fs/btrfs/qgroup.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2011 STRATO.  All rights reserved.
   4  */
   5
   6 #include <linux/sched.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/writeback.h>
   9 #include <linux/blkdev.h>
  10 #include <linux/rbtree.h>
  11 #include <linux/slab.h>
  12 #include <linux/workqueue.h>
  13 #include <linux/btrfs.h>
  14 #include <linux/sizes.h>
  15
  16 #include "ctree.h"
  17 #include "transaction.h"
  18 #include "disk-io.h"
  19 #include "locking.h"
  20 #include "ulist.h"
  21 #include "backref.h"
  22 #include "extent_io.h"
  23 #include "qgroup.h"
  24
  25
  26 /* TODO XXX FIXME
  27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  28  *  - reorganize keys
  29  *  - compressed
  30  *  - sync
  31  *  - copy also limits on subvol creation
  32  *  - limit
  33  *  - caches for ulists
  34  *  - performance benchmarks
  35  *  - check all ioctl parameters
  36  */
  37
  38 /*
  39  * Helpers to access qgroup reservation
  40  *
  41  * Callers should ensure the lock context and type are valid
  42  */
  43
  44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
  45 {
  46         u64 ret = 0;
  47         int i;
  48
  49         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
  50                 ret += qgroup->rsv.values[i];
  51
  52         return ret;
  53 }
  54
  55 #ifdef CONFIG_BTRFS_DEBUG
  56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
  57 {
  58         if (type == BTRFS_QGROUP_RSV_DATA)
  59                 return "data";
  60         if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
  61                 return "meta_pertrans";
  62         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
  63                 return "meta_prealloc";
  64         return NULL;
  65 }
  66 #endif
  67
  68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
  69                            struct btrfs_qgroup *qgroup, u64 num_bytes,
  70                            enum btrfs_qgroup_rsv_type type)
  71 {
  72         trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
  73         qgroup->rsv.values[type] += num_bytes;
  74 }
  75
  76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
  77                                struct btrfs_qgroup *qgroup, u64 num_bytes,
  78                                enum btrfs_qgroup_rsv_type type)
  79 {
  80         trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
  81         if (qgroup->rsv.values[type] >= num_bytes) {
  82                 qgroup->rsv.values[type] -= num_bytes;
  83                 return;
  84         }
  85 #ifdef CONFIG_BTRFS_DEBUG
  86         WARN_RATELIMIT(1,
  87                 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
  88                 qgroup->qgroupid, qgroup_rsv_type_str(type),
  89                 qgroup->rsv.values[type], num_bytes);
  90 #endif
  91         qgroup->rsv.values[type] = 0;
  92 }
  93
  94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
  95                                      struct btrfs_qgroup *dest,
  96                                      struct btrfs_qgroup *src)
  97 {
  98         int i;
  99
 100         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 101                 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
 102 }
 103
 104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
 105                                          struct btrfs_qgroup *dest,
 106                                           struct btrfs_qgroup *src)
 107 {
 108         int i;
 109
 110         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 111                 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
 112 }
 113
 114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
 115                                            int mod)
 116 {
 117         if (qg->old_refcnt < seq)
 118                 qg->old_refcnt = seq;
 119         qg->old_refcnt += mod;
 120 }
 121
 122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
 123                                            int mod)
 124 {
 125         if (qg->new_refcnt < seq)
 126                 qg->new_refcnt = seq;
 127         qg->new_refcnt += mod;
 128 }
 129
 130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
 131 {
 132         if (qg->old_refcnt < seq)
 133                 return 0;
 134         return qg->old_refcnt - seq;
 135 }
 136
 137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
 138 {
 139         if (qg->new_refcnt < seq)
 140                 return 0;
 141         return qg->new_refcnt - seq;
 142 }
 143
 144 /*
 145  * glue structure to represent the relations between qgroups.
 146  */
 147 struct btrfs_qgroup_list {
 148         struct list_head next_group;
 149         struct list_head next_member;
 150         struct btrfs_qgroup *group;
 151         struct btrfs_qgroup *member;
 152 };
 153
 154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
 155 {
 156         return (u64)(uintptr_t)qg;
 157 }
 158
 159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
 160 {
 161         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
 162 }
 163
 164 static int
 165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 166                    int init_flags);
 167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 168
 169 /* must be called with qgroup_ioctl_lock held */
 170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 171                                            u64 qgroupid)
 172 {
 173         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 174         struct btrfs_qgroup *qgroup;
 175
 176         while (n) {
 177                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 178                 if (qgroup->qgroupid < qgroupid)
 179                         n = n->rb_left;
 180                 else if (qgroup->qgroupid > qgroupid)
 181                         n = n->rb_right;
 182                 else
 183                         return qgroup;
 184         }
 185         return NULL;
 186 }
 187
 188 /* must be called with qgroup_lock held */
 189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 190                                           u64 qgroupid)
 191 {
 192         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 193         struct rb_node *parent = NULL;
 194         struct btrfs_qgroup *qgroup;
 195
 196         while (*p) {
 197                 parent = *p;
 198                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 199
 200                 if (qgroup->qgroupid < qgroupid)
 201                         p = &(*p)->rb_left;
 202                 else if (qgroup->qgroupid > qgroupid)
 203                         p = &(*p)->rb_right;
 204                 else
 205                         return qgroup;
 206         }
 207
 208         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 209         if (!qgroup)
 210                 return ERR_PTR(-ENOMEM);
 211
 212         qgroup->qgroupid = qgroupid;
 213         INIT_LIST_HEAD(&qgroup->groups);
 214         INIT_LIST_HEAD(&qgroup->members);
 215         INIT_LIST_HEAD(&qgroup->dirty);
 216
 217         rb_link_node(&qgroup->node, parent, p);
 218         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 219
 220         return qgroup;
 221 }
 222
 223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
 224 {
 225         struct btrfs_qgroup_list *list;
 226
 227         list_del(&qgroup->dirty);
 228         while (!list_empty(&qgroup->groups)) {
 229                 list = list_first_entry(&qgroup->groups,
 230                                         struct btrfs_qgroup_list, next_group);
 231                 list_del(&list->next_group);
 232                 list_del(&list->next_member);
 233                 kfree(list);
 234         }
 235
 236         while (!list_empty(&qgroup->members)) {
 237                 list = list_first_entry(&qgroup->members,
 238                                         struct btrfs_qgroup_list, next_member);
 239                 list_del(&list->next_group);
 240                 list_del(&list->next_member);
 241                 kfree(list);
 242         }
 243         kfree(qgroup);
 244 }
 245
 246 /* must be called with qgroup_lock held */
 247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 248 {
 249         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 250
 251         if (!qgroup)
 252                 return -ENOENT;
 253
 254         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 255         __del_qgroup_rb(qgroup);
 256         return 0;
 257 }
 258
 259 /* must be called with qgroup_lock held */
 260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 261                            u64 memberid, u64 parentid)
 262 {
 263         struct btrfs_qgroup *member;
 264         struct btrfs_qgroup *parent;
 265         struct btrfs_qgroup_list *list;
 266
 267         member = find_qgroup_rb(fs_info, memberid);
 268         parent = find_qgroup_rb(fs_info, parentid);
 269         if (!member || !parent)
 270                 return -ENOENT;
 271
 272         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 273         if (!list)
 274                 return -ENOMEM;
 275
 276         list->group = parent;
 277         list->member = member;
 278         list_add_tail(&list->next_group, &member->groups);
 279         list_add_tail(&list->next_member, &parent->members);
 280
 281         return 0;
 282 }
 283
 284 /* must be called with qgroup_lock held */
 285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 286                            u64 memberid, u64 parentid)
 287 {
 288         struct btrfs_qgroup *member;
 289         struct btrfs_qgroup *parent;
 290         struct btrfs_qgroup_list *list;
 291
 292         member = find_qgroup_rb(fs_info, memberid);
 293         parent = find_qgroup_rb(fs_info, parentid);
 294         if (!member || !parent)
 295                 return -ENOENT;
 296
 297         list_for_each_entry(list, &member->groups, next_group) {
 298                 if (list->group == parent) {
 299                         list_del(&list->next_group);
 300                         list_del(&list->next_member);
 301                         kfree(list);
 302                         return 0;
 303                 }
 304         }
 305         return -ENOENT;
 306 }
 307
 308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 310                                u64 rfer, u64 excl)
 311 {
 312         struct btrfs_qgroup *qgroup;
 313
 314         qgroup = find_qgroup_rb(fs_info, qgroupid);
 315         if (!qgroup)
 316                 return -EINVAL;
 317         if (qgroup->rfer != rfer || qgroup->excl != excl)
 318                 return -EINVAL;
 319         return 0;
 320 }
 321 #endif
 322
 323 /*
 324  * The full config is read in one go, only called from open_ctree()
 325  * It doesn't use any locking, as at this point we're still single-threaded
 326  */
 327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 328 {
 329         struct btrfs_key key;
 330         struct btrfs_key found_key;
 331         struct btrfs_root *quota_root = fs_info->quota_root;
 332         struct btrfs_path *path = NULL;
 333         struct extent_buffer *l;
 334         int slot;
 335         int ret = 0;
 336         u64 flags = 0;
 337         u64 rescan_progress = 0;
 338
 339         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 340                 return 0;
 341
 342         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 343         if (!fs_info->qgroup_ulist) {
 344                 ret = -ENOMEM;
 345                 goto out;
 346         }
 347
 348         path = btrfs_alloc_path();
 349         if (!path) {
 350                 ret = -ENOMEM;
 351                 goto out;
 352         }
 353
 354         /* default this to quota off, in case no status key is found */
 355         fs_info->qgroup_flags = 0;
 356
 357         /*
 358          * pass 1: read status, all qgroup infos and limits
 359          */
 360         key.objectid = 0;
 361         key.type = 0;
 362         key.offset = 0;
 363         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 364         if (ret)
 365                 goto out;
 366
 367         while (1) {
 368                 struct btrfs_qgroup *qgroup;
 369
 370                 slot = path->slots[0];
 371                 l = path->nodes[0];
 372                 btrfs_item_key_to_cpu(l, &found_key, slot);
 373
 374                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 375                         struct btrfs_qgroup_status_item *ptr;
 376
 377                         ptr = btrfs_item_ptr(l, slot,
 378                                              struct btrfs_qgroup_status_item);
 379
 380                         if (btrfs_qgroup_status_version(l, ptr) !=
 381                             BTRFS_QGROUP_STATUS_VERSION) {
 382                                 btrfs_err(fs_info,
 383                                  "old qgroup version, quota disabled");
 384                                 goto out;
 385                         }
 386                         if (btrfs_qgroup_status_generation(l, ptr) !=
 387                             fs_info->generation) {
 388                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 389                                 btrfs_err(fs_info,
 390                                         "qgroup generation mismatch, marked as inconsistent");
 391                         }
 392                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 393                                                                           ptr);
 394                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 395                         goto next1;
 396                 }
 397
 398                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 399                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 400                         goto next1;
 401
 402                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 403                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 404                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 405                         btrfs_err(fs_info, "inconsistent qgroup config");
 406                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 407                 }
 408                 if (!qgroup) {
 409                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 410                         if (IS_ERR(qgroup)) {
 411                                 ret = PTR_ERR(qgroup);
 412                                 goto out;
 413                         }
 414                 }
 415                 switch (found_key.type) {
 416                 case BTRFS_QGROUP_INFO_KEY: {
 417                         struct btrfs_qgroup_info_item *ptr;
 418
 419                         ptr = btrfs_item_ptr(l, slot,
 420                                              struct btrfs_qgroup_info_item);
 421                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 422                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 423                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 424                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 425                         /* generation currently unused */
 426                         break;
 427                 }
 428                 case BTRFS_QGROUP_LIMIT_KEY: {
 429                         struct btrfs_qgroup_limit_item *ptr;
 430
 431                         ptr = btrfs_item_ptr(l, slot,
 432                                              struct btrfs_qgroup_limit_item);
 433                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 434                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 435                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 436                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 437                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 438                         break;
 439                 }
 440                 }
 441 next1:
 442                 ret = btrfs_next_item(quota_root, path);
 443                 if (ret < 0)
 444                         goto out;
 445                 if (ret)
 446                         break;
 447         }
 448         btrfs_release_path(path);
 449
 450         /*
 451          * pass 2: read all qgroup relations
 452          */
 453         key.objectid = 0;
 454         key.type = BTRFS_QGROUP_RELATION_KEY;
 455         key.offset = 0;
 456         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 457         if (ret)
 458                 goto out;
 459         while (1) {
 460                 slot = path->slots[0];
 461                 l = path->nodes[0];
 462                 btrfs_item_key_to_cpu(l, &found_key, slot);
 463
 464                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 465                         goto next2;
 466
 467                 if (found_key.objectid > found_key.offset) {
 468                         /* parent <- member, not needed to build config */
 469                         /* FIXME should we omit the key completely? */
 470                         goto next2;
 471                 }
 472
 473                 ret = add_relation_rb(fs_info, found_key.objectid,
 474                                       found_key.offset);
 475                 if (ret == -ENOENT) {
 476                         btrfs_warn(fs_info,
 477                                 "orphan qgroup relation 0x%llx->0x%llx",
 478                                 found_key.objectid, found_key.offset);
 479                         ret = 0;        /* ignore the error */
 480                 }
 481                 if (ret)
 482                         goto out;
 483 next2:
 484                 ret = btrfs_next_item(quota_root, path);
 485                 if (ret < 0)
 486                         goto out;
 487                 if (ret)
 488                         break;
 489         }
 490 out:
 491         fs_info->qgroup_flags |= flags;
 492         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
 493                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 494         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 495                  ret >= 0)
 496                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 497         btrfs_free_path(path);
 498
 499         if (ret < 0) {
 500                 ulist_free(fs_info->qgroup_ulist);
 501                 fs_info->qgroup_ulist = NULL;
 502                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 503         }
 504
 505         return ret < 0 ? ret : 0;
 506 }
 507
 508 /*
 509  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 510  * first two are in single-threaded paths.And for the third one, we have set
 511  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 512  * up the in-memory structures without qgroup_lock held.
 513  */
 514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 515 {
 516         struct rb_node *n;
 517         struct btrfs_qgroup *qgroup;
 518
 519         while ((n = rb_first(&fs_info->qgroup_tree))) {
 520                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 521                 rb_erase(n, &fs_info->qgroup_tree);
 522                 __del_qgroup_rb(qgroup);
 523         }
 524         /*
 525          * We call btrfs_free_qgroup_config() when unmounting
 526          * filesystem and disabling quota, so we set qgroup_ulist
 527          * to be null here to avoid double free.
 528          */
 529         ulist_free(fs_info->qgroup_ulist);
 530         fs_info->qgroup_ulist = NULL;
 531 }
 532
 533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 534                                     u64 dst)
 535 {
 536         int ret;
 537         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 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_insert_empty_item(trans, quota_root, path, &key, 0);
 550
 551         btrfs_mark_buffer_dirty(path->nodes[0]);
 552
 553         btrfs_free_path(path);
 554         return ret;
 555 }
 556
 557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 558                                     u64 dst)
 559 {
 560         int ret;
 561         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 562         struct btrfs_path *path;
 563         struct btrfs_key key;
 564
 565         path = btrfs_alloc_path();
 566         if (!path)
 567                 return -ENOMEM;
 568
 569         key.objectid = src;
 570         key.type = BTRFS_QGROUP_RELATION_KEY;
 571         key.offset = dst;
 572
 573         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 574         if (ret < 0)
 575                 goto out;
 576
 577         if (ret > 0) {
 578                 ret = -ENOENT;
 579                 goto out;
 580         }
 581
 582         ret = btrfs_del_item(trans, quota_root, path);
 583 out:
 584         btrfs_free_path(path);
 585         return ret;
 586 }
 587
 588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 589                            struct btrfs_root *quota_root, u64 qgroupid)
 590 {
 591         int ret;
 592         struct btrfs_path *path;
 593         struct btrfs_qgroup_info_item *qgroup_info;
 594         struct btrfs_qgroup_limit_item *qgroup_limit;
 595         struct extent_buffer *leaf;
 596         struct btrfs_key key;
 597
 598         if (btrfs_is_testing(quota_root->fs_info))
 599                 return 0;
 600
 601         path = btrfs_alloc_path();
 602         if (!path)
 603                 return -ENOMEM;
 604
 605         key.objectid = 0;
 606         key.type = BTRFS_QGROUP_INFO_KEY;
 607         key.offset = qgroupid;
 608
 609         /*
 610          * Avoid a transaction abort by catching -EEXIST here. In that
 611          * case, we proceed by re-initializing the existing structure
 612          * on disk.
 613          */
 614
 615         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 616                                       sizeof(*qgroup_info));
 617         if (ret && ret != -EEXIST)
 618                 goto out;
 619
 620         leaf = path->nodes[0];
 621         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 622                                  struct btrfs_qgroup_info_item);
 623         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 624         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 625         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 626         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 627         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 628
 629         btrfs_mark_buffer_dirty(leaf);
 630
 631         btrfs_release_path(path);
 632
 633         key.type = BTRFS_QGROUP_LIMIT_KEY;
 634         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 635                                       sizeof(*qgroup_limit));
 636         if (ret && ret != -EEXIST)
 637                 goto out;
 638
 639         leaf = path->nodes[0];
 640         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 641                                   struct btrfs_qgroup_limit_item);
 642         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 643         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 644         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 645         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 646         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 647
 648         btrfs_mark_buffer_dirty(leaf);
 649
 650         ret = 0;
 651 out:
 652         btrfs_free_path(path);
 653         return ret;
 654 }
 655
 656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
 657 {
 658         int ret;
 659         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 660         struct btrfs_path *path;
 661         struct btrfs_key key;
 662
 663         path = btrfs_alloc_path();
 664         if (!path)
 665                 return -ENOMEM;
 666
 667         key.objectid = 0;
 668         key.type = BTRFS_QGROUP_INFO_KEY;
 669         key.offset = qgroupid;
 670         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 671         if (ret < 0)
 672                 goto out;
 673
 674         if (ret > 0) {
 675                 ret = -ENOENT;
 676                 goto out;
 677         }
 678
 679         ret = btrfs_del_item(trans, quota_root, path);
 680         if (ret)
 681                 goto out;
 682
 683         btrfs_release_path(path);
 684
 685         key.type = BTRFS_QGROUP_LIMIT_KEY;
 686         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 687         if (ret < 0)
 688                 goto out;
 689
 690         if (ret > 0) {
 691                 ret = -ENOENT;
 692                 goto out;
 693         }
 694
 695         ret = btrfs_del_item(trans, quota_root, path);
 696
 697 out:
 698         btrfs_free_path(path);
 699         return ret;
 700 }
 701
 702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 703                                     struct btrfs_qgroup *qgroup)
 704 {
 705         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 706         struct btrfs_path *path;
 707         struct btrfs_key key;
 708         struct extent_buffer *l;
 709         struct btrfs_qgroup_limit_item *qgroup_limit;
 710         int ret;
 711         int slot;
 712
 713         key.objectid = 0;
 714         key.type = BTRFS_QGROUP_LIMIT_KEY;
 715         key.offset = qgroup->qgroupid;
 716
 717         path = btrfs_alloc_path();
 718         if (!path)
 719                 return -ENOMEM;
 720
 721         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 722         if (ret > 0)
 723                 ret = -ENOENT;
 724
 725         if (ret)
 726                 goto out;
 727
 728         l = path->nodes[0];
 729         slot = path->slots[0];
 730         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 731         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 732         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 733         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 734         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 735         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 736
 737         btrfs_mark_buffer_dirty(l);
 738
 739 out:
 740         btrfs_free_path(path);
 741         return ret;
 742 }
 743
 744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 745                                    struct btrfs_qgroup *qgroup)
 746 {
 747         struct btrfs_fs_info *fs_info = trans->fs_info;
 748         struct btrfs_root *quota_root = fs_info->quota_root;
 749         struct btrfs_path *path;
 750         struct btrfs_key key;
 751         struct extent_buffer *l;
 752         struct btrfs_qgroup_info_item *qgroup_info;
 753         int ret;
 754         int slot;
 755
 756         if (btrfs_is_testing(fs_info))
 757                 return 0;
 758
 759         key.objectid = 0;
 760         key.type = BTRFS_QGROUP_INFO_KEY;
 761         key.offset = qgroup->qgroupid;
 762
 763         path = btrfs_alloc_path();
 764         if (!path)
 765                 return -ENOMEM;
 766
 767         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 768         if (ret > 0)
 769                 ret = -ENOENT;
 770
 771         if (ret)
 772                 goto out;
 773
 774         l = path->nodes[0];
 775         slot = path->slots[0];
 776         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 777         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 778         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 779         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 780         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 781         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 782
 783         btrfs_mark_buffer_dirty(l);
 784
 785 out:
 786         btrfs_free_path(path);
 787         return ret;
 788 }
 789
 790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
 791 {
 792         struct btrfs_fs_info *fs_info = trans->fs_info;
 793         struct btrfs_root *quota_root = fs_info->quota_root;
 794         struct btrfs_path *path;
 795         struct btrfs_key key;
 796         struct extent_buffer *l;
 797         struct btrfs_qgroup_status_item *ptr;
 798         int ret;
 799         int slot;
 800
 801         key.objectid = 0;
 802         key.type = BTRFS_QGROUP_STATUS_KEY;
 803         key.offset = 0;
 804
 805         path = btrfs_alloc_path();
 806         if (!path)
 807                 return -ENOMEM;
 808
 809         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 810         if (ret > 0)
 811                 ret = -ENOENT;
 812
 813         if (ret)
 814                 goto out;
 815
 816         l = path->nodes[0];
 817         slot = path->slots[0];
 818         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 819         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 820         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 821         btrfs_set_qgroup_status_rescan(l, ptr,
 822                                 fs_info->qgroup_rescan_progress.objectid);
 823
 824         btrfs_mark_buffer_dirty(l);
 825
 826 out:
 827         btrfs_free_path(path);
 828         return ret;
 829 }
 830
 831 /*
 832  * called with qgroup_lock held
 833  */
 834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 835                                   struct btrfs_root *root)
 836 {
 837         struct btrfs_path *path;
 838         struct btrfs_key key;
 839         struct extent_buffer *leaf = NULL;
 840         int ret;
 841         int nr = 0;
 842
 843         path = btrfs_alloc_path();
 844         if (!path)
 845                 return -ENOMEM;
 846
 847         path->leave_spinning = 1;
 848
 849         key.objectid = 0;
 850         key.offset = 0;
 851         key.type = 0;
 852
 853         while (1) {
 854                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 855                 if (ret < 0)
 856                         goto out;
 857                 leaf = path->nodes[0];
 858                 nr = btrfs_header_nritems(leaf);
 859                 if (!nr)
 860                         break;
 861                 /*
 862                  * delete the leaf one by one
 863                  * since the whole tree is going
 864                  * to be deleted.
 865                  */
 866                 path->slots[0] = 0;
 867                 ret = btrfs_del_items(trans, root, path, 0, nr);
 868                 if (ret)
 869                         goto out;
 870
 871                 btrfs_release_path(path);
 872         }
 873         ret = 0;
 874 out:
 875         btrfs_free_path(path);
 876         return ret;
 877 }
 878
 879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 880 {
 881         struct btrfs_root *quota_root;
 882         struct btrfs_root *tree_root = fs_info->tree_root;
 883         struct btrfs_path *path = NULL;
 884         struct btrfs_qgroup_status_item *ptr;
 885         struct extent_buffer *leaf;
 886         struct btrfs_key key;
 887         struct btrfs_key found_key;
 888         struct btrfs_qgroup *qgroup = NULL;
 889         struct btrfs_trans_handle *trans = NULL;
 890         int ret = 0;
 891         int slot;
 892
 893         mutex_lock(&fs_info->qgroup_ioctl_lock);
 894         if (fs_info->quota_root)
 895                 goto out;
 896
 897         /*
 898          * 1 for quota root item
 899          * 1 for BTRFS_QGROUP_STATUS item
 900          *
 901          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
 902          * per subvolume. However those are not currently reserved since it
 903          * would be a lot of overkill.
 904          */
 905         trans = btrfs_start_transaction(tree_root, 2);
 906         if (IS_ERR(trans)) {
 907                 ret = PTR_ERR(trans);
 908                 trans = NULL;
 909                 goto out;
 910         }
 911
 912         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 913         if (!fs_info->qgroup_ulist) {
 914                 ret = -ENOMEM;
 915                 btrfs_abort_transaction(trans, ret);
 916                 goto out;
 917         }
 918
 919         /*
 920          * initially create the quota tree
 921          */
 922         quota_root = btrfs_create_tree(trans, fs_info,
 923                                        BTRFS_QUOTA_TREE_OBJECTID);
 924         if (IS_ERR(quota_root)) {
 925                 ret =  PTR_ERR(quota_root);
 926                 btrfs_abort_transaction(trans, ret);
 927                 goto out;
 928         }
 929
 930         path = btrfs_alloc_path();
 931         if (!path) {
 932                 ret = -ENOMEM;
 933                 btrfs_abort_transaction(trans, ret);
 934                 goto out_free_root;
 935         }
 936
 937         key.objectid = 0;
 938         key.type = BTRFS_QGROUP_STATUS_KEY;
 939         key.offset = 0;
 940
 941         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 942                                       sizeof(*ptr));
 943         if (ret) {
 944                 btrfs_abort_transaction(trans, ret);
 945                 goto out_free_path;
 946         }
 947
 948         leaf = path->nodes[0];
 949         ptr = btrfs_item_ptr(leaf, path->slots[0],
 950                                  struct btrfs_qgroup_status_item);
 951         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 952         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 953         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 954                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 955         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 956         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 957
 958         btrfs_mark_buffer_dirty(leaf);
 959
 960         key.objectid = 0;
 961         key.type = BTRFS_ROOT_REF_KEY;
 962         key.offset = 0;
 963
 964         btrfs_release_path(path);
 965         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 966         if (ret > 0)
 967                 goto out_add_root;
 968         if (ret < 0) {
 969                 btrfs_abort_transaction(trans, ret);
 970                 goto out_free_path;
 971         }
 972
 973         while (1) {
 974                 slot = path->slots[0];
 975                 leaf = path->nodes[0];
 976                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 977
 978                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 979                         ret = add_qgroup_item(trans, quota_root,
 980                                               found_key.offset);
 981                         if (ret) {
 982                                 btrfs_abort_transaction(trans, ret);
 983                                 goto out_free_path;
 984                         }
 985
 986                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 987                         if (IS_ERR(qgroup)) {
 988                                 ret = PTR_ERR(qgroup);
 989                                 btrfs_abort_transaction(trans, ret);
 990                                 goto out_free_path;
 991                         }
 992                 }
 993                 ret = btrfs_next_item(tree_root, path);
 994                 if (ret < 0) {
 995                         btrfs_abort_transaction(trans, ret);
 996                         goto out_free_path;
 997                 }
 998                 if (ret)
 999                         break;
1000         }
1001
1002 out_add_root:
1003         btrfs_release_path(path);
1004         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1005         if (ret) {
1006                 btrfs_abort_transaction(trans, ret);
1007                 goto out_free_path;
1008         }
1009
1010         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011         if (IS_ERR(qgroup)) {
1012                 ret = PTR_ERR(qgroup);
1013                 btrfs_abort_transaction(trans, ret);
1014                 goto out_free_path;
1015         }
1016
1017         ret = btrfs_commit_transaction(trans);
1018         trans = NULL;
1019         if (ret)
1020                 goto out_free_path;
1021
1022         /*
1023          * Set quota enabled flag after committing the transaction, to avoid
1024          * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1025          * creation.
1026          */
1027         spin_lock(&fs_info->qgroup_lock);
1028         fs_info->quota_root = quota_root;
1029         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1030         spin_unlock(&fs_info->qgroup_lock);
1031
1032         ret = qgroup_rescan_init(fs_info, 0, 1);
1033         if (!ret) {
1034                 qgroup_rescan_zero_tracking(fs_info);
1035                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1036                                  &fs_info->qgroup_rescan_work);
1037         }
1038
1039 out_free_path:
1040         btrfs_free_path(path);
1041 out_free_root:
1042         if (ret) {
1043                 free_extent_buffer(quota_root->node);
1044                 free_extent_buffer(quota_root->commit_root);
1045                 kfree(quota_root);
1046         }
1047 out:
1048         if (ret) {
1049                 ulist_free(fs_info->qgroup_ulist);
1050                 fs_info->qgroup_ulist = NULL;
1051                 if (trans)
1052                         btrfs_end_transaction(trans);
1053         }
1054         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1055         return ret;
1056 }
1057
1058 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1059 {
1060         struct btrfs_root *quota_root;
1061         struct btrfs_trans_handle *trans = NULL;
1062         int ret = 0;
1063
1064         mutex_lock(&fs_info->qgroup_ioctl_lock);
1065         if (!fs_info->quota_root)
1066                 goto out;
1067
1068         /*
1069          * 1 For the root item
1070          *
1071          * We should also reserve enough items for the quota tree deletion in
1072          * btrfs_clean_quota_tree but this is not done.
1073          */
1074         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1075         if (IS_ERR(trans)) {
1076                 ret = PTR_ERR(trans);
1077                 goto out;
1078         }
1079
1080         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1081         btrfs_qgroup_wait_for_completion(fs_info, false);
1082         spin_lock(&fs_info->qgroup_lock);
1083         quota_root = fs_info->quota_root;
1084         fs_info->quota_root = NULL;
1085         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1086         spin_unlock(&fs_info->qgroup_lock);
1087
1088         btrfs_free_qgroup_config(fs_info);
1089
1090         ret = btrfs_clean_quota_tree(trans, quota_root);
1091         if (ret) {
1092                 btrfs_abort_transaction(trans, ret);
1093                 goto end_trans;
1094         }
1095
1096         ret = btrfs_del_root(trans, &quota_root->root_key);
1097         if (ret) {
1098                 btrfs_abort_transaction(trans, ret);
1099                 goto end_trans;
1100         }
1101
1102         list_del(&quota_root->dirty_list);
1103
1104         btrfs_tree_lock(quota_root->node);
1105         clean_tree_block(fs_info, quota_root->node);
1106         btrfs_tree_unlock(quota_root->node);
1107         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1108
1109         free_extent_buffer(quota_root->node);
1110         free_extent_buffer(quota_root->commit_root);
1111         kfree(quota_root);
1112
1113 end_trans:
1114         ret = btrfs_end_transaction(trans);
1115 out:
1116         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1117         return ret;
1118 }
1119
1120 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1121                          struct btrfs_qgroup *qgroup)
1122 {
1123         if (list_empty(&qgroup->dirty))
1124                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1125 }
1126
1127 /*
1128  * The easy accounting, we're updating qgroup relationship whose child qgroup
1129  * only has exclusive extents.
1130  *
1131  * In this case, all exclusive extents will also be exclusive for parent, so
1132  * excl/rfer just get added/removed.
1133  *
1134  * So is qgroup reservation space, which should also be added/removed to
1135  * parent.
1136  * Or when child tries to release reservation space, parent will underflow its
1137  * reservation (for relationship adding case).
1138  *
1139  * Caller should hold fs_info->qgroup_lock.
1140  */
1141 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1142                                     struct ulist *tmp, u64 ref_root,
1143                                     struct btrfs_qgroup *src, int sign)
1144 {
1145         struct btrfs_qgroup *qgroup;
1146         struct btrfs_qgroup_list *glist;
1147         struct ulist_node *unode;
1148         struct ulist_iterator uiter;
1149         u64 num_bytes = src->excl;
1150         int ret = 0;
1151
1152         qgroup = find_qgroup_rb(fs_info, ref_root);
1153         if (!qgroup)
1154                 goto out;
1155
1156         qgroup->rfer += sign * num_bytes;
1157         qgroup->rfer_cmpr += sign * num_bytes;
1158
1159         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1160         qgroup->excl += sign * num_bytes;
1161         qgroup->excl_cmpr += sign * num_bytes;
1162
1163         if (sign > 0)
1164                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1165         else
1166                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1167
1168         qgroup_dirty(fs_info, qgroup);
1169
1170         /* Get all of the parent groups that contain this qgroup */
1171         list_for_each_entry(glist, &qgroup->groups, next_group) {
1172                 ret = ulist_add(tmp, glist->group->qgroupid,
1173                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1174                 if (ret < 0)
1175                         goto out;
1176         }
1177
1178         /* Iterate all of the parents and adjust their reference counts */
1179         ULIST_ITER_INIT(&uiter);
1180         while ((unode = ulist_next(tmp, &uiter))) {
1181                 qgroup = unode_aux_to_qgroup(unode);
1182                 qgroup->rfer += sign * num_bytes;
1183                 qgroup->rfer_cmpr += sign * num_bytes;
1184                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1185                 qgroup->excl += sign * num_bytes;
1186                 if (sign > 0)
1187                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1188                 else
1189                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1190                 qgroup->excl_cmpr += sign * num_bytes;
1191                 qgroup_dirty(fs_info, qgroup);
1192
1193                 /* Add any parents of the parents */
1194                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1195                         ret = ulist_add(tmp, glist->group->qgroupid,
1196                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1197                         if (ret < 0)
1198                                 goto out;
1199                 }
1200         }
1201         ret = 0;
1202 out:
1203         return ret;
1204 }
1205
1206
1207 /*
1208  * Quick path for updating qgroup with only excl refs.
1209  *
1210  * In that case, just update all parent will be enough.
1211  * Or we needs to do a full rescan.
1212  * Caller should also hold fs_info->qgroup_lock.
1213  *
1214  * Return 0 for quick update, return >0 for need to full rescan
1215  * and mark INCONSISTENT flag.
1216  * Return < 0 for other error.
1217  */
1218 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1219                                    struct ulist *tmp, u64 src, u64 dst,
1220                                    int sign)
1221 {
1222         struct btrfs_qgroup *qgroup;
1223         int ret = 1;
1224         int err = 0;
1225
1226         qgroup = find_qgroup_rb(fs_info, src);
1227         if (!qgroup)
1228                 goto out;
1229         if (qgroup->excl == qgroup->rfer) {
1230                 ret = 0;
1231                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1232                                                qgroup, sign);
1233                 if (err < 0) {
1234                         ret = err;
1235                         goto out;
1236                 }
1237         }
1238 out:
1239         if (ret)
1240                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1241         return ret;
1242 }
1243
1244 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1245                               u64 dst)
1246 {
1247         struct btrfs_fs_info *fs_info = trans->fs_info;
1248         struct btrfs_root *quota_root;
1249         struct btrfs_qgroup *parent;
1250         struct btrfs_qgroup *member;
1251         struct btrfs_qgroup_list *list;
1252         struct ulist *tmp;
1253         int ret = 0;
1254
1255         /* Check the level of src and dst first */
1256         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1257                 return -EINVAL;
1258
1259         tmp = ulist_alloc(GFP_KERNEL);
1260         if (!tmp)
1261                 return -ENOMEM;
1262
1263         mutex_lock(&fs_info->qgroup_ioctl_lock);
1264         quota_root = fs_info->quota_root;
1265         if (!quota_root) {
1266                 ret = -EINVAL;
1267                 goto out;
1268         }
1269         member = find_qgroup_rb(fs_info, src);
1270         parent = find_qgroup_rb(fs_info, dst);
1271         if (!member || !parent) {
1272                 ret = -EINVAL;
1273                 goto out;
1274         }
1275
1276         /* check if such qgroup relation exist firstly */
1277         list_for_each_entry(list, &member->groups, next_group) {
1278                 if (list->group == parent) {
1279                         ret = -EEXIST;
1280                         goto out;
1281                 }
1282         }
1283
1284         ret = add_qgroup_relation_item(trans, src, dst);
1285         if (ret)
1286                 goto out;
1287
1288         ret = add_qgroup_relation_item(trans, dst, src);
1289         if (ret) {
1290                 del_qgroup_relation_item(trans, src, dst);
1291                 goto out;
1292         }
1293
1294         spin_lock(&fs_info->qgroup_lock);
1295         ret = add_relation_rb(fs_info, src, dst);
1296         if (ret < 0) {
1297                 spin_unlock(&fs_info->qgroup_lock);
1298                 goto out;
1299         }
1300         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1301         spin_unlock(&fs_info->qgroup_lock);
1302 out:
1303         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1304         ulist_free(tmp);
1305         return ret;
1306 }
1307
1308 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1309                                  u64 dst)
1310 {
1311         struct btrfs_fs_info *fs_info = trans->fs_info;
1312         struct btrfs_root *quota_root;
1313         struct btrfs_qgroup *parent;
1314         struct btrfs_qgroup *member;
1315         struct btrfs_qgroup_list *list;
1316         struct ulist *tmp;
1317         int ret = 0;
1318         int err;
1319
1320         tmp = ulist_alloc(GFP_KERNEL);
1321         if (!tmp)
1322                 return -ENOMEM;
1323
1324         quota_root = fs_info->quota_root;
1325         if (!quota_root) {
1326                 ret = -EINVAL;
1327                 goto out;
1328         }
1329
1330         member = find_qgroup_rb(fs_info, src);
1331         parent = find_qgroup_rb(fs_info, dst);
1332         if (!member || !parent) {
1333                 ret = -EINVAL;
1334                 goto out;
1335         }
1336
1337         /* check if such qgroup relation exist firstly */
1338         list_for_each_entry(list, &member->groups, next_group) {
1339                 if (list->group == parent)
1340                         goto exist;
1341         }
1342         ret = -ENOENT;
1343         goto out;
1344 exist:
1345         ret = del_qgroup_relation_item(trans, src, dst);
1346         err = del_qgroup_relation_item(trans, dst, src);
1347         if (err && !ret)
1348                 ret = err;
1349
1350         spin_lock(&fs_info->qgroup_lock);
1351         del_relation_rb(fs_info, src, dst);
1352         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1353         spin_unlock(&fs_info->qgroup_lock);
1354 out:
1355         ulist_free(tmp);
1356         return ret;
1357 }
1358
1359 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1360                               u64 dst)
1361 {
1362         struct btrfs_fs_info *fs_info = trans->fs_info;
1363         int ret = 0;
1364
1365         mutex_lock(&fs_info->qgroup_ioctl_lock);
1366         ret = __del_qgroup_relation(trans, src, dst);
1367         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1368
1369         return ret;
1370 }
1371
1372 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1373 {
1374         struct btrfs_fs_info *fs_info = trans->fs_info;
1375         struct btrfs_root *quota_root;
1376         struct btrfs_qgroup *qgroup;
1377         int ret = 0;
1378
1379         mutex_lock(&fs_info->qgroup_ioctl_lock);
1380         quota_root = fs_info->quota_root;
1381         if (!quota_root) {
1382                 ret = -EINVAL;
1383                 goto out;
1384         }
1385         qgroup = find_qgroup_rb(fs_info, qgroupid);
1386         if (qgroup) {
1387                 ret = -EEXIST;
1388                 goto out;
1389         }
1390
1391         ret = add_qgroup_item(trans, quota_root, qgroupid);
1392         if (ret)
1393                 goto out;
1394
1395         spin_lock(&fs_info->qgroup_lock);
1396         qgroup = add_qgroup_rb(fs_info, qgroupid);
1397         spin_unlock(&fs_info->qgroup_lock);
1398
1399         if (IS_ERR(qgroup))
1400                 ret = PTR_ERR(qgroup);
1401 out:
1402         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1403         return ret;
1404 }
1405
1406 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1407 {
1408         struct btrfs_fs_info *fs_info = trans->fs_info;
1409         struct btrfs_root *quota_root;
1410         struct btrfs_qgroup *qgroup;
1411         struct btrfs_qgroup_list *list;
1412         int ret = 0;
1413
1414         mutex_lock(&fs_info->qgroup_ioctl_lock);
1415         quota_root = fs_info->quota_root;
1416         if (!quota_root) {
1417                 ret = -EINVAL;
1418                 goto out;
1419         }
1420
1421         qgroup = find_qgroup_rb(fs_info, qgroupid);
1422         if (!qgroup) {
1423                 ret = -ENOENT;
1424                 goto out;
1425         }
1426
1427         /* Check if there are no children of this qgroup */
1428         if (!list_empty(&qgroup->members)) {
1429                 ret = -EBUSY;
1430                 goto out;
1431         }
1432
1433         ret = del_qgroup_item(trans, qgroupid);
1434         if (ret && ret != -ENOENT)
1435                 goto out;
1436
1437         while (!list_empty(&qgroup->groups)) {
1438                 list = list_first_entry(&qgroup->groups,
1439                                         struct btrfs_qgroup_list, next_group);
1440                 ret = __del_qgroup_relation(trans, qgroupid,
1441                                             list->group->qgroupid);
1442                 if (ret)
1443                         goto out;
1444         }
1445
1446         spin_lock(&fs_info->qgroup_lock);
1447         del_qgroup_rb(fs_info, qgroupid);
1448         spin_unlock(&fs_info->qgroup_lock);
1449 out:
1450         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1451         return ret;
1452 }
1453
1454 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1455                        struct btrfs_qgroup_limit *limit)
1456 {
1457         struct btrfs_fs_info *fs_info = trans->fs_info;
1458         struct btrfs_root *quota_root;
1459         struct btrfs_qgroup *qgroup;
1460         int ret = 0;
1461         /* Sometimes we would want to clear the limit on this qgroup.
1462          * To meet this requirement, we treat the -1 as a special value
1463          * which tell kernel to clear the limit on this qgroup.
1464          */
1465         const u64 CLEAR_VALUE = -1;
1466
1467         mutex_lock(&fs_info->qgroup_ioctl_lock);
1468         quota_root = fs_info->quota_root;
1469         if (!quota_root) {
1470                 ret = -EINVAL;
1471                 goto out;
1472         }
1473
1474         qgroup = find_qgroup_rb(fs_info, qgroupid);
1475         if (!qgroup) {
1476                 ret = -ENOENT;
1477                 goto out;
1478         }
1479
1480         spin_lock(&fs_info->qgroup_lock);
1481         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1482                 if (limit->max_rfer == CLEAR_VALUE) {
1483                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1484                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1485                         qgroup->max_rfer = 0;
1486                 } else {
1487                         qgroup->max_rfer = limit->max_rfer;
1488                 }
1489         }
1490         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1491                 if (limit->max_excl == CLEAR_VALUE) {
1492                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1493                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1494                         qgroup->max_excl = 0;
1495                 } else {
1496                         qgroup->max_excl = limit->max_excl;
1497                 }
1498         }
1499         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1500                 if (limit->rsv_rfer == CLEAR_VALUE) {
1501                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1502                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1503                         qgroup->rsv_rfer = 0;
1504                 } else {
1505                         qgroup->rsv_rfer = limit->rsv_rfer;
1506                 }
1507         }
1508         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1509                 if (limit->rsv_excl == CLEAR_VALUE) {
1510                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1511                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1512                         qgroup->rsv_excl = 0;
1513                 } else {
1514                         qgroup->rsv_excl = limit->rsv_excl;
1515                 }
1516         }
1517         qgroup->lim_flags |= limit->flags;
1518
1519         spin_unlock(&fs_info->qgroup_lock);
1520
1521         ret = update_qgroup_limit_item(trans, qgroup);
1522         if (ret) {
1523                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1524                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1525                        qgroupid);
1526         }
1527
1528 out:
1529         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1530         return ret;
1531 }
1532
1533 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1534                                 struct btrfs_delayed_ref_root *delayed_refs,
1535                                 struct btrfs_qgroup_extent_record *record)
1536 {
1537         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1538         struct rb_node *parent_node = NULL;
1539         struct btrfs_qgroup_extent_record *entry;
1540         u64 bytenr = record->bytenr;
1541
1542         lockdep_assert_held(&delayed_refs->lock);
1543         trace_btrfs_qgroup_trace_extent(fs_info, record);
1544
1545         while (*p) {
1546                 parent_node = *p;
1547                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1548                                  node);
1549                 if (bytenr < entry->bytenr) {
1550                         p = &(*p)->rb_left;
1551                 } else if (bytenr > entry->bytenr) {
1552                         p = &(*p)->rb_right;
1553                 } else {
1554                         if (record->data_rsv && !entry->data_rsv) {
1555                                 entry->data_rsv = record->data_rsv;
1556                                 entry->data_rsv_refroot =
1557                                         record->data_rsv_refroot;
1558                         }
1559                         return 1;
1560                 }
1561         }
1562
1563         rb_link_node(&record->node, parent_node, p);
1564         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1565         return 0;
1566 }
1567
1568 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1569                                    struct btrfs_qgroup_extent_record *qrecord)
1570 {
1571         struct ulist *old_root;
1572         u64 bytenr = qrecord->bytenr;
1573         int ret;
1574
1575         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1576         if (ret < 0) {
1577                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1578                 btrfs_warn(fs_info,
1579 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1580                         ret);
1581                 return 0;
1582         }
1583
1584         /*
1585          * Here we don't need to get the lock of
1586          * trans->transaction->delayed_refs, since inserted qrecord won't
1587          * be deleted, only qrecord->node may be modified (new qrecord insert)
1588          *
1589          * So modifying qrecord->old_roots is safe here
1590          */
1591         qrecord->old_roots = old_root;
1592         return 0;
1593 }
1594
1595 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1596                               u64 num_bytes, gfp_t gfp_flag)
1597 {
1598         struct btrfs_fs_info *fs_info = trans->fs_info;
1599         struct btrfs_qgroup_extent_record *record;
1600         struct btrfs_delayed_ref_root *delayed_refs;
1601         int ret;
1602
1603         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1604             || bytenr == 0 || num_bytes == 0)
1605                 return 0;
1606         record = kzalloc(sizeof(*record), gfp_flag);
1607         if (!record)
1608                 return -ENOMEM;
1609
1610         delayed_refs = &trans->transaction->delayed_refs;
1611         record->bytenr = bytenr;
1612         record->num_bytes = num_bytes;
1613         record->old_roots = NULL;
1614
1615         spin_lock(&delayed_refs->lock);
1616         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1617         spin_unlock(&delayed_refs->lock);
1618         if (ret > 0) {
1619                 kfree(record);
1620                 return 0;
1621         }
1622         return btrfs_qgroup_trace_extent_post(fs_info, record);
1623 }
1624
1625 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1626                                   struct extent_buffer *eb)
1627 {
1628         struct btrfs_fs_info *fs_info = trans->fs_info;
1629         int nr = btrfs_header_nritems(eb);
1630         int i, extent_type, ret;
1631         struct btrfs_key key;
1632         struct btrfs_file_extent_item *fi;
1633         u64 bytenr, num_bytes;
1634
1635         /* We can be called directly from walk_up_proc() */
1636         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1637                 return 0;
1638
1639         for (i = 0; i < nr; i++) {
1640                 btrfs_item_key_to_cpu(eb, &key, i);
1641
1642                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1643                         continue;
1644
1645                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1646                 /* filter out non qgroup-accountable extents  */
1647                 extent_type = btrfs_file_extent_type(eb, fi);
1648
1649                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1650                         continue;
1651
1652                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1653                 if (!bytenr)
1654                         continue;
1655
1656                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1657
1658                 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1659                                                 GFP_NOFS);
1660                 if (ret)
1661                         return ret;
1662         }
1663         cond_resched();
1664         return 0;
1665 }
1666
1667 /*
1668  * Walk up the tree from the bottom, freeing leaves and any interior
1669  * nodes which have had all slots visited. If a node (leaf or
1670  * interior) is freed, the node above it will have it's slot
1671  * incremented. The root node will never be freed.
1672  *
1673  * At the end of this function, we should have a path which has all
1674  * slots incremented to the next position for a search. If we need to
1675  * read a new node it will be NULL and the node above it will have the
1676  * correct slot selected for a later read.
1677  *
1678  * If we increment the root nodes slot counter past the number of
1679  * elements, 1 is returned to signal completion of the search.
1680  */
1681 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1682 {
1683         int level = 0;
1684         int nr, slot;
1685         struct extent_buffer *eb;
1686
1687         if (root_level == 0)
1688                 return 1;
1689
1690         while (level <= root_level) {
1691                 eb = path->nodes[level];
1692                 nr = btrfs_header_nritems(eb);
1693                 path->slots[level]++;
1694                 slot = path->slots[level];
1695                 if (slot >= nr || level == 0) {
1696                         /*
1697                          * Don't free the root -  we will detect this
1698                          * condition after our loop and return a
1699                          * positive value for caller to stop walking the tree.
1700                          */
1701                         if (level != root_level) {
1702                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1703                                 path->locks[level] = 0;
1704
1705                                 free_extent_buffer(eb);
1706                                 path->nodes[level] = NULL;
1707                                 path->slots[level] = 0;
1708                         }
1709                 } else {
1710                         /*
1711                          * We have a valid slot to walk back down
1712                          * from. Stop here so caller can process these
1713                          * new nodes.
1714                          */
1715                         break;
1716                 }
1717
1718                 level++;
1719         }
1720
1721         eb = path->nodes[root_level];
1722         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1723                 return 1;
1724
1725         return 0;
1726 }
1727
1728 /*
1729  * Helper function to trace a subtree tree block swap.
1730  *
1731  * The swap will happen in highest tree block, but there may be a lot of
1732  * tree blocks involved.
1733  *
1734  * For example:
1735  *  OO = Old tree blocks
1736  *  NN = New tree blocks allocated during balance
1737  *
1738  *           File tree (257)                  Reloc tree for 257
1739  * L2              OO                                NN
1740  *               /    \                            /    \
1741  * L1          OO      OO (a)                    OO      NN (a)
1742  *            / \     / \                       / \     / \
1743  * L0       OO   OO OO   OO                   OO   OO NN   NN
1744  *                  (b)  (c)                          (b)  (c)
1745  *
1746  * When calling qgroup_trace_extent_swap(), we will pass:
1747  * @src_eb = OO(a)
1748  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1749  * @dst_level = 0
1750  * @root_level = 1
1751  *
1752  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1753  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1754  *
1755  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1756  *
1757  * 1) Tree search from @src_eb
1758  *    It should acts as a simplified btrfs_search_slot().
1759  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1760  *    (first key).
1761  *
1762  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1763  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1764  *    They should be marked during previous (@dst_level = 1) iteration.
1765  *
1766  * 3) Mark file extents in leaves dirty
1767  *    We don't have good way to pick out new file extents only.
1768  *    So we still follow the old method by scanning all file extents in
1769  *    the leave.
1770  *
1771  * This function can free us from keeping two paths, thus later we only need
1772  * to care about how to iterate all new tree blocks in reloc tree.
1773  */
1774 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1775                                     struct extent_buffer *src_eb,
1776                                     struct btrfs_path *dst_path,
1777                                     int dst_level, int root_level,
1778                                     bool trace_leaf)
1779 {
1780         struct btrfs_key key;
1781         struct btrfs_path *src_path;
1782         struct btrfs_fs_info *fs_info = trans->fs_info;
1783         u32 nodesize = fs_info->nodesize;
1784         int cur_level = root_level;
1785         int ret;
1786
1787         BUG_ON(dst_level > root_level);
1788         /* Level mismatch */
1789         if (btrfs_header_level(src_eb) != root_level)
1790                 return -EINVAL;
1791
1792         src_path = btrfs_alloc_path();
1793         if (!src_path) {
1794                 ret = -ENOMEM;
1795                 goto out;
1796         }
1797
1798         if (dst_level)
1799                 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1800         else
1801                 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1802
1803         /* For src_path */
1804         extent_buffer_get(src_eb);
1805         src_path->nodes[root_level] = src_eb;
1806         src_path->slots[root_level] = dst_path->slots[root_level];
1807         src_path->locks[root_level] = 0;
1808
1809         /* A simplified version of btrfs_search_slot() */
1810         while (cur_level >= dst_level) {
1811                 struct btrfs_key src_key;
1812                 struct btrfs_key dst_key;
1813
1814                 if (src_path->nodes[cur_level] == NULL) {
1815                         struct btrfs_key first_key;
1816                         struct extent_buffer *eb;
1817                         int parent_slot;
1818                         u64 child_gen;
1819                         u64 child_bytenr;
1820
1821                         eb = src_path->nodes[cur_level + 1];
1822                         parent_slot = src_path->slots[cur_level + 1];
1823                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1824                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1825                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1826
1827                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1828                                              cur_level, &first_key);
1829                         if (IS_ERR(eb)) {
1830                                 ret = PTR_ERR(eb);
1831                                 goto out;
1832                         } else if (!extent_buffer_uptodate(eb)) {
1833                                 free_extent_buffer(eb);
1834                                 ret = -EIO;
1835                                 goto out;
1836                         }
1837
1838                         src_path->nodes[cur_level] = eb;
1839
1840                         btrfs_tree_read_lock(eb);
1841                         btrfs_set_lock_blocking_read(eb);
1842                         src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1843                 }
1844
1845                 src_path->slots[cur_level] = dst_path->slots[cur_level];
1846                 if (cur_level) {
1847                         btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1848                                         &dst_key, dst_path->slots[cur_level]);
1849                         btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1850                                         &src_key, src_path->slots[cur_level]);
1851                 } else {
1852                         btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1853                                         &dst_key, dst_path->slots[cur_level]);
1854                         btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1855                                         &src_key, src_path->slots[cur_level]);
1856                 }
1857                 /* Content mismatch, something went wrong */
1858                 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1859                         ret = -ENOENT;
1860                         goto out;
1861                 }
1862                 cur_level--;
1863         }
1864
1865         /*
1866          * Now both @dst_path and @src_path have been populated, record the tree
1867          * blocks for qgroup accounting.
1868          */
1869         ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1870                         nodesize, GFP_NOFS);
1871         if (ret < 0)
1872                 goto out;
1873         ret = btrfs_qgroup_trace_extent(trans,
1874                         dst_path->nodes[dst_level]->start,
1875                         nodesize, GFP_NOFS);
1876         if (ret < 0)
1877                 goto out;
1878
1879         /* Record leaf file extents */
1880         if (dst_level == 0 && trace_leaf) {
1881                 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1882                 if (ret < 0)
1883                         goto out;
1884                 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1885         }
1886 out:
1887         btrfs_free_path(src_path);
1888         return ret;
1889 }
1890
1891 /*
1892  * Helper function to do recursive generation-aware depth-first search, to
1893  * locate all new tree blocks in a subtree of reloc tree.
1894  *
1895  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1896  *         reloc tree
1897  * L2         NN (a)
1898  *          /    \
1899  * L1    OO        NN (b)
1900  *      /  \      /  \
1901  * L0  OO  OO    OO  NN
1902  *               (c) (d)
1903  * If we pass:
1904  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1905  * @cur_level = 1
1906  * @root_level = 1
1907  *
1908  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1909  * above tree blocks along with their counter parts in file tree.
1910  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1911  * won't affect OO(c).
1912  */
1913 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1914                                            struct extent_buffer *src_eb,
1915                                            struct btrfs_path *dst_path,
1916                                            int cur_level, int root_level,
1917                                            u64 last_snapshot, bool trace_leaf)
1918 {
1919         struct btrfs_fs_info *fs_info = trans->fs_info;
1920         struct extent_buffer *eb;
1921         bool need_cleanup = false;
1922         int ret = 0;
1923         int i;
1924
1925         /* Level sanity check */
1926         if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL ||
1927             root_level < 0 || root_level >= BTRFS_MAX_LEVEL ||
1928             root_level < cur_level) {
1929                 btrfs_err_rl(fs_info,
1930                         "%s: bad levels, cur_level=%d root_level=%d",
1931                         __func__, cur_level, root_level);
1932                 return -EUCLEAN;
1933         }
1934
1935         /* Read the tree block if needed */
1936         if (dst_path->nodes[cur_level] == NULL) {
1937                 struct btrfs_key first_key;
1938                 int parent_slot;
1939                 u64 child_gen;
1940                 u64 child_bytenr;
1941
1942                 /*
1943                  * dst_path->nodes[root_level] must be initialized before
1944                  * calling this function.
1945                  */
1946                 if (cur_level == root_level) {
1947                         btrfs_err_rl(fs_info,
1948         "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1949                                 __func__, root_level, root_level, cur_level);
1950                         return -EUCLEAN;
1951                 }
1952
1953                 /*
1954                  * We need to get child blockptr/gen from parent before we can
1955                  * read it.
1956                   */
1957                 eb = dst_path->nodes[cur_level + 1];
1958                 parent_slot = dst_path->slots[cur_level + 1];
1959                 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1960                 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1961                 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1962
1963                 /* This node is old, no need to trace */
1964                 if (child_gen < last_snapshot)
1965                         goto out;
1966
1967                 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1968                                      cur_level, &first_key);
1969                 if (IS_ERR(eb)) {
1970                         ret = PTR_ERR(eb);
1971                         goto out;
1972                 } else if (!extent_buffer_uptodate(eb)) {
1973                         free_extent_buffer(eb);
1974                         ret = -EIO;
1975                         goto out;
1976                 }
1977
1978                 dst_path->nodes[cur_level] = eb;
1979                 dst_path->slots[cur_level] = 0;
1980
1981                 btrfs_tree_read_lock(eb);
1982                 btrfs_set_lock_blocking_read(eb);
1983                 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1984                 need_cleanup = true;
1985         }
1986
1987         /* Now record this tree block and its counter part for qgroups */
1988         ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1989                                        root_level, trace_leaf);
1990         if (ret < 0)
1991                 goto cleanup;
1992
1993         eb = dst_path->nodes[cur_level];
1994
1995         if (cur_level > 0) {
1996                 /* Iterate all child tree blocks */
1997                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1998                         /* Skip old tree blocks as they won't be swapped */
1999                         if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2000                                 continue;
2001                         dst_path->slots[cur_level] = i;
2002
2003                         /* Recursive call (at most 7 times) */
2004                         ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2005                                         dst_path, cur_level - 1, root_level,
2006                                         last_snapshot, trace_leaf);
2007                         if (ret < 0)
2008                                 goto cleanup;
2009                 }
2010         }
2011
2012 cleanup:
2013         if (need_cleanup) {
2014                 /* Clean up */
2015                 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2016                                      dst_path->locks[cur_level]);
2017                 free_extent_buffer(dst_path->nodes[cur_level]);
2018                 dst_path->nodes[cur_level] = NULL;
2019                 dst_path->slots[cur_level] = 0;
2020                 dst_path->locks[cur_level] = 0;
2021         }
2022 out:
2023         return ret;
2024 }
2025
2026 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2027                                 struct extent_buffer *src_eb,
2028                                 struct extent_buffer *dst_eb,
2029                                 u64 last_snapshot, bool trace_leaf)
2030 {
2031         struct btrfs_fs_info *fs_info = trans->fs_info;
2032         struct btrfs_path *dst_path = NULL;
2033         int level;
2034         int ret;
2035
2036         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2037                 return 0;
2038
2039         /* Wrong parameter order */
2040         if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2041                 btrfs_err_rl(fs_info,
2042                 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2043                              btrfs_header_generation(src_eb),
2044                              btrfs_header_generation(dst_eb));
2045                 return -EUCLEAN;
2046         }
2047
2048         if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2049                 ret = -EIO;
2050                 goto out;
2051         }
2052
2053         level = btrfs_header_level(dst_eb);
2054         dst_path = btrfs_alloc_path();
2055         if (!dst_path) {
2056                 ret = -ENOMEM;
2057                 goto out;
2058         }
2059         /* For dst_path */
2060         extent_buffer_get(dst_eb);
2061         dst_path->nodes[level] = dst_eb;
2062         dst_path->slots[level] = 0;
2063         dst_path->locks[level] = 0;
2064
2065         /* Do the generation aware breadth-first search */
2066         ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2067                                               level, last_snapshot, trace_leaf);
2068         if (ret < 0)
2069                 goto out;
2070         ret = 0;
2071
2072 out:
2073         btrfs_free_path(dst_path);
2074         if (ret < 0)
2075                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2076         return ret;
2077 }
2078
2079 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2080                                struct extent_buffer *root_eb,
2081                                u64 root_gen, int root_level)
2082 {
2083         struct btrfs_fs_info *fs_info = trans->fs_info;
2084         int ret = 0;
2085         int level;
2086         struct extent_buffer *eb = root_eb;
2087         struct btrfs_path *path = NULL;
2088
2089         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2090         BUG_ON(root_eb == NULL);
2091
2092         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2093                 return 0;
2094
2095         if (!extent_buffer_uptodate(root_eb)) {
2096                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2097                 if (ret)
2098                         goto out;
2099         }
2100
2101         if (root_level == 0) {
2102                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2103                 goto out;
2104         }
2105
2106         path = btrfs_alloc_path();
2107         if (!path)
2108                 return -ENOMEM;
2109
2110         /*
2111          * Walk down the tree.  Missing extent blocks are filled in as
2112          * we go. Metadata is accounted every time we read a new
2113          * extent block.
2114          *
2115          * When we reach a leaf, we account for file extent items in it,
2116          * walk back up the tree (adjusting slot pointers as we go)
2117          * and restart the search process.
2118          */
2119         extent_buffer_get(root_eb); /* For path */
2120         path->nodes[root_level] = root_eb;
2121         path->slots[root_level] = 0;
2122         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2123 walk_down:
2124         level = root_level;
2125         while (level >= 0) {
2126                 if (path->nodes[level] == NULL) {
2127                         struct btrfs_key first_key;
2128                         int parent_slot;
2129                         u64 child_gen;
2130                         u64 child_bytenr;
2131
2132                         /*
2133                          * We need to get child blockptr/gen from parent before
2134                          * we can read it.
2135                           */
2136                         eb = path->nodes[level + 1];
2137                         parent_slot = path->slots[level + 1];
2138                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2139                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2140                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2141
2142                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
2143                                              level, &first_key);
2144                         if (IS_ERR(eb)) {
2145                                 ret = PTR_ERR(eb);
2146                                 goto out;
2147                         } else if (!extent_buffer_uptodate(eb)) {
2148                                 free_extent_buffer(eb);
2149                                 ret = -EIO;
2150                                 goto out;
2151                         }
2152
2153                         path->nodes[level] = eb;
2154                         path->slots[level] = 0;
2155
2156                         btrfs_tree_read_lock(eb);
2157                         btrfs_set_lock_blocking_read(eb);
2158                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2159
2160                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2161                                                         fs_info->nodesize,
2162                                                         GFP_NOFS);
2163                         if (ret)
2164                                 goto out;
2165                 }
2166
2167                 if (level == 0) {
2168                         ret = btrfs_qgroup_trace_leaf_items(trans,
2169                                                             path->nodes[level]);
2170                         if (ret)
2171                                 goto out;
2172
2173                         /* Nonzero return here means we completed our search */
2174                         ret = adjust_slots_upwards(path, root_level);
2175                         if (ret)
2176                                 break;
2177
2178                         /* Restart search with new slots */
2179                         goto walk_down;
2180                 }
2181
2182                 level--;
2183         }
2184
2185         ret = 0;
2186 out:
2187         btrfs_free_path(path);
2188
2189         return ret;
2190 }
2191
2192 #define UPDATE_NEW      0
2193 #define UPDATE_OLD      1
2194 /*
2195  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2196  */
2197 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2198                                 struct ulist *roots, struct ulist *tmp,
2199                                 struct ulist *qgroups, u64 seq, int update_old)
2200 {
2201         struct ulist_node *unode;
2202         struct ulist_iterator uiter;
2203         struct ulist_node *tmp_unode;
2204         struct ulist_iterator tmp_uiter;
2205         struct btrfs_qgroup *qg;
2206         int ret = 0;
2207
2208         if (!roots)
2209                 return 0;
2210         ULIST_ITER_INIT(&uiter);
2211         while ((unode = ulist_next(roots, &uiter))) {
2212                 qg = find_qgroup_rb(fs_info, unode->val);
2213                 if (!qg)
2214                         continue;
2215
2216                 ulist_reinit(tmp);
2217                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2218                                 GFP_ATOMIC);
2219                 if (ret < 0)
2220                         return ret;
2221                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2222                 if (ret < 0)
2223                         return ret;
2224                 ULIST_ITER_INIT(&tmp_uiter);
2225                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2226                         struct btrfs_qgroup_list *glist;
2227
2228                         qg = unode_aux_to_qgroup(tmp_unode);
2229                         if (update_old)
2230                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2231                         else
2232                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2233                         list_for_each_entry(glist, &qg->groups, next_group) {
2234                                 ret = ulist_add(qgroups, glist->group->qgroupid,
2235                                                 qgroup_to_aux(glist->group),
2236                                                 GFP_ATOMIC);
2237                                 if (ret < 0)
2238                                         return ret;
2239                                 ret = ulist_add(tmp, glist->group->qgroupid,
2240                                                 qgroup_to_aux(glist->group),
2241                                                 GFP_ATOMIC);
2242                                 if (ret < 0)
2243                                         return ret;
2244                         }
2245                 }
2246         }
2247         return 0;
2248 }
2249
2250 /*
2251  * Update qgroup rfer/excl counters.
2252  * Rfer update is easy, codes can explain themselves.
2253  *
2254  * Excl update is tricky, the update is split into 2 part.
2255  * Part 1: Possible exclusive <-> sharing detect:
2256  *      |       A       |       !A      |
2257  *  -------------------------------------
2258  *  B   |       *       |       -       |
2259  *  -------------------------------------
2260  *  !B  |       +       |       **      |
2261  *  -------------------------------------
2262  *
2263  * Conditions:
2264  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
2265  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
2266  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
2267  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
2268  *
2269  * Results:
2270  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
2271  * *: Definitely not changed.           **: Possible unchanged.
2272  *
2273  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2274  *
2275  * To make the logic clear, we first use condition A and B to split
2276  * combination into 4 results.
2277  *
2278  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2279  * only on variant maybe 0.
2280  *
2281  * Lastly, check result **, since there are 2 variants maybe 0, split them
2282  * again(2x2).
2283  * But this time we don't need to consider other things, the codes and logic
2284  * is easy to understand now.
2285  */
2286 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2287                                   struct ulist *qgroups,
2288                                   u64 nr_old_roots,
2289                                   u64 nr_new_roots,
2290                                   u64 num_bytes, u64 seq)
2291 {
2292         struct ulist_node *unode;
2293         struct ulist_iterator uiter;
2294         struct btrfs_qgroup *qg;
2295         u64 cur_new_count, cur_old_count;
2296
2297         ULIST_ITER_INIT(&uiter);
2298         while ((unode = ulist_next(qgroups, &uiter))) {
2299                 bool dirty = false;
2300
2301                 qg = unode_aux_to_qgroup(unode);
2302                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2303                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2304
2305                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2306                                              cur_new_count);
2307
2308                 /* Rfer update part */
2309                 if (cur_old_count == 0 && cur_new_count > 0) {
2310                         qg->rfer += num_bytes;
2311                         qg->rfer_cmpr += num_bytes;
2312                         dirty = true;
2313                 }
2314                 if (cur_old_count > 0 && cur_new_count == 0) {
2315                         qg->rfer -= num_bytes;
2316                         qg->rfer_cmpr -= num_bytes;
2317                         dirty = true;
2318                 }
2319
2320                 /* Excl update part */
2321                 /* Exclusive/none -> shared case */
2322                 if (cur_old_count == nr_old_roots &&
2323                     cur_new_count < nr_new_roots) {
2324                         /* Exclusive -> shared */
2325                         if (cur_old_count != 0) {
2326                                 qg->excl -= num_bytes;
2327                                 qg->excl_cmpr -= num_bytes;
2328                                 dirty = true;
2329                         }
2330                 }
2331
2332                 /* Shared -> exclusive/none case */
2333                 if (cur_old_count < nr_old_roots &&
2334                     cur_new_count == nr_new_roots) {
2335                         /* Shared->exclusive */
2336                         if (cur_new_count != 0) {
2337                                 qg->excl += num_bytes;
2338                                 qg->excl_cmpr += num_bytes;
2339                                 dirty = true;
2340                         }
2341                 }
2342
2343                 /* Exclusive/none -> exclusive/none case */
2344                 if (cur_old_count == nr_old_roots &&
2345                     cur_new_count == nr_new_roots) {
2346                         if (cur_old_count == 0) {
2347                                 /* None -> exclusive/none */
2348
2349                                 if (cur_new_count != 0) {
2350                                         /* None -> exclusive */
2351                                         qg->excl += num_bytes;
2352                                         qg->excl_cmpr += num_bytes;
2353                                         dirty = true;
2354                                 }
2355                                 /* None -> none, nothing changed */
2356                         } else {
2357                                 /* Exclusive -> exclusive/none */
2358
2359                                 if (cur_new_count == 0) {
2360                                         /* Exclusive -> none */
2361                                         qg->excl -= num_bytes;
2362                                         qg->excl_cmpr -= num_bytes;
2363                                         dirty = true;
2364                                 }
2365                                 /* Exclusive -> exclusive, nothing changed */
2366                         }
2367                 }
2368
2369                 if (dirty)
2370                         qgroup_dirty(fs_info, qg);
2371         }
2372         return 0;
2373 }
2374
2375 /*
2376  * Check if the @roots potentially is a list of fs tree roots
2377  *
2378  * Return 0 for definitely not a fs/subvol tree roots ulist
2379  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2380  *          one as well)
2381  */
2382 static int maybe_fs_roots(struct ulist *roots)
2383 {
2384         struct ulist_node *unode;
2385         struct ulist_iterator uiter;
2386
2387         /* Empty one, still possible for fs roots */
2388         if (!roots || roots->nnodes == 0)
2389                 return 1;
2390
2391         ULIST_ITER_INIT(&uiter);
2392         unode = ulist_next(roots, &uiter);
2393         if (!unode)
2394                 return 1;
2395
2396         /*
2397          * If it contains fs tree roots, then it must belong to fs/subvol
2398          * trees.
2399          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2400          */
2401         return is_fstree(unode->val);
2402 }
2403
2404 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2405                                 u64 num_bytes, struct ulist *old_roots,
2406                                 struct ulist *new_roots)
2407 {
2408         struct btrfs_fs_info *fs_info = trans->fs_info;
2409         struct ulist *qgroups = NULL;
2410         struct ulist *tmp = NULL;
2411         u64 seq;
2412         u64 nr_new_roots = 0;
2413         u64 nr_old_roots = 0;
2414         int ret = 0;
2415
2416         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2417                 return 0;
2418
2419         if (new_roots) {
2420                 if (!maybe_fs_roots(new_roots))
2421                         goto out_free;
2422                 nr_new_roots = new_roots->nnodes;
2423         }
2424         if (old_roots) {
2425                 if (!maybe_fs_roots(old_roots))
2426                         goto out_free;
2427                 nr_old_roots = old_roots->nnodes;
2428         }
2429
2430         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2431         if (nr_old_roots == 0 && nr_new_roots == 0)
2432                 goto out_free;
2433
2434         BUG_ON(!fs_info->quota_root);
2435
2436         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2437                                         num_bytes, nr_old_roots, nr_new_roots);
2438
2439         qgroups = ulist_alloc(GFP_NOFS);
2440         if (!qgroups) {
2441                 ret = -ENOMEM;
2442                 goto out_free;
2443         }
2444         tmp = ulist_alloc(GFP_NOFS);
2445         if (!tmp) {
2446                 ret = -ENOMEM;
2447                 goto out_free;
2448         }
2449
2450         mutex_lock(&fs_info->qgroup_rescan_lock);
2451         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2452                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2453                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2454                         ret = 0;
2455                         goto out_free;
2456                 }
2457         }
2458         mutex_unlock(&fs_info->qgroup_rescan_lock);
2459
2460         spin_lock(&fs_info->qgroup_lock);
2461         seq = fs_info->qgroup_seq;
2462
2463         /* Update old refcnts using old_roots */
2464         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2465                                    UPDATE_OLD);
2466         if (ret < 0)
2467                 goto out;
2468
2469         /* Update new refcnts using new_roots */
2470         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2471                                    UPDATE_NEW);
2472         if (ret < 0)
2473                 goto out;
2474
2475         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2476                                num_bytes, seq);
2477
2478         /*
2479          * Bump qgroup_seq to avoid seq overlap
2480          */
2481         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2482 out:
2483         spin_unlock(&fs_info->qgroup_lock);
2484 out_free:
2485         ulist_free(tmp);
2486         ulist_free(qgroups);
2487         ulist_free(old_roots);
2488         ulist_free(new_roots);
2489         return ret;
2490 }
2491
2492 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2493 {
2494         struct btrfs_fs_info *fs_info = trans->fs_info;
2495         struct btrfs_qgroup_extent_record *record;
2496         struct btrfs_delayed_ref_root *delayed_refs;
2497         struct ulist *new_roots = NULL;
2498         struct rb_node *node;
2499         u64 num_dirty_extents = 0;
2500         u64 qgroup_to_skip;
2501         int ret = 0;
2502
2503         delayed_refs = &trans->transaction->delayed_refs;
2504         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2505         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2506                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2507                                   node);
2508
2509                 num_dirty_extents++;
2510                 trace_btrfs_qgroup_account_extents(fs_info, record);
2511
2512                 if (!ret) {
2513                         /*
2514                          * Old roots should be searched when inserting qgroup
2515                          * extent record
2516                          */
2517                         if (WARN_ON(!record->old_roots)) {
2518                                 /* Search commit root to find old_roots */
2519                                 ret = btrfs_find_all_roots(NULL, fs_info,
2520                                                 record->bytenr, 0,
2521                                                 &record->old_roots, false);
2522                                 if (ret < 0)
2523                                         goto cleanup;
2524                         }
2525
2526                         /* Free the reserved data space */
2527                         btrfs_qgroup_free_refroot(fs_info,
2528                                         record->data_rsv_refroot,
2529                                         record->data_rsv,
2530                                         BTRFS_QGROUP_RSV_DATA);
2531                         /*
2532                          * Use SEQ_LAST as time_seq to do special search, which
2533                          * doesn't lock tree or delayed_refs and search current
2534                          * root. It's safe inside commit_transaction().
2535                          */
2536                         ret = btrfs_find_all_roots(trans, fs_info,
2537                                 record->bytenr, SEQ_LAST, &new_roots, false);
2538                         if (ret < 0)
2539                                 goto cleanup;
2540                         if (qgroup_to_skip) {
2541                                 ulist_del(new_roots, qgroup_to_skip, 0);
2542                                 ulist_del(record->old_roots, qgroup_to_skip,
2543                                           0);
2544                         }
2545                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2546                                                           record->num_bytes,
2547                                                           record->old_roots,
2548                                                           new_roots);
2549                         record->old_roots = NULL;
2550                         new_roots = NULL;
2551                 }
2552 cleanup:
2553                 ulist_free(record->old_roots);
2554                 ulist_free(new_roots);
2555                 new_roots = NULL;
2556                 rb_erase(node, &delayed_refs->dirty_extent_root);
2557                 kfree(record);
2558
2559         }
2560         trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2561                                        num_dirty_extents);
2562         return ret;
2563 }
2564
2565 /*
2566  * called from commit_transaction. Writes all changed qgroups to disk.
2567  */
2568 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2569 {
2570         struct btrfs_fs_info *fs_info = trans->fs_info;
2571         struct btrfs_root *quota_root = fs_info->quota_root;
2572         int ret = 0;
2573
2574         if (!quota_root)
2575                 return ret;
2576
2577         spin_lock(&fs_info->qgroup_lock);
2578         while (!list_empty(&fs_info->dirty_qgroups)) {
2579                 struct btrfs_qgroup *qgroup;
2580                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2581                                           struct btrfs_qgroup, dirty);
2582                 list_del_init(&qgroup->dirty);
2583                 spin_unlock(&fs_info->qgroup_lock);
2584                 ret = update_qgroup_info_item(trans, qgroup);
2585                 if (ret)
2586                         fs_info->qgroup_flags |=
2587                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2588                 ret = update_qgroup_limit_item(trans, qgroup);
2589                 if (ret)
2590                         fs_info->qgroup_flags |=
2591                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2592                 spin_lock(&fs_info->qgroup_lock);
2593         }
2594         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2595                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2596         else
2597                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2598         spin_unlock(&fs_info->qgroup_lock);
2599
2600         ret = update_qgroup_status_item(trans);
2601         if (ret)
2602                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2603
2604         return ret;
2605 }
2606
2607 /*
2608  * Copy the accounting information between qgroups. This is necessary
2609  * when a snapshot or a subvolume is created. Throwing an error will
2610  * cause a transaction abort so we take extra care here to only error
2611  * when a readonly fs is a reasonable outcome.
2612  */
2613 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2614                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2615 {
2616         int ret = 0;
2617         int i;
2618         u64 *i_qgroups;
2619         struct btrfs_fs_info *fs_info = trans->fs_info;
2620         struct btrfs_root *quota_root;
2621         struct btrfs_qgroup *srcgroup;
2622         struct btrfs_qgroup *dstgroup;
2623         u32 level_size = 0;
2624         u64 nums;
2625
2626         mutex_lock(&fs_info->qgroup_ioctl_lock);
2627         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2628                 goto out;
2629
2630         quota_root = fs_info->quota_root;
2631         if (!quota_root) {
2632                 ret = -EINVAL;
2633                 goto out;
2634         }
2635
2636         if (inherit) {
2637                 i_qgroups = (u64 *)(inherit + 1);
2638                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2639                        2 * inherit->num_excl_copies;
2640                 for (i = 0; i < nums; ++i) {
2641                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2642
2643                         /*
2644                          * Zero out invalid groups so we can ignore
2645                          * them later.
2646                          */
2647                         if (!srcgroup ||
2648                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2649                                 *i_qgroups = 0ULL;
2650
2651                         ++i_qgroups;
2652                 }
2653         }
2654
2655         /*
2656          * create a tracking group for the subvol itself
2657          */
2658         ret = add_qgroup_item(trans, quota_root, objectid);
2659         if (ret)
2660                 goto out;
2661
2662         /*
2663          * add qgroup to all inherited groups
2664          */
2665         if (inherit) {
2666                 i_qgroups = (u64 *)(inherit + 1);
2667                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2668                         if (*i_qgroups == 0)
2669                                 continue;
2670                         ret = add_qgroup_relation_item(trans, objectid,
2671                                                        *i_qgroups);
2672                         if (ret && ret != -EEXIST)
2673                                 goto out;
2674                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2675                                                        objectid);
2676                         if (ret && ret != -EEXIST)
2677                                 goto out;
2678                 }
2679                 ret = 0;
2680         }
2681
2682
2683         spin_lock(&fs_info->qgroup_lock);
2684
2685         dstgroup = add_qgroup_rb(fs_info, objectid);
2686         if (IS_ERR(dstgroup)) {
2687                 ret = PTR_ERR(dstgroup);
2688                 goto unlock;
2689         }
2690
2691         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2692                 dstgroup->lim_flags = inherit->lim.flags;
2693                 dstgroup->max_rfer = inherit->lim.max_rfer;
2694                 dstgroup->max_excl = inherit->lim.max_excl;
2695                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2696                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2697
2698                 ret = update_qgroup_limit_item(trans, dstgroup);
2699                 if (ret) {
2700                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2701                         btrfs_info(fs_info,
2702                                    "unable to update quota limit for %llu",
2703                                    dstgroup->qgroupid);
2704                         goto unlock;
2705                 }
2706         }
2707
2708         if (srcid) {
2709                 srcgroup = find_qgroup_rb(fs_info, srcid);
2710                 if (!srcgroup)
2711                         goto unlock;
2712
2713                 /*
2714                  * We call inherit after we clone the root in order to make sure
2715                  * our counts don't go crazy, so at this point the only
2716                  * difference between the two roots should be the root node.
2717                  */
2718                 level_size = fs_info->nodesize;
2719                 dstgroup->rfer = srcgroup->rfer;
2720                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2721                 dstgroup->excl = level_size;
2722                 dstgroup->excl_cmpr = level_size;
2723                 srcgroup->excl = level_size;
2724                 srcgroup->excl_cmpr = level_size;
2725
2726                 /* inherit the limit info */
2727                 dstgroup->lim_flags = srcgroup->lim_flags;
2728                 dstgroup->max_rfer = srcgroup->max_rfer;
2729                 dstgroup->max_excl = srcgroup->max_excl;
2730                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2731                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2732
2733                 qgroup_dirty(fs_info, dstgroup);
2734                 qgroup_dirty(fs_info, srcgroup);
2735         }
2736
2737         if (!inherit)
2738                 goto unlock;
2739
2740         i_qgroups = (u64 *)(inherit + 1);
2741         for (i = 0; i < inherit->num_qgroups; ++i) {
2742                 if (*i_qgroups) {
2743                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2744                         if (ret)
2745                                 goto unlock;
2746                 }
2747                 ++i_qgroups;
2748         }
2749
2750         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2751                 struct btrfs_qgroup *src;
2752                 struct btrfs_qgroup *dst;
2753
2754                 if (!i_qgroups[0] || !i_qgroups[1])
2755                         continue;
2756
2757                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2758                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2759
2760                 if (!src || !dst) {
2761                         ret = -EINVAL;
2762                         goto unlock;
2763                 }
2764
2765                 dst->rfer = src->rfer - level_size;
2766                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2767         }
2768         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2769                 struct btrfs_qgroup *src;
2770                 struct btrfs_qgroup *dst;
2771
2772                 if (!i_qgroups[0] || !i_qgroups[1])
2773                         continue;
2774
2775                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2776                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2777
2778                 if (!src || !dst) {
2779                         ret = -EINVAL;
2780                         goto unlock;
2781                 }
2782
2783                 dst->excl = src->excl + level_size;
2784                 dst->excl_cmpr = src->excl_cmpr + level_size;
2785         }
2786
2787 unlock:
2788         spin_unlock(&fs_info->qgroup_lock);
2789 out:
2790         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2791         return ret;
2792 }
2793
2794 /*
2795  * Two limits to commit transaction in advance.
2796  *
2797  * For RATIO, it will be 1/RATIO of the remaining limit
2798  * (excluding data and prealloc meta) as threshold.
2799  * For SIZE, it will be in byte unit as threshold.
2800  */
2801 #define QGROUP_PERTRANS_RATIO           32
2802 #define QGROUP_PERTRANS_SIZE            SZ_32M
2803 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2804                                 const struct btrfs_qgroup *qg, u64 num_bytes)
2805 {
2806         u64 limit;
2807         u64 threshold;
2808
2809         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2810             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2811                 return false;
2812
2813         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2814             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2815                 return false;
2816
2817         /*
2818          * Even if we passed the check, it's better to check if reservation
2819          * for meta_pertrans is pushing us near limit.
2820          * If there is too much pertrans reservation or it's near the limit,
2821          * let's try commit transaction to free some, using transaction_kthread
2822          */
2823         if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2824                               BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2825                 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2826                         limit = qg->max_excl;
2827                 else
2828                         limit = qg->max_rfer;
2829                 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2830                             qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2831                             QGROUP_PERTRANS_RATIO;
2832                 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2833
2834                 /*
2835                  * Use transaction_kthread to commit transaction, so we no
2836                  * longer need to bother nested transaction nor lock context.
2837                  */
2838                 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2839                         btrfs_commit_transaction_locksafe(fs_info);
2840         }
2841
2842         return true;
2843 }
2844
2845 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2846                           enum btrfs_qgroup_rsv_type type)
2847 {
2848         struct btrfs_root *quota_root;
2849         struct btrfs_qgroup *qgroup;
2850         struct btrfs_fs_info *fs_info = root->fs_info;
2851         u64 ref_root = root->root_key.objectid;
2852         int ret = 0;
2853         struct ulist_node *unode;
2854         struct ulist_iterator uiter;
2855
2856         if (!is_fstree(ref_root))
2857                 return 0;
2858
2859         if (num_bytes == 0)
2860                 return 0;
2861
2862         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2863             capable(CAP_SYS_RESOURCE))
2864                 enforce = false;
2865
2866         spin_lock(&fs_info->qgroup_lock);
2867         quota_root = fs_info->quota_root;
2868         if (!quota_root)
2869                 goto out;
2870
2871         qgroup = find_qgroup_rb(fs_info, ref_root);
2872         if (!qgroup)
2873                 goto out;
2874
2875         /*
2876          * in a first step, we check all affected qgroups if any limits would
2877          * be exceeded
2878          */
2879         ulist_reinit(fs_info->qgroup_ulist);
2880         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2881                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2882         if (ret < 0)
2883                 goto out;
2884         ULIST_ITER_INIT(&uiter);
2885         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2886                 struct btrfs_qgroup *qg;
2887                 struct btrfs_qgroup_list *glist;
2888
2889                 qg = unode_aux_to_qgroup(unode);
2890
2891                 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2892                         ret = -EDQUOT;
2893                         goto out;
2894                 }
2895
2896                 list_for_each_entry(glist, &qg->groups, next_group) {
2897                         ret = ulist_add(fs_info->qgroup_ulist,
2898                                         glist->group->qgroupid,
2899                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2900                         if (ret < 0)
2901                                 goto out;
2902                 }
2903         }
2904         ret = 0;
2905         /*
2906          * no limits exceeded, now record the reservation into all qgroups
2907          */
2908         ULIST_ITER_INIT(&uiter);
2909         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2910                 struct btrfs_qgroup *qg;
2911
2912                 qg = unode_aux_to_qgroup(unode);
2913
2914                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2915         }
2916
2917 out:
2918         spin_unlock(&fs_info->qgroup_lock);
2919         return ret;
2920 }
2921
2922 /*
2923  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2924  * qgroup).
2925  *
2926  * Will handle all higher level qgroup too.
2927  *
2928  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2929  * This special case is only used for META_PERTRANS type.
2930  */
2931 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2932                                u64 ref_root, u64 num_bytes,
2933                                enum btrfs_qgroup_rsv_type type)
2934 {
2935         struct btrfs_root *quota_root;
2936         struct btrfs_qgroup *qgroup;
2937         struct ulist_node *unode;
2938         struct ulist_iterator uiter;
2939         int ret = 0;
2940
2941         if (!is_fstree(ref_root))
2942                 return;
2943
2944         if (num_bytes == 0)
2945                 return;
2946
2947         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2948                 WARN(1, "%s: Invalid type to free", __func__);
2949                 return;
2950         }
2951         spin_lock(&fs_info->qgroup_lock);
2952
2953         quota_root = fs_info->quota_root;
2954         if (!quota_root)
2955                 goto out;
2956
2957         qgroup = find_qgroup_rb(fs_info, ref_root);
2958         if (!qgroup)
2959                 goto out;
2960
2961         if (num_bytes == (u64)-1)
2962                 /*
2963                  * We're freeing all pertrans rsv, get reserved value from
2964                  * level 0 qgroup as real num_bytes to free.
2965                  */
2966                 num_bytes = qgroup->rsv.values[type];
2967
2968         ulist_reinit(fs_info->qgroup_ulist);
2969         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2970                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2971         if (ret < 0)
2972                 goto out;
2973         ULIST_ITER_INIT(&uiter);
2974         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2975                 struct btrfs_qgroup *qg;
2976                 struct btrfs_qgroup_list *glist;
2977
2978                 qg = unode_aux_to_qgroup(unode);
2979
2980                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2981
2982                 list_for_each_entry(glist, &qg->groups, next_group) {
2983                         ret = ulist_add(fs_info->qgroup_ulist,
2984                                         glist->group->qgroupid,
2985                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2986                         if (ret < 0)
2987                                 goto out;
2988                 }
2989         }
2990
2991 out:
2992         spin_unlock(&fs_info->qgroup_lock);
2993 }
2994
2995 /*
2996  * Check if the leaf is the last leaf. Which means all node pointers
2997  * are at their last position.
2998  */
2999 static bool is_last_leaf(struct btrfs_path *path)
3000 {
3001         int i;
3002
3003         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3004                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3005                         return false;
3006         }
3007         return true;
3008 }
3009
3010 /*
3011  * returns < 0 on error, 0 when more leafs are to be scanned.
3012  * returns 1 when done.
3013  */
3014 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3015                               struct btrfs_path *path)
3016 {
3017         struct btrfs_fs_info *fs_info = trans->fs_info;
3018         struct btrfs_key found;
3019         struct extent_buffer *scratch_leaf = NULL;
3020         struct ulist *roots = NULL;
3021         u64 num_bytes;
3022         bool done;
3023         int slot;
3024         int ret;
3025
3026         mutex_lock(&fs_info->qgroup_rescan_lock);
3027         ret = btrfs_search_slot_for_read(fs_info->extent_root,
3028                                          &fs_info->qgroup_rescan_progress,
3029                                          path, 1, 0);
3030
3031         btrfs_debug(fs_info,
3032                 "current progress key (%llu %u %llu), search_slot ret %d",
3033                 fs_info->qgroup_rescan_progress.objectid,
3034                 fs_info->qgroup_rescan_progress.type,
3035                 fs_info->qgroup_rescan_progress.offset, ret);
3036
3037         if (ret) {
3038                 /*
3039                  * The rescan is about to end, we will not be scanning any
3040                  * further blocks. We cannot unset the RESCAN flag here, because
3041                  * we want to commit the transaction if everything went well.
3042                  * To make the live accounting work in this phase, we set our
3043                  * scan progress pointer such that every real extent objectid
3044                  * will be smaller.
3045                  */
3046                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3047                 btrfs_release_path(path);
3048                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3049                 return ret;
3050         }
3051         done = is_last_leaf(path);
3052
3053         btrfs_item_key_to_cpu(path->nodes[0], &found,
3054                               btrfs_header_nritems(path->nodes[0]) - 1);
3055         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3056
3057         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3058         if (!scratch_leaf) {
3059                 ret = -ENOMEM;
3060                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3061                 goto out;
3062         }
3063         slot = path->slots[0];
3064         btrfs_release_path(path);
3065         mutex_unlock(&fs_info->qgroup_rescan_lock);
3066
3067         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3068                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3069                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3070                     found.type != BTRFS_METADATA_ITEM_KEY)
3071                         continue;
3072                 if (found.type == BTRFS_METADATA_ITEM_KEY)
3073                         num_bytes = fs_info->nodesize;
3074                 else
3075                         num_bytes = found.offset;
3076
3077                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3078                                            &roots, false);
3079                 if (ret < 0)
3080                         goto out;
3081                 /* For rescan, just pass old_roots as NULL */
3082                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3083                                                   num_bytes, NULL, roots);
3084                 if (ret < 0)
3085                         goto out;
3086         }
3087 out:
3088         if (scratch_leaf)
3089                 free_extent_buffer(scratch_leaf);
3090
3091         if (done && !ret) {
3092                 ret = 1;
3093                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3094         }
3095         return ret;
3096 }
3097
3098 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3099 {
3100         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3101                                                      qgroup_rescan_work);
3102         struct btrfs_path *path;
3103         struct btrfs_trans_handle *trans = NULL;
3104         int err = -ENOMEM;
3105         int ret = 0;
3106
3107         path = btrfs_alloc_path();
3108         if (!path)
3109                 goto out;
3110         /*
3111          * Rescan should only search for commit root, and any later difference
3112          * should be recorded by qgroup
3113          */
3114         path->search_commit_root = 1;
3115         path->skip_locking = 1;
3116
3117         err = 0;
3118         while (!err && !btrfs_fs_closing(fs_info)) {
3119                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3120                 if (IS_ERR(trans)) {
3121                         err = PTR_ERR(trans);
3122                         break;
3123                 }
3124                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3125                         err = -EINTR;
3126                 } else {
3127                         err = qgroup_rescan_leaf(trans, path);
3128                 }
3129                 if (err > 0)
3130                         btrfs_commit_transaction(trans);
3131                 else
3132                         btrfs_end_transaction(trans);
3133         }
3134
3135 out:
3136         btrfs_free_path(path);
3137
3138         mutex_lock(&fs_info->qgroup_rescan_lock);
3139         if (!btrfs_fs_closing(fs_info))
3140                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3141
3142         if (err > 0 &&
3143             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3144                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3145         } else if (err < 0) {
3146                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3147         }
3148         mutex_unlock(&fs_info->qgroup_rescan_lock);
3149
3150         /*
3151          * only update status, since the previous part has already updated the
3152          * qgroup info.
3153          */
3154         trans = btrfs_start_transaction(fs_info->quota_root, 1);
3155         if (IS_ERR(trans)) {
3156                 err = PTR_ERR(trans);
3157                 btrfs_err(fs_info,
3158                           "fail to start transaction for status update: %d",
3159                           err);
3160                 goto done;
3161         }
3162         ret = update_qgroup_status_item(trans);
3163         if (ret < 0) {
3164                 err = ret;
3165                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3166         }
3167         btrfs_end_transaction(trans);
3168
3169         if (btrfs_fs_closing(fs_info)) {
3170                 btrfs_info(fs_info, "qgroup scan paused");
3171         } else if (err >= 0) {
3172                 btrfs_info(fs_info, "qgroup scan completed%s",
3173                         err > 0 ? " (inconsistency flag cleared)" : "");
3174         } else {
3175                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3176         }
3177
3178 done:
3179         mutex_lock(&fs_info->qgroup_rescan_lock);
3180         fs_info->qgroup_rescan_running = false;
3181         mutex_unlock(&fs_info->qgroup_rescan_lock);
3182         complete_all(&fs_info->qgroup_rescan_completion);
3183 }
3184
3185 /*
3186  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3187  * memory required for the rescan context.
3188  */
3189 static int
3190 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3191                    int init_flags)
3192 {
3193         int ret = 0;
3194
3195         if (!init_flags) {
3196                 /* we're resuming qgroup rescan at mount time */
3197                 if (!(fs_info->qgroup_flags &
3198                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3199                         btrfs_warn(fs_info,
3200                         "qgroup rescan init failed, qgroup is not enabled");
3201                         ret = -EINVAL;
3202                 } else if (!(fs_info->qgroup_flags &
3203                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3204                         btrfs_warn(fs_info,
3205                         "qgroup rescan init failed, qgroup rescan is not queued");
3206                         ret = -EINVAL;
3207                 }
3208
3209                 if (ret)
3210                         return ret;
3211         }
3212
3213         mutex_lock(&fs_info->qgroup_rescan_lock);
3214         spin_lock(&fs_info->qgroup_lock);
3215
3216         if (init_flags) {
3217                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3218                         btrfs_warn(fs_info,
3219                                    "qgroup rescan is already in progress");
3220                         ret = -EINPROGRESS;
3221                 } else if (!(fs_info->qgroup_flags &
3222                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3223                         btrfs_warn(fs_info,
3224                         "qgroup rescan init failed, qgroup is not enabled");
3225                         ret = -EINVAL;
3226                 }
3227
3228                 if (ret) {
3229                         spin_unlock(&fs_info->qgroup_lock);
3230                         mutex_unlock(&fs_info->qgroup_rescan_lock);
3231                         return ret;
3232                 }
3233                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3234         }
3235
3236         memset(&fs_info->qgroup_rescan_progress, 0,
3237                 sizeof(fs_info->qgroup_rescan_progress));
3238         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3239         init_completion(&fs_info->qgroup_rescan_completion);
3240         fs_info->qgroup_rescan_running = true;
3241
3242         spin_unlock(&fs_info->qgroup_lock);
3243         mutex_unlock(&fs_info->qgroup_rescan_lock);
3244
3245         memset(&fs_info->qgroup_rescan_work, 0,
3246                sizeof(fs_info->qgroup_rescan_work));
3247         btrfs_init_work(&fs_info->qgroup_rescan_work,
3248                         btrfs_qgroup_rescan_helper,
3249                         btrfs_qgroup_rescan_worker, NULL, NULL);
3250         return 0;
3251 }
3252
3253 static void
3254 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3255 {
3256         struct rb_node *n;
3257         struct btrfs_qgroup *qgroup;
3258
3259         spin_lock(&fs_info->qgroup_lock);
3260         /* clear all current qgroup tracking information */
3261         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3262                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3263                 qgroup->rfer = 0;
3264                 qgroup->rfer_cmpr = 0;
3265                 qgroup->excl = 0;
3266                 qgroup->excl_cmpr = 0;
3267                 qgroup_dirty(fs_info, qgroup);
3268         }
3269         spin_unlock(&fs_info->qgroup_lock);
3270 }
3271
3272 int
3273 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3274 {
3275         int ret = 0;
3276         struct btrfs_trans_handle *trans;
3277
3278         ret = qgroup_rescan_init(fs_info, 0, 1);
3279         if (ret)
3280                 return ret;
3281
3282         /*
3283          * We have set the rescan_progress to 0, which means no more
3284          * delayed refs will be accounted by btrfs_qgroup_account_ref.
3285          * However, btrfs_qgroup_account_ref may be right after its call
3286          * to btrfs_find_all_roots, in which case it would still do the
3287          * accounting.
3288          * To solve this, we're committing the transaction, which will
3289          * ensure we run all delayed refs and only after that, we are
3290          * going to clear all tracking information for a clean start.
3291          */
3292
3293         trans = btrfs_join_transaction(fs_info->fs_root);
3294         if (IS_ERR(trans)) {
3295                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3296                 return PTR_ERR(trans);
3297         }
3298         ret = btrfs_commit_transaction(trans);
3299         if (ret) {
3300                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3301                 return ret;
3302         }
3303
3304         qgroup_rescan_zero_tracking(fs_info);
3305
3306         btrfs_queue_work(fs_info->qgroup_rescan_workers,
3307                          &fs_info->qgroup_rescan_work);
3308
3309         return 0;
3310 }
3311
3312 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3313                                      bool interruptible)
3314 {
3315         int running;
3316         int ret = 0;
3317
3318         mutex_lock(&fs_info->qgroup_rescan_lock);
3319         spin_lock(&fs_info->qgroup_lock);
3320         running = fs_info->qgroup_rescan_running;
3321         spin_unlock(&fs_info->qgroup_lock);
3322         mutex_unlock(&fs_info->qgroup_rescan_lock);
3323
3324         if (!running)
3325                 return 0;
3326
3327         if (interruptible)
3328                 ret = wait_for_completion_interruptible(
3329                                         &fs_info->qgroup_rescan_completion);
3330         else
3331                 wait_for_completion(&fs_info->qgroup_rescan_completion);
3332
3333         return ret;
3334 }
3335
3336 /*
3337  * this is only called from open_ctree where we're still single threaded, thus
3338  * locking is omitted here.
3339  */
3340 void
3341 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3342 {
3343         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3344                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3345                                  &fs_info->qgroup_rescan_work);
3346 }
3347
3348 /*
3349  * Reserve qgroup space for range [start, start + len).
3350  *
3351  * This function will either reserve space from related qgroups or doing
3352  * nothing if the range is already reserved.
3353  *
3354  * Return 0 for successful reserve
3355  * Return <0 for error (including -EQUOT)
3356  *
3357  * NOTE: this function may sleep for memory allocation.
3358  *       if btrfs_qgroup_reserve_data() is called multiple times with
3359  *       same @reserved, caller must ensure when error happens it's OK
3360  *       to free *ALL* reserved space.
3361  */
3362 int btrfs_qgroup_reserve_data(struct inode *inode,
3363                         struct extent_changeset **reserved_ret, u64 start,
3364                         u64 len)
3365 {
3366         struct btrfs_root *root = BTRFS_I(inode)->root;
3367         struct ulist_node *unode;
3368         struct ulist_iterator uiter;
3369         struct extent_changeset *reserved;
3370         u64 orig_reserved;
3371         u64 to_reserve;
3372         int ret;
3373
3374         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3375             !is_fstree(root->root_key.objectid) || len == 0)
3376                 return 0;
3377
3378         /* @reserved parameter is mandatory for qgroup */
3379         if (WARN_ON(!reserved_ret))
3380                 return -EINVAL;
3381         if (!*reserved_ret) {
3382                 *reserved_ret = extent_changeset_alloc();
3383                 if (!*reserved_ret)
3384                         return -ENOMEM;
3385         }
3386         reserved = *reserved_ret;
3387         /* Record already reserved space */
3388         orig_reserved = reserved->bytes_changed;
3389         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3390                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3391
3392         /* Newly reserved space */
3393         to_reserve = reserved->bytes_changed - orig_reserved;
3394         trace_btrfs_qgroup_reserve_data(inode, start, len,
3395                                         to_reserve, QGROUP_RESERVE);
3396         if (ret < 0)
3397                 goto cleanup;
3398         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3399         if (ret < 0)
3400                 goto cleanup;
3401
3402         return ret;
3403
3404 cleanup:
3405         /* cleanup *ALL* already reserved ranges */
3406         ULIST_ITER_INIT(&uiter);
3407         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3408                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3409                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3410         extent_changeset_release(reserved);
3411         return ret;
3412 }
3413
3414 /* Free ranges specified by @reserved, normally in error path */
3415 static int qgroup_free_reserved_data(struct inode *inode,
3416                         struct extent_changeset *reserved, u64 start, u64 len)
3417 {
3418         struct btrfs_root *root = BTRFS_I(inode)->root;
3419         struct ulist_node *unode;
3420         struct ulist_iterator uiter;
3421         struct extent_changeset changeset;
3422         int freed = 0;
3423         int ret;
3424
3425         extent_changeset_init(&changeset);
3426         len = round_up(start + len, root->fs_info->sectorsize);
3427         start = round_down(start, root->fs_info->sectorsize);
3428
3429         ULIST_ITER_INIT(&uiter);
3430         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3431                 u64 range_start = unode->val;
3432                 /* unode->aux is the inclusive end */
3433                 u64 range_len = unode->aux - range_start + 1;
3434                 u64 free_start;
3435                 u64 free_len;
3436
3437                 extent_changeset_release(&changeset);
3438
3439                 /* Only free range in range [start, start + len) */
3440                 if (range_start >= start + len ||
3441                     range_start + range_len <= start)
3442                         continue;
3443                 free_start = max(range_start, start);
3444                 free_len = min(start + len, range_start + range_len) -
3445                            free_start;
3446                 /*
3447                  * TODO: To also modify reserved->ranges_reserved to reflect
3448                  * the modification.
3449                  *
3450                  * However as long as we free qgroup reserved according to
3451                  * EXTENT_QGROUP_RESERVED, we won't double free.
3452                  * So not need to rush.
3453                  */
3454                 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3455                                 free_start, free_start + free_len - 1,
3456                                 EXTENT_QGROUP_RESERVED, &changeset);
3457                 if (ret < 0)
3458                         goto out;
3459                 freed += changeset.bytes_changed;
3460         }
3461         btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3462                                   BTRFS_QGROUP_RSV_DATA);
3463         ret = freed;
3464 out:
3465         extent_changeset_release(&changeset);
3466         return ret;
3467 }
3468
3469 static int __btrfs_qgroup_release_data(struct inode *inode,
3470                         struct extent_changeset *reserved, u64 start, u64 len,
3471                         int free)
3472 {
3473         struct extent_changeset changeset;
3474         int trace_op = QGROUP_RELEASE;
3475         int ret;
3476
3477         if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3478                       &BTRFS_I(inode)->root->fs_info->flags))
3479                 return 0;
3480
3481         /* In release case, we shouldn't have @reserved */
3482         WARN_ON(!free && reserved);
3483         if (free && reserved)
3484                 return qgroup_free_reserved_data(inode, reserved, start, len);
3485         extent_changeset_init(&changeset);
3486         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3487                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3488         if (ret < 0)
3489                 goto out;
3490
3491         if (free)
3492                 trace_op = QGROUP_FREE;
3493         trace_btrfs_qgroup_release_data(inode, start, len,
3494                                         changeset.bytes_changed, trace_op);
3495         if (free)
3496                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3497                                 BTRFS_I(inode)->root->root_key.objectid,
3498                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3499         ret = changeset.bytes_changed;
3500 out:
3501         extent_changeset_release(&changeset);
3502         return ret;
3503 }
3504
3505 /*
3506  * Free a reserved space range from io_tree and related qgroups
3507  *
3508  * Should be called when a range of pages get invalidated before reaching disk.
3509  * Or for error cleanup case.
3510  * if @reserved is given, only reserved range in [@start, @start + @len) will
3511  * be freed.
3512  *
3513  * For data written to disk, use btrfs_qgroup_release_data().
3514  *
3515  * NOTE: This function may sleep for memory allocation.
3516  */
3517 int btrfs_qgroup_free_data(struct inode *inode,
3518                         struct extent_changeset *reserved, u64 start, u64 len)
3519 {
3520         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3521 }
3522
3523 /*
3524  * Release a reserved space range from io_tree only.
3525  *
3526  * Should be called when a range of pages get written to disk and corresponding
3527  * FILE_EXTENT is inserted into corresponding root.
3528  *
3529  * Since new qgroup accounting framework will only update qgroup numbers at
3530  * commit_transaction() time, its reserved space shouldn't be freed from
3531  * related qgroups.
3532  *
3533  * But we should release the range from io_tree, to allow further write to be
3534  * COWed.
3535  *
3536  * NOTE: This function may sleep for memory allocation.
3537  */
3538 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3539 {
3540         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3541 }
3542
3543 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3544                               enum btrfs_qgroup_rsv_type type)
3545 {
3546         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3547             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3548                 return;
3549         if (num_bytes == 0)
3550                 return;
3551
3552         spin_lock(&root->qgroup_meta_rsv_lock);
3553         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3554                 root->qgroup_meta_rsv_prealloc += num_bytes;
3555         else
3556                 root->qgroup_meta_rsv_pertrans += num_bytes;
3557         spin_unlock(&root->qgroup_meta_rsv_lock);
3558 }
3559
3560 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3561                              enum btrfs_qgroup_rsv_type type)
3562 {
3563         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3564             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3565                 return 0;
3566         if (num_bytes == 0)
3567                 return 0;
3568
3569         spin_lock(&root->qgroup_meta_rsv_lock);
3570         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3571                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3572                                   num_bytes);
3573                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3574         } else {
3575                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3576                                   num_bytes);
3577                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3578         }
3579         spin_unlock(&root->qgroup_meta_rsv_lock);
3580         return num_bytes;
3581 }
3582
3583 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3584                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3585 {
3586         struct btrfs_fs_info *fs_info = root->fs_info;
3587         int ret;
3588
3589         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3590             !is_fstree(root->root_key.objectid) || num_bytes == 0)
3591                 return 0;
3592
3593         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3594         trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3595         ret = qgroup_reserve(root, num_bytes, enforce, type);
3596         if (ret < 0)
3597                 return ret;
3598         /*
3599          * Record what we have reserved into root.
3600          *
3601          * To avoid quota disabled->enabled underflow.
3602          * In that case, we may try to free space we haven't reserved
3603          * (since quota was disabled), so record what we reserved into root.
3604          * And ensure later release won't underflow this number.
3605          */
3606         add_root_meta_rsv(root, num_bytes, type);
3607         return ret;
3608 }
3609
3610 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3611 {
3612         struct btrfs_fs_info *fs_info = root->fs_info;
3613
3614         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3615             !is_fstree(root->root_key.objectid))
3616                 return;
3617
3618         /* TODO: Update trace point to handle such free */
3619         trace_qgroup_meta_free_all_pertrans(root);
3620         /* Special value -1 means to free all reserved space */
3621         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3622                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3623 }
3624
3625 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3626                               enum btrfs_qgroup_rsv_type type)
3627 {
3628         struct btrfs_fs_info *fs_info = root->fs_info;
3629
3630         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3631             !is_fstree(root->root_key.objectid))
3632                 return;
3633
3634         /*
3635          * reservation for META_PREALLOC can happen before quota is enabled,
3636          * which can lead to underflow.
3637          * Here ensure we will only free what we really have reserved.
3638          */
3639         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3640         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3641         trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3642         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3643                                   num_bytes, type);
3644 }
3645
3646 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3647                                 int num_bytes)
3648 {
3649         struct btrfs_root *quota_root = fs_info->quota_root;
3650         struct btrfs_qgroup *qgroup;
3651         struct ulist_node *unode;
3652         struct ulist_iterator uiter;
3653         int ret = 0;
3654
3655         if (num_bytes == 0)
3656                 return;
3657         if (!quota_root)
3658                 return;
3659
3660         spin_lock(&fs_info->qgroup_lock);
3661         qgroup = find_qgroup_rb(fs_info, ref_root);
3662         if (!qgroup)
3663                 goto out;
3664         ulist_reinit(fs_info->qgroup_ulist);
3665         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3666                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3667         if (ret < 0)
3668                 goto out;
3669         ULIST_ITER_INIT(&uiter);
3670         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3671                 struct btrfs_qgroup *qg;
3672                 struct btrfs_qgroup_list *glist;
3673
3674                 qg = unode_aux_to_qgroup(unode);
3675
3676                 qgroup_rsv_release(fs_info, qg, num_bytes,
3677                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3678                 qgroup_rsv_add(fs_info, qg, num_bytes,
3679                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3680                 list_for_each_entry(glist, &qg->groups, next_group) {
3681                         ret = ulist_add(fs_info->qgroup_ulist,
3682                                         glist->group->qgroupid,
3683                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3684                         if (ret < 0)
3685                                 goto out;
3686                 }
3687         }
3688 out:
3689         spin_unlock(&fs_info->qgroup_lock);
3690 }
3691
3692 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3693 {
3694         struct btrfs_fs_info *fs_info = root->fs_info;
3695
3696         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3697             !is_fstree(root->root_key.objectid))
3698                 return;
3699         /* Same as btrfs_qgroup_free_meta_prealloc() */
3700         num_bytes = sub_root_meta_rsv(root, num_bytes,
3701                                       BTRFS_QGROUP_RSV_META_PREALLOC);
3702         trace_qgroup_meta_convert(root, num_bytes);
3703         qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3704 }
3705
3706 /*
3707  * Check qgroup reserved space leaking, normally at destroy inode
3708  * time
3709  */
3710 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3711 {
3712         struct extent_changeset changeset;
3713         struct ulist_node *unode;
3714         struct ulist_iterator iter;
3715         int ret;
3716
3717         extent_changeset_init(&changeset);
3718         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3719                         EXTENT_QGROUP_RESERVED, &changeset);
3720
3721         WARN_ON(ret < 0);
3722         if (WARN_ON(changeset.bytes_changed)) {
3723                 ULIST_ITER_INIT(&iter);
3724                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3725                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3726                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3727                                 inode->i_ino, unode->val, unode->aux);
3728                 }
3729                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3730                                 BTRFS_I(inode)->root->root_key.objectid,
3731                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3732
3733         }
3734         extent_changeset_release(&changeset);
3735 }
3736
3737 void btrfs_qgroup_init_swapped_blocks(
3738         struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3739 {
3740         int i;
3741
3742         spin_lock_init(&swapped_blocks->lock);
3743         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3744                 swapped_blocks->blocks[i] = RB_ROOT;
3745         swapped_blocks->swapped = false;
3746 }
3747
3748 /*
3749  * Delete all swapped blocks record of @root.
3750  * Every record here means we skipped a full subtree scan for qgroup.
3751  *
3752  * Gets called when committing one transaction.
3753  */
3754 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3755 {
3756         struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3757         int i;
3758
3759         swapped_blocks = &root->swapped_blocks;
3760
3761         spin_lock(&swapped_blocks->lock);
3762         if (!swapped_blocks->swapped)
3763                 goto out;
3764         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3765                 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3766                 struct btrfs_qgroup_swapped_block *entry;
3767                 struct btrfs_qgroup_swapped_block *next;
3768
3769                 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3770                                                      node)
3771                         kfree(entry);
3772                 swapped_blocks->blocks[i] = RB_ROOT;
3773         }
3774         swapped_blocks->swapped = false;
3775 out:
3776         spin_unlock(&swapped_blocks->lock);
3777 }
3778
3779 /*
3780  * Add subtree roots record into @subvol_root.
3781  *
3782  * @subvol_root:        tree root of the subvolume tree get swapped
3783  * @bg:                 block group under balance
3784  * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3785  * @reloc_parent/slot:  pointer to the subtree root in reloc tree
3786  *                      BOTH POINTERS ARE BEFORE TREE SWAP
3787  * @last_snapshot:      last snapshot generation of the subvolume tree
3788  */
3789 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3790                 struct btrfs_root *subvol_root,
3791                 struct btrfs_block_group_cache *bg,
3792                 struct extent_buffer *subvol_parent, int subvol_slot,
3793                 struct extent_buffer *reloc_parent, int reloc_slot,
3794                 u64 last_snapshot)
3795 {
3796         struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3797         struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3798         struct btrfs_qgroup_swapped_block *block;
3799         struct rb_node **cur;
3800         struct rb_node *parent = NULL;
3801         int level = btrfs_header_level(subvol_parent) - 1;
3802         int ret = 0;
3803
3804         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3805                 return 0;
3806
3807         if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3808             btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3809                 btrfs_err_rl(fs_info,
3810                 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3811                         __func__,
3812                         btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3813                         btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3814                 return -EUCLEAN;
3815         }
3816
3817         block = kmalloc(sizeof(*block), GFP_NOFS);
3818         if (!block) {
3819                 ret = -ENOMEM;
3820                 goto out;
3821         }
3822
3823         /*
3824          * @reloc_parent/slot is still before swap, while @block is going to
3825          * record the bytenr after swap, so we do the swap here.
3826          */
3827         block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3828         block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3829                                                              reloc_slot);
3830         block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3831         block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3832                                                             subvol_slot);
3833         block->last_snapshot = last_snapshot;
3834         block->level = level;
3835         if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
3836                 block->trace_leaf = true;
3837         else
3838                 block->trace_leaf = false;
3839         btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3840
3841         /* Insert @block into @blocks */
3842         spin_lock(&blocks->lock);
3843         cur = &blocks->blocks[level].rb_node;
3844         while (*cur) {
3845                 struct btrfs_qgroup_swapped_block *entry;
3846
3847                 parent = *cur;
3848                 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3849                                  node);
3850
3851                 if (entry->subvol_bytenr < block->subvol_bytenr) {
3852                         cur = &(*cur)->rb_left;
3853                 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3854                         cur = &(*cur)->rb_right;
3855                 } else {
3856                         if (entry->subvol_generation !=
3857                                         block->subvol_generation ||
3858                             entry->reloc_bytenr != block->reloc_bytenr ||
3859                             entry->reloc_generation !=
3860                                         block->reloc_generation) {
3861                                 /*
3862                                  * Duplicated but mismatch entry found.
3863                                  * Shouldn't happen.
3864                                  *
3865                                  * Marking qgroup inconsistent should be enough
3866                                  * for end users.
3867                                  */
3868                                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3869                                 ret = -EEXIST;
3870                         }
3871                         kfree(block);
3872                         goto out_unlock;
3873                 }
3874         }
3875         rb_link_node(&block->node, parent, cur);
3876         rb_insert_color(&block->node, &blocks->blocks[level]);
3877         blocks->swapped = true;
3878 out_unlock:
3879         spin_unlock(&blocks->lock);
3880 out:
3881         if (ret < 0)
3882                 fs_info->qgroup_flags |=
3883                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3884         return ret;
3885 }
3886
3887 /*
3888  * Check if the tree block is a subtree root, and if so do the needed
3889  * delayed subtree trace for qgroup.
3890  *
3891  * This is called during btrfs_cow_block().
3892  */
3893 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3894                                          struct btrfs_root *root,
3895                                          struct extent_buffer *subvol_eb)
3896 {
3897         struct btrfs_fs_info *fs_info = root->fs_info;
3898         struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3899         struct btrfs_qgroup_swapped_block *block;
3900         struct extent_buffer *reloc_eb = NULL;
3901         struct rb_node *node;
3902         bool found = false;
3903         bool swapped = false;
3904         int level = btrfs_header_level(subvol_eb);
3905         int ret = 0;
3906         int i;
3907
3908         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3909                 return 0;
3910         if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3911                 return 0;
3912
3913         spin_lock(&blocks->lock);
3914         if (!blocks->swapped) {
3915                 spin_unlock(&blocks->lock);
3916                 return 0;
3917         }
3918         node = blocks->blocks[level].rb_node;
3919
3920         while (node) {
3921                 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3922                 if (block->subvol_bytenr < subvol_eb->start) {
3923                         node = node->rb_left;
3924                 } else if (block->subvol_bytenr > subvol_eb->start) {
3925                         node = node->rb_right;
3926                 } else {
3927                         found = true;
3928                         break;
3929                 }
3930         }
3931         if (!found) {
3932                 spin_unlock(&blocks->lock);
3933                 goto out;
3934         }
3935         /* Found one, remove it from @blocks first and update blocks->swapped */
3936         rb_erase(&block->node, &blocks->blocks[level]);
3937         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3938                 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3939                         swapped = true;
3940                         break;
3941                 }
3942         }
3943         blocks->swapped = swapped;
3944         spin_unlock(&blocks->lock);
3945
3946         /* Read out reloc subtree root */
3947         reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3948                                    block->reloc_generation, block->level,
3949                                    &block->first_key);
3950         if (IS_ERR(reloc_eb)) {
3951                 ret = PTR_ERR(reloc_eb);
3952                 reloc_eb = NULL;
3953                 goto free_out;
3954         }
3955         if (!extent_buffer_uptodate(reloc_eb)) {
3956                 ret = -EIO;
3957                 goto free_out;
3958         }
3959
3960         ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3961                         block->last_snapshot, block->trace_leaf);
3962 free_out:
3963         kfree(block);
3964         free_extent_buffer(reloc_eb);
3965 out:
3966         if (ret < 0) {
3967                 btrfs_err_rl(fs_info,
3968                              "failed to account subtree at bytenr %llu: %d",
3969                              subvol_eb->start, ret);
3970                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3971         }
3972         return ret;
3973 }