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 #define ptr_to_u64(x) ((u64)(uintptr_t)x)
 135 #define u64_to_ptr(x) ((struct btrfs_qgroup *)(uintptr_t)x)
 136
 137 static int
 138 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 139                    int init_flags);
 140 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 141
 142 /* must be called with qgroup_ioctl_lock held */
 143 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 144                                            u64 qgroupid)
 145 {
 146         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 147         struct btrfs_qgroup *qgroup;
 148
 149         while (n) {
 150                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 151                 if (qgroup->qgroupid < qgroupid)
 152                         n = n->rb_left;
 153                 else if (qgroup->qgroupid > qgroupid)
 154                         n = n->rb_right;
 155                 else
 156                         return qgroup;
 157         }
 158         return NULL;
 159 }
 160
 161 /* must be called with qgroup_lock held */
 162 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 163                                           u64 qgroupid)
 164 {
 165         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 166         struct rb_node *parent = NULL;
 167         struct btrfs_qgroup *qgroup;
 168
 169         while (*p) {
 170                 parent = *p;
 171                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 172
 173                 if (qgroup->qgroupid < qgroupid)
 174                         p = &(*p)->rb_left;
 175                 else if (qgroup->qgroupid > qgroupid)
 176                         p = &(*p)->rb_right;
 177                 else
 178                         return qgroup;
 179         }
 180
 181         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 182         if (!qgroup)
 183                 return ERR_PTR(-ENOMEM);
 184
 185         qgroup->qgroupid = qgroupid;
 186         INIT_LIST_HEAD(&qgroup->groups);
 187         INIT_LIST_HEAD(&qgroup->members);
 188         INIT_LIST_HEAD(&qgroup->dirty);
 189
 190         rb_link_node(&qgroup->node, parent, p);
 191         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 192
 193         return qgroup;
 194 }
 195
 196 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
 197 {
 198         struct btrfs_qgroup_list *list;
 199
 200         list_del(&qgroup->dirty);
 201         while (!list_empty(&qgroup->groups)) {
 202                 list = list_first_entry(&qgroup->groups,
 203                                         struct btrfs_qgroup_list, next_group);
 204                 list_del(&list->next_group);
 205                 list_del(&list->next_member);
 206                 kfree(list);
 207         }
 208
 209         while (!list_empty(&qgroup->members)) {
 210                 list = list_first_entry(&qgroup->members,
 211                                         struct btrfs_qgroup_list, next_member);
 212                 list_del(&list->next_group);
 213                 list_del(&list->next_member);
 214                 kfree(list);
 215         }
 216         kfree(qgroup);
 217 }
 218
 219 /* must be called with qgroup_lock held */
 220 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 221 {
 222         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 223
 224         if (!qgroup)
 225                 return -ENOENT;
 226
 227         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 228         __del_qgroup_rb(qgroup);
 229         return 0;
 230 }
 231
 232 /* must be called with qgroup_lock held */
 233 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 234                            u64 memberid, u64 parentid)
 235 {
 236         struct btrfs_qgroup *member;
 237         struct btrfs_qgroup *parent;
 238         struct btrfs_qgroup_list *list;
 239
 240         member = find_qgroup_rb(fs_info, memberid);
 241         parent = find_qgroup_rb(fs_info, parentid);
 242         if (!member || !parent)
 243                 return -ENOENT;
 244
 245         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 246         if (!list)
 247                 return -ENOMEM;
 248
 249         list->group = parent;
 250         list->member = member;
 251         list_add_tail(&list->next_group, &member->groups);
 252         list_add_tail(&list->next_member, &parent->members);
 253
 254         return 0;
 255 }
 256
 257 /* must be called with qgroup_lock held */
 258 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 259                            u64 memberid, u64 parentid)
 260 {
 261         struct btrfs_qgroup *member;
 262         struct btrfs_qgroup *parent;
 263         struct btrfs_qgroup_list *list;
 264
 265         member = find_qgroup_rb(fs_info, memberid);
 266         parent = find_qgroup_rb(fs_info, parentid);
 267         if (!member || !parent)
 268                 return -ENOENT;
 269
 270         list_for_each_entry(list, &member->groups, next_group) {
 271                 if (list->group == parent) {
 272                         list_del(&list->next_group);
 273                         list_del(&list->next_member);
 274                         kfree(list);
 275                         return 0;
 276                 }
 277         }
 278         return -ENOENT;
 279 }
 280
 281 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 282 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 283                                u64 rfer, u64 excl)
 284 {
 285         struct btrfs_qgroup *qgroup;
 286
 287         qgroup = find_qgroup_rb(fs_info, qgroupid);
 288         if (!qgroup)
 289                 return -EINVAL;
 290         if (qgroup->rfer != rfer || qgroup->excl != excl)
 291                 return -EINVAL;
 292         return 0;
 293 }
 294 #endif
 295
 296 /*
 297  * The full config is read in one go, only called from open_ctree()
 298  * It doesn't use any locking, as at this point we're still single-threaded
 299  */
 300 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 301 {
 302         struct btrfs_key key;
 303         struct btrfs_key found_key;
 304         struct btrfs_root *quota_root = fs_info->quota_root;
 305         struct btrfs_path *path = NULL;
 306         struct extent_buffer *l;
 307         int slot;
 308         int ret = 0;
 309         u64 flags = 0;
 310         u64 rescan_progress = 0;
 311
 312         if (!fs_info->quota_enabled)
 313                 return 0;
 314
 315         fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
 316         if (!fs_info->qgroup_ulist) {
 317                 ret = -ENOMEM;
 318                 goto out;
 319         }
 320
 321         path = btrfs_alloc_path();
 322         if (!path) {
 323                 ret = -ENOMEM;
 324                 goto out;
 325         }
 326
 327         /* default this to quota off, in case no status key is found */
 328         fs_info->qgroup_flags = 0;
 329
 330         /*
 331          * pass 1: read status, all qgroup infos and limits
 332          */
 333         key.objectid = 0;
 334         key.type = 0;
 335         key.offset = 0;
 336         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 337         if (ret)
 338                 goto out;
 339
 340         while (1) {
 341                 struct btrfs_qgroup *qgroup;
 342
 343                 slot = path->slots[0];
 344                 l = path->nodes[0];
 345                 btrfs_item_key_to_cpu(l, &found_key, slot);
 346
 347                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 348                         struct btrfs_qgroup_status_item *ptr;
 349
 350                         ptr = btrfs_item_ptr(l, slot,
 351                                              struct btrfs_qgroup_status_item);
 352
 353                         if (btrfs_qgroup_status_version(l, ptr) !=
 354                             BTRFS_QGROUP_STATUS_VERSION) {
 355                                 btrfs_err(fs_info,
 356                                  "old qgroup version, quota disabled");
 357                                 goto out;
 358                         }
 359                         if (btrfs_qgroup_status_generation(l, ptr) !=
 360                             fs_info->generation) {
 361                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 362                                 btrfs_err(fs_info,
 363                                         "qgroup generation mismatch, "
 364                                         "marked as inconsistent");
 365                         }
 366                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 367                                                                           ptr);
 368                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 369                         goto next1;
 370                 }
 371
 372                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 373                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 374                         goto next1;
 375
 376                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 377                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 378                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 379                         btrfs_err(fs_info, "inconsistent qgroup config");
 380                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 381                 }
 382                 if (!qgroup) {
 383                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 384                         if (IS_ERR(qgroup)) {
 385                                 ret = PTR_ERR(qgroup);
 386                                 goto out;
 387                         }
 388                 }
 389                 switch (found_key.type) {
 390                 case BTRFS_QGROUP_INFO_KEY: {
 391                         struct btrfs_qgroup_info_item *ptr;
 392
 393                         ptr = btrfs_item_ptr(l, slot,
 394                                              struct btrfs_qgroup_info_item);
 395                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 396                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 397                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 398                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 399                         /* generation currently unused */
 400                         break;
 401                 }
 402                 case BTRFS_QGROUP_LIMIT_KEY: {
 403                         struct btrfs_qgroup_limit_item *ptr;
 404
 405                         ptr = btrfs_item_ptr(l, slot,
 406                                              struct btrfs_qgroup_limit_item);
 407                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 408                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 409                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 410                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 411                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 412                         break;
 413                 }
 414                 }
 415 next1:
 416                 ret = btrfs_next_item(quota_root, path);
 417                 if (ret < 0)
 418                         goto out;
 419                 if (ret)
 420                         break;
 421         }
 422         btrfs_release_path(path);
 423
 424         /*
 425          * pass 2: read all qgroup relations
 426          */
 427         key.objectid = 0;
 428         key.type = BTRFS_QGROUP_RELATION_KEY;
 429         key.offset = 0;
 430         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 431         if (ret)
 432                 goto out;
 433         while (1) {
 434                 slot = path->slots[0];
 435                 l = path->nodes[0];
 436                 btrfs_item_key_to_cpu(l, &found_key, slot);
 437
 438                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 439                         goto next2;
 440
 441                 if (found_key.objectid > found_key.offset) {
 442                         /* parent <- member, not needed to build config */
 443                         /* FIXME should we omit the key completely? */
 444                         goto next2;
 445                 }
 446
 447                 ret = add_relation_rb(fs_info, found_key.objectid,
 448                                       found_key.offset);
 449                 if (ret == -ENOENT) {
 450                         btrfs_warn(fs_info,
 451                                 "orphan qgroup relation 0x%llx->0x%llx",
 452                                 found_key.objectid, found_key.offset);
 453                         ret = 0;        /* ignore the error */
 454                 }
 455                 if (ret)
 456                         goto out;
 457 next2:
 458                 ret = btrfs_next_item(quota_root, path);
 459                 if (ret < 0)
 460                         goto out;
 461                 if (ret)
 462                         break;
 463         }
 464 out:
 465         fs_info->qgroup_flags |= flags;
 466         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
 467                 fs_info->quota_enabled = 0;
 468                 fs_info->pending_quota_state = 0;
 469         } else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 470                    ret >= 0) {
 471                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 472         }
 473         btrfs_free_path(path);
 474
 475         if (ret < 0) {
 476                 ulist_free(fs_info->qgroup_ulist);
 477                 fs_info->qgroup_ulist = NULL;
 478                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 479         }
 480
 481         return ret < 0 ? ret : 0;
 482 }
 483
 484 /*
 485  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 486  * first two are in single-threaded paths.And for the third one, we have set
 487  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 488  * up the in-memory structures without qgroup_lock held.
 489  */
 490 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 491 {
 492         struct rb_node *n;
 493         struct btrfs_qgroup *qgroup;
 494
 495         while ((n = rb_first(&fs_info->qgroup_tree))) {
 496                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 497                 rb_erase(n, &fs_info->qgroup_tree);
 498                 __del_qgroup_rb(qgroup);
 499         }
 500         /*
 501          * we call btrfs_free_qgroup_config() when umounting
 502          * filesystem and disabling quota, so we set qgroup_ulist
 503          * to be null here to avoid double free.
 504          */
 505         ulist_free(fs_info->qgroup_ulist);
 506         fs_info->qgroup_ulist = NULL;
 507 }
 508
 509 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
 510                                     struct btrfs_root *quota_root,
 511                                     u64 src, u64 dst)
 512 {
 513         int ret;
 514         struct btrfs_path *path;
 515         struct btrfs_key key;
 516
 517         path = btrfs_alloc_path();
 518         if (!path)
 519                 return -ENOMEM;
 520
 521         key.objectid = src;
 522         key.type = BTRFS_QGROUP_RELATION_KEY;
 523         key.offset = dst;
 524
 525         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 526
 527         btrfs_mark_buffer_dirty(path->nodes[0]);
 528
 529         btrfs_free_path(path);
 530         return ret;
 531 }
 532
 533 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
 534                                     struct btrfs_root *quota_root,
 535                                     u64 src, u64 dst)
 536 {
 537         int ret;
 538         struct btrfs_path *path;
 539         struct btrfs_key key;
 540
 541         path = btrfs_alloc_path();
 542         if (!path)
 543                 return -ENOMEM;
 544
 545         key.objectid = src;
 546         key.type = BTRFS_QGROUP_RELATION_KEY;
 547         key.offset = dst;
 548
 549         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 550         if (ret < 0)
 551                 goto out;
 552
 553         if (ret > 0) {
 554                 ret = -ENOENT;
 555                 goto out;
 556         }
 557
 558         ret = btrfs_del_item(trans, quota_root, path);
 559 out:
 560         btrfs_free_path(path);
 561         return ret;
 562 }
 563
 564 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 565                            struct btrfs_root *quota_root, u64 qgroupid)
 566 {
 567         int ret;
 568         struct btrfs_path *path;
 569         struct btrfs_qgroup_info_item *qgroup_info;
 570         struct btrfs_qgroup_limit_item *qgroup_limit;
 571         struct extent_buffer *leaf;
 572         struct btrfs_key key;
 573
 574         if (btrfs_is_testing(quota_root->fs_info))
 575                 return 0;
 576
 577         path = btrfs_alloc_path();
 578         if (!path)
 579                 return -ENOMEM;
 580
 581         key.objectid = 0;
 582         key.type = BTRFS_QGROUP_INFO_KEY;
 583         key.offset = qgroupid;
 584
 585         /*
 586          * Avoid a transaction abort by catching -EEXIST here. In that
 587          * case, we proceed by re-initializing the existing structure
 588          * on disk.
 589          */
 590
 591         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 592                                       sizeof(*qgroup_info));
 593         if (ret && ret != -EEXIST)
 594                 goto out;
 595
 596         leaf = path->nodes[0];
 597         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 598                                  struct btrfs_qgroup_info_item);
 599         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 600         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 601         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 602         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 603         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 604
 605         btrfs_mark_buffer_dirty(leaf);
 606
 607         btrfs_release_path(path);
 608
 609         key.type = BTRFS_QGROUP_LIMIT_KEY;
 610         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 611                                       sizeof(*qgroup_limit));
 612         if (ret && ret != -EEXIST)
 613                 goto out;
 614
 615         leaf = path->nodes[0];
 616         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 617                                   struct btrfs_qgroup_limit_item);
 618         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 619         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 620         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 621         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 622         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 623
 624         btrfs_mark_buffer_dirty(leaf);
 625
 626         ret = 0;
 627 out:
 628         btrfs_free_path(path);
 629         return ret;
 630 }
 631
 632 static int del_qgroup_item(struct btrfs_trans_handle *trans,
 633                            struct btrfs_root *quota_root, u64 qgroupid)
 634 {
 635         int ret;
 636         struct btrfs_path *path;
 637         struct btrfs_key key;
 638
 639         path = btrfs_alloc_path();
 640         if (!path)
 641                 return -ENOMEM;
 642
 643         key.objectid = 0;
 644         key.type = BTRFS_QGROUP_INFO_KEY;
 645         key.offset = qgroupid;
 646         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 647         if (ret < 0)
 648                 goto out;
 649
 650         if (ret > 0) {
 651                 ret = -ENOENT;
 652                 goto out;
 653         }
 654
 655         ret = btrfs_del_item(trans, quota_root, path);
 656         if (ret)
 657                 goto out;
 658
 659         btrfs_release_path(path);
 660
 661         key.type = BTRFS_QGROUP_LIMIT_KEY;
 662         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 663         if (ret < 0)
 664                 goto out;
 665
 666         if (ret > 0) {
 667                 ret = -ENOENT;
 668                 goto out;
 669         }
 670
 671         ret = btrfs_del_item(trans, quota_root, path);
 672
 673 out:
 674         btrfs_free_path(path);
 675         return ret;
 676 }
 677
 678 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 679                                     struct btrfs_root *root,
 680                                     struct btrfs_qgroup *qgroup)
 681 {
 682         struct btrfs_path *path;
 683         struct btrfs_key key;
 684         struct extent_buffer *l;
 685         struct btrfs_qgroup_limit_item *qgroup_limit;
 686         int ret;
 687         int slot;
 688
 689         key.objectid = 0;
 690         key.type = BTRFS_QGROUP_LIMIT_KEY;
 691         key.offset = qgroup->qgroupid;
 692
 693         path = btrfs_alloc_path();
 694         if (!path)
 695                 return -ENOMEM;
 696
 697         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 698         if (ret > 0)
 699                 ret = -ENOENT;
 700
 701         if (ret)
 702                 goto out;
 703
 704         l = path->nodes[0];
 705         slot = path->slots[0];
 706         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 707         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 708         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 709         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 710         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 711         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 712
 713         btrfs_mark_buffer_dirty(l);
 714
 715 out:
 716         btrfs_free_path(path);
 717         return ret;
 718 }
 719
 720 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 721                                    struct btrfs_root *root,
 722                                    struct btrfs_qgroup *qgroup)
 723 {
 724         struct btrfs_path *path;
 725         struct btrfs_key key;
 726         struct extent_buffer *l;
 727         struct btrfs_qgroup_info_item *qgroup_info;
 728         int ret;
 729         int slot;
 730
 731         if (btrfs_is_testing(root->fs_info))
 732                 return 0;
 733
 734         key.objectid = 0;
 735         key.type = BTRFS_QGROUP_INFO_KEY;
 736         key.offset = qgroup->qgroupid;
 737
 738         path = btrfs_alloc_path();
 739         if (!path)
 740                 return -ENOMEM;
 741
 742         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 743         if (ret > 0)
 744                 ret = -ENOENT;
 745
 746         if (ret)
 747                 goto out;
 748
 749         l = path->nodes[0];
 750         slot = path->slots[0];
 751         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 752         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 753         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 754         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 755         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 756         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 757
 758         btrfs_mark_buffer_dirty(l);
 759
 760 out:
 761         btrfs_free_path(path);
 762         return ret;
 763 }
 764
 765 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 766                                      struct btrfs_fs_info *fs_info,
 767                                     struct btrfs_root *root)
 768 {
 769         struct btrfs_path *path;
 770         struct btrfs_key key;
 771         struct extent_buffer *l;
 772         struct btrfs_qgroup_status_item *ptr;
 773         int ret;
 774         int slot;
 775
 776         key.objectid = 0;
 777         key.type = BTRFS_QGROUP_STATUS_KEY;
 778         key.offset = 0;
 779
 780         path = btrfs_alloc_path();
 781         if (!path)
 782                 return -ENOMEM;
 783
 784         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 785         if (ret > 0)
 786                 ret = -ENOENT;
 787
 788         if (ret)
 789                 goto out;
 790
 791         l = path->nodes[0];
 792         slot = path->slots[0];
 793         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 794         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 795         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 796         btrfs_set_qgroup_status_rescan(l, ptr,
 797                                 fs_info->qgroup_rescan_progress.objectid);
 798
 799         btrfs_mark_buffer_dirty(l);
 800
 801 out:
 802         btrfs_free_path(path);
 803         return ret;
 804 }
 805
 806 /*
 807  * called with qgroup_lock held
 808  */
 809 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 810                                   struct btrfs_root *root)
 811 {
 812         struct btrfs_path *path;
 813         struct btrfs_key key;
 814         struct extent_buffer *leaf = NULL;
 815         int ret;
 816         int nr = 0;
 817
 818         path = btrfs_alloc_path();
 819         if (!path)
 820                 return -ENOMEM;
 821
 822         path->leave_spinning = 1;
 823
 824         key.objectid = 0;
 825         key.offset = 0;
 826         key.type = 0;
 827
 828         while (1) {
 829                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 830                 if (ret < 0)
 831                         goto out;
 832                 leaf = path->nodes[0];
 833                 nr = btrfs_header_nritems(leaf);
 834                 if (!nr)
 835                         break;
 836                 /*
 837                  * delete the leaf one by one
 838                  * since the whole tree is going
 839                  * to be deleted.
 840                  */
 841                 path->slots[0] = 0;
 842                 ret = btrfs_del_items(trans, root, path, 0, nr);
 843                 if (ret)
 844                         goto out;
 845
 846                 btrfs_release_path(path);
 847         }
 848         ret = 0;
 849 out:
 850         root->fs_info->pending_quota_state = 0;
 851         btrfs_free_path(path);
 852         return ret;
 853 }
 854
 855 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 856                        struct btrfs_fs_info *fs_info)
 857 {
 858         struct btrfs_root *quota_root;
 859         struct btrfs_root *tree_root = fs_info->tree_root;
 860         struct btrfs_path *path = NULL;
 861         struct btrfs_qgroup_status_item *ptr;
 862         struct extent_buffer *leaf;
 863         struct btrfs_key key;
 864         struct btrfs_key found_key;
 865         struct btrfs_qgroup *qgroup = NULL;
 866         int ret = 0;
 867         int slot;
 868
 869         mutex_lock(&fs_info->qgroup_ioctl_lock);
 870         if (fs_info->quota_root) {
 871                 fs_info->pending_quota_state = 1;
 872                 goto out;
 873         }
 874
 875         fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
 876         if (!fs_info->qgroup_ulist) {
 877                 ret = -ENOMEM;
 878                 goto out;
 879         }
 880
 881         /*
 882          * initially create the quota tree
 883          */
 884         quota_root = btrfs_create_tree(trans, fs_info,
 885                                        BTRFS_QUOTA_TREE_OBJECTID);
 886         if (IS_ERR(quota_root)) {
 887                 ret =  PTR_ERR(quota_root);
 888                 goto out;
 889         }
 890
 891         path = btrfs_alloc_path();
 892         if (!path) {
 893                 ret = -ENOMEM;
 894                 goto out_free_root;
 895         }
 896
 897         key.objectid = 0;
 898         key.type = BTRFS_QGROUP_STATUS_KEY;
 899         key.offset = 0;
 900
 901         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 902                                       sizeof(*ptr));
 903         if (ret)
 904                 goto out_free_path;
 905
 906         leaf = path->nodes[0];
 907         ptr = btrfs_item_ptr(leaf, path->slots[0],
 908                                  struct btrfs_qgroup_status_item);
 909         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 910         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 911         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 912                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 913         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 914         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 915
 916         btrfs_mark_buffer_dirty(leaf);
 917
 918         key.objectid = 0;
 919         key.type = BTRFS_ROOT_REF_KEY;
 920         key.offset = 0;
 921
 922         btrfs_release_path(path);
 923         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 924         if (ret > 0)
 925                 goto out_add_root;
 926         if (ret < 0)
 927                 goto out_free_path;
 928
 929
 930         while (1) {
 931                 slot = path->slots[0];
 932                 leaf = path->nodes[0];
 933                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 934
 935                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 936                         ret = add_qgroup_item(trans, quota_root,
 937                                               found_key.offset);
 938                         if (ret)
 939                                 goto out_free_path;
 940
 941                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 942                         if (IS_ERR(qgroup)) {
 943                                 ret = PTR_ERR(qgroup);
 944                                 goto out_free_path;
 945                         }
 946                 }
 947                 ret = btrfs_next_item(tree_root, path);
 948                 if (ret < 0)
 949                         goto out_free_path;
 950                 if (ret)
 951                         break;
 952         }
 953
 954 out_add_root:
 955         btrfs_release_path(path);
 956         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
 957         if (ret)
 958                 goto out_free_path;
 959
 960         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
 961         if (IS_ERR(qgroup)) {
 962                 ret = PTR_ERR(qgroup);
 963                 goto out_free_path;
 964         }
 965         spin_lock(&fs_info->qgroup_lock);
 966         fs_info->quota_root = quota_root;
 967         fs_info->pending_quota_state = 1;
 968         spin_unlock(&fs_info->qgroup_lock);
 969 out_free_path:
 970         btrfs_free_path(path);
 971 out_free_root:
 972         if (ret) {
 973                 free_extent_buffer(quota_root->node);
 974                 free_extent_buffer(quota_root->commit_root);
 975                 kfree(quota_root);
 976         }
 977 out:
 978         if (ret) {
 979                 ulist_free(fs_info->qgroup_ulist);
 980                 fs_info->qgroup_ulist = NULL;
 981         }
 982         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 983         return ret;
 984 }
 985
 986 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
 987                         struct btrfs_fs_info *fs_info)
 988 {
 989         struct btrfs_root *tree_root = fs_info->tree_root;
 990         struct btrfs_root *quota_root;
 991         int ret = 0;
 992
 993         mutex_lock(&fs_info->qgroup_ioctl_lock);
 994         if (!fs_info->quota_root)
 995                 goto out;
 996         fs_info->quota_enabled = 0;
 997         fs_info->pending_quota_state = 0;
 998         btrfs_qgroup_wait_for_completion(fs_info);
 999         spin_lock(&fs_info->qgroup_lock);
1000         quota_root = fs_info->quota_root;
1001         fs_info->quota_root = NULL;
1002         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1003         spin_unlock(&fs_info->qgroup_lock);
1004
1005         btrfs_free_qgroup_config(fs_info);
1006
1007         ret = btrfs_clean_quota_tree(trans, quota_root);
1008         if (ret)
1009                 goto out;
1010
1011         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
1012         if (ret)
1013                 goto out;
1014
1015         list_del(&quota_root->dirty_list);
1016
1017         btrfs_tree_lock(quota_root->node);
1018         clean_tree_block(trans, tree_root->fs_info, quota_root->node);
1019         btrfs_tree_unlock(quota_root->node);
1020         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1021
1022         free_extent_buffer(quota_root->node);
1023         free_extent_buffer(quota_root->commit_root);
1024         kfree(quota_root);
1025 out:
1026         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1027         return ret;
1028 }
1029
1030 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1031                          struct btrfs_qgroup *qgroup)
1032 {
1033         if (list_empty(&qgroup->dirty))
1034                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1035 }
1036
1037 /*
1038  * The easy accounting, if we are adding/removing the only ref for an extent
1039  * then this qgroup and all of the parent qgroups get their reference and
1040  * exclusive counts adjusted.
1041  *
1042  * Caller should hold fs_info->qgroup_lock.
1043  */
1044 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1045                                     struct ulist *tmp, u64 ref_root,
1046                                     u64 num_bytes, int sign)
1047 {
1048         struct btrfs_qgroup *qgroup;
1049         struct btrfs_qgroup_list *glist;
1050         struct ulist_node *unode;
1051         struct ulist_iterator uiter;
1052         int ret = 0;
1053
1054         qgroup = find_qgroup_rb(fs_info, ref_root);
1055         if (!qgroup)
1056                 goto out;
1057
1058         qgroup->rfer += sign * num_bytes;
1059         qgroup->rfer_cmpr += sign * num_bytes;
1060
1061         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1062         qgroup->excl += sign * num_bytes;
1063         qgroup->excl_cmpr += sign * num_bytes;
1064         if (sign > 0)
1065                 qgroup->reserved -= num_bytes;
1066
1067         qgroup_dirty(fs_info, qgroup);
1068
1069         /* Get all of the parent groups that contain this qgroup */
1070         list_for_each_entry(glist, &qgroup->groups, next_group) {
1071                 ret = ulist_add(tmp, glist->group->qgroupid,
1072                                 ptr_to_u64(glist->group), GFP_ATOMIC);
1073                 if (ret < 0)
1074                         goto out;
1075         }
1076
1077         /* Iterate all of the parents and adjust their reference counts */
1078         ULIST_ITER_INIT(&uiter);
1079         while ((unode = ulist_next(tmp, &uiter))) {
1080                 qgroup = u64_to_ptr(unode->aux);
1081                 qgroup->rfer += sign * num_bytes;
1082                 qgroup->rfer_cmpr += sign * num_bytes;
1083                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1084                 qgroup->excl += sign * num_bytes;
1085                 if (sign > 0)
1086                         qgroup->reserved -= num_bytes;
1087                 qgroup->excl_cmpr += sign * num_bytes;
1088                 qgroup_dirty(fs_info, qgroup);
1089
1090                 /* Add any parents of the parents */
1091                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1092                         ret = ulist_add(tmp, glist->group->qgroupid,
1093                                         ptr_to_u64(glist->group), GFP_ATOMIC);
1094                         if (ret < 0)
1095                                 goto out;
1096                 }
1097         }
1098         ret = 0;
1099 out:
1100         return ret;
1101 }
1102
1103
1104 /*
1105  * Quick path for updating qgroup with only excl refs.
1106  *
1107  * In that case, just update all parent will be enough.
1108  * Or we needs to do a full rescan.
1109  * Caller should also hold fs_info->qgroup_lock.
1110  *
1111  * Return 0 for quick update, return >0 for need to full rescan
1112  * and mark INCONSISTENT flag.
1113  * Return < 0 for other error.
1114  */
1115 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1116                                    struct ulist *tmp, u64 src, u64 dst,
1117                                    int sign)
1118 {
1119         struct btrfs_qgroup *qgroup;
1120         int ret = 1;
1121         int err = 0;
1122
1123         qgroup = find_qgroup_rb(fs_info, src);
1124         if (!qgroup)
1125                 goto out;
1126         if (qgroup->excl == qgroup->rfer) {
1127                 ret = 0;
1128                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1129                                                qgroup->excl, sign);
1130                 if (err < 0) {
1131                         ret = err;
1132                         goto out;
1133                 }
1134         }
1135 out:
1136         if (ret)
1137                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1138         return ret;
1139 }
1140
1141 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1142                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1143 {
1144         struct btrfs_root *quota_root;
1145         struct btrfs_qgroup *parent;
1146         struct btrfs_qgroup *member;
1147         struct btrfs_qgroup_list *list;
1148         struct ulist *tmp;
1149         int ret = 0;
1150
1151         /* Check the level of src and dst first */
1152         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1153                 return -EINVAL;
1154
1155         tmp = ulist_alloc(GFP_NOFS);
1156         if (!tmp)
1157                 return -ENOMEM;
1158
1159         mutex_lock(&fs_info->qgroup_ioctl_lock);
1160         quota_root = fs_info->quota_root;
1161         if (!quota_root) {
1162                 ret = -EINVAL;
1163                 goto out;
1164         }
1165         member = find_qgroup_rb(fs_info, src);
1166         parent = find_qgroup_rb(fs_info, dst);
1167         if (!member || !parent) {
1168                 ret = -EINVAL;
1169                 goto out;
1170         }
1171
1172         /* check if such qgroup relation exist firstly */
1173         list_for_each_entry(list, &member->groups, next_group) {
1174                 if (list->group == parent) {
1175                         ret = -EEXIST;
1176                         goto out;
1177                 }
1178         }
1179
1180         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1181         if (ret)
1182                 goto out;
1183
1184         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1185         if (ret) {
1186                 del_qgroup_relation_item(trans, quota_root, src, dst);
1187                 goto out;
1188         }
1189
1190         spin_lock(&fs_info->qgroup_lock);
1191         ret = add_relation_rb(quota_root->fs_info, src, dst);
1192         if (ret < 0) {
1193                 spin_unlock(&fs_info->qgroup_lock);
1194                 goto out;
1195         }
1196         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1197         spin_unlock(&fs_info->qgroup_lock);
1198 out:
1199         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1200         ulist_free(tmp);
1201         return ret;
1202 }
1203
1204 int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1205                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1206 {
1207         struct btrfs_root *quota_root;
1208         struct btrfs_qgroup *parent;
1209         struct btrfs_qgroup *member;
1210         struct btrfs_qgroup_list *list;
1211         struct ulist *tmp;
1212         int ret = 0;
1213         int err;
1214
1215         tmp = ulist_alloc(GFP_NOFS);
1216         if (!tmp)
1217                 return -ENOMEM;
1218
1219         quota_root = fs_info->quota_root;
1220         if (!quota_root) {
1221                 ret = -EINVAL;
1222                 goto out;
1223         }
1224
1225         member = find_qgroup_rb(fs_info, src);
1226         parent = find_qgroup_rb(fs_info, dst);
1227         if (!member || !parent) {
1228                 ret = -EINVAL;
1229                 goto out;
1230         }
1231
1232         /* check if such qgroup relation exist firstly */
1233         list_for_each_entry(list, &member->groups, next_group) {
1234                 if (list->group == parent)
1235                         goto exist;
1236         }
1237         ret = -ENOENT;
1238         goto out;
1239 exist:
1240         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1241         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1242         if (err && !ret)
1243                 ret = err;
1244
1245         spin_lock(&fs_info->qgroup_lock);
1246         del_relation_rb(fs_info, src, dst);
1247         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1248         spin_unlock(&fs_info->qgroup_lock);
1249 out:
1250         ulist_free(tmp);
1251         return ret;
1252 }
1253
1254 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1255                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1256 {
1257         int ret = 0;
1258
1259         mutex_lock(&fs_info->qgroup_ioctl_lock);
1260         ret = __del_qgroup_relation(trans, fs_info, src, dst);
1261         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1262
1263         return ret;
1264 }
1265
1266 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1267                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1268 {
1269         struct btrfs_root *quota_root;
1270         struct btrfs_qgroup *qgroup;
1271         int ret = 0;
1272
1273         mutex_lock(&fs_info->qgroup_ioctl_lock);
1274         quota_root = fs_info->quota_root;
1275         if (!quota_root) {
1276                 ret = -EINVAL;
1277                 goto out;
1278         }
1279         qgroup = find_qgroup_rb(fs_info, qgroupid);
1280         if (qgroup) {
1281                 ret = -EEXIST;
1282                 goto out;
1283         }
1284
1285         ret = add_qgroup_item(trans, quota_root, qgroupid);
1286         if (ret)
1287                 goto out;
1288
1289         spin_lock(&fs_info->qgroup_lock);
1290         qgroup = add_qgroup_rb(fs_info, qgroupid);
1291         spin_unlock(&fs_info->qgroup_lock);
1292
1293         if (IS_ERR(qgroup))
1294                 ret = PTR_ERR(qgroup);
1295 out:
1296         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1297         return ret;
1298 }
1299
1300 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1301                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1302 {
1303         struct btrfs_root *quota_root;
1304         struct btrfs_qgroup *qgroup;
1305         struct btrfs_qgroup_list *list;
1306         int ret = 0;
1307
1308         mutex_lock(&fs_info->qgroup_ioctl_lock);
1309         quota_root = fs_info->quota_root;
1310         if (!quota_root) {
1311                 ret = -EINVAL;
1312                 goto out;
1313         }
1314
1315         qgroup = find_qgroup_rb(fs_info, qgroupid);
1316         if (!qgroup) {
1317                 ret = -ENOENT;
1318                 goto out;
1319         } else {
1320                 /* check if there are no children of this qgroup */
1321                 if (!list_empty(&qgroup->members)) {
1322                         ret = -EBUSY;
1323                         goto out;
1324                 }
1325         }
1326         ret = del_qgroup_item(trans, quota_root, qgroupid);
1327
1328         while (!list_empty(&qgroup->groups)) {
1329                 list = list_first_entry(&qgroup->groups,
1330                                         struct btrfs_qgroup_list, next_group);
1331                 ret = __del_qgroup_relation(trans, fs_info,
1332                                            qgroupid,
1333                                            list->group->qgroupid);
1334                 if (ret)
1335                         goto out;
1336         }
1337
1338         spin_lock(&fs_info->qgroup_lock);
1339         del_qgroup_rb(quota_root->fs_info, qgroupid);
1340         spin_unlock(&fs_info->qgroup_lock);
1341 out:
1342         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1343         return ret;
1344 }
1345
1346 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1347                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1348                        struct btrfs_qgroup_limit *limit)
1349 {
1350         struct btrfs_root *quota_root;
1351         struct btrfs_qgroup *qgroup;
1352         int ret = 0;
1353         /* Sometimes we would want to clear the limit on this qgroup.
1354          * To meet this requirement, we treat the -1 as a special value
1355          * which tell kernel to clear the limit on this qgroup.
1356          */
1357         const u64 CLEAR_VALUE = -1;
1358
1359         mutex_lock(&fs_info->qgroup_ioctl_lock);
1360         quota_root = fs_info->quota_root;
1361         if (!quota_root) {
1362                 ret = -EINVAL;
1363                 goto out;
1364         }
1365
1366         qgroup = find_qgroup_rb(fs_info, qgroupid);
1367         if (!qgroup) {
1368                 ret = -ENOENT;
1369                 goto out;
1370         }
1371
1372         spin_lock(&fs_info->qgroup_lock);
1373         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1374                 if (limit->max_rfer == CLEAR_VALUE) {
1375                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1376                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1377                         qgroup->max_rfer = 0;
1378                 } else {
1379                         qgroup->max_rfer = limit->max_rfer;
1380                 }
1381         }
1382         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1383                 if (limit->max_excl == CLEAR_VALUE) {
1384                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1385                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1386                         qgroup->max_excl = 0;
1387                 } else {
1388                         qgroup->max_excl = limit->max_excl;
1389                 }
1390         }
1391         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1392                 if (limit->rsv_rfer == CLEAR_VALUE) {
1393                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1394                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1395                         qgroup->rsv_rfer = 0;
1396                 } else {
1397                         qgroup->rsv_rfer = limit->rsv_rfer;
1398                 }
1399         }
1400         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1401                 if (limit->rsv_excl == CLEAR_VALUE) {
1402                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1403                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1404                         qgroup->rsv_excl = 0;
1405                 } else {
1406                         qgroup->rsv_excl = limit->rsv_excl;
1407                 }
1408         }
1409         qgroup->lim_flags |= limit->flags;
1410
1411         spin_unlock(&fs_info->qgroup_lock);
1412
1413         ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1414         if (ret) {
1415                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1416                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1417                        qgroupid);
1418         }
1419
1420 out:
1421         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1422         return ret;
1423 }
1424
1425 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1426                                          struct btrfs_fs_info *fs_info)
1427 {
1428         struct btrfs_qgroup_extent_record *record;
1429         struct btrfs_delayed_ref_root *delayed_refs;
1430         struct rb_node *node;
1431         u64 qgroup_to_skip;
1432         int ret = 0;
1433
1434         delayed_refs = &trans->transaction->delayed_refs;
1435         qgroup_to_skip = delayed_refs->qgroup_to_skip;
1436
1437         /*
1438          * No need to do lock, since this function will only be called in
1439          * btrfs_commit_transaction().
1440          */
1441         node = rb_first(&delayed_refs->dirty_extent_root);
1442         while (node) {
1443                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1444                                   node);
1445                 ret = btrfs_find_all_roots(NULL, fs_info, record->bytenr, 0,
1446                                            &record->old_roots);
1447                 if (ret < 0)
1448                         break;
1449                 if (qgroup_to_skip)
1450                         ulist_del(record->old_roots, qgroup_to_skip, 0);
1451                 node = rb_next(node);
1452         }
1453         return ret;
1454 }
1455
1456 struct btrfs_qgroup_extent_record *
1457 btrfs_qgroup_insert_dirty_extent(struct btrfs_fs_info *fs_info,
1458                                  struct btrfs_delayed_ref_root *delayed_refs,
1459                                  struct btrfs_qgroup_extent_record *record)
1460 {
1461         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1462         struct rb_node *parent_node = NULL;
1463         struct btrfs_qgroup_extent_record *entry;
1464         u64 bytenr = record->bytenr;
1465
1466         assert_spin_locked(&delayed_refs->lock);
1467         trace_btrfs_qgroup_insert_dirty_extent(fs_info, record);
1468
1469         while (*p) {
1470                 parent_node = *p;
1471                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1472                                  node);
1473                 if (bytenr < entry->bytenr)
1474                         p = &(*p)->rb_left;
1475                 else if (bytenr > entry->bytenr)
1476                         p = &(*p)->rb_right;
1477                 else
1478                         return entry;
1479         }
1480
1481         rb_link_node(&record->node, parent_node, p);
1482         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1483         return NULL;
1484 }
1485
1486 #define UPDATE_NEW      0
1487 #define UPDATE_OLD      1
1488 /*
1489  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1490  */
1491 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1492                                 struct ulist *roots, struct ulist *tmp,
1493                                 struct ulist *qgroups, u64 seq, int update_old)
1494 {
1495         struct ulist_node *unode;
1496         struct ulist_iterator uiter;
1497         struct ulist_node *tmp_unode;
1498         struct ulist_iterator tmp_uiter;
1499         struct btrfs_qgroup *qg;
1500         int ret = 0;
1501
1502         if (!roots)
1503                 return 0;
1504         ULIST_ITER_INIT(&uiter);
1505         while ((unode = ulist_next(roots, &uiter))) {
1506                 qg = find_qgroup_rb(fs_info, unode->val);
1507                 if (!qg)
1508                         continue;
1509
1510                 ulist_reinit(tmp);
1511                 ret = ulist_add(qgroups, qg->qgroupid, ptr_to_u64(qg),
1512                                 GFP_ATOMIC);
1513                 if (ret < 0)
1514                         return ret;
1515                 ret = ulist_add(tmp, qg->qgroupid, ptr_to_u64(qg), GFP_ATOMIC);
1516                 if (ret < 0)
1517                         return ret;
1518                 ULIST_ITER_INIT(&tmp_uiter);
1519                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1520                         struct btrfs_qgroup_list *glist;
1521
1522                         qg = u64_to_ptr(tmp_unode->aux);
1523                         if (update_old)
1524                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1525                         else
1526                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1527                         list_for_each_entry(glist, &qg->groups, next_group) {
1528                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1529                                                 ptr_to_u64(glist->group),
1530                                                 GFP_ATOMIC);
1531                                 if (ret < 0)
1532                                         return ret;
1533                                 ret = ulist_add(tmp, glist->group->qgroupid,
1534                                                 ptr_to_u64(glist->group),
1535                                                 GFP_ATOMIC);
1536                                 if (ret < 0)
1537                                         return ret;
1538                         }
1539                 }
1540         }
1541         return 0;
1542 }
1543
1544 /*
1545  * Update qgroup rfer/excl counters.
1546  * Rfer update is easy, codes can explain themselves.
1547  *
1548  * Excl update is tricky, the update is split into 2 part.
1549  * Part 1: Possible exclusive <-> sharing detect:
1550  *      |       A       |       !A      |
1551  *  -------------------------------------
1552  *  B   |       *       |       -       |
1553  *  -------------------------------------
1554  *  !B  |       +       |       **      |
1555  *  -------------------------------------
1556  *
1557  * Conditions:
1558  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1559  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1560  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1561  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1562  *
1563  * Results:
1564  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1565  * *: Definitely not changed.           **: Possible unchanged.
1566  *
1567  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1568  *
1569  * To make the logic clear, we first use condition A and B to split
1570  * combination into 4 results.
1571  *
1572  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1573  * only on variant maybe 0.
1574  *
1575  * Lastly, check result **, since there are 2 variants maybe 0, split them
1576  * again(2x2).
1577  * But this time we don't need to consider other things, the codes and logic
1578  * is easy to understand now.
1579  */
1580 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1581                                   struct ulist *qgroups,
1582                                   u64 nr_old_roots,
1583                                   u64 nr_new_roots,
1584                                   u64 num_bytes, u64 seq)
1585 {
1586         struct ulist_node *unode;
1587         struct ulist_iterator uiter;
1588         struct btrfs_qgroup *qg;
1589         u64 cur_new_count, cur_old_count;
1590
1591         ULIST_ITER_INIT(&uiter);
1592         while ((unode = ulist_next(qgroups, &uiter))) {
1593                 bool dirty = false;
1594
1595                 qg = u64_to_ptr(unode->aux);
1596                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1597                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1598
1599                 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1600                                              cur_old_count, cur_new_count);
1601
1602                 /* Rfer update part */
1603                 if (cur_old_count == 0 && cur_new_count > 0) {
1604                         qg->rfer += num_bytes;
1605                         qg->rfer_cmpr += num_bytes;
1606                         dirty = true;
1607                 }
1608                 if (cur_old_count > 0 && cur_new_count == 0) {
1609                         qg->rfer -= num_bytes;
1610                         qg->rfer_cmpr -= num_bytes;
1611                         dirty = true;
1612                 }
1613
1614                 /* Excl update part */
1615                 /* Exclusive/none -> shared case */
1616                 if (cur_old_count == nr_old_roots &&
1617                     cur_new_count < nr_new_roots) {
1618                         /* Exclusive -> shared */
1619                         if (cur_old_count != 0) {
1620                                 qg->excl -= num_bytes;
1621                                 qg->excl_cmpr -= num_bytes;
1622                                 dirty = true;
1623                         }
1624                 }
1625
1626                 /* Shared -> exclusive/none case */
1627                 if (cur_old_count < nr_old_roots &&
1628                     cur_new_count == nr_new_roots) {
1629                         /* Shared->exclusive */
1630                         if (cur_new_count != 0) {
1631                                 qg->excl += num_bytes;
1632                                 qg->excl_cmpr += num_bytes;
1633                                 dirty = true;
1634                         }
1635                 }
1636
1637                 /* Exclusive/none -> exclusive/none case */
1638                 if (cur_old_count == nr_old_roots &&
1639                     cur_new_count == nr_new_roots) {
1640                         if (cur_old_count == 0) {
1641                                 /* None -> exclusive/none */
1642
1643                                 if (cur_new_count != 0) {
1644                                         /* None -> exclusive */
1645                                         qg->excl += num_bytes;
1646                                         qg->excl_cmpr += num_bytes;
1647                                         dirty = true;
1648                                 }
1649                                 /* None -> none, nothing changed */
1650                         } else {
1651                                 /* Exclusive -> exclusive/none */
1652
1653                                 if (cur_new_count == 0) {
1654                                         /* Exclusive -> none */
1655                                         qg->excl -= num_bytes;
1656                                         qg->excl_cmpr -= num_bytes;
1657                                         dirty = true;
1658                                 }
1659                                 /* Exclusive -> exclusive, nothing changed */
1660                         }
1661                 }
1662
1663                 if (dirty)
1664                         qgroup_dirty(fs_info, qg);
1665         }
1666         return 0;
1667 }
1668
1669 int
1670 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1671                             struct btrfs_fs_info *fs_info,
1672                             u64 bytenr, u64 num_bytes,
1673                             struct ulist *old_roots, struct ulist *new_roots)
1674 {
1675         struct ulist *qgroups = NULL;
1676         struct ulist *tmp = NULL;
1677         u64 seq;
1678         u64 nr_new_roots = 0;
1679         u64 nr_old_roots = 0;
1680         int ret = 0;
1681
1682         if (new_roots)
1683                 nr_new_roots = new_roots->nnodes;
1684         if (old_roots)
1685                 nr_old_roots = old_roots->nnodes;
1686
1687         if (!fs_info->quota_enabled)
1688                 goto out_free;
1689         BUG_ON(!fs_info->quota_root);
1690
1691         trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1692                                           nr_old_roots, nr_new_roots);
1693
1694         qgroups = ulist_alloc(GFP_NOFS);
1695         if (!qgroups) {
1696                 ret = -ENOMEM;
1697                 goto out_free;
1698         }
1699         tmp = ulist_alloc(GFP_NOFS);
1700         if (!tmp) {
1701                 ret = -ENOMEM;
1702                 goto out_free;
1703         }
1704
1705         mutex_lock(&fs_info->qgroup_rescan_lock);
1706         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1707                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1708                         mutex_unlock(&fs_info->qgroup_rescan_lock);
1709                         ret = 0;
1710                         goto out_free;
1711                 }
1712         }
1713         mutex_unlock(&fs_info->qgroup_rescan_lock);
1714
1715         spin_lock(&fs_info->qgroup_lock);
1716         seq = fs_info->qgroup_seq;
1717
1718         /* Update old refcnts using old_roots */
1719         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1720                                    UPDATE_OLD);
1721         if (ret < 0)
1722                 goto out;
1723
1724         /* Update new refcnts using new_roots */
1725         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1726                                    UPDATE_NEW);
1727         if (ret < 0)
1728                 goto out;
1729
1730         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1731                                num_bytes, seq);
1732
1733         /*
1734          * Bump qgroup_seq to avoid seq overlap
1735          */
1736         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1737 out:
1738         spin_unlock(&fs_info->qgroup_lock);
1739 out_free:
1740         ulist_free(tmp);
1741         ulist_free(qgroups);
1742         ulist_free(old_roots);
1743         ulist_free(new_roots);
1744         return ret;
1745 }
1746
1747 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
1748                                  struct btrfs_fs_info *fs_info)
1749 {
1750         struct btrfs_qgroup_extent_record *record;
1751         struct btrfs_delayed_ref_root *delayed_refs;
1752         struct ulist *new_roots = NULL;
1753         struct rb_node *node;
1754         u64 qgroup_to_skip;
1755         int ret = 0;
1756
1757         delayed_refs = &trans->transaction->delayed_refs;
1758         qgroup_to_skip = delayed_refs->qgroup_to_skip;
1759         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
1760                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1761                                   node);
1762
1763                 trace_btrfs_qgroup_account_extents(fs_info, record);
1764
1765                 if (!ret) {
1766                         /*
1767                          * Use (u64)-1 as time_seq to do special search, which
1768                          * doesn't lock tree or delayed_refs and search current
1769                          * root. It's safe inside commit_transaction().
1770                          */
1771                         ret = btrfs_find_all_roots(trans, fs_info,
1772                                         record->bytenr, (u64)-1, &new_roots);
1773                         if (ret < 0)
1774                                 goto cleanup;
1775                         if (qgroup_to_skip)
1776                                 ulist_del(new_roots, qgroup_to_skip, 0);
1777                         ret = btrfs_qgroup_account_extent(trans, fs_info,
1778                                         record->bytenr, record->num_bytes,
1779                                         record->old_roots, new_roots);
1780                         record->old_roots = NULL;
1781                         new_roots = NULL;
1782                 }
1783 cleanup:
1784                 ulist_free(record->old_roots);
1785                 ulist_free(new_roots);
1786                 new_roots = NULL;
1787                 rb_erase(node, &delayed_refs->dirty_extent_root);
1788                 kfree(record);
1789
1790         }
1791         return ret;
1792 }
1793
1794 /*
1795  * called from commit_transaction. Writes all changed qgroups to disk.
1796  */
1797 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1798                       struct btrfs_fs_info *fs_info)
1799 {
1800         struct btrfs_root *quota_root = fs_info->quota_root;
1801         int ret = 0;
1802         int start_rescan_worker = 0;
1803
1804         if (!quota_root)
1805                 goto out;
1806
1807         if (!fs_info->quota_enabled && fs_info->pending_quota_state)
1808                 start_rescan_worker = 1;
1809
1810         fs_info->quota_enabled = fs_info->pending_quota_state;
1811
1812         spin_lock(&fs_info->qgroup_lock);
1813         while (!list_empty(&fs_info->dirty_qgroups)) {
1814                 struct btrfs_qgroup *qgroup;
1815                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1816                                           struct btrfs_qgroup, dirty);
1817                 list_del_init(&qgroup->dirty);
1818                 spin_unlock(&fs_info->qgroup_lock);
1819                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1820                 if (ret)
1821                         fs_info->qgroup_flags |=
1822                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1823                 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1824                 if (ret)
1825                         fs_info->qgroup_flags |=
1826                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1827                 spin_lock(&fs_info->qgroup_lock);
1828         }
1829         if (fs_info->quota_enabled)
1830                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1831         else
1832                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1833         spin_unlock(&fs_info->qgroup_lock);
1834
1835         ret = update_qgroup_status_item(trans, fs_info, quota_root);
1836         if (ret)
1837                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1838
1839         if (!ret && start_rescan_worker) {
1840                 ret = qgroup_rescan_init(fs_info, 0, 1);
1841                 if (!ret) {
1842                         qgroup_rescan_zero_tracking(fs_info);
1843                         btrfs_queue_work(fs_info->qgroup_rescan_workers,
1844                                          &fs_info->qgroup_rescan_work);
1845                 }
1846                 ret = 0;
1847         }
1848
1849 out:
1850
1851         return ret;
1852 }
1853
1854 /*
1855  * Copy the accounting information between qgroups. This is necessary
1856  * when a snapshot or a subvolume is created. Throwing an error will
1857  * cause a transaction abort so we take extra care here to only error
1858  * when a readonly fs is a reasonable outcome.
1859  */
1860 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1861                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1862                          struct btrfs_qgroup_inherit *inherit)
1863 {
1864         int ret = 0;
1865         int i;
1866         u64 *i_qgroups;
1867         struct btrfs_root *quota_root = fs_info->quota_root;
1868         struct btrfs_qgroup *srcgroup;
1869         struct btrfs_qgroup *dstgroup;
1870         u32 level_size = 0;
1871         u64 nums;
1872
1873         mutex_lock(&fs_info->qgroup_ioctl_lock);
1874         if (!fs_info->quota_enabled)
1875                 goto out;
1876
1877         if (!quota_root) {
1878                 ret = -EINVAL;
1879                 goto out;
1880         }
1881
1882         if (inherit) {
1883                 i_qgroups = (u64 *)(inherit + 1);
1884                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1885                        2 * inherit->num_excl_copies;
1886                 for (i = 0; i < nums; ++i) {
1887                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1888
1889                         /*
1890                          * Zero out invalid groups so we can ignore
1891                          * them later.
1892                          */
1893                         if (!srcgroup ||
1894                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
1895                                 *i_qgroups = 0ULL;
1896
1897                         ++i_qgroups;
1898                 }
1899         }
1900
1901         /*
1902          * create a tracking group for the subvol itself
1903          */
1904         ret = add_qgroup_item(trans, quota_root, objectid);
1905         if (ret)
1906                 goto out;
1907
1908         if (srcid) {
1909                 struct btrfs_root *srcroot;
1910                 struct btrfs_key srckey;
1911
1912                 srckey.objectid = srcid;
1913                 srckey.type = BTRFS_ROOT_ITEM_KEY;
1914                 srckey.offset = (u64)-1;
1915                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1916                 if (IS_ERR(srcroot)) {
1917                         ret = PTR_ERR(srcroot);
1918                         goto out;
1919                 }
1920
1921                 rcu_read_lock();
1922                 level_size = srcroot->nodesize;
1923                 rcu_read_unlock();
1924         }
1925
1926         /*
1927          * add qgroup to all inherited groups
1928          */
1929         if (inherit) {
1930                 i_qgroups = (u64 *)(inherit + 1);
1931                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
1932                         if (*i_qgroups == 0)
1933                                 continue;
1934                         ret = add_qgroup_relation_item(trans, quota_root,
1935                                                        objectid, *i_qgroups);
1936                         if (ret && ret != -EEXIST)
1937                                 goto out;
1938                         ret = add_qgroup_relation_item(trans, quota_root,
1939                                                        *i_qgroups, objectid);
1940                         if (ret && ret != -EEXIST)
1941                                 goto out;
1942                 }
1943                 ret = 0;
1944         }
1945
1946
1947         spin_lock(&fs_info->qgroup_lock);
1948
1949         dstgroup = add_qgroup_rb(fs_info, objectid);
1950         if (IS_ERR(dstgroup)) {
1951                 ret = PTR_ERR(dstgroup);
1952                 goto unlock;
1953         }
1954
1955         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1956                 dstgroup->lim_flags = inherit->lim.flags;
1957                 dstgroup->max_rfer = inherit->lim.max_rfer;
1958                 dstgroup->max_excl = inherit->lim.max_excl;
1959                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
1960                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
1961
1962                 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
1963                 if (ret) {
1964                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1965                         btrfs_info(fs_info, "unable to update quota limit for %llu",
1966                                dstgroup->qgroupid);
1967                         goto unlock;
1968                 }
1969         }
1970
1971         if (srcid) {
1972                 srcgroup = find_qgroup_rb(fs_info, srcid);
1973                 if (!srcgroup)
1974                         goto unlock;
1975
1976                 /*
1977                  * We call inherit after we clone the root in order to make sure
1978                  * our counts don't go crazy, so at this point the only
1979                  * difference between the two roots should be the root node.
1980                  */
1981                 dstgroup->rfer = srcgroup->rfer;
1982                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
1983                 dstgroup->excl = level_size;
1984                 dstgroup->excl_cmpr = level_size;
1985                 srcgroup->excl = level_size;
1986                 srcgroup->excl_cmpr = level_size;
1987
1988                 /* inherit the limit info */
1989                 dstgroup->lim_flags = srcgroup->lim_flags;
1990                 dstgroup->max_rfer = srcgroup->max_rfer;
1991                 dstgroup->max_excl = srcgroup->max_excl;
1992                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
1993                 dstgroup->rsv_excl = srcgroup->rsv_excl;
1994
1995                 qgroup_dirty(fs_info, dstgroup);
1996                 qgroup_dirty(fs_info, srcgroup);
1997         }
1998
1999         if (!inherit)
2000                 goto unlock;
2001
2002         i_qgroups = (u64 *)(inherit + 1);
2003         for (i = 0; i < inherit->num_qgroups; ++i) {
2004                 if (*i_qgroups) {
2005                         ret = add_relation_rb(quota_root->fs_info, objectid,
2006                                               *i_qgroups);
2007                         if (ret)
2008                                 goto unlock;
2009                 }
2010                 ++i_qgroups;
2011         }
2012
2013         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2014                 struct btrfs_qgroup *src;
2015                 struct btrfs_qgroup *dst;
2016
2017                 if (!i_qgroups[0] || !i_qgroups[1])
2018                         continue;
2019
2020                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2021                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2022
2023                 if (!src || !dst) {
2024                         ret = -EINVAL;
2025                         goto unlock;
2026                 }
2027
2028                 dst->rfer = src->rfer - level_size;
2029                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2030         }
2031         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2032                 struct btrfs_qgroup *src;
2033                 struct btrfs_qgroup *dst;
2034
2035                 if (!i_qgroups[0] || !i_qgroups[1])
2036                         continue;
2037
2038                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2039                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2040
2041                 if (!src || !dst) {
2042                         ret = -EINVAL;
2043                         goto unlock;
2044                 }
2045
2046                 dst->excl = src->excl + level_size;
2047                 dst->excl_cmpr = src->excl_cmpr + level_size;
2048         }
2049
2050 unlock:
2051         spin_unlock(&fs_info->qgroup_lock);
2052 out:
2053         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2054         return ret;
2055 }
2056
2057 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
2058 {
2059         struct btrfs_root *quota_root;
2060         struct btrfs_qgroup *qgroup;
2061         struct btrfs_fs_info *fs_info = root->fs_info;
2062         u64 ref_root = root->root_key.objectid;
2063         int ret = 0;
2064         struct ulist_node *unode;
2065         struct ulist_iterator uiter;
2066
2067         if (!is_fstree(ref_root))
2068                 return 0;
2069
2070         if (num_bytes == 0)
2071                 return 0;
2072
2073         spin_lock(&fs_info->qgroup_lock);
2074         quota_root = fs_info->quota_root;
2075         if (!quota_root)
2076                 goto out;
2077
2078         qgroup = find_qgroup_rb(fs_info, ref_root);
2079         if (!qgroup)
2080                 goto out;
2081
2082         /*
2083          * in a first step, we check all affected qgroups if any limits would
2084          * be exceeded
2085          */
2086         ulist_reinit(fs_info->qgroup_ulist);
2087         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2088                         (uintptr_t)qgroup, GFP_ATOMIC);
2089         if (ret < 0)
2090                 goto out;
2091         ULIST_ITER_INIT(&uiter);
2092         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2093                 struct btrfs_qgroup *qg;
2094                 struct btrfs_qgroup_list *glist;
2095
2096                 qg = u64_to_ptr(unode->aux);
2097
2098                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2099                     qg->reserved + (s64)qg->rfer + num_bytes >
2100                     qg->max_rfer) {
2101                         ret = -EDQUOT;
2102                         goto out;
2103                 }
2104
2105                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2106                     qg->reserved + (s64)qg->excl + num_bytes >
2107                     qg->max_excl) {
2108                         ret = -EDQUOT;
2109                         goto out;
2110                 }
2111
2112                 list_for_each_entry(glist, &qg->groups, next_group) {
2113                         ret = ulist_add(fs_info->qgroup_ulist,
2114                                         glist->group->qgroupid,
2115                                         (uintptr_t)glist->group, GFP_ATOMIC);
2116                         if (ret < 0)
2117                                 goto out;
2118                 }
2119         }
2120         ret = 0;
2121         /*
2122          * no limits exceeded, now record the reservation into all qgroups
2123          */
2124         ULIST_ITER_INIT(&uiter);
2125         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2126                 struct btrfs_qgroup *qg;
2127
2128                 qg = u64_to_ptr(unode->aux);
2129
2130                 qg->reserved += num_bytes;
2131         }
2132
2133 out:
2134         spin_unlock(&fs_info->qgroup_lock);
2135         return ret;
2136 }
2137
2138 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2139                                u64 ref_root, u64 num_bytes)
2140 {
2141         struct btrfs_root *quota_root;
2142         struct btrfs_qgroup *qgroup;
2143         struct ulist_node *unode;
2144         struct ulist_iterator uiter;
2145         int ret = 0;
2146
2147         if (!is_fstree(ref_root))
2148                 return;
2149
2150         if (num_bytes == 0)
2151                 return;
2152
2153         spin_lock(&fs_info->qgroup_lock);
2154
2155         quota_root = fs_info->quota_root;
2156         if (!quota_root)
2157                 goto out;
2158
2159         qgroup = find_qgroup_rb(fs_info, ref_root);
2160         if (!qgroup)
2161                 goto out;
2162
2163         ulist_reinit(fs_info->qgroup_ulist);
2164         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2165                         (uintptr_t)qgroup, GFP_ATOMIC);
2166         if (ret < 0)
2167                 goto out;
2168         ULIST_ITER_INIT(&uiter);
2169         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2170                 struct btrfs_qgroup *qg;
2171                 struct btrfs_qgroup_list *glist;
2172
2173                 qg = u64_to_ptr(unode->aux);
2174
2175                 qg->reserved -= num_bytes;
2176
2177                 list_for_each_entry(glist, &qg->groups, next_group) {
2178                         ret = ulist_add(fs_info->qgroup_ulist,
2179                                         glist->group->qgroupid,
2180                                         (uintptr_t)glist->group, GFP_ATOMIC);
2181                         if (ret < 0)
2182                                 goto out;
2183                 }
2184         }
2185
2186 out:
2187         spin_unlock(&fs_info->qgroup_lock);
2188 }
2189
2190 static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes)
2191 {
2192         return btrfs_qgroup_free_refroot(root->fs_info, root->objectid,
2193                                          num_bytes);
2194 }
2195 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2196 {
2197         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2198                 return;
2199         btrfs_err(trans->fs_info,
2200                 "qgroups not uptodate in trans handle %p:  list is%s empty, "
2201                 "seq is %#x.%x",
2202                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2203                 (u32)(trans->delayed_ref_elem.seq >> 32),
2204                 (u32)trans->delayed_ref_elem.seq);
2205         BUG();
2206 }
2207
2208 /*
2209  * returns < 0 on error, 0 when more leafs are to be scanned.
2210  * returns 1 when done.
2211  */
2212 static int
2213 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2214                    struct btrfs_trans_handle *trans)
2215 {
2216         struct btrfs_key found;
2217         struct extent_buffer *scratch_leaf = NULL;
2218         struct ulist *roots = NULL;
2219         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2220         u64 num_bytes;
2221         int slot;
2222         int ret;
2223
2224         mutex_lock(&fs_info->qgroup_rescan_lock);
2225         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2226                                          &fs_info->qgroup_rescan_progress,
2227                                          path, 1, 0);
2228
2229         pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
2230                  fs_info->qgroup_rescan_progress.objectid,
2231                  fs_info->qgroup_rescan_progress.type,
2232                  fs_info->qgroup_rescan_progress.offset, ret);
2233
2234         if (ret) {
2235                 /*
2236                  * The rescan is about to end, we will not be scanning any
2237                  * further blocks. We cannot unset the RESCAN flag here, because
2238                  * we want to commit the transaction if everything went well.
2239                  * To make the live accounting work in this phase, we set our
2240                  * scan progress pointer such that every real extent objectid
2241                  * will be smaller.
2242                  */
2243                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2244                 btrfs_release_path(path);
2245                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2246                 return ret;
2247         }
2248
2249         btrfs_item_key_to_cpu(path->nodes[0], &found,
2250                               btrfs_header_nritems(path->nodes[0]) - 1);
2251         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2252
2253         btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2254         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2255         if (!scratch_leaf) {
2256                 ret = -ENOMEM;
2257                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2258                 goto out;
2259         }
2260         extent_buffer_get(scratch_leaf);
2261         btrfs_tree_read_lock(scratch_leaf);
2262         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2263         slot = path->slots[0];
2264         btrfs_release_path(path);
2265         mutex_unlock(&fs_info->qgroup_rescan_lock);
2266
2267         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2268                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2269                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2270                     found.type != BTRFS_METADATA_ITEM_KEY)
2271                         continue;
2272                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2273                         num_bytes = fs_info->extent_root->nodesize;
2274                 else
2275                         num_bytes = found.offset;
2276
2277                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2278                                            &roots);
2279                 if (ret < 0)
2280                         goto out;
2281                 /* For rescan, just pass old_roots as NULL */
2282                 ret = btrfs_qgroup_account_extent(trans, fs_info,
2283                                 found.objectid, num_bytes, NULL, roots);
2284                 if (ret < 0)
2285                         goto out;
2286         }
2287 out:
2288         if (scratch_leaf) {
2289                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2290                 free_extent_buffer(scratch_leaf);
2291         }
2292         btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2293
2294         return ret;
2295 }
2296
2297 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2298 {
2299         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2300                                                      qgroup_rescan_work);
2301         struct btrfs_path *path;
2302         struct btrfs_trans_handle *trans = NULL;
2303         int err = -ENOMEM;
2304         int ret = 0;
2305
2306         path = btrfs_alloc_path();
2307         if (!path)
2308                 goto out;
2309
2310         err = 0;
2311         while (!err && !btrfs_fs_closing(fs_info)) {
2312                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2313                 if (IS_ERR(trans)) {
2314                         err = PTR_ERR(trans);
2315                         break;
2316                 }
2317                 if (!fs_info->quota_enabled) {
2318                         err = -EINTR;
2319                 } else {
2320                         err = qgroup_rescan_leaf(fs_info, path, trans);
2321                 }
2322                 if (err > 0)
2323                         btrfs_commit_transaction(trans, fs_info->fs_root);
2324                 else
2325                         btrfs_end_transaction(trans, fs_info->fs_root);
2326         }
2327
2328 out:
2329         btrfs_free_path(path);
2330
2331         mutex_lock(&fs_info->qgroup_rescan_lock);
2332         if (!btrfs_fs_closing(fs_info))
2333                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2334
2335         if (err > 0 &&
2336             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2337                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2338         } else if (err < 0) {
2339                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2340         }
2341         mutex_unlock(&fs_info->qgroup_rescan_lock);
2342
2343         /*
2344          * only update status, since the previous part has already updated the
2345          * qgroup info.
2346          */
2347         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2348         if (IS_ERR(trans)) {
2349                 err = PTR_ERR(trans);
2350                 btrfs_err(fs_info,
2351                           "fail to start transaction for status update: %d\n",
2352                           err);
2353                 goto done;
2354         }
2355         ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2356         if (ret < 0) {
2357                 err = ret;
2358                 btrfs_err(fs_info, "fail to update qgroup status: %d\n", err);
2359         }
2360         btrfs_end_transaction(trans, fs_info->quota_root);
2361
2362         if (btrfs_fs_closing(fs_info)) {
2363                 btrfs_info(fs_info, "qgroup scan paused");
2364         } else if (err >= 0) {
2365                 btrfs_info(fs_info, "qgroup scan completed%s",
2366                         err > 0 ? " (inconsistency flag cleared)" : "");
2367         } else {
2368                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2369         }
2370
2371 done:
2372         complete_all(&fs_info->qgroup_rescan_completion);
2373 }
2374
2375 /*
2376  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2377  * memory required for the rescan context.
2378  */
2379 static int
2380 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2381                    int init_flags)
2382 {
2383         int ret = 0;
2384
2385         if (!init_flags &&
2386             (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2387              !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2388                 ret = -EINVAL;
2389                 goto err;
2390         }
2391
2392         mutex_lock(&fs_info->qgroup_rescan_lock);
2393         spin_lock(&fs_info->qgroup_lock);
2394
2395         if (init_flags) {
2396                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2397                         ret = -EINPROGRESS;
2398                 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2399                         ret = -EINVAL;
2400
2401                 if (ret) {
2402                         spin_unlock(&fs_info->qgroup_lock);
2403                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2404                         goto err;
2405                 }
2406                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2407         }
2408
2409         memset(&fs_info->qgroup_rescan_progress, 0,
2410                 sizeof(fs_info->qgroup_rescan_progress));
2411         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2412         init_completion(&fs_info->qgroup_rescan_completion);
2413
2414         spin_unlock(&fs_info->qgroup_lock);
2415         mutex_unlock(&fs_info->qgroup_rescan_lock);
2416
2417         memset(&fs_info->qgroup_rescan_work, 0,
2418                sizeof(fs_info->qgroup_rescan_work));
2419         btrfs_init_work(&fs_info->qgroup_rescan_work,
2420                         btrfs_qgroup_rescan_helper,
2421                         btrfs_qgroup_rescan_worker, NULL, NULL);
2422
2423         if (ret) {
2424 err:
2425                 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2426                 return ret;
2427         }
2428
2429         return 0;
2430 }
2431
2432 static void
2433 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2434 {
2435         struct rb_node *n;
2436         struct btrfs_qgroup *qgroup;
2437
2438         spin_lock(&fs_info->qgroup_lock);
2439         /* clear all current qgroup tracking information */
2440         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2441                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2442                 qgroup->rfer = 0;
2443                 qgroup->rfer_cmpr = 0;
2444                 qgroup->excl = 0;
2445                 qgroup->excl_cmpr = 0;
2446         }
2447         spin_unlock(&fs_info->qgroup_lock);
2448 }
2449
2450 int
2451 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2452 {
2453         int ret = 0;
2454         struct btrfs_trans_handle *trans;
2455
2456         ret = qgroup_rescan_init(fs_info, 0, 1);
2457         if (ret)
2458                 return ret;
2459
2460         /*
2461          * We have set the rescan_progress to 0, which means no more
2462          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2463          * However, btrfs_qgroup_account_ref may be right after its call
2464          * to btrfs_find_all_roots, in which case it would still do the
2465          * accounting.
2466          * To solve this, we're committing the transaction, which will
2467          * ensure we run all delayed refs and only after that, we are
2468          * going to clear all tracking information for a clean start.
2469          */
2470
2471         trans = btrfs_join_transaction(fs_info->fs_root);
2472         if (IS_ERR(trans)) {
2473                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2474                 return PTR_ERR(trans);
2475         }
2476         ret = btrfs_commit_transaction(trans, fs_info->fs_root);
2477         if (ret) {
2478                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2479                 return ret;
2480         }
2481
2482         qgroup_rescan_zero_tracking(fs_info);
2483
2484         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2485                          &fs_info->qgroup_rescan_work);
2486
2487         return 0;
2488 }
2489
2490 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info)
2491 {
2492         int running;
2493         int ret = 0;
2494
2495         mutex_lock(&fs_info->qgroup_rescan_lock);
2496         spin_lock(&fs_info->qgroup_lock);
2497         running = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2498         spin_unlock(&fs_info->qgroup_lock);
2499         mutex_unlock(&fs_info->qgroup_rescan_lock);
2500
2501         if (running)
2502                 ret = wait_for_completion_interruptible(
2503                                         &fs_info->qgroup_rescan_completion);
2504
2505         return ret;
2506 }
2507
2508 /*
2509  * this is only called from open_ctree where we're still single threaded, thus
2510  * locking is omitted here.
2511  */
2512 void
2513 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2514 {
2515         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2516                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2517                                  &fs_info->qgroup_rescan_work);
2518 }
2519
2520 /*
2521  * Reserve qgroup space for range [start, start + len).
2522  *
2523  * This function will either reserve space from related qgroups or doing
2524  * nothing if the range is already reserved.
2525  *
2526  * Return 0 for successful reserve
2527  * Return <0 for error (including -EQUOT)
2528  *
2529  * NOTE: this function may sleep for memory allocation.
2530  */
2531 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2532 {
2533         struct btrfs_root *root = BTRFS_I(inode)->root;
2534         struct extent_changeset changeset;
2535         struct ulist_node *unode;
2536         struct ulist_iterator uiter;
2537         int ret;
2538
2539         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2540             len == 0)
2541                 return 0;
2542
2543         changeset.bytes_changed = 0;
2544         changeset.range_changed = ulist_alloc(GFP_NOFS);
2545         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2546                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2547         trace_btrfs_qgroup_reserve_data(inode, start, len,
2548                                         changeset.bytes_changed,
2549                                         QGROUP_RESERVE);
2550         if (ret < 0)
2551                 goto cleanup;
2552         ret = qgroup_reserve(root, changeset.bytes_changed);
2553         if (ret < 0)
2554                 goto cleanup;
2555
2556         ulist_free(changeset.range_changed);
2557         return ret;
2558
2559 cleanup:
2560         /* cleanup already reserved ranges */
2561         ULIST_ITER_INIT(&uiter);
2562         while ((unode = ulist_next(changeset.range_changed, &uiter)))
2563                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2564                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2565                                  GFP_NOFS);
2566         ulist_free(changeset.range_changed);
2567         return ret;
2568 }
2569
2570 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2571                                        int free)
2572 {
2573         struct extent_changeset changeset;
2574         int trace_op = QGROUP_RELEASE;
2575         int ret;
2576
2577         changeset.bytes_changed = 0;
2578         changeset.range_changed = ulist_alloc(GFP_NOFS);
2579         if (!changeset.range_changed)
2580                 return -ENOMEM;
2581
2582         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2583                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2584         if (ret < 0)
2585                 goto out;
2586
2587         if (free) {
2588                 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2589                 trace_op = QGROUP_FREE;
2590         }
2591         trace_btrfs_qgroup_release_data(inode, start, len,
2592                                         changeset.bytes_changed, trace_op);
2593 out:
2594         ulist_free(changeset.range_changed);
2595         return ret;
2596 }
2597
2598 /*
2599  * Free a reserved space range from io_tree and related qgroups
2600  *
2601  * Should be called when a range of pages get invalidated before reaching disk.
2602  * Or for error cleanup case.
2603  *
2604  * For data written to disk, use btrfs_qgroup_release_data().
2605  *
2606  * NOTE: This function may sleep for memory allocation.
2607  */
2608 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2609 {
2610         return __btrfs_qgroup_release_data(inode, start, len, 1);
2611 }
2612
2613 /*
2614  * Release a reserved space range from io_tree only.
2615  *
2616  * Should be called when a range of pages get written to disk and corresponding
2617  * FILE_EXTENT is inserted into corresponding root.
2618  *
2619  * Since new qgroup accounting framework will only update qgroup numbers at
2620  * commit_transaction() time, its reserved space shouldn't be freed from
2621  * related qgroups.
2622  *
2623  * But we should release the range from io_tree, to allow further write to be
2624  * COWed.
2625  *
2626  * NOTE: This function may sleep for memory allocation.
2627  */
2628 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2629 {
2630         return __btrfs_qgroup_release_data(inode, start, len, 0);
2631 }
2632
2633 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes)
2634 {
2635         int ret;
2636
2637         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2638             num_bytes == 0)
2639                 return 0;
2640
2641         BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2642         ret = qgroup_reserve(root, num_bytes);
2643         if (ret < 0)
2644                 return ret;
2645         atomic_add(num_bytes, &root->qgroup_meta_rsv);
2646         return ret;
2647 }
2648
2649 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2650 {
2651         int reserved;
2652
2653         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2654                 return;
2655
2656         reserved = atomic_xchg(&root->qgroup_meta_rsv, 0);
2657         if (reserved == 0)
2658                 return;
2659         qgroup_free(root, reserved);
2660 }
2661
2662 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2663 {
2664         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2665                 return;
2666
2667         BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2668         WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes);
2669         atomic_sub(num_bytes, &root->qgroup_meta_rsv);
2670         qgroup_free(root, num_bytes);
2671 }
2672
2673 /*
2674  * Check qgroup reserved space leaking, normally at destroy inode
2675  * time
2676  */
2677 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2678 {
2679         struct extent_changeset changeset;
2680         struct ulist_node *unode;
2681         struct ulist_iterator iter;
2682         int ret;
2683
2684         changeset.bytes_changed = 0;
2685         changeset.range_changed = ulist_alloc(GFP_NOFS);
2686         if (WARN_ON(!changeset.range_changed))
2687                 return;
2688
2689         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
2690                         EXTENT_QGROUP_RESERVED, &changeset);
2691
2692         WARN_ON(ret < 0);
2693         if (WARN_ON(changeset.bytes_changed)) {
2694                 ULIST_ITER_INIT(&iter);
2695                 while ((unode = ulist_next(changeset.range_changed, &iter))) {
2696                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
2697                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2698                                 inode->i_ino, unode->val, unode->aux);
2699                 }
2700                 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2701         }
2702         ulist_free(changeset.range_changed);
2703 }