb_left;
 182                 else if (qgroup->qgroupid > qgroupid)
 183                         p = &(*p)->rb_right;
 184                 else
 185                         return qgroup;
 186         }
 187
 188         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 189         if (!qgroup)
 190                 return ERR_PTR(-ENOMEM);
 191
 192         qgroup->qgroupid = qgroupid;
 193         INIT_LIST_HEAD(&qgroup->groups);
 194         INIT_LIST_HEAD(&qgroup->members);
 195         INIT_LIST_HEAD(&qgroup->dirty);
 196
 197         rb_link_node(&qgroup->node, parent, p);
 198         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 199
 200         return qgroup;
 201 }
 202
 203 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
 204 {
 205         struct btrfs_qgroup_list *list;
 206
 207         list_del(&qgroup->dirty);
 208         while (!list_empty(&qgroup->groups)) {
 209                 list = list_first_entry(&qgroup->groups,
 210                                         struct btrfs_qgroup_list, next_group);
 211                 list_del(&list->next_group);
 212                 list_del(&list->next_member);
 213                 kfree(list);
 214         }
 215
 216         while (!list_empty(&qgroup->members)) {
 217                 list = list_first_entry(&qgroup->members,
 218                                         struct btrfs_qgroup_list, next_member);
 219                 list_del(&list->next_group);
 220                 list_del(&list->next_member);
 221                 kfree(list);
 222         }
 223         kfree(qgroup);
 224 }
 225
 226 /* must be called with qgroup_lock held */
 227 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 228 {
 229         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 230
 231         if (!qgroup)
 232                 return -ENOENT;
 233
 234         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 235         __del_qgroup_rb(qgroup);
 236         return 0;
 237 }
 238
 239 /* must be called with qgroup_lock held */
 240 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 241                            u64 memberid, u64 parentid)
 242 {
 243         struct btrfs_qgroup *member;
 244         struct btrfs_qgroup *parent;
 245         struct btrfs_qgroup_list *list;
 246
 247         member = find_qgroup_rb(fs_info, memberid);
 248         parent = find_qgroup_rb(fs_info, parentid);
 249         if (!member || !parent)
 250                 return -ENOENT;
 251
 252         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 253         if (!list)
 254                 return -ENOMEM;
 255
 256         list->group = parent;
 257         list->member = member;
 258         list_add_tail(&list->next_group, &member->groups);
 259         list_add_tail(&list->next_member, &parent->members);
 260
 261         return 0;
 262 }
 263
 264 /* must be called with qgroup_lock held */
 265 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 266                            u64 memberid, u64 parentid)
 267 {
 268         struct btrfs_qgroup *member;
 269         struct btrfs_qgroup *parent;
 270         struct btrfs_qgroup_list *list;
 271
 272         member = find_qgroup_rb(fs_info, memberid);
 273         parent = find_qgroup_rb(fs_info, parentid);
 274         if (!member || !parent)
 275                 return -ENOENT;
 276
 277         list_for_each_entry(list, &member->groups, next_group) {
 278                 if (list->group == parent) {
 279                         list_del(&list->next_group);
 280                         list_del(&list->next_member);
 281                         kfree(list);
 282                         return 0;
 283                 }
 284         }
 285         return -ENOENT;
 286 }
 287
 288 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 289 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 290                                u64 rfer, u64 excl)
 291 {
 292         struct btrfs_qgroup *qgroup;
 293
 294         qgroup = find_qgroup_rb(fs_info, qgroupid);
 295         if (!qgroup)
 296                 return -EINVAL;
 297         if (qgroup->rfer != rfer || qgroup->excl != excl)
 298                 return -EINVAL;
 299         return 0;
 300 }
 301 #endif
 302
 303 /*
 304  * The full config is read in one go, only called from open_ctree()
 305  * It doesn't use any locking, as at this point we're still single-threaded
 306  */
 307 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 308 {
 309         struct btrfs_key key;
 310         struct btrfs_key found_key;
 311         struct btrfs_root *quota_root = fs_info->quota_root;
 312         struct btrfs_path *path = NULL;
 313         struct extent_buffer *l;
 314         int slot;
 315         int ret = 0;
 316         u64 flags = 0;
 317         u64 rescan_progress = 0;
 318
 319         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 320                 return 0;
 321
 322         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 323         if (!fs_info->qgroup_ulist) {
 324                 ret = -ENOMEM;
 325                 goto out;
 326         }
 327
 328         path = btrfs_alloc_path();
 329         if (!path) {
 330                 ret = -ENOMEM;
 331                 goto out;
 332         }
 333
 334         /* default this to quota off, in case no status key is found */
 335         fs_info->qgroup_flags = 0;
 336
 337         /*
 338          * pass 1: read status, all qgroup infos and limits
 339          */
 340         key.objectid = 0;
 341         key.type = 0;
 342         key.offset = 0;
 343         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 344         if (ret)
 345                 goto out;
 346
 347         while (1) {
 348                 struct btrfs_qgroup *qgroup;
 349
 350                 slot = path->slots[0];
 351                 l = path->nodes[0];
 352                 btrfs_item_key_to_cpu(l, &found_key, slot);
 353
 354                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 355                         struct btrfs_qgroup_status_item *ptr;
 356
 357                         ptr = btrfs_item_ptr(l, slot,
 358                                              struct btrfs_qgroup_status_item);
 359
 360                         if (btrfs_qgroup_status_version(l, ptr) !=
 361                             BTRFS_QGROUP_STATUS_VERSION) {
 362                                 btrfs_err(fs_info,
 363                                  "old qgroup version, quota disabled");
 364                                 goto out;
 365                         }
 366                         if (btrfs_qgroup_status_generation(l, ptr) !=
 367                             fs_info->generation) {
 368                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 369                                 btrfs_err(fs_info,
 370                                         "qgroup generation mismatch, marked as inconsistent");
 371                         }
 372                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 373                                                                           ptr);
 374                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 375                         goto next1;
 376                 }
 377
 378                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 379                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 380                         goto next1;
 381
 382                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 383                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 384                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 385                         btrfs_err(fs_info, "inconsistent qgroup config");
 386                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 387                 }
 388                 if (!qgroup) {
 389                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 390                         if (IS_ERR(qgroup)) {
 391                                 ret = PTR_ERR(qgroup);
 392                                 goto out;
 393                         }
 394                 }
 395                 switch (found_key.type) {
 396                 case BTRFS_QGROUP_INFO_KEY: {
 397                         struct btrfs_qgroup_info_item *ptr;
 398
 399                         ptr = btrfs_item_ptr(l, slot,
 400                                              struct btrfs_qgroup_info_item);
 401                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 402                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 403                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 404                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 405                         /* generation currently unused */
 406                         break;
 407                 }
 408                 case BTRFS_QGROUP_LIMIT_KEY: {
 409                         struct btrfs_qgroup_limit_item *ptr;
 410
 411                         ptr = btrfs_item_ptr(l, slot,
 412                                              struct btrfs_qgroup_limit_item);
 413                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 414                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 415                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 416                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 417                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 418                         break;
 419                 }
 420                 }
 421 next1:
 422                 ret = btrfs_next_item(quota_root, path);
 423                 if (ret < 0)
 424                         goto out;
 425                 if (ret)
 426                         break;
 427         }
 428         btrfs_release_path(path);
 429
 430         /*
 431          * pass 2: read all qgroup relations
 432          */
 433         key.objectid = 0;
 434         key.type = BTRFS_QGROUP_RELATION_KEY;
 435         key.offset = 0;
 436         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 437         if (ret)
 438                 goto out;
 439         while (1) {
 440                 slot = path->slots[0];
 441                 l = path->nodes[0];
 442                 btrfs_item_key_to_cpu(l, &found_key, slot);
 443
 444                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 445                         goto next2;
 446
 447                 if (found_key.objectid > found_key.offset) {
 448                         /* parent <- member, not needed to build config */
 449                         /* FIXME should we omit the key completely? */
 450                         goto next2;
 451                 }
 452
 453                 ret = add_relation_rb(fs_info, found_key.objectid,
 454                                       found_key.offset);
 455                 if (ret == -ENOENT) {
 456                         btrfs_warn(fs_info,
 457                                 "orphan qgroup relation 0x%llx->0x%llx",
 458                                 found_key.objectid, found_key.offset);
 459                         ret = 0;        /* ignore the error */
 460                 }
 461                 if (ret)
 462                         goto out;
 463 next2:
 464                 ret = btrfs_next_item(quota_root, path);
 465                 if (ret < 0)
 466                         goto out;
 467                 if (ret)
 468                         break;
 469         }
 470 out:
 471         fs_info->qgroup_flags |= flags;
 472         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
 473                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 474         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 475                  ret >= 0)
 476                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 477         btrfs_free_path(path);
 478
 479         if (ret < 0) {
 480                 ulist_free(fs_info->qgroup_ulist);
 481                 fs_info->qgroup_ulist = NULL;
 482                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 483         }
 484
 485         return ret < 0 ? ret : 0;
 486 }
 487
 488 /*
 489  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 490  * first two are in single-threaded paths.And for the third one, we have set
 491  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 492  * up the in-memory structures without qgroup_lock held.
 493  */
 494 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 495 {
 496         struct rb_node *n;
 497         struct btrfs_qgroup *qgroup;
 498
 499         while ((n = rb_first(&fs_info->qgroup_tree))) {
 500                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 501                 rb_erase(n, &fs_info->qgroup_tree);
 502                 __del_qgroup_rb(qgroup);
 503         }
 504         /*
 505          * we call btrfs_free_qgroup_config() when umounting
 506          * filesystem and disabling quota, so we set qgroup_ulist
 507          * to be null here to avoid double free.
 508          */
 509         ulist_free(fs_info->qgroup_ulist);
 510         fs_info->qgroup_ulist = NULL;
 511 }
 512
 513 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
 514                                     struct btrfs_root *quota_root,
 515                                     u64 src, u64 dst)
 516 {
 517         int ret;
 518         struct btrfs_path *path;
 519         struct btrfs_key key;
 520
 521         path = btrfs_alloc_path();
 522         if (!path)
 523                 return -ENOMEM;
 524
 525         key.objectid = src;
 526         key.type = BTRFS_QGROUP_RELATION_KEY;
 527         key.offset = dst;
 528
 529         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 530
 531         btrfs_mark_buffer_dirty(path->nodes[0]);
 532
 533         btrfs_free_path(path);
 534         return ret;
 535 }
 536
 537 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
 538                                     struct btrfs_root *quota_root,
 539                                     u64 src, u64 dst)
 540 {
 541         int ret;
 542         struct btrfs_path *path;
 543         struct btrfs_key key;
 544
 545         path = btrfs_alloc_path();
 546         if (!path)
 547                 return -ENOMEM;
 548
 549         key.objectid = src;
 550         key.type = BTRFS_QGROUP_RELATION_KEY;
 551         key.offset = dst;
 552
 553         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 554         if (ret < 0)
 555                 goto out;
 556
 557         if (ret > 0) {
 558                 ret = -ENOENT;
 559                 goto out;
 560         }
 561
 562         ret = btrfs_del_item(trans, quota_root, path);
 563 out:
 564         btrfs_free_path(path);
 565         return ret;
 566 }
 567
 568 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 569                            struct btrfs_root *quota_root, u64 qgroupid)
 570 {
 571         int ret;
 572         struct btrfs_path *path;
 573         struct btrfs_qgroup_info_item *qgroup_info;
 574         struct btrfs_qgroup_limit_item *qgroup_limit;
 575         struct extent_buffer *leaf;
 576         struct btrfs_key key;
 577
 578         if (btrfs_is_testing(quota_root->fs_info))
 579                 return 0;
 580
 581         path = btrfs_alloc_path();
 582         if (!path)
 583                 return -ENOMEM;
 584
 585         key.objectid = 0;
 586         key.type = BTRFS_QGROUP_INFO_KEY;
 587         key.offset = qgroupid;
 588
 589         /*
 590          * Avoid a transaction abort by catching -EEXIST here. In that
 591          * case, we proceed by re-initializing the existing structure
 592          * on disk.
 593          */
 594
 595         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 596                                       sizeof(*qgroup_info));
 597         if (ret && ret != -EEXIST)
 598                 goto out;
 599
 600         leaf = path->nodes[0];
 601         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 602                                  struct btrfs_qgroup_info_item);
 603         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 604         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 605         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 606         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 607         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 608
 609         btrfs_mark_buffer_dirty(leaf);
 610
 611         btrfs_release_path(path);
 612
 613         key.type = BTRFS_QGROUP_LIMIT_KEY;
 614         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 615                                       sizeof(*qgroup_limit));
 616         if (ret && ret != -EEXIST)
 617                 goto out;
 618
 619         leaf = path->nodes[0];
 620         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 621                                   struct btrfs_qgroup_limit_item);
 622         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 623         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 624         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 625         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 626         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 627
 628         btrfs_mark_buffer_dirty(leaf);
 629
 630         ret = 0;
 631 out:
 632         btrfs_free_path(path);
 633         return ret;
 634 }
 635
 636 static int del_qgroup_item(struct btrfs_trans_handle *trans,
 637                            struct btrfs_root *quota_root, u64 qgroupid)
 638 {
 639         int ret;
 640         struct btrfs_path *path;
 641         struct btrfs_key key;
 642
 643         path = btrfs_alloc_path();
 644         if (!path)
 645                 return -ENOMEM;
 646
 647         key.objectid = 0;
 648         key.type = BTRFS_QGROUP_INFO_KEY;
 649         key.offset = qgroupid;
 650         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 651         if (ret < 0)
 652                 goto out;
 653
 654         if (ret > 0) {
 655                 ret = -ENOENT;
 656                 goto out;
 657         }
 658
 659         ret = btrfs_del_item(trans, quota_root, path);
 660         if (ret)
 661                 goto out;
 662
 663         btrfs_release_path(path);
 664
 665         key.type = BTRFS_QGROUP_LIMIT_KEY;
 666         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 667         if (ret < 0)
 668                 goto out;
 669
 670         if (ret > 0) {
 671                 ret = -ENOENT;
 672                 goto out;
 673         }
 674
 675         ret = btrfs_del_item(trans, quota_root, path);
 676
 677 out:
 678         btrfs_free_path(path);
 679         return ret;
 680 }
 681
 682 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 683                                     struct btrfs_root *root,
 684                                     struct btrfs_qgroup *qgroup)
 685 {
 686         struct btrfs_path *path;
 687         struct btrfs_key key;
 688         struct extent_buffer *l;
 689         struct btrfs_qgroup_limit_item *qgroup_limit;
 690         int ret;
 691         int slot;
 692
 693         key.objectid = 0;
 694         key.type = BTRFS_QGROUP_LIMIT_KEY;
 695         key.offset = qgroup->qgroupid;
 696
 697         path = btrfs_alloc_path();
 698         if (!path)
 699                 return -ENOMEM;
 700
 701         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 702         if (ret > 0)
 703                 ret = -ENOENT;
 704
 705         if (ret)
 706                 goto out;
 707
 708         l = path->nodes[0];
 709         slot = path->slots[0];
 710         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 711         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 712         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 713         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 714         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 715         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 716
 717         btrfs_mark_buffer_dirty(l);
 718
 719 out:
 720         btrfs_free_path(path);
 721         return ret;
 722 }
 723
 724 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 725                                    struct btrfs_root *root,
 726                                    struct btrfs_qgroup *qgroup)
 727 {
 728         struct btrfs_path *path;
 729         struct btrfs_key key;
 730         struct extent_buffer *l;
 731         struct btrfs_qgroup_info_item *qgroup_info;
 732         int ret;
 733         int slot;
 734
 735         if (btrfs_is_testing(root->fs_info))
 736                 return 0;
 737
 738         key.objectid = 0;
 739         key.type = BTRFS_QGROUP_INFO_KEY;
 740         key.offset = qgroup->qgroupid;
 741
 742         path = btrfs_alloc_path();
 743         if (!path)
 744                 return -ENOMEM;
 745
 746         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 747         if (ret > 0)
 748                 ret = -ENOENT;
 749
 750         if (ret)
 751                 goto out;
 752
 753         l = path->nodes[0];
 754         slot = path->slots[0];
 755         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 756         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 757         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 758         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 759         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 760         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 761
 762         btrfs_mark_buffer_dirty(l);
 763
 764 out:
 765         btrfs_free_path(path);
 766         return ret;
 767 }
 768
 769 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 770                                      struct btrfs_fs_info *fs_info,
 771                                     struct btrfs_root *root)
 772 {
 773         struct btrfs_path *path;
 774         struct btrfs_key key;
 775         struct extent_buffer *l;
 776         struct btrfs_qgroup_status_item *ptr;
 777         int ret;
 778         int slot;
 779
 780         key.objectid = 0;
 781         key.type = BTRFS_QGROUP_STATUS_KEY;
 782         key.offset = 0;
 783
 784         path = btrfs_alloc_path();
 785         if (!path)
 786                 return -ENOMEM;
 787
 788         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 789         if (ret > 0)
 790                 ret = -ENOENT;
 791
 792         if (ret)
 793                 goto out;
 794
 795         l = path->nodes[0];
 796         slot = path->slots[0];
 797         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 798         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 799         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 800         btrfs_set_qgroup_status_rescan(l, ptr,
 801                                 fs_info->qgroup_rescan_progress.objectid);
 802
 803         btrfs_mark_buffer_dirty(l);
 804
 805 out:
 806         btrfs_free_path(path);
 807         return ret;
 808 }
 809
 810 /*
 811  * called with qgroup_lock held
 812  */
 813 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 814                                   struct btrfs_root *root)
 815 {
 816         struct btrfs_path *path;
 817         struct btrfs_key key;
 818         struct extent_buffer *leaf = NULL;
 819         int ret;
 820         int nr = 0;
 821
 822         path = btrfs_alloc_path();
 823         if (!path)
 824                 return -ENOMEM;
 825
 826         path->leave_spinning = 1;
 827
 828         key.objectid = 0;
 829         key.offset = 0;
 830         key.type = 0;
 831
 832         while (1) {
 833                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 834                 if (ret < 0)
 835                         goto out;
 836                 leaf = path->nodes[0];
 837                 nr = btrfs_header_nritems(leaf);
 838                 if (!nr)
 839                         break;
 840                 /*
 841                  * delete the leaf one by one
 842                  * since the whole tree is going
 843                  * to be deleted.
 844                  */
 845                 path->slots[0] = 0;
 846                 ret = btrfs_del_items(trans, root, path, 0, nr);
 847                 if (ret)
 848                         goto out;
 849
 850                 btrfs_release_path(path);
 851         }
 852         ret = 0;
 853 out:
 854         set_bit(BTRFS_FS_QUOTA_DISABLING, &root->fs_info->flags);
 855         btrfs_free_path(path);
 856         return ret;
 857 }
 858
 859 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 860                        struct btrfs_fs_info *fs_info)
 861 {
 862         struct btrfs_root *quota_root;
 863         struct btrfs_root *tree_root = fs_info->tree_root;
 864         struct btrfs_path *path = NULL;
 865         struct btrfs_qgroup_status_item *ptr;
 866         struct extent_buffer *leaf;
 867         struct btrfs_key key;
 868         struct btrfs_key found_key;
 869         struct btrfs_qgroup *qgroup = NULL;
 870         int ret = 0;
 871         int slot;
 872
 873         mutex_lock(&fs_info->qgroup_ioctl_lock);
 874         if (fs_info->quota_root) {
 875                 set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
 876                 goto out;
 877         }
 878
 879         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 880         if (!fs_info->qgroup_ulist) {
 881                 ret = -ENOMEM;
 882                 goto out;
 883         }
 884
 885         /*
 886          * initially create the quota tree
 887          */
 888         quota_root = btrfs_create_tree(trans, fs_info,
 889                                        BTRFS_QUOTA_TREE_OBJECTID);
 890         if (IS_ERR(quota_root)) {
 891                 ret =  PTR_ERR(quota_root);
 892                 goto out;
 893         }
 894
 895         path = btrfs_alloc_path();
 896         if (!path) {
 897                 ret = -ENOMEM;
 898                 goto out_free_root;
 899         }
 900
 901         key.objectid = 0;
 902         key.type = BTRFS_QGROUP_STATUS_KEY;
 903         key.offset = 0;
 904
 905         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 906                                       sizeof(*ptr));
 907         if (ret)
 908                 goto out_free_path;
 909
 910         leaf = path->nodes[0];
 911         ptr = btrfs_item_ptr(leaf, path->slots[0],
 912                                  struct btrfs_qgroup_status_item);
 913         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 914         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 915         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 916                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 917         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 918         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 919
 920         btrfs_mark_buffer_dirty(leaf);
 921
 922         key.objectid = 0;
 923         key.type = BTRFS_ROOT_REF_KEY;
 924         key.offset = 0;
 925
 926         btrfs_release_path(path);
 927         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 928         if (ret > 0)
 929                 goto out_add_root;
 930         if (ret < 0)
 931                 goto out_free_path;
 932
 933
 934         while (1) {
 935                 slot = path->slots[0];
 936                 leaf = path->nodes[0];
 937                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 938
 939                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 940                         ret = add_qgroup_item(trans, quota_root,
 941                                               found_key.offset);
 942                         if (ret)
 943                                 goto out_free_path;
 944
 945                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 946                         if (IS_ERR(qgroup)) {
 947                                 ret = PTR_ERR(qgroup);
 948                                 goto out_free_path;
 949                         }
 950                 }
 951                 ret = btrfs_next_item(tree_root, path);
 952                 if (ret < 0)
 953                         goto out_free_path;
 954                 if (ret)
 955                         break;
 956         }
 957
 958 out_add_root:
 959         btrfs_release_path(path);
 960         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
 961         if (ret)
 962                 goto out_free_path;
 963
 964         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
 965         if (IS_ERR(qgroup)) {
 966                 ret = PTR_ERR(qgroup);
 967                 goto out_free_path;
 968         }
 969         spin_lock(&fs_info->qgroup_lock);
 970         fs_info->quota_root = quota_root;
 971         set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
 972         spin_unlock(&fs_info->qgroup_lock);
 973 out_free_path:
 974         btrfs_free_path(path);
 975 out_free_root:
 976         if (ret) {
 977                 free_extent_buffer(quota_root->node);
 978                 free_extent_buffer(quota_root->commit_root);
 979                 kfree(quota_root);
 980         }
 981 out:
 982         if (ret) {
 983                 ulist_free(fs_info->qgroup_ulist);
 984                 fs_info->qgroup_ulist = NULL;
 985         }
 986         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 987         return ret;
 988 }
 989
 990 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
 991                         struct btrfs_fs_info *fs_info)
 992 {
 993         struct btrfs_root *tree_root = fs_info->tree_root;
 994         struct btrfs_root *quota_root;
 995         int ret = 0;
 996
 997         mutex_lock(&fs_info->qgroup_ioctl_lock);
 998         if (!fs_info->quota_root)
 999                 goto out;
1000         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1001         set_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags);
1002         btrfs_qgroup_wait_for_completion(fs_info, false);
1003         spin_lock(&fs_info->qgroup_lock);
1004         quota_root = fs_info->quota_root;
1005         fs_info->quota_root = NULL;
1006         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1007         spin_unlock(&fs_info->qgroup_lock);
1008
1009         btrfs_free_qgroup_config(fs_info);
1010
1011         ret = btrfs_clean_quota_tree(trans, quota_root);
1012         if (ret)
1013                 goto out;
1014
1015         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
1016         if (ret)
1017                 goto out;
1018
1019         list_del(&quota_root->dirty_list);
1020
1021         btrfs_tree_lock(quota_root->node);
1022         clean_tree_block(fs_info, quota_root->node);
1023         btrfs_tree_unlock(quota_root->node);
1024         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1025
1026         free_extent_buffer(quota_root->node);
1027         free_extent_buffer(quota_root->commit_root);
1028         kfree(quota_root);
1029 out:
1030         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1031         return ret;
1032 }
1033
1034 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1035                          struct btrfs_qgroup *qgroup)
1036 {
1037         if (list_empty(&qgroup->dirty))
1038                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1039 }
1040
1041 static void report_reserved_underflow(struct btrfs_fs_info *fs_info,
1042                                       struct btrfs_qgroup *qgroup,
1043                                       u64 num_bytes)
1044 {
1045         btrfs_warn(fs_info,
1046                 "qgroup %llu reserved space underflow, have: %llu, to free: %llu",
1047                 qgroup->qgroupid, qgroup->reserved, num_bytes);
1048         qgroup->reserved = 0;
1049 }
1050 /*
1051  * The easy accounting, if we are adding/removing the only ref for an extent
1052  * then this qgroup and all of the parent qgroups get their reference and
1053  * exclusive counts adjusted.
1054  *
1055  * Caller should hold fs_info->qgroup_lock.
1056  */
1057 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1058                                     struct ulist *tmp, u64 ref_root,
1059                                     u64 num_bytes, int sign)
1060 {
1061         struct btrfs_qgroup *qgroup;
1062         struct btrfs_qgroup_list *glist;
1063         struct ulist_node *unode;
1064         struct ulist_iterator uiter;
1065         int ret = 0;
1066
1067         qgroup = find_qgroup_rb(fs_info, ref_root);
1068         if (!qgroup)
1069                 goto out;
1070
1071         qgroup->rfer += sign * num_bytes;
1072         qgroup->rfer_cmpr += sign * num_bytes;
1073
1074         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1075         qgroup->excl += sign * num_bytes;
1076         qgroup->excl_cmpr += sign * num_bytes;
1077         if (sign > 0) {
1078                 if (WARN_ON(qgroup->reserved < num_bytes))
1079                         report_reserved_underflow(fs_info, qgroup, num_bytes);
1080                 else
1081                         qgroup->reserved -= num_bytes;
1082         }
1083
1084         qgroup_dirty(fs_info, qgroup);
1085
1086         /* Get all of the parent groups that contain this qgroup */
1087         list_for_each_entry(glist, &qgroup->groups, next_group) {
1088                 ret = ulist_add(tmp, glist->group->qgroupid,
1089                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1090                 if (ret < 0)
1091                         goto out;
1092         }
1093
1094         /* Iterate all of the parents and adjust their reference counts */
1095         ULIST_ITER_INIT(&uiter);
1096         while ((unode = ulist_next(tmp, &uiter))) {
1097                 qgroup = unode_aux_to_qgroup(unode);
1098                 qgroup->rfer += sign * num_bytes;
1099                 qgroup->rfer_cmpr += sign * num_bytes;
1100                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1101                 qgroup->excl += sign * num_bytes;
1102                 if (sign > 0) {
1103                         if (WARN_ON(qgroup->reserved < num_bytes))
1104                                 report_reserved_underflow(fs_info, qgroup,
1105                                                           num_bytes);
1106                         else
1107                                 qgroup->reserved -= num_bytes;
1108                 }
1109                 qgroup->excl_cmpr += sign * num_bytes;
1110                 qgroup_dirty(fs_info, qgroup);
1111
1112                 /* Add any parents of the parents */
1113                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1114                         ret = ulist_add(tmp, glist->group->qgroupid,
1115                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1116                         if (ret < 0)
1117                                 goto out;
1118                 }
1119         }
1120         ret = 0;
1121 out:
1122         return ret;
1123 }
1124
1125
1126 /*
1127  * Quick path for updating qgroup with only excl refs.
1128  *
1129  * In that case, just update all parent will be enough.
1130  * Or we needs to do a full rescan.
1131  * Caller should also hold fs_info->qgroup_lock.
1132  *
1133  * Return 0 for quick update, return >0 for need to full rescan
1134  * and mark INCONSISTENT flag.
1135  * Return < 0 for other error.
1136  */
1137 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1138                                    struct ulist *tmp, u64 src, u64 dst,
1139                                    int sign)
1140 {
1141         struct btrfs_qgroup *qgroup;
1142         int ret = 1;
1143         int err = 0;
1144
1145         qgroup = find_qgroup_rb(fs_info, src);
1146         if (!qgroup)
1147                 goto out;
1148         if (qgroup->excl == qgroup->rfer) {
1149                 ret = 0;
1150                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1151                                                qgroup->excl, sign);
1152                 if (err < 0) {
1153                         ret = err;
1154                         goto out;
1155                 }
1156         }
1157 out:
1158         if (ret)
1159                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1160         return ret;
1161 }
1162
1163 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1164                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1165 {
1166         struct btrfs_root *quota_root;
1167         struct btrfs_qgroup *parent;
1168         struct btrfs_qgroup *member;
1169         struct btrfs_qgroup_list *list;
1170         struct ulist *tmp;
1171         int ret = 0;
1172
1173         /* Check the level of src and dst first */
1174         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1175                 return -EINVAL;
1176
1177         tmp = ulist_alloc(GFP_KERNEL);
1178         if (!tmp)
1179                 return -ENOMEM;
1180
1181         mutex_lock(&fs_info->qgroup_ioctl_lock);
1182         quota_root = fs_info->quota_root;
1183         if (!quota_root) {
1184                 ret = -EINVAL;
1185                 goto out;
1186         }
1187         member = find_qgroup_rb(fs_info, src);
1188         parent = find_qgroup_rb(fs_info, dst);
1189         if (!member || !parent) {
1190                 ret = -EINVAL;
1191                 goto out;
1192         }
1193
1194         /* check if such qgroup relation exist firstly */
1195         list_for_each_entry(list, &member->groups, next_group) {
1196                 if (list->group == parent) {
1197                         ret = -EEXIST;
1198                         goto out;
1199                 }
1200         }
1201
1202         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1203         if (ret)
1204                 goto out;
1205
1206         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1207         if (ret) {
1208                 del_qgroup_relation_item(trans, quota_root, src, dst);
1209                 goto out;
1210         }
1211
1212         spin_lock(&fs_info->qgroup_lock);
1213         ret = add_relation_rb(fs_info, src, dst);
1214         if (ret < 0) {
1215                 spin_unlock(&fs_info->qgroup_lock);
1216                 goto out;
1217         }
1218         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1219         spin_unlock(&fs_info->qgroup_lock);
1220 out:
1221         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1222         ulist_free(tmp);
1223         return ret;
1224 }
1225
1226 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1227                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1228 {
1229         struct btrfs_root *quota_root;
1230         struct btrfs_qgroup *parent;
1231         struct btrfs_qgroup *member;
1232         struct btrfs_qgroup_list *list;
1233         struct ulist *tmp;
1234         int ret = 0;
1235         int err;
1236
1237         tmp = ulist_alloc(GFP_KERNEL);
1238         if (!tmp)
1239                 return -ENOMEM;
1240
1241         quota_root = fs_info->quota_root;
1242         if (!quota_root) {
1243                 ret = -EINVAL;
1244                 goto out;
1245         }
1246
1247         member = find_qgroup_rb(fs_info, src);
1248         parent = find_qgroup_rb(fs_info, dst);
1249         if (!member || !parent) {
1250                 ret = -EINVAL;
1251                 goto out;
1252         }
1253
1254         /* check if such qgroup relation exist firstly */
1255         list_for_each_entry(list, &member->groups, next_group) {
1256                 if (list->group == parent)
1257                         goto exist;
1258         }
1259         ret = -ENOENT;
1260         goto out;
1261 exist:
1262         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1263         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1264         if (err && !ret)
1265                 ret = err;
1266
1267         spin_lock(&fs_info->qgroup_lock);
1268         del_relation_rb(fs_info, src, dst);
1269         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1270         spin_unlock(&fs_info->qgroup_lock);
1271 out:
1272         ulist_free(tmp);
1273         return ret;
1274 }
1275
1276 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1277                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1278 {
1279         int ret = 0;
1280
1281         mutex_lock(&fs_info->qgroup_ioctl_lock);
1282         ret = __del_qgroup_relation(trans, fs_info, src, dst);
1283         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1284
1285         return ret;
1286 }
1287
1288 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1289                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1290 {
1291         struct btrfs_root *quota_root;
1292         struct btrfs_qgroup *qgroup;
1293         int ret = 0;
1294
1295         mutex_lock(&fs_info->qgroup_ioctl_lock);
1296         quota_root = fs_info->quota_root;
1297         if (!quota_root) {
1298                 ret = -EINVAL;
1299                 goto out;
1300         }
1301         qgroup = find_qgroup_rb(fs_info, qgroupid);
1302         if (qgroup) {
1303                 ret = -EEXIST;
1304                 goto out;
1305         }
1306
1307         ret = add_qgroup_item(trans, quota_root, qgroupid);
1308         if (ret)
1309                 goto out;
1310
1311         spin_lock(&fs_info->qgroup_lock);
1312         qgroup = add_qgroup_rb(fs_info, qgroupid);
1313         spin_unlock(&fs_info->qgroup_lock);
1314
1315         if (IS_ERR(qgroup))
1316                 ret = PTR_ERR(qgroup);
1317 out:
1318         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1319         return ret;
1320 }
1321
1322 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1323                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1324 {
1325         struct btrfs_root *quota_root;
1326         struct btrfs_qgroup *qgroup;
1327         struct btrfs_qgroup_list *list;
1328         int ret = 0;
1329
1330         mutex_lock(&fs_info->qgroup_ioctl_lock);
1331         quota_root = fs_info->quota_root;
1332         if (!quota_root) {
1333                 ret = -EINVAL;
1334                 goto out;
1335         }
1336
1337         qgroup = find_qgroup_rb(fs_info, qgroupid);
1338         if (!qgroup) {
1339                 ret = -ENOENT;
1340                 goto out;
1341         } else {
1342                 /* check if there are no children of this qgroup */
1343                 if (!list_empty(&qgroup->members)) {
1344                         ret = -EBUSY;
1345                         goto out;
1346                 }
1347         }
1348         ret = del_qgroup_item(trans, quota_root, qgroupid);
1349
1350         while (!list_empty(&qgroup->groups)) {
1351                 list = list_first_entry(&qgroup->groups,
1352                                         struct btrfs_qgroup_list, next_group);
1353                 ret = __del_qgroup_relation(trans, fs_info,
1354                                            qgroupid,
1355                                            list->group->qgroupid);
1356                 if (ret)
1357                         goto out;
1358         }
1359
1360         spin_lock(&fs_info->qgroup_lock);
1361         del_qgroup_rb(fs_info, qgroupid);
1362         spin_unlock(&fs_info->qgroup_lock);
1363 out:
1364         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1365         return ret;
1366 }
1367
1368 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1369                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1370                        struct btrfs_qgroup_limit *limit)
1371 {
1372         struct btrfs_root *quota_root;
1373         struct btrfs_qgroup *qgroup;
1374         int ret = 0;
1375         /* Sometimes we would want to clear the limit on this qgroup.
1376          * To meet this requirement, we treat the -1 as a special value
1377          * which tell kernel to clear the limit on this qgroup.
1378          */
1379         const u64 CLEAR_VALUE = -1;
1380
1381         mutex_lock(&fs_info->qgroup_ioctl_lock);
1382         quota_root = fs_info->quota_root;
1383         if (!quota_root) {
1384                 ret = -EINVAL;
1385                 goto out;
1386         }
1387
1388         qgroup = find_qgroup_rb(fs_info, qgroupid);
1389         if (!qgroup) {
1390                 ret = -ENOENT;
1391                 goto out;
1392         }
1393
1394         spin_lock(&fs_info->qgroup_lock);
1395         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1396                 if (limit->max_rfer == CLEAR_VALUE) {
1397                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1398                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1399                         qgroup->max_rfer = 0;
1400                 } else {
1401                         qgroup->max_rfer = limit->max_rfer;
1402                 }
1403         }
1404         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1405                 if (limit->max_excl == CLEAR_VALUE) {
1406                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1407                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1408                         qgroup->max_excl = 0;
1409                 } else {
1410                         qgroup->max_excl = limit->max_excl;
1411                 }
1412         }
1413         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1414                 if (limit->rsv_rfer == CLEAR_VALUE) {
1415                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1416                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1417                         qgroup->rsv_rfer = 0;
1418                 } else {
1419                         qgroup->rsv_rfer = limit->rsv_rfer;
1420                 }
1421         }
1422         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1423                 if (limit->rsv_excl == CLEAR_VALUE) {
1424                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1425                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1426                         qgroup->rsv_excl = 0;
1427                 } else {
1428                         qgroup->rsv_excl = limit->rsv_excl;
1429                 }
1430         }
1431         qgroup->lim_flags |= limit->flags;
1432
1433         spin_unlock(&fs_info->qgroup_lock);
1434
1435         ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1436         if (ret) {
1437                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1438                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1439                        qgroupid);
1440         }
1441
1442 out:
1443         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1444         return ret;
1445 }
1446
1447 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1448                                          struct btrfs_fs_info *fs_info)
1449 {
1450         struct btrfs_qgroup_extent_record *record;
1451         struct btrfs_delayed_ref_root *delayed_refs;
1452         struct rb_node *node;
1453         u64 qgroup_to_skip;
1454         int ret = 0;
1455
1456         delayed_refs = &trans->transaction->delayed_refs;
1457         qgroup_to_skip = delayed_refs->qgroup_to_skip;
1458
1459         /*
1460          * No need to do lock, since this function will only be called in
1461          * btrfs_commit_transaction().
1462          */
1463         node = rb_first(&delayed_refs->dirty_extent_root);
1464         while (node) {
1465                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1466                                   node);
1467                 if (WARN_ON(!record->old_roots))
1468                         ret = btrfs_find_all_roots(NULL, fs_info,
1469                                         record->bytenr, 0, &record->old_roots);
1470                 if (ret < 0)
1471                         break;
1472                 if (qgroup_to_skip)
1473                         ulist_del(record->old_roots, qgroup_to_skip, 0);
1474                 node = rb_next(node);
1475         }
1476         return ret;
1477 }
1478
1479 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1480                                 struct btrfs_delayed_ref_root *delayed_refs,
1481                                 struct btrfs_qgroup_extent_record *record)
1482 {
1483         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1484         struct rb_node *parent_node = NULL;
1485         struct btrfs_qgroup_extent_record *entry;
1486         u64 bytenr = record->bytenr;
1487
1488         assert_spin_locked(&delayed_refs->lock);
1489         trace_btrfs_qgroup_trace_extent(fs_info, record);
1490
1491         while (*p) {
1492                 parent_node = *p;
1493                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1494                                  node);
1495                 if (bytenr < entry->bytenr)
1496                         p = &(*p)->rb_left;
1497                 else if (bytenr > entry->bytenr)
1498                         p = &(*p)->rb_right;
1499                 else
1500                         return 1;
1501         }
1502
1503         rb_link_node(&record->node, parent_node, p);
1504         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1505         return 0;
1506 }
1507
1508 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1509                                    struct btrfs_qgroup_extent_record *qrecord)
1510 {
1511         struct ulist *old_root;
1512         u64 bytenr = qrecord->bytenr;
1513         int ret;
1514
1515         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root);
1516         if (ret < 0)
1517                 return ret;
1518
1519         /*
1520          * Here we don't need to get the lock of
1521          * trans->transaction->delayed_refs, since inserted qrecord won't
1522          * be deleted, only qrecord->node may be modified (new qrecord insert)
1523          *
1524          * So modifying qrecord->old_roots is safe here
1525          */
1526         qrecord->old_roots = old_root;
1527         return 0;
1528 }
1529
1530 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1531                 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1532                 gfp_t gfp_flag)
1533 {
1534         struct btrfs_qgroup_extent_record *record;
1535         struct btrfs_delayed_ref_root *delayed_refs;
1536         int ret;
1537
1538         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1539             || bytenr == 0 || num_bytes == 0)
1540                 return 0;
1541         if (WARN_ON(trans == NULL))
1542                 return -EINVAL;
1543         record = kmalloc(sizeof(*record), gfp_flag);
1544         if (!record)
1545                 return -ENOMEM;
1546
1547         delayed_refs = &trans->transaction->delayed_refs;
1548         record->bytenr = bytenr;
1549         record->num_bytes = num_bytes;
1550         record->old_roots = NULL;
1551
1552         spin_lock(&delayed_refs->lock);
1553         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1554         spin_unlock(&delayed_refs->lock);
1555         if (ret > 0) {
1556                 kfree(record);
1557                 return 0;
1558         }
1559         return btrfs_qgroup_trace_extent_post(fs_info, record);
1560 }
1561
1562 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1563                                   struct btrfs_fs_info *fs_info,
1564                                   struct extent_buffer *eb)
1565 {
1566         int nr = btrfs_header_nritems(eb);
1567         int i, extent_type, ret;
1568         struct btrfs_key key;
1569         struct btrfs_file_extent_item *fi;
1570         u64 bytenr, num_bytes;
1571
1572         /* We can be called directly from walk_up_proc() */
1573         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1574                 return 0;
1575
1576         for (i = 0; i < nr; i++) {
1577                 btrfs_item_key_to_cpu(eb, &key, i);
1578
1579                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1580                         continue;
1581
1582                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1583                 /* filter out non qgroup-accountable extents  */
1584                 extent_type = btrfs_file_extent_type(eb, fi);
1585
1586                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1587                         continue;
1588
1589                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1590                 if (!bytenr)
1591                         continue;
1592
1593                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1594
1595                 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1596                                                 num_bytes, GFP_NOFS);
1597                 if (ret)
1598                         return ret;
1599         }
1600         return 0;
1601 }
1602
1603 /*
1604  * Walk up the tree from the bottom, freeing leaves and any interior
1605  * nodes which have had all slots visited. If a node (leaf or
1606  * interior) is freed, the node above it will have it's slot
1607  * incremented. The root node will never be freed.
1608  *
1609  * At the end of this function, we should have a path which has all
1610  * slots incremented to the next position for a search. If we need to
1611  * read a new node it will be NULL and the node above it will have the
1612  * correct slot selected for a later read.
1613  *
1614  * If we increment the root nodes slot counter past the number of
1615  * elements, 1 is returned to signal completion of the search.
1616  */
1617 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1618 {
1619         int level = 0;
1620         int nr, slot;
1621         struct extent_buffer *eb;
1622
1623         if (root_level == 0)
1624                 return 1;
1625
1626         while (level <= root_level) {
1627                 eb = path->nodes[level];
1628                 nr = btrfs_header_nritems(eb);
1629                 path->slots[level]++;
1630                 slot = path->slots[level];
1631                 if (slot >= nr || level == 0) {
1632                         /*
1633                          * Don't free the root -  we will detect this
1634                          * condition after our loop and return a
1635                          * positive value for caller to stop walking the tree.
1636                          */
1637                         if (level != root_level) {
1638                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1639                                 path->locks[level] = 0;
1640
1641                                 free_extent_buffer(eb);
1642                                 path->nodes[level] = NULL;
1643                                 path->slots[level] = 0;
1644                         }
1645                 } else {
1646                         /*
1647                          * We have a valid slot to walk back down
1648                          * from. Stop here so caller can process these
1649                          * new nodes.
1650                          */
1651                         break;
1652                 }
1653
1654                 level++;
1655         }
1656
1657         eb = path->nodes[root_level];
1658         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1659                 return 1;
1660
1661         return 0;
1662 }
1663
1664 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1665                                struct btrfs_root *root,
1666                                struct extent_buffer *root_eb,
1667                                u64 root_gen, int root_level)
1668 {
1669         struct btrfs_fs_info *fs_info = root->fs_info;
1670         int ret = 0;
1671         int level;
1672         struct extent_buffer *eb = root_eb;
1673         struct btrfs_path *path = NULL;
1674
1675         BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
1676         BUG_ON(root_eb == NULL);
1677
1678         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1679                 return 0;
1680
1681         if (!extent_buffer_uptodate(root_eb)) {
1682                 ret = btrfs_read_buffer(root_eb, root_gen);
1683                 if (ret)
1684                         goto out;
1685         }
1686
1687         if (root_level == 0) {
1688                 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1689                 goto out;
1690         }
1691
1692         path = btrfs_alloc_path();
1693         if (!path)
1694                 return -ENOMEM;
1695
1696         /*
1697          * Walk down the tree.  Missing extent blocks are filled in as
1698          * we go. Metadata is accounted every time we read a new
1699          * extent block.
1700          *
1701          * When we reach a leaf, we account for file extent items in it,
1702          * walk back up the tree (adjusting slot pointers as we go)
1703          * and restart the search process.
1704          */
1705         extent_buffer_get(root_eb); /* For path */
1706         path->nodes[root_level] = root_eb;
1707         path->slots[root_level] = 0;
1708         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1709 walk_down:
1710         level = root_level;
1711         while (level >= 0) {
1712                 if (path->nodes[level] == NULL) {
1713                         int parent_slot;
1714                         u64 child_gen;
1715                         u64 child_bytenr;
1716
1717                         /*
1718                          * We need to get child blockptr/gen from parent before
1719                          * we can read it.
1720                           */
1721                         eb = path->nodes[level + 1];
1722                         parent_slot = path->slots[level + 1];
1723                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1724                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1725
1726                         eb = read_tree_block(fs_info, child_bytenr, child_gen);
1727                         if (IS_ERR(eb)) {
1728                                 ret = PTR_ERR(eb);
1729                                 goto out;
1730                         } else if (!extent_buffer_uptodate(eb)) {
1731                                 free_extent_buffer(eb);
1732                                 ret = -EIO;
1733                                 goto out;
1734                         }
1735
1736                         path->nodes[level] = eb;
1737                         path->slots[level] = 0;
1738
1739                         btrfs_tree_read_lock(eb);
1740                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1741                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1742
1743                         ret = btrfs_qgroup_trace_extent(trans, fs_info,
1744                                                         child_bytenr,
1745                                                         fs_info->nodesize,
1746                                                         GFP_NOFS);
1747                         if (ret)
1748                                 goto out;
1749                 }
1750
1751                 if (level == 0) {
1752                         ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1753                                                            path->nodes[level]);
1754                         if (ret)
1755                                 goto out;
1756
1757                         /* Nonzero return here means we completed our search */
1758                         ret = adjust_slots_upwards(path, root_level);
1759                         if (ret)
1760                                 break;
1761
1762                         /* Restart search with new slots */
1763                         goto walk_down;
1764                 }
1765
1766                 level--;
1767         }
1768
1769         ret = 0;
1770 out:
1771         btrfs_free_path(path);
1772
1773         return ret;
1774 }
1775
1776 #define UPDATE_NEW      0
1777 #define UPDATE_OLD      1
1778 /*
1779  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1780  */
1781 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1782                                 struct ulist *roots, struct ulist *tmp,
1783                                 struct ulist *qgroups, u64 seq, int update_old)
1784 {
1785         struct ulist_node *unode;
1786         struct ulist_iterator uiter;
1787         struct ulist_node *tmp_unode;
1788         struct ulist_iterator tmp_uiter;
1789         struct btrfs_qgroup *qg;
1790         int ret = 0;
1791
1792         if (!roots)
1793                 return 0;
1794         ULIST_ITER_INIT(&uiter);
1795         while ((unode = ulist_next(roots, &uiter))) {
1796                 qg = find_qgroup_rb(fs_info, unode->val);
1797                 if (!qg)
1798                         continue;
1799
1800                 ulist_reinit(tmp);
1801                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1802                                 GFP_ATOMIC);
1803                 if (ret < 0)
1804                         return ret;
1805                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1806                 if (ret < 0)
1807                         return ret;
1808                 ULIST_ITER_INIT(&tmp_uiter);
1809                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1810                         struct btrfs_qgroup_list *glist;
1811
1812                         qg = unode_aux_to_qgroup(tmp_unode);
1813                         if (update_old)
1814                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1815                         else
1816                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1817                         list_for_each_entry(glist, &qg->groups, next_group) {
1818                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1819                                                 qgroup_to_aux(glist->group),
1820                                                 GFP_ATOMIC);
1821                                 if (ret < 0)
1822                                         return ret;
1823                                 ret = ulist_add(tmp, glist->group->qgroupid,
1824                                                 qgroup_to_aux(glist->group),
1825                                                 GFP_ATOMIC);
1826                                 if (ret < 0)
1827                                         return ret;
1828                         }
1829                 }
1830         }
1831         return 0;
1832 }
1833
1834 /*
1835  * Update qgroup rfer/excl counters.
1836  * Rfer update is easy, codes can explain themselves.
1837  *
1838  * Excl update is tricky, the update is split into 2 part.
1839  * Part 1: Possible exclusive <-> sharing detect:
1840  *      |       A       |       !A      |
1841  *  -------------------------------------
1842  *  B   |       *       |       -       |
1843  *  -------------------------------------
1844  *  !B  |       +       |       **      |
1845  *  -------------------------------------
1846  *
1847  * Conditions:
1848  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1849  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1850  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1851  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1852  *
1853  * Results:
1854  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1855  * *: Definitely not changed.           **: Possible unchanged.
1856  *
1857  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1858  *
1859  * To make the logic clear, we first use condition A and B to split
1860  * combination into 4 results.
1861  *
1862  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1863  * only on variant maybe 0.
1864  *
1865  * Lastly, check result **, since there are 2 variants maybe 0, split them
1866  * again(2x2).
1867  * But this time we don't need to consider other things, the codes and logic
1868  * is easy to understand now.
1869  */
1870 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1871                                   struct ulist *qgroups,
1872                                   u64 nr_old_roots,
1873                                   u64 nr_new_roots,
1874                                   u64 num_bytes, u64 seq)
1875 {
1876         struct ulist_node *unode;
1877         struct ulist_iterator uiter;
1878         struct btrfs_qgroup *qg;
1879         u64 cur_new_count, cur_old_count;
1880
1881         ULIST_ITER_INIT(&uiter);
1882         while ((unode = ulist_next(qgroups, &uiter))) {
1883                 bool dirty = false;
1884
1885                 qg = unode_aux_to_qgroup(unode);
1886                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1887                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1888
1889                 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1890                                              cur_old_count, cur_new_count);
1891
1892                 /* Rfer update part */
1893                 if (cur_old_count == 0 && cur_new_count > 0) {
1894                         qg->rfer += num_bytes;
1895                         qg->rfer_cmpr += num_bytes;
1896                         dirty = true;
1897                 }
1898                 if (cur_old_count > 0 && cur_new_count == 0) {
1899                         qg->rfer -= num_bytes;
1900                         qg->rfer_cmpr -= num_bytes;
1901                         dirty = true;
1902                 }
1903
1904                 /* Excl update part */
1905                 /* Exclusive/none -> shared case */
1906                 if (cur_old_count == nr_old_roots &&
1907                     cur_new_count < nr_new_roots) {
1908                         /* Exclusive -> shared */
1909                         if (cur_old_count != 0) {
1910                                 qg->excl -= num_bytes;
1911                                 qg->excl_cmpr -= num_bytes;
1912                                 dirty = true;
1913                         }
1914                 }
1915
1916                 /* Shared -> exclusive/none case */
1917                 if (cur_old_count < nr_old_roots &&
1918                     cur_new_count == nr_new_roots) {
1919                         /* Shared->exclusive */
1920                         if (cur_new_count != 0) {
1921                                 qg->excl += num_bytes;
1922                                 qg->excl_cmpr += num_bytes;
1923                                 dirty = true;
1924                         }
1925                 }
1926
1927                 /* Exclusive/none -> exclusive/none case */
1928                 if (cur_old_count == nr_old_roots &&
1929                     cur_new_count == nr_new_roots) {
1930                         if (cur_old_count == 0) {
1931                                 /* None -> exclusive/none */
1932
1933                                 if (cur_new_count != 0) {
1934                                         /* None -> exclusive */
1935                                         qg->excl += num_bytes;
1936                                         qg->excl_cmpr += num_bytes;
1937                                         dirty = true;
1938                                 }
1939                                 /* None -> none, nothing changed */
1940                         } else {
1941                                 /* Exclusive -> exclusive/none */
1942
1943                                 if (cur_new_count == 0) {
1944                                         /* Exclusive -> none */
1945                                         qg->excl -= num_bytes;
1946                                         qg->excl_cmpr -= num_bytes;
1947                                         dirty = true;
1948                                 }
1949                                 /* Exclusive -> exclusive, nothing changed */
1950                         }
1951                 }
1952
1953                 if (dirty)
1954                         qgroup_dirty(fs_info, qg);
1955         }
1956         return 0;
1957 }
1958
1959 int
1960 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1961                             struct btrfs_fs_info *fs_info,
1962                             u64 bytenr, u64 num_bytes,
1963                             struct ulist *old_roots, struct ulist *new_roots)
1964 {
1965         struct ulist *qgroups = NULL;
1966         struct ulist *tmp = NULL;
1967         u64 seq;
1968         u64 nr_new_roots = 0;
1969         u64 nr_old_roots = 0;
1970         int ret = 0;
1971
1972         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1973                 return 0;
1974
1975         if (new_roots)
1976                 nr_new_roots = new_roots->nnodes;
1977         if (old_roots)
1978                 nr_old_roots = old_roots->nnodes;
1979
1980         BUG_ON(!fs_info->quota_root);
1981
1982         trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1983                                           nr_old_roots, nr_new_roots);
1984
1985         qgroups = ulist_alloc(GFP_NOFS);
1986         if (!qgroups) {
1987                 ret = -ENOMEM;
1988                 goto out_free;
1989         }
1990         tmp = ulist_alloc(GFP_NOFS);
1991         if (!tmp) {
1992                 ret = -ENOMEM;
1993                 goto out_free;
1994         }
1995
1996         mutex_lock(&fs_info->qgroup_rescan_lock);
1997         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1998                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1999                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2000                         ret = 0;
2001                         goto out_free;
2002                 }
2003         }
2004         mutex_unlock(&fs_info->qgroup_rescan_lock);
2005
2006         spin_lock(&fs_info->qgroup_lock);
2007         seq = fs_info->qgroup_seq;
2008
2009         /* Update old refcnts using old_roots */
2010         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2011                                    UPDATE_OLD);
2012         if (ret < 0)
2013                 goto out;
2014
2015         /* Update new refcnts using new_roots */
2016         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2017                                    UPDATE_NEW);
2018         if (ret < 0)
2019                 goto out;
2020
2021         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2022                                num_bytes, seq);
2023
2024         /*
2025          * Bump qgroup_seq to avoid seq overlap
2026          */
2027         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2028 out:
2029         spin_unlock(&fs_info->qgroup_lock);
2030 out_free:
2031         ulist_free(tmp);
2032         ulist_free(qgroups);
2033         ulist_free(old_roots);
2034         ulist_free(new_roots);
2035         return ret;
2036 }
2037
2038 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
2039                                  struct btrfs_fs_info *fs_info)
2040 {
2041         struct btrfs_qgroup_extent_record *record;
2042         struct btrfs_delayed_ref_root *delayed_refs;
2043         struct ulist *new_roots = NULL;
2044         struct rb_node *node;
2045         u64 qgroup_to_skip;
2046         int ret = 0;
2047
2048         delayed_refs = &trans->transaction->delayed_refs;
2049         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2050         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2051                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2052                                   node);
2053
2054                 trace_btrfs_qgroup_account_extents(fs_info, record);
2055
2056                 if (!ret) {
2057                         /*
2058                          * Use (u64)-1 as time_seq to do special search, which
2059                          * doesn't lock tree or delayed_refs and search current
2060                          * root. It's safe inside commit_transaction().
2061                          */
2062                         ret = btrfs_find_all_roots(trans, fs_info,
2063                                         record->bytenr, (u64)-1, &new_roots);
2064                         if (ret < 0)
2065                                 goto cleanup;
2066                         if (qgroup_to_skip)
2067                                 ulist_del(new_roots, qgroup_to_skip, 0);
2068                         ret = btrfs_qgroup_account_extent(trans, fs_info,
2069                                         record->bytenr, record->num_bytes,
2070                                         record->old_roots, new_roots);
2071                         record->old_roots = NULL;
2072                         new_roots = NULL;
2073                 }
2074 cleanup:
2075                 ulist_free(record->old_roots);
2076                 ulist_free(new_roots);
2077                 new_roots = NULL;
2078                 rb_erase(node, &delayed_refs->dirty_extent_root);
2079                 kfree(record);
2080
2081         }
2082         return ret;
2083 }
2084
2085 /*
2086  * called from commit_transaction. Writes all changed qgroups to disk.
2087  */
2088 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2089                       struct btrfs_fs_info *fs_info)
2090 {
2091         struct btrfs_root *quota_root = fs_info->quota_root;
2092         int ret = 0;
2093         int start_rescan_worker = 0;
2094
2095         if (!quota_root)
2096                 goto out;
2097
2098         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
2099             test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2100                 start_rescan_worker = 1;
2101
2102         if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2103                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2104         if (test_and_clear_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags))
2105                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2106
2107         spin_lock(&fs_info->qgroup_lock);
2108         while (!list_empty(&fs_info->dirty_qgroups)) {
2109                 struct btrfs_qgroup *qgroup;
2110                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2111                                           struct btrfs_qgroup, dirty);
2112                 list_del_init(&qgroup->dirty);
2113                 spin_unlock(&fs_info->qgroup_lock);
2114                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2115                 if (ret)
2116                         fs_info->qgroup_flags |=
2117                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2118                 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2119                 if (ret)
2120                         fs_info->qgroup_flags |=
2121                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2122                 spin_lock(&fs_info->qgroup_lock);
2123         }
2124         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2125                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2126         else
2127                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2128         spin_unlock(&fs_info->qgroup_lock);
2129
2130         ret = update_qgroup_status_item(trans, fs_info, quota_root);
2131         if (ret)
2132                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2133
2134         if (!ret && start_rescan_worker) {
2135                 ret = qgroup_rescan_init(fs_info, 0, 1);
2136                 if (!ret) {
2137                         qgroup_rescan_zero_tracking(fs_info);
2138                         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2139                                          &fs_info->qgroup_rescan_work);
2140                 }
2141                 ret = 0;
2142         }
2143
2144 out:
2145
2146         return ret;
2147 }
2148
2149 /*
2150  * Copy the accounting information between qgroups. This is necessary
2151  * when a snapshot or a subvolume is created. Throwing an error will
2152  * cause a transaction abort so we take extra care here to only error
2153  * when a readonly fs is a reasonable outcome.
2154  */
2155 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2156                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2157                          struct btrfs_qgroup_inherit *inherit)
2158 {
2159         int ret = 0;
2160         int i;
2161         u64 *i_qgroups;
2162         struct btrfs_root *quota_root = fs_info->quota_root;
2163         struct btrfs_qgroup *srcgroup;
2164         struct btrfs_qgroup *dstgroup;
2165         u32 level_size = 0;
2166         u64 nums;
2167
2168         mutex_lock(&fs_info->qgroup_ioctl_lock);
2169         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2170                 goto out;
2171
2172         if (!quota_root) {
2173                 ret = -EINVAL;
2174                 goto out;
2175         }
2176
2177         if (inherit) {
2178                 i_qgroups = (u64 *)(inherit + 1);
2179                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2180                        2 * inherit->num_excl_copies;
2181                 for (i = 0; i < nums; ++i) {
2182                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2183
2184                         /*
2185                          * Zero out invalid groups so we can ignore
2186                          * them later.
2187                          */
2188                         if (!srcgroup ||
2189                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2190                                 *i_qgroups = 0ULL;
2191
2192                         ++i_qgroups;
2193                 }
2194         }
2195
2196         /*
2197          * create a tracking group for the subvol itself
2198          */
2199         ret = add_qgroup_item(trans, quota_root, objectid);
2200         if (ret)
2201                 goto out;
2202
2203         if (srcid) {
2204                 struct btrfs_root *srcroot;
2205                 struct btrfs_key srckey;
2206
2207                 srckey.objectid = srcid;
2208                 srckey.type = BTRFS_ROOT_ITEM_KEY;
2209                 srckey.offset = (u64)-1;
2210                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2211                 if (IS_ERR(srcroot)) {
2212                         ret = PTR_ERR(srcroot);
2213                         goto out;
2214                 }
2215
2216                 level_size = fs_info->nodesize;
2217         }
2218
2219         /*
2220          * add qgroup to all inherited groups
2221          */
2222         if (inherit) {
2223                 i_qgroups = (u64 *)(inherit + 1);
2224                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2225                         if (*i_qgroups == 0)
2226                                 continue;
2227                         ret = add_qgroup_relation_item(trans, quota_root,
2228                                                        objectid, *i_qgroups);
2229                         if (ret && ret != -EEXIST)
2230                                 goto out;
2231                         ret = add_qgroup_relation_item(trans, quota_root,
2232                                                        *i_qgroups, objectid);
2233                         if (ret && ret != -EEXIST)
2234                                 goto out;
2235                 }
2236                 ret = 0;
2237         }
2238
2239
2240         spin_lock(&fs_info->qgroup_lock);
2241
2242         dstgroup = add_qgroup_rb(fs_info, objectid);
2243         if (IS_ERR(dstgroup)) {
2244                 ret = PTR_ERR(dstgroup);
2245                 goto unlock;
2246         }
2247
2248         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2249                 dstgroup->lim_flags = inherit->lim.flags;
2250                 dstgroup->max_rfer = inherit->lim.max_rfer;
2251                 dstgroup->max_excl = inherit->lim.max_excl;
2252                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2253                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2254
2255                 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2256                 if (ret) {
2257                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2258                         btrfs_info(fs_info,
2259                                    "unable to update quota limit for %llu",
2260                                    dstgroup->qgroupid);
2261                         goto unlock;
2262                 }
2263         }
2264
2265         if (srcid) {
2266                 srcgroup = find_qgroup_rb(fs_info, srcid);
2267                 if (!srcgroup)
2268                         goto unlock;
2269
2270                 /*
2271                  * We call inherit after we clone the root in order to make sure
2272                  * our counts don't go crazy, so at this point the only
2273                  * difference between the two roots should be the root node.
2274                  */
2275                 dstgroup->rfer = srcgroup->rfer;
2276                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2277                 dstgroup->excl = level_size;
2278                 dstgroup->excl_cmpr = level_size;
2279                 srcgroup->excl = level_size;
2280                 srcgroup->excl_cmpr = level_size;
2281
2282                 /* inherit the limit info */
2283                 dstgroup->lim_flags = srcgroup->lim_flags;
2284                 dstgroup->max_rfer = srcgroup->max_rfer;
2285                 dstgroup->max_excl = srcgroup->max_excl;
2286                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2287                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2288
2289                 qgroup_dirty(fs_info, dstgroup);
2290                 qgroup_dirty(fs_info, srcgroup);
2291         }
2292
2293         if (!inherit)
2294                 goto unlock;
2295
2296         i_qgroups = (u64 *)(inherit + 1);
2297         for (i = 0; i < inherit->num_qgroups; ++i) {
2298                 if (*i_qgroups) {
2299                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2300                         if (ret)
2301                                 goto unlock;
2302                 }
2303                 ++i_qgroups;
2304         }
2305
2306         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2307                 struct btrfs_qgroup *src;
2308                 struct btrfs_qgroup *dst;
2309
2310                 if (!i_qgroups[0] || !i_qgroups[1])
2311                         continue;
2312
2313                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2314                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2315
2316                 if (!src || !dst) {
2317                         ret = -EINVAL;
2318                         goto unlock;
2319                 }
2320
2321                 dst->rfer = src->rfer - level_size;
2322                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2323         }
2324         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2325                 struct btrfs_qgroup *src;
2326                 struct btrfs_qgroup *dst;
2327
2328                 if (!i_qgroups[0] || !i_qgroups[1])
2329                         continue;
2330
2331                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2332                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2333
2334                 if (!src || !dst) {
2335                         ret = -EINVAL;
2336                         goto unlock;
2337                 }
2338
2339                 dst->excl = src->excl + level_size;
2340                 dst->excl_cmpr = src->excl_cmpr + level_size;
2341         }
2342
2343 unlock:
2344         spin_unlock(&fs_info->qgroup_lock);
2345 out:
2346         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2347         return ret;
2348 }
2349
2350 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2351 {
2352         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2353             qg->reserved + (s64)qg->rfer + num_bytes > qg->max_rfer)
2354                 return false;
2355
2356         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2357             qg->reserved + (s64)qg->excl + num_bytes > qg->max_excl)
2358                 return false;
2359
2360         return true;
2361 }
2362
2363 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce)
2364 {
2365         struct btrfs_root *quota_root;
2366         struct btrfs_qgroup *qgroup;
2367         struct btrfs_fs_info *fs_info = root->fs_info;
2368         u64 ref_root = root->root_key.objectid;
2369         int ret = 0;
2370         struct ulist_node *unode;
2371         struct ulist_iterator uiter;
2372
2373         if (!is_fstree(ref_root))
2374                 return 0;
2375
2376         if (num_bytes == 0)
2377                 return 0;
2378
2379         spin_lock(&fs_info->qgroup_lock);
2380         quota_root = fs_info->quota_root;
2381         if (!quota_root)
2382                 goto out;
2383
2384         qgroup = find_qgroup_rb(fs_info, ref_root);
2385         if (!qgroup)
2386                 goto out;
2387
2388         /*
2389          * in a first step, we check all affected qgroups if any limits would
2390          * be exceeded
2391          */
2392         ulist_reinit(fs_info->qgroup_ulist);
2393         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2394                         (uintptr_t)qgroup, GFP_ATOMIC);
2395         if (ret < 0)
2396                 goto out;
2397         ULIST_ITER_INIT(&uiter);
2398         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2399                 struct btrfs_qgroup *qg;
2400                 struct btrfs_qgroup_list *glist;
2401
2402                 qg = unode_aux_to_qgroup(unode);
2403
2404                 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2405                         ret = -EDQUOT;
2406                         goto out;
2407                 }
2408
2409                 list_for_each_entry(glist, &qg->groups, next_group) {
2410                         ret = ulist_add(fs_info->qgroup_ulist,
2411                                         glist->group->qgroupid,
2412                                         (uintptr_t)glist->group, GFP_ATOMIC);
2413                         if (ret < 0)
2414                                 goto out;
2415                 }
2416         }
2417         ret = 0;
2418         /*
2419          * no limits exceeded, now record the reservation into all qgroups
2420          */
2421         ULIST_ITER_INIT(&uiter);
2422         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2423                 struct btrfs_qgroup *qg;
2424
2425                 qg = unode_aux_to_qgroup(unode);
2426
2427                 qg->reserved += num_bytes;
2428         }
2429
2430 out:
2431         spin_unlock(&fs_info->qgroup_lock);
2432         return ret;
2433 }
2434
2435 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2436                                u64 ref_root, u64 num_bytes)
2437 {
2438         struct btrfs_root *quota_root;
2439         struct btrfs_qgroup *qgroup;
2440         struct ulist_node *unode;
2441         struct ulist_iterator uiter;
2442         int ret = 0;
2443
2444         if (!is_fstree(ref_root))
2445                 return;
2446
2447         if (num_bytes == 0)
2448                 return;
2449
2450         spin_lock(&fs_info->qgroup_lock);
2451
2452         quota_root = fs_info->quota_root;
2453         if (!quota_root)
2454                 goto out;
2455
2456         qgroup = find_qgroup_rb(fs_info, ref_root);
2457         if (!qgroup)
2458                 goto out;
2459
2460         ulist_reinit(fs_info->qgroup_ulist);
2461         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2462                         (uintptr_t)qgroup, GFP_ATOMIC);
2463         if (ret < 0)
2464                 goto out;
2465         ULIST_ITER_INIT(&uiter);
2466         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2467                 struct btrfs_qgroup *qg;
2468                 struct btrfs_qgroup_list *glist;
2469
2470                 qg = unode_aux_to_qgroup(unode);
2471
2472                 if (WARN_ON(qg->reserved < num_bytes))
2473                         report_reserved_underflow(fs_info, qg, num_bytes);
2474                 else
2475                         qg->reserved -= num_bytes;
2476
2477                 list_for_each_entry(glist, &qg->groups, next_group) {
2478                         ret = ulist_add(fs_info->qgroup_ulist,
2479                                         glist->group->qgroupid,
2480                                         (uintptr_t)glist->group, GFP_ATOMIC);
2481                         if (ret < 0)
2482                                 goto out;
2483                 }
2484         }
2485
2486 out:
2487         spin_unlock(&fs_info->qgroup_lock);
2488 }
2489
2490 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2491 {
2492         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2493                 return;
2494         btrfs_err(trans->fs_info,
2495                 "qgroups not uptodate in trans handle %p:  list is%s empty, seq is %#x.%x",
2496                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2497                 (u32)(trans->delayed_ref_elem.seq >> 32),
2498                 (u32)trans->delayed_ref_elem.seq);
2499         BUG();
2500 }
2501
2502 /*
2503  * returns < 0 on error, 0 when more leafs are to be scanned.
2504  * returns 1 when done.
2505  */
2506 static int
2507 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2508                    struct btrfs_trans_handle *trans)
2509 {
2510         struct btrfs_key found;
2511         struct extent_buffer *scratch_leaf = NULL;
2512         struct ulist *roots = NULL;
2513         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2514         u64 num_bytes;
2515         int slot;
2516         int ret;
2517
2518         mutex_lock(&fs_info->qgroup_rescan_lock);
2519         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2520                                          &fs_info->qgroup_rescan_progress,
2521                                          path, 1, 0);
2522
2523         btrfs_debug(fs_info,
2524                 "current progress key (%llu %u %llu), search_slot ret %d",
2525                 fs_info->qgroup_rescan_progress.objectid,
2526                 fs_info->qgroup_rescan_progress.type,
2527                 fs_info->qgroup_rescan_progress.offset, ret);
2528
2529         if (ret) {
2530                 /*
2531                  * The rescan is about to end, we will not be scanning any
2532                  * further blocks. We cannot unset the RESCAN flag here, because
2533                  * we want to commit the transaction if everything went well.
2534                  * To make the live accounting work in this phase, we set our
2535                  * scan progress pointer such that every real extent objectid
2536                  * will be smaller.
2537                  */
2538                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2539                 btrfs_release_path(path);
2540                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2541                 return ret;
2542         }
2543
2544         btrfs_item_key_to_cpu(path->nodes[0], &found,
2545                               btrfs_header_nritems(path->nodes[0]) - 1);
2546         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2547
2548         btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2549         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2550         if (!scratch_leaf) {
2551                 ret = -ENOMEM;
2552                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2553                 goto out;
2554         }
2555         extent_buffer_get(scratch_leaf);
2556         btrfs_tree_read_lock(scratch_leaf);
2557         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2558         slot = path->slots[0];
2559         btrfs_release_path(path);
2560         mutex_unlock(&fs_info->qgroup_rescan_lock);
2561
2562         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2563                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2564                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2565                     found.type != BTRFS_METADATA_ITEM_KEY)
2566                         continue;
2567                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2568                         num_bytes = fs_info->nodesize;
2569                 else
2570                         num_bytes = found.offset;
2571
2572                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2573                                            &roots);
2574                 if (ret < 0)
2575                         goto out;
2576                 /* For rescan, just pass old_roots as NULL */
2577                 ret = btrfs_qgroup_account_extent(trans, fs_info,
2578                                 found.objectid, num_bytes, NULL, roots);
2579                 if (ret < 0)
2580                         goto out;
2581         }
2582 out:
2583         if (scratch_leaf) {
2584                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2585                 free_extent_buffer(scratch_leaf);
2586         }
2587         btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2588
2589         return ret;
2590 }
2591
2592 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2593 {
2594         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2595                                                      qgroup_rescan_work);
2596         struct btrfs_path *path;
2597         struct btrfs_trans_handle *trans = NULL;
2598         int err = -ENOMEM;
2599         int ret = 0;
2600
2601         path = btrfs_alloc_path();
2602         if (!path)
2603                 goto out;
2604
2605         err = 0;
2606         while (!err && !btrfs_fs_closing(fs_info)) {
2607                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2608                 if (IS_ERR(trans)) {
2609                         err = PTR_ERR(trans);
2610                         break;
2611                 }
2612                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2613                         err = -EINTR;
2614                 } else {
2615                         err = qgroup_rescan_leaf(fs_info, path, trans);
2616                 }
2617                 if (err > 0)
2618                         btrfs_commit_transaction(trans);
2619                 else
2620                         btrfs_end_transaction(trans);
2621         }
2622
2623 out:
2624         btrfs_free_path(path);
2625
2626         mutex_lock(&fs_info->qgroup_rescan_lock);
2627         if (!btrfs_fs_closing(fs_info))
2628                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2629
2630         if (err > 0 &&
2631             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2632                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2633         } else if (err < 0) {
2634                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2635         }
2636         mutex_unlock(&fs_info->qgroup_rescan_lock);
2637
2638         /*
2639          * only update status, since the previous part has already updated the
2640          * qgroup info.
2641          */
2642         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2643         if (IS_ERR(trans)) {
2644                 err = PTR_ERR(trans);
2645                 btrfs_err(fs_info,
2646                           "fail to start transaction for status update: %d\n",
2647                           err);
2648                 goto done;
2649         }
2650         ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2651         if (ret < 0) {
2652                 err = ret;
2653                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2654         }
2655         btrfs_end_transaction(trans);
2656
2657         if (btrfs_fs_closing(fs_info)) {
2658                 btrfs_info(fs_info, "qgroup scan paused");
2659         } else if (err >= 0) {
2660                 btrfs_info(fs_info, "qgroup scan completed%s",
2661                         err > 0 ? " (inconsistency flag cleared)" : "");
2662         } else {
2663                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2664         }
2665
2666 done:
2667         mutex_lock(&fs_info->qgroup_rescan_lock);
2668         fs_info->qgroup_rescan_running = false;
2669         mutex_unlock(&fs_info->qgroup_rescan_lock);
2670         complete_all(&fs_info->qgroup_rescan_completion);
2671 }
2672
2673 /*
2674  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2675  * memory required for the rescan context.
2676  */
2677 static int
2678 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2679                    int init_flags)
2680 {
2681         int ret = 0;
2682
2683         if (!init_flags &&
2684             (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2685              !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2686                 ret = -EINVAL;
2687                 goto err;
2688         }
2689
2690         mutex_lock(&fs_info->qgroup_rescan_lock);
2691         spin_lock(&fs_info->qgroup_lock);
2692
2693         if (init_flags) {
2694                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2695                         ret = -EINPROGRESS;
2696                 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2697                         ret = -EINVAL;
2698
2699                 if (ret) {
2700                         spin_unlock(&fs_info->qgroup_lock);
2701                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2702                         goto err;
2703                 }
2704                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2705         }
2706
2707         memset(&fs_info->qgroup_rescan_progress, 0,
2708                 sizeof(fs_info->qgroup_rescan_progress));
2709         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2710         init_completion(&fs_info->qgroup_rescan_completion);
2711         fs_info->qgroup_rescan_running = true;
2712
2713         spin_unlock(&fs_info->qgroup_lock);
2714         mutex_unlock(&fs_info->qgroup_rescan_lock);
2715
2716         memset(&fs_info->qgroup_rescan_work, 0,
2717                sizeof(fs_info->qgroup_rescan_work));
2718         btrfs_init_work(&fs_info->qgroup_rescan_work,
2719                         btrfs_qgroup_rescan_helper,
2720                         btrfs_qgroup_rescan_worker, NULL, NULL);
2721
2722         if (ret) {
2723 err:
2724                 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2725                 return ret;
2726         }
2727
2728         return 0;
2729 }
2730
2731 static void
2732 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2733 {
2734         struct rb_node *n;
2735         struct btrfs_qgroup *qgroup;
2736
2737         spin_lock(&fs_info->qgroup_lock);
2738         /* clear all current qgroup tracking information */
2739         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2740                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2741                 qgroup->rfer = 0;
2742                 qgroup->rfer_cmpr = 0;
2743                 qgroup->excl = 0;
2744                 qgroup->excl_cmpr = 0;
2745         }
2746         spin_unlock(&fs_info->qgroup_lock);
2747 }
2748
2749 int
2750 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2751 {
2752         int ret = 0;
2753         struct btrfs_trans_handle *trans;
2754
2755         ret = qgroup_rescan_init(fs_info, 0, 1);
2756         if (ret)
2757                 return ret;
2758
2759         /*
2760          * We have set the rescan_progress to 0, which means no more
2761          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2762          * However, btrfs_qgroup_account_ref may be right after its call
2763          * to btrfs_find_all_roots, in which case it would still do the
2764          * accounting.
2765          * To solve this, we're committing the transaction, which will
2766          * ensure we run all delayed refs and only after that, we are
2767          * going to clear all tracking information for a clean start.
2768          */
2769
2770         trans = btrfs_join_transaction(fs_info->fs_root);
2771         if (IS_ERR(trans)) {
2772                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2773                 return PTR_ERR(trans);
2774         }
2775         ret = btrfs_commit_transaction(trans);
2776         if (ret) {
2777                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2778                 return ret;
2779         }
2780
2781         qgroup_rescan_zero_tracking(fs_info);
2782
2783         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2784                          &fs_info->qgroup_rescan_work);
2785
2786         return 0;
2787 }
2788
2789 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2790                                      bool interruptible)
2791 {
2792         int running;
2793         int ret = 0;
2794
2795         mutex_lock(&fs_info->qgroup_rescan_lock);
2796         spin_lock(&fs_info->qgroup_lock);
2797         running = fs_info->qgroup_rescan_running;
2798         spin_unlock(&fs_info->qgroup_lock);
2799         mutex_unlock(&fs_info->qgroup_rescan_lock);
2800
2801         if (!running)
2802                 return 0;
2803
2804         if (interruptible)
2805                 ret = wait_for_completion_interruptible(
2806                                         &fs_info->qgroup_rescan_completion);
2807         else
2808                 wait_for_completion(&fs_info->qgroup_rescan_completion);
2809
2810         return ret;
2811 }
2812
2813 /*
2814  * this is only called from open_ctree where we're still single threaded, thus
2815  * locking is omitted here.
2816  */
2817 void
2818 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2819 {
2820         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2821                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2822                                  &fs_info->qgroup_rescan_work);
2823 }
2824
2825 /*
2826  * Reserve qgroup space for range [start, start + len).
2827  *
2828  * This function will either reserve space from related qgroups or doing
2829  * nothing if the range is already reserved.
2830  *
2831  * Return 0 for successful reserve
2832  * Return <0 for error (including -EQUOT)
2833  *
2834  * NOTE: this function may sleep for memory allocation.
2835  */
2836 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2837 {
2838         struct btrfs_root *root = BTRFS_I(inode)->root;
2839         struct extent_changeset changeset;
2840         struct ulist_node *unode;
2841         struct ulist_iterator uiter;
2842         int ret;
2843
2844         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2845             !is_fstree(root->objectid) || len == 0)
2846                 return 0;
2847
2848         changeset.bytes_changed = 0;
2849         ulist_init(&changeset.range_changed);
2850         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2851                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2852         trace_btrfs_qgroup_reserve_data(inode, start, len,
2853                                         changeset.bytes_changed,
2854                                         QGROUP_RESERVE);
2855         if (ret < 0)
2856                 goto cleanup;
2857         ret = qgroup_reserve(root, changeset.bytes_changed, true);
2858         if (ret < 0)
2859                 goto cleanup;
2860
2861         ulist_release(&changeset.range_changed);
2862         return ret;
2863
2864 cleanup:
2865         /* cleanup already reserved ranges */
2866         ULIST_ITER_INIT(&uiter);
2867         while ((unode = ulist_next(&changeset.range_changed, &uiter)))
2868                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2869                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2870                                  GFP_NOFS);
2871         ulist_release(&changeset.range_changed);
2872         return ret;
2873 }
2874
2875 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2876                                        int free)
2877 {
2878         struct extent_changeset changeset;
2879         int trace_op = QGROUP_RELEASE;
2880         int ret;
2881
2882         changeset.bytes_changed = 0;
2883         ulist_init(&changeset.range_changed);
2884         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2885                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2886         if (ret < 0)
2887                 goto out;
2888
2889         if (free) {
2890                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
2891                                 BTRFS_I(inode)->root->objectid,
2892                                 changeset.bytes_changed);
2893                 trace_op = QGROUP_FREE;
2894         }
2895         trace_btrfs_qgroup_release_data(inode, start, len,
2896                                         changeset.bytes_changed, trace_op);
2897 out:
2898         ulist_release(&changeset.range_changed);
2899         return ret;
2900 }
2901
2902 /*
2903  * Free a reserved space range from io_tree and related qgroups
2904  *
2905  * Should be called when a range of pages get invalidated before reaching disk.
2906  * Or for error cleanup case.
2907  *
2908  * For data written to disk, use btrfs_qgroup_release_data().
2909  *
2910  * NOTE: This function may sleep for memory allocation.
2911  */
2912 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2913 {
2914         return __btrfs_qgroup_release_data(inode, start, len, 1);
2915 }
2916
2917 /*
2918  * Release a reserved space range from io_tree only.
2919  *
2920  * Should be called when a range of pages get written to disk and corresponding
2921  * FILE_EXTENT is inserted into corresponding root.
2922  *
2923  * Since new qgroup accounting framework will only update qgroup numbers at
2924  * commit_transaction() time, its reserved space shouldn't be freed from
2925  * related qgroups.
2926  *
2927  * But we should release the range from io_tree, to allow further write to be
2928  * COWed.
2929  *
2930  * NOTE: This function may sleep for memory allocation.
2931  */
2932 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2933 {
2934         return __btrfs_qgroup_release_data(inode, start, len, 0);
2935 }
2936
2937 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
2938                               bool enforce)
2939 {
2940         struct btrfs_fs_info *fs_info = root->fs_info;
2941         int ret;
2942
2943         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2944             !is_fstree(root->objectid) || num_bytes == 0)
2945                 return 0;
2946
2947         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2948         ret = qgroup_reserve(root, num_bytes, enforce);
2949         if (ret < 0)
2950                 return ret;
2951         atomic64_add(num_bytes, &root->qgroup_meta_rsv);
2952         return ret;
2953 }
2954
2955 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2956 {
2957         struct btrfs_fs_info *fs_info = root->fs_info;
2958         u64 reserved;
2959
2960         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2961             !is_fstree(root->objectid))
2962                 return;
2963
2964         reserved = atomic64_xchg(&root->qgroup_meta_rsv, 0);
2965         if (reserved == 0)
2966                 return;
2967         btrfs_qgroup_free_refroot(fs_info, root->objectid, reserved);
2968 }
2969
2970 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2971 {
2972         struct btrfs_fs_info *fs_info = root->fs_info;
2973
2974         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2975             !is_fstree(root->objectid))
2976                 return;
2977
2978         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2979         WARN_ON(atomic64_read(&root->qgroup_meta_rsv) < num_bytes);
2980         atomic64_sub(num_bytes, &root->qgroup_meta_rsv);
2981         btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes);
2982 }
2983
2984 /*
2985  * Check qgroup reserved space leaking, normally at destroy inode
2986  * time
2987  */
2988 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2989 {
2990         struct extent_changeset changeset;
2991         struct ulist_node *unode;
2992         struct ulist_iterator iter;
2993         int ret;
2994
2995         changeset.bytes_changed = 0;
2996         ulist_init(&changeset.range_changed);
2997         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
2998                         EXTENT_QGROUP_RESERVED, &changeset);
2999
3000         WARN_ON(ret < 0);
3001         if (WARN_ON(changeset.bytes_changed)) {
3002                 ULIST_ITER_INIT(&iter);
3003                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3004                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3005                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3006                                 inode->i_ino, unode->val, unode->aux);
3007                 }
3008                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3009                                 BTRFS_I(inode)->root->objectid,
3010                                 changeset.bytes_changed);
3011
3012         }
3013         ulist_release(&changeset.range_changed);
3014 }