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, u64 src,
 534                                     u64 dst)
 535 {
 536         int ret;
 537         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 538         struct btrfs_path *path;
 539         struct btrfs_key key;
 540
 541         path = btrfs_alloc_path();
 542         if (!path)
 543                 return -ENOMEM;
 544
 545         key.objectid = src;
 546         key.type = BTRFS_QGROUP_RELATION_KEY;
 547         key.offset = dst;
 548
 549         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 550
 551         btrfs_mark_buffer_dirty(path->nodes[0]);
 552
 553         btrfs_free_path(path);
 554         return ret;
 555 }
 556
 557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 558                                     u64 dst)
 559 {
 560         int ret;
 561         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 562         struct btrfs_path *path;
 563         struct btrfs_key key;
 564
 565         path = btrfs_alloc_path();
 566         if (!path)
 567                 return -ENOMEM;
 568
 569         key.objectid = src;
 570         key.type = BTRFS_QGROUP_RELATION_KEY;
 571         key.offset = dst;
 572
 573         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 574         if (ret < 0)
 575                 goto out;
 576
 577         if (ret > 0) {
 578                 ret = -ENOENT;
 579                 goto out;
 580         }
 581
 582         ret = btrfs_del_item(trans, quota_root, path);
 583 out:
 584         btrfs_free_path(path);
 585         return ret;
 586 }
 587
 588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 589                            struct btrfs_root *quota_root, u64 qgroupid)
 590 {
 591         int ret;
 592         struct btrfs_path *path;
 593         struct btrfs_qgroup_info_item *qgroup_info;
 594         struct btrfs_qgroup_limit_item *qgroup_limit;
 595         struct extent_buffer *leaf;
 596         struct btrfs_key key;
 597
 598         if (btrfs_is_testing(quota_root->fs_info))
 599                 return 0;
 600
 601         path = btrfs_alloc_path();
 602         if (!path)
 603                 return -ENOMEM;
 604
 605         key.objectid = 0;
 606         key.type = BTRFS_QGROUP_INFO_KEY;
 607         key.offset = qgroupid;
 608
 609         /*
 610          * Avoid a transaction abort by catching -EEXIST here. In that
 611          * case, we proceed by re-initializing the existing structure
 612          * on disk.
 613          */
 614
 615         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 616                                       sizeof(*qgroup_info));
 617         if (ret && ret != -EEXIST)
 618                 goto out;
 619
 620         leaf = path->nodes[0];
 621         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 622                                  struct btrfs_qgroup_info_item);
 623         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 624         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 625         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 626         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 627         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 628
 629         btrfs_mark_buffer_dirty(leaf);
 630
 631         btrfs_release_path(path);
 632
 633         key.type = BTRFS_QGROUP_LIMIT_KEY;
 634         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 635                                       sizeof(*qgroup_limit));
 636         if (ret && ret != -EEXIST)
 637                 goto out;
 638
 639         leaf = path->nodes[0];
 640         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 641                                   struct btrfs_qgroup_limit_item);
 642         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 643         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 644         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 645         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 646         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 647
 648         btrfs_mark_buffer_dirty(leaf);
 649
 650         ret = 0;
 651 out:
 652         btrfs_free_path(path);
 653         return ret;
 654 }
 655
 656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
 657 {
 658         int ret;
 659         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 660         struct btrfs_path *path;
 661         struct btrfs_key key;
 662
 663         path = btrfs_alloc_path();
 664         if (!path)
 665                 return -ENOMEM;
 666
 667         key.objectid = 0;
 668         key.type = BTRFS_QGROUP_INFO_KEY;
 669         key.offset = qgroupid;
 670         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 671         if (ret < 0)
 672                 goto out;
 673
 674         if (ret > 0) {
 675                 ret = -ENOENT;
 676                 goto out;
 677         }
 678
 679         ret = btrfs_del_item(trans, quota_root, path);
 680         if (ret)
 681                 goto out;
 682
 683         btrfs_release_path(path);
 684
 685         key.type = BTRFS_QGROUP_LIMIT_KEY;
 686         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 687         if (ret < 0)
 688                 goto out;
 689
 690         if (ret > 0) {
 691                 ret = -ENOENT;
 692                 goto out;
 693         }
 694
 695         ret = btrfs_del_item(trans, quota_root, path);
 696
 697 out:
 698         btrfs_free_path(path);
 699         return ret;
 700 }
 701
 702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 703                                     struct btrfs_qgroup *qgroup)
 704 {
 705         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 706         struct btrfs_path *path;
 707         struct btrfs_key key;
 708         struct extent_buffer *l;
 709         struct btrfs_qgroup_limit_item *qgroup_limit;
 710         int ret;
 711         int slot;
 712
 713         key.objectid = 0;
 714         key.type = BTRFS_QGROUP_LIMIT_KEY;
 715         key.offset = qgroup->qgroupid;
 716
 717         path = btrfs_alloc_path();
 718         if (!path)
 719                 return -ENOMEM;
 720
 721         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 722         if (ret > 0)
 723                 ret = -ENOENT;
 724
 725         if (ret)
 726                 goto out;
 727
 728         l = path->nodes[0];
 729         slot = path->slots[0];
 730         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 731         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 732         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 733         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 734         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 735         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 736
 737         btrfs_mark_buffer_dirty(l);
 738
 739 out:
 740         btrfs_free_path(path);
 741         return ret;
 742 }
 743
 744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 745                                    struct btrfs_qgroup *qgroup)
 746 {
 747         struct btrfs_fs_info *fs_info = trans->fs_info;
 748         struct btrfs_root *quota_root = fs_info->quota_root;
 749         struct btrfs_path *path;
 750         struct btrfs_key key;
 751         struct extent_buffer *l;
 752         struct btrfs_qgroup_info_item *qgroup_info;
 753         int ret;
 754         int slot;
 755
 756         if (btrfs_is_testing(fs_info))
 757                 return 0;
 758
 759         key.objectid = 0;
 760         key.type = BTRFS_QGROUP_INFO_KEY;
 761         key.offset = qgroup->qgroupid;
 762
 763         path = btrfs_alloc_path();
 764         if (!path)
 765                 return -ENOMEM;
 766
 767         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 768         if (ret > 0)
 769                 ret = -ENOENT;
 770
 771         if (ret)
 772                 goto out;
 773
 774         l = path->nodes[0];
 775         slot = path->slots[0];
 776         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 777         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 778         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 779         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 780         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 781         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 782
 783         btrfs_mark_buffer_dirty(l);
 784
 785 out:
 786         btrfs_free_path(path);
 787         return ret;
 788 }
 789
 790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
 791 {
 792         struct btrfs_fs_info *fs_info = trans->fs_info;
 793         struct btrfs_root *quota_root = fs_info->quota_root;
 794         struct btrfs_path *path;
 795         struct btrfs_key key;
 796         struct extent_buffer *l;
 797         struct btrfs_qgroup_status_item *ptr;
 798         int ret;
 799         int slot;
 800
 801         key.objectid = 0;
 802         key.type = BTRFS_QGROUP_STATUS_KEY;
 803         key.offset = 0;
 804
 805         path = btrfs_alloc_path();
 806         if (!path)
 807                 return -ENOMEM;
 808
 809         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 810         if (ret > 0)
 811                 ret = -ENOENT;
 812
 813         if (ret)
 814                 goto out;
 815
 816         l = path->nodes[0];
 817         slot = path->slots[0];
 818         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 819         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 820         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 821         btrfs_set_qgroup_status_rescan(l, ptr,
 822                                 fs_info->qgroup_rescan_progress.objectid);
 823
 824         btrfs_mark_buffer_dirty(l);
 825
 826 out:
 827         btrfs_free_path(path);
 828         return ret;
 829 }
 830
 831 /*
 832  * called with qgroup_lock held
 833  */
 834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 835                                   struct btrfs_root *root)
 836 {
 837         struct btrfs_path *path;
 838         struct btrfs_key key;
 839         struct extent_buffer *leaf = NULL;
 840         int ret;
 841         int nr = 0;
 842
 843         path = btrfs_alloc_path();
 844         if (!path)
 845                 return -ENOMEM;
 846
 847         path->leave_spinning = 1;
 848
 849         key.objectid = 0;
 850         key.offset = 0;
 851         key.type = 0;
 852
 853         while (1) {
 854                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 855                 if (ret < 0)
 856                         goto out;
 857                 leaf = path->nodes[0];
 858                 nr = btrfs_header_nritems(leaf);
 859                 if (!nr)
 860                         break;
 861                 /*
 862                  * delete the leaf one by one
 863                  * since the whole tree is going
 864                  * to be deleted.
 865                  */
 866                 path->slots[0] = 0;
 867                 ret = btrfs_del_items(trans, root, path, 0, nr);
 868                 if (ret)
 869                         goto out;
 870
 871                 btrfs_release_path(path);
 872         }
 873         ret = 0;
 874 out:
 875         btrfs_free_path(path);
 876         return ret;
 877 }
 878
 879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 880 {
 881         struct btrfs_root *quota_root;
 882         struct btrfs_root *tree_root = fs_info->tree_root;
 883         struct btrfs_path *path = NULL;
 884         struct btrfs_qgroup_status_item *ptr;
 885         struct extent_buffer *leaf;
 886         struct btrfs_key key;
 887         struct btrfs_key found_key;
 888         struct btrfs_qgroup *qgroup = NULL;
 889         struct btrfs_trans_handle *trans = NULL;
 890         int ret = 0;
 891         int slot;
 892
 893         mutex_lock(&fs_info->qgroup_ioctl_lock);
 894         if (fs_info->quota_root)
 895                 goto out;
 896
 897         /*
 898          * 1 for quota root item
 899          * 1 for BTRFS_QGROUP_STATUS item
 900          *
 901          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
 902          * per subvolume. However those are not currently reserved since it
 903          * would be a lot of overkill.
 904          */
 905         trans = btrfs_start_transaction(tree_root, 2);
 906         if (IS_ERR(trans)) {
 907                 ret = PTR_ERR(trans);
 908                 trans = NULL;
 909                 goto out;
 910         }
 911
 912         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 913         if (!fs_info->qgroup_ulist) {
 914                 ret = -ENOMEM;
 915                 btrfs_abort_transaction(trans, ret);
 916                 goto out;
 917         }
 918
 919         /*
 920          * initially create the quota tree
 921          */
 922         quota_root = btrfs_create_tree(trans, fs_info,
 923                                        BTRFS_QUOTA_TREE_OBJECTID);
 924         if (IS_ERR(quota_root)) {
 925                 ret =  PTR_ERR(quota_root);
 926                 btrfs_abort_transaction(trans, ret);
 927                 goto out;
 928         }
 929
 930         path = btrfs_alloc_path();
 931         if (!path) {
 932                 ret = -ENOMEM;
 933                 btrfs_abort_transaction(trans, ret);
 934                 goto out_free_root;
 935         }
 936
 937         key.objectid = 0;
 938         key.type = BTRFS_QGROUP_STATUS_KEY;
 939         key.offset = 0;
 940
 941         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 942                                       sizeof(*ptr));
 943         if (ret) {
 944                 btrfs_abort_transaction(trans, ret);
 945                 goto out_free_path;
 946         }
 947
 948         leaf = path->nodes[0];
 949         ptr = btrfs_item_ptr(leaf, path->slots[0],
 950                                  struct btrfs_qgroup_status_item);
 951         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 952         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 953         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 954                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 955         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 956         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 957
 958         btrfs_mark_buffer_dirty(leaf);
 959
 960         key.objectid = 0;
 961         key.type = BTRFS_ROOT_REF_KEY;
 962         key.offset = 0;
 963
 964         btrfs_release_path(path);
 965         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 966         if (ret > 0)
 967                 goto out_add_root;
 968         if (ret < 0) {
 969                 btrfs_abort_transaction(trans, ret);
 970                 goto out_free_path;
 971         }
 972
 973         while (1) {
 974                 slot = path->slots[0];
 975                 leaf = path->nodes[0];
 976                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 977
 978                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 979                         ret = add_qgroup_item(trans, quota_root,
 980                                               found_key.offset);
 981                         if (ret) {
 982                                 btrfs_abort_transaction(trans, ret);
 983                                 goto out_free_path;
 984                         }
 985
 986                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 987                         if (IS_ERR(qgroup)) {
 988                                 ret = PTR_ERR(qgroup);
 989                                 btrfs_abort_transaction(trans, ret);
 990                                 goto out_free_path;
 991                         }
 992                 }
 993                 ret = btrfs_next_item(tree_root, path);
 994                 if (ret < 0) {
 995                         btrfs_abort_transaction(trans, ret);
 996                         goto out_free_path;
 997                 }
 998                 if (ret)
 999                         break;
1000         }
1001
1002 out_add_root:
1003         btrfs_release_path(path);
1004         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1005         if (ret) {
1006                 btrfs_abort_transaction(trans, ret);
1007                 goto out_free_path;
1008         }
1009
1010         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011         if (IS_ERR(qgroup)) {
1012                 ret = PTR_ERR(qgroup);
1013                 btrfs_abort_transaction(trans, ret);
1014                 goto out_free_path;
1015         }
1016         spin_lock(&fs_info->qgroup_lock);
1017         fs_info->quota_root = quota_root;
1018         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1019         spin_unlock(&fs_info->qgroup_lock);
1020
1021         ret = btrfs_commit_transaction(trans);
1022         trans = NULL;
1023         if (ret)
1024                 goto out_free_path;
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, &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, u64 src,
1239                               u64 dst)
1240 {
1241         struct btrfs_fs_info *fs_info = trans->fs_info;
1242         struct btrfs_root *quota_root;
1243         struct btrfs_qgroup *parent;
1244         struct btrfs_qgroup *member;
1245         struct btrfs_qgroup_list *list;
1246         struct ulist *tmp;
1247         int ret = 0;
1248
1249         /* Check the level of src and dst first */
1250         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1251                 return -EINVAL;
1252
1253         tmp = ulist_alloc(GFP_KERNEL);
1254         if (!tmp)
1255                 return -ENOMEM;
1256
1257         mutex_lock(&fs_info->qgroup_ioctl_lock);
1258         quota_root = fs_info->quota_root;
1259         if (!quota_root) {
1260                 ret = -EINVAL;
1261                 goto out;
1262         }
1263         member = find_qgroup_rb(fs_info, src);
1264         parent = find_qgroup_rb(fs_info, dst);
1265         if (!member || !parent) {
1266                 ret = -EINVAL;
1267                 goto out;
1268         }
1269
1270         /* check if such qgroup relation exist firstly */
1271         list_for_each_entry(list, &member->groups, next_group) {
1272                 if (list->group == parent) {
1273                         ret = -EEXIST;
1274                         goto out;
1275                 }
1276         }
1277
1278         ret = add_qgroup_relation_item(trans, src, dst);
1279         if (ret)
1280                 goto out;
1281
1282         ret = add_qgroup_relation_item(trans, dst, src);
1283         if (ret) {
1284                 del_qgroup_relation_item(trans, src, dst);
1285                 goto out;
1286         }
1287
1288         spin_lock(&fs_info->qgroup_lock);
1289         ret = add_relation_rb(fs_info, src, dst);
1290         if (ret < 0) {
1291                 spin_unlock(&fs_info->qgroup_lock);
1292                 goto out;
1293         }
1294         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1295         spin_unlock(&fs_info->qgroup_lock);
1296 out:
1297         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1298         ulist_free(tmp);
1299         return ret;
1300 }
1301
1302 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1303                                  u64 dst)
1304 {
1305         struct btrfs_fs_info *fs_info = trans->fs_info;
1306         struct btrfs_root *quota_root;
1307         struct btrfs_qgroup *parent;
1308         struct btrfs_qgroup *member;
1309         struct btrfs_qgroup_list *list;
1310         struct ulist *tmp;
1311         int ret = 0;
1312         int err;
1313
1314         tmp = ulist_alloc(GFP_KERNEL);
1315         if (!tmp)
1316                 return -ENOMEM;
1317
1318         quota_root = fs_info->quota_root;
1319         if (!quota_root) {
1320                 ret = -EINVAL;
1321                 goto out;
1322         }
1323
1324         member = find_qgroup_rb(fs_info, src);
1325         parent = find_qgroup_rb(fs_info, dst);
1326         if (!member || !parent) {
1327                 ret = -EINVAL;
1328                 goto out;
1329         }
1330
1331         /* check if such qgroup relation exist firstly */
1332         list_for_each_entry(list, &member->groups, next_group) {
1333                 if (list->group == parent)
1334                         goto exist;
1335         }
1336         ret = -ENOENT;
1337         goto out;
1338 exist:
1339         ret = del_qgroup_relation_item(trans, src, dst);
1340         err = del_qgroup_relation_item(trans, dst, src);
1341         if (err && !ret)
1342                 ret = err;
1343
1344         spin_lock(&fs_info->qgroup_lock);
1345         del_relation_rb(fs_info, src, dst);
1346         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1347         spin_unlock(&fs_info->qgroup_lock);
1348 out:
1349         ulist_free(tmp);
1350         return ret;
1351 }
1352
1353 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1354                               u64 dst)
1355 {
1356         struct btrfs_fs_info *fs_info = trans->fs_info;
1357         int ret = 0;
1358
1359         mutex_lock(&fs_info->qgroup_ioctl_lock);
1360         ret = __del_qgroup_relation(trans, src, dst);
1361         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1362
1363         return ret;
1364 }
1365
1366 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1367 {
1368         struct btrfs_fs_info *fs_info = trans->fs_info;
1369         struct btrfs_root *quota_root;
1370         struct btrfs_qgroup *qgroup;
1371         int ret = 0;
1372
1373         mutex_lock(&fs_info->qgroup_ioctl_lock);
1374         quota_root = fs_info->quota_root;
1375         if (!quota_root) {
1376                 ret = -EINVAL;
1377                 goto out;
1378         }
1379         qgroup = find_qgroup_rb(fs_info, qgroupid);
1380         if (qgroup) {
1381                 ret = -EEXIST;
1382                 goto out;
1383         }
1384
1385         ret = add_qgroup_item(trans, quota_root, qgroupid);
1386         if (ret)
1387                 goto out;
1388
1389         spin_lock(&fs_info->qgroup_lock);
1390         qgroup = add_qgroup_rb(fs_info, qgroupid);
1391         spin_unlock(&fs_info->qgroup_lock);
1392
1393         if (IS_ERR(qgroup))
1394                 ret = PTR_ERR(qgroup);
1395 out:
1396         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1397         return ret;
1398 }
1399
1400 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1401 {
1402         struct btrfs_fs_info *fs_info = trans->fs_info;
1403         struct btrfs_root *quota_root;
1404         struct btrfs_qgroup *qgroup;
1405         struct btrfs_qgroup_list *list;
1406         int ret = 0;
1407
1408         mutex_lock(&fs_info->qgroup_ioctl_lock);
1409         quota_root = fs_info->quota_root;
1410         if (!quota_root) {
1411                 ret = -EINVAL;
1412                 goto out;
1413         }
1414
1415         qgroup = find_qgroup_rb(fs_info, qgroupid);
1416         if (!qgroup) {
1417                 ret = -ENOENT;
1418                 goto out;
1419         } else {
1420                 /* check if there are no children of this qgroup */
1421                 if (!list_empty(&qgroup->members)) {
1422                         ret = -EBUSY;
1423                         goto out;
1424                 }
1425         }
1426         ret = del_qgroup_item(trans, qgroupid);
1427         if (ret && ret != -ENOENT)
1428                 goto out;
1429
1430         while (!list_empty(&qgroup->groups)) {
1431                 list = list_first_entry(&qgroup->groups,
1432                                         struct btrfs_qgroup_list, next_group);
1433                 ret = __del_qgroup_relation(trans, qgroupid,
1434                                             list->group->qgroupid);
1435                 if (ret)
1436                         goto out;
1437         }
1438
1439         spin_lock(&fs_info->qgroup_lock);
1440         del_qgroup_rb(fs_info, qgroupid);
1441         spin_unlock(&fs_info->qgroup_lock);
1442 out:
1443         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1444         return ret;
1445 }
1446
1447 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1448                        struct btrfs_qgroup_limit *limit)
1449 {
1450         struct btrfs_fs_info *fs_info = trans->fs_info;
1451         struct btrfs_root *quota_root;
1452         struct btrfs_qgroup *qgroup;
1453         int ret = 0;
1454         /* Sometimes we would want to clear the limit on this qgroup.
1455          * To meet this requirement, we treat the -1 as a special value
1456          * which tell kernel to clear the limit on this qgroup.
1457          */
1458         const u64 CLEAR_VALUE = -1;
1459
1460         mutex_lock(&fs_info->qgroup_ioctl_lock);
1461         quota_root = fs_info->quota_root;
1462         if (!quota_root) {
1463                 ret = -EINVAL;
1464                 goto out;
1465         }
1466
1467         qgroup = find_qgroup_rb(fs_info, qgroupid);
1468         if (!qgroup) {
1469                 ret = -ENOENT;
1470                 goto out;
1471         }
1472
1473         spin_lock(&fs_info->qgroup_lock);
1474         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1475                 if (limit->max_rfer == CLEAR_VALUE) {
1476                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1477                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1478                         qgroup->max_rfer = 0;
1479                 } else {
1480                         qgroup->max_rfer = limit->max_rfer;
1481                 }
1482         }
1483         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1484                 if (limit->max_excl == CLEAR_VALUE) {
1485                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1486                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1487                         qgroup->max_excl = 0;
1488                 } else {
1489                         qgroup->max_excl = limit->max_excl;
1490                 }
1491         }
1492         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1493                 if (limit->rsv_rfer == CLEAR_VALUE) {
1494                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1495                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1496                         qgroup->rsv_rfer = 0;
1497                 } else {
1498                         qgroup->rsv_rfer = limit->rsv_rfer;
1499                 }
1500         }
1501         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1502                 if (limit->rsv_excl == CLEAR_VALUE) {
1503                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1504                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1505                         qgroup->rsv_excl = 0;
1506                 } else {
1507                         qgroup->rsv_excl = limit->rsv_excl;
1508                 }
1509         }
1510         qgroup->lim_flags |= limit->flags;
1511
1512         spin_unlock(&fs_info->qgroup_lock);
1513
1514         ret = update_qgroup_limit_item(trans, qgroup);
1515         if (ret) {
1516                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1517                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1518                        qgroupid);
1519         }
1520
1521 out:
1522         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1523         return ret;
1524 }
1525
1526 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1527                                 struct btrfs_delayed_ref_root *delayed_refs,
1528                                 struct btrfs_qgroup_extent_record *record)
1529 {
1530         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1531         struct rb_node *parent_node = NULL;
1532         struct btrfs_qgroup_extent_record *entry;
1533         u64 bytenr = record->bytenr;
1534
1535         lockdep_assert_held(&delayed_refs->lock);
1536         trace_btrfs_qgroup_trace_extent(fs_info, record);
1537
1538         while (*p) {
1539                 parent_node = *p;
1540                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1541                                  node);
1542                 if (bytenr < entry->bytenr)
1543                         p = &(*p)->rb_left;
1544                 else if (bytenr > entry->bytenr)
1545                         p = &(*p)->rb_right;
1546                 else
1547                         return 1;
1548         }
1549
1550         rb_link_node(&record->node, parent_node, p);
1551         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1552         return 0;
1553 }
1554
1555 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1556                                    struct btrfs_qgroup_extent_record *qrecord)
1557 {
1558         struct ulist *old_root;
1559         u64 bytenr = qrecord->bytenr;
1560         int ret;
1561
1562         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1563         if (ret < 0) {
1564                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1565                 btrfs_warn(fs_info,
1566 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1567                         ret);
1568                 return 0;
1569         }
1570
1571         /*
1572          * Here we don't need to get the lock of
1573          * trans->transaction->delayed_refs, since inserted qrecord won't
1574          * be deleted, only qrecord->node may be modified (new qrecord insert)
1575          *
1576          * So modifying qrecord->old_roots is safe here
1577          */
1578         qrecord->old_roots = old_root;
1579         return 0;
1580 }
1581
1582 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1583                               u64 num_bytes, gfp_t gfp_flag)
1584 {
1585         struct btrfs_fs_info *fs_info = trans->fs_info;
1586         struct btrfs_qgroup_extent_record *record;
1587         struct btrfs_delayed_ref_root *delayed_refs;
1588         int ret;
1589
1590         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1591             || bytenr == 0 || num_bytes == 0)
1592                 return 0;
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 extent_buffer *eb)
1614 {
1615         struct btrfs_fs_info *fs_info = trans->fs_info;
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, bytenr, num_bytes,
1646                                                 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 extent_buffer *root_eb,
1717                                u64 root_gen, int root_level)
1718 {
1719         struct btrfs_fs_info *fs_info = trans->fs_info;
1720         int ret = 0;
1721         int level;
1722         struct extent_buffer *eb = root_eb;
1723         struct btrfs_path *path = NULL;
1724
1725         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1726         BUG_ON(root_eb == NULL);
1727
1728         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1729                 return 0;
1730
1731         if (!extent_buffer_uptodate(root_eb)) {
1732                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
1733                 if (ret)
1734                         goto out;
1735         }
1736
1737         if (root_level == 0) {
1738                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
1739                 goto out;
1740         }
1741
1742         path = btrfs_alloc_path();
1743         if (!path)
1744                 return -ENOMEM;
1745
1746         /*
1747          * Walk down the tree.  Missing extent blocks are filled in as
1748          * we go. Metadata is accounted every time we read a new
1749          * extent block.
1750          *
1751          * When we reach a leaf, we account for file extent items in it,
1752          * walk back up the tree (adjusting slot pointers as we go)
1753          * and restart the search process.
1754          */
1755         extent_buffer_get(root_eb); /* For path */
1756         path->nodes[root_level] = root_eb;
1757         path->slots[root_level] = 0;
1758         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1759 walk_down:
1760         level = root_level;
1761         while (level >= 0) {
1762                 if (path->nodes[level] == NULL) {
1763                         struct btrfs_key first_key;
1764                         int parent_slot;
1765                         u64 child_gen;
1766                         u64 child_bytenr;
1767
1768                         /*
1769                          * We need to get child blockptr/gen from parent before
1770                          * we can read it.
1771                           */
1772                         eb = path->nodes[level + 1];
1773                         parent_slot = path->slots[level + 1];
1774                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1775                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1776                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1777
1778                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1779                                              level, &first_key);
1780                         if (IS_ERR(eb)) {
1781                                 ret = PTR_ERR(eb);
1782                                 goto out;
1783                         } else if (!extent_buffer_uptodate(eb)) {
1784                                 free_extent_buffer(eb);
1785                                 ret = -EIO;
1786                                 goto out;
1787                         }
1788
1789                         path->nodes[level] = eb;
1790                         path->slots[level] = 0;
1791
1792                         btrfs_tree_read_lock(eb);
1793                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1794                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1795
1796                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
1797                                                         fs_info->nodesize,
1798                                                         GFP_NOFS);
1799                         if (ret)
1800                                 goto out;
1801                 }
1802
1803                 if (level == 0) {
1804                         ret = btrfs_qgroup_trace_leaf_items(trans,
1805                                                             path->nodes[level]);
1806                         if (ret)
1807                                 goto out;
1808
1809                         /* Nonzero return here means we completed our search */
1810                         ret = adjust_slots_upwards(path, root_level);
1811                         if (ret)
1812                                 break;
1813
1814                         /* Restart search with new slots */
1815                         goto walk_down;
1816                 }
1817
1818                 level--;
1819         }
1820
1821         ret = 0;
1822 out:
1823         btrfs_free_path(path);
1824
1825         return ret;
1826 }
1827
1828 #define UPDATE_NEW      0
1829 #define UPDATE_OLD      1
1830 /*
1831  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1832  */
1833 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1834                                 struct ulist *roots, struct ulist *tmp,
1835                                 struct ulist *qgroups, u64 seq, int update_old)
1836 {
1837         struct ulist_node *unode;
1838         struct ulist_iterator uiter;
1839         struct ulist_node *tmp_unode;
1840         struct ulist_iterator tmp_uiter;
1841         struct btrfs_qgroup *qg;
1842         int ret = 0;
1843
1844         if (!roots)
1845                 return 0;
1846         ULIST_ITER_INIT(&uiter);
1847         while ((unode = ulist_next(roots, &uiter))) {
1848                 qg = find_qgroup_rb(fs_info, unode->val);
1849                 if (!qg)
1850                         continue;
1851
1852                 ulist_reinit(tmp);
1853                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1854                                 GFP_ATOMIC);
1855                 if (ret < 0)
1856                         return ret;
1857                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1858                 if (ret < 0)
1859                         return ret;
1860                 ULIST_ITER_INIT(&tmp_uiter);
1861                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1862                         struct btrfs_qgroup_list *glist;
1863
1864                         qg = unode_aux_to_qgroup(tmp_unode);
1865                         if (update_old)
1866                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1867                         else
1868                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1869                         list_for_each_entry(glist, &qg->groups, next_group) {
1870                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1871                                                 qgroup_to_aux(glist->group),
1872                                                 GFP_ATOMIC);
1873                                 if (ret < 0)
1874                                         return ret;
1875                                 ret = ulist_add(tmp, glist->group->qgroupid,
1876                                                 qgroup_to_aux(glist->group),
1877                                                 GFP_ATOMIC);
1878                                 if (ret < 0)
1879                                         return ret;
1880                         }
1881                 }
1882         }
1883         return 0;
1884 }
1885
1886 /*
1887  * Update qgroup rfer/excl counters.
1888  * Rfer update is easy, codes can explain themselves.
1889  *
1890  * Excl update is tricky, the update is split into 2 part.
1891  * Part 1: Possible exclusive <-> sharing detect:
1892  *      |       A       |       !A      |
1893  *  -------------------------------------
1894  *  B   |       *       |       -       |
1895  *  -------------------------------------
1896  *  !B  |       +       |       **      |
1897  *  -------------------------------------
1898  *
1899  * Conditions:
1900  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1901  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1902  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1903  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1904  *
1905  * Results:
1906  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1907  * *: Definitely not changed.           **: Possible unchanged.
1908  *
1909  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1910  *
1911  * To make the logic clear, we first use condition A and B to split
1912  * combination into 4 results.
1913  *
1914  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1915  * only on variant maybe 0.
1916  *
1917  * Lastly, check result **, since there are 2 variants maybe 0, split them
1918  * again(2x2).
1919  * But this time we don't need to consider other things, the codes and logic
1920  * is easy to understand now.
1921  */
1922 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1923                                   struct ulist *qgroups,
1924                                   u64 nr_old_roots,
1925                                   u64 nr_new_roots,
1926                                   u64 num_bytes, u64 seq)
1927 {
1928         struct ulist_node *unode;
1929         struct ulist_iterator uiter;
1930         struct btrfs_qgroup *qg;
1931         u64 cur_new_count, cur_old_count;
1932
1933         ULIST_ITER_INIT(&uiter);
1934         while ((unode = ulist_next(qgroups, &uiter))) {
1935                 bool dirty = false;
1936
1937                 qg = unode_aux_to_qgroup(unode);
1938                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1939                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1940
1941                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
1942                                              cur_new_count);
1943
1944                 /* Rfer update part */
1945                 if (cur_old_count == 0 && cur_new_count > 0) {
1946                         qg->rfer += num_bytes;
1947                         qg->rfer_cmpr += num_bytes;
1948                         dirty = true;
1949                 }
1950                 if (cur_old_count > 0 && cur_new_count == 0) {
1951                         qg->rfer -= num_bytes;
1952                         qg->rfer_cmpr -= num_bytes;
1953                         dirty = true;
1954                 }
1955
1956                 /* Excl update part */
1957                 /* Exclusive/none -> shared case */
1958                 if (cur_old_count == nr_old_roots &&
1959                     cur_new_count < nr_new_roots) {
1960                         /* Exclusive -> shared */
1961                         if (cur_old_count != 0) {
1962                                 qg->excl -= num_bytes;
1963                                 qg->excl_cmpr -= num_bytes;
1964                                 dirty = true;
1965                         }
1966                 }
1967
1968                 /* Shared -> exclusive/none case */
1969                 if (cur_old_count < nr_old_roots &&
1970                     cur_new_count == nr_new_roots) {
1971                         /* Shared->exclusive */
1972                         if (cur_new_count != 0) {
1973                                 qg->excl += num_bytes;
1974                                 qg->excl_cmpr += num_bytes;
1975                                 dirty = true;
1976                         }
1977                 }
1978
1979                 /* Exclusive/none -> exclusive/none case */
1980                 if (cur_old_count == nr_old_roots &&
1981                     cur_new_count == nr_new_roots) {
1982                         if (cur_old_count == 0) {
1983                                 /* None -> exclusive/none */
1984
1985                                 if (cur_new_count != 0) {
1986                                         /* None -> exclusive */
1987                                         qg->excl += num_bytes;
1988                                         qg->excl_cmpr += num_bytes;
1989                                         dirty = true;
1990                                 }
1991                                 /* None -> none, nothing changed */
1992                         } else {
1993                                 /* Exclusive -> exclusive/none */
1994
1995                                 if (cur_new_count == 0) {
1996                                         /* Exclusive -> none */
1997                                         qg->excl -= num_bytes;
1998                                         qg->excl_cmpr -= num_bytes;
1999                                         dirty = true;
2000                                 }
2001                                 /* Exclusive -> exclusive, nothing changed */
2002                         }
2003                 }
2004
2005                 if (dirty)
2006                         qgroup_dirty(fs_info, qg);
2007         }
2008         return 0;
2009 }
2010
2011 /*
2012  * Check if the @roots potentially is a list of fs tree roots
2013  *
2014  * Return 0 for definitely not a fs/subvol tree roots ulist
2015  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2016  *          one as well)
2017  */
2018 static int maybe_fs_roots(struct ulist *roots)
2019 {
2020         struct ulist_node *unode;
2021         struct ulist_iterator uiter;
2022
2023         /* Empty one, still possible for fs roots */
2024         if (!roots || roots->nnodes == 0)
2025                 return 1;
2026
2027         ULIST_ITER_INIT(&uiter);
2028         unode = ulist_next(roots, &uiter);
2029         if (!unode)
2030                 return 1;
2031
2032         /*
2033          * If it contains fs tree roots, then it must belong to fs/subvol
2034          * trees.
2035          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2036          */
2037         return is_fstree(unode->val);
2038 }
2039
2040 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2041                                 u64 num_bytes, struct ulist *old_roots,
2042                                 struct ulist *new_roots)
2043 {
2044         struct btrfs_fs_info *fs_info = trans->fs_info;
2045         struct ulist *qgroups = NULL;
2046         struct ulist *tmp = NULL;
2047         u64 seq;
2048         u64 nr_new_roots = 0;
2049         u64 nr_old_roots = 0;
2050         int ret = 0;
2051
2052         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2053                 return 0;
2054
2055         if (new_roots) {
2056                 if (!maybe_fs_roots(new_roots))
2057                         goto out_free;
2058                 nr_new_roots = new_roots->nnodes;
2059         }
2060         if (old_roots) {
2061                 if (!maybe_fs_roots(old_roots))
2062                         goto out_free;
2063                 nr_old_roots = old_roots->nnodes;
2064         }
2065
2066         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2067         if (nr_old_roots == 0 && nr_new_roots == 0)
2068                 goto out_free;
2069
2070         BUG_ON(!fs_info->quota_root);
2071
2072         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2073                                         num_bytes, nr_old_roots, nr_new_roots);
2074
2075         qgroups = ulist_alloc(GFP_NOFS);
2076         if (!qgroups) {
2077                 ret = -ENOMEM;
2078                 goto out_free;
2079         }
2080         tmp = ulist_alloc(GFP_NOFS);
2081         if (!tmp) {
2082                 ret = -ENOMEM;
2083                 goto out_free;
2084         }
2085
2086         mutex_lock(&fs_info->qgroup_rescan_lock);
2087         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2088                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2089                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2090                         ret = 0;
2091                         goto out_free;
2092                 }
2093         }
2094         mutex_unlock(&fs_info->qgroup_rescan_lock);
2095
2096         spin_lock(&fs_info->qgroup_lock);
2097         seq = fs_info->qgroup_seq;
2098
2099         /* Update old refcnts using old_roots */
2100         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2101                                    UPDATE_OLD);
2102         if (ret < 0)
2103                 goto out;
2104
2105         /* Update new refcnts using new_roots */
2106         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2107                                    UPDATE_NEW);
2108         if (ret < 0)
2109                 goto out;
2110
2111         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2112                                num_bytes, seq);
2113
2114         /*
2115          * Bump qgroup_seq to avoid seq overlap
2116          */
2117         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2118 out:
2119         spin_unlock(&fs_info->qgroup_lock);
2120 out_free:
2121         ulist_free(tmp);
2122         ulist_free(qgroups);
2123         ulist_free(old_roots);
2124         ulist_free(new_roots);
2125         return ret;
2126 }
2127
2128 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2129 {
2130         struct btrfs_fs_info *fs_info = trans->fs_info;
2131         struct btrfs_qgroup_extent_record *record;
2132         struct btrfs_delayed_ref_root *delayed_refs;
2133         struct ulist *new_roots = NULL;
2134         struct rb_node *node;
2135         u64 qgroup_to_skip;
2136         int ret = 0;
2137
2138         delayed_refs = &trans->transaction->delayed_refs;
2139         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2140         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2141                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2142                                   node);
2143
2144                 trace_btrfs_qgroup_account_extents(fs_info, record);
2145
2146                 if (!ret) {
2147                         /*
2148                          * Old roots should be searched when inserting qgroup
2149                          * extent record
2150                          */
2151                         if (WARN_ON(!record->old_roots)) {
2152                                 /* Search commit root to find old_roots */
2153                                 ret = btrfs_find_all_roots(NULL, fs_info,
2154                                                 record->bytenr, 0,
2155                                                 &record->old_roots, false);
2156                                 if (ret < 0)
2157                                         goto cleanup;
2158                         }
2159
2160                         /*
2161                          * Use SEQ_LAST as time_seq to do special search, which
2162                          * doesn't lock tree or delayed_refs and search current
2163                          * root. It's safe inside commit_transaction().
2164                          */
2165                         ret = btrfs_find_all_roots(trans, fs_info,
2166                                 record->bytenr, SEQ_LAST, &new_roots, false);
2167                         if (ret < 0)
2168                                 goto cleanup;
2169                         if (qgroup_to_skip) {
2170                                 ulist_del(new_roots, qgroup_to_skip, 0);
2171                                 ulist_del(record->old_roots, qgroup_to_skip,
2172                                           0);
2173                         }
2174                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2175                                                           record->num_bytes,
2176                                                           record->old_roots,
2177                                                           new_roots);
2178                         record->old_roots = NULL;
2179                         new_roots = NULL;
2180                 }
2181 cleanup:
2182                 ulist_free(record->old_roots);
2183                 ulist_free(new_roots);
2184                 new_roots = NULL;
2185                 rb_erase(node, &delayed_refs->dirty_extent_root);
2186                 kfree(record);
2187
2188         }
2189         return ret;
2190 }
2191
2192 /*
2193  * called from commit_transaction. Writes all changed qgroups to disk.
2194  */
2195 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2196 {
2197         struct btrfs_fs_info *fs_info = trans->fs_info;
2198         struct btrfs_root *quota_root = fs_info->quota_root;
2199         int ret = 0;
2200
2201         if (!quota_root)
2202                 return ret;
2203
2204         spin_lock(&fs_info->qgroup_lock);
2205         while (!list_empty(&fs_info->dirty_qgroups)) {
2206                 struct btrfs_qgroup *qgroup;
2207                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2208                                           struct btrfs_qgroup, dirty);
2209                 list_del_init(&qgroup->dirty);
2210                 spin_unlock(&fs_info->qgroup_lock);
2211                 ret = update_qgroup_info_item(trans, qgroup);
2212                 if (ret)
2213                         fs_info->qgroup_flags |=
2214                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2215                 ret = update_qgroup_limit_item(trans, qgroup);
2216                 if (ret)
2217                         fs_info->qgroup_flags |=
2218                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2219                 spin_lock(&fs_info->qgroup_lock);
2220         }
2221         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2222                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2223         else
2224                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2225         spin_unlock(&fs_info->qgroup_lock);
2226
2227         ret = update_qgroup_status_item(trans);
2228         if (ret)
2229                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2230
2231         return ret;
2232 }
2233
2234 /*
2235  * Copy the accounting information between qgroups. This is necessary
2236  * when a snapshot or a subvolume is created. Throwing an error will
2237  * cause a transaction abort so we take extra care here to only error
2238  * when a readonly fs is a reasonable outcome.
2239  */
2240 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2241                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2242 {
2243         int ret = 0;
2244         int i;
2245         u64 *i_qgroups;
2246         struct btrfs_fs_info *fs_info = trans->fs_info;
2247         struct btrfs_root *quota_root = fs_info->quota_root;
2248         struct btrfs_qgroup *srcgroup;
2249         struct btrfs_qgroup *dstgroup;
2250         u32 level_size = 0;
2251         u64 nums;
2252
2253         mutex_lock(&fs_info->qgroup_ioctl_lock);
2254         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2255                 goto out;
2256
2257         if (!quota_root) {
2258                 ret = -EINVAL;
2259                 goto out;
2260         }
2261
2262         if (inherit) {
2263                 i_qgroups = (u64 *)(inherit + 1);
2264                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2265                        2 * inherit->num_excl_copies;
2266                 for (i = 0; i < nums; ++i) {
2267                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2268
2269                         /*
2270                          * Zero out invalid groups so we can ignore
2271                          * them later.
2272                          */
2273                         if (!srcgroup ||
2274                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2275                                 *i_qgroups = 0ULL;
2276
2277                         ++i_qgroups;
2278                 }
2279         }
2280
2281         /*
2282          * create a tracking group for the subvol itself
2283          */
2284         ret = add_qgroup_item(trans, quota_root, objectid);
2285         if (ret)
2286                 goto out;
2287
2288         /*
2289          * add qgroup to all inherited groups
2290          */
2291         if (inherit) {
2292                 i_qgroups = (u64 *)(inherit + 1);
2293                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2294                         if (*i_qgroups == 0)
2295                                 continue;
2296                         ret = add_qgroup_relation_item(trans, objectid,
2297                                                        *i_qgroups);
2298                         if (ret && ret != -EEXIST)
2299                                 goto out;
2300                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2301                                                        objectid);
2302                         if (ret && ret != -EEXIST)
2303                                 goto out;
2304                 }
2305                 ret = 0;
2306         }
2307
2308
2309         spin_lock(&fs_info->qgroup_lock);
2310
2311         dstgroup = add_qgroup_rb(fs_info, objectid);
2312         if (IS_ERR(dstgroup)) {
2313                 ret = PTR_ERR(dstgroup);
2314                 goto unlock;
2315         }
2316
2317         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2318                 dstgroup->lim_flags = inherit->lim.flags;
2319                 dstgroup->max_rfer = inherit->lim.max_rfer;
2320                 dstgroup->max_excl = inherit->lim.max_excl;
2321                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2322                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2323
2324                 ret = update_qgroup_limit_item(trans, dstgroup);
2325                 if (ret) {
2326                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2327                         btrfs_info(fs_info,
2328                                    "unable to update quota limit for %llu",
2329                                    dstgroup->qgroupid);
2330                         goto unlock;
2331                 }
2332         }
2333
2334         if (srcid) {
2335                 srcgroup = find_qgroup_rb(fs_info, srcid);
2336                 if (!srcgroup)
2337                         goto unlock;
2338
2339                 /*
2340                  * We call inherit after we clone the root in order to make sure
2341                  * our counts don't go crazy, so at this point the only
2342                  * difference between the two roots should be the root node.
2343                  */
2344                 level_size = fs_info->nodesize;
2345                 dstgroup->rfer = srcgroup->rfer;
2346                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2347                 dstgroup->excl = level_size;
2348                 dstgroup->excl_cmpr = level_size;
2349                 srcgroup->excl = level_size;
2350                 srcgroup->excl_cmpr = level_size;
2351
2352                 /* inherit the limit info */
2353                 dstgroup->lim_flags = srcgroup->lim_flags;
2354                 dstgroup->max_rfer = srcgroup->max_rfer;
2355                 dstgroup->max_excl = srcgroup->max_excl;
2356                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2357                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2358
2359                 qgroup_dirty(fs_info, dstgroup);
2360                 qgroup_dirty(fs_info, srcgroup);
2361         }
2362
2363         if (!inherit)
2364                 goto unlock;
2365
2366         i_qgroups = (u64 *)(inherit + 1);
2367         for (i = 0; i < inherit->num_qgroups; ++i) {
2368                 if (*i_qgroups) {
2369                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2370                         if (ret)
2371                                 goto unlock;
2372                 }
2373                 ++i_qgroups;
2374         }
2375
2376         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2377                 struct btrfs_qgroup *src;
2378                 struct btrfs_qgroup *dst;
2379
2380                 if (!i_qgroups[0] || !i_qgroups[1])
2381                         continue;
2382
2383                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2384                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2385
2386                 if (!src || !dst) {
2387                         ret = -EINVAL;
2388                         goto unlock;
2389                 }
2390
2391                 dst->rfer = src->rfer - level_size;
2392                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2393         }
2394         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2395                 struct btrfs_qgroup *src;
2396                 struct btrfs_qgroup *dst;
2397
2398                 if (!i_qgroups[0] || !i_qgroups[1])
2399                         continue;
2400
2401                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2402                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2403
2404                 if (!src || !dst) {
2405                         ret = -EINVAL;
2406                         goto unlock;
2407                 }
2408
2409                 dst->excl = src->excl + level_size;
2410                 dst->excl_cmpr = src->excl_cmpr + level_size;
2411         }
2412
2413 unlock:
2414         spin_unlock(&fs_info->qgroup_lock);
2415 out:
2416         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2417         return ret;
2418 }
2419
2420 /*
2421  * Two limits to commit transaction in advance.
2422  *
2423  * For RATIO, it will be 1/RATIO of the remaining limit
2424  * (excluding data and prealloc meta) as threshold.
2425  * For SIZE, it will be in byte unit as threshold.
2426  */
2427 #define QGROUP_PERTRANS_RATIO           32
2428 #define QGROUP_PERTRANS_SIZE            SZ_32M
2429 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2430                                 const struct btrfs_qgroup *qg, u64 num_bytes)
2431 {
2432         u64 limit;
2433         u64 threshold;
2434
2435         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2436             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2437                 return false;
2438
2439         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2440             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2441                 return false;
2442
2443         /*
2444          * Even if we passed the check, it's better to check if reservation
2445          * for meta_pertrans is pushing us near limit.
2446          * If there is too much pertrans reservation or it's near the limit,
2447          * let's try commit transaction to free some, using transaction_kthread
2448          */
2449         if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2450                               BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2451                 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2452                         limit = qg->max_excl;
2453                 else
2454                         limit = qg->max_rfer;
2455                 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2456                             qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2457                             QGROUP_PERTRANS_RATIO;
2458                 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2459
2460                 /*
2461                  * Use transaction_kthread to commit transaction, so we no
2462                  * longer need to bother nested transaction nor lock context.
2463                  */
2464                 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2465                         btrfs_commit_transaction_locksafe(fs_info);
2466         }
2467
2468         return true;
2469 }
2470
2471 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2472                           enum btrfs_qgroup_rsv_type type)
2473 {
2474         struct btrfs_root *quota_root;
2475         struct btrfs_qgroup *qgroup;
2476         struct btrfs_fs_info *fs_info = root->fs_info;
2477         u64 ref_root = root->root_key.objectid;
2478         int ret = 0;
2479         struct ulist_node *unode;
2480         struct ulist_iterator uiter;
2481
2482         if (!is_fstree(ref_root))
2483                 return 0;
2484
2485         if (num_bytes == 0)
2486                 return 0;
2487
2488         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2489             capable(CAP_SYS_RESOURCE))
2490                 enforce = false;
2491
2492         spin_lock(&fs_info->qgroup_lock);
2493         quota_root = fs_info->quota_root;
2494         if (!quota_root)
2495                 goto out;
2496
2497         qgroup = find_qgroup_rb(fs_info, ref_root);
2498         if (!qgroup)
2499                 goto out;
2500
2501         /*
2502          * in a first step, we check all affected qgroups if any limits would
2503          * be exceeded
2504          */
2505         ulist_reinit(fs_info->qgroup_ulist);
2506         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2507                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2508         if (ret < 0)
2509                 goto out;
2510         ULIST_ITER_INIT(&uiter);
2511         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2512                 struct btrfs_qgroup *qg;
2513                 struct btrfs_qgroup_list *glist;
2514
2515                 qg = unode_aux_to_qgroup(unode);
2516
2517                 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2518                         ret = -EDQUOT;
2519                         goto out;
2520                 }
2521
2522                 list_for_each_entry(glist, &qg->groups, next_group) {
2523                         ret = ulist_add(fs_info->qgroup_ulist,
2524                                         glist->group->qgroupid,
2525                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2526                         if (ret < 0)
2527                                 goto out;
2528                 }
2529         }
2530         ret = 0;
2531         /*
2532          * no limits exceeded, now record the reservation into all qgroups
2533          */
2534         ULIST_ITER_INIT(&uiter);
2535         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2536                 struct btrfs_qgroup *qg;
2537
2538                 qg = unode_aux_to_qgroup(unode);
2539
2540                 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2541                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2542         }
2543
2544 out:
2545         spin_unlock(&fs_info->qgroup_lock);
2546         return ret;
2547 }
2548
2549 /*
2550  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2551  * qgroup).
2552  *
2553  * Will handle all higher level qgroup too.
2554  *
2555  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2556  * This special case is only used for META_PERTRANS type.
2557  */
2558 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2559                                u64 ref_root, u64 num_bytes,
2560                                enum btrfs_qgroup_rsv_type type)
2561 {
2562         struct btrfs_root *quota_root;
2563         struct btrfs_qgroup *qgroup;
2564         struct ulist_node *unode;
2565         struct ulist_iterator uiter;
2566         int ret = 0;
2567
2568         if (!is_fstree(ref_root))
2569                 return;
2570
2571         if (num_bytes == 0)
2572                 return;
2573
2574         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2575                 WARN(1, "%s: Invalid type to free", __func__);
2576                 return;
2577         }
2578         spin_lock(&fs_info->qgroup_lock);
2579
2580         quota_root = fs_info->quota_root;
2581         if (!quota_root)
2582                 goto out;
2583
2584         qgroup = find_qgroup_rb(fs_info, ref_root);
2585         if (!qgroup)
2586                 goto out;
2587
2588         if (num_bytes == (u64)-1)
2589                 /*
2590                  * We're freeing all pertrans rsv, get reserved value from
2591                  * level 0 qgroup as real num_bytes to free.
2592                  */
2593                 num_bytes = qgroup->rsv.values[type];
2594
2595         ulist_reinit(fs_info->qgroup_ulist);
2596         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2597                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2598         if (ret < 0)
2599                 goto out;
2600         ULIST_ITER_INIT(&uiter);
2601         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2602                 struct btrfs_qgroup *qg;
2603                 struct btrfs_qgroup_list *glist;
2604
2605                 qg = unode_aux_to_qgroup(unode);
2606
2607                 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2608                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2609
2610                 list_for_each_entry(glist, &qg->groups, next_group) {
2611                         ret = ulist_add(fs_info->qgroup_ulist,
2612                                         glist->group->qgroupid,
2613                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2614                         if (ret < 0)
2615                                 goto out;
2616                 }
2617         }
2618
2619 out:
2620         spin_unlock(&fs_info->qgroup_lock);
2621 }
2622
2623 /*
2624  * Check if the leaf is the last leaf. Which means all node pointers
2625  * are at their last position.
2626  */
2627 static bool is_last_leaf(struct btrfs_path *path)
2628 {
2629         int i;
2630
2631         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2632                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2633                         return false;
2634         }
2635         return true;
2636 }
2637
2638 /*
2639  * returns < 0 on error, 0 when more leafs are to be scanned.
2640  * returns 1 when done.
2641  */
2642 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
2643                               struct btrfs_path *path)
2644 {
2645         struct btrfs_fs_info *fs_info = trans->fs_info;
2646         struct btrfs_key found;
2647         struct extent_buffer *scratch_leaf = NULL;
2648         struct ulist *roots = NULL;
2649         u64 num_bytes;
2650         bool done;
2651         int slot;
2652         int ret;
2653
2654         mutex_lock(&fs_info->qgroup_rescan_lock);
2655         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2656                                          &fs_info->qgroup_rescan_progress,
2657                                          path, 1, 0);
2658
2659         btrfs_debug(fs_info,
2660                 "current progress key (%llu %u %llu), search_slot ret %d",
2661                 fs_info->qgroup_rescan_progress.objectid,
2662                 fs_info->qgroup_rescan_progress.type,
2663                 fs_info->qgroup_rescan_progress.offset, ret);
2664
2665         if (ret) {
2666                 /*
2667                  * The rescan is about to end, we will not be scanning any
2668                  * further blocks. We cannot unset the RESCAN flag here, because
2669                  * we want to commit the transaction if everything went well.
2670                  * To make the live accounting work in this phase, we set our
2671                  * scan progress pointer such that every real extent objectid
2672                  * will be smaller.
2673                  */
2674                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2675                 btrfs_release_path(path);
2676                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2677                 return ret;
2678         }
2679         done = is_last_leaf(path);
2680
2681         btrfs_item_key_to_cpu(path->nodes[0], &found,
2682                               btrfs_header_nritems(path->nodes[0]) - 1);
2683         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2684
2685         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2686         if (!scratch_leaf) {
2687                 ret = -ENOMEM;
2688                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2689                 goto out;
2690         }
2691         extent_buffer_get(scratch_leaf);
2692         btrfs_tree_read_lock(scratch_leaf);
2693         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2694         slot = path->slots[0];
2695         btrfs_release_path(path);
2696         mutex_unlock(&fs_info->qgroup_rescan_lock);
2697
2698         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2699                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2700                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2701                     found.type != BTRFS_METADATA_ITEM_KEY)
2702                         continue;
2703                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2704                         num_bytes = fs_info->nodesize;
2705                 else
2706                         num_bytes = found.offset;
2707
2708                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2709                                            &roots, false);
2710                 if (ret < 0)
2711                         goto out;
2712                 /* For rescan, just pass old_roots as NULL */
2713                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
2714                                                   num_bytes, NULL, roots);
2715                 if (ret < 0)
2716                         goto out;
2717         }
2718 out:
2719         if (scratch_leaf) {
2720                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2721                 free_extent_buffer(scratch_leaf);
2722         }
2723
2724         if (done && !ret) {
2725                 ret = 1;
2726                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2727         }
2728         return ret;
2729 }
2730
2731 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2732 {
2733         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2734                                                      qgroup_rescan_work);
2735         struct btrfs_path *path;
2736         struct btrfs_trans_handle *trans = NULL;
2737         int err = -ENOMEM;
2738         int ret = 0;
2739
2740         path = btrfs_alloc_path();
2741         if (!path)
2742                 goto out;
2743         /*
2744          * Rescan should only search for commit root, and any later difference
2745          * should be recorded by qgroup
2746          */
2747         path->search_commit_root = 1;
2748         path->skip_locking = 1;
2749
2750         err = 0;
2751         while (!err && !btrfs_fs_closing(fs_info)) {
2752                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2753                 if (IS_ERR(trans)) {
2754                         err = PTR_ERR(trans);
2755                         break;
2756                 }
2757                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2758                         err = -EINTR;
2759                 } else {
2760                         err = qgroup_rescan_leaf(trans, path);
2761                 }
2762                 if (err > 0)
2763                         btrfs_commit_transaction(trans);
2764                 else
2765                         btrfs_end_transaction(trans);
2766         }
2767
2768 out:
2769         btrfs_free_path(path);
2770
2771         mutex_lock(&fs_info->qgroup_rescan_lock);
2772         if (!btrfs_fs_closing(fs_info))
2773                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2774
2775         if (err > 0 &&
2776             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2777                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2778         } else if (err < 0) {
2779                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2780         }
2781         mutex_unlock(&fs_info->qgroup_rescan_lock);
2782
2783         /*
2784          * only update status, since the previous part has already updated the
2785          * qgroup info.
2786          */
2787         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2788         if (IS_ERR(trans)) {
2789                 err = PTR_ERR(trans);
2790                 btrfs_err(fs_info,
2791                           "fail to start transaction for status update: %d",
2792                           err);
2793                 goto done;
2794         }
2795         ret = update_qgroup_status_item(trans);
2796         if (ret < 0) {
2797                 err = ret;
2798                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2799         }
2800         btrfs_end_transaction(trans);
2801
2802         if (btrfs_fs_closing(fs_info)) {
2803                 btrfs_info(fs_info, "qgroup scan paused");
2804         } else if (err >= 0) {
2805                 btrfs_info(fs_info, "qgroup scan completed%s",
2806                         err > 0 ? " (inconsistency flag cleared)" : "");
2807         } else {
2808                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2809         }
2810
2811 done:
2812         mutex_lock(&fs_info->qgroup_rescan_lock);
2813         fs_info->qgroup_rescan_running = false;
2814         mutex_unlock(&fs_info->qgroup_rescan_lock);
2815         complete_all(&fs_info->qgroup_rescan_completion);
2816 }
2817
2818 /*
2819  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2820  * memory required for the rescan context.
2821  */
2822 static int
2823 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2824                    int init_flags)
2825 {
2826         int ret = 0;
2827
2828         if (!init_flags) {
2829                 /* we're resuming qgroup rescan at mount time */
2830                 if (!(fs_info->qgroup_flags &
2831                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
2832                         btrfs_warn(fs_info,
2833                         "qgroup rescan init failed, qgroup is not enabled");
2834                         ret = -EINVAL;
2835                 } else if (!(fs_info->qgroup_flags &
2836                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
2837                         btrfs_warn(fs_info,
2838                         "qgroup rescan init failed, qgroup rescan is not queued");
2839                         ret = -EINVAL;
2840                 }
2841
2842                 if (ret)
2843                         return ret;
2844         }
2845
2846         mutex_lock(&fs_info->qgroup_rescan_lock);
2847         spin_lock(&fs_info->qgroup_lock);
2848
2849         if (init_flags) {
2850                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2851                         btrfs_warn(fs_info,
2852                                    "qgroup rescan is already in progress");
2853                         ret = -EINPROGRESS;
2854                 } else if (!(fs_info->qgroup_flags &
2855                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
2856                         btrfs_warn(fs_info,
2857                         "qgroup rescan init failed, qgroup is not enabled");
2858                         ret = -EINVAL;
2859                 }
2860
2861                 if (ret) {
2862                         spin_unlock(&fs_info->qgroup_lock);
2863                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2864                         return ret;
2865                 }
2866                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2867         }
2868
2869         memset(&fs_info->qgroup_rescan_progress, 0,
2870                 sizeof(fs_info->qgroup_rescan_progress));
2871         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2872         init_completion(&fs_info->qgroup_rescan_completion);
2873         fs_info->qgroup_rescan_running = true;
2874
2875         spin_unlock(&fs_info->qgroup_lock);
2876         mutex_unlock(&fs_info->qgroup_rescan_lock);
2877
2878         memset(&fs_info->qgroup_rescan_work, 0,
2879                sizeof(fs_info->qgroup_rescan_work));
2880         btrfs_init_work(&fs_info->qgroup_rescan_work,
2881                         btrfs_qgroup_rescan_helper,
2882                         btrfs_qgroup_rescan_worker, NULL, NULL);
2883         return 0;
2884 }
2885
2886 static void
2887 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2888 {
2889         struct rb_node *n;
2890         struct btrfs_qgroup *qgroup;
2891
2892         spin_lock(&fs_info->qgroup_lock);
2893         /* clear all current qgroup tracking information */
2894         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2895                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2896                 qgroup->rfer = 0;
2897                 qgroup->rfer_cmpr = 0;
2898                 qgroup->excl = 0;
2899                 qgroup->excl_cmpr = 0;
2900         }
2901         spin_unlock(&fs_info->qgroup_lock);
2902 }
2903
2904 int
2905 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2906 {
2907         int ret = 0;
2908         struct btrfs_trans_handle *trans;
2909
2910         ret = qgroup_rescan_init(fs_info, 0, 1);
2911         if (ret)
2912                 return ret;
2913
2914         /*
2915          * We have set the rescan_progress to 0, which means no more
2916          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2917          * However, btrfs_qgroup_account_ref may be right after its call
2918          * to btrfs_find_all_roots, in which case it would still do the
2919          * accounting.
2920          * To solve this, we're committing the transaction, which will
2921          * ensure we run all delayed refs and only after that, we are
2922          * going to clear all tracking information for a clean start.
2923          */
2924
2925         trans = btrfs_join_transaction(fs_info->fs_root);
2926         if (IS_ERR(trans)) {
2927                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2928                 return PTR_ERR(trans);
2929         }
2930         ret = btrfs_commit_transaction(trans);
2931         if (ret) {
2932                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2933                 return ret;
2934         }
2935
2936         qgroup_rescan_zero_tracking(fs_info);
2937
2938         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2939                          &fs_info->qgroup_rescan_work);
2940
2941         return 0;
2942 }
2943
2944 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2945                                      bool interruptible)
2946 {
2947         int running;
2948         int ret = 0;
2949
2950         mutex_lock(&fs_info->qgroup_rescan_lock);
2951         spin_lock(&fs_info->qgroup_lock);
2952         running = fs_info->qgroup_rescan_running;
2953         spin_unlock(&fs_info->qgroup_lock);
2954         mutex_unlock(&fs_info->qgroup_rescan_lock);
2955
2956         if (!running)
2957                 return 0;
2958
2959         if (interruptible)
2960                 ret = wait_for_completion_interruptible(
2961                                         &fs_info->qgroup_rescan_completion);
2962         else
2963                 wait_for_completion(&fs_info->qgroup_rescan_completion);
2964
2965         return ret;
2966 }
2967
2968 /*
2969  * this is only called from open_ctree where we're still single threaded, thus
2970  * locking is omitted here.
2971  */
2972 void
2973 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2974 {
2975         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2976                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2977                                  &fs_info->qgroup_rescan_work);
2978 }
2979
2980 /*
2981  * Reserve qgroup space for range [start, start + len).
2982  *
2983  * This function will either reserve space from related qgroups or doing
2984  * nothing if the range is already reserved.
2985  *
2986  * Return 0 for successful reserve
2987  * Return <0 for error (including -EQUOT)
2988  *
2989  * NOTE: this function may sleep for memory allocation.
2990  *       if btrfs_qgroup_reserve_data() is called multiple times with
2991  *       same @reserved, caller must ensure when error happens it's OK
2992  *       to free *ALL* reserved space.
2993  */
2994 int btrfs_qgroup_reserve_data(struct inode *inode,
2995                         struct extent_changeset **reserved_ret, u64 start,
2996                         u64 len)
2997 {
2998         struct btrfs_root *root = BTRFS_I(inode)->root;
2999         struct ulist_node *unode;
3000         struct ulist_iterator uiter;
3001         struct extent_changeset *reserved;
3002         u64 orig_reserved;
3003         u64 to_reserve;
3004         int ret;
3005
3006         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3007             !is_fstree(root->objectid) || len == 0)
3008                 return 0;
3009
3010         /* @reserved parameter is mandatory for qgroup */
3011         if (WARN_ON(!reserved_ret))
3012                 return -EINVAL;
3013         if (!*reserved_ret) {
3014                 *reserved_ret = extent_changeset_alloc();
3015                 if (!*reserved_ret)
3016                         return -ENOMEM;
3017         }
3018         reserved = *reserved_ret;
3019         /* Record already reserved space */
3020         orig_reserved = reserved->bytes_changed;
3021         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3022                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3023
3024         /* Newly reserved space */
3025         to_reserve = reserved->bytes_changed - orig_reserved;
3026         trace_btrfs_qgroup_reserve_data(inode, start, len,
3027                                         to_reserve, QGROUP_RESERVE);
3028         if (ret < 0)
3029                 goto cleanup;
3030         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3031         if (ret < 0)
3032                 goto cleanup;
3033
3034         return ret;
3035
3036 cleanup:
3037         /* cleanup *ALL* already reserved ranges */
3038         ULIST_ITER_INIT(&uiter);
3039         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3040                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3041                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3042         extent_changeset_release(reserved);
3043         return ret;
3044 }
3045
3046 /* Free ranges specified by @reserved, normally in error path */
3047 static int qgroup_free_reserved_data(struct inode *inode,
3048                         struct extent_changeset *reserved, u64 start, u64 len)
3049 {
3050         struct btrfs_root *root = BTRFS_I(inode)->root;
3051         struct ulist_node *unode;
3052         struct ulist_iterator uiter;
3053         struct extent_changeset changeset;
3054         int freed = 0;
3055         int ret;
3056
3057         extent_changeset_init(&changeset);
3058         len = round_up(start + len, root->fs_info->sectorsize);
3059         start = round_down(start, root->fs_info->sectorsize);
3060
3061         ULIST_ITER_INIT(&uiter);
3062         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3063                 u64 range_start = unode->val;
3064                 /* unode->aux is the inclusive end */
3065                 u64 range_len = unode->aux - range_start + 1;
3066                 u64 free_start;
3067                 u64 free_len;
3068
3069                 extent_changeset_release(&changeset);
3070
3071                 /* Only free range in range [start, start + len) */
3072                 if (range_start >= start + len ||
3073                     range_start + range_len <= start)
3074                         continue;
3075                 free_start = max(range_start, start);
3076                 free_len = min(start + len, range_start + range_len) -
3077                            free_start;
3078                 /*
3079                  * TODO: To also modify reserved->ranges_reserved to reflect
3080                  * the modification.
3081                  *
3082                  * However as long as we free qgroup reserved according to
3083                  * EXTENT_QGROUP_RESERVED, we won't double free.
3084                  * So not need to rush.
3085                  */
3086                 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3087                                 free_start, free_start + free_len - 1,
3088                                 EXTENT_QGROUP_RESERVED, &changeset);
3089                 if (ret < 0)
3090                         goto out;
3091                 freed += changeset.bytes_changed;
3092         }
3093         btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed,
3094                                   BTRFS_QGROUP_RSV_DATA);
3095         ret = freed;
3096 out:
3097         extent_changeset_release(&changeset);
3098         return ret;
3099 }
3100
3101 static int __btrfs_qgroup_release_data(struct inode *inode,
3102                         struct extent_changeset *reserved, u64 start, u64 len,
3103                         int free)
3104 {
3105         struct extent_changeset changeset;
3106         int trace_op = QGROUP_RELEASE;
3107         int ret;
3108
3109         /* In release case, we shouldn't have @reserved */
3110         WARN_ON(!free && reserved);
3111         if (free && reserved)
3112                 return qgroup_free_reserved_data(inode, reserved, start, len);
3113         extent_changeset_init(&changeset);
3114         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3115                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3116         if (ret < 0)
3117                 goto out;
3118
3119         if (free)
3120                 trace_op = QGROUP_FREE;
3121         trace_btrfs_qgroup_release_data(inode, start, len,
3122                                         changeset.bytes_changed, trace_op);
3123         if (free)
3124                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3125                                 BTRFS_I(inode)->root->objectid,
3126                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3127         ret = changeset.bytes_changed;
3128 out:
3129         extent_changeset_release(&changeset);
3130         return ret;
3131 }
3132
3133 /*
3134  * Free a reserved space range from io_tree and related qgroups
3135  *
3136  * Should be called when a range of pages get invalidated before reaching disk.
3137  * Or for error cleanup case.
3138  * if @reserved is given, only reserved range in [@start, @start + @len) will
3139  * be freed.
3140  *
3141  * For data written to disk, use btrfs_qgroup_release_data().
3142  *
3143  * NOTE: This function may sleep for memory allocation.
3144  */
3145 int btrfs_qgroup_free_data(struct inode *inode,
3146                         struct extent_changeset *reserved, u64 start, u64 len)
3147 {
3148         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3149 }
3150
3151 /*
3152  * Release a reserved space range from io_tree only.
3153  *
3154  * Should be called when a range of pages get written to disk and corresponding
3155  * FILE_EXTENT is inserted into corresponding root.
3156  *
3157  * Since new qgroup accounting framework will only update qgroup numbers at
3158  * commit_transaction() time, its reserved space shouldn't be freed from
3159  * related qgroups.
3160  *
3161  * But we should release the range from io_tree, to allow further write to be
3162  * COWed.
3163  *
3164  * NOTE: This function may sleep for memory allocation.
3165  */
3166 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3167 {
3168         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3169 }
3170
3171 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3172                               enum btrfs_qgroup_rsv_type type)
3173 {
3174         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3175             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3176                 return;
3177         if (num_bytes == 0)
3178                 return;
3179
3180         spin_lock(&root->qgroup_meta_rsv_lock);
3181         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3182                 root->qgroup_meta_rsv_prealloc += num_bytes;
3183         else
3184                 root->qgroup_meta_rsv_pertrans += num_bytes;
3185         spin_unlock(&root->qgroup_meta_rsv_lock);
3186 }
3187
3188 static int sub_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 0;
3194         if (num_bytes == 0)
3195                 return 0;
3196
3197         spin_lock(&root->qgroup_meta_rsv_lock);
3198         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3199                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3200                                   num_bytes);
3201                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3202         } else {
3203                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3204                                   num_bytes);
3205                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3206         }
3207         spin_unlock(&root->qgroup_meta_rsv_lock);
3208         return num_bytes;
3209 }
3210
3211 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3212                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3213 {
3214         struct btrfs_fs_info *fs_info = root->fs_info;
3215         int ret;
3216
3217         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3218             !is_fstree(root->objectid) || num_bytes == 0)
3219                 return 0;
3220
3221         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3222         trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3223         ret = qgroup_reserve(root, num_bytes, enforce, type);
3224         if (ret < 0)
3225                 return ret;
3226         /*
3227          * Record what we have reserved into root.
3228          *
3229          * To avoid quota disabled->enabled underflow.
3230          * In that case, we may try to free space we haven't reserved
3231          * (since quota was disabled), so record what we reserved into root.
3232          * And ensure later release won't underflow this number.
3233          */
3234         add_root_meta_rsv(root, num_bytes, type);
3235         return ret;
3236 }
3237
3238 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3239 {
3240         struct btrfs_fs_info *fs_info = root->fs_info;
3241
3242         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3243             !is_fstree(root->objectid))
3244                 return;
3245
3246         /* TODO: Update trace point to handle such free */
3247         trace_qgroup_meta_free_all_pertrans(root);
3248         /* Special value -1 means to free all reserved space */
3249         btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1,
3250                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3251 }
3252
3253 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3254                               enum btrfs_qgroup_rsv_type type)
3255 {
3256         struct btrfs_fs_info *fs_info = root->fs_info;
3257
3258         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3259             !is_fstree(root->objectid))
3260                 return;
3261
3262         /*
3263          * reservation for META_PREALLOC can happen before quota is enabled,
3264          * which can lead to underflow.
3265          * Here ensure we will only free what we really have reserved.
3266          */
3267         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3268         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3269         trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3270         btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type);
3271 }
3272
3273 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3274                                 int num_bytes)
3275 {
3276         struct btrfs_root *quota_root = fs_info->quota_root;
3277         struct btrfs_qgroup *qgroup;
3278         struct ulist_node *unode;
3279         struct ulist_iterator uiter;
3280         int ret = 0;
3281
3282         if (num_bytes == 0)
3283                 return;
3284         if (!quota_root)
3285                 return;
3286
3287         spin_lock(&fs_info->qgroup_lock);
3288         qgroup = find_qgroup_rb(fs_info, ref_root);
3289         if (!qgroup)
3290                 goto out;
3291         ulist_reinit(fs_info->qgroup_ulist);
3292         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3293                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3294         if (ret < 0)
3295                 goto out;
3296         ULIST_ITER_INIT(&uiter);
3297         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3298                 struct btrfs_qgroup *qg;
3299                 struct btrfs_qgroup_list *glist;
3300
3301                 qg = unode_aux_to_qgroup(unode);
3302
3303                 qgroup_rsv_release(fs_info, qg, num_bytes,
3304                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3305                 qgroup_rsv_add(fs_info, qg, num_bytes,
3306                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3307                 list_for_each_entry(glist, &qg->groups, next_group) {
3308                         ret = ulist_add(fs_info->qgroup_ulist,
3309                                         glist->group->qgroupid,
3310                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3311                         if (ret < 0)
3312                                 goto out;
3313                 }
3314         }
3315 out:
3316         spin_unlock(&fs_info->qgroup_lock);
3317 }
3318
3319 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3320 {
3321         struct btrfs_fs_info *fs_info = root->fs_info;
3322
3323         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3324             !is_fstree(root->objectid))
3325                 return;
3326         /* Same as btrfs_qgroup_free_meta_prealloc() */
3327         num_bytes = sub_root_meta_rsv(root, num_bytes,
3328                                       BTRFS_QGROUP_RSV_META_PREALLOC);
3329         trace_qgroup_meta_convert(root, num_bytes);
3330         qgroup_convert_meta(fs_info, root->objectid, num_bytes);
3331 }
3332
3333 /*
3334  * Check qgroup reserved space leaking, normally at destroy inode
3335  * time
3336  */
3337 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3338 {
3339         struct extent_changeset changeset;
3340         struct ulist_node *unode;
3341         struct ulist_iterator iter;
3342         int ret;
3343
3344         extent_changeset_init(&changeset);
3345         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3346                         EXTENT_QGROUP_RESERVED, &changeset);
3347
3348         WARN_ON(ret < 0);
3349         if (WARN_ON(changeset.bytes_changed)) {
3350                 ULIST_ITER_INIT(&iter);
3351                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3352                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3353                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3354                                 inode->i_ino, unode->val, unode->aux);
3355                 }
3356                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3357                                 BTRFS_I(inode)->root->objectid,
3358                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3359
3360         }
3361         extent_changeset_release(&changeset);
3362 }