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
  36 /* TODO XXX FIXME
  37  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  38  *  - reorganize keys
  39  *  - compressed
  40  *  - sync
  41  *  - copy also limits on subvol creation
  42  *  - limit
  43  *  - caches fuer ulists
  44  *  - performance benchmarks
  45  *  - check all ioctl parameters
  46  */
  47
  48 /*
  49  * one struct for each qgroup, organized in fs_info->qgroup_tree.
  50  */
  51 struct btrfs_qgroup {
  52         u64 qgroupid;
  53
  54         /*
  55          * state
  56          */
  57         u64 rfer;       /* referenced */
  58         u64 rfer_cmpr;  /* referenced compressed */
  59         u64 excl;       /* exclusive */
  60         u64 excl_cmpr;  /* exclusive compressed */
  61
  62         /*
  63          * limits
  64          */
  65         u64 lim_flags;  /* which limits are set */
  66         u64 max_rfer;
  67         u64 max_excl;
  68         u64 rsv_rfer;
  69         u64 rsv_excl;
  70
  71         /*
  72          * reservation tracking
  73          */
  74         u64 reserved;
  75
  76         /*
  77          * lists
  78          */
  79         struct list_head groups;  /* groups this group is member of */
  80         struct list_head members; /* groups that are members of this group */
  81         struct list_head dirty;   /* dirty groups */
  82         struct rb_node node;      /* tree of qgroups */
  83
  84         /*
  85          * temp variables for accounting operations
  86          */
  87         u64 tag;
  88         u64 refcnt;
  89 };
  90
  91 /*
  92  * glue structure to represent the relations between qgroups.
  93  */
  94 struct btrfs_qgroup_list {
  95         struct list_head next_group;
  96         struct list_head next_member;
  97         struct btrfs_qgroup *group;
  98         struct btrfs_qgroup *member;
  99 };
 100
 101 static int
 102 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 103                    int init_flags);
 104 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 105
 106 /* must be called with qgroup_ioctl_lock held */
 107 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 108                                            u64 qgroupid)
 109 {
 110         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 111         struct btrfs_qgroup *qgroup;
 112
 113         while (n) {
 114                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 115                 if (qgroup->qgroupid < qgroupid)
 116                         n = n->rb_left;
 117                 else if (qgroup->qgroupid > qgroupid)
 118                         n = n->rb_right;
 119                 else
 120                         return qgroup;
 121         }
 122         return NULL;
 123 }
 124
 125 /* must be called with qgroup_lock held */
 126 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 127                                           u64 qgroupid)
 128 {
 129         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 130         struct rb_node *parent = NULL;
 131         struct btrfs_qgroup *qgroup;
 132
 133         while (*p) {
 134                 parent = *p;
 135                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 136
 137                 if (qgroup->qgroupid < qgroupid)
 138                         p = &(*p)->rb_left;
 139                 else if (qgroup->qgroupid > qgroupid)
 140                         p = &(*p)->rb_right;
 141                 else
 142                         return qgroup;
 143         }
 144
 145         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 146         if (!qgroup)
 147                 return ERR_PTR(-ENOMEM);
 148
 149         qgroup->qgroupid = qgroupid;
 150         INIT_LIST_HEAD(&qgroup->groups);
 151         INIT_LIST_HEAD(&qgroup->members);
 152         INIT_LIST_HEAD(&qgroup->dirty);
 153
 154         rb_link_node(&qgroup->node, parent, p);
 155         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 156
 157         return qgroup;
 158 }
 159
 160 /* must be called with qgroup_lock held */
 161 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 162 {
 163         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 164         struct btrfs_qgroup_list *list;
 165
 166         if (!qgroup)
 167                 return -ENOENT;
 168
 169         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 170         list_del(&qgroup->dirty);
 171
 172         while (!list_empty(&qgroup->groups)) {
 173                 list = list_first_entry(&qgroup->groups,
 174                                         struct btrfs_qgroup_list, next_group);
 175                 list_del(&list->next_group);
 176                 list_del(&list->next_member);
 177                 kfree(list);
 178         }
 179
 180         while (!list_empty(&qgroup->members)) {
 181                 list = list_first_entry(&qgroup->members,
 182                                         struct btrfs_qgroup_list, next_member);
 183                 list_del(&list->next_group);
 184                 list_del(&list->next_member);
 185                 kfree(list);
 186         }
 187         kfree(qgroup);
 188
 189         return 0;
 190 }
 191
 192 /* must be called with qgroup_lock held */
 193 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 194                            u64 memberid, u64 parentid)
 195 {
 196         struct btrfs_qgroup *member;
 197         struct btrfs_qgroup *parent;
 198         struct btrfs_qgroup_list *list;
 199
 200         member = find_qgroup_rb(fs_info, memberid);
 201         parent = find_qgroup_rb(fs_info, parentid);
 202         if (!member || !parent)
 203                 return -ENOENT;
 204
 205         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 206         if (!list)
 207                 return -ENOMEM;
 208
 209         list->group = parent;
 210         list->member = member;
 211         list_add_tail(&list->next_group, &member->groups);
 212         list_add_tail(&list->next_member, &parent->members);
 213
 214         return 0;
 215 }
 216
 217 /* must be called with qgroup_lock held */
 218 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 219                            u64 memberid, u64 parentid)
 220 {
 221         struct btrfs_qgroup *member;
 222         struct btrfs_qgroup *parent;
 223         struct btrfs_qgroup_list *list;
 224
 225         member = find_qgroup_rb(fs_info, memberid);
 226         parent = find_qgroup_rb(fs_info, parentid);
 227         if (!member || !parent)
 228                 return -ENOENT;
 229
 230         list_for_each_entry(list, &member->groups, next_group) {
 231                 if (list->group == parent) {
 232                         list_del(&list->next_group);
 233                         list_del(&list->next_member);
 234                         kfree(list);
 235                         return 0;
 236                 }
 237         }
 238         return -ENOENT;
 239 }
 240
 241 /*
 242  * The full config is read in one go, only called from open_ctree()
 243  * It doesn't use any locking, as at this point we're still single-threaded
 244  */
 245 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 246 {
 247         struct btrfs_key key;
 248         struct btrfs_key found_key;
 249         struct btrfs_root *quota_root = fs_info->quota_root;
 250         struct btrfs_path *path = NULL;
 251         struct extent_buffer *l;
 252         int slot;
 253         int ret = 0;
 254         u64 flags = 0;
 255         u64 rescan_progress = 0;
 256
 257         if (!fs_info->quota_enabled)
 258                 return 0;
 259
 260         fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
 261         if (!fs_info->qgroup_ulist) {
 262                 ret = -ENOMEM;
 263                 goto out;
 264         }
 265
 266         path = btrfs_alloc_path();
 267         if (!path) {
 268                 ret = -ENOMEM;
 269                 goto out;
 270         }
 271
 272         /* default this to quota off, in case no status key is found */
 273         fs_info->qgroup_flags = 0;
 274
 275         /*
 276          * pass 1: read status, all qgroup infos and limits
 277          */
 278         key.objectid = 0;
 279         key.type = 0;
 280         key.offset = 0;
 281         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 282         if (ret)
 283                 goto out;
 284
 285         while (1) {
 286                 struct btrfs_qgroup *qgroup;
 287
 288                 slot = path->slots[0];
 289                 l = path->nodes[0];
 290                 btrfs_item_key_to_cpu(l, &found_key, slot);
 291
 292                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 293                         struct btrfs_qgroup_status_item *ptr;
 294
 295                         ptr = btrfs_item_ptr(l, slot,
 296                                              struct btrfs_qgroup_status_item);
 297
 298                         if (btrfs_qgroup_status_version(l, ptr) !=
 299                             BTRFS_QGROUP_STATUS_VERSION) {
 300                                 printk(KERN_ERR
 301                                  "btrfs: old qgroup version, quota disabled\n");
 302                                 goto out;
 303                         }
 304                         if (btrfs_qgroup_status_generation(l, ptr) !=
 305                             fs_info->generation) {
 306                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 307                                 printk(KERN_ERR
 308                                         "btrfs: qgroup generation mismatch, "
 309                                         "marked as inconsistent\n");
 310                         }
 311                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 312                                                                           ptr);
 313                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 314                         goto next1;
 315                 }
 316
 317                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 318                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 319                         goto next1;
 320
 321                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 322                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 323                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 324                         printk(KERN_ERR "btrfs: inconsitent qgroup config\n");
 325                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 326                 }
 327                 if (!qgroup) {
 328                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 329                         if (IS_ERR(qgroup)) {
 330                                 ret = PTR_ERR(qgroup);
 331                                 goto out;
 332                         }
 333                 }
 334                 switch (found_key.type) {
 335                 case BTRFS_QGROUP_INFO_KEY: {
 336                         struct btrfs_qgroup_info_item *ptr;
 337
 338                         ptr = btrfs_item_ptr(l, slot,
 339                                              struct btrfs_qgroup_info_item);
 340                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 341                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 342                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 343                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 344                         /* generation currently unused */
 345                         break;
 346                 }
 347                 case BTRFS_QGROUP_LIMIT_KEY: {
 348                         struct btrfs_qgroup_limit_item *ptr;
 349
 350                         ptr = btrfs_item_ptr(l, slot,
 351                                              struct btrfs_qgroup_limit_item);
 352                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 353                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 354                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 355                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 356                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 357                         break;
 358                 }
 359                 }
 360 next1:
 361                 ret = btrfs_next_item(quota_root, path);
 362                 if (ret < 0)
 363                         goto out;
 364                 if (ret)
 365                         break;
 366         }
 367         btrfs_release_path(path);
 368
 369         /*
 370          * pass 2: read all qgroup relations
 371          */
 372         key.objectid = 0;
 373         key.type = BTRFS_QGROUP_RELATION_KEY;
 374         key.offset = 0;
 375         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 376         if (ret)
 377                 goto out;
 378         while (1) {
 379                 slot = path->slots[0];
 380                 l = path->nodes[0];
 381                 btrfs_item_key_to_cpu(l, &found_key, slot);
 382
 383                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 384                         goto next2;
 385
 386                 if (found_key.objectid > found_key.offset) {
 387                         /* parent <- member, not needed to build config */
 388                         /* FIXME should we omit the key completely? */
 389                         goto next2;
 390                 }
 391
 392                 ret = add_relation_rb(fs_info, found_key.objectid,
 393                                       found_key.offset);
 394                 if (ret == -ENOENT) {
 395                         printk(KERN_WARNING
 396                                 "btrfs: orphan qgroup relation 0x%llx->0x%llx\n",
 397                                 (unsigned long long)found_key.objectid,
 398                                 (unsigned long long)found_key.offset);
 399                         ret = 0;        /* ignore the error */
 400                 }
 401                 if (ret)
 402                         goto out;
 403 next2:
 404                 ret = btrfs_next_item(quota_root, path);
 405                 if (ret < 0)
 406                         goto out;
 407                 if (ret)
 408                         break;
 409         }
 410 out:
 411         fs_info->qgroup_flags |= flags;
 412         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
 413                 fs_info->quota_enabled = 0;
 414                 fs_info->pending_quota_state = 0;
 415         } else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 416                    ret >= 0) {
 417                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 418         }
 419         btrfs_free_path(path);
 420
 421         if (ret < 0) {
 422                 ulist_free(fs_info->qgroup_ulist);
 423                 fs_info->qgroup_ulist = NULL;
 424                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 425         }
 426
 427         return ret < 0 ? ret : 0;
 428 }
 429
 430 /*
 431  * This is only called from close_ctree() or open_ctree(), both in single-
 432  * treaded paths. Clean up the in-memory structures. No locking needed.
 433  */
 434 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 435 {
 436         struct rb_node *n;
 437         struct btrfs_qgroup *qgroup;
 438         struct btrfs_qgroup_list *list;
 439
 440         while ((n = rb_first(&fs_info->qgroup_tree))) {
 441                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 442                 rb_erase(n, &fs_info->qgroup_tree);
 443
 444                 while (!list_empty(&qgroup->groups)) {
 445                         list = list_first_entry(&qgroup->groups,
 446                                                 struct btrfs_qgroup_list,
 447                                                 next_group);
 448                         list_del(&list->next_group);
 449                         list_del(&list->next_member);
 450                         kfree(list);
 451                 }
 452
 453                 while (!list_empty(&qgroup->members)) {
 454                         list = list_first_entry(&qgroup->members,
 455                                                 struct btrfs_qgroup_list,
 456                                                 next_member);
 457                         list_del(&list->next_group);
 458                         list_del(&list->next_member);
 459                         kfree(list);
 460                 }
 461                 kfree(qgroup);
 462         }
 463         /*
 464          * we call btrfs_free_qgroup_config() when umounting
 465          * filesystem and disabling quota, so we set qgroup_ulit
 466          * to be null here to avoid double free.
 467          */
 468         ulist_free(fs_info->qgroup_ulist);
 469         fs_info->qgroup_ulist = NULL;
 470 }
 471
 472 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
 473                                     struct btrfs_root *quota_root,
 474                                     u64 src, u64 dst)
 475 {
 476         int ret;
 477         struct btrfs_path *path;
 478         struct btrfs_key key;
 479
 480         path = btrfs_alloc_path();
 481         if (!path)
 482                 return -ENOMEM;
 483
 484         key.objectid = src;
 485         key.type = BTRFS_QGROUP_RELATION_KEY;
 486         key.offset = dst;
 487
 488         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 489
 490         btrfs_mark_buffer_dirty(path->nodes[0]);
 491
 492         btrfs_free_path(path);
 493         return ret;
 494 }
 495
 496 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
 497                                     struct btrfs_root *quota_root,
 498                                     u64 src, u64 dst)
 499 {
 500         int ret;
 501         struct btrfs_path *path;
 502         struct btrfs_key key;
 503
 504         path = btrfs_alloc_path();
 505         if (!path)
 506                 return -ENOMEM;
 507
 508         key.objectid = src;
 509         key.type = BTRFS_QGROUP_RELATION_KEY;
 510         key.offset = dst;
 511
 512         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 513         if (ret < 0)
 514                 goto out;
 515
 516         if (ret > 0) {
 517                 ret = -ENOENT;
 518                 goto out;
 519         }
 520
 521         ret = btrfs_del_item(trans, quota_root, path);
 522 out:
 523         btrfs_free_path(path);
 524         return ret;
 525 }
 526
 527 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 528                            struct btrfs_root *quota_root, u64 qgroupid)
 529 {
 530         int ret;
 531         struct btrfs_path *path;
 532         struct btrfs_qgroup_info_item *qgroup_info;
 533         struct btrfs_qgroup_limit_item *qgroup_limit;
 534         struct extent_buffer *leaf;
 535         struct btrfs_key key;
 536
 537         path = btrfs_alloc_path();
 538         if (!path)
 539                 return -ENOMEM;
 540
 541         key.objectid = 0;
 542         key.type = BTRFS_QGROUP_INFO_KEY;
 543         key.offset = qgroupid;
 544
 545         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 546                                       sizeof(*qgroup_info));
 547         if (ret)
 548                 goto out;
 549
 550         leaf = path->nodes[0];
 551         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 552                                  struct btrfs_qgroup_info_item);
 553         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 554         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 555         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 556         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 557         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 558
 559         btrfs_mark_buffer_dirty(leaf);
 560
 561         btrfs_release_path(path);
 562
 563         key.type = BTRFS_QGROUP_LIMIT_KEY;
 564         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 565                                       sizeof(*qgroup_limit));
 566         if (ret)
 567                 goto out;
 568
 569         leaf = path->nodes[0];
 570         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 571                                   struct btrfs_qgroup_limit_item);
 572         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 573         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 574         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 575         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 576         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 577
 578         btrfs_mark_buffer_dirty(leaf);
 579
 580         ret = 0;
 581 out:
 582         btrfs_free_path(path);
 583         return ret;
 584 }
 585
 586 static int del_qgroup_item(struct btrfs_trans_handle *trans,
 587                            struct btrfs_root *quota_root, u64 qgroupid)
 588 {
 589         int ret;
 590         struct btrfs_path *path;
 591         struct btrfs_key key;
 592
 593         path = btrfs_alloc_path();
 594         if (!path)
 595                 return -ENOMEM;
 596
 597         key.objectid = 0;
 598         key.type = BTRFS_QGROUP_INFO_KEY;
 599         key.offset = qgroupid;
 600         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 601         if (ret < 0)
 602                 goto out;
 603
 604         if (ret > 0) {
 605                 ret = -ENOENT;
 606                 goto out;
 607         }
 608
 609         ret = btrfs_del_item(trans, quota_root, path);
 610         if (ret)
 611                 goto out;
 612
 613         btrfs_release_path(path);
 614
 615         key.type = BTRFS_QGROUP_LIMIT_KEY;
 616         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 617         if (ret < 0)
 618                 goto out;
 619
 620         if (ret > 0) {
 621                 ret = -ENOENT;
 622                 goto out;
 623         }
 624
 625         ret = btrfs_del_item(trans, quota_root, path);
 626
 627 out:
 628         btrfs_free_path(path);
 629         return ret;
 630 }
 631
 632 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 633                                     struct btrfs_root *root, u64 qgroupid,
 634                                     u64 flags, u64 max_rfer, u64 max_excl,
 635                                     u64 rsv_rfer, u64 rsv_excl)
 636 {
 637         struct btrfs_path *path;
 638         struct btrfs_key key;
 639         struct extent_buffer *l;
 640         struct btrfs_qgroup_limit_item *qgroup_limit;
 641         int ret;
 642         int slot;
 643
 644         key.objectid = 0;
 645         key.type = BTRFS_QGROUP_LIMIT_KEY;
 646         key.offset = qgroupid;
 647
 648         path = btrfs_alloc_path();
 649         if (!path)
 650                 return -ENOMEM;
 651
 652         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 653         if (ret > 0)
 654                 ret = -ENOENT;
 655
 656         if (ret)
 657                 goto out;
 658
 659         l = path->nodes[0];
 660         slot = path->slots[0];
 661         qgroup_limit = btrfs_item_ptr(l, path->slots[0],
 662                                       struct btrfs_qgroup_limit_item);
 663         btrfs_set_qgroup_limit_flags(l, qgroup_limit, flags);
 664         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, max_rfer);
 665         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, max_excl);
 666         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, rsv_rfer);
 667         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, rsv_excl);
 668
 669         btrfs_mark_buffer_dirty(l);
 670
 671 out:
 672         btrfs_free_path(path);
 673         return ret;
 674 }
 675
 676 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 677                                    struct btrfs_root *root,
 678                                    struct btrfs_qgroup *qgroup)
 679 {
 680         struct btrfs_path *path;
 681         struct btrfs_key key;
 682         struct extent_buffer *l;
 683         struct btrfs_qgroup_info_item *qgroup_info;
 684         int ret;
 685         int slot;
 686
 687         key.objectid = 0;
 688         key.type = BTRFS_QGROUP_INFO_KEY;
 689         key.offset = qgroup->qgroupid;
 690
 691         path = btrfs_alloc_path();
 692         if (!path)
 693                 return -ENOMEM;
 694
 695         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 696         if (ret > 0)
 697                 ret = -ENOENT;
 698
 699         if (ret)
 700                 goto out;
 701
 702         l = path->nodes[0];
 703         slot = path->slots[0];
 704         qgroup_info = btrfs_item_ptr(l, path->slots[0],
 705                                  struct btrfs_qgroup_info_item);
 706         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 707         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 708         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 709         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 710         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 711
 712         btrfs_mark_buffer_dirty(l);
 713
 714 out:
 715         btrfs_free_path(path);
 716         return ret;
 717 }
 718
 719 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 720                                      struct btrfs_fs_info *fs_info,
 721                                     struct btrfs_root *root)
 722 {
 723         struct btrfs_path *path;
 724         struct btrfs_key key;
 725         struct extent_buffer *l;
 726         struct btrfs_qgroup_status_item *ptr;
 727         int ret;
 728         int slot;
 729
 730         key.objectid = 0;
 731         key.type = BTRFS_QGROUP_STATUS_KEY;
 732         key.offset = 0;
 733
 734         path = btrfs_alloc_path();
 735         if (!path)
 736                 return -ENOMEM;
 737
 738         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 739         if (ret > 0)
 740                 ret = -ENOENT;
 741
 742         if (ret)
 743                 goto out;
 744
 745         l = path->nodes[0];
 746         slot = path->slots[0];
 747         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 748         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 749         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 750         btrfs_set_qgroup_status_rescan(l, ptr,
 751                                 fs_info->qgroup_rescan_progress.objectid);
 752
 753         btrfs_mark_buffer_dirty(l);
 754
 755 out:
 756         btrfs_free_path(path);
 757         return ret;
 758 }
 759
 760 /*
 761  * called with qgroup_lock held
 762  */
 763 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 764                                   struct btrfs_root *root)
 765 {
 766         struct btrfs_path *path;
 767         struct btrfs_key key;
 768         struct extent_buffer *leaf = NULL;
 769         int ret;
 770         int nr = 0;
 771
 772         path = btrfs_alloc_path();
 773         if (!path)
 774                 return -ENOMEM;
 775
 776         path->leave_spinning = 1;
 777
 778         key.objectid = 0;
 779         key.offset = 0;
 780         key.type = 0;
 781
 782         while (1) {
 783                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 784                 if (ret < 0)
 785                         goto out;
 786                 leaf = path->nodes[0];
 787                 nr = btrfs_header_nritems(leaf);
 788                 if (!nr)
 789                         break;
 790                 /*
 791                  * delete the leaf one by one
 792                  * since the whole tree is going
 793                  * to be deleted.
 794                  */
 795                 path->slots[0] = 0;
 796                 ret = btrfs_del_items(trans, root, path, 0, nr);
 797                 if (ret)
 798                         goto out;
 799
 800                 btrfs_release_path(path);
 801         }
 802         ret = 0;
 803 out:
 804         root->fs_info->pending_quota_state = 0;
 805         btrfs_free_path(path);
 806         return ret;
 807 }
 808
 809 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 810                        struct btrfs_fs_info *fs_info)
 811 {
 812         struct btrfs_root *quota_root;
 813         struct btrfs_root *tree_root = fs_info->tree_root;
 814         struct btrfs_path *path = NULL;
 815         struct btrfs_qgroup_status_item *ptr;
 816         struct extent_buffer *leaf;
 817         struct btrfs_key key;
 818         struct btrfs_key found_key;
 819         struct btrfs_qgroup *qgroup = NULL;
 820         int ret = 0;
 821         int slot;
 822
 823         mutex_lock(&fs_info->qgroup_ioctl_lock);
 824         if (fs_info->quota_root) {
 825                 fs_info->pending_quota_state = 1;
 826                 goto out;
 827         }
 828
 829         fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
 830         if (!fs_info->qgroup_ulist) {
 831                 ret = -ENOMEM;
 832                 goto out;
 833         }
 834
 835         /*
 836          * initially create the quota tree
 837          */
 838         quota_root = btrfs_create_tree(trans, fs_info,
 839                                        BTRFS_QUOTA_TREE_OBJECTID);
 840         if (IS_ERR(quota_root)) {
 841                 ret =  PTR_ERR(quota_root);
 842                 goto out;
 843         }
 844
 845         path = btrfs_alloc_path();
 846         if (!path) {
 847                 ret = -ENOMEM;
 848                 goto out_free_root;
 849         }
 850
 851         key.objectid = 0;
 852         key.type = BTRFS_QGROUP_STATUS_KEY;
 853         key.offset = 0;
 854
 855         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 856                                       sizeof(*ptr));
 857         if (ret)
 858                 goto out_free_path;
 859
 860         leaf = path->nodes[0];
 861         ptr = btrfs_item_ptr(leaf, path->slots[0],
 862                                  struct btrfs_qgroup_status_item);
 863         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 864         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 865         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 866                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 867         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 868         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 869
 870         btrfs_mark_buffer_dirty(leaf);
 871
 872         key.objectid = 0;
 873         key.type = BTRFS_ROOT_REF_KEY;
 874         key.offset = 0;
 875
 876         btrfs_release_path(path);
 877         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 878         if (ret > 0)
 879                 goto out_add_root;
 880         if (ret < 0)
 881                 goto out_free_path;
 882
 883
 884         while (1) {
 885                 slot = path->slots[0];
 886                 leaf = path->nodes[0];
 887                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 888
 889                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 890                         ret = add_qgroup_item(trans, quota_root,
 891                                               found_key.offset);
 892                         if (ret)
 893                                 goto out_free_path;
 894
 895                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 896                         if (IS_ERR(qgroup)) {
 897                                 ret = PTR_ERR(qgroup);
 898                                 goto out_free_path;
 899                         }
 900                 }
 901                 ret = btrfs_next_item(tree_root, path);
 902                 if (ret < 0)
 903                         goto out_free_path;
 904                 if (ret)
 905                         break;
 906         }
 907
 908 out_add_root:
 909         btrfs_release_path(path);
 910         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
 911         if (ret)
 912                 goto out_free_path;
 913
 914         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
 915         if (IS_ERR(qgroup)) {
 916                 ret = PTR_ERR(qgroup);
 917                 goto out_free_path;
 918         }
 919         spin_lock(&fs_info->qgroup_lock);
 920         fs_info->quota_root = quota_root;
 921         fs_info->pending_quota_state = 1;
 922         spin_unlock(&fs_info->qgroup_lock);
 923 out_free_path:
 924         btrfs_free_path(path);
 925 out_free_root:
 926         if (ret) {
 927                 free_extent_buffer(quota_root->node);
 928                 free_extent_buffer(quota_root->commit_root);
 929                 kfree(quota_root);
 930         }
 931 out:
 932         if (ret) {
 933                 ulist_free(fs_info->qgroup_ulist);
 934                 fs_info->qgroup_ulist = NULL;
 935         }
 936         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 937         return ret;
 938 }
 939
 940 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
 941                         struct btrfs_fs_info *fs_info)
 942 {
 943         struct btrfs_root *tree_root = fs_info->tree_root;
 944         struct btrfs_root *quota_root;
 945         int ret = 0;
 946
 947         mutex_lock(&fs_info->qgroup_ioctl_lock);
 948         if (!fs_info->quota_root)
 949                 goto out;
 950         spin_lock(&fs_info->qgroup_lock);
 951         fs_info->quota_enabled = 0;
 952         fs_info->pending_quota_state = 0;
 953         quota_root = fs_info->quota_root;
 954         fs_info->quota_root = NULL;
 955         btrfs_free_qgroup_config(fs_info);
 956         spin_unlock(&fs_info->qgroup_lock);
 957
 958         if (!quota_root) {
 959                 ret = -EINVAL;
 960                 goto out;
 961         }
 962
 963         ret = btrfs_clean_quota_tree(trans, quota_root);
 964         if (ret)
 965                 goto out;
 966
 967         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
 968         if (ret)
 969                 goto out;
 970
 971         list_del(&quota_root->dirty_list);
 972
 973         btrfs_tree_lock(quota_root->node);
 974         clean_tree_block(trans, tree_root, quota_root->node);
 975         btrfs_tree_unlock(quota_root->node);
 976         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
 977
 978         free_extent_buffer(quota_root->node);
 979         free_extent_buffer(quota_root->commit_root);
 980         kfree(quota_root);
 981 out:
 982         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 983         return ret;
 984 }
 985
 986 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
 987                          struct btrfs_qgroup *qgroup)
 988 {
 989         if (list_empty(&qgroup->dirty))
 990                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
 991 }
 992
 993 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
 994                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
 995 {
 996         struct btrfs_root *quota_root;
 997         struct btrfs_qgroup *parent;
 998         struct btrfs_qgroup *member;
 999         struct btrfs_qgroup_list *list;
1000         int ret = 0;
1001
1002         mutex_lock(&fs_info->qgroup_ioctl_lock);
1003         quota_root = fs_info->quota_root;
1004         if (!quota_root) {
1005                 ret = -EINVAL;
1006                 goto out;
1007         }
1008         member = find_qgroup_rb(fs_info, src);
1009         parent = find_qgroup_rb(fs_info, dst);
1010         if (!member || !parent) {
1011                 ret = -EINVAL;
1012                 goto out;
1013         }
1014
1015         /* check if such qgroup relation exist firstly */
1016         list_for_each_entry(list, &member->groups, next_group) {
1017                 if (list->group == parent) {
1018                         ret = -EEXIST;
1019                         goto out;
1020                 }
1021         }
1022
1023         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1024         if (ret)
1025                 goto out;
1026
1027         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1028         if (ret) {
1029                 del_qgroup_relation_item(trans, quota_root, src, dst);
1030                 goto out;
1031         }
1032
1033         spin_lock(&fs_info->qgroup_lock);
1034         ret = add_relation_rb(quota_root->fs_info, src, dst);
1035         spin_unlock(&fs_info->qgroup_lock);
1036 out:
1037         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1038         return ret;
1039 }
1040
1041 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1042                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1043 {
1044         struct btrfs_root *quota_root;
1045         struct btrfs_qgroup *parent;
1046         struct btrfs_qgroup *member;
1047         struct btrfs_qgroup_list *list;
1048         int ret = 0;
1049         int err;
1050
1051         mutex_lock(&fs_info->qgroup_ioctl_lock);
1052         quota_root = fs_info->quota_root;
1053         if (!quota_root) {
1054                 ret = -EINVAL;
1055                 goto out;
1056         }
1057
1058         member = find_qgroup_rb(fs_info, src);
1059         parent = find_qgroup_rb(fs_info, dst);
1060         if (!member || !parent) {
1061                 ret = -EINVAL;
1062                 goto out;
1063         }
1064
1065         /* check if such qgroup relation exist firstly */
1066         list_for_each_entry(list, &member->groups, next_group) {
1067                 if (list->group == parent)
1068                         goto exist;
1069         }
1070         ret = -ENOENT;
1071         goto out;
1072 exist:
1073         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1074         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1075         if (err && !ret)
1076                 ret = err;
1077
1078         spin_lock(&fs_info->qgroup_lock);
1079         del_relation_rb(fs_info, src, dst);
1080         spin_unlock(&fs_info->qgroup_lock);
1081 out:
1082         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1083         return ret;
1084 }
1085
1086 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1087                         struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
1088 {
1089         struct btrfs_root *quota_root;
1090         struct btrfs_qgroup *qgroup;
1091         int ret = 0;
1092
1093         mutex_lock(&fs_info->qgroup_ioctl_lock);
1094         quota_root = fs_info->quota_root;
1095         if (!quota_root) {
1096                 ret = -EINVAL;
1097                 goto out;
1098         }
1099         qgroup = find_qgroup_rb(fs_info, qgroupid);
1100         if (qgroup) {
1101                 ret = -EEXIST;
1102                 goto out;
1103         }
1104
1105         ret = add_qgroup_item(trans, quota_root, qgroupid);
1106         if (ret)
1107                 goto out;
1108
1109         spin_lock(&fs_info->qgroup_lock);
1110         qgroup = add_qgroup_rb(fs_info, qgroupid);
1111         spin_unlock(&fs_info->qgroup_lock);
1112
1113         if (IS_ERR(qgroup))
1114                 ret = PTR_ERR(qgroup);
1115 out:
1116         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1117         return ret;
1118 }
1119
1120 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1121                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1122 {
1123         struct btrfs_root *quota_root;
1124         struct btrfs_qgroup *qgroup;
1125         int ret = 0;
1126
1127         mutex_lock(&fs_info->qgroup_ioctl_lock);
1128         quota_root = fs_info->quota_root;
1129         if (!quota_root) {
1130                 ret = -EINVAL;
1131                 goto out;
1132         }
1133
1134         qgroup = find_qgroup_rb(fs_info, qgroupid);
1135         if (!qgroup) {
1136                 ret = -ENOENT;
1137                 goto out;
1138         } else {
1139                 /* check if there are no relations to this qgroup */
1140                 if (!list_empty(&qgroup->groups) ||
1141                     !list_empty(&qgroup->members)) {
1142                         ret = -EBUSY;
1143                         goto out;
1144                 }
1145         }
1146         ret = del_qgroup_item(trans, quota_root, qgroupid);
1147
1148         spin_lock(&fs_info->qgroup_lock);
1149         del_qgroup_rb(quota_root->fs_info, qgroupid);
1150         spin_unlock(&fs_info->qgroup_lock);
1151 out:
1152         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1153         return ret;
1154 }
1155
1156 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1157                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1158                        struct btrfs_qgroup_limit *limit)
1159 {
1160         struct btrfs_root *quota_root;
1161         struct btrfs_qgroup *qgroup;
1162         int ret = 0;
1163
1164         mutex_lock(&fs_info->qgroup_ioctl_lock);
1165         quota_root = fs_info->quota_root;
1166         if (!quota_root) {
1167                 ret = -EINVAL;
1168                 goto out;
1169         }
1170
1171         qgroup = find_qgroup_rb(fs_info, qgroupid);
1172         if (!qgroup) {
1173                 ret = -ENOENT;
1174                 goto out;
1175         }
1176         ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1177                                        limit->flags, limit->max_rfer,
1178                                        limit->max_excl, limit->rsv_rfer,
1179                                        limit->rsv_excl);
1180         if (ret) {
1181                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1182                 printk(KERN_INFO "unable to update quota limit for %llu\n",
1183                        (unsigned long long)qgroupid);
1184         }
1185
1186         spin_lock(&fs_info->qgroup_lock);
1187         qgroup->lim_flags = limit->flags;
1188         qgroup->max_rfer = limit->max_rfer;
1189         qgroup->max_excl = limit->max_excl;
1190         qgroup->rsv_rfer = limit->rsv_rfer;
1191         qgroup->rsv_excl = limit->rsv_excl;
1192         spin_unlock(&fs_info->qgroup_lock);
1193 out:
1194         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1195         return ret;
1196 }
1197
1198 /*
1199  * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1200  * the modification into a list that's later used by btrfs_end_transaction to
1201  * pass the recorded modifications on to btrfs_qgroup_account_ref.
1202  */
1203 int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1204                             struct btrfs_delayed_ref_node *node,
1205                             struct btrfs_delayed_extent_op *extent_op)
1206 {
1207         struct qgroup_update *u;
1208
1209         BUG_ON(!trans->delayed_ref_elem.seq);
1210         u = kmalloc(sizeof(*u), GFP_NOFS);
1211         if (!u)
1212                 return -ENOMEM;
1213
1214         u->node = node;
1215         u->extent_op = extent_op;
1216         list_add_tail(&u->list, &trans->qgroup_ref_list);
1217
1218         return 0;
1219 }
1220
1221 static int qgroup_account_ref_step1(struct btrfs_fs_info *fs_info,
1222                                     struct ulist *roots, struct ulist *tmp,
1223                                     u64 seq)
1224 {
1225         struct ulist_node *unode;
1226         struct ulist_iterator uiter;
1227         struct ulist_node *tmp_unode;
1228         struct ulist_iterator tmp_uiter;
1229         struct btrfs_qgroup *qg;
1230         int ret;
1231
1232         ULIST_ITER_INIT(&uiter);
1233         while ((unode = ulist_next(roots, &uiter))) {
1234                 qg = find_qgroup_rb(fs_info, unode->val);
1235                 if (!qg)
1236                         continue;
1237
1238                 ulist_reinit(tmp);
1239                                                 /* XXX id not needed */
1240                 ret = ulist_add(tmp, qg->qgroupid,
1241                                 (u64)(uintptr_t)qg, GFP_ATOMIC);
1242                 if (ret < 0)
1243                         return ret;
1244                 ULIST_ITER_INIT(&tmp_uiter);
1245                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1246                         struct btrfs_qgroup_list *glist;
1247
1248                         qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1249                         if (qg->refcnt < seq)
1250                                 qg->refcnt = seq + 1;
1251                         else
1252                                 ++qg->refcnt;
1253
1254                         list_for_each_entry(glist, &qg->groups, next_group) {
1255                                 ret = ulist_add(tmp, glist->group->qgroupid,
1256                                                 (u64)(uintptr_t)glist->group,
1257                                                 GFP_ATOMIC);
1258                                 if (ret < 0)
1259                                         return ret;
1260                         }
1261                 }
1262         }
1263
1264         return 0;
1265 }
1266
1267 static int qgroup_account_ref_step2(struct btrfs_fs_info *fs_info,
1268                                     struct ulist *roots, struct ulist *tmp,
1269                                     u64 seq, int sgn, u64 num_bytes,
1270                                     struct btrfs_qgroup *qgroup)
1271 {
1272         struct ulist_node *unode;
1273         struct ulist_iterator uiter;
1274         struct btrfs_qgroup *qg;
1275         struct btrfs_qgroup_list *glist;
1276         int ret;
1277
1278         ulist_reinit(tmp);
1279         ret = ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
1280         if (ret < 0)
1281                 return ret;
1282
1283         ULIST_ITER_INIT(&uiter);
1284         while ((unode = ulist_next(tmp, &uiter))) {
1285                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1286                 if (qg->refcnt < seq) {
1287                         /* not visited by step 1 */
1288                         qg->rfer += sgn * num_bytes;
1289                         qg->rfer_cmpr += sgn * num_bytes;
1290                         if (roots->nnodes == 0) {
1291                                 qg->excl += sgn * num_bytes;
1292                                 qg->excl_cmpr += sgn * num_bytes;
1293                         }
1294                         qgroup_dirty(fs_info, qg);
1295                 }
1296                 WARN_ON(qg->tag >= seq);
1297                 qg->tag = seq;
1298
1299                 list_for_each_entry(glist, &qg->groups, next_group) {
1300                         ret = ulist_add(tmp, glist->group->qgroupid,
1301                                         (uintptr_t)glist->group, GFP_ATOMIC);
1302                         if (ret < 0)
1303                                 return ret;
1304                 }
1305         }
1306
1307         return 0;
1308 }
1309
1310 static int qgroup_account_ref_step3(struct btrfs_fs_info *fs_info,
1311                                     struct ulist *roots, struct ulist *tmp,
1312                                     u64 seq, int sgn, u64 num_bytes)
1313 {
1314         struct ulist_node *unode;
1315         struct ulist_iterator uiter;
1316         struct btrfs_qgroup *qg;
1317         struct ulist_node *tmp_unode;
1318         struct ulist_iterator tmp_uiter;
1319         int ret;
1320
1321         ULIST_ITER_INIT(&uiter);
1322         while ((unode = ulist_next(roots, &uiter))) {
1323                 qg = find_qgroup_rb(fs_info, unode->val);
1324                 if (!qg)
1325                         continue;
1326
1327                 ulist_reinit(tmp);
1328                 ret = ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
1329                 if (ret < 0)
1330                         return ret;
1331
1332                 ULIST_ITER_INIT(&tmp_uiter);
1333                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1334                         struct btrfs_qgroup_list *glist;
1335
1336                         qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1337                         if (qg->tag == seq)
1338                                 continue;
1339
1340                         if (qg->refcnt - seq == roots->nnodes) {
1341                                 qg->excl -= sgn * num_bytes;
1342                                 qg->excl_cmpr -= sgn * num_bytes;
1343                                 qgroup_dirty(fs_info, qg);
1344                         }
1345
1346                         list_for_each_entry(glist, &qg->groups, next_group) {
1347                                 ret = ulist_add(tmp, glist->group->qgroupid,
1348                                                 (uintptr_t)glist->group,
1349                                                 GFP_ATOMIC);
1350                                 if (ret < 0)
1351                                         return ret;
1352                         }
1353                 }
1354         }
1355
1356         return 0;
1357 }
1358
1359 /*
1360  * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1361  * from the fs. First, all roots referencing the extent are searched, and
1362  * then the space is accounted accordingly to the different roots. The
1363  * accounting algorithm works in 3 steps documented inline.
1364  */
1365 int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1366                              struct btrfs_fs_info *fs_info,
1367                              struct btrfs_delayed_ref_node *node,
1368                              struct btrfs_delayed_extent_op *extent_op)
1369 {
1370         struct btrfs_key ins;
1371         struct btrfs_root *quota_root;
1372         u64 ref_root;
1373         struct btrfs_qgroup *qgroup;
1374         struct ulist *roots = NULL;
1375         u64 seq;
1376         int ret = 0;
1377         int sgn;
1378
1379         if (!fs_info->quota_enabled)
1380                 return 0;
1381
1382         BUG_ON(!fs_info->quota_root);
1383
1384         ins.objectid = node->bytenr;
1385         ins.offset = node->num_bytes;
1386         ins.type = BTRFS_EXTENT_ITEM_KEY;
1387
1388         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1389             node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1390                 struct btrfs_delayed_tree_ref *ref;
1391                 ref = btrfs_delayed_node_to_tree_ref(node);
1392                 ref_root = ref->root;
1393         } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1394                    node->type == BTRFS_SHARED_DATA_REF_KEY) {
1395                 struct btrfs_delayed_data_ref *ref;
1396                 ref = btrfs_delayed_node_to_data_ref(node);
1397                 ref_root = ref->root;
1398         } else {
1399                 BUG();
1400         }
1401
1402         if (!is_fstree(ref_root)) {
1403                 /*
1404                  * non-fs-trees are not being accounted
1405                  */
1406                 return 0;
1407         }
1408
1409         switch (node->action) {
1410         case BTRFS_ADD_DELAYED_REF:
1411         case BTRFS_ADD_DELAYED_EXTENT:
1412                 sgn = 1;
1413                 seq = btrfs_tree_mod_seq_prev(node->seq);
1414                 break;
1415         case BTRFS_DROP_DELAYED_REF:
1416                 sgn = -1;
1417                 seq = node->seq;
1418                 break;
1419         case BTRFS_UPDATE_DELAYED_HEAD:
1420                 return 0;
1421         default:
1422                 BUG();
1423         }
1424
1425         mutex_lock(&fs_info->qgroup_rescan_lock);
1426         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1427                 if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
1428                         mutex_unlock(&fs_info->qgroup_rescan_lock);
1429                         return 0;
1430                 }
1431         }
1432         mutex_unlock(&fs_info->qgroup_rescan_lock);
1433
1434         /*
1435          * the delayed ref sequence number we pass depends on the direction of
1436          * the operation. for add operations, we pass
1437          * tree_mod_log_prev_seq(node->seq) to skip
1438          * the delayed ref's current sequence number, because we need the state
1439          * of the tree before the add operation. for delete operations, we pass
1440          * (node->seq) to include the delayed ref's current sequence number,
1441          * because we need the state of the tree after the delete operation.
1442          */
1443         ret = btrfs_find_all_roots(trans, fs_info, node->bytenr, seq, &roots);
1444         if (ret < 0)
1445                 return ret;
1446
1447         spin_lock(&fs_info->qgroup_lock);
1448
1449         quota_root = fs_info->quota_root;
1450         if (!quota_root)
1451                 goto unlock;
1452
1453         qgroup = find_qgroup_rb(fs_info, ref_root);
1454         if (!qgroup)
1455                 goto unlock;
1456
1457         /*
1458          * step 1: for each old ref, visit all nodes once and inc refcnt
1459          */
1460         ulist_reinit(fs_info->qgroup_ulist);
1461         seq = fs_info->qgroup_seq;
1462         fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1463
1464         ret = qgroup_account_ref_step1(fs_info, roots, fs_info->qgroup_ulist,
1465                                        seq);
1466         if (ret)
1467                 goto unlock;
1468
1469         /*
1470          * step 2: walk from the new root
1471          */
1472         ret = qgroup_account_ref_step2(fs_info, roots, fs_info->qgroup_ulist,
1473                                        seq, sgn, node->num_bytes, qgroup);
1474         if (ret)
1475                 goto unlock;
1476
1477         /*
1478          * step 3: walk again from old refs
1479          */
1480         ret = qgroup_account_ref_step3(fs_info, roots, fs_info->qgroup_ulist,
1481                                        seq, sgn, node->num_bytes);
1482         if (ret)
1483                 goto unlock;
1484
1485 unlock:
1486         spin_unlock(&fs_info->qgroup_lock);
1487         ulist_free(roots);
1488
1489         return ret;
1490 }
1491
1492 /*
1493  * called from commit_transaction. Writes all changed qgroups to disk.
1494  */
1495 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1496                       struct btrfs_fs_info *fs_info)
1497 {
1498         struct btrfs_root *quota_root = fs_info->quota_root;
1499         int ret = 0;
1500         int start_rescan_worker = 0;
1501
1502         if (!quota_root)
1503                 goto out;
1504
1505         if (!fs_info->quota_enabled && fs_info->pending_quota_state)
1506                 start_rescan_worker = 1;
1507
1508         fs_info->quota_enabled = fs_info->pending_quota_state;
1509
1510         spin_lock(&fs_info->qgroup_lock);
1511         while (!list_empty(&fs_info->dirty_qgroups)) {
1512                 struct btrfs_qgroup *qgroup;
1513                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1514                                           struct btrfs_qgroup, dirty);
1515                 list_del_init(&qgroup->dirty);
1516                 spin_unlock(&fs_info->qgroup_lock);
1517                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1518                 if (ret)
1519                         fs_info->qgroup_flags |=
1520                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1521                 spin_lock(&fs_info->qgroup_lock);
1522         }
1523         if (fs_info->quota_enabled)
1524                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1525         else
1526                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1527         spin_unlock(&fs_info->qgroup_lock);
1528
1529         ret = update_qgroup_status_item(trans, fs_info, quota_root);
1530         if (ret)
1531                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1532
1533         if (!ret && start_rescan_worker) {
1534                 ret = qgroup_rescan_init(fs_info, 0, 1);
1535                 if (!ret) {
1536                         qgroup_rescan_zero_tracking(fs_info);
1537                         btrfs_queue_worker(&fs_info->qgroup_rescan_workers,
1538                                            &fs_info->qgroup_rescan_work);
1539                 }
1540                 ret = 0;
1541         }
1542
1543 out:
1544
1545         return ret;
1546 }
1547
1548 /*
1549  * copy the acounting information between qgroups. This is necessary when a
1550  * snapshot or a subvolume is created
1551  */
1552 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1553                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1554                          struct btrfs_qgroup_inherit *inherit)
1555 {
1556         int ret = 0;
1557         int i;
1558         u64 *i_qgroups;
1559         struct btrfs_root *quota_root = fs_info->quota_root;
1560         struct btrfs_qgroup *srcgroup;
1561         struct btrfs_qgroup *dstgroup;
1562         u32 level_size = 0;
1563         u64 nums;
1564
1565         mutex_lock(&fs_info->qgroup_ioctl_lock);
1566         if (!fs_info->quota_enabled)
1567                 goto out;
1568
1569         if (!quota_root) {
1570                 ret = -EINVAL;
1571                 goto out;
1572         }
1573
1574         if (inherit) {
1575                 i_qgroups = (u64 *)(inherit + 1);
1576                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1577                        2 * inherit->num_excl_copies;
1578                 for (i = 0; i < nums; ++i) {
1579                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1580                         if (!srcgroup) {
1581                                 ret = -EINVAL;
1582                                 goto out;
1583                         }
1584                         ++i_qgroups;
1585                 }
1586         }
1587
1588         /*
1589          * create a tracking group for the subvol itself
1590          */
1591         ret = add_qgroup_item(trans, quota_root, objectid);
1592         if (ret)
1593                 goto out;
1594
1595         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1596                 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1597                                                inherit->lim.flags,
1598                                                inherit->lim.max_rfer,
1599                                                inherit->lim.max_excl,
1600                                                inherit->lim.rsv_rfer,
1601                                                inherit->lim.rsv_excl);
1602                 if (ret)
1603                         goto out;
1604         }
1605
1606         if (srcid) {
1607                 struct btrfs_root *srcroot;
1608                 struct btrfs_key srckey;
1609                 int srcroot_level;
1610
1611                 srckey.objectid = srcid;
1612                 srckey.type = BTRFS_ROOT_ITEM_KEY;
1613                 srckey.offset = (u64)-1;
1614                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1615                 if (IS_ERR(srcroot)) {
1616                         ret = PTR_ERR(srcroot);
1617                         goto out;
1618                 }
1619
1620                 rcu_read_lock();
1621                 srcroot_level = btrfs_header_level(srcroot->node);
1622                 level_size = btrfs_level_size(srcroot, srcroot_level);
1623                 rcu_read_unlock();
1624         }
1625
1626         /*
1627          * add qgroup to all inherited groups
1628          */
1629         if (inherit) {
1630                 i_qgroups = (u64 *)(inherit + 1);
1631                 for (i = 0; i < inherit->num_qgroups; ++i) {
1632                         ret = add_qgroup_relation_item(trans, quota_root,
1633                                                        objectid, *i_qgroups);
1634                         if (ret)
1635                                 goto out;
1636                         ret = add_qgroup_relation_item(trans, quota_root,
1637                                                        *i_qgroups, objectid);
1638                         if (ret)
1639                                 goto out;
1640                         ++i_qgroups;
1641                 }
1642         }
1643
1644
1645         spin_lock(&fs_info->qgroup_lock);
1646
1647         dstgroup = add_qgroup_rb(fs_info, objectid);
1648         if (IS_ERR(dstgroup)) {
1649                 ret = PTR_ERR(dstgroup);
1650                 goto unlock;
1651         }
1652
1653         if (srcid) {
1654                 srcgroup = find_qgroup_rb(fs_info, srcid);
1655                 if (!srcgroup)
1656                         goto unlock;
1657                 dstgroup->rfer = srcgroup->rfer - level_size;
1658                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1659                 srcgroup->excl = level_size;
1660                 srcgroup->excl_cmpr = level_size;
1661                 qgroup_dirty(fs_info, dstgroup);
1662                 qgroup_dirty(fs_info, srcgroup);
1663         }
1664
1665         if (!inherit)
1666                 goto unlock;
1667
1668         i_qgroups = (u64 *)(inherit + 1);
1669         for (i = 0; i < inherit->num_qgroups; ++i) {
1670                 ret = add_relation_rb(quota_root->fs_info, objectid,
1671                                       *i_qgroups);
1672                 if (ret)
1673                         goto unlock;
1674                 ++i_qgroups;
1675         }
1676
1677         for (i = 0; i <  inherit->num_ref_copies; ++i) {
1678                 struct btrfs_qgroup *src;
1679                 struct btrfs_qgroup *dst;
1680
1681                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1682                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1683
1684                 if (!src || !dst) {
1685                         ret = -EINVAL;
1686                         goto unlock;
1687                 }
1688
1689                 dst->rfer = src->rfer - level_size;
1690                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1691                 i_qgroups += 2;
1692         }
1693         for (i = 0; i <  inherit->num_excl_copies; ++i) {
1694                 struct btrfs_qgroup *src;
1695                 struct btrfs_qgroup *dst;
1696
1697                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1698                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1699
1700                 if (!src || !dst) {
1701                         ret = -EINVAL;
1702                         goto unlock;
1703                 }
1704
1705                 dst->excl = src->excl + level_size;
1706                 dst->excl_cmpr = src->excl_cmpr + level_size;
1707                 i_qgroups += 2;
1708         }
1709
1710 unlock:
1711         spin_unlock(&fs_info->qgroup_lock);
1712 out:
1713         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1714         return ret;
1715 }
1716
1717 /*
1718  * reserve some space for a qgroup and all its parents. The reservation takes
1719  * place with start_transaction or dealloc_reserve, similar to ENOSPC
1720  * accounting. If not enough space is available, EDQUOT is returned.
1721  * We assume that the requested space is new for all qgroups.
1722  */
1723 int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1724 {
1725         struct btrfs_root *quota_root;
1726         struct btrfs_qgroup *qgroup;
1727         struct btrfs_fs_info *fs_info = root->fs_info;
1728         u64 ref_root = root->root_key.objectid;
1729         int ret = 0;
1730         struct ulist_node *unode;
1731         struct ulist_iterator uiter;
1732
1733         if (!is_fstree(ref_root))
1734                 return 0;
1735
1736         if (num_bytes == 0)
1737                 return 0;
1738
1739         spin_lock(&fs_info->qgroup_lock);
1740         quota_root = fs_info->quota_root;
1741         if (!quota_root)
1742                 goto out;
1743
1744         qgroup = find_qgroup_rb(fs_info, ref_root);
1745         if (!qgroup)
1746                 goto out;
1747
1748         /*
1749          * in a first step, we check all affected qgroups if any limits would
1750          * be exceeded
1751          */
1752         ulist_reinit(fs_info->qgroup_ulist);
1753         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
1754                         (uintptr_t)qgroup, GFP_ATOMIC);
1755         if (ret < 0)
1756                 goto out;
1757         ULIST_ITER_INIT(&uiter);
1758         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
1759                 struct btrfs_qgroup *qg;
1760                 struct btrfs_qgroup_list *glist;
1761
1762                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1763
1764                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1765                     qg->reserved + (s64)qg->rfer + num_bytes >
1766                     qg->max_rfer) {
1767                         ret = -EDQUOT;
1768                         goto out;
1769                 }
1770
1771                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1772                     qg->reserved + (s64)qg->excl + num_bytes >
1773                     qg->max_excl) {
1774                         ret = -EDQUOT;
1775                         goto out;
1776                 }
1777
1778                 list_for_each_entry(glist, &qg->groups, next_group) {
1779                         ret = ulist_add(fs_info->qgroup_ulist,
1780                                         glist->group->qgroupid,
1781                                         (uintptr_t)glist->group, GFP_ATOMIC);
1782                         if (ret < 0)
1783                                 goto out;
1784                 }
1785         }
1786         ret = 0;
1787         /*
1788          * no limits exceeded, now record the reservation into all qgroups
1789          */
1790         ULIST_ITER_INIT(&uiter);
1791         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
1792                 struct btrfs_qgroup *qg;
1793
1794                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1795
1796                 qg->reserved += num_bytes;
1797         }
1798
1799 out:
1800         spin_unlock(&fs_info->qgroup_lock);
1801         return ret;
1802 }
1803
1804 void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1805 {
1806         struct btrfs_root *quota_root;
1807         struct btrfs_qgroup *qgroup;
1808         struct btrfs_fs_info *fs_info = root->fs_info;
1809         struct ulist_node *unode;
1810         struct ulist_iterator uiter;
1811         u64 ref_root = root->root_key.objectid;
1812         int ret = 0;
1813
1814         if (!is_fstree(ref_root))
1815                 return;
1816
1817         if (num_bytes == 0)
1818                 return;
1819
1820         spin_lock(&fs_info->qgroup_lock);
1821
1822         quota_root = fs_info->quota_root;
1823         if (!quota_root)
1824                 goto out;
1825
1826         qgroup = find_qgroup_rb(fs_info, ref_root);
1827         if (!qgroup)
1828                 goto out;
1829
1830         ulist_reinit(fs_info->qgroup_ulist);
1831         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
1832                         (uintptr_t)qgroup, GFP_ATOMIC);
1833         if (ret < 0)
1834                 goto out;
1835         ULIST_ITER_INIT(&uiter);
1836         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
1837                 struct btrfs_qgroup *qg;
1838                 struct btrfs_qgroup_list *glist;
1839
1840                 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1841
1842                 qg->reserved -= num_bytes;
1843
1844                 list_for_each_entry(glist, &qg->groups, next_group) {
1845                         ret = ulist_add(fs_info->qgroup_ulist,
1846                                         glist->group->qgroupid,
1847                                         (uintptr_t)glist->group, GFP_ATOMIC);
1848                         if (ret < 0)
1849                                 goto out;
1850                 }
1851         }
1852
1853 out:
1854         spin_unlock(&fs_info->qgroup_lock);
1855 }
1856
1857 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1858 {
1859         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1860                 return;
1861         pr_err("btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %#x.%x\n",
1862                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1863                 (u32)(trans->delayed_ref_elem.seq >> 32),
1864                 (u32)trans->delayed_ref_elem.seq);
1865         BUG();
1866 }
1867
1868 /*
1869  * returns < 0 on error, 0 when more leafs are to be scanned.
1870  * returns 1 when done, 2 when done and FLAG_INCONSISTENT was cleared.
1871  */
1872 static int
1873 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1874                    struct btrfs_trans_handle *trans, struct ulist *tmp,
1875                    struct extent_buffer *scratch_leaf)
1876 {
1877         struct btrfs_key found;
1878         struct ulist *roots = NULL;
1879         struct ulist_node *unode;
1880         struct ulist_iterator uiter;
1881         struct seq_list tree_mod_seq_elem = {};
1882         u64 seq;
1883         int slot;
1884         int ret;
1885
1886         path->leave_spinning = 1;
1887         mutex_lock(&fs_info->qgroup_rescan_lock);
1888         ret = btrfs_search_slot_for_read(fs_info->extent_root,
1889                                          &fs_info->qgroup_rescan_progress,
1890                                          path, 1, 0);
1891
1892         pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
1893                  (unsigned long long)fs_info->qgroup_rescan_progress.objectid,
1894                  fs_info->qgroup_rescan_progress.type,
1895                  (unsigned long long)fs_info->qgroup_rescan_progress.offset,
1896                  ret);
1897
1898         if (ret) {
1899                 /*
1900                  * The rescan is about to end, we will not be scanning any
1901                  * further blocks. We cannot unset the RESCAN flag here, because
1902                  * we want to commit the transaction if everything went well.
1903                  * To make the live accounting work in this phase, we set our
1904                  * scan progress pointer such that every real extent objectid
1905                  * will be smaller.
1906                  */
1907                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
1908                 btrfs_release_path(path);
1909                 mutex_unlock(&fs_info->qgroup_rescan_lock);
1910                 return ret;
1911         }
1912
1913         btrfs_item_key_to_cpu(path->nodes[0], &found,
1914                               btrfs_header_nritems(path->nodes[0]) - 1);
1915         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
1916
1917         btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1918         memcpy(scratch_leaf, path->nodes[0], sizeof(*scratch_leaf));
1919         slot = path->slots[0];
1920         btrfs_release_path(path);
1921         mutex_unlock(&fs_info->qgroup_rescan_lock);
1922
1923         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
1924                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
1925                 if (found.type != BTRFS_EXTENT_ITEM_KEY)
1926                         continue;
1927                 ret = btrfs_find_all_roots(trans, fs_info, found.objectid,
1928                                            tree_mod_seq_elem.seq, &roots);
1929                 if (ret < 0)
1930                         goto out;
1931                 spin_lock(&fs_info->qgroup_lock);
1932                 seq = fs_info->qgroup_seq;
1933                 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1934
1935                 ret = qgroup_account_ref_step1(fs_info, roots, tmp, seq);
1936                 if (ret) {
1937                         spin_unlock(&fs_info->qgroup_lock);
1938                         ulist_free(roots);
1939                         goto out;
1940                 }
1941
1942                 /*
1943                  * step2 of btrfs_qgroup_account_ref works from a single root,
1944                  * we're doing all at once here.
1945                  */
1946                 ulist_reinit(tmp);
1947                 ULIST_ITER_INIT(&uiter);
1948                 while ((unode = ulist_next(roots, &uiter))) {
1949                         struct btrfs_qgroup *qg;
1950
1951                         qg = find_qgroup_rb(fs_info, unode->val);
1952                         if (!qg)
1953                                 continue;
1954
1955                         ret = ulist_add(tmp, qg->qgroupid, (uintptr_t)qg,
1956                                         GFP_ATOMIC);
1957                         if (ret < 0) {
1958                                 spin_unlock(&fs_info->qgroup_lock);
1959                                 ulist_free(roots);
1960                                 goto out;
1961                         }
1962                 }
1963
1964                 /* this loop is similar to step 2 of btrfs_qgroup_account_ref */
1965                 ULIST_ITER_INIT(&uiter);
1966                 while ((unode = ulist_next(tmp, &uiter))) {
1967                         struct btrfs_qgroup *qg;
1968                         struct btrfs_qgroup_list *glist;
1969
1970                         qg = (struct btrfs_qgroup *)(uintptr_t) unode->aux;
1971                         qg->rfer += found.offset;
1972                         qg->rfer_cmpr += found.offset;
1973                         WARN_ON(qg->tag >= seq);
1974                         if (qg->refcnt - seq == roots->nnodes) {
1975                                 qg->excl += found.offset;
1976                                 qg->excl_cmpr += found.offset;
1977                         }
1978                         qgroup_dirty(fs_info, qg);
1979
1980                         list_for_each_entry(glist, &qg->groups, next_group) {
1981                                 ret = ulist_add(tmp, glist->group->qgroupid,
1982                                                 (uintptr_t)glist->group,
1983                                                 GFP_ATOMIC);
1984                                 if (ret < 0) {
1985                                         spin_unlock(&fs_info->qgroup_lock);
1986                                         ulist_free(roots);
1987                                         goto out;
1988                                 }
1989                         }
1990                 }
1991
1992                 spin_unlock(&fs_info->qgroup_lock);
1993                 ulist_free(roots);
1994                 ret = 0;
1995         }
1996
1997 out:
1998         btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1999
2000         return ret;
2001 }
2002
2003 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2004 {
2005         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2006                                                      qgroup_rescan_work);
2007         struct btrfs_path *path;
2008         struct btrfs_trans_handle *trans = NULL;
2009         struct ulist *tmp = NULL;
2010         struct extent_buffer *scratch_leaf = NULL;
2011         int err = -ENOMEM;
2012
2013         path = btrfs_alloc_path();
2014         if (!path)
2015                 goto out;
2016         tmp = ulist_alloc(GFP_NOFS);
2017         if (!tmp)
2018                 goto out;
2019         scratch_leaf = kmalloc(sizeof(*scratch_leaf), GFP_NOFS);
2020         if (!scratch_leaf)
2021                 goto out;
2022
2023         err = 0;
2024         while (!err) {
2025                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2026                 if (IS_ERR(trans)) {
2027                         err = PTR_ERR(trans);
2028                         break;
2029                 }
2030                 if (!fs_info->quota_enabled) {
2031                         err = -EINTR;
2032                 } else {
2033                         err = qgroup_rescan_leaf(fs_info, path, trans,
2034                                                  tmp, scratch_leaf);
2035                 }
2036                 if (err > 0)
2037                         btrfs_commit_transaction(trans, fs_info->fs_root);
2038                 else
2039                         btrfs_end_transaction(trans, fs_info->fs_root);
2040         }
2041
2042 out:
2043         kfree(scratch_leaf);
2044         ulist_free(tmp);
2045         btrfs_free_path(path);
2046
2047         mutex_lock(&fs_info->qgroup_rescan_lock);
2048         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2049
2050         if (err == 2 &&
2051             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2052                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2053         } else if (err < 0) {
2054                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2055         }
2056         mutex_unlock(&fs_info->qgroup_rescan_lock);
2057
2058         if (err >= 0) {
2059                 pr_info("btrfs: qgroup scan completed%s\n",
2060                         err == 2 ? " (inconsistency flag cleared)" : "");
2061         } else {
2062                 pr_err("btrfs: qgroup scan failed with %d\n", err);
2063         }
2064
2065         complete_all(&fs_info->qgroup_rescan_completion);
2066 }
2067
2068 /*
2069  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2070  * memory required for the rescan context.
2071  */
2072 static int
2073 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2074                    int init_flags)
2075 {
2076         int ret = 0;
2077
2078         if (!init_flags &&
2079             (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2080              !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2081                 ret = -EINVAL;
2082                 goto err;
2083         }
2084
2085         mutex_lock(&fs_info->qgroup_rescan_lock);
2086         spin_lock(&fs_info->qgroup_lock);
2087
2088         if (init_flags) {
2089                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2090                         ret = -EINPROGRESS;
2091                 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2092                         ret = -EINVAL;
2093
2094                 if (ret) {
2095                         spin_unlock(&fs_info->qgroup_lock);
2096                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2097                         goto err;
2098                 }
2099
2100                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2101         }
2102
2103         memset(&fs_info->qgroup_rescan_progress, 0,
2104                 sizeof(fs_info->qgroup_rescan_progress));
2105         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2106
2107         spin_unlock(&fs_info->qgroup_lock);
2108         mutex_unlock(&fs_info->qgroup_rescan_lock);
2109
2110         init_completion(&fs_info->qgroup_rescan_completion);
2111
2112         memset(&fs_info->qgroup_rescan_work, 0,
2113                sizeof(fs_info->qgroup_rescan_work));
2114         fs_info->qgroup_rescan_work.func = btrfs_qgroup_rescan_worker;
2115
2116         if (ret) {
2117 err:
2118                 pr_info("btrfs: qgroup_rescan_init failed with %d\n", ret);
2119                 return ret;
2120         }
2121
2122         return 0;
2123 }
2124
2125 static void
2126 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2127 {
2128         struct rb_node *n;
2129         struct btrfs_qgroup *qgroup;
2130
2131         spin_lock(&fs_info->qgroup_lock);
2132         /* clear all current qgroup tracking information */
2133         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2134                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2135                 qgroup->rfer = 0;
2136                 qgroup->rfer_cmpr = 0;
2137                 qgroup->excl = 0;
2138                 qgroup->excl_cmpr = 0;
2139         }
2140         spin_unlock(&fs_info->qgroup_lock);
2141 }
2142
2143 int
2144 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2145 {
2146         int ret = 0;
2147         struct btrfs_trans_handle *trans;
2148
2149         ret = qgroup_rescan_init(fs_info, 0, 1);
2150         if (ret)
2151                 return ret;
2152
2153         /*
2154          * We have set the rescan_progress to 0, which means no more
2155          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2156          * However, btrfs_qgroup_account_ref may be right after its call
2157          * to btrfs_find_all_roots, in which case it would still do the
2158          * accounting.
2159          * To solve this, we're committing the transaction, which will
2160          * ensure we run all delayed refs and only after that, we are
2161          * going to clear all tracking information for a clean start.
2162          */
2163
2164         trans = btrfs_join_transaction(fs_info->fs_root);
2165         if (IS_ERR(trans)) {
2166                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2167                 return PTR_ERR(trans);
2168         }
2169         ret = btrfs_commit_transaction(trans, fs_info->fs_root);
2170         if (ret) {
2171                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2172                 return ret;
2173         }
2174
2175         qgroup_rescan_zero_tracking(fs_info);
2176
2177         btrfs_queue_worker(&fs_info->qgroup_rescan_workers,
2178                            &fs_info->qgroup_rescan_work);
2179
2180         return 0;
2181 }
2182
2183 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info)
2184 {
2185         int running;
2186         int ret = 0;
2187
2188         mutex_lock(&fs_info->qgroup_rescan_lock);
2189         spin_lock(&fs_info->qgroup_lock);
2190         running = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2191         spin_unlock(&fs_info->qgroup_lock);
2192         mutex_unlock(&fs_info->qgroup_rescan_lock);
2193
2194         if (running)
2195                 ret = wait_for_completion_interruptible(
2196                                         &fs_info->qgroup_rescan_completion);
2197
2198         return ret;
2199 }
2200
2201 /*
2202  * this is only called from open_ctree where we're still single threaded, thus
2203  * locking is omitted here.
2204  */
2205 void
2206 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2207 {
2208         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2209                 btrfs_queue_worker(&fs_info->qgroup_rescan_workers,
2210                                    &fs_info->qgroup_rescan_work);
2211 }