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_NOFS);
 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_NOFS);
 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(trans, 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 /*
1042  * The easy accounting, if we are adding/removing the only ref for an extent
1043  * then this qgroup and all of the parent qgroups get their reference and
1044  * exclusive counts adjusted.
1045  *
1046  * Caller should hold fs_info->qgroup_lock.
1047  */
1048 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1049                                     struct ulist *tmp, u64 ref_root,
1050                                     u64 num_bytes, int sign)
1051 {
1052         struct btrfs_qgroup *qgroup;
1053         struct btrfs_qgroup_list *glist;
1054         struct ulist_node *unode;
1055         struct ulist_iterator uiter;
1056         int ret = 0;
1057
1058         qgroup = find_qgroup_rb(fs_info, ref_root);
1059         if (!qgroup)
1060                 goto out;
1061
1062         qgroup->rfer += sign * num_bytes;
1063         qgroup->rfer_cmpr += sign * num_bytes;
1064
1065         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1066         qgroup->excl += sign * num_bytes;
1067         qgroup->excl_cmpr += sign * num_bytes;
1068         if (sign > 0)
1069                 qgroup->reserved -= num_bytes;
1070
1071         qgroup_dirty(fs_info, qgroup);
1072
1073         /* Get all of the parent groups that contain this qgroup */
1074         list_for_each_entry(glist, &qgroup->groups, next_group) {
1075                 ret = ulist_add(tmp, glist->group->qgroupid,
1076                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1077                 if (ret < 0)
1078                         goto out;
1079         }
1080
1081         /* Iterate all of the parents and adjust their reference counts */
1082         ULIST_ITER_INIT(&uiter);
1083         while ((unode = ulist_next(tmp, &uiter))) {
1084                 qgroup = unode_aux_to_qgroup(unode);
1085                 qgroup->rfer += sign * num_bytes;
1086                 qgroup->rfer_cmpr += sign * num_bytes;
1087                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1088                 qgroup->excl += sign * num_bytes;
1089                 if (sign > 0)
1090                         qgroup->reserved -= num_bytes;
1091                 qgroup->excl_cmpr += sign * num_bytes;
1092                 qgroup_dirty(fs_info, qgroup);
1093
1094                 /* Add any parents of the parents */
1095                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1096                         ret = ulist_add(tmp, glist->group->qgroupid,
1097                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1098                         if (ret < 0)
1099                                 goto out;
1100                 }
1101         }
1102         ret = 0;
1103 out:
1104         return ret;
1105 }
1106
1107
1108 /*
1109  * Quick path for updating qgroup with only excl refs.
1110  *
1111  * In that case, just update all parent will be enough.
1112  * Or we needs to do a full rescan.
1113  * Caller should also hold fs_info->qgroup_lock.
1114  *
1115  * Return 0 for quick update, return >0 for need to full rescan
1116  * and mark INCONSISTENT flag.
1117  * Return < 0 for other error.
1118  */
1119 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1120                                    struct ulist *tmp, u64 src, u64 dst,
1121                                    int sign)
1122 {
1123         struct btrfs_qgroup *qgroup;
1124         int ret = 1;
1125         int err = 0;
1126
1127         qgroup = find_qgroup_rb(fs_info, src);
1128         if (!qgroup)
1129                 goto out;
1130         if (qgroup->excl == qgroup->rfer) {
1131                 ret = 0;
1132                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1133                                                qgroup->excl, sign);
1134                 if (err < 0) {
1135                         ret = err;
1136                         goto out;
1137                 }
1138         }
1139 out:
1140         if (ret)
1141                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1142         return ret;
1143 }
1144
1145 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1146                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1147 {
1148         struct btrfs_root *quota_root;
1149         struct btrfs_qgroup *parent;
1150         struct btrfs_qgroup *member;
1151         struct btrfs_qgroup_list *list;
1152         struct ulist *tmp;
1153         int ret = 0;
1154
1155         /* Check the level of src and dst first */
1156         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1157                 return -EINVAL;
1158
1159         tmp = ulist_alloc(GFP_NOFS);
1160         if (!tmp)
1161                 return -ENOMEM;
1162
1163         mutex_lock(&fs_info->qgroup_ioctl_lock);
1164         quota_root = fs_info->quota_root;
1165         if (!quota_root) {
1166                 ret = -EINVAL;
1167                 goto out;
1168         }
1169         member = find_qgroup_rb(fs_info, src);
1170         parent = find_qgroup_rb(fs_info, dst);
1171         if (!member || !parent) {
1172                 ret = -EINVAL;
1173                 goto out;
1174         }
1175
1176         /* check if such qgroup relation exist firstly */
1177         list_for_each_entry(list, &member->groups, next_group) {
1178                 if (list->group == parent) {
1179                         ret = -EEXIST;
1180                         goto out;
1181                 }
1182         }
1183
1184         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1185         if (ret)
1186                 goto out;
1187
1188         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1189         if (ret) {
1190                 del_qgroup_relation_item(trans, quota_root, src, dst);
1191                 goto out;
1192         }
1193
1194         spin_lock(&fs_info->qgroup_lock);
1195         ret = add_relation_rb(fs_info, src, dst);
1196         if (ret < 0) {
1197                 spin_unlock(&fs_info->qgroup_lock);
1198                 goto out;
1199         }
1200         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1201         spin_unlock(&fs_info->qgroup_lock);
1202 out:
1203         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1204         ulist_free(tmp);
1205         return ret;
1206 }
1207
1208 int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1209                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1210 {
1211         struct btrfs_root *quota_root;
1212         struct btrfs_qgroup *parent;
1213         struct btrfs_qgroup *member;
1214         struct btrfs_qgroup_list *list;
1215         struct ulist *tmp;
1216         int ret = 0;
1217         int err;
1218
1219         tmp = ulist_alloc(GFP_NOFS);
1220         if (!tmp)
1221                 return -ENOMEM;
1222
1223         quota_root = fs_info->quota_root;
1224         if (!quota_root) {
1225                 ret = -EINVAL;
1226                 goto out;
1227         }
1228
1229         member = find_qgroup_rb(fs_info, src);
1230         parent = find_qgroup_rb(fs_info, dst);
1231         if (!member || !parent) {
1232                 ret = -EINVAL;
1233                 goto out;
1234         }
1235
1236         /* check if such qgroup relation exist firstly */
1237         list_for_each_entry(list, &member->groups, next_group) {
1238                 if (list->group == parent)
1239                         goto exist;
1240         }
1241         ret = -ENOENT;
1242         goto out;
1243 exist:
1244         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1245         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1246         if (err && !ret)
1247                 ret = err;
1248
1249         spin_lock(&fs_info->qgroup_lock);
1250         del_relation_rb(fs_info, src, dst);
1251         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1252         spin_unlock(&fs_info->qgroup_lock);
1253 out:
1254         ulist_free(tmp);
1255         return ret;
1256 }
1257
1258 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1259                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1260 {
1261         int ret = 0;
1262
1263         mutex_lock(&fs_info->qgroup_ioctl_lock);
1264         ret = __del_qgroup_relation(trans, fs_info, src, dst);
1265         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1266
1267         return ret;
1268 }
1269
1270 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1271                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1272 {
1273         struct btrfs_root *quota_root;
1274         struct btrfs_qgroup *qgroup;
1275         int ret = 0;
1276
1277         mutex_lock(&fs_info->qgroup_ioctl_lock);
1278         quota_root = fs_info->quota_root;
1279         if (!quota_root) {
1280                 ret = -EINVAL;
1281                 goto out;
1282         }
1283         qgroup = find_qgroup_rb(fs_info, qgroupid);
1284         if (qgroup) {
1285                 ret = -EEXIST;
1286                 goto out;
1287         }
1288
1289         ret = add_qgroup_item(trans, quota_root, qgroupid);
1290         if (ret)
1291                 goto out;
1292
1293         spin_lock(&fs_info->qgroup_lock);
1294         qgroup = add_qgroup_rb(fs_info, qgroupid);
1295         spin_unlock(&fs_info->qgroup_lock);
1296
1297         if (IS_ERR(qgroup))
1298                 ret = PTR_ERR(qgroup);
1299 out:
1300         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1301         return ret;
1302 }
1303
1304 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1305                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1306 {
1307         struct btrfs_root *quota_root;
1308         struct btrfs_qgroup *qgroup;
1309         struct btrfs_qgroup_list *list;
1310         int ret = 0;
1311
1312         mutex_lock(&fs_info->qgroup_ioctl_lock);
1313         quota_root = fs_info->quota_root;
1314         if (!quota_root) {
1315                 ret = -EINVAL;
1316                 goto out;
1317         }
1318
1319         qgroup = find_qgroup_rb(fs_info, qgroupid);
1320         if (!qgroup) {
1321                 ret = -ENOENT;
1322                 goto out;
1323         } else {
1324                 /* check if there are no children of this qgroup */
1325                 if (!list_empty(&qgroup->members)) {
1326                         ret = -EBUSY;
1327                         goto out;
1328                 }
1329         }
1330         ret = del_qgroup_item(trans, quota_root, qgroupid);
1331
1332         while (!list_empty(&qgroup->groups)) {
1333                 list = list_first_entry(&qgroup->groups,
1334                                         struct btrfs_qgroup_list, next_group);
1335                 ret = __del_qgroup_relation(trans, fs_info,
1336                                            qgroupid,
1337                                            list->group->qgroupid);
1338                 if (ret)
1339                         goto out;
1340         }
1341
1342         spin_lock(&fs_info->qgroup_lock);
1343         del_qgroup_rb(fs_info, qgroupid);
1344         spin_unlock(&fs_info->qgroup_lock);
1345 out:
1346         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1347         return ret;
1348 }
1349
1350 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1351                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1352                        struct btrfs_qgroup_limit *limit)
1353 {
1354         struct btrfs_root *quota_root;
1355         struct btrfs_qgroup *qgroup;
1356         int ret = 0;
1357         /* Sometimes we would want to clear the limit on this qgroup.
1358          * To meet this requirement, we treat the -1 as a special value
1359          * which tell kernel to clear the limit on this qgroup.
1360          */
1361         const u64 CLEAR_VALUE = -1;
1362
1363         mutex_lock(&fs_info->qgroup_ioctl_lock);
1364         quota_root = fs_info->quota_root;
1365         if (!quota_root) {
1366                 ret = -EINVAL;
1367                 goto out;
1368         }
1369
1370         qgroup = find_qgroup_rb(fs_info, qgroupid);
1371         if (!qgroup) {
1372                 ret = -ENOENT;
1373                 goto out;
1374         }
1375
1376         spin_lock(&fs_info->qgroup_lock);
1377         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1378                 if (limit->max_rfer == CLEAR_VALUE) {
1379                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1380                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1381                         qgroup->max_rfer = 0;
1382                 } else {
1383                         qgroup->max_rfer = limit->max_rfer;
1384                 }
1385         }
1386         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1387                 if (limit->max_excl == CLEAR_VALUE) {
1388                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1389                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1390                         qgroup->max_excl = 0;
1391                 } else {
1392                         qgroup->max_excl = limit->max_excl;
1393                 }
1394         }
1395         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1396                 if (limit->rsv_rfer == CLEAR_VALUE) {
1397                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1398                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1399                         qgroup->rsv_rfer = 0;
1400                 } else {
1401                         qgroup->rsv_rfer = limit->rsv_rfer;
1402                 }
1403         }
1404         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1405                 if (limit->rsv_excl == CLEAR_VALUE) {
1406                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1407                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1408                         qgroup->rsv_excl = 0;
1409                 } else {
1410                         qgroup->rsv_excl = limit->rsv_excl;
1411                 }
1412         }
1413         qgroup->lim_flags |= limit->flags;
1414
1415         spin_unlock(&fs_info->qgroup_lock);
1416
1417         ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1418         if (ret) {
1419                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1420                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1421                        qgroupid);
1422         }
1423
1424 out:
1425         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1426         return ret;
1427 }
1428
1429 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1430                                          struct btrfs_fs_info *fs_info)
1431 {
1432         struct btrfs_qgroup_extent_record *record;
1433         struct btrfs_delayed_ref_root *delayed_refs;
1434         struct rb_node *node;
1435         u64 qgroup_to_skip;
1436         int ret = 0;
1437
1438         delayed_refs = &trans->transaction->delayed_refs;
1439         qgroup_to_skip = delayed_refs->qgroup_to_skip;
1440
1441         /*
1442          * No need to do lock, since this function will only be called in
1443          * btrfs_commit_transaction().
1444          */
1445         node = rb_first(&delayed_refs->dirty_extent_root);
1446         while (node) {
1447                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1448                                   node);
1449                 ret = btrfs_find_all_roots(NULL, fs_info, record->bytenr, 0,
1450                                            &record->old_roots);
1451                 if (ret < 0)
1452                         break;
1453                 if (qgroup_to_skip)
1454                         ulist_del(record->old_roots, qgroup_to_skip, 0);
1455                 node = rb_next(node);
1456         }
1457         return ret;
1458 }
1459
1460 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1461                                 struct btrfs_delayed_ref_root *delayed_refs,
1462                                 struct btrfs_qgroup_extent_record *record)
1463 {
1464         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1465         struct rb_node *parent_node = NULL;
1466         struct btrfs_qgroup_extent_record *entry;
1467         u64 bytenr = record->bytenr;
1468
1469         assert_spin_locked(&delayed_refs->lock);
1470         trace_btrfs_qgroup_trace_extent(fs_info, record);
1471
1472         while (*p) {
1473                 parent_node = *p;
1474                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1475                                  node);
1476                 if (bytenr < entry->bytenr)
1477                         p = &(*p)->rb_left;
1478                 else if (bytenr > entry->bytenr)
1479                         p = &(*p)->rb_right;
1480                 else
1481                         return 1;
1482         }
1483
1484         rb_link_node(&record->node, parent_node, p);
1485         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1486         return 0;
1487 }
1488
1489 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1490                 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1491                 gfp_t gfp_flag)
1492 {
1493         struct btrfs_qgroup_extent_record *record;
1494         struct btrfs_delayed_ref_root *delayed_refs;
1495         int ret;
1496
1497         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1498             || bytenr == 0 || num_bytes == 0)
1499                 return 0;
1500         if (WARN_ON(trans == NULL))
1501                 return -EINVAL;
1502         record = kmalloc(sizeof(*record), gfp_flag);
1503         if (!record)
1504                 return -ENOMEM;
1505
1506         delayed_refs = &trans->transaction->delayed_refs;
1507         record->bytenr = bytenr;
1508         record->num_bytes = num_bytes;
1509         record->old_roots = NULL;
1510
1511         spin_lock(&delayed_refs->lock);
1512         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1513         spin_unlock(&delayed_refs->lock);
1514         if (ret > 0)
1515                 kfree(record);
1516         return 0;
1517 }
1518
1519 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1520                                   struct btrfs_fs_info *fs_info,
1521                                   struct extent_buffer *eb)
1522 {
1523         int nr = btrfs_header_nritems(eb);
1524         int i, extent_type, ret;
1525         struct btrfs_key key;
1526         struct btrfs_file_extent_item *fi;
1527         u64 bytenr, num_bytes;
1528
1529         /* We can be called directly from walk_up_proc() */
1530         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1531                 return 0;
1532
1533         for (i = 0; i < nr; i++) {
1534                 btrfs_item_key_to_cpu(eb, &key, i);
1535
1536                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1537                         continue;
1538
1539                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1540                 /* filter out non qgroup-accountable extents  */
1541                 extent_type = btrfs_file_extent_type(eb, fi);
1542
1543                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1544                         continue;
1545
1546                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1547                 if (!bytenr)
1548                         continue;
1549
1550                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1551
1552                 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1553                                                 num_bytes, GFP_NOFS);
1554                 if (ret)
1555                         return ret;
1556         }
1557         return 0;
1558 }
1559
1560 /*
1561  * Walk up the tree from the bottom, freeing leaves and any interior
1562  * nodes which have had all slots visited. If a node (leaf or
1563  * interior) is freed, the node above it will have it's slot
1564  * incremented. The root node will never be freed.
1565  *
1566  * At the end of this function, we should have a path which has all
1567  * slots incremented to the next position for a search. If we need to
1568  * read a new node it will be NULL and the node above it will have the
1569  * correct slot selected for a later read.
1570  *
1571  * If we increment the root nodes slot counter past the number of
1572  * elements, 1 is returned to signal completion of the search.
1573  */
1574 static int adjust_slots_upwards(struct btrfs_root *root,
1575                                 struct btrfs_path *path, int root_level)
1576 {
1577         int level = 0;
1578         int nr, slot;
1579         struct extent_buffer *eb;
1580
1581         if (root_level == 0)
1582                 return 1;
1583
1584         while (level <= root_level) {
1585                 eb = path->nodes[level];
1586                 nr = btrfs_header_nritems(eb);
1587                 path->slots[level]++;
1588                 slot = path->slots[level];
1589                 if (slot >= nr || level == 0) {
1590                         /*
1591                          * Don't free the root -  we will detect this
1592                          * condition after our loop and return a
1593                          * positive value for caller to stop walking the tree.
1594                          */
1595                         if (level != root_level) {
1596                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1597                                 path->locks[level] = 0;
1598
1599                                 free_extent_buffer(eb);
1600                                 path->nodes[level] = NULL;
1601                                 path->slots[level] = 0;
1602                         }
1603                 } else {
1604                         /*
1605                          * We have a valid slot to walk back down
1606                          * from. Stop here so caller can process these
1607                          * new nodes.
1608                          */
1609                         break;
1610                 }
1611
1612                 level++;
1613         }
1614
1615         eb = path->nodes[root_level];
1616         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1617                 return 1;
1618
1619         return 0;
1620 }
1621
1622 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1623                                struct btrfs_root *root,
1624                                struct extent_buffer *root_eb,
1625                                u64 root_gen, int root_level)
1626 {
1627         struct btrfs_fs_info *fs_info = root->fs_info;
1628         int ret = 0;
1629         int level;
1630         struct extent_buffer *eb = root_eb;
1631         struct btrfs_path *path = NULL;
1632
1633         BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
1634         BUG_ON(root_eb == NULL);
1635
1636         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1637                 return 0;
1638
1639         if (!extent_buffer_uptodate(root_eb)) {
1640                 ret = btrfs_read_buffer(root_eb, root_gen);
1641                 if (ret)
1642                         goto out;
1643         }
1644
1645         if (root_level == 0) {
1646                 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1647                 goto out;
1648         }
1649
1650         path = btrfs_alloc_path();
1651         if (!path)
1652                 return -ENOMEM;
1653
1654         /*
1655          * Walk down the tree.  Missing extent blocks are filled in as
1656          * we go. Metadata is accounted every time we read a new
1657          * extent block.
1658          *
1659          * When we reach a leaf, we account for file extent items in it,
1660          * walk back up the tree (adjusting slot pointers as we go)
1661          * and restart the search process.
1662          */
1663         extent_buffer_get(root_eb); /* For path */
1664         path->nodes[root_level] = root_eb;
1665         path->slots[root_level] = 0;
1666         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1667 walk_down:
1668         level = root_level;
1669         while (level >= 0) {
1670                 if (path->nodes[level] == NULL) {
1671                         int parent_slot;
1672                         u64 child_gen;
1673                         u64 child_bytenr;
1674
1675                         /*
1676                          * We need to get child blockptr/gen from parent before
1677                          * we can read it.
1678                           */
1679                         eb = path->nodes[level + 1];
1680                         parent_slot = path->slots[level + 1];
1681                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1682                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1683
1684                         eb = read_tree_block(fs_info, child_bytenr, child_gen);
1685                         if (IS_ERR(eb)) {
1686                                 ret = PTR_ERR(eb);
1687                                 goto out;
1688                         } else if (!extent_buffer_uptodate(eb)) {
1689                                 free_extent_buffer(eb);
1690                                 ret = -EIO;
1691                                 goto out;
1692                         }
1693
1694                         path->nodes[level] = eb;
1695                         path->slots[level] = 0;
1696
1697                         btrfs_tree_read_lock(eb);
1698                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1699                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1700
1701                         ret = btrfs_qgroup_trace_extent(trans, fs_info,
1702                                                         child_bytenr,
1703                                                         fs_info->nodesize,
1704                                                         GFP_NOFS);
1705                         if (ret)
1706                                 goto out;
1707                 }
1708
1709                 if (level == 0) {
1710                         ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1711                                                            path->nodes[level]);
1712                         if (ret)
1713                                 goto out;
1714
1715                         /* Nonzero return here means we completed our search */
1716                         ret = adjust_slots_upwards(root, path, root_level);
1717                         if (ret)
1718                                 break;
1719
1720                         /* Restart search with new slots */
1721                         goto walk_down;
1722                 }
1723
1724                 level--;
1725         }
1726
1727         ret = 0;
1728 out:
1729         btrfs_free_path(path);
1730
1731         return ret;
1732 }
1733
1734 #define UPDATE_NEW      0
1735 #define UPDATE_OLD      1
1736 /*
1737  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1738  */
1739 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1740                                 struct ulist *roots, struct ulist *tmp,
1741                                 struct ulist *qgroups, u64 seq, int update_old)
1742 {
1743         struct ulist_node *unode;
1744         struct ulist_iterator uiter;
1745         struct ulist_node *tmp_unode;
1746         struct ulist_iterator tmp_uiter;
1747         struct btrfs_qgroup *qg;
1748         int ret = 0;
1749
1750         if (!roots)
1751                 return 0;
1752         ULIST_ITER_INIT(&uiter);
1753         while ((unode = ulist_next(roots, &uiter))) {
1754                 qg = find_qgroup_rb(fs_info, unode->val);
1755                 if (!qg)
1756                         continue;
1757
1758                 ulist_reinit(tmp);
1759                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1760                                 GFP_ATOMIC);
1761                 if (ret < 0)
1762                         return ret;
1763                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1764                 if (ret < 0)
1765                         return ret;
1766                 ULIST_ITER_INIT(&tmp_uiter);
1767                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1768                         struct btrfs_qgroup_list *glist;
1769
1770                         qg = unode_aux_to_qgroup(tmp_unode);
1771                         if (update_old)
1772                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1773                         else
1774                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1775                         list_for_each_entry(glist, &qg->groups, next_group) {
1776                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1777                                                 qgroup_to_aux(glist->group),
1778                                                 GFP_ATOMIC);
1779                                 if (ret < 0)
1780                                         return ret;
1781                                 ret = ulist_add(tmp, glist->group->qgroupid,
1782                                                 qgroup_to_aux(glist->group),
1783                                                 GFP_ATOMIC);
1784                                 if (ret < 0)
1785                                         return ret;
1786                         }
1787                 }
1788         }
1789         return 0;
1790 }
1791
1792 /*
1793  * Update qgroup rfer/excl counters.
1794  * Rfer update is easy, codes can explain themselves.
1795  *
1796  * Excl update is tricky, the update is split into 2 part.
1797  * Part 1: Possible exclusive <-> sharing detect:
1798  *      |       A       |       !A      |
1799  *  -------------------------------------
1800  *  B   |       *       |       -       |
1801  *  -------------------------------------
1802  *  !B  |       +       |       **      |
1803  *  -------------------------------------
1804  *
1805  * Conditions:
1806  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1807  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1808  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1809  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1810  *
1811  * Results:
1812  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1813  * *: Definitely not changed.           **: Possible unchanged.
1814  *
1815  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1816  *
1817  * To make the logic clear, we first use condition A and B to split
1818  * combination into 4 results.
1819  *
1820  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1821  * only on variant maybe 0.
1822  *
1823  * Lastly, check result **, since there are 2 variants maybe 0, split them
1824  * again(2x2).
1825  * But this time we don't need to consider other things, the codes and logic
1826  * is easy to understand now.
1827  */
1828 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1829                                   struct ulist *qgroups,
1830                                   u64 nr_old_roots,
1831                                   u64 nr_new_roots,
1832                                   u64 num_bytes, u64 seq)
1833 {
1834         struct ulist_node *unode;
1835         struct ulist_iterator uiter;
1836         struct btrfs_qgroup *qg;
1837         u64 cur_new_count, cur_old_count;
1838
1839         ULIST_ITER_INIT(&uiter);
1840         while ((unode = ulist_next(qgroups, &uiter))) {
1841                 bool dirty = false;
1842
1843                 qg = unode_aux_to_qgroup(unode);
1844                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1845                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1846
1847                 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1848                                              cur_old_count, cur_new_count);
1849
1850                 /* Rfer update part */
1851                 if (cur_old_count == 0 && cur_new_count > 0) {
1852                         qg->rfer += num_bytes;
1853                         qg->rfer_cmpr += num_bytes;
1854                         dirty = true;
1855                 }
1856                 if (cur_old_count > 0 && cur_new_count == 0) {
1857                         qg->rfer -= num_bytes;
1858                         qg->rfer_cmpr -= num_bytes;
1859                         dirty = true;
1860                 }
1861
1862                 /* Excl update part */
1863                 /* Exclusive/none -> shared case */
1864                 if (cur_old_count == nr_old_roots &&
1865                     cur_new_count < nr_new_roots) {
1866                         /* Exclusive -> shared */
1867                         if (cur_old_count != 0) {
1868                                 qg->excl -= num_bytes;
1869                                 qg->excl_cmpr -= num_bytes;
1870                                 dirty = true;
1871                         }
1872                 }
1873
1874                 /* Shared -> exclusive/none case */
1875                 if (cur_old_count < nr_old_roots &&
1876                     cur_new_count == nr_new_roots) {
1877                         /* Shared->exclusive */
1878                         if (cur_new_count != 0) {
1879                                 qg->excl += num_bytes;
1880                                 qg->excl_cmpr += num_bytes;
1881                                 dirty = true;
1882                         }
1883                 }
1884
1885                 /* Exclusive/none -> exclusive/none case */
1886                 if (cur_old_count == nr_old_roots &&
1887                     cur_new_count == nr_new_roots) {
1888                         if (cur_old_count == 0) {
1889                                 /* None -> exclusive/none */
1890
1891                                 if (cur_new_count != 0) {
1892                                         /* None -> exclusive */
1893                                         qg->excl += num_bytes;
1894                                         qg->excl_cmpr += num_bytes;
1895                                         dirty = true;
1896                                 }
1897                                 /* None -> none, nothing changed */
1898                         } else {
1899                                 /* Exclusive -> exclusive/none */
1900
1901                                 if (cur_new_count == 0) {
1902                                         /* Exclusive -> none */
1903                                         qg->excl -= num_bytes;
1904                                         qg->excl_cmpr -= num_bytes;
1905                                         dirty = true;
1906                                 }
1907                                 /* Exclusive -> exclusive, nothing changed */
1908                         }
1909                 }
1910
1911                 if (dirty)
1912                         qgroup_dirty(fs_info, qg);
1913         }
1914         return 0;
1915 }
1916
1917 int
1918 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1919                             struct btrfs_fs_info *fs_info,
1920                             u64 bytenr, u64 num_bytes,
1921                             struct ulist *old_roots, struct ulist *new_roots)
1922 {
1923         struct ulist *qgroups = NULL;
1924         struct ulist *tmp = NULL;
1925         u64 seq;
1926         u64 nr_new_roots = 0;
1927         u64 nr_old_roots = 0;
1928         int ret = 0;
1929
1930         if (new_roots)
1931                 nr_new_roots = new_roots->nnodes;
1932         if (old_roots)
1933                 nr_old_roots = old_roots->nnodes;
1934
1935         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1936                 goto out_free;
1937         BUG_ON(!fs_info->quota_root);
1938
1939         trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1940                                           nr_old_roots, nr_new_roots);
1941
1942         qgroups = ulist_alloc(GFP_NOFS);
1943         if (!qgroups) {
1944                 ret = -ENOMEM;
1945                 goto out_free;
1946         }
1947         tmp = ulist_alloc(GFP_NOFS);
1948         if (!tmp) {
1949                 ret = -ENOMEM;
1950                 goto out_free;
1951         }
1952
1953         mutex_lock(&fs_info->qgroup_rescan_lock);
1954         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1955                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1956                         mutex_unlock(&fs_info->qgroup_rescan_lock);
1957                         ret = 0;
1958                         goto out_free;
1959                 }
1960         }
1961         mutex_unlock(&fs_info->qgroup_rescan_lock);
1962
1963         spin_lock(&fs_info->qgroup_lock);
1964         seq = fs_info->qgroup_seq;
1965
1966         /* Update old refcnts using old_roots */
1967         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1968                                    UPDATE_OLD);
1969         if (ret < 0)
1970                 goto out;
1971
1972         /* Update new refcnts using new_roots */
1973         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1974                                    UPDATE_NEW);
1975         if (ret < 0)
1976                 goto out;
1977
1978         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1979                                num_bytes, seq);
1980
1981         /*
1982          * Bump qgroup_seq to avoid seq overlap
1983          */
1984         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1985 out:
1986         spin_unlock(&fs_info->qgroup_lock);
1987 out_free:
1988         ulist_free(tmp);
1989         ulist_free(qgroups);
1990         ulist_free(old_roots);
1991         ulist_free(new_roots);
1992         return ret;
1993 }
1994
1995 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
1996                                  struct btrfs_fs_info *fs_info)
1997 {
1998         struct btrfs_qgroup_extent_record *record;
1999         struct btrfs_delayed_ref_root *delayed_refs;
2000         struct ulist *new_roots = NULL;
2001         struct rb_node *node;
2002         u64 qgroup_to_skip;
2003         int ret = 0;
2004
2005         delayed_refs = &trans->transaction->delayed_refs;
2006         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2007         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2008                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2009                                   node);
2010
2011                 trace_btrfs_qgroup_account_extents(fs_info, record);
2012
2013                 if (!ret) {
2014                         /*
2015                          * Use (u64)-1 as time_seq to do special search, which
2016                          * doesn't lock tree or delayed_refs and search current
2017                          * root. It's safe inside commit_transaction().
2018                          */
2019                         ret = btrfs_find_all_roots(trans, fs_info,
2020                                         record->bytenr, (u64)-1, &new_roots);
2021                         if (ret < 0)
2022                                 goto cleanup;
2023                         if (qgroup_to_skip)
2024                                 ulist_del(new_roots, qgroup_to_skip, 0);
2025                         ret = btrfs_qgroup_account_extent(trans, fs_info,
2026                                         record->bytenr, record->num_bytes,
2027                                         record->old_roots, new_roots);
2028                         record->old_roots = NULL;
2029                         new_roots = NULL;
2030                 }
2031 cleanup:
2032                 ulist_free(record->old_roots);
2033                 ulist_free(new_roots);
2034                 new_roots = NULL;
2035                 rb_erase(node, &delayed_refs->dirty_extent_root);
2036                 kfree(record);
2037
2038         }
2039         return ret;
2040 }
2041
2042 /*
2043  * called from commit_transaction. Writes all changed qgroups to disk.
2044  */
2045 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2046                       struct btrfs_fs_info *fs_info)
2047 {
2048         struct btrfs_root *quota_root = fs_info->quota_root;
2049         int ret = 0;
2050         int start_rescan_worker = 0;
2051
2052         if (!quota_root)
2053                 goto out;
2054
2055         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
2056             test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2057                 start_rescan_worker = 1;
2058
2059         if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2060                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2061         if (test_and_clear_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags))
2062                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2063
2064         spin_lock(&fs_info->qgroup_lock);
2065         while (!list_empty(&fs_info->dirty_qgroups)) {
2066                 struct btrfs_qgroup *qgroup;
2067                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2068                                           struct btrfs_qgroup, dirty);
2069                 list_del_init(&qgroup->dirty);
2070                 spin_unlock(&fs_info->qgroup_lock);
2071                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2072                 if (ret)
2073                         fs_info->qgroup_flags |=
2074                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2075                 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2076                 if (ret)
2077                         fs_info->qgroup_flags |=
2078                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2079                 spin_lock(&fs_info->qgroup_lock);
2080         }
2081         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2082                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2083         else
2084                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2085         spin_unlock(&fs_info->qgroup_lock);
2086
2087         ret = update_qgroup_status_item(trans, fs_info, quota_root);
2088         if (ret)
2089                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2090
2091         if (!ret && start_rescan_worker) {
2092                 ret = qgroup_rescan_init(fs_info, 0, 1);
2093                 if (!ret) {
2094                         qgroup_rescan_zero_tracking(fs_info);
2095                         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2096                                          &fs_info->qgroup_rescan_work);
2097                 }
2098                 ret = 0;
2099         }
2100
2101 out:
2102
2103         return ret;
2104 }
2105
2106 /*
2107  * Copy the accounting information between qgroups. This is necessary
2108  * when a snapshot or a subvolume is created. Throwing an error will
2109  * cause a transaction abort so we take extra care here to only error
2110  * when a readonly fs is a reasonable outcome.
2111  */
2112 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2113                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2114                          struct btrfs_qgroup_inherit *inherit)
2115 {
2116         int ret = 0;
2117         int i;
2118         u64 *i_qgroups;
2119         struct btrfs_root *quota_root = fs_info->quota_root;
2120         struct btrfs_qgroup *srcgroup;
2121         struct btrfs_qgroup *dstgroup;
2122         u32 level_size = 0;
2123         u64 nums;
2124
2125         mutex_lock(&fs_info->qgroup_ioctl_lock);
2126         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2127                 goto out;
2128
2129         if (!quota_root) {
2130                 ret = -EINVAL;
2131                 goto out;
2132         }
2133
2134         if (inherit) {
2135                 i_qgroups = (u64 *)(inherit + 1);
2136                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2137                        2 * inherit->num_excl_copies;
2138                 for (i = 0; i < nums; ++i) {
2139                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2140
2141                         /*
2142                          * Zero out invalid groups so we can ignore
2143                          * them later.
2144                          */
2145                         if (!srcgroup ||
2146                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2147                                 *i_qgroups = 0ULL;
2148
2149                         ++i_qgroups;
2150                 }
2151         }
2152
2153         /*
2154          * create a tracking group for the subvol itself
2155          */
2156         ret = add_qgroup_item(trans, quota_root, objectid);
2157         if (ret)
2158                 goto out;
2159
2160         if (srcid) {
2161                 struct btrfs_root *srcroot;
2162                 struct btrfs_key srckey;
2163
2164                 srckey.objectid = srcid;
2165                 srckey.type = BTRFS_ROOT_ITEM_KEY;
2166                 srckey.offset = (u64)-1;
2167                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2168                 if (IS_ERR(srcroot)) {
2169                         ret = PTR_ERR(srcroot);
2170                         goto out;
2171                 }
2172
2173                 rcu_read_lock();
2174                 level_size = fs_info->nodesize;
2175                 rcu_read_unlock();
2176         }
2177
2178         /*
2179          * add qgroup to all inherited groups
2180          */
2181         if (inherit) {
2182                 i_qgroups = (u64 *)(inherit + 1);
2183                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2184                         if (*i_qgroups == 0)
2185                                 continue;
2186                         ret = add_qgroup_relation_item(trans, quota_root,
2187                                                        objectid, *i_qgroups);
2188                         if (ret && ret != -EEXIST)
2189                                 goto out;
2190                         ret = add_qgroup_relation_item(trans, quota_root,
2191                                                        *i_qgroups, objectid);
2192                         if (ret && ret != -EEXIST)
2193                                 goto out;
2194                 }
2195                 ret = 0;
2196         }
2197
2198
2199         spin_lock(&fs_info->qgroup_lock);
2200
2201         dstgroup = add_qgroup_rb(fs_info, objectid);
2202         if (IS_ERR(dstgroup)) {
2203                 ret = PTR_ERR(dstgroup);
2204                 goto unlock;
2205         }
2206
2207         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2208                 dstgroup->lim_flags = inherit->lim.flags;
2209                 dstgroup->max_rfer = inherit->lim.max_rfer;
2210                 dstgroup->max_excl = inherit->lim.max_excl;
2211                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2212                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2213
2214                 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2215                 if (ret) {
2216                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2217                         btrfs_info(fs_info,
2218                                    "unable to update quota limit for %llu",
2219                                    dstgroup->qgroupid);
2220                         goto unlock;
2221                 }
2222         }
2223
2224         if (srcid) {
2225                 srcgroup = find_qgroup_rb(fs_info, srcid);
2226                 if (!srcgroup)
2227                         goto unlock;
2228
2229                 /*
2230                  * We call inherit after we clone the root in order to make sure
2231                  * our counts don't go crazy, so at this point the only
2232                  * difference between the two roots should be the root node.
2233                  */
2234                 dstgroup->rfer = srcgroup->rfer;
2235                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2236                 dstgroup->excl = level_size;
2237                 dstgroup->excl_cmpr = level_size;
2238                 srcgroup->excl = level_size;
2239                 srcgroup->excl_cmpr = level_size;
2240
2241                 /* inherit the limit info */
2242                 dstgroup->lim_flags = srcgroup->lim_flags;
2243                 dstgroup->max_rfer = srcgroup->max_rfer;
2244                 dstgroup->max_excl = srcgroup->max_excl;
2245                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2246                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2247
2248                 qgroup_dirty(fs_info, dstgroup);
2249                 qgroup_dirty(fs_info, srcgroup);
2250         }
2251
2252         if (!inherit)
2253                 goto unlock;
2254
2255         i_qgroups = (u64 *)(inherit + 1);
2256         for (i = 0; i < inherit->num_qgroups; ++i) {
2257                 if (*i_qgroups) {
2258                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2259                         if (ret)
2260                                 goto unlock;
2261                 }
2262                 ++i_qgroups;
2263         }
2264
2265         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2266                 struct btrfs_qgroup *src;
2267                 struct btrfs_qgroup *dst;
2268
2269                 if (!i_qgroups[0] || !i_qgroups[1])
2270                         continue;
2271
2272                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2273                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2274
2275                 if (!src || !dst) {
2276                         ret = -EINVAL;
2277                         goto unlock;
2278                 }
2279
2280                 dst->rfer = src->rfer - level_size;
2281                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2282         }
2283         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2284                 struct btrfs_qgroup *src;
2285                 struct btrfs_qgroup *dst;
2286
2287                 if (!i_qgroups[0] || !i_qgroups[1])
2288                         continue;
2289
2290                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2291                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2292
2293                 if (!src || !dst) {
2294                         ret = -EINVAL;
2295                         goto unlock;
2296                 }
2297
2298                 dst->excl = src->excl + level_size;
2299                 dst->excl_cmpr = src->excl_cmpr + level_size;
2300         }
2301
2302 unlock:
2303         spin_unlock(&fs_info->qgroup_lock);
2304 out:
2305         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2306         return ret;
2307 }
2308
2309 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
2310 {
2311         struct btrfs_root *quota_root;
2312         struct btrfs_qgroup *qgroup;
2313         struct btrfs_fs_info *fs_info = root->fs_info;
2314         u64 ref_root = root->root_key.objectid;
2315         int ret = 0;
2316         struct ulist_node *unode;
2317         struct ulist_iterator uiter;
2318
2319         if (!is_fstree(ref_root))
2320                 return 0;
2321
2322         if (num_bytes == 0)
2323                 return 0;
2324
2325         spin_lock(&fs_info->qgroup_lock);
2326         quota_root = fs_info->quota_root;
2327         if (!quota_root)
2328                 goto out;
2329
2330         qgroup = find_qgroup_rb(fs_info, ref_root);
2331         if (!qgroup)
2332                 goto out;
2333
2334         /*
2335          * in a first step, we check all affected qgroups if any limits would
2336          * be exceeded
2337          */
2338         ulist_reinit(fs_info->qgroup_ulist);
2339         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2340                         (uintptr_t)qgroup, GFP_ATOMIC);
2341         if (ret < 0)
2342                 goto out;
2343         ULIST_ITER_INIT(&uiter);
2344         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2345                 struct btrfs_qgroup *qg;
2346                 struct btrfs_qgroup_list *glist;
2347
2348                 qg = unode_aux_to_qgroup(unode);
2349
2350                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2351                     qg->reserved + (s64)qg->rfer + num_bytes >
2352                     qg->max_rfer) {
2353                         ret = -EDQUOT;
2354                         goto out;
2355                 }
2356
2357                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2358                     qg->reserved + (s64)qg->excl + num_bytes >
2359                     qg->max_excl) {
2360                         ret = -EDQUOT;
2361                         goto out;
2362                 }
2363
2364                 list_for_each_entry(glist, &qg->groups, next_group) {
2365                         ret = ulist_add(fs_info->qgroup_ulist,
2366                                         glist->group->qgroupid,
2367                                         (uintptr_t)glist->group, GFP_ATOMIC);
2368                         if (ret < 0)
2369                                 goto out;
2370                 }
2371         }
2372         ret = 0;
2373         /*
2374          * no limits exceeded, now record the reservation into all qgroups
2375          */
2376         ULIST_ITER_INIT(&uiter);
2377         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2378                 struct btrfs_qgroup *qg;
2379
2380                 qg = unode_aux_to_qgroup(unode);
2381
2382                 qg->reserved += num_bytes;
2383         }
2384
2385 out:
2386         spin_unlock(&fs_info->qgroup_lock);
2387         return ret;
2388 }
2389
2390 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2391                                u64 ref_root, u64 num_bytes)
2392 {
2393         struct btrfs_root *quota_root;
2394         struct btrfs_qgroup *qgroup;
2395         struct ulist_node *unode;
2396         struct ulist_iterator uiter;
2397         int ret = 0;
2398
2399         if (!is_fstree(ref_root))
2400                 return;
2401
2402         if (num_bytes == 0)
2403                 return;
2404
2405         spin_lock(&fs_info->qgroup_lock);
2406
2407         quota_root = fs_info->quota_root;
2408         if (!quota_root)
2409                 goto out;
2410
2411         qgroup = find_qgroup_rb(fs_info, ref_root);
2412         if (!qgroup)
2413                 goto out;
2414
2415         ulist_reinit(fs_info->qgroup_ulist);
2416         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2417                         (uintptr_t)qgroup, GFP_ATOMIC);
2418         if (ret < 0)
2419                 goto out;
2420         ULIST_ITER_INIT(&uiter);
2421         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2422                 struct btrfs_qgroup *qg;
2423                 struct btrfs_qgroup_list *glist;
2424
2425                 qg = unode_aux_to_qgroup(unode);
2426
2427                 qg->reserved -= num_bytes;
2428
2429                 list_for_each_entry(glist, &qg->groups, next_group) {
2430                         ret = ulist_add(fs_info->qgroup_ulist,
2431                                         glist->group->qgroupid,
2432                                         (uintptr_t)glist->group, GFP_ATOMIC);
2433                         if (ret < 0)
2434                                 goto out;
2435                 }
2436         }
2437
2438 out:
2439         spin_unlock(&fs_info->qgroup_lock);
2440 }
2441
2442 static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes)
2443 {
2444         return btrfs_qgroup_free_refroot(root->fs_info, root->objectid,
2445                                          num_bytes);
2446 }
2447 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2448 {
2449         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2450                 return;
2451         btrfs_err(trans->fs_info,
2452                 "qgroups not uptodate in trans handle %p:  list is%s empty, seq is %#x.%x",
2453                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2454                 (u32)(trans->delayed_ref_elem.seq >> 32),
2455                 (u32)trans->delayed_ref_elem.seq);
2456         BUG();
2457 }
2458
2459 /*
2460  * returns < 0 on error, 0 when more leafs are to be scanned.
2461  * returns 1 when done.
2462  */
2463 static int
2464 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2465                    struct btrfs_trans_handle *trans)
2466 {
2467         struct btrfs_key found;
2468         struct extent_buffer *scratch_leaf = NULL;
2469         struct ulist *roots = NULL;
2470         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2471         u64 num_bytes;
2472         int slot;
2473         int ret;
2474
2475         mutex_lock(&fs_info->qgroup_rescan_lock);
2476         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2477                                          &fs_info->qgroup_rescan_progress,
2478                                          path, 1, 0);
2479
2480         btrfs_debug(fs_info,
2481                 "current progress key (%llu %u %llu), search_slot ret %d",
2482                 fs_info->qgroup_rescan_progress.objectid,
2483                 fs_info->qgroup_rescan_progress.type,
2484                 fs_info->qgroup_rescan_progress.offset, ret);
2485
2486         if (ret) {
2487                 /*
2488                  * The rescan is about to end, we will not be scanning any
2489                  * further blocks. We cannot unset the RESCAN flag here, because
2490                  * we want to commit the transaction if everything went well.
2491                  * To make the live accounting work in this phase, we set our
2492                  * scan progress pointer such that every real extent objectid
2493                  * will be smaller.
2494                  */
2495                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2496                 btrfs_release_path(path);
2497                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2498                 return ret;
2499         }
2500
2501         btrfs_item_key_to_cpu(path->nodes[0], &found,
2502                               btrfs_header_nritems(path->nodes[0]) - 1);
2503         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2504
2505         btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2506         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2507         if (!scratch_leaf) {
2508                 ret = -ENOMEM;
2509                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2510                 goto out;
2511         }
2512         extent_buffer_get(scratch_leaf);
2513         btrfs_tree_read_lock(scratch_leaf);
2514         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2515         slot = path->slots[0];
2516         btrfs_release_path(path);
2517         mutex_unlock(&fs_info->qgroup_rescan_lock);
2518
2519         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2520                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2521                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2522                     found.type != BTRFS_METADATA_ITEM_KEY)
2523                         continue;
2524                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2525                         num_bytes = fs_info->nodesize;
2526                 else
2527                         num_bytes = found.offset;
2528
2529                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2530                                            &roots);
2531                 if (ret < 0)
2532                         goto out;
2533                 /* For rescan, just pass old_roots as NULL */
2534                 ret = btrfs_qgroup_account_extent(trans, fs_info,
2535                                 found.objectid, num_bytes, NULL, roots);
2536                 if (ret < 0)
2537                         goto out;
2538         }
2539 out:
2540         if (scratch_leaf) {
2541                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2542                 free_extent_buffer(scratch_leaf);
2543         }
2544         btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2545
2546         return ret;
2547 }
2548
2549 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2550 {
2551         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2552                                                      qgroup_rescan_work);
2553         struct btrfs_path *path;
2554         struct btrfs_trans_handle *trans = NULL;
2555         int err = -ENOMEM;
2556         int ret = 0;
2557
2558         path = btrfs_alloc_path();
2559         if (!path)
2560                 goto out;
2561
2562         err = 0;
2563         while (!err && !btrfs_fs_closing(fs_info)) {
2564                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2565                 if (IS_ERR(trans)) {
2566                         err = PTR_ERR(trans);
2567                         break;
2568                 }
2569                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2570                         err = -EINTR;
2571                 } else {
2572                         err = qgroup_rescan_leaf(fs_info, path, trans);
2573                 }
2574                 if (err > 0)
2575                         btrfs_commit_transaction(trans);
2576                 else
2577                         btrfs_end_transaction(trans);
2578         }
2579
2580 out:
2581         btrfs_free_path(path);
2582
2583         mutex_lock(&fs_info->qgroup_rescan_lock);
2584         if (!btrfs_fs_closing(fs_info))
2585                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2586
2587         if (err > 0 &&
2588             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2589                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2590         } else if (err < 0) {
2591                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2592         }
2593         mutex_unlock(&fs_info->qgroup_rescan_lock);
2594
2595         /*
2596          * only update status, since the previous part has already updated the
2597          * qgroup info.
2598          */
2599         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2600         if (IS_ERR(trans)) {
2601                 err = PTR_ERR(trans);
2602                 btrfs_err(fs_info,
2603                           "fail to start transaction for status update: %d\n",
2604                           err);
2605                 goto done;
2606         }
2607         ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2608         if (ret < 0) {
2609                 err = ret;
2610                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2611         }
2612         btrfs_end_transaction(trans);
2613
2614         if (btrfs_fs_closing(fs_info)) {
2615                 btrfs_info(fs_info, "qgroup scan paused");
2616         } else if (err >= 0) {
2617                 btrfs_info(fs_info, "qgroup scan completed%s",
2618                         err > 0 ? " (inconsistency flag cleared)" : "");
2619         } else {
2620                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2621         }
2622
2623 done:
2624         mutex_lock(&fs_info->qgroup_rescan_lock);
2625         fs_info->qgroup_rescan_running = false;
2626         mutex_unlock(&fs_info->qgroup_rescan_lock);
2627         complete_all(&fs_info->qgroup_rescan_completion);
2628 }
2629
2630 /*
2631  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2632  * memory required for the rescan context.
2633  */
2634 static int
2635 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2636                    int init_flags)
2637 {
2638         int ret = 0;
2639
2640         if (!init_flags &&
2641             (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2642              !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2643                 ret = -EINVAL;
2644                 goto err;
2645         }
2646
2647         mutex_lock(&fs_info->qgroup_rescan_lock);
2648         spin_lock(&fs_info->qgroup_lock);
2649
2650         if (init_flags) {
2651                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2652                         ret = -EINPROGRESS;
2653                 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2654                         ret = -EINVAL;
2655
2656                 if (ret) {
2657                         spin_unlock(&fs_info->qgroup_lock);
2658                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2659                         goto err;
2660                 }
2661                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2662         }
2663
2664         memset(&fs_info->qgroup_rescan_progress, 0,
2665                 sizeof(fs_info->qgroup_rescan_progress));
2666         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2667         init_completion(&fs_info->qgroup_rescan_completion);
2668         fs_info->qgroup_rescan_running = true;
2669
2670         spin_unlock(&fs_info->qgroup_lock);
2671         mutex_unlock(&fs_info->qgroup_rescan_lock);
2672
2673         memset(&fs_info->qgroup_rescan_work, 0,
2674                sizeof(fs_info->qgroup_rescan_work));
2675         btrfs_init_work(&fs_info->qgroup_rescan_work,
2676                         btrfs_qgroup_rescan_helper,
2677                         btrfs_qgroup_rescan_worker, NULL, NULL);
2678
2679         if (ret) {
2680 err:
2681                 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2682                 return ret;
2683         }
2684
2685         return 0;
2686 }
2687
2688 static void
2689 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2690 {
2691         struct rb_node *n;
2692         struct btrfs_qgroup *qgroup;
2693
2694         spin_lock(&fs_info->qgroup_lock);
2695         /* clear all current qgroup tracking information */
2696         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2697                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2698                 qgroup->rfer = 0;
2699                 qgroup->rfer_cmpr = 0;
2700                 qgroup->excl = 0;
2701                 qgroup->excl_cmpr = 0;
2702         }
2703         spin_unlock(&fs_info->qgroup_lock);
2704 }
2705
2706 int
2707 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2708 {
2709         int ret = 0;
2710         struct btrfs_trans_handle *trans;
2711
2712         ret = qgroup_rescan_init(fs_info, 0, 1);
2713         if (ret)
2714                 return ret;
2715
2716         /*
2717          * We have set the rescan_progress to 0, which means no more
2718          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2719          * However, btrfs_qgroup_account_ref may be right after its call
2720          * to btrfs_find_all_roots, in which case it would still do the
2721          * accounting.
2722          * To solve this, we're committing the transaction, which will
2723          * ensure we run all delayed refs and only after that, we are
2724          * going to clear all tracking information for a clean start.
2725          */
2726
2727         trans = btrfs_join_transaction(fs_info->fs_root);
2728         if (IS_ERR(trans)) {
2729                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2730                 return PTR_ERR(trans);
2731         }
2732         ret = btrfs_commit_transaction(trans);
2733         if (ret) {
2734                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2735                 return ret;
2736         }
2737
2738         qgroup_rescan_zero_tracking(fs_info);
2739
2740         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2741                          &fs_info->qgroup_rescan_work);
2742
2743         return 0;
2744 }
2745
2746 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2747                                      bool interruptible)
2748 {
2749         int running;
2750         int ret = 0;
2751
2752         mutex_lock(&fs_info->qgroup_rescan_lock);
2753         spin_lock(&fs_info->qgroup_lock);
2754         running = fs_info->qgroup_rescan_running;
2755         spin_unlock(&fs_info->qgroup_lock);
2756         mutex_unlock(&fs_info->qgroup_rescan_lock);
2757
2758         if (!running)
2759                 return 0;
2760
2761         if (interruptible)
2762                 ret = wait_for_completion_interruptible(
2763                                         &fs_info->qgroup_rescan_completion);
2764         else
2765                 wait_for_completion(&fs_info->qgroup_rescan_completion);
2766
2767         return ret;
2768 }
2769
2770 /*
2771  * this is only called from open_ctree where we're still single threaded, thus
2772  * locking is omitted here.
2773  */
2774 void
2775 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2776 {
2777         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2778                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2779                                  &fs_info->qgroup_rescan_work);
2780 }
2781
2782 /*
2783  * Reserve qgroup space for range [start, start + len).
2784  *
2785  * This function will either reserve space from related qgroups or doing
2786  * nothing if the range is already reserved.
2787  *
2788  * Return 0 for successful reserve
2789  * Return <0 for error (including -EQUOT)
2790  *
2791  * NOTE: this function may sleep for memory allocation.
2792  */
2793 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2794 {
2795         struct btrfs_root *root = BTRFS_I(inode)->root;
2796         struct extent_changeset changeset;
2797         struct ulist_node *unode;
2798         struct ulist_iterator uiter;
2799         int ret;
2800
2801         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2802             !is_fstree(root->objectid) || len == 0)
2803                 return 0;
2804
2805         changeset.bytes_changed = 0;
2806         changeset.range_changed = ulist_alloc(GFP_NOFS);
2807         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2808                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2809         trace_btrfs_qgroup_reserve_data(inode, start, len,
2810                                         changeset.bytes_changed,
2811                                         QGROUP_RESERVE);
2812         if (ret < 0)
2813                 goto cleanup;
2814         ret = qgroup_reserve(root, changeset.bytes_changed);
2815         if (ret < 0)
2816                 goto cleanup;
2817
2818         ulist_free(changeset.range_changed);
2819         return ret;
2820
2821 cleanup:
2822         /* cleanup already reserved ranges */
2823         ULIST_ITER_INIT(&uiter);
2824         while ((unode = ulist_next(changeset.range_changed, &uiter)))
2825                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2826                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2827                                  GFP_NOFS);
2828         ulist_free(changeset.range_changed);
2829         return ret;
2830 }
2831
2832 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2833                                        int free)
2834 {
2835         struct extent_changeset changeset;
2836         int trace_op = QGROUP_RELEASE;
2837         int ret;
2838
2839         changeset.bytes_changed = 0;
2840         changeset.range_changed = ulist_alloc(GFP_NOFS);
2841         if (!changeset.range_changed)
2842                 return -ENOMEM;
2843
2844         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2845                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2846         if (ret < 0)
2847                 goto out;
2848
2849         if (free) {
2850                 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2851                 trace_op = QGROUP_FREE;
2852         }
2853         trace_btrfs_qgroup_release_data(inode, start, len,
2854                                         changeset.bytes_changed, trace_op);
2855 out:
2856         ulist_free(changeset.range_changed);
2857         return ret;
2858 }
2859
2860 /*
2861  * Free a reserved space range from io_tree and related qgroups
2862  *
2863  * Should be called when a range of pages get invalidated before reaching disk.
2864  * Or for error cleanup case.
2865  *
2866  * For data written to disk, use btrfs_qgroup_release_data().
2867  *
2868  * NOTE: This function may sleep for memory allocation.
2869  */
2870 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2871 {
2872         return __btrfs_qgroup_release_data(inode, start, len, 1);
2873 }
2874
2875 /*
2876  * Release a reserved space range from io_tree only.
2877  *
2878  * Should be called when a range of pages get written to disk and corresponding
2879  * FILE_EXTENT is inserted into corresponding root.
2880  *
2881  * Since new qgroup accounting framework will only update qgroup numbers at
2882  * commit_transaction() time, its reserved space shouldn't be freed from
2883  * related qgroups.
2884  *
2885  * But we should release the range from io_tree, to allow further write to be
2886  * COWed.
2887  *
2888  * NOTE: This function may sleep for memory allocation.
2889  */
2890 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2891 {
2892         return __btrfs_qgroup_release_data(inode, start, len, 0);
2893 }
2894
2895 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes)
2896 {
2897         struct btrfs_fs_info *fs_info = root->fs_info;
2898         int ret;
2899
2900         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2901             !is_fstree(root->objectid) || num_bytes == 0)
2902                 return 0;
2903
2904         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2905         ret = qgroup_reserve(root, num_bytes);
2906         if (ret < 0)
2907                 return ret;
2908         atomic_add(num_bytes, &root->qgroup_meta_rsv);
2909         return ret;
2910 }
2911
2912 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2913 {
2914         struct btrfs_fs_info *fs_info = root->fs_info;
2915         int reserved;
2916
2917         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2918             !is_fstree(root->objectid))
2919                 return;
2920
2921         reserved = atomic_xchg(&root->qgroup_meta_rsv, 0);
2922         if (reserved == 0)
2923                 return;
2924         qgroup_free(root, reserved);
2925 }
2926
2927 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2928 {
2929         struct btrfs_fs_info *fs_info = root->fs_info;
2930
2931         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2932             !is_fstree(root->objectid))
2933                 return;
2934
2935         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2936         WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes);
2937         atomic_sub(num_bytes, &root->qgroup_meta_rsv);
2938         qgroup_free(root, num_bytes);
2939 }
2940
2941 /*
2942  * Check qgroup reserved space leaking, normally at destroy inode
2943  * time
2944  */
2945 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2946 {
2947         struct extent_changeset changeset;
2948         struct ulist_node *unode;
2949         struct ulist_iterator iter;
2950         int ret;
2951
2952         changeset.bytes_changed = 0;
2953         changeset.range_changed = ulist_alloc(GFP_NOFS);
2954         if (WARN_ON(!changeset.range_changed))
2955                 return;
2956
2957         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
2958                         EXTENT_QGROUP_RESERVED, &changeset);
2959
2960         WARN_ON(ret < 0);
2961         if (WARN_ON(changeset.bytes_changed)) {
2962                 ULIST_ITER_INIT(&iter);
2963                 while ((unode = ulist_next(changeset.range_changed, &iter))) {
2964                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
2965                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2966                                 inode->i_ino, unode->val, unode->aux);
2967                 }
2968                 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2969         }
2970         ulist_free(changeset.range_changed);
2971 }