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 fuer 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 umounting
 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,
 534                                     struct btrfs_root *quota_root,
 535                                     u64 src, u64 dst)
 536 {
 537         int ret;
 538         struct btrfs_path *path;
 539         struct btrfs_key key;
 540
 541         path = btrfs_alloc_path();
 542         if (!path)
 543                 return -ENOMEM;
 544
 545         key.objectid = src;
 546         key.type = BTRFS_QGROUP_RELATION_KEY;
 547         key.offset = dst;
 548
 549         ret = btrfs_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,
 558                                     struct btrfs_root *quota_root,
 559                                     u64 src, u64 dst)
 560 {
 561         int ret;
 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,
 657                            struct btrfs_root *quota_root, u64 qgroupid)
 658 {
 659         int ret;
 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_root *root,
 704                                     struct btrfs_qgroup *qgroup)
 705 {
 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, 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_root *root,
 746                                    struct btrfs_qgroup *qgroup)
 747 {
 748         struct btrfs_path *path;
 749         struct btrfs_key key;
 750         struct extent_buffer *l;
 751         struct btrfs_qgroup_info_item *qgroup_info;
 752         int ret;
 753         int slot;
 754
 755         if (btrfs_is_testing(root->fs_info))
 756                 return 0;
 757
 758         key.objectid = 0;
 759         key.type = BTRFS_QGROUP_INFO_KEY;
 760         key.offset = qgroup->qgroupid;
 761
 762         path = btrfs_alloc_path();
 763         if (!path)
 764                 return -ENOMEM;
 765
 766         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 767         if (ret > 0)
 768                 ret = -ENOENT;
 769
 770         if (ret)
 771                 goto out;
 772
 773         l = path->nodes[0];
 774         slot = path->slots[0];
 775         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 776         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 777         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 778         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 779         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 780         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 781
 782         btrfs_mark_buffer_dirty(l);
 783
 784 out:
 785         btrfs_free_path(path);
 786         return ret;
 787 }
 788
 789 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 790                                      struct btrfs_fs_info *fs_info,
 791                                     struct btrfs_root *root)
 792 {
 793         struct btrfs_path *path;
 794         struct btrfs_key key;
 795         struct extent_buffer *l;
 796         struct btrfs_qgroup_status_item *ptr;
 797         int ret;
 798         int slot;
 799
 800         key.objectid = 0;
 801         key.type = BTRFS_QGROUP_STATUS_KEY;
 802         key.offset = 0;
 803
 804         path = btrfs_alloc_path();
 805         if (!path)
 806                 return -ENOMEM;
 807
 808         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 809         if (ret > 0)
 810                 ret = -ENOENT;
 811
 812         if (ret)
 813                 goto out;
 814
 815         l = path->nodes[0];
 816         slot = path->slots[0];
 817         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 818         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 819         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 820         btrfs_set_qgroup_status_rescan(l, ptr,
 821                                 fs_info->qgroup_rescan_progress.objectid);
 822
 823         btrfs_mark_buffer_dirty(l);
 824
 825 out:
 826         btrfs_free_path(path);
 827         return ret;
 828 }
 829
 830 /*
 831  * called with qgroup_lock held
 832  */
 833 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 834                                   struct btrfs_root *root)
 835 {
 836         struct btrfs_path *path;
 837         struct btrfs_key key;
 838         struct extent_buffer *leaf = NULL;
 839         int ret;
 840         int nr = 0;
 841
 842         path = btrfs_alloc_path();
 843         if (!path)
 844                 return -ENOMEM;
 845
 846         path->leave_spinning = 1;
 847
 848         key.objectid = 0;
 849         key.offset = 0;
 850         key.type = 0;
 851
 852         while (1) {
 853                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 854                 if (ret < 0)
 855                         goto out;
 856                 leaf = path->nodes[0];
 857                 nr = btrfs_header_nritems(leaf);
 858                 if (!nr)
 859                         break;
 860                 /*
 861                  * delete the leaf one by one
 862                  * since the whole tree is going
 863                  * to be deleted.
 864                  */
 865                 path->slots[0] = 0;
 866                 ret = btrfs_del_items(trans, root, path, 0, nr);
 867                 if (ret)
 868                         goto out;
 869
 870                 btrfs_release_path(path);
 871         }
 872         ret = 0;
 873 out:
 874         btrfs_free_path(path);
 875         return ret;
 876 }
 877
 878 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 879 {
 880         struct btrfs_root *quota_root;
 881         struct btrfs_root *tree_root = fs_info->tree_root;
 882         struct btrfs_path *path = NULL;
 883         struct btrfs_qgroup_status_item *ptr;
 884         struct extent_buffer *leaf;
 885         struct btrfs_key key;
 886         struct btrfs_key found_key;
 887         struct btrfs_qgroup *qgroup = NULL;
 888         struct btrfs_trans_handle *trans = NULL;
 889         int ret = 0;
 890         int slot;
 891
 892         mutex_lock(&fs_info->qgroup_ioctl_lock);
 893         if (fs_info->quota_root)
 894                 goto out;
 895
 896         /*
 897          * 1 for quota root item
 898          * 1 for BTRFS_QGROUP_STATUS item
 899          *
 900          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
 901          * per subvolume. However those are not currently reserved since it
 902          * would be a lot of overkill.
 903          */
 904         trans = btrfs_start_transaction(tree_root, 2);
 905         if (IS_ERR(trans)) {
 906                 ret = PTR_ERR(trans);
 907                 trans = NULL;
 908                 goto out;
 909         }
 910
 911         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 912         if (!fs_info->qgroup_ulist) {
 913                 ret = -ENOMEM;
 914                 btrfs_abort_transaction(trans, ret);
 915                 goto out;
 916         }
 917
 918         /*
 919          * initially create the quota tree
 920          */
 921         quota_root = btrfs_create_tree(trans, fs_info,
 922                                        BTRFS_QUOTA_TREE_OBJECTID);
 923         if (IS_ERR(quota_root)) {
 924                 ret =  PTR_ERR(quota_root);
 925                 btrfs_abort_transaction(trans, ret);
 926                 goto out;
 927         }
 928
 929         path = btrfs_alloc_path();
 930         if (!path) {
 931                 ret = -ENOMEM;
 932                 btrfs_abort_transaction(trans, ret);
 933                 goto out_free_root;
 934         }
 935
 936         key.objectid = 0;
 937         key.type = BTRFS_QGROUP_STATUS_KEY;
 938         key.offset = 0;
 939
 940         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 941                                       sizeof(*ptr));
 942         if (ret) {
 943                 btrfs_abort_transaction(trans, ret);
 944                 goto out_free_path;
 945         }
 946
 947         leaf = path->nodes[0];
 948         ptr = btrfs_item_ptr(leaf, path->slots[0],
 949                                  struct btrfs_qgroup_status_item);
 950         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 951         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 952         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 953                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 954         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 955         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 956
 957         btrfs_mark_buffer_dirty(leaf);
 958
 959         key.objectid = 0;
 960         key.type = BTRFS_ROOT_REF_KEY;
 961         key.offset = 0;
 962
 963         btrfs_release_path(path);
 964         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 965         if (ret > 0)
 966                 goto out_add_root;
 967         if (ret < 0) {
 968                 btrfs_abort_transaction(trans, ret);
 969                 goto out_free_path;
 970         }
 971
 972         while (1) {
 973                 slot = path->slots[0];
 974                 leaf = path->nodes[0];
 975                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 976
 977                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 978                         ret = add_qgroup_item(trans, quota_root,
 979                                               found_key.offset);
 980                         if (ret) {
 981                                 btrfs_abort_transaction(trans, ret);
 982                                 goto out_free_path;
 983                         }
 984
 985                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 986                         if (IS_ERR(qgroup)) {
 987                                 ret = PTR_ERR(qgroup);
 988                                 btrfs_abort_transaction(trans, ret);
 989                                 goto out_free_path;
 990                         }
 991                 }
 992                 ret = btrfs_next_item(tree_root, path);
 993                 if (ret < 0) {
 994                         btrfs_abort_transaction(trans, ret);
 995                         goto out_free_path;
 996                 }
 997                 if (ret)
 998                         break;
 999         }
1000
1001 out_add_root:
1002         btrfs_release_path(path);
1003         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1004         if (ret) {
1005                 btrfs_abort_transaction(trans, ret);
1006                 goto out_free_path;
1007         }
1008
1009         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1010         if (IS_ERR(qgroup)) {
1011                 ret = PTR_ERR(qgroup);
1012                 btrfs_abort_transaction(trans, ret);
1013                 goto out_free_path;
1014         }
1015         spin_lock(&fs_info->qgroup_lock);
1016         fs_info->quota_root = quota_root;
1017         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1018         spin_unlock(&fs_info->qgroup_lock);
1019
1020         ret = btrfs_commit_transaction(trans);
1021         if (ret) {
1022                 trans = NULL;
1023                 goto out_free_path;
1024         }
1025
1026         ret = qgroup_rescan_init(fs_info, 0, 1);
1027         if (!ret) {
1028                 qgroup_rescan_zero_tracking(fs_info);
1029                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1030                                  &fs_info->qgroup_rescan_work);
1031         }
1032
1033 out_free_path:
1034         btrfs_free_path(path);
1035 out_free_root:
1036         if (ret) {
1037                 free_extent_buffer(quota_root->node);
1038                 free_extent_buffer(quota_root->commit_root);
1039                 kfree(quota_root);
1040         }
1041 out:
1042         if (ret) {
1043                 ulist_free(fs_info->qgroup_ulist);
1044                 fs_info->qgroup_ulist = NULL;
1045                 if (trans)
1046                         btrfs_end_transaction(trans);
1047         }
1048         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1049         return ret;
1050 }
1051
1052 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1053 {
1054         struct btrfs_root *quota_root;
1055         struct btrfs_trans_handle *trans = NULL;
1056         int ret = 0;
1057
1058         mutex_lock(&fs_info->qgroup_ioctl_lock);
1059         if (!fs_info->quota_root)
1060                 goto out;
1061
1062         /*
1063          * 1 For the root item
1064          *
1065          * We should also reserve enough items for the quota tree deletion in
1066          * btrfs_clean_quota_tree but this is not done.
1067          */
1068         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1069         if (IS_ERR(trans)) {
1070                 ret = PTR_ERR(trans);
1071                 goto out;
1072         }
1073
1074         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1075         btrfs_qgroup_wait_for_completion(fs_info, false);
1076         spin_lock(&fs_info->qgroup_lock);
1077         quota_root = fs_info->quota_root;
1078         fs_info->quota_root = NULL;
1079         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1080         spin_unlock(&fs_info->qgroup_lock);
1081
1082         btrfs_free_qgroup_config(fs_info);
1083
1084         ret = btrfs_clean_quota_tree(trans, quota_root);
1085         if (ret) {
1086                 btrfs_abort_transaction(trans, ret);
1087                 goto end_trans;
1088         }
1089
1090         ret = btrfs_del_root(trans, fs_info, &quota_root->root_key);
1091         if (ret) {
1092                 btrfs_abort_transaction(trans, ret);
1093                 goto end_trans;
1094         }
1095
1096         list_del(&quota_root->dirty_list);
1097
1098         btrfs_tree_lock(quota_root->node);
1099         clean_tree_block(fs_info, quota_root->node);
1100         btrfs_tree_unlock(quota_root->node);
1101         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1102
1103         free_extent_buffer(quota_root->node);
1104         free_extent_buffer(quota_root->commit_root);
1105         kfree(quota_root);
1106
1107 end_trans:
1108         ret = btrfs_end_transaction(trans);
1109 out:
1110         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1111         return ret;
1112 }
1113
1114 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1115                          struct btrfs_qgroup *qgroup)
1116 {
1117         if (list_empty(&qgroup->dirty))
1118                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1119 }
1120
1121 /*
1122  * The easy accounting, we're updating qgroup relationship whose child qgroup
1123  * only has exclusive extents.
1124  *
1125  * In this case, all exclsuive extents will also be exlusive for parent, so
1126  * excl/rfer just get added/removed.
1127  *
1128  * So is qgroup reservation space, which should also be added/removed to
1129  * parent.
1130  * Or when child tries to release reservation space, parent will underflow its
1131  * reservation (for relationship adding case).
1132  *
1133  * Caller should hold fs_info->qgroup_lock.
1134  */
1135 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1136                                     struct ulist *tmp, u64 ref_root,
1137                                     struct btrfs_qgroup *src, int sign)
1138 {
1139         struct btrfs_qgroup *qgroup;
1140         struct btrfs_qgroup_list *glist;
1141         struct ulist_node *unode;
1142         struct ulist_iterator uiter;
1143         u64 num_bytes = src->excl;
1144         int ret = 0;
1145
1146         qgroup = find_qgroup_rb(fs_info, ref_root);
1147         if (!qgroup)
1148                 goto out;
1149
1150         qgroup->rfer += sign * num_bytes;
1151         qgroup->rfer_cmpr += sign * num_bytes;
1152
1153         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1154         qgroup->excl += sign * num_bytes;
1155         qgroup->excl_cmpr += sign * num_bytes;
1156
1157         if (sign > 0)
1158                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1159         else
1160                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1161
1162         qgroup_dirty(fs_info, qgroup);
1163
1164         /* Get all of the parent groups that contain this qgroup */
1165         list_for_each_entry(glist, &qgroup->groups, next_group) {
1166                 ret = ulist_add(tmp, glist->group->qgroupid,
1167                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1168                 if (ret < 0)
1169                         goto out;
1170         }
1171
1172         /* Iterate all of the parents and adjust their reference counts */
1173         ULIST_ITER_INIT(&uiter);
1174         while ((unode = ulist_next(tmp, &uiter))) {
1175                 qgroup = unode_aux_to_qgroup(unode);
1176                 qgroup->rfer += sign * num_bytes;
1177                 qgroup->rfer_cmpr += sign * num_bytes;
1178                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1179                 qgroup->excl += sign * num_bytes;
1180                 if (sign > 0)
1181                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1182                 else
1183                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1184                 qgroup->excl_cmpr += sign * num_bytes;
1185                 qgroup_dirty(fs_info, qgroup);
1186
1187                 /* Add any parents of the parents */
1188                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1189                         ret = ulist_add(tmp, glist->group->qgroupid,
1190                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1191                         if (ret < 0)
1192                                 goto out;
1193                 }
1194         }
1195         ret = 0;
1196 out:
1197         return ret;
1198 }
1199
1200
1201 /*
1202  * Quick path for updating qgroup with only excl refs.
1203  *
1204  * In that case, just update all parent will be enough.
1205  * Or we needs to do a full rescan.
1206  * Caller should also hold fs_info->qgroup_lock.
1207  *
1208  * Return 0 for quick update, return >0 for need to full rescan
1209  * and mark INCONSISTENT flag.
1210  * Return < 0 for other error.
1211  */
1212 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1213                                    struct ulist *tmp, u64 src, u64 dst,
1214                                    int sign)
1215 {
1216         struct btrfs_qgroup *qgroup;
1217         int ret = 1;
1218         int err = 0;
1219
1220         qgroup = find_qgroup_rb(fs_info, src);
1221         if (!qgroup)
1222                 goto out;
1223         if (qgroup->excl == qgroup->rfer) {
1224                 ret = 0;
1225                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1226                                                qgroup, sign);
1227                 if (err < 0) {
1228                         ret = err;
1229                         goto out;
1230                 }
1231         }
1232 out:
1233         if (ret)
1234                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1235         return ret;
1236 }
1237
1238 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1239                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1240 {
1241         struct btrfs_root *quota_root;
1242         struct btrfs_qgroup *parent;
1243         struct btrfs_qgroup *member;
1244         struct btrfs_qgroup_list *list;
1245         struct ulist *tmp;
1246         int ret = 0;
1247
1248         /* Check the level of src and dst first */
1249         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1250                 return -EINVAL;
1251
1252         tmp = ulist_alloc(GFP_KERNEL);
1253         if (!tmp)
1254                 return -ENOMEM;
1255
1256         mutex_lock(&fs_info->qgroup_ioctl_lock);
1257         quota_root = fs_info->quota_root;
1258         if (!quota_root) {
1259                 ret = -EINVAL;
1260                 goto out;
1261         }
1262         member = find_qgroup_rb(fs_info, src);
1263         parent = find_qgroup_rb(fs_info, dst);
1264         if (!member || !parent) {
1265                 ret = -EINVAL;
1266                 goto out;
1267         }
1268
1269         /* check if such qgroup relation exist firstly */
1270         list_for_each_entry(list, &member->groups, next_group) {
1271                 if (list->group == parent) {
1272                         ret = -EEXIST;
1273                         goto out;
1274                 }
1275         }
1276
1277         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1278         if (ret)
1279                 goto out;
1280
1281         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1282         if (ret) {
1283                 del_qgroup_relation_item(trans, quota_root, src, dst);
1284                 goto out;
1285         }
1286
1287         spin_lock(&fs_info->qgroup_lock);
1288         ret = add_relation_rb(fs_info, src, dst);
1289         if (ret < 0) {
1290                 spin_unlock(&fs_info->qgroup_lock);
1291                 goto out;
1292         }
1293         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1294         spin_unlock(&fs_info->qgroup_lock);
1295 out:
1296         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1297         ulist_free(tmp);
1298         return ret;
1299 }
1300
1301 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1302                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1303 {
1304         struct btrfs_root *quota_root;
1305         struct btrfs_qgroup *parent;
1306         struct btrfs_qgroup *member;
1307         struct btrfs_qgroup_list *list;
1308         struct ulist *tmp;
1309         int ret = 0;
1310         int err;
1311
1312         tmp = ulist_alloc(GFP_KERNEL);
1313         if (!tmp)
1314                 return -ENOMEM;
1315
1316         quota_root = fs_info->quota_root;
1317         if (!quota_root) {
1318                 ret = -EINVAL;
1319                 goto out;
1320         }
1321
1322         member = find_qgroup_rb(fs_info, src);
1323         parent = find_qgroup_rb(fs_info, dst);
1324         if (!member || !parent) {
1325                 ret = -EINVAL;
1326                 goto out;
1327         }
1328
1329         /* check if such qgroup relation exist firstly */
1330         list_for_each_entry(list, &member->groups, next_group) {
1331                 if (list->group == parent)
1332                         goto exist;
1333         }
1334         ret = -ENOENT;
1335         goto out;
1336 exist:
1337         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1338         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1339         if (err && !ret)
1340                 ret = err;
1341
1342         spin_lock(&fs_info->qgroup_lock);
1343         del_relation_rb(fs_info, src, dst);
1344         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1345         spin_unlock(&fs_info->qgroup_lock);
1346 out:
1347         ulist_free(tmp);
1348         return ret;
1349 }
1350
1351 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1352                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1353 {
1354         int ret = 0;
1355
1356         mutex_lock(&fs_info->qgroup_ioctl_lock);
1357         ret = __del_qgroup_relation(trans, fs_info, src, dst);
1358         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1359
1360         return ret;
1361 }
1362
1363 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1364                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1365 {
1366         struct btrfs_root *quota_root;
1367         struct btrfs_qgroup *qgroup;
1368         int ret = 0;
1369
1370         mutex_lock(&fs_info->qgroup_ioctl_lock);
1371         quota_root = fs_info->quota_root;
1372         if (!quota_root) {
1373                 ret = -EINVAL;
1374                 goto out;
1375         }
1376         qgroup = find_qgroup_rb(fs_info, qgroupid);
1377         if (qgroup) {
1378                 ret = -EEXIST;
1379                 goto out;
1380         }
1381
1382         ret = add_qgroup_item(trans, quota_root, qgroupid);
1383         if (ret)
1384                 goto out;
1385
1386         spin_lock(&fs_info->qgroup_lock);
1387         qgroup = add_qgroup_rb(fs_info, qgroupid);
1388         spin_unlock(&fs_info->qgroup_lock);
1389
1390         if (IS_ERR(qgroup))
1391                 ret = PTR_ERR(qgroup);
1392 out:
1393         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1394         return ret;
1395 }
1396
1397 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1398                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1399 {
1400         struct btrfs_root *quota_root;
1401         struct btrfs_qgroup *qgroup;
1402         struct btrfs_qgroup_list *list;
1403         int ret = 0;
1404
1405         mutex_lock(&fs_info->qgroup_ioctl_lock);
1406         quota_root = fs_info->quota_root;
1407         if (!quota_root) {
1408                 ret = -EINVAL;
1409                 goto out;
1410         }
1411
1412         qgroup = find_qgroup_rb(fs_info, qgroupid);
1413         if (!qgroup) {
1414                 ret = -ENOENT;
1415                 goto out;
1416         } else {
1417                 /* check if there are no children of this qgroup */
1418                 if (!list_empty(&qgroup->members)) {
1419                         ret = -EBUSY;
1420                         goto out;
1421                 }
1422         }
1423         ret = del_qgroup_item(trans, quota_root, qgroupid);
1424         if (ret && ret != -ENOENT)
1425                 goto out;
1426
1427         while (!list_empty(&qgroup->groups)) {
1428                 list = list_first_entry(&qgroup->groups,
1429                                         struct btrfs_qgroup_list, next_group);
1430                 ret = __del_qgroup_relation(trans, fs_info,
1431                                            qgroupid,
1432                                            list->group->qgroupid);
1433                 if (ret)
1434                         goto out;
1435         }
1436
1437         spin_lock(&fs_info->qgroup_lock);
1438         del_qgroup_rb(fs_info, qgroupid);
1439         spin_unlock(&fs_info->qgroup_lock);
1440 out:
1441         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1442         return ret;
1443 }
1444
1445 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1446                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1447                        struct btrfs_qgroup_limit *limit)
1448 {
1449         struct btrfs_root *quota_root;
1450         struct btrfs_qgroup *qgroup;
1451         int ret = 0;
1452         /* Sometimes we would want to clear the limit on this qgroup.
1453          * To meet this requirement, we treat the -1 as a special value
1454          * which tell kernel to clear the limit on this qgroup.
1455          */
1456         const u64 CLEAR_VALUE = -1;
1457
1458         mutex_lock(&fs_info->qgroup_ioctl_lock);
1459         quota_root = fs_info->quota_root;
1460         if (!quota_root) {
1461                 ret = -EINVAL;
1462                 goto out;
1463         }
1464
1465         qgroup = find_qgroup_rb(fs_info, qgroupid);
1466         if (!qgroup) {
1467                 ret = -ENOENT;
1468                 goto out;
1469         }
1470
1471         spin_lock(&fs_info->qgroup_lock);
1472         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1473                 if (limit->max_rfer == CLEAR_VALUE) {
1474                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1475                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1476                         qgroup->max_rfer = 0;
1477                 } else {
1478                         qgroup->max_rfer = limit->max_rfer;
1479                 }
1480         }
1481         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1482                 if (limit->max_excl == CLEAR_VALUE) {
1483                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1484                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1485                         qgroup->max_excl = 0;
1486                 } else {
1487                         qgroup->max_excl = limit->max_excl;
1488                 }
1489         }
1490         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1491                 if (limit->rsv_rfer == CLEAR_VALUE) {
1492                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1493                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1494                         qgroup->rsv_rfer = 0;
1495                 } else {
1496                         qgroup->rsv_rfer = limit->rsv_rfer;
1497                 }
1498         }
1499         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1500                 if (limit->rsv_excl == CLEAR_VALUE) {
1501                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1502                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1503                         qgroup->rsv_excl = 0;
1504                 } else {
1505                         qgroup->rsv_excl = limit->rsv_excl;
1506                 }
1507         }
1508         qgroup->lim_flags |= limit->flags;
1509
1510         spin_unlock(&fs_info->qgroup_lock);
1511
1512         ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1513         if (ret) {
1514                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1515                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1516                        qgroupid);
1517         }
1518
1519 out:
1520         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1521         return ret;
1522 }
1523
1524 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1525                                 struct btrfs_delayed_ref_root *delayed_refs,
1526                                 struct btrfs_qgroup_extent_record *record)
1527 {
1528         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1529         struct rb_node *parent_node = NULL;
1530         struct btrfs_qgroup_extent_record *entry;
1531         u64 bytenr = record->bytenr;
1532
1533         lockdep_assert_held(&delayed_refs->lock);
1534         trace_btrfs_qgroup_trace_extent(fs_info, record);
1535
1536         while (*p) {
1537                 parent_node = *p;
1538                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1539                                  node);
1540                 if (bytenr < entry->bytenr)
1541                         p = &(*p)->rb_left;
1542                 else if (bytenr > entry->bytenr)
1543                         p = &(*p)->rb_right;
1544                 else
1545                         return 1;
1546         }
1547
1548         rb_link_node(&record->node, parent_node, p);
1549         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1550         return 0;
1551 }
1552
1553 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1554                                    struct btrfs_qgroup_extent_record *qrecord)
1555 {
1556         struct ulist *old_root;
1557         u64 bytenr = qrecord->bytenr;
1558         int ret;
1559
1560         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1561         if (ret < 0) {
1562                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1563                 btrfs_warn(fs_info,
1564 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1565                         ret);
1566                 return 0;
1567         }
1568
1569         /*
1570          * Here we don't need to get the lock of
1571          * trans->transaction->delayed_refs, since inserted qrecord won't
1572          * be deleted, only qrecord->node may be modified (new qrecord insert)
1573          *
1574          * So modifying qrecord->old_roots is safe here
1575          */
1576         qrecord->old_roots = old_root;
1577         return 0;
1578 }
1579
1580 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1581                 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1582                 gfp_t gfp_flag)
1583 {
1584         struct btrfs_qgroup_extent_record *record;
1585         struct btrfs_delayed_ref_root *delayed_refs;
1586         int ret;
1587
1588         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1589             || bytenr == 0 || num_bytes == 0)
1590                 return 0;
1591         if (WARN_ON(trans == NULL))
1592                 return -EINVAL;
1593         record = kmalloc(sizeof(*record), gfp_flag);
1594         if (!record)
1595                 return -ENOMEM;
1596
1597         delayed_refs = &trans->transaction->delayed_refs;
1598         record->bytenr = bytenr;
1599         record->num_bytes = num_bytes;
1600         record->old_roots = NULL;
1601
1602         spin_lock(&delayed_refs->lock);
1603         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1604         spin_unlock(&delayed_refs->lock);
1605         if (ret > 0) {
1606                 kfree(record);
1607                 return 0;
1608         }
1609         return btrfs_qgroup_trace_extent_post(fs_info, record);
1610 }
1611
1612 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1613                                   struct btrfs_fs_info *fs_info,
1614                                   struct extent_buffer *eb)
1615 {
1616         int nr = btrfs_header_nritems(eb);
1617         int i, extent_type, ret;
1618         struct btrfs_key key;
1619         struct btrfs_file_extent_item *fi;
1620         u64 bytenr, num_bytes;
1621
1622         /* We can be called directly from walk_up_proc() */
1623         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1624                 return 0;
1625
1626         for (i = 0; i < nr; i++) {
1627                 btrfs_item_key_to_cpu(eb, &key, i);
1628
1629                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1630                         continue;
1631
1632                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1633                 /* filter out non qgroup-accountable extents  */
1634                 extent_type = btrfs_file_extent_type(eb, fi);
1635
1636                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1637                         continue;
1638
1639                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1640                 if (!bytenr)
1641                         continue;
1642
1643                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1644
1645                 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1646                                                 num_bytes, GFP_NOFS);
1647                 if (ret)
1648                         return ret;
1649         }
1650         cond_resched();
1651         return 0;
1652 }
1653
1654 /*
1655  * Walk up the tree from the bottom, freeing leaves and any interior
1656  * nodes which have had all slots visited. If a node (leaf or
1657  * interior) is freed, the node above it will have it's slot
1658  * incremented. The root node will never be freed.
1659  *
1660  * At the end of this function, we should have a path which has all
1661  * slots incremented to the next position for a search. If we need to
1662  * read a new node it will be NULL and the node above it will have the
1663  * correct slot selected for a later read.
1664  *
1665  * If we increment the root nodes slot counter past the number of
1666  * elements, 1 is returned to signal completion of the search.
1667  */
1668 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1669 {
1670         int level = 0;
1671         int nr, slot;
1672         struct extent_buffer *eb;
1673
1674         if (root_level == 0)
1675                 return 1;
1676
1677         while (level <= root_level) {
1678                 eb = path->nodes[level];
1679                 nr = btrfs_header_nritems(eb);
1680                 path->slots[level]++;
1681                 slot = path->slots[level];
1682                 if (slot >= nr || level == 0) {
1683                         /*
1684                          * Don't free the root -  we will detect this
1685                          * condition after our loop and return a
1686                          * positive value for caller to stop walking the tree.
1687                          */
1688                         if (level != root_level) {
1689                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1690                                 path->locks[level] = 0;
1691
1692                                 free_extent_buffer(eb);
1693                                 path->nodes[level] = NULL;
1694                                 path->slots[level] = 0;
1695                         }
1696                 } else {
1697                         /*
1698                          * We have a valid slot to walk back down
1699                          * from. Stop here so caller can process these
1700                          * new nodes.
1701                          */
1702                         break;
1703                 }
1704
1705                 level++;
1706         }
1707
1708         eb = path->nodes[root_level];
1709         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1710                 return 1;
1711
1712         return 0;
1713 }
1714
1715 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1716                                struct btrfs_root *root,
1717                                struct extent_buffer *root_eb,
1718                                u64 root_gen, int root_level)
1719 {
1720         struct btrfs_fs_info *fs_info = root->fs_info;
1721         int ret = 0;
1722         int level;
1723         struct extent_buffer *eb = root_eb;
1724         struct btrfs_path *path = NULL;
1725
1726         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1727         BUG_ON(root_eb == NULL);
1728
1729         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1730                 return 0;
1731
1732         if (!extent_buffer_uptodate(root_eb)) {
1733                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
1734                 if (ret)
1735                         goto out;
1736         }
1737
1738         if (root_level == 0) {
1739                 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1740                 goto out;
1741         }
1742
1743         path = btrfs_alloc_path();
1744         if (!path)
1745                 return -ENOMEM;
1746
1747         /*
1748          * Walk down the tree.  Missing extent blocks are filled in as
1749          * we go. Metadata is accounted every time we read a new
1750          * extent block.
1751          *
1752          * When we reach a leaf, we account for file extent items in it,
1753          * walk back up the tree (adjusting slot pointers as we go)
1754          * and restart the search process.
1755          */
1756         extent_buffer_get(root_eb); /* For path */
1757         path->nodes[root_level] = root_eb;
1758         path->slots[root_level] = 0;
1759         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1760 walk_down:
1761         level = root_level;
1762         while (level >= 0) {
1763                 if (path->nodes[level] == NULL) {
1764                         struct btrfs_key first_key;
1765                         int parent_slot;
1766                         u64 child_gen;
1767                         u64 child_bytenr;
1768
1769                         /*
1770                          * We need to get child blockptr/gen from parent before
1771                          * we can read it.
1772                           */
1773                         eb = path->nodes[level + 1];
1774                         parent_slot = path->slots[level + 1];
1775                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1776                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1777                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1778
1779                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1780                                              level, &first_key);
1781                         if (IS_ERR(eb)) {
1782                                 ret = PTR_ERR(eb);
1783                                 goto out;
1784                         } else if (!extent_buffer_uptodate(eb)) {
1785                                 free_extent_buffer(eb);
1786                                 ret = -EIO;
1787                                 goto out;
1788                         }
1789
1790                         path->nodes[level] = eb;
1791                         path->slots[level] = 0;
1792
1793                         btrfs_tree_read_lock(eb);
1794                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1795                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1796
1797                         ret = btrfs_qgroup_trace_extent(trans, fs_info,
1798                                                         child_bytenr,
1799                                                         fs_info->nodesize,
1800                                                         GFP_NOFS);
1801                         if (ret)
1802                                 goto out;
1803                 }
1804
1805                 if (level == 0) {
1806                         ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1807                                                            path->nodes[level]);
1808                         if (ret)
1809                                 goto out;
1810
1811                         /* Nonzero return here means we completed our search */
1812                         ret = adjust_slots_upwards(path, root_level);
1813                         if (ret)
1814                                 break;
1815
1816                         /* Restart search with new slots */
1817                         goto walk_down;
1818                 }
1819
1820                 level--;
1821         }
1822
1823         ret = 0;
1824 out:
1825         btrfs_free_path(path);
1826
1827         return ret;
1828 }
1829
1830 #define UPDATE_NEW      0
1831 #define UPDATE_OLD      1
1832 /*
1833  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1834  */
1835 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1836                                 struct ulist *roots, struct ulist *tmp,
1837                                 struct ulist *qgroups, u64 seq, int update_old)
1838 {
1839         struct ulist_node *unode;
1840         struct ulist_iterator uiter;
1841         struct ulist_node *tmp_unode;
1842         struct ulist_iterator tmp_uiter;
1843         struct btrfs_qgroup *qg;
1844         int ret = 0;
1845
1846         if (!roots)
1847                 return 0;
1848         ULIST_ITER_INIT(&uiter);
1849         while ((unode = ulist_next(roots, &uiter))) {
1850                 qg = find_qgroup_rb(fs_info, unode->val);
1851                 if (!qg)
1852                         continue;
1853
1854                 ulist_reinit(tmp);
1855                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1856                                 GFP_ATOMIC);
1857                 if (ret < 0)
1858                         return ret;
1859                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1860                 if (ret < 0)
1861                         return ret;
1862                 ULIST_ITER_INIT(&tmp_uiter);
1863                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1864                         struct btrfs_qgroup_list *glist;
1865
1866                         qg = unode_aux_to_qgroup(tmp_unode);
1867                         if (update_old)
1868                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1869                         else
1870                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1871                         list_for_each_entry(glist, &qg->groups, next_group) {
1872                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1873                                                 qgroup_to_aux(glist->group),
1874                                                 GFP_ATOMIC);
1875                                 if (ret < 0)
1876                                         return ret;
1877                                 ret = ulist_add(tmp, glist->group->qgroupid,
1878                                                 qgroup_to_aux(glist->group),
1879                                                 GFP_ATOMIC);
1880                                 if (ret < 0)
1881                                         return ret;
1882                         }
1883                 }
1884         }
1885         return 0;
1886 }
1887
1888 /*
1889  * Update qgroup rfer/excl counters.
1890  * Rfer update is easy, codes can explain themselves.
1891  *
1892  * Excl update is tricky, the update is split into 2 part.
1893  * Part 1: Possible exclusive <-> sharing detect:
1894  *      |       A       |       !A      |
1895  *  -------------------------------------
1896  *  B   |       *       |       -       |
1897  *  -------------------------------------
1898  *  !B  |       +       |       **      |
1899  *  -------------------------------------
1900  *
1901  * Conditions:
1902  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1903  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1904  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1905  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1906  *
1907  * Results:
1908  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1909  * *: Definitely not changed.           **: Possible unchanged.
1910  *
1911  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1912  *
1913  * To make the logic clear, we first use condition A and B to split
1914  * combination into 4 results.
1915  *
1916  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1917  * only on variant maybe 0.
1918  *
1919  * Lastly, check result **, since there are 2 variants maybe 0, split them
1920  * again(2x2).
1921  * But this time we don't need to consider other things, the codes and logic
1922  * is easy to understand now.
1923  */
1924 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1925                                   struct ulist *qgroups,
1926                                   u64 nr_old_roots,
1927                                   u64 nr_new_roots,
1928                                   u64 num_bytes, u64 seq)
1929 {
1930         struct ulist_node *unode;
1931         struct ulist_iterator uiter;
1932         struct btrfs_qgroup *qg;
1933         u64 cur_new_count, cur_old_count;
1934
1935         ULIST_ITER_INIT(&uiter);
1936         while ((unode = ulist_next(qgroups, &uiter))) {
1937                 bool dirty = false;
1938
1939                 qg = unode_aux_to_qgroup(unode);
1940                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1941                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1942
1943                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
1944                                              cur_new_count);
1945
1946                 /* Rfer update part */
1947                 if (cur_old_count == 0 && cur_new_count > 0) {
1948                         qg->rfer += num_bytes;
1949                         qg->rfer_cmpr += num_bytes;
1950                         dirty = true;
1951                 }
1952                 if (cur_old_count > 0 && cur_new_count == 0) {
1953                         qg->rfer -= num_bytes;
1954                         qg->rfer_cmpr -= num_bytes;
1955                         dirty = true;
1956                 }
1957
1958                 /* Excl update part */
1959                 /* Exclusive/none -> shared case */
1960                 if (cur_old_count == nr_old_roots &&
1961                     cur_new_count < nr_new_roots) {
1962                         /* Exclusive -> shared */
1963                         if (cur_old_count != 0) {
1964                                 qg->excl -= num_bytes;
1965                                 qg->excl_cmpr -= num_bytes;
1966                                 dirty = true;
1967                         }
1968                 }
1969
1970                 /* Shared -> exclusive/none case */
1971                 if (cur_old_count < nr_old_roots &&
1972                     cur_new_count == nr_new_roots) {
1973                         /* Shared->exclusive */
1974                         if (cur_new_count != 0) {
1975                                 qg->excl += num_bytes;
1976                                 qg->excl_cmpr += num_bytes;
1977                                 dirty = true;
1978                         }
1979                 }
1980
1981                 /* Exclusive/none -> exclusive/none case */
1982                 if (cur_old_count == nr_old_roots &&
1983                     cur_new_count == nr_new_roots) {
1984                         if (cur_old_count == 0) {
1985                                 /* None -> exclusive/none */
1986
1987                                 if (cur_new_count != 0) {
1988                                         /* None -> exclusive */
1989                                         qg->excl += num_bytes;
1990                                         qg->excl_cmpr += num_bytes;
1991                                         dirty = true;
1992                                 }
1993                                 /* None -> none, nothing changed */
1994                         } else {
1995                                 /* Exclusive -> exclusive/none */
1996
1997                                 if (cur_new_count == 0) {
1998                                         /* Exclusive -> none */
1999                                         qg->excl -= num_bytes;
2000                                         qg->excl_cmpr -= num_bytes;
2001                                         dirty = true;
2002                                 }
2003                                 /* Exclusive -> exclusive, nothing changed */
2004                         }
2005                 }
2006
2007                 if (dirty)
2008                         qgroup_dirty(fs_info, qg);
2009         }
2010         return 0;
2011 }
2012
2013 /*
2014  * Check if the @roots potentially is a list of fs tree roots
2015  *
2016  * Return 0 for definitely not a fs/subvol tree roots ulist
2017  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2018  *          one as well)
2019  */
2020 static int maybe_fs_roots(struct ulist *roots)
2021 {
2022         struct ulist_node *unode;
2023         struct ulist_iterator uiter;
2024
2025         /* Empty one, still possible for fs roots */
2026         if (!roots || roots->nnodes == 0)
2027                 return 1;
2028
2029         ULIST_ITER_INIT(&uiter);
2030         unode = ulist_next(roots, &uiter);
2031         if (!unode)
2032                 return 1;
2033
2034         /*
2035          * If it contains fs tree roots, then it must belong to fs/subvol
2036          * trees.
2037          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2038          */
2039         return is_fstree(unode->val);
2040 }
2041
2042 int
2043 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
2044                             struct btrfs_fs_info *fs_info,
2045                             u64 bytenr, u64 num_bytes,
2046                             struct ulist *old_roots, struct ulist *new_roots)
2047 {
2048         struct ulist *qgroups = NULL;
2049         struct ulist *tmp = NULL;
2050         u64 seq;
2051         u64 nr_new_roots = 0;
2052         u64 nr_old_roots = 0;
2053         int ret = 0;
2054
2055         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2056                 return 0;
2057
2058         if (new_roots) {
2059                 if (!maybe_fs_roots(new_roots))
2060                         goto out_free;
2061                 nr_new_roots = new_roots->nnodes;
2062         }
2063         if (old_roots) {
2064                 if (!maybe_fs_roots(old_roots))
2065                         goto out_free;
2066                 nr_old_roots = old_roots->nnodes;
2067         }
2068
2069         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2070         if (nr_old_roots == 0 && nr_new_roots == 0)
2071                 goto out_free;
2072
2073         BUG_ON(!fs_info->quota_root);
2074
2075         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2076                                         num_bytes, nr_old_roots, nr_new_roots);
2077
2078         qgroups = ulist_alloc(GFP_NOFS);
2079         if (!qgroups) {
2080                 ret = -ENOMEM;
2081                 goto out_free;
2082         }
2083         tmp = ulist_alloc(GFP_NOFS);
2084         if (!tmp) {
2085                 ret = -ENOMEM;
2086                 goto out_free;
2087         }
2088
2089         mutex_lock(&fs_info->qgroup_rescan_lock);
2090         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2091                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2092                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2093                         ret = 0;
2094                         goto out_free;
2095                 }
2096         }
2097         mutex_unlock(&fs_info->qgroup_rescan_lock);
2098
2099         spin_lock(&fs_info->qgroup_lock);
2100         seq = fs_info->qgroup_seq;
2101
2102         /* Update old refcnts using old_roots */
2103         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2104                                    UPDATE_OLD);
2105         if (ret < 0)
2106                 goto out;
2107
2108         /* Update new refcnts using new_roots */
2109         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2110                                    UPDATE_NEW);
2111         if (ret < 0)
2112                 goto out;
2113
2114         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2115                                num_bytes, seq);
2116
2117         /*
2118          * Bump qgroup_seq to avoid seq overlap
2119          */
2120         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2121 out:
2122         spin_unlock(&fs_info->qgroup_lock);
2123 out_free:
2124         ulist_free(tmp);
2125         ulist_free(qgroups);
2126         ulist_free(old_roots);
2127         ulist_free(new_roots);
2128         return ret;
2129 }
2130
2131 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2132 {
2133         struct btrfs_fs_info *fs_info = trans->fs_info;
2134         struct btrfs_qgroup_extent_record *record;
2135         struct btrfs_delayed_ref_root *delayed_refs;
2136         struct ulist *new_roots = NULL;
2137         struct rb_node *node;
2138         u64 qgroup_to_skip;
2139         int ret = 0;
2140
2141         delayed_refs = &trans->transaction->delayed_refs;
2142         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2143         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2144                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2145                                   node);
2146
2147                 trace_btrfs_qgroup_account_extents(fs_info, record);
2148
2149                 if (!ret) {
2150                         /*
2151                          * Old roots should be searched when inserting qgroup
2152                          * extent record
2153                          */
2154                         if (WARN_ON(!record->old_roots)) {
2155                                 /* Search commit root to find old_roots */
2156                                 ret = btrfs_find_all_roots(NULL, fs_info,
2157                                                 record->bytenr, 0,
2158                                                 &record->old_roots, false);
2159                                 if (ret < 0)
2160                                         goto cleanup;
2161                         }
2162
2163                         /*
2164                          * Use SEQ_LAST as time_seq to do special search, which
2165                          * doesn't lock tree or delayed_refs and search current
2166                          * root. It's safe inside commit_transaction().
2167                          */
2168                         ret = btrfs_find_all_roots(trans, fs_info,
2169                                 record->bytenr, SEQ_LAST, &new_roots, false);
2170                         if (ret < 0)
2171                                 goto cleanup;
2172                         if (qgroup_to_skip) {
2173                                 ulist_del(new_roots, qgroup_to_skip, 0);
2174                                 ulist_del(record->old_roots, qgroup_to_skip,
2175                                           0);
2176                         }
2177                         ret = btrfs_qgroup_account_extent(trans, fs_info,
2178                                         record->bytenr, record->num_bytes,
2179                                         record->old_roots, new_roots);
2180                         record->old_roots = NULL;
2181                         new_roots = NULL;
2182                 }
2183 cleanup:
2184                 ulist_free(record->old_roots);
2185                 ulist_free(new_roots);
2186                 new_roots = NULL;
2187                 rb_erase(node, &delayed_refs->dirty_extent_root);
2188                 kfree(record);
2189
2190         }
2191         return ret;
2192 }
2193
2194 /*
2195  * called from commit_transaction. Writes all changed qgroups to disk.
2196  */
2197 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2198                       struct btrfs_fs_info *fs_info)
2199 {
2200         struct btrfs_root *quota_root = fs_info->quota_root;
2201         int ret = 0;
2202
2203         if (!quota_root)
2204                 return ret;
2205
2206         spin_lock(&fs_info->qgroup_lock);
2207         while (!list_empty(&fs_info->dirty_qgroups)) {
2208                 struct btrfs_qgroup *qgroup;
2209                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2210                                           struct btrfs_qgroup, dirty);
2211                 list_del_init(&qgroup->dirty);
2212                 spin_unlock(&fs_info->qgroup_lock);
2213                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2214                 if (ret)
2215                         fs_info->qgroup_flags |=
2216                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2217                 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2218                 if (ret)
2219                         fs_info->qgroup_flags |=
2220                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2221                 spin_lock(&fs_info->qgroup_lock);
2222         }
2223         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2224                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2225         else
2226                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2227         spin_unlock(&fs_info->qgroup_lock);
2228
2229         ret = update_qgroup_status_item(trans, fs_info, quota_root);
2230         if (ret)
2231                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2232
2233         return ret;
2234 }
2235
2236 /*
2237  * Copy the accounting information between qgroups. This is necessary
2238  * when a snapshot or a subvolume is created. Throwing an error will
2239  * cause a transaction abort so we take extra care here to only error
2240  * when a readonly fs is a reasonable outcome.
2241  */
2242 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2243                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2244                          struct btrfs_qgroup_inherit *inherit)
2245 {
2246         int ret = 0;
2247         int i;
2248         u64 *i_qgroups;
2249         struct btrfs_root *quota_root = fs_info->quota_root;
2250         struct btrfs_qgroup *srcgroup;
2251         struct btrfs_qgroup *dstgroup;
2252         u32 level_size = 0;
2253         u64 nums;
2254
2255         mutex_lock(&fs_info->qgroup_ioctl_lock);
2256         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2257                 goto out;
2258
2259         if (!quota_root) {
2260                 ret = -EINVAL;
2261                 goto out;
2262         }
2263
2264         if (inherit) {
2265                 i_qgroups = (u64 *)(inherit + 1);
2266                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2267                        2 * inherit->num_excl_copies;
2268                 for (i = 0; i < nums; ++i) {
2269                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2270
2271                         /*
2272                          * Zero out invalid groups so we can ignore
2273                          * them later.
2274                          */
2275                         if (!srcgroup ||
2276                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2277                                 *i_qgroups = 0ULL;
2278
2279                         ++i_qgroups;
2280                 }
2281         }
2282
2283         /*
2284          * create a tracking group for the subvol itself
2285          */
2286         ret = add_qgroup_item(trans, quota_root, objectid);
2287         if (ret)
2288                 goto out;
2289
2290         if (srcid) {
2291                 struct btrfs_root *srcroot;
2292                 struct btrfs_key srckey;
2293
2294                 srckey.objectid = srcid;
2295                 srckey.type = BTRFS_ROOT_ITEM_KEY;
2296                 srckey.offset = (u64)-1;
2297                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2298                 if (IS_ERR(srcroot)) {
2299                         ret = PTR_ERR(srcroot);
2300                         goto out;
2301                 }
2302
2303                 level_size = fs_info->nodesize;
2304         }
2305
2306         /*
2307          * add qgroup to all inherited groups
2308          */
2309         if (inherit) {
2310                 i_qgroups = (u64 *)(inherit + 1);
2311                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2312                         if (*i_qgroups == 0)
2313                                 continue;
2314                         ret = add_qgroup_relation_item(trans, quota_root,
2315                                                        objectid, *i_qgroups);
2316                         if (ret && ret != -EEXIST)
2317                                 goto out;
2318                         ret = add_qgroup_relation_item(trans, quota_root,
2319                                                        *i_qgroups, objectid);
2320                         if (ret && ret != -EEXIST)
2321                                 goto out;
2322                 }
2323                 ret = 0;
2324         }
2325
2326
2327         spin_lock(&fs_info->qgroup_lock);
2328
2329         dstgroup = add_qgroup_rb(fs_info, objectid);
2330         if (IS_ERR(dstgroup)) {
2331                 ret = PTR_ERR(dstgroup);
2332                 goto unlock;
2333         }
2334
2335         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2336                 dstgroup->lim_flags = inherit->lim.flags;
2337                 dstgroup->max_rfer = inherit->lim.max_rfer;
2338                 dstgroup->max_excl = inherit->lim.max_excl;
2339                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2340                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2341
2342                 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2343                 if (ret) {
2344                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2345                         btrfs_info(fs_info,
2346                                    "unable to update quota limit for %llu",
2347                                    dstgroup->qgroupid);
2348                         goto unlock;
2349                 }
2350         }
2351
2352         if (srcid) {
2353                 srcgroup = find_qgroup_rb(fs_info, srcid);
2354                 if (!srcgroup)
2355                         goto unlock;
2356
2357                 /*
2358                  * We call inherit after we clone the root in order to make sure
2359                  * our counts don't go crazy, so at this point the only
2360                  * difference between the two roots should be the root node.
2361                  */
2362                 dstgroup->rfer = srcgroup->rfer;
2363                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2364                 dstgroup->excl = level_size;
2365                 dstgroup->excl_cmpr = level_size;
2366                 srcgroup->excl = level_size;
2367                 srcgroup->excl_cmpr = level_size;
2368
2369                 /* inherit the limit info */
2370                 dstgroup->lim_flags = srcgroup->lim_flags;
2371                 dstgroup->max_rfer = srcgroup->max_rfer;
2372                 dstgroup->max_excl = srcgroup->max_excl;
2373                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2374                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2375
2376                 qgroup_dirty(fs_info, dstgroup);
2377                 qgroup_dirty(fs_info, srcgroup);
2378         }
2379
2380         if (!inherit)
2381                 goto unlock;
2382
2383         i_qgroups = (u64 *)(inherit + 1);
2384         for (i = 0; i < inherit->num_qgroups; ++i) {
2385                 if (*i_qgroups) {
2386                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2387                         if (ret)
2388                                 goto unlock;
2389                 }
2390                 ++i_qgroups;
2391         }
2392
2393         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2394                 struct btrfs_qgroup *src;
2395                 struct btrfs_qgroup *dst;
2396
2397                 if (!i_qgroups[0] || !i_qgroups[1])
2398                         continue;
2399
2400                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2401                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2402
2403                 if (!src || !dst) {
2404                         ret = -EINVAL;
2405                         goto unlock;
2406                 }
2407
2408                 dst->rfer = src->rfer - level_size;
2409                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2410         }
2411         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2412                 struct btrfs_qgroup *src;
2413                 struct btrfs_qgroup *dst;
2414
2415                 if (!i_qgroups[0] || !i_qgroups[1])
2416                         continue;
2417
2418                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2419                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2420
2421                 if (!src || !dst) {
2422                         ret = -EINVAL;
2423                         goto unlock;
2424                 }
2425
2426                 dst->excl = src->excl + level_size;
2427                 dst->excl_cmpr = src->excl_cmpr + level_size;
2428         }
2429
2430 unlock:
2431         spin_unlock(&fs_info->qgroup_lock);
2432 out:
2433         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2434         return ret;
2435 }
2436
2437 /*
2438  * Two limits to commit transaction in advance.
2439  *
2440  * For RATIO, it will be 1/RATIO of the remaining limit
2441  * (excluding data and prealloc meta) as threshold.
2442  * For SIZE, it will be in byte unit as threshold.
2443  */
2444 #define QGROUP_PERTRANS_RATIO           32
2445 #define QGROUP_PERTRANS_SIZE            SZ_32M
2446 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2447                                 const struct btrfs_qgroup *qg, u64 num_bytes)
2448 {
2449         u64 limit;
2450         u64 threshold;
2451
2452         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2453             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2454                 return false;
2455
2456         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2457             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2458                 return false;
2459
2460         /*
2461          * Even if we passed the check, it's better to check if reservation
2462          * for meta_pertrans is pushing us near limit.
2463          * If there is too much pertrans reservation or it's near the limit,
2464          * let's try commit transaction to free some, using transaction_kthread
2465          */
2466         if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2467                               BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2468                 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2469                         limit = qg->max_excl;
2470                 else
2471                         limit = qg->max_rfer;
2472                 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2473                             qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2474                             QGROUP_PERTRANS_RATIO;
2475                 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2476
2477                 /*
2478                  * Use transaction_kthread to commit transaction, so we no
2479                  * longer need to bother nested transaction nor lock context.
2480                  */
2481                 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2482                         btrfs_commit_transaction_locksafe(fs_info);
2483         }
2484
2485         return true;
2486 }
2487
2488 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2489                           enum btrfs_qgroup_rsv_type type)
2490 {
2491         struct btrfs_root *quota_root;
2492         struct btrfs_qgroup *qgroup;
2493         struct btrfs_fs_info *fs_info = root->fs_info;
2494         u64 ref_root = root->root_key.objectid;
2495         int ret = 0;
2496         struct ulist_node *unode;
2497         struct ulist_iterator uiter;
2498
2499         if (!is_fstree(ref_root))
2500                 return 0;
2501
2502         if (num_bytes == 0)
2503                 return 0;
2504
2505         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2506             capable(CAP_SYS_RESOURCE))
2507                 enforce = false;
2508
2509         spin_lock(&fs_info->qgroup_lock);
2510         quota_root = fs_info->quota_root;
2511         if (!quota_root)
2512                 goto out;
2513
2514         qgroup = find_qgroup_rb(fs_info, ref_root);
2515         if (!qgroup)
2516                 goto out;
2517
2518         /*
2519          * in a first step, we check all affected qgroups if any limits would
2520          * be exceeded
2521          */
2522         ulist_reinit(fs_info->qgroup_ulist);
2523         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2524                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2525         if (ret < 0)
2526                 goto out;
2527         ULIST_ITER_INIT(&uiter);
2528         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2529                 struct btrfs_qgroup *qg;
2530                 struct btrfs_qgroup_list *glist;
2531
2532                 qg = unode_aux_to_qgroup(unode);
2533
2534                 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2535                         ret = -EDQUOT;
2536                         goto out;
2537                 }
2538
2539                 list_for_each_entry(glist, &qg->groups, next_group) {
2540                         ret = ulist_add(fs_info->qgroup_ulist,
2541                                         glist->group->qgroupid,
2542                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2543                         if (ret < 0)
2544                                 goto out;
2545                 }
2546         }
2547         ret = 0;
2548         /*
2549          * no limits exceeded, now record the reservation into all qgroups
2550          */
2551         ULIST_ITER_INIT(&uiter);
2552         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2553                 struct btrfs_qgroup *qg;
2554
2555                 qg = unode_aux_to_qgroup(unode);
2556
2557                 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2558                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2559         }
2560
2561 out:
2562         spin_unlock(&fs_info->qgroup_lock);
2563         return ret;
2564 }
2565
2566 /*
2567  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2568  * qgroup).
2569  *
2570  * Will handle all higher level qgroup too.
2571  *
2572  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2573  * This special case is only used for META_PERTRANS type.
2574  */
2575 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2576                                u64 ref_root, u64 num_bytes,
2577                                enum btrfs_qgroup_rsv_type type)
2578 {
2579         struct btrfs_root *quota_root;
2580         struct btrfs_qgroup *qgroup;
2581         struct ulist_node *unode;
2582         struct ulist_iterator uiter;
2583         int ret = 0;
2584
2585         if (!is_fstree(ref_root))
2586                 return;
2587
2588         if (num_bytes == 0)
2589                 return;
2590
2591         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2592                 WARN(1, "%s: Invalid type to free", __func__);
2593                 return;
2594         }
2595         spin_lock(&fs_info->qgroup_lock);
2596
2597         quota_root = fs_info->quota_root;
2598         if (!quota_root)
2599                 goto out;
2600
2601         qgroup = find_qgroup_rb(fs_info, ref_root);
2602         if (!qgroup)
2603                 goto out;
2604
2605         if (num_bytes == (u64)-1)
2606                 /*
2607                  * We're freeing all pertrans rsv, get reserved value from
2608                  * level 0 qgroup as real num_bytes to free.
2609                  */
2610                 num_bytes = qgroup->rsv.values[type];
2611
2612         ulist_reinit(fs_info->qgroup_ulist);
2613         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2614                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2615         if (ret < 0)
2616                 goto out;
2617         ULIST_ITER_INIT(&uiter);
2618         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2619                 struct btrfs_qgroup *qg;
2620                 struct btrfs_qgroup_list *glist;
2621
2622                 qg = unode_aux_to_qgroup(unode);
2623
2624                 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2625                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2626
2627                 list_for_each_entry(glist, &qg->groups, next_group) {
2628                         ret = ulist_add(fs_info->qgroup_ulist,
2629                                         glist->group->qgroupid,
2630                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2631                         if (ret < 0)
2632                                 goto out;
2633                 }
2634         }
2635
2636 out:
2637         spin_unlock(&fs_info->qgroup_lock);
2638 }
2639
2640 /*
2641  * Check if the leaf is the last leaf. Which means all node pointers
2642  * are at their last position.
2643  */
2644 static bool is_last_leaf(struct btrfs_path *path)
2645 {
2646         int i;
2647
2648         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2649                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2650                         return false;
2651         }
2652         return true;
2653 }
2654
2655 /*
2656  * returns < 0 on error, 0 when more leafs are to be scanned.
2657  * returns 1 when done.
2658  */
2659 static int
2660 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2661                    struct btrfs_trans_handle *trans)
2662 {
2663         struct btrfs_key found;
2664         struct extent_buffer *scratch_leaf = NULL;
2665         struct ulist *roots = NULL;
2666         u64 num_bytes;
2667         bool done;
2668         int slot;
2669         int ret;
2670
2671         mutex_lock(&fs_info->qgroup_rescan_lock);
2672         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2673                                          &fs_info->qgroup_rescan_progress,
2674                                          path, 1, 0);
2675
2676         btrfs_debug(fs_info,
2677                 "current progress key (%llu %u %llu), search_slot ret %d",
2678                 fs_info->qgroup_rescan_progress.objectid,
2679                 fs_info->qgroup_rescan_progress.type,
2680                 fs_info->qgroup_rescan_progress.offset, ret);
2681
2682         if (ret) {
2683                 /*
2684                  * The rescan is about to end, we will not be scanning any
2685                  * further blocks. We cannot unset the RESCAN flag here, because
2686                  * we want to commit the transaction if everything went well.
2687                  * To make the live accounting work in this phase, we set our
2688                  * scan progress pointer such that every real extent objectid
2689                  * will be smaller.
2690                  */
2691                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2692                 btrfs_release_path(path);
2693                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2694                 return ret;
2695         }
2696         done = is_last_leaf(path);
2697
2698         btrfs_item_key_to_cpu(path->nodes[0], &found,
2699                               btrfs_header_nritems(path->nodes[0]) - 1);
2700         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2701
2702         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2703         if (!scratch_leaf) {
2704                 ret = -ENOMEM;
2705                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2706                 goto out;
2707         }
2708         extent_buffer_get(scratch_leaf);
2709         btrfs_tree_read_lock(scratch_leaf);
2710         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2711         slot = path->slots[0];
2712         btrfs_release_path(path);
2713         mutex_unlock(&fs_info->qgroup_rescan_lock);
2714
2715         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2716                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2717                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2718                     found.type != BTRFS_METADATA_ITEM_KEY)
2719                         continue;
2720                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2721                         num_bytes = fs_info->nodesize;
2722                 else
2723                         num_bytes = found.offset;
2724
2725                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2726                                            &roots, false);
2727                 if (ret < 0)
2728                         goto out;
2729                 /* For rescan, just pass old_roots as NULL */
2730                 ret = btrfs_qgroup_account_extent(trans, fs_info,
2731                                 found.objectid, num_bytes, NULL, roots);
2732                 if (ret < 0)
2733                         goto out;
2734         }
2735 out:
2736         if (scratch_leaf) {
2737                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2738                 free_extent_buffer(scratch_leaf);
2739         }
2740
2741         if (done && !ret) {
2742                 ret = 1;
2743                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2744         }
2745         return ret;
2746 }
2747
2748 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2749 {
2750         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2751                                                      qgroup_rescan_work);
2752         struct btrfs_path *path;
2753         struct btrfs_trans_handle *trans = NULL;
2754         int err = -ENOMEM;
2755         int ret = 0;
2756
2757         path = btrfs_alloc_path();
2758         if (!path)
2759                 goto out;
2760         /*
2761          * Rescan should only search for commit root, and any later difference
2762          * should be recorded by qgroup
2763          */
2764         path->search_commit_root = 1;
2765         path->skip_locking = 1;
2766
2767         err = 0;
2768         while (!err && !btrfs_fs_closing(fs_info)) {
2769                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2770                 if (IS_ERR(trans)) {
2771                         err = PTR_ERR(trans);
2772                         break;
2773                 }
2774                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2775                         err = -EINTR;
2776                 } else {
2777                         err = qgroup_rescan_leaf(fs_info, path, trans);
2778                 }
2779                 if (err > 0)
2780                         btrfs_commit_transaction(trans);
2781                 else
2782                         btrfs_end_transaction(trans);
2783         }
2784
2785 out:
2786         btrfs_free_path(path);
2787
2788         mutex_lock(&fs_info->qgroup_rescan_lock);
2789         if (!btrfs_fs_closing(fs_info))
2790                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2791
2792         if (err > 0 &&
2793             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2794                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2795         } else if (err < 0) {
2796                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2797         }
2798         mutex_unlock(&fs_info->qgroup_rescan_lock);
2799
2800         /*
2801          * only update status, since the previous part has already updated the
2802          * qgroup info.
2803          */
2804         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2805         if (IS_ERR(trans)) {
2806                 err = PTR_ERR(trans);
2807                 btrfs_err(fs_info,
2808                           "fail to start transaction for status update: %d",
2809                           err);
2810                 goto done;
2811         }
2812         ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2813         if (ret < 0) {
2814                 err = ret;
2815                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2816         }
2817         btrfs_end_transaction(trans);
2818
2819         if (btrfs_fs_closing(fs_info)) {
2820                 btrfs_info(fs_info, "qgroup scan paused");
2821         } else if (err >= 0) {
2822                 btrfs_info(fs_info, "qgroup scan completed%s",
2823                         err > 0 ? " (inconsistency flag cleared)" : "");
2824         } else {
2825                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2826         }
2827
2828 done:
2829         mutex_lock(&fs_info->qgroup_rescan_lock);
2830         fs_info->qgroup_rescan_running = false;
2831         mutex_unlock(&fs_info->qgroup_rescan_lock);
2832         complete_all(&fs_info->qgroup_rescan_completion);
2833 }
2834
2835 /*
2836  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2837  * memory required for the rescan context.
2838  */
2839 static int
2840 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2841                    int init_flags)
2842 {
2843         int ret = 0;
2844
2845         if (!init_flags) {
2846                 /* we're resuming qgroup rescan at mount time */
2847                 if (!(fs_info->qgroup_flags &
2848                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
2849                         btrfs_warn(fs_info,
2850                         "qgroup rescan init failed, qgroup is not enabled");
2851                         ret = -EINVAL;
2852                 } else if (!(fs_info->qgroup_flags &
2853                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
2854                         btrfs_warn(fs_info,
2855                         "qgroup rescan init failed, qgroup rescan is not queued");
2856                         ret = -EINVAL;
2857                 }
2858
2859                 if (ret)
2860                         return ret;
2861         }
2862
2863         mutex_lock(&fs_info->qgroup_rescan_lock);
2864         spin_lock(&fs_info->qgroup_lock);
2865
2866         if (init_flags) {
2867                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2868                         btrfs_warn(fs_info,
2869                                    "qgroup rescan is already in progress");
2870                         ret = -EINPROGRESS;
2871                 } else if (!(fs_info->qgroup_flags &
2872                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
2873                         btrfs_warn(fs_info,
2874                         "qgroup rescan init failed, qgroup is not enabled");
2875                         ret = -EINVAL;
2876                 }
2877
2878                 if (ret) {
2879                         spin_unlock(&fs_info->qgroup_lock);
2880                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2881                         return ret;
2882                 }
2883                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2884         }
2885
2886         memset(&fs_info->qgroup_rescan_progress, 0,
2887                 sizeof(fs_info->qgroup_rescan_progress));
2888         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2889         init_completion(&fs_info->qgroup_rescan_completion);
2890         fs_info->qgroup_rescan_running = true;
2891
2892         spin_unlock(&fs_info->qgroup_lock);
2893         mutex_unlock(&fs_info->qgroup_rescan_lock);
2894
2895         memset(&fs_info->qgroup_rescan_work, 0,
2896                sizeof(fs_info->qgroup_rescan_work));
2897         btrfs_init_work(&fs_info->qgroup_rescan_work,
2898                         btrfs_qgroup_rescan_helper,
2899                         btrfs_qgroup_rescan_worker, NULL, NULL);
2900         return 0;
2901 }
2902
2903 static void
2904 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2905 {
2906         struct rb_node *n;
2907         struct btrfs_qgroup *qgroup;
2908
2909         spin_lock(&fs_info->qgroup_lock);
2910         /* clear all current qgroup tracking information */
2911         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2912                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2913                 qgroup->rfer = 0;
2914                 qgroup->rfer_cmpr = 0;
2915                 qgroup->excl = 0;
2916                 qgroup->excl_cmpr = 0;
2917         }
2918         spin_unlock(&fs_info->qgroup_lock);
2919 }
2920
2921 int
2922 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2923 {
2924         int ret = 0;
2925         struct btrfs_trans_handle *trans;
2926
2927         ret = qgroup_rescan_init(fs_info, 0, 1);
2928         if (ret)
2929                 return ret;
2930
2931         /*
2932          * We have set the rescan_progress to 0, which means no more
2933          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2934          * However, btrfs_qgroup_account_ref may be right after its call
2935          * to btrfs_find_all_roots, in which case it would still do the
2936          * accounting.
2937          * To solve this, we're committing the transaction, which will
2938          * ensure we run all delayed refs and only after that, we are
2939          * going to clear all tracking information for a clean start.
2940          */
2941
2942         trans = btrfs_join_transaction(fs_info->fs_root);
2943         if (IS_ERR(trans)) {
2944                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2945                 return PTR_ERR(trans);
2946         }
2947         ret = btrfs_commit_transaction(trans);
2948         if (ret) {
2949                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2950                 return ret;
2951         }
2952
2953         qgroup_rescan_zero_tracking(fs_info);
2954
2955         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2956                          &fs_info->qgroup_rescan_work);
2957
2958         return 0;
2959 }
2960
2961 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2962                                      bool interruptible)
2963 {
2964         int running;
2965         int ret = 0;
2966
2967         mutex_lock(&fs_info->qgroup_rescan_lock);
2968         spin_lock(&fs_info->qgroup_lock);
2969         running = fs_info->qgroup_rescan_running;
2970         spin_unlock(&fs_info->qgroup_lock);
2971         mutex_unlock(&fs_info->qgroup_rescan_lock);
2972
2973         if (!running)
2974                 return 0;
2975
2976         if (interruptible)
2977                 ret = wait_for_completion_interruptible(
2978                                         &fs_info->qgroup_rescan_completion);
2979         else
2980                 wait_for_completion(&fs_info->qgroup_rescan_completion);
2981
2982         return ret;
2983 }
2984
2985 /*
2986  * this is only called from open_ctree where we're still single threaded, thus
2987  * locking is omitted here.
2988  */
2989 void
2990 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2991 {
2992         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2993                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2994                                  &fs_info->qgroup_rescan_work);
2995 }
2996
2997 /*
2998  * Reserve qgroup space for range [start, start + len).
2999  *
3000  * This function will either reserve space from related qgroups or doing
3001  * nothing if the range is already reserved.
3002  *
3003  * Return 0 for successful reserve
3004  * Return <0 for error (including -EQUOT)
3005  *
3006  * NOTE: this function may sleep for memory allocation.
3007  *       if btrfs_qgroup_reserve_data() is called multiple times with
3008  *       same @reserved, caller must ensure when error happens it's OK
3009  *       to free *ALL* reserved space.
3010  */
3011 int btrfs_qgroup_reserve_data(struct inode *inode,
3012                         struct extent_changeset **reserved_ret, u64 start,
3013                         u64 len)
3014 {
3015         struct btrfs_root *root = BTRFS_I(inode)->root;
3016         struct ulist_node *unode;
3017         struct ulist_iterator uiter;
3018         struct extent_changeset *reserved;
3019         u64 orig_reserved;
3020         u64 to_reserve;
3021         int ret;
3022
3023         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3024             !is_fstree(root->objectid) || len == 0)
3025                 return 0;
3026
3027         /* @reserved parameter is mandatory for qgroup */
3028         if (WARN_ON(!reserved_ret))
3029                 return -EINVAL;
3030         if (!*reserved_ret) {
3031                 *reserved_ret = extent_changeset_alloc();
3032                 if (!*reserved_ret)
3033                         return -ENOMEM;
3034         }
3035         reserved = *reserved_ret;
3036         /* Record already reserved space */
3037         orig_reserved = reserved->bytes_changed;
3038         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3039                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3040
3041         /* Newly reserved space */
3042         to_reserve = reserved->bytes_changed - orig_reserved;
3043         trace_btrfs_qgroup_reserve_data(inode, start, len,
3044                                         to_reserve, QGROUP_RESERVE);
3045         if (ret < 0)
3046                 goto cleanup;
3047         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3048         if (ret < 0)
3049                 goto cleanup;
3050
3051         return ret;
3052
3053 cleanup:
3054         /* cleanup *ALL* already reserved ranges */
3055         ULIST_ITER_INIT(&uiter);
3056         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3057                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3058                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3059         extent_changeset_release(reserved);
3060         return ret;
3061 }
3062
3063 /* Free ranges specified by @reserved, normally in error path */
3064 static int qgroup_free_reserved_data(struct inode *inode,
3065                         struct extent_changeset *reserved, u64 start, u64 len)
3066 {
3067         struct btrfs_root *root = BTRFS_I(inode)->root;
3068         struct ulist_node *unode;
3069         struct ulist_iterator uiter;
3070         struct extent_changeset changeset;
3071         int freed = 0;
3072         int ret;
3073
3074         extent_changeset_init(&changeset);
3075         len = round_up(start + len, root->fs_info->sectorsize);
3076         start = round_down(start, root->fs_info->sectorsize);
3077
3078         ULIST_ITER_INIT(&uiter);
3079         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3080                 u64 range_start = unode->val;
3081                 /* unode->aux is the inclusive end */
3082                 u64 range_len = unode->aux - range_start + 1;
3083                 u64 free_start;
3084                 u64 free_len;
3085
3086                 extent_changeset_release(&changeset);
3087
3088                 /* Only free range in range [start, start + len) */
3089                 if (range_start >= start + len ||
3090                     range_start + range_len <= start)
3091                         continue;
3092                 free_start = max(range_start, start);
3093                 free_len = min(start + len, range_start + range_len) -
3094                            free_start;
3095                 /*
3096                  * TODO: To also modify reserved->ranges_reserved to reflect
3097                  * the modification.
3098                  *
3099                  * However as long as we free qgroup reserved according to
3100                  * EXTENT_QGROUP_RESERVED, we won't double free.
3101                  * So not need to rush.
3102                  */
3103                 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3104                                 free_start, free_start + free_len - 1,
3105                                 EXTENT_QGROUP_RESERVED, &changeset);
3106                 if (ret < 0)
3107                         goto out;
3108                 freed += changeset.bytes_changed;
3109         }
3110         btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed,
3111                                   BTRFS_QGROUP_RSV_DATA);
3112         ret = freed;
3113 out:
3114         extent_changeset_release(&changeset);
3115         return ret;
3116 }
3117
3118 static int __btrfs_qgroup_release_data(struct inode *inode,
3119                         struct extent_changeset *reserved, u64 start, u64 len,
3120                         int free)
3121 {
3122         struct extent_changeset changeset;
3123         int trace_op = QGROUP_RELEASE;
3124         int ret;
3125
3126         /* In release case, we shouldn't have @reserved */
3127         WARN_ON(!free && reserved);
3128         if (free && reserved)
3129                 return qgroup_free_reserved_data(inode, reserved, start, len);
3130         extent_changeset_init(&changeset);
3131         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3132                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3133         if (ret < 0)
3134                 goto out;
3135
3136         if (free)
3137                 trace_op = QGROUP_FREE;
3138         trace_btrfs_qgroup_release_data(inode, start, len,
3139                                         changeset.bytes_changed, trace_op);
3140         if (free)
3141                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3142                                 BTRFS_I(inode)->root->objectid,
3143                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3144         ret = changeset.bytes_changed;
3145 out:
3146         extent_changeset_release(&changeset);
3147         return ret;
3148 }
3149
3150 /*
3151  * Free a reserved space range from io_tree and related qgroups
3152  *
3153  * Should be called when a range of pages get invalidated before reaching disk.
3154  * Or for error cleanup case.
3155  * if @reserved is given, only reserved range in [@start, @start + @len) will
3156  * be freed.
3157  *
3158  * For data written to disk, use btrfs_qgroup_release_data().
3159  *
3160  * NOTE: This function may sleep for memory allocation.
3161  */
3162 int btrfs_qgroup_free_data(struct inode *inode,
3163                         struct extent_changeset *reserved, u64 start, u64 len)
3164 {
3165         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3166 }
3167
3168 /*
3169  * Release a reserved space range from io_tree only.
3170  *
3171  * Should be called when a range of pages get written to disk and corresponding
3172  * FILE_EXTENT is inserted into corresponding root.
3173  *
3174  * Since new qgroup accounting framework will only update qgroup numbers at
3175  * commit_transaction() time, its reserved space shouldn't be freed from
3176  * related qgroups.
3177  *
3178  * But we should release the range from io_tree, to allow further write to be
3179  * COWed.
3180  *
3181  * NOTE: This function may sleep for memory allocation.
3182  */
3183 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3184 {
3185         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3186 }
3187
3188 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3189                               enum btrfs_qgroup_rsv_type type)
3190 {
3191         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3192             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3193                 return;
3194         if (num_bytes == 0)
3195                 return;
3196
3197         spin_lock(&root->qgroup_meta_rsv_lock);
3198         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3199                 root->qgroup_meta_rsv_prealloc += num_bytes;
3200         else
3201                 root->qgroup_meta_rsv_pertrans += num_bytes;
3202         spin_unlock(&root->qgroup_meta_rsv_lock);
3203 }
3204
3205 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3206                              enum btrfs_qgroup_rsv_type type)
3207 {
3208         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3209             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3210                 return 0;
3211         if (num_bytes == 0)
3212                 return 0;
3213
3214         spin_lock(&root->qgroup_meta_rsv_lock);
3215         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3216                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3217                                   num_bytes);
3218                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3219         } else {
3220                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3221                                   num_bytes);
3222                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3223         }
3224         spin_unlock(&root->qgroup_meta_rsv_lock);
3225         return num_bytes;
3226 }
3227
3228 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3229                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3230 {
3231         struct btrfs_fs_info *fs_info = root->fs_info;
3232         int ret;
3233
3234         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3235             !is_fstree(root->objectid) || num_bytes == 0)
3236                 return 0;
3237
3238         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3239         trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3240         ret = qgroup_reserve(root, num_bytes, enforce, type);
3241         if (ret < 0)
3242                 return ret;
3243         /*
3244          * Record what we have reserved into root.
3245          *
3246          * To avoid quota disabled->enabled underflow.
3247          * In that case, we may try to free space we haven't reserved
3248          * (since quota was disabled), so record what we reserved into root.
3249          * And ensure later release won't underflow this number.
3250          */
3251         add_root_meta_rsv(root, num_bytes, type);
3252         return ret;
3253 }
3254
3255 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3256 {
3257         struct btrfs_fs_info *fs_info = root->fs_info;
3258
3259         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3260             !is_fstree(root->objectid))
3261                 return;
3262
3263         /* TODO: Update trace point to handle such free */
3264         trace_qgroup_meta_free_all_pertrans(root);
3265         /* Special value -1 means to free all reserved space */
3266         btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1,
3267                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3268 }
3269
3270 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3271                               enum btrfs_qgroup_rsv_type type)
3272 {
3273         struct btrfs_fs_info *fs_info = root->fs_info;
3274
3275         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3276             !is_fstree(root->objectid))
3277                 return;
3278
3279         /*
3280          * reservation for META_PREALLOC can happen before quota is enabled,
3281          * which can lead to underflow.
3282          * Here ensure we will only free what we really have reserved.
3283          */
3284         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3285         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3286         trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3287         btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type);
3288 }
3289
3290 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3291                                 int num_bytes)
3292 {
3293         struct btrfs_root *quota_root = fs_info->quota_root;
3294         struct btrfs_qgroup *qgroup;
3295         struct ulist_node *unode;
3296         struct ulist_iterator uiter;
3297         int ret = 0;
3298
3299         if (num_bytes == 0)
3300                 return;
3301         if (!quota_root)
3302                 return;
3303
3304         spin_lock(&fs_info->qgroup_lock);
3305         qgroup = find_qgroup_rb(fs_info, ref_root);
3306         if (!qgroup)
3307                 goto out;
3308         ulist_reinit(fs_info->qgroup_ulist);
3309         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3310                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3311         if (ret < 0)
3312                 goto out;
3313         ULIST_ITER_INIT(&uiter);
3314         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3315                 struct btrfs_qgroup *qg;
3316                 struct btrfs_qgroup_list *glist;
3317
3318                 qg = unode_aux_to_qgroup(unode);
3319
3320                 qgroup_rsv_release(fs_info, qg, num_bytes,
3321                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3322                 qgroup_rsv_add(fs_info, qg, num_bytes,
3323                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3324                 list_for_each_entry(glist, &qg->groups, next_group) {
3325                         ret = ulist_add(fs_info->qgroup_ulist,
3326                                         glist->group->qgroupid,
3327                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3328                         if (ret < 0)
3329                                 goto out;
3330                 }
3331         }
3332 out:
3333         spin_unlock(&fs_info->qgroup_lock);
3334 }
3335
3336 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3337 {
3338         struct btrfs_fs_info *fs_info = root->fs_info;
3339
3340         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3341             !is_fstree(root->objectid))
3342                 return;
3343         /* Same as btrfs_qgroup_free_meta_prealloc() */
3344         num_bytes = sub_root_meta_rsv(root, num_bytes,
3345                                       BTRFS_QGROUP_RSV_META_PREALLOC);
3346         trace_qgroup_meta_convert(root, num_bytes);
3347         qgroup_convert_meta(fs_info, root->objectid, num_bytes);
3348 }
3349
3350 /*
3351  * Check qgroup reserved space leaking, normally at destroy inode
3352  * time
3353  */
3354 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3355 {
3356         struct extent_changeset changeset;
3357         struct ulist_node *unode;
3358         struct ulist_iterator iter;
3359         int ret;
3360
3361         extent_changeset_init(&changeset);
3362         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3363                         EXTENT_QGROUP_RESERVED, &changeset);
3364
3365         WARN_ON(ret < 0);
3366         if (WARN_ON(changeset.bytes_changed)) {
3367                 ULIST_ITER_INIT(&iter);
3368                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3369                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3370                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3371                                 inode->i_ino, unode->val, unode->aux);
3372                 }
3373                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3374                                 BTRFS_I(inode)->root->objectid,
3375                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3376
3377         }
3378         extent_changeset_release(&changeset);
3379 }