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 #include "block-group.h"
  25
  26 /* TODO XXX FIXME
  27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  28  *  - reorganize keys
  29  *  - compressed
  30  *  - sync
  31  *  - copy also limits on subvol creation
  32  *  - limit
  33  *  - caches for ulists
  34  *  - performance benchmarks
  35  *  - check all ioctl parameters
  36  */
  37
  38 /*
  39  * Helpers to access qgroup reservation
  40  *
  41  * Callers should ensure the lock context and type are valid
  42  */
  43
  44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
  45 {
  46         u64 ret = 0;
  47         int i;
  48
  49         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
  50                 ret += qgroup->rsv.values[i];
  51
  52         return ret;
  53 }
  54
  55 #ifdef CONFIG_BTRFS_DEBUG
  56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
  57 {
  58         if (type == BTRFS_QGROUP_RSV_DATA)
  59                 return "data";
  60         if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
  61                 return "meta_pertrans";
  62         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
  63                 return "meta_prealloc";
  64         return NULL;
  65 }
  66 #endif
  67
  68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
  69                            struct btrfs_qgroup *qgroup, u64 num_bytes,
  70                            enum btrfs_qgroup_rsv_type type)
  71 {
  72         trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
  73         qgroup->rsv.values[type] += num_bytes;
  74 }
  75
  76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
  77                                struct btrfs_qgroup *qgroup, u64 num_bytes,
  78                                enum btrfs_qgroup_rsv_type type)
  79 {
  80         trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
  81         if (qgroup->rsv.values[type] >= num_bytes) {
  82                 qgroup->rsv.values[type] -= num_bytes;
  83                 return;
  84         }
  85 #ifdef CONFIG_BTRFS_DEBUG
  86         WARN_RATELIMIT(1,
  87                 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
  88                 qgroup->qgroupid, qgroup_rsv_type_str(type),
  89                 qgroup->rsv.values[type], num_bytes);
  90 #endif
  91         qgroup->rsv.values[type] = 0;
  92 }
  93
  94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
  95                                      struct btrfs_qgroup *dest,
  96                                      struct btrfs_qgroup *src)
  97 {
  98         int i;
  99
 100         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 101                 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
 102 }
 103
 104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
 105                                          struct btrfs_qgroup *dest,
 106                                           struct btrfs_qgroup *src)
 107 {
 108         int i;
 109
 110         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 111                 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
 112 }
 113
 114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
 115                                            int mod)
 116 {
 117         if (qg->old_refcnt < seq)
 118                 qg->old_refcnt = seq;
 119         qg->old_refcnt += mod;
 120 }
 121
 122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
 123                                            int mod)
 124 {
 125         if (qg->new_refcnt < seq)
 126                 qg->new_refcnt = seq;
 127         qg->new_refcnt += mod;
 128 }
 129
 130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
 131 {
 132         if (qg->old_refcnt < seq)
 133                 return 0;
 134         return qg->old_refcnt - seq;
 135 }
 136
 137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
 138 {
 139         if (qg->new_refcnt < seq)
 140                 return 0;
 141         return qg->new_refcnt - seq;
 142 }
 143
 144 /*
 145  * glue structure to represent the relations between qgroups.
 146  */
 147 struct btrfs_qgroup_list {
 148         struct list_head next_group;
 149         struct list_head next_member;
 150         struct btrfs_qgroup *group;
 151         struct btrfs_qgroup *member;
 152 };
 153
 154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
 155 {
 156         return (u64)(uintptr_t)qg;
 157 }
 158
 159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
 160 {
 161         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
 162 }
 163
 164 static int
 165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 166                    int init_flags);
 167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 168
 169 /* must be called with qgroup_ioctl_lock held */
 170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 171                                            u64 qgroupid)
 172 {
 173         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 174         struct btrfs_qgroup *qgroup;
 175
 176         while (n) {
 177                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 178                 if (qgroup->qgroupid < qgroupid)
 179                         n = n->rb_left;
 180                 else if (qgroup->qgroupid > qgroupid)
 181                         n = n->rb_right;
 182                 else
 183                         return qgroup;
 184         }
 185         return NULL;
 186 }
 187
 188 /* must be called with qgroup_lock held */
 189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 190                                           u64 qgroupid)
 191 {
 192         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 193         struct rb_node *parent = NULL;
 194         struct btrfs_qgroup *qgroup;
 195
 196         while (*p) {
 197                 parent = *p;
 198                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 199
 200                 if (qgroup->qgroupid < qgroupid)
 201                         p = &(*p)->rb_left;
 202                 else if (qgroup->qgroupid > qgroupid)
 203                         p = &(*p)->rb_right;
 204                 else
 205                         return qgroup;
 206         }
 207
 208         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 209         if (!qgroup)
 210                 return ERR_PTR(-ENOMEM);
 211
 212         qgroup->qgroupid = qgroupid;
 213         INIT_LIST_HEAD(&qgroup->groups);
 214         INIT_LIST_HEAD(&qgroup->members);
 215         INIT_LIST_HEAD(&qgroup->dirty);
 216
 217         rb_link_node(&qgroup->node, parent, p);
 218         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 219
 220         return qgroup;
 221 }
 222
 223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
 224 {
 225         struct btrfs_qgroup_list *list;
 226
 227         list_del(&qgroup->dirty);
 228         while (!list_empty(&qgroup->groups)) {
 229                 list = list_first_entry(&qgroup->groups,
 230                                         struct btrfs_qgroup_list, next_group);
 231                 list_del(&list->next_group);
 232                 list_del(&list->next_member);
 233                 kfree(list);
 234         }
 235
 236         while (!list_empty(&qgroup->members)) {
 237                 list = list_first_entry(&qgroup->members,
 238                                         struct btrfs_qgroup_list, next_member);
 239                 list_del(&list->next_group);
 240                 list_del(&list->next_member);
 241                 kfree(list);
 242         }
 243         kfree(qgroup);
 244 }
 245
 246 /* must be called with qgroup_lock held */
 247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 248 {
 249         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 250
 251         if (!qgroup)
 252                 return -ENOENT;
 253
 254         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 255         __del_qgroup_rb(qgroup);
 256         return 0;
 257 }
 258
 259 /* must be called with qgroup_lock held */
 260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 261                            u64 memberid, u64 parentid)
 262 {
 263         struct btrfs_qgroup *member;
 264         struct btrfs_qgroup *parent;
 265         struct btrfs_qgroup_list *list;
 266
 267         member = find_qgroup_rb(fs_info, memberid);
 268         parent = find_qgroup_rb(fs_info, parentid);
 269         if (!member || !parent)
 270                 return -ENOENT;
 271
 272         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 273         if (!list)
 274                 return -ENOMEM;
 275
 276         list->group = parent;
 277         list->member = member;
 278         list_add_tail(&list->next_group, &member->groups);
 279         list_add_tail(&list->next_member, &parent->members);
 280
 281         return 0;
 282 }
 283
 284 /* must be called with qgroup_lock held */
 285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 286                            u64 memberid, u64 parentid)
 287 {
 288         struct btrfs_qgroup *member;
 289         struct btrfs_qgroup *parent;
 290         struct btrfs_qgroup_list *list;
 291
 292         member = find_qgroup_rb(fs_info, memberid);
 293         parent = find_qgroup_rb(fs_info, parentid);
 294         if (!member || !parent)
 295                 return -ENOENT;
 296
 297         list_for_each_entry(list, &member->groups, next_group) {
 298                 if (list->group == parent) {
 299                         list_del(&list->next_group);
 300                         list_del(&list->next_member);
 301                         kfree(list);
 302                         return 0;
 303                 }
 304         }
 305         return -ENOENT;
 306 }
 307
 308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 310                                u64 rfer, u64 excl)
 311 {
 312         struct btrfs_qgroup *qgroup;
 313
 314         qgroup = find_qgroup_rb(fs_info, qgroupid);
 315         if (!qgroup)
 316                 return -EINVAL;
 317         if (qgroup->rfer != rfer || qgroup->excl != excl)
 318                 return -EINVAL;
 319         return 0;
 320 }
 321 #endif
 322
 323 /*
 324  * The full config is read in one go, only called from open_ctree()
 325  * It doesn't use any locking, as at this point we're still single-threaded
 326  */
 327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 328 {
 329         struct btrfs_key key;
 330         struct btrfs_key found_key;
 331         struct btrfs_root *quota_root = fs_info->quota_root;
 332         struct btrfs_path *path = NULL;
 333         struct extent_buffer *l;
 334         int slot;
 335         int ret = 0;
 336         u64 flags = 0;
 337         u64 rescan_progress = 0;
 338
 339         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 340                 return 0;
 341
 342         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 343         if (!fs_info->qgroup_ulist) {
 344                 ret = -ENOMEM;
 345                 goto out;
 346         }
 347
 348         path = btrfs_alloc_path();
 349         if (!path) {
 350                 ret = -ENOMEM;
 351                 goto out;
 352         }
 353
 354         /* default this to quota off, in case no status key is found */
 355         fs_info->qgroup_flags = 0;
 356
 357         /*
 358          * pass 1: read status, all qgroup infos and limits
 359          */
 360         key.objectid = 0;
 361         key.type = 0;
 362         key.offset = 0;
 363         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 364         if (ret)
 365                 goto out;
 366
 367         while (1) {
 368                 struct btrfs_qgroup *qgroup;
 369
 370                 slot = path->slots[0];
 371                 l = path->nodes[0];
 372                 btrfs_item_key_to_cpu(l, &found_key, slot);
 373
 374                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 375                         struct btrfs_qgroup_status_item *ptr;
 376
 377                         ptr = btrfs_item_ptr(l, slot,
 378                                              struct btrfs_qgroup_status_item);
 379
 380                         if (btrfs_qgroup_status_version(l, ptr) !=
 381                             BTRFS_QGROUP_STATUS_VERSION) {
 382                                 btrfs_err(fs_info,
 383                                  "old qgroup version, quota disabled");
 384                                 goto out;
 385                         }
 386                         if (btrfs_qgroup_status_generation(l, ptr) !=
 387                             fs_info->generation) {
 388                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 389                                 btrfs_err(fs_info,
 390                                         "qgroup generation mismatch, marked as inconsistent");
 391                         }
 392                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 393                                                                           ptr);
 394                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 395                         goto next1;
 396                 }
 397
 398                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 399                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 400                         goto next1;
 401
 402                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 403                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 404                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 405                         btrfs_err(fs_info, "inconsistent qgroup config");
 406                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 407                 }
 408                 if (!qgroup) {
 409                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 410                         if (IS_ERR(qgroup)) {
 411                                 ret = PTR_ERR(qgroup);
 412                                 goto out;
 413                         }
 414                 }
 415                 switch (found_key.type) {
 416                 case BTRFS_QGROUP_INFO_KEY: {
 417                         struct btrfs_qgroup_info_item *ptr;
 418
 419                         ptr = btrfs_item_ptr(l, slot,
 420                                              struct btrfs_qgroup_info_item);
 421                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 422                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 423                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 424                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 425                         /* generation currently unused */
 426                         break;
 427                 }
 428                 case BTRFS_QGROUP_LIMIT_KEY: {
 429                         struct btrfs_qgroup_limit_item *ptr;
 430
 431                         ptr = btrfs_item_ptr(l, slot,
 432                                              struct btrfs_qgroup_limit_item);
 433                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 434                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 435                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 436                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 437                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 438                         break;
 439                 }
 440                 }
 441 next1:
 442                 ret = btrfs_next_item(quota_root, path);
 443                 if (ret < 0)
 444                         goto out;
 445                 if (ret)
 446                         break;
 447         }
 448         btrfs_release_path(path);
 449
 450         /*
 451          * pass 2: read all qgroup relations
 452          */
 453         key.objectid = 0;
 454         key.type = BTRFS_QGROUP_RELATION_KEY;
 455         key.offset = 0;
 456         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 457         if (ret)
 458                 goto out;
 459         while (1) {
 460                 slot = path->slots[0];
 461                 l = path->nodes[0];
 462                 btrfs_item_key_to_cpu(l, &found_key, slot);
 463
 464                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 465                         goto next2;
 466
 467                 if (found_key.objectid > found_key.offset) {
 468                         /* parent <- member, not needed to build config */
 469                         /* FIXME should we omit the key completely? */
 470                         goto next2;
 471                 }
 472
 473                 ret = add_relation_rb(fs_info, found_key.objectid,
 474                                       found_key.offset);
 475                 if (ret == -ENOENT) {
 476                         btrfs_warn(fs_info,
 477                                 "orphan qgroup relation 0x%llx->0x%llx",
 478                                 found_key.objectid, found_key.offset);
 479                         ret = 0;        /* ignore the error */
 480                 }
 481                 if (ret)
 482                         goto out;
 483 next2:
 484                 ret = btrfs_next_item(quota_root, path);
 485                 if (ret < 0)
 486                         goto out;
 487                 if (ret)
 488                         break;
 489         }
 490 out:
 491         fs_info->qgroup_flags |= flags;
 492         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
 493                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 494         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 495                  ret >= 0)
 496                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 497         btrfs_free_path(path);
 498
 499         if (ret < 0) {
 500                 ulist_free(fs_info->qgroup_ulist);
 501                 fs_info->qgroup_ulist = NULL;
 502                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 503         }
 504
 505         return ret < 0 ? ret : 0;
 506 }
 507
 508 static u64 btrfs_qgroup_subvolid(u64 qgroupid)
 509 {
 510         return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
 511 }
 512
 513 /*
 514  * Called in close_ctree() when quota is still enabled.  This verifies we don't
 515  * leak some reserved space.
 516  *
 517  * Return false if no reserved space is left.
 518  * Return true if some reserved space is leaked.
 519  */
 520 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
 521 {
 522         struct rb_node *node;
 523         bool ret = false;
 524
 525         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 526                 return ret;
 527         /*
 528          * Since we're unmounting, there is no race and no need to grab qgroup
 529          * lock.  And here we don't go post-order to provide a more user
 530          * friendly sorted result.
 531          */
 532         for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
 533                 struct btrfs_qgroup *qgroup;
 534                 int i;
 535
 536                 qgroup = rb_entry(node, struct btrfs_qgroup, node);
 537                 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
 538                         if (qgroup->rsv.values[i]) {
 539                                 ret = true;
 540                                 btrfs_warn(fs_info,
 541                 "qgroup %llu/%llu has unreleased space, type %d rsv %llu",
 542                                    btrfs_qgroup_level(qgroup->qgroupid),
 543                                    btrfs_qgroup_subvolid(qgroup->qgroupid),
 544                                    i, qgroup->rsv.values[i]);
 545                         }
 546                 }
 547         }
 548         return ret;
 549 }
 550
 551 /*
 552  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 553  * first two are in single-threaded paths.And for the third one, we have set
 554  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 555  * up the in-memory structures without qgroup_lock held.
 556  */
 557 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 558 {
 559         struct rb_node *n;
 560         struct btrfs_qgroup *qgroup;
 561
 562         while ((n = rb_first(&fs_info->qgroup_tree))) {
 563                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 564                 rb_erase(n, &fs_info->qgroup_tree);
 565                 __del_qgroup_rb(qgroup);
 566         }
 567         /*
 568          * We call btrfs_free_qgroup_config() when unmounting
 569          * filesystem and disabling quota, so we set qgroup_ulist
 570          * to be null here to avoid double free.
 571          */
 572         ulist_free(fs_info->qgroup_ulist);
 573         fs_info->qgroup_ulist = NULL;
 574 }
 575
 576 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 577                                     u64 dst)
 578 {
 579         int ret;
 580         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 581         struct btrfs_path *path;
 582         struct btrfs_key key;
 583
 584         path = btrfs_alloc_path();
 585         if (!path)
 586                 return -ENOMEM;
 587
 588         key.objectid = src;
 589         key.type = BTRFS_QGROUP_RELATION_KEY;
 590         key.offset = dst;
 591
 592         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 593
 594         btrfs_mark_buffer_dirty(path->nodes[0]);
 595
 596         btrfs_free_path(path);
 597         return ret;
 598 }
 599
 600 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 601                                     u64 dst)
 602 {
 603         int ret;
 604         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 605         struct btrfs_path *path;
 606         struct btrfs_key key;
 607
 608         path = btrfs_alloc_path();
 609         if (!path)
 610                 return -ENOMEM;
 611
 612         key.objectid = src;
 613         key.type = BTRFS_QGROUP_RELATION_KEY;
 614         key.offset = dst;
 615
 616         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 617         if (ret < 0)
 618                 goto out;
 619
 620         if (ret > 0) {
 621                 ret = -ENOENT;
 622                 goto out;
 623         }
 624
 625         ret = btrfs_del_item(trans, quota_root, path);
 626 out:
 627         btrfs_free_path(path);
 628         return ret;
 629 }
 630
 631 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 632                            struct btrfs_root *quota_root, u64 qgroupid)
 633 {
 634         int ret;
 635         struct btrfs_path *path;
 636         struct btrfs_qgroup_info_item *qgroup_info;
 637         struct btrfs_qgroup_limit_item *qgroup_limit;
 638         struct extent_buffer *leaf;
 639         struct btrfs_key key;
 640
 641         if (btrfs_is_testing(quota_root->fs_info))
 642                 return 0;
 643
 644         path = btrfs_alloc_path();
 645         if (!path)
 646                 return -ENOMEM;
 647
 648         key.objectid = 0;
 649         key.type = BTRFS_QGROUP_INFO_KEY;
 650         key.offset = qgroupid;
 651
 652         /*
 653          * Avoid a transaction abort by catching -EEXIST here. In that
 654          * case, we proceed by re-initializing the existing structure
 655          * on disk.
 656          */
 657
 658         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 659                                       sizeof(*qgroup_info));
 660         if (ret && ret != -EEXIST)
 661                 goto out;
 662
 663         leaf = path->nodes[0];
 664         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 665                                  struct btrfs_qgroup_info_item);
 666         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 667         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 668         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 669         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 670         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 671
 672         btrfs_mark_buffer_dirty(leaf);
 673
 674         btrfs_release_path(path);
 675
 676         key.type = BTRFS_QGROUP_LIMIT_KEY;
 677         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 678                                       sizeof(*qgroup_limit));
 679         if (ret && ret != -EEXIST)
 680                 goto out;
 681
 682         leaf = path->nodes[0];
 683         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 684                                   struct btrfs_qgroup_limit_item);
 685         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 686         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 687         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 688         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 689         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 690
 691         btrfs_mark_buffer_dirty(leaf);
 692
 693         ret = 0;
 694 out:
 695         btrfs_free_path(path);
 696         return ret;
 697 }
 698
 699 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
 700 {
 701         int ret;
 702         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 703         struct btrfs_path *path;
 704         struct btrfs_key key;
 705
 706         path = btrfs_alloc_path();
 707         if (!path)
 708                 return -ENOMEM;
 709
 710         key.objectid = 0;
 711         key.type = BTRFS_QGROUP_INFO_KEY;
 712         key.offset = qgroupid;
 713         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 714         if (ret < 0)
 715                 goto out;
 716
 717         if (ret > 0) {
 718                 ret = -ENOENT;
 719                 goto out;
 720         }
 721
 722         ret = btrfs_del_item(trans, quota_root, path);
 723         if (ret)
 724                 goto out;
 725
 726         btrfs_release_path(path);
 727
 728         key.type = BTRFS_QGROUP_LIMIT_KEY;
 729         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 730         if (ret < 0)
 731                 goto out;
 732
 733         if (ret > 0) {
 734                 ret = -ENOENT;
 735                 goto out;
 736         }
 737
 738         ret = btrfs_del_item(trans, quota_root, path);
 739
 740 out:
 741         btrfs_free_path(path);
 742         return ret;
 743 }
 744
 745 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 746                                     struct btrfs_qgroup *qgroup)
 747 {
 748         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 749         struct btrfs_path *path;
 750         struct btrfs_key key;
 751         struct extent_buffer *l;
 752         struct btrfs_qgroup_limit_item *qgroup_limit;
 753         int ret;
 754         int slot;
 755
 756         key.objectid = 0;
 757         key.type = BTRFS_QGROUP_LIMIT_KEY;
 758         key.offset = qgroup->qgroupid;
 759
 760         path = btrfs_alloc_path();
 761         if (!path)
 762                 return -ENOMEM;
 763
 764         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 765         if (ret > 0)
 766                 ret = -ENOENT;
 767
 768         if (ret)
 769                 goto out;
 770
 771         l = path->nodes[0];
 772         slot = path->slots[0];
 773         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 774         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 775         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 776         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 777         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 778         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 779
 780         btrfs_mark_buffer_dirty(l);
 781
 782 out:
 783         btrfs_free_path(path);
 784         return ret;
 785 }
 786
 787 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 788                                    struct btrfs_qgroup *qgroup)
 789 {
 790         struct btrfs_fs_info *fs_info = trans->fs_info;
 791         struct btrfs_root *quota_root = fs_info->quota_root;
 792         struct btrfs_path *path;
 793         struct btrfs_key key;
 794         struct extent_buffer *l;
 795         struct btrfs_qgroup_info_item *qgroup_info;
 796         int ret;
 797         int slot;
 798
 799         if (btrfs_is_testing(fs_info))
 800                 return 0;
 801
 802         key.objectid = 0;
 803         key.type = BTRFS_QGROUP_INFO_KEY;
 804         key.offset = qgroup->qgroupid;
 805
 806         path = btrfs_alloc_path();
 807         if (!path)
 808                 return -ENOMEM;
 809
 810         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 811         if (ret > 0)
 812                 ret = -ENOENT;
 813
 814         if (ret)
 815                 goto out;
 816
 817         l = path->nodes[0];
 818         slot = path->slots[0];
 819         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 820         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 821         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 822         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 823         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 824         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 825
 826         btrfs_mark_buffer_dirty(l);
 827
 828 out:
 829         btrfs_free_path(path);
 830         return ret;
 831 }
 832
 833 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
 834 {
 835         struct btrfs_fs_info *fs_info = trans->fs_info;
 836         struct btrfs_root *quota_root = fs_info->quota_root;
 837         struct btrfs_path *path;
 838         struct btrfs_key key;
 839         struct extent_buffer *l;
 840         struct btrfs_qgroup_status_item *ptr;
 841         int ret;
 842         int slot;
 843
 844         key.objectid = 0;
 845         key.type = BTRFS_QGROUP_STATUS_KEY;
 846         key.offset = 0;
 847
 848         path = btrfs_alloc_path();
 849         if (!path)
 850                 return -ENOMEM;
 851
 852         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 853         if (ret > 0)
 854                 ret = -ENOENT;
 855
 856         if (ret)
 857                 goto out;
 858
 859         l = path->nodes[0];
 860         slot = path->slots[0];
 861         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 862         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 863         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 864         btrfs_set_qgroup_status_rescan(l, ptr,
 865                                 fs_info->qgroup_rescan_progress.objectid);
 866
 867         btrfs_mark_buffer_dirty(l);
 868
 869 out:
 870         btrfs_free_path(path);
 871         return ret;
 872 }
 873
 874 /*
 875  * called with qgroup_lock held
 876  */
 877 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 878                                   struct btrfs_root *root)
 879 {
 880         struct btrfs_path *path;
 881         struct btrfs_key key;
 882         struct extent_buffer *leaf = NULL;
 883         int ret;
 884         int nr = 0;
 885
 886         path = btrfs_alloc_path();
 887         if (!path)
 888                 return -ENOMEM;
 889
 890         path->leave_spinning = 1;
 891
 892         key.objectid = 0;
 893         key.offset = 0;
 894         key.type = 0;
 895
 896         while (1) {
 897                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 898                 if (ret < 0)
 899                         goto out;
 900                 leaf = path->nodes[0];
 901                 nr = btrfs_header_nritems(leaf);
 902                 if (!nr)
 903                         break;
 904                 /*
 905                  * delete the leaf one by one
 906                  * since the whole tree is going
 907                  * to be deleted.
 908                  */
 909                 path->slots[0] = 0;
 910                 ret = btrfs_del_items(trans, root, path, 0, nr);
 911                 if (ret)
 912                         goto out;
 913
 914                 btrfs_release_path(path);
 915         }
 916         ret = 0;
 917 out:
 918         btrfs_free_path(path);
 919         return ret;
 920 }
 921
 922 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 923 {
 924         struct btrfs_root *quota_root;
 925         struct btrfs_root *tree_root = fs_info->tree_root;
 926         struct btrfs_path *path = NULL;
 927         struct btrfs_qgroup_status_item *ptr;
 928         struct extent_buffer *leaf;
 929         struct btrfs_key key;
 930         struct btrfs_key found_key;
 931         struct btrfs_qgroup *qgroup = NULL;
 932         struct btrfs_trans_handle *trans = NULL;
 933         int ret = 0;
 934         int slot;
 935
 936         mutex_lock(&fs_info->qgroup_ioctl_lock);
 937         if (fs_info->quota_root)
 938                 goto out;
 939
 940         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 941         if (!fs_info->qgroup_ulist) {
 942                 ret = -ENOMEM;
 943                 goto out;
 944         }
 945
 946         /*
 947          * 1 for quota root item
 948          * 1 for BTRFS_QGROUP_STATUS item
 949          *
 950          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
 951          * per subvolume. However those are not currently reserved since it
 952          * would be a lot of overkill.
 953          */
 954         trans = btrfs_start_transaction(tree_root, 2);
 955         if (IS_ERR(trans)) {
 956                 ret = PTR_ERR(trans);
 957                 trans = NULL;
 958                 goto out;
 959         }
 960
 961         /*
 962          * initially create the quota tree
 963          */
 964         quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
 965         if (IS_ERR(quota_root)) {
 966                 ret =  PTR_ERR(quota_root);
 967                 btrfs_abort_transaction(trans, ret);
 968                 goto out;
 969         }
 970
 971         path = btrfs_alloc_path();
 972         if (!path) {
 973                 ret = -ENOMEM;
 974                 btrfs_abort_transaction(trans, ret);
 975                 goto out_free_root;
 976         }
 977
 978         key.objectid = 0;
 979         key.type = BTRFS_QGROUP_STATUS_KEY;
 980         key.offset = 0;
 981
 982         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 983                                       sizeof(*ptr));
 984         if (ret) {
 985                 btrfs_abort_transaction(trans, ret);
 986                 goto out_free_path;
 987         }
 988
 989         leaf = path->nodes[0];
 990         ptr = btrfs_item_ptr(leaf, path->slots[0],
 991                                  struct btrfs_qgroup_status_item);
 992         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 993         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 994         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 995                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 996         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 997         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 998
 999         btrfs_mark_buffer_dirty(leaf);
1000
1001         key.objectid = 0;
1002         key.type = BTRFS_ROOT_REF_KEY;
1003         key.offset = 0;
1004
1005         btrfs_release_path(path);
1006         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1007         if (ret > 0)
1008                 goto out_add_root;
1009         if (ret < 0) {
1010                 btrfs_abort_transaction(trans, ret);
1011                 goto out_free_path;
1012         }
1013
1014         while (1) {
1015                 slot = path->slots[0];
1016                 leaf = path->nodes[0];
1017                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1018
1019                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1020                         ret = add_qgroup_item(trans, quota_root,
1021                                               found_key.offset);
1022                         if (ret) {
1023                                 btrfs_abort_transaction(trans, ret);
1024                                 goto out_free_path;
1025                         }
1026
1027                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
1028                         if (IS_ERR(qgroup)) {
1029                                 ret = PTR_ERR(qgroup);
1030                                 btrfs_abort_transaction(trans, ret);
1031                                 goto out_free_path;
1032                         }
1033                 }
1034                 ret = btrfs_next_item(tree_root, path);
1035                 if (ret < 0) {
1036                         btrfs_abort_transaction(trans, ret);
1037                         goto out_free_path;
1038                 }
1039                 if (ret)
1040                         break;
1041         }
1042
1043 out_add_root:
1044         btrfs_release_path(path);
1045         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1046         if (ret) {
1047                 btrfs_abort_transaction(trans, ret);
1048                 goto out_free_path;
1049         }
1050
1051         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1052         if (IS_ERR(qgroup)) {
1053                 ret = PTR_ERR(qgroup);
1054                 btrfs_abort_transaction(trans, ret);
1055                 goto out_free_path;
1056         }
1057
1058         ret = btrfs_commit_transaction(trans);
1059         trans = NULL;
1060         if (ret)
1061                 goto out_free_path;
1062
1063         /*
1064          * Set quota enabled flag after committing the transaction, to avoid
1065          * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1066          * creation.
1067          */
1068         spin_lock(&fs_info->qgroup_lock);
1069         fs_info->quota_root = quota_root;
1070         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1071         spin_unlock(&fs_info->qgroup_lock);
1072
1073         ret = qgroup_rescan_init(fs_info, 0, 1);
1074         if (!ret) {
1075                 qgroup_rescan_zero_tracking(fs_info);
1076                 fs_info->qgroup_rescan_running = true;
1077                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1078                                  &fs_info->qgroup_rescan_work);
1079         }
1080
1081 out_free_path:
1082         btrfs_free_path(path);
1083 out_free_root:
1084         if (ret)
1085                 btrfs_put_root(quota_root);
1086 out:
1087         if (ret) {
1088                 ulist_free(fs_info->qgroup_ulist);
1089                 fs_info->qgroup_ulist = NULL;
1090                 if (trans)
1091                         btrfs_end_transaction(trans);
1092         }
1093         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1094         return ret;
1095 }
1096
1097 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1098 {
1099         struct btrfs_root *quota_root;
1100         struct btrfs_trans_handle *trans = NULL;
1101         int ret = 0;
1102
1103         mutex_lock(&fs_info->qgroup_ioctl_lock);
1104         if (!fs_info->quota_root)
1105                 goto out;
1106
1107         /*
1108          * 1 For the root item
1109          *
1110          * We should also reserve enough items for the quota tree deletion in
1111          * btrfs_clean_quota_tree but this is not done.
1112          */
1113         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1114         if (IS_ERR(trans)) {
1115                 ret = PTR_ERR(trans);
1116                 goto out;
1117         }
1118
1119         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1120         btrfs_qgroup_wait_for_completion(fs_info, false);
1121         spin_lock(&fs_info->qgroup_lock);
1122         quota_root = fs_info->quota_root;
1123         fs_info->quota_root = NULL;
1124         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1125         spin_unlock(&fs_info->qgroup_lock);
1126
1127         btrfs_free_qgroup_config(fs_info);
1128
1129         ret = btrfs_clean_quota_tree(trans, quota_root);
1130         if (ret) {
1131                 btrfs_abort_transaction(trans, ret);
1132                 goto end_trans;
1133         }
1134
1135         ret = btrfs_del_root(trans, &quota_root->root_key);
1136         if (ret) {
1137                 btrfs_abort_transaction(trans, ret);
1138                 goto end_trans;
1139         }
1140
1141         list_del(&quota_root->dirty_list);
1142
1143         btrfs_tree_lock(quota_root->node);
1144         btrfs_clean_tree_block(quota_root->node);
1145         btrfs_tree_unlock(quota_root->node);
1146         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1147
1148         btrfs_put_root(quota_root);
1149
1150 end_trans:
1151         ret = btrfs_end_transaction(trans);
1152 out:
1153         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1154         return ret;
1155 }
1156
1157 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1158                          struct btrfs_qgroup *qgroup)
1159 {
1160         if (list_empty(&qgroup->dirty))
1161                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1162 }
1163
1164 /*
1165  * The easy accounting, we're updating qgroup relationship whose child qgroup
1166  * only has exclusive extents.
1167  *
1168  * In this case, all exclusive extents will also be exclusive for parent, so
1169  * excl/rfer just get added/removed.
1170  *
1171  * So is qgroup reservation space, which should also be added/removed to
1172  * parent.
1173  * Or when child tries to release reservation space, parent will underflow its
1174  * reservation (for relationship adding case).
1175  *
1176  * Caller should hold fs_info->qgroup_lock.
1177  */
1178 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1179                                     struct ulist *tmp, u64 ref_root,
1180                                     struct btrfs_qgroup *src, int sign)
1181 {
1182         struct btrfs_qgroup *qgroup;
1183         struct btrfs_qgroup_list *glist;
1184         struct ulist_node *unode;
1185         struct ulist_iterator uiter;
1186         u64 num_bytes = src->excl;
1187         int ret = 0;
1188
1189         qgroup = find_qgroup_rb(fs_info, ref_root);
1190         if (!qgroup)
1191                 goto out;
1192
1193         qgroup->rfer += sign * num_bytes;
1194         qgroup->rfer_cmpr += sign * num_bytes;
1195
1196         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1197         qgroup->excl += sign * num_bytes;
1198         qgroup->excl_cmpr += sign * num_bytes;
1199
1200         if (sign > 0)
1201                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1202         else
1203                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1204
1205         qgroup_dirty(fs_info, qgroup);
1206
1207         /* Get all of the parent groups that contain this qgroup */
1208         list_for_each_entry(glist, &qgroup->groups, next_group) {
1209                 ret = ulist_add(tmp, glist->group->qgroupid,
1210                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1211                 if (ret < 0)
1212                         goto out;
1213         }
1214
1215         /* Iterate all of the parents and adjust their reference counts */
1216         ULIST_ITER_INIT(&uiter);
1217         while ((unode = ulist_next(tmp, &uiter))) {
1218                 qgroup = unode_aux_to_qgroup(unode);
1219                 qgroup->rfer += sign * num_bytes;
1220                 qgroup->rfer_cmpr += sign * num_bytes;
1221                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1222                 qgroup->excl += sign * num_bytes;
1223                 if (sign > 0)
1224                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1225                 else
1226                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1227                 qgroup->excl_cmpr += sign * num_bytes;
1228                 qgroup_dirty(fs_info, qgroup);
1229
1230                 /* Add any parents of the parents */
1231                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1232                         ret = ulist_add(tmp, glist->group->qgroupid,
1233                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1234                         if (ret < 0)
1235                                 goto out;
1236                 }
1237         }
1238         ret = 0;
1239 out:
1240         return ret;
1241 }
1242
1243
1244 /*
1245  * Quick path for updating qgroup with only excl refs.
1246  *
1247  * In that case, just update all parent will be enough.
1248  * Or we needs to do a full rescan.
1249  * Caller should also hold fs_info->qgroup_lock.
1250  *
1251  * Return 0 for quick update, return >0 for need to full rescan
1252  * and mark INCONSISTENT flag.
1253  * Return < 0 for other error.
1254  */
1255 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1256                                    struct ulist *tmp, u64 src, u64 dst,
1257                                    int sign)
1258 {
1259         struct btrfs_qgroup *qgroup;
1260         int ret = 1;
1261         int err = 0;
1262
1263         qgroup = find_qgroup_rb(fs_info, src);
1264         if (!qgroup)
1265                 goto out;
1266         if (qgroup->excl == qgroup->rfer) {
1267                 ret = 0;
1268                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1269                                                qgroup, sign);
1270                 if (err < 0) {
1271                         ret = err;
1272                         goto out;
1273                 }
1274         }
1275 out:
1276         if (ret)
1277                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1278         return ret;
1279 }
1280
1281 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1282                               u64 dst)
1283 {
1284         struct btrfs_fs_info *fs_info = trans->fs_info;
1285         struct btrfs_qgroup *parent;
1286         struct btrfs_qgroup *member;
1287         struct btrfs_qgroup_list *list;
1288         struct ulist *tmp;
1289         int ret = 0;
1290
1291         /* Check the level of src and dst first */
1292         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1293                 return -EINVAL;
1294
1295         tmp = ulist_alloc(GFP_KERNEL);
1296         if (!tmp)
1297                 return -ENOMEM;
1298
1299         mutex_lock(&fs_info->qgroup_ioctl_lock);
1300         if (!fs_info->quota_root) {
1301                 ret = -ENOTCONN;
1302                 goto out;
1303         }
1304         member = find_qgroup_rb(fs_info, src);
1305         parent = find_qgroup_rb(fs_info, dst);
1306         if (!member || !parent) {
1307                 ret = -EINVAL;
1308                 goto out;
1309         }
1310
1311         /* check if such qgroup relation exist firstly */
1312         list_for_each_entry(list, &member->groups, next_group) {
1313                 if (list->group == parent) {
1314                         ret = -EEXIST;
1315                         goto out;
1316                 }
1317         }
1318
1319         ret = add_qgroup_relation_item(trans, src, dst);
1320         if (ret)
1321                 goto out;
1322
1323         ret = add_qgroup_relation_item(trans, dst, src);
1324         if (ret) {
1325                 del_qgroup_relation_item(trans, src, dst);
1326                 goto out;
1327         }
1328
1329         spin_lock(&fs_info->qgroup_lock);
1330         ret = add_relation_rb(fs_info, src, dst);
1331         if (ret < 0) {
1332                 spin_unlock(&fs_info->qgroup_lock);
1333                 goto out;
1334         }
1335         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1336         spin_unlock(&fs_info->qgroup_lock);
1337 out:
1338         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1339         ulist_free(tmp);
1340         return ret;
1341 }
1342
1343 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1344                                  u64 dst)
1345 {
1346         struct btrfs_fs_info *fs_info = trans->fs_info;
1347         struct btrfs_qgroup *parent;
1348         struct btrfs_qgroup *member;
1349         struct btrfs_qgroup_list *list;
1350         struct ulist *tmp;
1351         bool found = false;
1352         int ret = 0;
1353         int ret2;
1354
1355         tmp = ulist_alloc(GFP_KERNEL);
1356         if (!tmp)
1357                 return -ENOMEM;
1358
1359         if (!fs_info->quota_root) {
1360                 ret = -ENOTCONN;
1361                 goto out;
1362         }
1363
1364         member = find_qgroup_rb(fs_info, src);
1365         parent = find_qgroup_rb(fs_info, dst);
1366         /*
1367          * The parent/member pair doesn't exist, then try to delete the dead
1368          * relation items only.
1369          */
1370         if (!member || !parent)
1371                 goto delete_item;
1372
1373         /* check if such qgroup relation exist firstly */
1374         list_for_each_entry(list, &member->groups, next_group) {
1375                 if (list->group == parent) {
1376                         found = true;
1377                         break;
1378                 }
1379         }
1380
1381 delete_item:
1382         ret = del_qgroup_relation_item(trans, src, dst);
1383         if (ret < 0 && ret != -ENOENT)
1384                 goto out;
1385         ret2 = del_qgroup_relation_item(trans, dst, src);
1386         if (ret2 < 0 && ret2 != -ENOENT)
1387                 goto out;
1388
1389         /* At least one deletion succeeded, return 0 */
1390         if (!ret || !ret2)
1391                 ret = 0;
1392
1393         if (found) {
1394                 spin_lock(&fs_info->qgroup_lock);
1395                 del_relation_rb(fs_info, src, dst);
1396                 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1397                 spin_unlock(&fs_info->qgroup_lock);
1398         }
1399 out:
1400         ulist_free(tmp);
1401         return ret;
1402 }
1403
1404 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1405                               u64 dst)
1406 {
1407         struct btrfs_fs_info *fs_info = trans->fs_info;
1408         int ret = 0;
1409
1410         mutex_lock(&fs_info->qgroup_ioctl_lock);
1411         ret = __del_qgroup_relation(trans, src, dst);
1412         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1413
1414         return ret;
1415 }
1416
1417 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1418 {
1419         struct btrfs_fs_info *fs_info = trans->fs_info;
1420         struct btrfs_root *quota_root;
1421         struct btrfs_qgroup *qgroup;
1422         int ret = 0;
1423
1424         mutex_lock(&fs_info->qgroup_ioctl_lock);
1425         if (!fs_info->quota_root) {
1426                 ret = -ENOTCONN;
1427                 goto out;
1428         }
1429         quota_root = fs_info->quota_root;
1430         qgroup = find_qgroup_rb(fs_info, qgroupid);
1431         if (qgroup) {
1432                 ret = -EEXIST;
1433                 goto out;
1434         }
1435
1436         ret = add_qgroup_item(trans, quota_root, qgroupid);
1437         if (ret)
1438                 goto out;
1439
1440         spin_lock(&fs_info->qgroup_lock);
1441         qgroup = add_qgroup_rb(fs_info, qgroupid);
1442         spin_unlock(&fs_info->qgroup_lock);
1443
1444         if (IS_ERR(qgroup))
1445                 ret = PTR_ERR(qgroup);
1446 out:
1447         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1448         return ret;
1449 }
1450
1451 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1452 {
1453         struct btrfs_fs_info *fs_info = trans->fs_info;
1454         struct btrfs_qgroup *qgroup;
1455         struct btrfs_qgroup_list *list;
1456         int ret = 0;
1457
1458         mutex_lock(&fs_info->qgroup_ioctl_lock);
1459         if (!fs_info->quota_root) {
1460                 ret = -ENOTCONN;
1461                 goto out;
1462         }
1463
1464         qgroup = find_qgroup_rb(fs_info, qgroupid);
1465         if (!qgroup) {
1466                 ret = -ENOENT;
1467                 goto out;
1468         }
1469
1470         /* Check if there are no children of this qgroup */
1471         if (!list_empty(&qgroup->members)) {
1472                 ret = -EBUSY;
1473                 goto out;
1474         }
1475
1476         ret = del_qgroup_item(trans, qgroupid);
1477         if (ret && ret != -ENOENT)
1478                 goto out;
1479
1480         while (!list_empty(&qgroup->groups)) {
1481                 list = list_first_entry(&qgroup->groups,
1482                                         struct btrfs_qgroup_list, next_group);
1483                 ret = __del_qgroup_relation(trans, qgroupid,
1484                                             list->group->qgroupid);
1485                 if (ret)
1486                         goto out;
1487         }
1488
1489         spin_lock(&fs_info->qgroup_lock);
1490         del_qgroup_rb(fs_info, qgroupid);
1491         spin_unlock(&fs_info->qgroup_lock);
1492 out:
1493         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1494         return ret;
1495 }
1496
1497 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1498                        struct btrfs_qgroup_limit *limit)
1499 {
1500         struct btrfs_fs_info *fs_info = trans->fs_info;
1501         struct btrfs_qgroup *qgroup;
1502         int ret = 0;
1503         /* Sometimes we would want to clear the limit on this qgroup.
1504          * To meet this requirement, we treat the -1 as a special value
1505          * which tell kernel to clear the limit on this qgroup.
1506          */
1507         const u64 CLEAR_VALUE = -1;
1508
1509         mutex_lock(&fs_info->qgroup_ioctl_lock);
1510         if (!fs_info->quota_root) {
1511                 ret = -ENOTCONN;
1512                 goto out;
1513         }
1514
1515         qgroup = find_qgroup_rb(fs_info, qgroupid);
1516         if (!qgroup) {
1517                 ret = -ENOENT;
1518                 goto out;
1519         }
1520
1521         spin_lock(&fs_info->qgroup_lock);
1522         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1523                 if (limit->max_rfer == CLEAR_VALUE) {
1524                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1525                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1526                         qgroup->max_rfer = 0;
1527                 } else {
1528                         qgroup->max_rfer = limit->max_rfer;
1529                 }
1530         }
1531         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1532                 if (limit->max_excl == CLEAR_VALUE) {
1533                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1534                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1535                         qgroup->max_excl = 0;
1536                 } else {
1537                         qgroup->max_excl = limit->max_excl;
1538                 }
1539         }
1540         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1541                 if (limit->rsv_rfer == CLEAR_VALUE) {
1542                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1543                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1544                         qgroup->rsv_rfer = 0;
1545                 } else {
1546                         qgroup->rsv_rfer = limit->rsv_rfer;
1547                 }
1548         }
1549         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1550                 if (limit->rsv_excl == CLEAR_VALUE) {
1551                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1552                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1553                         qgroup->rsv_excl = 0;
1554                 } else {
1555                         qgroup->rsv_excl = limit->rsv_excl;
1556                 }
1557         }
1558         qgroup->lim_flags |= limit->flags;
1559
1560         spin_unlock(&fs_info->qgroup_lock);
1561
1562         ret = update_qgroup_limit_item(trans, qgroup);
1563         if (ret) {
1564                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1565                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1566                        qgroupid);
1567         }
1568
1569 out:
1570         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1571         return ret;
1572 }
1573
1574 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1575                                 struct btrfs_delayed_ref_root *delayed_refs,
1576                                 struct btrfs_qgroup_extent_record *record)
1577 {
1578         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1579         struct rb_node *parent_node = NULL;
1580         struct btrfs_qgroup_extent_record *entry;
1581         u64 bytenr = record->bytenr;
1582
1583         lockdep_assert_held(&delayed_refs->lock);
1584         trace_btrfs_qgroup_trace_extent(fs_info, record);
1585
1586         while (*p) {
1587                 parent_node = *p;
1588                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1589                                  node);
1590                 if (bytenr < entry->bytenr) {
1591                         p = &(*p)->rb_left;
1592                 } else if (bytenr > entry->bytenr) {
1593                         p = &(*p)->rb_right;
1594                 } else {
1595                         if (record->data_rsv && !entry->data_rsv) {
1596                                 entry->data_rsv = record->data_rsv;
1597                                 entry->data_rsv_refroot =
1598                                         record->data_rsv_refroot;
1599                         }
1600                         return 1;
1601                 }
1602         }
1603
1604         rb_link_node(&record->node, parent_node, p);
1605         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1606         return 0;
1607 }
1608
1609 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1610                                    struct btrfs_qgroup_extent_record *qrecord)
1611 {
1612         struct ulist *old_root;
1613         u64 bytenr = qrecord->bytenr;
1614         int ret;
1615
1616         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1617         if (ret < 0) {
1618                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1619                 btrfs_warn(fs_info,
1620 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1621                         ret);
1622                 return 0;
1623         }
1624
1625         /*
1626          * Here we don't need to get the lock of
1627          * trans->transaction->delayed_refs, since inserted qrecord won't
1628          * be deleted, only qrecord->node may be modified (new qrecord insert)
1629          *
1630          * So modifying qrecord->old_roots is safe here
1631          */
1632         qrecord->old_roots = old_root;
1633         return 0;
1634 }
1635
1636 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1637                               u64 num_bytes, gfp_t gfp_flag)
1638 {
1639         struct btrfs_fs_info *fs_info = trans->fs_info;
1640         struct btrfs_qgroup_extent_record *record;
1641         struct btrfs_delayed_ref_root *delayed_refs;
1642         int ret;
1643
1644         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1645             || bytenr == 0 || num_bytes == 0)
1646                 return 0;
1647         record = kzalloc(sizeof(*record), gfp_flag);
1648         if (!record)
1649                 return -ENOMEM;
1650
1651         delayed_refs = &trans->transaction->delayed_refs;
1652         record->bytenr = bytenr;
1653         record->num_bytes = num_bytes;
1654         record->old_roots = NULL;
1655
1656         spin_lock(&delayed_refs->lock);
1657         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1658         spin_unlock(&delayed_refs->lock);
1659         if (ret > 0) {
1660                 kfree(record);
1661                 return 0;
1662         }
1663         return btrfs_qgroup_trace_extent_post(fs_info, record);
1664 }
1665
1666 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1667                                   struct extent_buffer *eb)
1668 {
1669         struct btrfs_fs_info *fs_info = trans->fs_info;
1670         int nr = btrfs_header_nritems(eb);
1671         int i, extent_type, ret;
1672         struct btrfs_key key;
1673         struct btrfs_file_extent_item *fi;
1674         u64 bytenr, num_bytes;
1675
1676         /* We can be called directly from walk_up_proc() */
1677         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1678                 return 0;
1679
1680         for (i = 0; i < nr; i++) {
1681                 btrfs_item_key_to_cpu(eb, &key, i);
1682
1683                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1684                         continue;
1685
1686                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1687                 /* filter out non qgroup-accountable extents  */
1688                 extent_type = btrfs_file_extent_type(eb, fi);
1689
1690                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1691                         continue;
1692
1693                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1694                 if (!bytenr)
1695                         continue;
1696
1697                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1698
1699                 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1700                                                 GFP_NOFS);
1701                 if (ret)
1702                         return ret;
1703         }
1704         cond_resched();
1705         return 0;
1706 }
1707
1708 /*
1709  * Walk up the tree from the bottom, freeing leaves and any interior
1710  * nodes which have had all slots visited. If a node (leaf or
1711  * interior) is freed, the node above it will have it's slot
1712  * incremented. The root node will never be freed.
1713  *
1714  * At the end of this function, we should have a path which has all
1715  * slots incremented to the next position for a search. If we need to
1716  * read a new node it will be NULL and the node above it will have the
1717  * correct slot selected for a later read.
1718  *
1719  * If we increment the root nodes slot counter past the number of
1720  * elements, 1 is returned to signal completion of the search.
1721  */
1722 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1723 {
1724         int level = 0;
1725         int nr, slot;
1726         struct extent_buffer *eb;
1727
1728         if (root_level == 0)
1729                 return 1;
1730
1731         while (level <= root_level) {
1732                 eb = path->nodes[level];
1733                 nr = btrfs_header_nritems(eb);
1734                 path->slots[level]++;
1735                 slot = path->slots[level];
1736                 if (slot >= nr || level == 0) {
1737                         /*
1738                          * Don't free the root -  we will detect this
1739                          * condition after our loop and return a
1740                          * positive value for caller to stop walking the tree.
1741                          */
1742                         if (level != root_level) {
1743                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1744                                 path->locks[level] = 0;
1745
1746                                 free_extent_buffer(eb);
1747                                 path->nodes[level] = NULL;
1748                                 path->slots[level] = 0;
1749                         }
1750                 } else {
1751                         /*
1752                          * We have a valid slot to walk back down
1753                          * from. Stop here so caller can process these
1754                          * new nodes.
1755                          */
1756                         break;
1757                 }
1758
1759                 level++;
1760         }
1761
1762         eb = path->nodes[root_level];
1763         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1764                 return 1;
1765
1766         return 0;
1767 }
1768
1769 /*
1770  * Helper function to trace a subtree tree block swap.
1771  *
1772  * The swap will happen in highest tree block, but there may be a lot of
1773  * tree blocks involved.
1774  *
1775  * For example:
1776  *  OO = Old tree blocks
1777  *  NN = New tree blocks allocated during balance
1778  *
1779  *           File tree (257)                  Reloc tree for 257
1780  * L2              OO                                NN
1781  *               /    \                            /    \
1782  * L1          OO      OO (a)                    OO      NN (a)
1783  *            / \     / \                       / \     / \
1784  * L0       OO   OO OO   OO                   OO   OO NN   NN
1785  *                  (b)  (c)                          (b)  (c)
1786  *
1787  * When calling qgroup_trace_extent_swap(), we will pass:
1788  * @src_eb = OO(a)
1789  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1790  * @dst_level = 0
1791  * @root_level = 1
1792  *
1793  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1794  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1795  *
1796  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1797  *
1798  * 1) Tree search from @src_eb
1799  *    It should acts as a simplified btrfs_search_slot().
1800  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1801  *    (first key).
1802  *
1803  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1804  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1805  *    They should be marked during previous (@dst_level = 1) iteration.
1806  *
1807  * 3) Mark file extents in leaves dirty
1808  *    We don't have good way to pick out new file extents only.
1809  *    So we still follow the old method by scanning all file extents in
1810  *    the leave.
1811  *
1812  * This function can free us from keeping two paths, thus later we only need
1813  * to care about how to iterate all new tree blocks in reloc tree.
1814  */
1815 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1816                                     struct extent_buffer *src_eb,
1817                                     struct btrfs_path *dst_path,
1818                                     int dst_level, int root_level,
1819                                     bool trace_leaf)
1820 {
1821         struct btrfs_key key;
1822         struct btrfs_path *src_path;
1823         struct btrfs_fs_info *fs_info = trans->fs_info;
1824         u32 nodesize = fs_info->nodesize;
1825         int cur_level = root_level;
1826         int ret;
1827
1828         BUG_ON(dst_level > root_level);
1829         /* Level mismatch */
1830         if (btrfs_header_level(src_eb) != root_level)
1831                 return -EINVAL;
1832
1833         src_path = btrfs_alloc_path();
1834         if (!src_path) {
1835                 ret = -ENOMEM;
1836                 goto out;
1837         }
1838
1839         if (dst_level)
1840                 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1841         else
1842                 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1843
1844         /* For src_path */
1845         atomic_inc(&src_eb->refs);
1846         src_path->nodes[root_level] = src_eb;
1847         src_path->slots[root_level] = dst_path->slots[root_level];
1848         src_path->locks[root_level] = 0;
1849
1850         /* A simplified version of btrfs_search_slot() */
1851         while (cur_level >= dst_level) {
1852                 struct btrfs_key src_key;
1853                 struct btrfs_key dst_key;
1854
1855                 if (src_path->nodes[cur_level] == NULL) {
1856                         struct btrfs_key first_key;
1857                         struct extent_buffer *eb;
1858                         int parent_slot;
1859                         u64 child_gen;
1860                         u64 child_bytenr;
1861
1862                         eb = src_path->nodes[cur_level + 1];
1863                         parent_slot = src_path->slots[cur_level + 1];
1864                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1865                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1866                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1867
1868                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1869                                              cur_level, &first_key);
1870                         if (IS_ERR(eb)) {
1871                                 ret = PTR_ERR(eb);
1872                                 goto out;
1873                         } else if (!extent_buffer_uptodate(eb)) {
1874                                 free_extent_buffer(eb);
1875                                 ret = -EIO;
1876                                 goto out;
1877                         }
1878
1879                         src_path->nodes[cur_level] = eb;
1880
1881                         btrfs_tree_read_lock(eb);
1882                         btrfs_set_lock_blocking_read(eb);
1883                         src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1884                 }
1885
1886                 src_path->slots[cur_level] = dst_path->slots[cur_level];
1887                 if (cur_level) {
1888                         btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1889                                         &dst_key, dst_path->slots[cur_level]);
1890                         btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1891                                         &src_key, src_path->slots[cur_level]);
1892                 } else {
1893                         btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1894                                         &dst_key, dst_path->slots[cur_level]);
1895                         btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1896                                         &src_key, src_path->slots[cur_level]);
1897                 }
1898                 /* Content mismatch, something went wrong */
1899                 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1900                         ret = -ENOENT;
1901                         goto out;
1902                 }
1903                 cur_level--;
1904         }
1905
1906         /*
1907          * Now both @dst_path and @src_path have been populated, record the tree
1908          * blocks for qgroup accounting.
1909          */
1910         ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1911                         nodesize, GFP_NOFS);
1912         if (ret < 0)
1913                 goto out;
1914         ret = btrfs_qgroup_trace_extent(trans,
1915                         dst_path->nodes[dst_level]->start,
1916                         nodesize, GFP_NOFS);
1917         if (ret < 0)
1918                 goto out;
1919
1920         /* Record leaf file extents */
1921         if (dst_level == 0 && trace_leaf) {
1922                 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1923                 if (ret < 0)
1924                         goto out;
1925                 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1926         }
1927 out:
1928         btrfs_free_path(src_path);
1929         return ret;
1930 }
1931
1932 /*
1933  * Helper function to do recursive generation-aware depth-first search, to
1934  * locate all new tree blocks in a subtree of reloc tree.
1935  *
1936  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1937  *         reloc tree
1938  * L2         NN (a)
1939  *          /    \
1940  * L1    OO        NN (b)
1941  *      /  \      /  \
1942  * L0  OO  OO    OO  NN
1943  *               (c) (d)
1944  * If we pass:
1945  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1946  * @cur_level = 1
1947  * @root_level = 1
1948  *
1949  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1950  * above tree blocks along with their counter parts in file tree.
1951  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1952  * won't affect OO(c).
1953  */
1954 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1955                                            struct extent_buffer *src_eb,
1956                                            struct btrfs_path *dst_path,
1957                                            int cur_level, int root_level,
1958                                            u64 last_snapshot, bool trace_leaf)
1959 {
1960         struct btrfs_fs_info *fs_info = trans->fs_info;
1961         struct extent_buffer *eb;
1962         bool need_cleanup = false;
1963         int ret = 0;
1964         int i;
1965
1966         /* Level sanity check */
1967         if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1968             root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1969             root_level < cur_level) {
1970                 btrfs_err_rl(fs_info,
1971                         "%s: bad levels, cur_level=%d root_level=%d",
1972                         __func__, cur_level, root_level);
1973                 return -EUCLEAN;
1974         }
1975
1976         /* Read the tree block if needed */
1977         if (dst_path->nodes[cur_level] == NULL) {
1978                 struct btrfs_key first_key;
1979                 int parent_slot;
1980                 u64 child_gen;
1981                 u64 child_bytenr;
1982
1983                 /*
1984                  * dst_path->nodes[root_level] must be initialized before
1985                  * calling this function.
1986                  */
1987                 if (cur_level == root_level) {
1988                         btrfs_err_rl(fs_info,
1989         "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1990                                 __func__, root_level, root_level, cur_level);
1991                         return -EUCLEAN;
1992                 }
1993
1994                 /*
1995                  * We need to get child blockptr/gen from parent before we can
1996                  * read it.
1997                   */
1998                 eb = dst_path->nodes[cur_level + 1];
1999                 parent_slot = dst_path->slots[cur_level + 1];
2000                 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2001                 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2002                 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2003
2004                 /* This node is old, no need to trace */
2005                 if (child_gen < last_snapshot)
2006                         goto out;
2007
2008                 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2009                                      cur_level, &first_key);
2010                 if (IS_ERR(eb)) {
2011                         ret = PTR_ERR(eb);
2012                         goto out;
2013                 } else if (!extent_buffer_uptodate(eb)) {
2014                         free_extent_buffer(eb);
2015                         ret = -EIO;
2016                         goto out;
2017                 }
2018
2019                 dst_path->nodes[cur_level] = eb;
2020                 dst_path->slots[cur_level] = 0;
2021
2022                 btrfs_tree_read_lock(eb);
2023                 btrfs_set_lock_blocking_read(eb);
2024                 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
2025                 need_cleanup = true;
2026         }
2027
2028         /* Now record this tree block and its counter part for qgroups */
2029         ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2030                                        root_level, trace_leaf);
2031         if (ret < 0)
2032                 goto cleanup;
2033
2034         eb = dst_path->nodes[cur_level];
2035
2036         if (cur_level > 0) {
2037                 /* Iterate all child tree blocks */
2038                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2039                         /* Skip old tree blocks as they won't be swapped */
2040                         if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2041                                 continue;
2042                         dst_path->slots[cur_level] = i;
2043
2044                         /* Recursive call (at most 7 times) */
2045                         ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2046                                         dst_path, cur_level - 1, root_level,
2047                                         last_snapshot, trace_leaf);
2048                         if (ret < 0)
2049                                 goto cleanup;
2050                 }
2051         }
2052
2053 cleanup:
2054         if (need_cleanup) {
2055                 /* Clean up */
2056                 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2057                                      dst_path->locks[cur_level]);
2058                 free_extent_buffer(dst_path->nodes[cur_level]);
2059                 dst_path->nodes[cur_level] = NULL;
2060                 dst_path->slots[cur_level] = 0;
2061                 dst_path->locks[cur_level] = 0;
2062         }
2063 out:
2064         return ret;
2065 }
2066
2067 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2068                                 struct extent_buffer *src_eb,
2069                                 struct extent_buffer *dst_eb,
2070                                 u64 last_snapshot, bool trace_leaf)
2071 {
2072         struct btrfs_fs_info *fs_info = trans->fs_info;
2073         struct btrfs_path *dst_path = NULL;
2074         int level;
2075         int ret;
2076
2077         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2078                 return 0;
2079
2080         /* Wrong parameter order */
2081         if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2082                 btrfs_err_rl(fs_info,
2083                 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2084                              btrfs_header_generation(src_eb),
2085                              btrfs_header_generation(dst_eb));
2086                 return -EUCLEAN;
2087         }
2088
2089         if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2090                 ret = -EIO;
2091                 goto out;
2092         }
2093
2094         level = btrfs_header_level(dst_eb);
2095         dst_path = btrfs_alloc_path();
2096         if (!dst_path) {
2097                 ret = -ENOMEM;
2098                 goto out;
2099         }
2100         /* For dst_path */
2101         atomic_inc(&dst_eb->refs);
2102         dst_path->nodes[level] = dst_eb;
2103         dst_path->slots[level] = 0;
2104         dst_path->locks[level] = 0;
2105
2106         /* Do the generation aware breadth-first search */
2107         ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2108                                               level, last_snapshot, trace_leaf);
2109         if (ret < 0)
2110                 goto out;
2111         ret = 0;
2112
2113 out:
2114         btrfs_free_path(dst_path);
2115         if (ret < 0)
2116                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2117         return ret;
2118 }
2119
2120 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2121                                struct extent_buffer *root_eb,
2122                                u64 root_gen, int root_level)
2123 {
2124         struct btrfs_fs_info *fs_info = trans->fs_info;
2125         int ret = 0;
2126         int level;
2127         struct extent_buffer *eb = root_eb;
2128         struct btrfs_path *path = NULL;
2129
2130         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2131         BUG_ON(root_eb == NULL);
2132
2133         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2134                 return 0;
2135
2136         if (!extent_buffer_uptodate(root_eb)) {
2137                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2138                 if (ret)
2139                         goto out;
2140         }
2141
2142         if (root_level == 0) {
2143                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2144                 goto out;
2145         }
2146
2147         path = btrfs_alloc_path();
2148         if (!path)
2149                 return -ENOMEM;
2150
2151         /*
2152          * Walk down the tree.  Missing extent blocks are filled in as
2153          * we go. Metadata is accounted every time we read a new
2154          * extent block.
2155          *
2156          * When we reach a leaf, we account for file extent items in it,
2157          * walk back up the tree (adjusting slot pointers as we go)
2158          * and restart the search process.
2159          */
2160         atomic_inc(&root_eb->refs);     /* For path */
2161         path->nodes[root_level] = root_eb;
2162         path->slots[root_level] = 0;
2163         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2164 walk_down:
2165         level = root_level;
2166         while (level >= 0) {
2167                 if (path->nodes[level] == NULL) {
2168                         struct btrfs_key first_key;
2169                         int parent_slot;
2170                         u64 child_gen;
2171                         u64 child_bytenr;
2172
2173                         /*
2174                          * We need to get child blockptr/gen from parent before
2175                          * we can read it.
2176                           */
2177                         eb = path->nodes[level + 1];
2178                         parent_slot = path->slots[level + 1];
2179                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2180                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2181                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2182
2183                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
2184                                              level, &first_key);
2185                         if (IS_ERR(eb)) {
2186                                 ret = PTR_ERR(eb);
2187                                 goto out;
2188                         } else if (!extent_buffer_uptodate(eb)) {
2189                                 free_extent_buffer(eb);
2190                                 ret = -EIO;
2191                                 goto out;
2192                         }
2193
2194                         path->nodes[level] = eb;
2195                         path->slots[level] = 0;
2196
2197                         btrfs_tree_read_lock(eb);
2198                         btrfs_set_lock_blocking_read(eb);
2199                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2200
2201                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2202                                                         fs_info->nodesize,
2203                                                         GFP_NOFS);
2204                         if (ret)
2205                                 goto out;
2206                 }
2207
2208                 if (level == 0) {
2209                         ret = btrfs_qgroup_trace_leaf_items(trans,
2210                                                             path->nodes[level]);
2211                         if (ret)
2212                                 goto out;
2213
2214                         /* Nonzero return here means we completed our search */
2215                         ret = adjust_slots_upwards(path, root_level);
2216                         if (ret)
2217                                 break;
2218
2219                         /* Restart search with new slots */
2220                         goto walk_down;
2221                 }
2222
2223                 level--;
2224         }
2225
2226         ret = 0;
2227 out:
2228         btrfs_free_path(path);
2229
2230         return ret;
2231 }
2232
2233 #define UPDATE_NEW      0
2234 #define UPDATE_OLD      1
2235 /*
2236  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2237  */
2238 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2239                                 struct ulist *roots, struct ulist *tmp,
2240                                 struct ulist *qgroups, u64 seq, int update_old)
2241 {
2242         struct ulist_node *unode;
2243         struct ulist_iterator uiter;
2244         struct ulist_node *tmp_unode;
2245         struct ulist_iterator tmp_uiter;
2246         struct btrfs_qgroup *qg;
2247         int ret = 0;
2248
2249         if (!roots)
2250                 return 0;
2251         ULIST_ITER_INIT(&uiter);
2252         while ((unode = ulist_next(roots, &uiter))) {
2253                 qg = find_qgroup_rb(fs_info, unode->val);
2254                 if (!qg)
2255                         continue;
2256
2257                 ulist_reinit(tmp);
2258                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2259                                 GFP_ATOMIC);
2260                 if (ret < 0)
2261                         return ret;
2262                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2263                 if (ret < 0)
2264                         return ret;
2265                 ULIST_ITER_INIT(&tmp_uiter);
2266                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2267                         struct btrfs_qgroup_list *glist;
2268
2269                         qg = unode_aux_to_qgroup(tmp_unode);
2270                         if (update_old)
2271                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2272                         else
2273                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2274                         list_for_each_entry(glist, &qg->groups, next_group) {
2275                                 ret = ulist_add(qgroups, glist->group->qgroupid,
2276                                                 qgroup_to_aux(glist->group),
2277                                                 GFP_ATOMIC);
2278                                 if (ret < 0)
2279                                         return ret;
2280                                 ret = ulist_add(tmp, glist->group->qgroupid,
2281                                                 qgroup_to_aux(glist->group),
2282                                                 GFP_ATOMIC);
2283                                 if (ret < 0)
2284                                         return ret;
2285                         }
2286                 }
2287         }
2288         return 0;
2289 }
2290
2291 /*
2292  * Update qgroup rfer/excl counters.
2293  * Rfer update is easy, codes can explain themselves.
2294  *
2295  * Excl update is tricky, the update is split into 2 part.
2296  * Part 1: Possible exclusive <-> sharing detect:
2297  *      |       A       |       !A      |
2298  *  -------------------------------------
2299  *  B   |       *       |       -       |
2300  *  -------------------------------------
2301  *  !B  |       +       |       **      |
2302  *  -------------------------------------
2303  *
2304  * Conditions:
2305  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
2306  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
2307  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
2308  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
2309  *
2310  * Results:
2311  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
2312  * *: Definitely not changed.           **: Possible unchanged.
2313  *
2314  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2315  *
2316  * To make the logic clear, we first use condition A and B to split
2317  * combination into 4 results.
2318  *
2319  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2320  * only on variant maybe 0.
2321  *
2322  * Lastly, check result **, since there are 2 variants maybe 0, split them
2323  * again(2x2).
2324  * But this time we don't need to consider other things, the codes and logic
2325  * is easy to understand now.
2326  */
2327 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2328                                   struct ulist *qgroups,
2329                                   u64 nr_old_roots,
2330                                   u64 nr_new_roots,
2331                                   u64 num_bytes, u64 seq)
2332 {
2333         struct ulist_node *unode;
2334         struct ulist_iterator uiter;
2335         struct btrfs_qgroup *qg;
2336         u64 cur_new_count, cur_old_count;
2337
2338         ULIST_ITER_INIT(&uiter);
2339         while ((unode = ulist_next(qgroups, &uiter))) {
2340                 bool dirty = false;
2341
2342                 qg = unode_aux_to_qgroup(unode);
2343                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2344                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2345
2346                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2347                                              cur_new_count);
2348
2349                 /* Rfer update part */
2350                 if (cur_old_count == 0 && cur_new_count > 0) {
2351                         qg->rfer += num_bytes;
2352                         qg->rfer_cmpr += num_bytes;
2353                         dirty = true;
2354                 }
2355                 if (cur_old_count > 0 && cur_new_count == 0) {
2356                         qg->rfer -= num_bytes;
2357                         qg->rfer_cmpr -= num_bytes;
2358                         dirty = true;
2359                 }
2360
2361                 /* Excl update part */
2362                 /* Exclusive/none -> shared case */
2363                 if (cur_old_count == nr_old_roots &&
2364                     cur_new_count < nr_new_roots) {
2365                         /* Exclusive -> shared */
2366                         if (cur_old_count != 0) {
2367                                 qg->excl -= num_bytes;
2368                                 qg->excl_cmpr -= num_bytes;
2369                                 dirty = true;
2370                         }
2371                 }
2372
2373                 /* Shared -> exclusive/none case */
2374                 if (cur_old_count < nr_old_roots &&
2375                     cur_new_count == nr_new_roots) {
2376                         /* Shared->exclusive */
2377                         if (cur_new_count != 0) {
2378                                 qg->excl += num_bytes;
2379                                 qg->excl_cmpr += num_bytes;
2380                                 dirty = true;
2381                         }
2382                 }
2383
2384                 /* Exclusive/none -> exclusive/none case */
2385                 if (cur_old_count == nr_old_roots &&
2386                     cur_new_count == nr_new_roots) {
2387                         if (cur_old_count == 0) {
2388                                 /* None -> exclusive/none */
2389
2390                                 if (cur_new_count != 0) {
2391                                         /* None -> exclusive */
2392                                         qg->excl += num_bytes;
2393                                         qg->excl_cmpr += num_bytes;
2394                                         dirty = true;
2395                                 }
2396                                 /* None -> none, nothing changed */
2397                         } else {
2398                                 /* Exclusive -> exclusive/none */
2399
2400                                 if (cur_new_count == 0) {
2401                                         /* Exclusive -> none */
2402                                         qg->excl -= num_bytes;
2403                                         qg->excl_cmpr -= num_bytes;
2404                                         dirty = true;
2405                                 }
2406                                 /* Exclusive -> exclusive, nothing changed */
2407                         }
2408                 }
2409
2410                 if (dirty)
2411                         qgroup_dirty(fs_info, qg);
2412         }
2413         return 0;
2414 }
2415
2416 /*
2417  * Check if the @roots potentially is a list of fs tree roots
2418  *
2419  * Return 0 for definitely not a fs/subvol tree roots ulist
2420  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2421  *          one as well)
2422  */
2423 static int maybe_fs_roots(struct ulist *roots)
2424 {
2425         struct ulist_node *unode;
2426         struct ulist_iterator uiter;
2427
2428         /* Empty one, still possible for fs roots */
2429         if (!roots || roots->nnodes == 0)
2430                 return 1;
2431
2432         ULIST_ITER_INIT(&uiter);
2433         unode = ulist_next(roots, &uiter);
2434         if (!unode)
2435                 return 1;
2436
2437         /*
2438          * If it contains fs tree roots, then it must belong to fs/subvol
2439          * trees.
2440          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2441          */
2442         return is_fstree(unode->val);
2443 }
2444
2445 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2446                                 u64 num_bytes, struct ulist *old_roots,
2447                                 struct ulist *new_roots)
2448 {
2449         struct btrfs_fs_info *fs_info = trans->fs_info;
2450         struct ulist *qgroups = NULL;
2451         struct ulist *tmp = NULL;
2452         u64 seq;
2453         u64 nr_new_roots = 0;
2454         u64 nr_old_roots = 0;
2455         int ret = 0;
2456
2457         /*
2458          * If quotas get disabled meanwhile, the resouces need to be freed and
2459          * we can't just exit here.
2460          */
2461         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2462                 goto out_free;
2463
2464         if (new_roots) {
2465                 if (!maybe_fs_roots(new_roots))
2466                         goto out_free;
2467                 nr_new_roots = new_roots->nnodes;
2468         }
2469         if (old_roots) {
2470                 if (!maybe_fs_roots(old_roots))
2471                         goto out_free;
2472                 nr_old_roots = old_roots->nnodes;
2473         }
2474
2475         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2476         if (nr_old_roots == 0 && nr_new_roots == 0)
2477                 goto out_free;
2478
2479         BUG_ON(!fs_info->quota_root);
2480
2481         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2482                                         num_bytes, nr_old_roots, nr_new_roots);
2483
2484         qgroups = ulist_alloc(GFP_NOFS);
2485         if (!qgroups) {
2486                 ret = -ENOMEM;
2487                 goto out_free;
2488         }
2489         tmp = ulist_alloc(GFP_NOFS);
2490         if (!tmp) {
2491                 ret = -ENOMEM;
2492                 goto out_free;
2493         }
2494
2495         mutex_lock(&fs_info->qgroup_rescan_lock);
2496         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2497                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2498                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2499                         ret = 0;
2500                         goto out_free;
2501                 }
2502         }
2503         mutex_unlock(&fs_info->qgroup_rescan_lock);
2504
2505         spin_lock(&fs_info->qgroup_lock);
2506         seq = fs_info->qgroup_seq;
2507
2508         /* Update old refcnts using old_roots */
2509         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2510                                    UPDATE_OLD);
2511         if (ret < 0)
2512                 goto out;
2513
2514         /* Update new refcnts using new_roots */
2515         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2516                                    UPDATE_NEW);
2517         if (ret < 0)
2518                 goto out;
2519
2520         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2521                                num_bytes, seq);
2522
2523         /*
2524          * Bump qgroup_seq to avoid seq overlap
2525          */
2526         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2527 out:
2528         spin_unlock(&fs_info->qgroup_lock);
2529 out_free:
2530         ulist_free(tmp);
2531         ulist_free(qgroups);
2532         ulist_free(old_roots);
2533         ulist_free(new_roots);
2534         return ret;
2535 }
2536
2537 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2538 {
2539         struct btrfs_fs_info *fs_info = trans->fs_info;
2540         struct btrfs_qgroup_extent_record *record;
2541         struct btrfs_delayed_ref_root *delayed_refs;
2542         struct ulist *new_roots = NULL;
2543         struct rb_node *node;
2544         u64 num_dirty_extents = 0;
2545         u64 qgroup_to_skip;
2546         int ret = 0;
2547
2548         delayed_refs = &trans->transaction->delayed_refs;
2549         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2550         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2551                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2552                                   node);
2553
2554                 num_dirty_extents++;
2555                 trace_btrfs_qgroup_account_extents(fs_info, record);
2556
2557                 if (!ret) {
2558                         /*
2559                          * Old roots should be searched when inserting qgroup
2560                          * extent record
2561                          */
2562                         if (WARN_ON(!record->old_roots)) {
2563                                 /* Search commit root to find old_roots */
2564                                 ret = btrfs_find_all_roots(NULL, fs_info,
2565                                                 record->bytenr, 0,
2566                                                 &record->old_roots, false);
2567                                 if (ret < 0)
2568                                         goto cleanup;
2569                         }
2570
2571                         /* Free the reserved data space */
2572                         btrfs_qgroup_free_refroot(fs_info,
2573                                         record->data_rsv_refroot,
2574                                         record->data_rsv,
2575                                         BTRFS_QGROUP_RSV_DATA);
2576                         /*
2577                          * Use SEQ_LAST as time_seq to do special search, which
2578                          * doesn't lock tree or delayed_refs and search current
2579                          * root. It's safe inside commit_transaction().
2580                          */
2581                         ret = btrfs_find_all_roots(trans, fs_info,
2582                                 record->bytenr, SEQ_LAST, &new_roots, false);
2583                         if (ret < 0)
2584                                 goto cleanup;
2585                         if (qgroup_to_skip) {
2586                                 ulist_del(new_roots, qgroup_to_skip, 0);
2587                                 ulist_del(record->old_roots, qgroup_to_skip,
2588                                           0);
2589                         }
2590                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2591                                                           record->num_bytes,
2592                                                           record->old_roots,
2593                                                           new_roots);
2594                         record->old_roots = NULL;
2595                         new_roots = NULL;
2596                 }
2597 cleanup:
2598                 ulist_free(record->old_roots);
2599                 ulist_free(new_roots);
2600                 new_roots = NULL;
2601                 rb_erase(node, &delayed_refs->dirty_extent_root);
2602                 kfree(record);
2603
2604         }
2605         trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2606                                        num_dirty_extents);
2607         return ret;
2608 }
2609
2610 /*
2611  * called from commit_transaction. Writes all changed qgroups to disk.
2612  */
2613 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2614 {
2615         struct btrfs_fs_info *fs_info = trans->fs_info;
2616         int ret = 0;
2617
2618         if (!fs_info->quota_root)
2619                 return ret;
2620
2621         spin_lock(&fs_info->qgroup_lock);
2622         while (!list_empty(&fs_info->dirty_qgroups)) {
2623                 struct btrfs_qgroup *qgroup;
2624                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2625                                           struct btrfs_qgroup, dirty);
2626                 list_del_init(&qgroup->dirty);
2627                 spin_unlock(&fs_info->qgroup_lock);
2628                 ret = update_qgroup_info_item(trans, qgroup);
2629                 if (ret)
2630                         fs_info->qgroup_flags |=
2631                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2632                 ret = update_qgroup_limit_item(trans, qgroup);
2633                 if (ret)
2634                         fs_info->qgroup_flags |=
2635                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2636                 spin_lock(&fs_info->qgroup_lock);
2637         }
2638         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2639                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2640         else
2641                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2642         spin_unlock(&fs_info->qgroup_lock);
2643
2644         ret = update_qgroup_status_item(trans);
2645         if (ret)
2646                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2647
2648         return ret;
2649 }
2650
2651 /*
2652  * Copy the accounting information between qgroups. This is necessary
2653  * when a snapshot or a subvolume is created. Throwing an error will
2654  * cause a transaction abort so we take extra care here to only error
2655  * when a readonly fs is a reasonable outcome.
2656  */
2657 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2658                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2659 {
2660         int ret = 0;
2661         int i;
2662         u64 *i_qgroups;
2663         bool committing = false;
2664         struct btrfs_fs_info *fs_info = trans->fs_info;
2665         struct btrfs_root *quota_root;
2666         struct btrfs_qgroup *srcgroup;
2667         struct btrfs_qgroup *dstgroup;
2668         bool need_rescan = false;
2669         u32 level_size = 0;
2670         u64 nums;
2671
2672         /*
2673          * There are only two callers of this function.
2674          *
2675          * One in create_subvol() in the ioctl context, which needs to hold
2676          * the qgroup_ioctl_lock.
2677          *
2678          * The other one in create_pending_snapshot() where no other qgroup
2679          * code can modify the fs as they all need to either start a new trans
2680          * or hold a trans handler, thus we don't need to hold
2681          * qgroup_ioctl_lock.
2682          * This would avoid long and complex lock chain and make lockdep happy.
2683          */
2684         spin_lock(&fs_info->trans_lock);
2685         if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2686                 committing = true;
2687         spin_unlock(&fs_info->trans_lock);
2688
2689         if (!committing)
2690                 mutex_lock(&fs_info->qgroup_ioctl_lock);
2691         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2692                 goto out;
2693
2694         quota_root = fs_info->quota_root;
2695         if (!quota_root) {
2696                 ret = -EINVAL;
2697                 goto out;
2698         }
2699
2700         if (inherit) {
2701                 i_qgroups = (u64 *)(inherit + 1);
2702                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2703                        2 * inherit->num_excl_copies;
2704                 for (i = 0; i < nums; ++i) {
2705                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2706
2707                         /*
2708                          * Zero out invalid groups so we can ignore
2709                          * them later.
2710                          */
2711                         if (!srcgroup ||
2712                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2713                                 *i_qgroups = 0ULL;
2714
2715                         ++i_qgroups;
2716                 }
2717         }
2718
2719         /*
2720          * create a tracking group for the subvol itself
2721          */
2722         ret = add_qgroup_item(trans, quota_root, objectid);
2723         if (ret)
2724                 goto out;
2725
2726         /*
2727          * add qgroup to all inherited groups
2728          */
2729         if (inherit) {
2730                 i_qgroups = (u64 *)(inherit + 1);
2731                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2732                         if (*i_qgroups == 0)
2733                                 continue;
2734                         ret = add_qgroup_relation_item(trans, objectid,
2735                                                        *i_qgroups);
2736                         if (ret && ret != -EEXIST)
2737                                 goto out;
2738                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2739                                                        objectid);
2740                         if (ret && ret != -EEXIST)
2741                                 goto out;
2742                 }
2743                 ret = 0;
2744         }
2745
2746
2747         spin_lock(&fs_info->qgroup_lock);
2748
2749         dstgroup = add_qgroup_rb(fs_info, objectid);
2750         if (IS_ERR(dstgroup)) {
2751                 ret = PTR_ERR(dstgroup);
2752                 goto unlock;
2753         }
2754
2755         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2756                 dstgroup->lim_flags = inherit->lim.flags;
2757                 dstgroup->max_rfer = inherit->lim.max_rfer;
2758                 dstgroup->max_excl = inherit->lim.max_excl;
2759                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2760                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2761
2762                 ret = update_qgroup_limit_item(trans, dstgroup);
2763                 if (ret) {
2764                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2765                         btrfs_info(fs_info,
2766                                    "unable to update quota limit for %llu",
2767                                    dstgroup->qgroupid);
2768                         goto unlock;
2769                 }
2770         }
2771
2772         if (srcid) {
2773                 srcgroup = find_qgroup_rb(fs_info, srcid);
2774                 if (!srcgroup)
2775                         goto unlock;
2776
2777                 /*
2778                  * We call inherit after we clone the root in order to make sure
2779                  * our counts don't go crazy, so at this point the only
2780                  * difference between the two roots should be the root node.
2781                  */
2782                 level_size = fs_info->nodesize;
2783                 dstgroup->rfer = srcgroup->rfer;
2784                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2785                 dstgroup->excl = level_size;
2786                 dstgroup->excl_cmpr = level_size;
2787                 srcgroup->excl = level_size;
2788                 srcgroup->excl_cmpr = level_size;
2789
2790                 /* inherit the limit info */
2791                 dstgroup->lim_flags = srcgroup->lim_flags;
2792                 dstgroup->max_rfer = srcgroup->max_rfer;
2793                 dstgroup->max_excl = srcgroup->max_excl;
2794                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2795                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2796
2797                 qgroup_dirty(fs_info, dstgroup);
2798                 qgroup_dirty(fs_info, srcgroup);
2799         }
2800
2801         if (!inherit)
2802                 goto unlock;
2803
2804         i_qgroups = (u64 *)(inherit + 1);
2805         for (i = 0; i < inherit->num_qgroups; ++i) {
2806                 if (*i_qgroups) {
2807                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2808                         if (ret)
2809                                 goto unlock;
2810                 }
2811                 ++i_qgroups;
2812
2813                 /*
2814                  * If we're doing a snapshot, and adding the snapshot to a new
2815                  * qgroup, the numbers are guaranteed to be incorrect.
2816                  */
2817                 if (srcid)
2818                         need_rescan = true;
2819         }
2820
2821         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2822                 struct btrfs_qgroup *src;
2823                 struct btrfs_qgroup *dst;
2824
2825                 if (!i_qgroups[0] || !i_qgroups[1])
2826                         continue;
2827
2828                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2829                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2830
2831                 if (!src || !dst) {
2832                         ret = -EINVAL;
2833                         goto unlock;
2834                 }
2835
2836                 dst->rfer = src->rfer - level_size;
2837                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2838
2839                 /* Manually tweaking numbers certainly needs a rescan */
2840                 need_rescan = true;
2841         }
2842         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2843                 struct btrfs_qgroup *src;
2844                 struct btrfs_qgroup *dst;
2845
2846                 if (!i_qgroups[0] || !i_qgroups[1])
2847                         continue;
2848
2849                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2850                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2851
2852                 if (!src || !dst) {
2853                         ret = -EINVAL;
2854                         goto unlock;
2855                 }
2856
2857                 dst->excl = src->excl + level_size;
2858                 dst->excl_cmpr = src->excl_cmpr + level_size;
2859                 need_rescan = true;
2860         }
2861
2862 unlock:
2863         spin_unlock(&fs_info->qgroup_lock);
2864 out:
2865         if (!committing)
2866                 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2867         if (need_rescan)
2868                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2869         return ret;
2870 }
2871
2872 /*
2873  * Two limits to commit transaction in advance.
2874  *
2875  * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2876  * For SIZE, it will be in byte unit as threshold.
2877  */
2878 #define QGROUP_FREE_RATIO               32
2879 #define QGROUP_FREE_SIZE                SZ_32M
2880 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2881                                 const struct btrfs_qgroup *qg, u64 num_bytes)
2882 {
2883         u64 free;
2884         u64 threshold;
2885
2886         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2887             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2888                 return false;
2889
2890         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2891             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2892                 return false;
2893
2894         /*
2895          * Even if we passed the check, it's better to check if reservation
2896          * for meta_pertrans is pushing us near limit.
2897          * If there is too much pertrans reservation or it's near the limit,
2898          * let's try commit transaction to free some, using transaction_kthread
2899          */
2900         if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2901                               BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2902                 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2903                         free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2904                         threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2905                                           QGROUP_FREE_SIZE);
2906                 } else {
2907                         free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2908                         threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2909                                           QGROUP_FREE_SIZE);
2910                 }
2911
2912                 /*
2913                  * Use transaction_kthread to commit transaction, so we no
2914                  * longer need to bother nested transaction nor lock context.
2915                  */
2916                 if (free < threshold)
2917                         btrfs_commit_transaction_locksafe(fs_info);
2918         }
2919
2920         return true;
2921 }
2922
2923 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2924                           enum btrfs_qgroup_rsv_type type)
2925 {
2926         struct btrfs_qgroup *qgroup;
2927         struct btrfs_fs_info *fs_info = root->fs_info;
2928         u64 ref_root = root->root_key.objectid;
2929         int ret = 0;
2930         struct ulist_node *unode;
2931         struct ulist_iterator uiter;
2932
2933         if (!is_fstree(ref_root))
2934                 return 0;
2935
2936         if (num_bytes == 0)
2937                 return 0;
2938
2939         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2940             capable(CAP_SYS_RESOURCE))
2941                 enforce = false;
2942
2943         spin_lock(&fs_info->qgroup_lock);
2944         if (!fs_info->quota_root)
2945                 goto out;
2946
2947         qgroup = find_qgroup_rb(fs_info, ref_root);
2948         if (!qgroup)
2949                 goto out;
2950
2951         /*
2952          * in a first step, we check all affected qgroups if any limits would
2953          * be exceeded
2954          */
2955         ulist_reinit(fs_info->qgroup_ulist);
2956         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2957                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2958         if (ret < 0)
2959                 goto out;
2960         ULIST_ITER_INIT(&uiter);
2961         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2962                 struct btrfs_qgroup *qg;
2963                 struct btrfs_qgroup_list *glist;
2964
2965                 qg = unode_aux_to_qgroup(unode);
2966
2967                 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2968                         ret = -EDQUOT;
2969                         goto out;
2970                 }
2971
2972                 list_for_each_entry(glist, &qg->groups, next_group) {
2973                         ret = ulist_add(fs_info->qgroup_ulist,
2974                                         glist->group->qgroupid,
2975                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2976                         if (ret < 0)
2977                                 goto out;
2978                 }
2979         }
2980         ret = 0;
2981         /*
2982          * no limits exceeded, now record the reservation into all qgroups
2983          */
2984         ULIST_ITER_INIT(&uiter);
2985         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2986                 struct btrfs_qgroup *qg;
2987
2988                 qg = unode_aux_to_qgroup(unode);
2989
2990                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2991         }
2992
2993 out:
2994         spin_unlock(&fs_info->qgroup_lock);
2995         return ret;
2996 }
2997
2998 /*
2999  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
3000  * qgroup).
3001  *
3002  * Will handle all higher level qgroup too.
3003  *
3004  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3005  * This special case is only used for META_PERTRANS type.
3006  */
3007 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3008                                u64 ref_root, u64 num_bytes,
3009                                enum btrfs_qgroup_rsv_type type)
3010 {
3011         struct btrfs_qgroup *qgroup;
3012         struct ulist_node *unode;
3013         struct ulist_iterator uiter;
3014         int ret = 0;
3015
3016         if (!is_fstree(ref_root))
3017                 return;
3018
3019         if (num_bytes == 0)
3020                 return;
3021
3022         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3023                 WARN(1, "%s: Invalid type to free", __func__);
3024                 return;
3025         }
3026         spin_lock(&fs_info->qgroup_lock);
3027
3028         if (!fs_info->quota_root)
3029                 goto out;
3030
3031         qgroup = find_qgroup_rb(fs_info, ref_root);
3032         if (!qgroup)
3033                 goto out;
3034
3035         if (num_bytes == (u64)-1)
3036                 /*
3037                  * We're freeing all pertrans rsv, get reserved value from
3038                  * level 0 qgroup as real num_bytes to free.
3039                  */
3040                 num_bytes = qgroup->rsv.values[type];
3041
3042         ulist_reinit(fs_info->qgroup_ulist);
3043         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3044                         qgroup_to_aux(qgroup), GFP_ATOMIC);
3045         if (ret < 0)
3046                 goto out;
3047         ULIST_ITER_INIT(&uiter);
3048         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3049                 struct btrfs_qgroup *qg;
3050                 struct btrfs_qgroup_list *glist;
3051
3052                 qg = unode_aux_to_qgroup(unode);
3053
3054                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3055
3056                 list_for_each_entry(glist, &qg->groups, next_group) {
3057                         ret = ulist_add(fs_info->qgroup_ulist,
3058                                         glist->group->qgroupid,
3059                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3060                         if (ret < 0)
3061                                 goto out;
3062                 }
3063         }
3064
3065 out:
3066         spin_unlock(&fs_info->qgroup_lock);
3067 }
3068
3069 /*
3070  * Check if the leaf is the last leaf. Which means all node pointers
3071  * are at their last position.
3072  */
3073 static bool is_last_leaf(struct btrfs_path *path)
3074 {
3075         int i;
3076
3077         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3078                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3079                         return false;
3080         }
3081         return true;
3082 }
3083
3084 /*
3085  * returns < 0 on error, 0 when more leafs are to be scanned.
3086  * returns 1 when done.
3087  */
3088 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3089                               struct btrfs_path *path)
3090 {
3091         struct btrfs_fs_info *fs_info = trans->fs_info;
3092         struct btrfs_key found;
3093         struct extent_buffer *scratch_leaf = NULL;
3094         struct ulist *roots = NULL;
3095         u64 num_bytes;
3096         bool done;
3097         int slot;
3098         int ret;
3099
3100         mutex_lock(&fs_info->qgroup_rescan_lock);
3101         ret = btrfs_search_slot_for_read(fs_info->extent_root,
3102                                          &fs_info->qgroup_rescan_progress,
3103                                          path, 1, 0);
3104
3105         btrfs_debug(fs_info,
3106                 "current progress key (%llu %u %llu), search_slot ret %d",
3107                 fs_info->qgroup_rescan_progress.objectid,
3108                 fs_info->qgroup_rescan_progress.type,
3109                 fs_info->qgroup_rescan_progress.offset, ret);
3110
3111         if (ret) {
3112                 /*
3113                  * The rescan is about to end, we will not be scanning any
3114                  * further blocks. We cannot unset the RESCAN flag here, because
3115                  * we want to commit the transaction if everything went well.
3116                  * To make the live accounting work in this phase, we set our
3117                  * scan progress pointer such that every real extent objectid
3118                  * will be smaller.
3119                  */
3120                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3121                 btrfs_release_path(path);
3122                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3123                 return ret;
3124         }
3125         done = is_last_leaf(path);
3126
3127         btrfs_item_key_to_cpu(path->nodes[0], &found,
3128                               btrfs_header_nritems(path->nodes[0]) - 1);
3129         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3130
3131         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3132         if (!scratch_leaf) {
3133                 ret = -ENOMEM;
3134                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3135                 goto out;
3136         }
3137         slot = path->slots[0];
3138         btrfs_release_path(path);
3139         mutex_unlock(&fs_info->qgroup_rescan_lock);
3140
3141         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3142                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3143                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3144                     found.type != BTRFS_METADATA_ITEM_KEY)
3145                         continue;
3146                 if (found.type == BTRFS_METADATA_ITEM_KEY)
3147                         num_bytes = fs_info->nodesize;
3148                 else
3149                         num_bytes = found.offset;
3150
3151                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3152                                            &roots, false);
3153                 if (ret < 0)
3154                         goto out;
3155                 /* For rescan, just pass old_roots as NULL */
3156                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3157                                                   num_bytes, NULL, roots);
3158                 if (ret < 0)
3159                         goto out;
3160         }
3161 out:
3162         if (scratch_leaf)
3163                 free_extent_buffer(scratch_leaf);
3164
3165         if (done && !ret) {
3166                 ret = 1;
3167                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3168         }
3169         return ret;
3170 }
3171
3172 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3173 {
3174         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3175                                                      qgroup_rescan_work);
3176         struct btrfs_path *path;
3177         struct btrfs_trans_handle *trans = NULL;
3178         int err = -ENOMEM;
3179         int ret = 0;
3180
3181         path = btrfs_alloc_path();
3182         if (!path)
3183                 goto out;
3184         /*
3185          * Rescan should only search for commit root, and any later difference
3186          * should be recorded by qgroup
3187          */
3188         path->search_commit_root = 1;
3189         path->skip_locking = 1;
3190
3191         err = 0;
3192         while (!err && !btrfs_fs_closing(fs_info)) {
3193                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3194                 if (IS_ERR(trans)) {
3195                         err = PTR_ERR(trans);
3196                         break;
3197                 }
3198                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3199                         err = -EINTR;
3200                 } else {
3201                         err = qgroup_rescan_leaf(trans, path);
3202                 }
3203                 if (err > 0)
3204                         btrfs_commit_transaction(trans);
3205                 else
3206                         btrfs_end_transaction(trans);
3207         }
3208
3209 out:
3210         btrfs_free_path(path);
3211
3212         mutex_lock(&fs_info->qgroup_rescan_lock);
3213         if (err > 0 &&
3214             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3215                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3216         } else if (err < 0) {
3217                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3218         }
3219         mutex_unlock(&fs_info->qgroup_rescan_lock);
3220
3221         /*
3222          * only update status, since the previous part has already updated the
3223          * qgroup info.
3224          */
3225         trans = btrfs_start_transaction(fs_info->quota_root, 1);
3226         if (IS_ERR(trans)) {
3227                 err = PTR_ERR(trans);
3228                 trans = NULL;
3229                 btrfs_err(fs_info,
3230                           "fail to start transaction for status update: %d",
3231                           err);
3232         }
3233
3234         mutex_lock(&fs_info->qgroup_rescan_lock);
3235         if (!btrfs_fs_closing(fs_info))
3236                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3237         if (trans) {
3238                 ret = update_qgroup_status_item(trans);
3239                 if (ret < 0) {
3240                         err = ret;
3241                         btrfs_err(fs_info, "fail to update qgroup status: %d",
3242                                   err);
3243                 }
3244         }
3245         fs_info->qgroup_rescan_running = false;
3246         complete_all(&fs_info->qgroup_rescan_completion);
3247         mutex_unlock(&fs_info->qgroup_rescan_lock);
3248
3249         if (!trans)
3250                 return;
3251
3252         btrfs_end_transaction(trans);
3253
3254         if (btrfs_fs_closing(fs_info)) {
3255                 btrfs_info(fs_info, "qgroup scan paused");
3256         } else if (err >= 0) {
3257                 btrfs_info(fs_info, "qgroup scan completed%s",
3258                         err > 0 ? " (inconsistency flag cleared)" : "");
3259         } else {
3260                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3261         }
3262 }
3263
3264 /*
3265  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3266  * memory required for the rescan context.
3267  */
3268 static int
3269 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3270                    int init_flags)
3271 {
3272         int ret = 0;
3273
3274         if (!init_flags) {
3275                 /* we're resuming qgroup rescan at mount time */
3276                 if (!(fs_info->qgroup_flags &
3277                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3278                         btrfs_warn(fs_info,
3279                         "qgroup rescan init failed, qgroup rescan is not queued");
3280                         ret = -EINVAL;
3281                 } else if (!(fs_info->qgroup_flags &
3282                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3283                         btrfs_warn(fs_info,
3284                         "qgroup rescan init failed, qgroup is not enabled");
3285                         ret = -EINVAL;
3286                 }
3287
3288                 if (ret)
3289                         return ret;
3290         }
3291
3292         mutex_lock(&fs_info->qgroup_rescan_lock);
3293
3294         if (init_flags) {
3295                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3296                         btrfs_warn(fs_info,
3297                                    "qgroup rescan is already in progress");
3298                         ret = -EINPROGRESS;
3299                 } else if (!(fs_info->qgroup_flags &
3300                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3301                         btrfs_warn(fs_info,
3302                         "qgroup rescan init failed, qgroup is not enabled");
3303                         ret = -EINVAL;
3304                 }
3305
3306                 if (ret) {
3307                         mutex_unlock(&fs_info->qgroup_rescan_lock);
3308                         return ret;
3309                 }
3310                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3311         }
3312
3313         memset(&fs_info->qgroup_rescan_progress, 0,
3314                 sizeof(fs_info->qgroup_rescan_progress));
3315         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3316         init_completion(&fs_info->qgroup_rescan_completion);
3317         mutex_unlock(&fs_info->qgroup_rescan_lock);
3318
3319         btrfs_init_work(&fs_info->qgroup_rescan_work,
3320                         btrfs_qgroup_rescan_worker, NULL, NULL);
3321         return 0;
3322 }
3323
3324 static void
3325 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3326 {
3327         struct rb_node *n;
3328         struct btrfs_qgroup *qgroup;
3329
3330         spin_lock(&fs_info->qgroup_lock);
3331         /* clear all current qgroup tracking information */
3332         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3333                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3334                 qgroup->rfer = 0;
3335                 qgroup->rfer_cmpr = 0;
3336                 qgroup->excl = 0;
3337                 qgroup->excl_cmpr = 0;
3338                 qgroup_dirty(fs_info, qgroup);
3339         }
3340         spin_unlock(&fs_info->qgroup_lock);
3341 }
3342
3343 int
3344 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3345 {
3346         int ret = 0;
3347         struct btrfs_trans_handle *trans;
3348
3349         ret = qgroup_rescan_init(fs_info, 0, 1);
3350         if (ret)
3351                 return ret;
3352
3353         /*
3354          * We have set the rescan_progress to 0, which means no more
3355          * delayed refs will be accounted by btrfs_qgroup_account_ref.
3356          * However, btrfs_qgroup_account_ref may be right after its call
3357          * to btrfs_find_all_roots, in which case it would still do the
3358          * accounting.
3359          * To solve this, we're committing the transaction, which will
3360          * ensure we run all delayed refs and only after that, we are
3361          * going to clear all tracking information for a clean start.
3362          */
3363
3364         trans = btrfs_join_transaction(fs_info->fs_root);
3365         if (IS_ERR(trans)) {
3366                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3367                 return PTR_ERR(trans);
3368         }
3369         ret = btrfs_commit_transaction(trans);
3370         if (ret) {
3371                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3372                 return ret;
3373         }
3374
3375         qgroup_rescan_zero_tracking(fs_info);
3376
3377         mutex_lock(&fs_info->qgroup_rescan_lock);
3378         fs_info->qgroup_rescan_running = true;
3379         btrfs_queue_work(fs_info->qgroup_rescan_workers,
3380                          &fs_info->qgroup_rescan_work);
3381         mutex_unlock(&fs_info->qgroup_rescan_lock);
3382
3383         return 0;
3384 }
3385
3386 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3387                                      bool interruptible)
3388 {
3389         int running;
3390         int ret = 0;
3391
3392         mutex_lock(&fs_info->qgroup_rescan_lock);
3393         running = fs_info->qgroup_rescan_running;
3394         mutex_unlock(&fs_info->qgroup_rescan_lock);
3395
3396         if (!running)
3397                 return 0;
3398
3399         if (interruptible)
3400                 ret = wait_for_completion_interruptible(
3401                                         &fs_info->qgroup_rescan_completion);
3402         else
3403                 wait_for_completion(&fs_info->qgroup_rescan_completion);
3404
3405         return ret;
3406 }
3407
3408 /*
3409  * this is only called from open_ctree where we're still single threaded, thus
3410  * locking is omitted here.
3411  */
3412 void
3413 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3414 {
3415         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3416                 mutex_lock(&fs_info->qgroup_rescan_lock);
3417                 fs_info->qgroup_rescan_running = true;
3418                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3419                                  &fs_info->qgroup_rescan_work);
3420                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3421         }
3422 }
3423
3424 /*
3425  * Reserve qgroup space for range [start, start + len).
3426  *
3427  * This function will either reserve space from related qgroups or doing
3428  * nothing if the range is already reserved.
3429  *
3430  * Return 0 for successful reserve
3431  * Return <0 for error (including -EQUOT)
3432  *
3433  * NOTE: this function may sleep for memory allocation.
3434  *       if btrfs_qgroup_reserve_data() is called multiple times with
3435  *       same @reserved, caller must ensure when error happens it's OK
3436  *       to free *ALL* reserved space.
3437  */
3438 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3439                         struct extent_changeset **reserved_ret, u64 start,
3440                         u64 len)
3441 {
3442         struct btrfs_root *root = inode->root;
3443         struct ulist_node *unode;
3444         struct ulist_iterator uiter;
3445         struct extent_changeset *reserved;
3446         u64 orig_reserved;
3447         u64 to_reserve;
3448         int ret;
3449
3450         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3451             !is_fstree(root->root_key.objectid) || len == 0)
3452                 return 0;
3453
3454         /* @reserved parameter is mandatory for qgroup */
3455         if (WARN_ON(!reserved_ret))
3456                 return -EINVAL;
3457         if (!*reserved_ret) {
3458                 *reserved_ret = extent_changeset_alloc();
3459                 if (!*reserved_ret)
3460                         return -ENOMEM;
3461         }
3462         reserved = *reserved_ret;
3463         /* Record already reserved space */
3464         orig_reserved = reserved->bytes_changed;
3465         ret = set_record_extent_bits(&inode->io_tree, start,
3466                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3467
3468         /* Newly reserved space */
3469         to_reserve = reserved->bytes_changed - orig_reserved;
3470         trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3471                                         to_reserve, QGROUP_RESERVE);
3472         if (ret < 0)
3473                 goto cleanup;
3474         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3475         if (ret < 0)
3476                 goto cleanup;
3477
3478         return ret;
3479
3480 cleanup:
3481         /* cleanup *ALL* already reserved ranges */
3482         ULIST_ITER_INIT(&uiter);
3483         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3484                 clear_extent_bit(&inode->io_tree, unode->val,
3485                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3486         /* Also free data bytes of already reserved one */
3487         btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid,
3488                                   orig_reserved, BTRFS_QGROUP_RSV_DATA);
3489         extent_changeset_release(reserved);
3490         return ret;
3491 }
3492
3493 /* Free ranges specified by @reserved, normally in error path */
3494 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3495                         struct extent_changeset *reserved, u64 start, u64 len)
3496 {
3497         struct btrfs_root *root = inode->root;
3498         struct ulist_node *unode;
3499         struct ulist_iterator uiter;
3500         struct extent_changeset changeset;
3501         int freed = 0;
3502         int ret;
3503
3504         extent_changeset_init(&changeset);
3505         len = round_up(start + len, root->fs_info->sectorsize);
3506         start = round_down(start, root->fs_info->sectorsize);
3507
3508         ULIST_ITER_INIT(&uiter);
3509         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3510                 u64 range_start = unode->val;
3511                 /* unode->aux is the inclusive end */
3512                 u64 range_len = unode->aux - range_start + 1;
3513                 u64 free_start;
3514                 u64 free_len;
3515
3516                 extent_changeset_release(&changeset);
3517
3518                 /* Only free range in range [start, start + len) */
3519                 if (range_start >= start + len ||
3520                     range_start + range_len <= start)
3521                         continue;
3522                 free_start = max(range_start, start);
3523                 free_len = min(start + len, range_start + range_len) -
3524                            free_start;
3525                 /*
3526                  * TODO: To also modify reserved->ranges_reserved to reflect
3527                  * the modification.
3528                  *
3529                  * However as long as we free qgroup reserved according to
3530                  * EXTENT_QGROUP_RESERVED, we won't double free.
3531                  * So not need to rush.
3532                  */
3533                 ret = clear_record_extent_bits(&inode->io_tree, free_start,
3534                                 free_start + free_len - 1,
3535                                 EXTENT_QGROUP_RESERVED, &changeset);
3536                 if (ret < 0)
3537                         goto out;
3538                 freed += changeset.bytes_changed;
3539         }
3540         btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3541                                   BTRFS_QGROUP_RSV_DATA);
3542         ret = freed;
3543 out:
3544         extent_changeset_release(&changeset);
3545         return ret;
3546 }
3547
3548 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3549                         struct extent_changeset *reserved, u64 start, u64 len,
3550                         int free)
3551 {
3552         struct extent_changeset changeset;
3553         int trace_op = QGROUP_RELEASE;
3554         int ret;
3555
3556         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3557                 return 0;
3558
3559         /* In release case, we shouldn't have @reserved */
3560         WARN_ON(!free && reserved);
3561         if (free && reserved)
3562                 return qgroup_free_reserved_data(inode, reserved, start, len);
3563         extent_changeset_init(&changeset);
3564         ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3565                                        EXTENT_QGROUP_RESERVED, &changeset);
3566         if (ret < 0)
3567                 goto out;
3568
3569         if (free)
3570                 trace_op = QGROUP_FREE;
3571         trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3572                                         changeset.bytes_changed, trace_op);
3573         if (free)
3574                 btrfs_qgroup_free_refroot(inode->root->fs_info,
3575                                 inode->root->root_key.objectid,
3576                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3577         ret = changeset.bytes_changed;
3578 out:
3579         extent_changeset_release(&changeset);
3580         return ret;
3581 }
3582
3583 /*
3584  * Free a reserved space range from io_tree and related qgroups
3585  *
3586  * Should be called when a range of pages get invalidated before reaching disk.
3587  * Or for error cleanup case.
3588  * if @reserved is given, only reserved range in [@start, @start + @len) will
3589  * be freed.
3590  *
3591  * For data written to disk, use btrfs_qgroup_release_data().
3592  *
3593  * NOTE: This function may sleep for memory allocation.
3594  */
3595 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3596                         struct extent_changeset *reserved, u64 start, u64 len)
3597 {
3598         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3599 }
3600
3601 /*
3602  * Release a reserved space range from io_tree only.
3603  *
3604  * Should be called when a range of pages get written to disk and corresponding
3605  * FILE_EXTENT is inserted into corresponding root.
3606  *
3607  * Since new qgroup accounting framework will only update qgroup numbers at
3608  * commit_transaction() time, its reserved space shouldn't be freed from
3609  * related qgroups.
3610  *
3611  * But we should release the range from io_tree, to allow further write to be
3612  * COWed.
3613  *
3614  * NOTE: This function may sleep for memory allocation.
3615  */
3616 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3617 {
3618         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3619 }
3620
3621 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3622                               enum btrfs_qgroup_rsv_type type)
3623 {
3624         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3625             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3626                 return;
3627         if (num_bytes == 0)
3628                 return;
3629
3630         spin_lock(&root->qgroup_meta_rsv_lock);
3631         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3632                 root->qgroup_meta_rsv_prealloc += num_bytes;
3633         else
3634                 root->qgroup_meta_rsv_pertrans += num_bytes;
3635         spin_unlock(&root->qgroup_meta_rsv_lock);
3636 }
3637
3638 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3639                              enum btrfs_qgroup_rsv_type type)
3640 {
3641         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3642             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3643                 return 0;
3644         if (num_bytes == 0)
3645                 return 0;
3646
3647         spin_lock(&root->qgroup_meta_rsv_lock);
3648         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3649                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3650                                   num_bytes);
3651                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3652         } else {
3653                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3654                                   num_bytes);
3655                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3656         }
3657         spin_unlock(&root->qgroup_meta_rsv_lock);
3658         return num_bytes;
3659 }
3660
3661 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3662                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3663 {
3664         struct btrfs_fs_info *fs_info = root->fs_info;
3665         int ret;
3666
3667         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3668             !is_fstree(root->root_key.objectid) || num_bytes == 0)
3669                 return 0;
3670
3671         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3672         trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3673         ret = qgroup_reserve(root, num_bytes, enforce, type);
3674         if (ret < 0)
3675                 return ret;
3676         /*
3677          * Record what we have reserved into root.
3678          *
3679          * To avoid quota disabled->enabled underflow.
3680          * In that case, we may try to free space we haven't reserved
3681          * (since quota was disabled), so record what we reserved into root.
3682          * And ensure later release won't underflow this number.
3683          */
3684         add_root_meta_rsv(root, num_bytes, type);
3685         return ret;
3686 }
3687
3688 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3689 {
3690         struct btrfs_fs_info *fs_info = root->fs_info;
3691
3692         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3693             !is_fstree(root->root_key.objectid))
3694                 return;
3695
3696         /* TODO: Update trace point to handle such free */
3697         trace_qgroup_meta_free_all_pertrans(root);
3698         /* Special value -1 means to free all reserved space */
3699         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3700                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3701 }
3702
3703 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3704                               enum btrfs_qgroup_rsv_type type)
3705 {
3706         struct btrfs_fs_info *fs_info = root->fs_info;
3707
3708         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3709             !is_fstree(root->root_key.objectid))
3710                 return;
3711
3712         /*
3713          * reservation for META_PREALLOC can happen before quota is enabled,
3714          * which can lead to underflow.
3715          * Here ensure we will only free what we really have reserved.
3716          */
3717         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3718         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3719         trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3720         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3721                                   num_bytes, type);
3722 }
3723
3724 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3725                                 int num_bytes)
3726 {
3727         struct btrfs_qgroup *qgroup;
3728         struct ulist_node *unode;
3729         struct ulist_iterator uiter;
3730         int ret = 0;
3731
3732         if (num_bytes == 0)
3733                 return;
3734         if (!fs_info->quota_root)
3735                 return;
3736
3737         spin_lock(&fs_info->qgroup_lock);
3738         qgroup = find_qgroup_rb(fs_info, ref_root);
3739         if (!qgroup)
3740                 goto out;
3741         ulist_reinit(fs_info->qgroup_ulist);
3742         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3743                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3744         if (ret < 0)
3745                 goto out;
3746         ULIST_ITER_INIT(&uiter);
3747         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3748                 struct btrfs_qgroup *qg;
3749                 struct btrfs_qgroup_list *glist;
3750
3751                 qg = unode_aux_to_qgroup(unode);
3752
3753                 qgroup_rsv_release(fs_info, qg, num_bytes,
3754                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3755                 qgroup_rsv_add(fs_info, qg, num_bytes,
3756                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3757                 list_for_each_entry(glist, &qg->groups, next_group) {
3758                         ret = ulist_add(fs_info->qgroup_ulist,
3759                                         glist->group->qgroupid,
3760                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3761                         if (ret < 0)
3762                                 goto out;
3763                 }
3764         }
3765 out:
3766         spin_unlock(&fs_info->qgroup_lock);
3767 }
3768
3769 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3770 {
3771         struct btrfs_fs_info *fs_info = root->fs_info;
3772
3773         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3774             !is_fstree(root->root_key.objectid))
3775                 return;
3776         /* Same as btrfs_qgroup_free_meta_prealloc() */
3777         num_bytes = sub_root_meta_rsv(root, num_bytes,
3778                                       BTRFS_QGROUP_RSV_META_PREALLOC);
3779         trace_qgroup_meta_convert(root, num_bytes);
3780         qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3781 }
3782
3783 /*
3784  * Check qgroup reserved space leaking, normally at destroy inode
3785  * time
3786  */
3787 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3788 {
3789         struct extent_changeset changeset;
3790         struct ulist_node *unode;
3791         struct ulist_iterator iter;
3792         int ret;
3793
3794         extent_changeset_init(&changeset);
3795         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3796                         EXTENT_QGROUP_RESERVED, &changeset);
3797
3798         WARN_ON(ret < 0);
3799         if (WARN_ON(changeset.bytes_changed)) {
3800                 ULIST_ITER_INIT(&iter);
3801                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3802                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3803                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3804                                 inode->i_ino, unode->val, unode->aux);
3805                 }
3806                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3807                                 BTRFS_I(inode)->root->root_key.objectid,
3808                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3809
3810         }
3811         extent_changeset_release(&changeset);
3812 }
3813
3814 void btrfs_qgroup_init_swapped_blocks(
3815         struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3816 {
3817         int i;
3818
3819         spin_lock_init(&swapped_blocks->lock);
3820         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3821                 swapped_blocks->blocks[i] = RB_ROOT;
3822         swapped_blocks->swapped = false;
3823 }
3824
3825 /*
3826  * Delete all swapped blocks record of @root.
3827  * Every record here means we skipped a full subtree scan for qgroup.
3828  *
3829  * Gets called when committing one transaction.
3830  */
3831 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3832 {
3833         struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3834         int i;
3835
3836         swapped_blocks = &root->swapped_blocks;
3837
3838         spin_lock(&swapped_blocks->lock);
3839         if (!swapped_blocks->swapped)
3840                 goto out;
3841         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3842                 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3843                 struct btrfs_qgroup_swapped_block *entry;
3844                 struct btrfs_qgroup_swapped_block *next;
3845
3846                 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3847                                                      node)
3848                         kfree(entry);
3849                 swapped_blocks->blocks[i] = RB_ROOT;
3850         }
3851         swapped_blocks->swapped = false;
3852 out:
3853         spin_unlock(&swapped_blocks->lock);
3854 }
3855
3856 /*
3857  * Add subtree roots record into @subvol_root.
3858  *
3859  * @subvol_root:        tree root of the subvolume tree get swapped
3860  * @bg:                 block group under balance
3861  * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3862  * @reloc_parent/slot:  pointer to the subtree root in reloc tree
3863  *                      BOTH POINTERS ARE BEFORE TREE SWAP
3864  * @last_snapshot:      last snapshot generation of the subvolume tree
3865  */
3866 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3867                 struct btrfs_root *subvol_root,
3868                 struct btrfs_block_group *bg,
3869                 struct extent_buffer *subvol_parent, int subvol_slot,
3870                 struct extent_buffer *reloc_parent, int reloc_slot,
3871                 u64 last_snapshot)
3872 {
3873         struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3874         struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3875         struct btrfs_qgroup_swapped_block *block;
3876         struct rb_node **cur;
3877         struct rb_node *parent = NULL;
3878         int level = btrfs_header_level(subvol_parent) - 1;
3879         int ret = 0;
3880
3881         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3882                 return 0;
3883
3884         if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3885             btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3886                 btrfs_err_rl(fs_info,
3887                 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3888                         __func__,
3889                         btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3890                         btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3891                 return -EUCLEAN;
3892         }
3893
3894         block = kmalloc(sizeof(*block), GFP_NOFS);
3895         if (!block) {
3896                 ret = -ENOMEM;
3897                 goto out;
3898         }
3899
3900         /*
3901          * @reloc_parent/slot is still before swap, while @block is going to
3902          * record the bytenr after swap, so we do the swap here.
3903          */
3904         block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3905         block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3906                                                              reloc_slot);
3907         block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3908         block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3909                                                             subvol_slot);
3910         block->last_snapshot = last_snapshot;
3911         block->level = level;
3912
3913         /*
3914          * If we have bg == NULL, we're called from btrfs_recover_relocation(),
3915          * no one else can modify tree blocks thus we qgroup will not change
3916          * no matter the value of trace_leaf.
3917          */
3918         if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
3919                 block->trace_leaf = true;
3920         else
3921                 block->trace_leaf = false;
3922         btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3923
3924         /* Insert @block into @blocks */
3925         spin_lock(&blocks->lock);
3926         cur = &blocks->blocks[level].rb_node;
3927         while (*cur) {
3928                 struct btrfs_qgroup_swapped_block *entry;
3929
3930                 parent = *cur;
3931                 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3932                                  node);
3933
3934                 if (entry->subvol_bytenr < block->subvol_bytenr) {
3935                         cur = &(*cur)->rb_left;
3936                 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3937                         cur = &(*cur)->rb_right;
3938                 } else {
3939                         if (entry->subvol_generation !=
3940                                         block->subvol_generation ||
3941                             entry->reloc_bytenr != block->reloc_bytenr ||
3942                             entry->reloc_generation !=
3943                                         block->reloc_generation) {
3944                                 /*
3945                                  * Duplicated but mismatch entry found.
3946                                  * Shouldn't happen.
3947                                  *
3948                                  * Marking qgroup inconsistent should be enough
3949                                  * for end users.
3950                                  */
3951                                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3952                                 ret = -EEXIST;
3953                         }
3954                         kfree(block);
3955                         goto out_unlock;
3956                 }
3957         }
3958         rb_link_node(&block->node, parent, cur);
3959         rb_insert_color(&block->node, &blocks->blocks[level]);
3960         blocks->swapped = true;
3961 out_unlock:
3962         spin_unlock(&blocks->lock);
3963 out:
3964         if (ret < 0)
3965                 fs_info->qgroup_flags |=
3966                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3967         return ret;
3968 }
3969
3970 /*
3971  * Check if the tree block is a subtree root, and if so do the needed
3972  * delayed subtree trace for qgroup.
3973  *
3974  * This is called during btrfs_cow_block().
3975  */
3976 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3977                                          struct btrfs_root *root,
3978                                          struct extent_buffer *subvol_eb)
3979 {
3980         struct btrfs_fs_info *fs_info = root->fs_info;
3981         struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3982         struct btrfs_qgroup_swapped_block *block;
3983         struct extent_buffer *reloc_eb = NULL;
3984         struct rb_node *node;
3985         bool found = false;
3986         bool swapped = false;
3987         int level = btrfs_header_level(subvol_eb);
3988         int ret = 0;
3989         int i;
3990
3991         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3992                 return 0;
3993         if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3994                 return 0;
3995
3996         spin_lock(&blocks->lock);
3997         if (!blocks->swapped) {
3998                 spin_unlock(&blocks->lock);
3999                 return 0;
4000         }
4001         node = blocks->blocks[level].rb_node;
4002
4003         while (node) {
4004                 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4005                 if (block->subvol_bytenr < subvol_eb->start) {
4006                         node = node->rb_left;
4007                 } else if (block->subvol_bytenr > subvol_eb->start) {
4008                         node = node->rb_right;
4009                 } else {
4010                         found = true;
4011                         break;
4012                 }
4013         }
4014         if (!found) {
4015                 spin_unlock(&blocks->lock);
4016                 goto out;
4017         }
4018         /* Found one, remove it from @blocks first and update blocks->swapped */
4019         rb_erase(&block->node, &blocks->blocks[level]);
4020         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4021                 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4022                         swapped = true;
4023                         break;
4024                 }
4025         }
4026         blocks->swapped = swapped;
4027         spin_unlock(&blocks->lock);
4028
4029         /* Read out reloc subtree root */
4030         reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
4031                                    block->reloc_generation, block->level,
4032                                    &block->first_key);
4033         if (IS_ERR(reloc_eb)) {
4034                 ret = PTR_ERR(reloc_eb);
4035                 reloc_eb = NULL;
4036                 goto free_out;
4037         }
4038         if (!extent_buffer_uptodate(reloc_eb)) {
4039                 ret = -EIO;
4040                 goto free_out;
4041         }
4042
4043         ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4044                         block->last_snapshot, block->trace_leaf);
4045 free_out:
4046         kfree(block);
4047         free_extent_buffer(reloc_eb);
4048 out:
4049         if (ret < 0) {
4050                 btrfs_err_rl(fs_info,
4051                              "failed to account subtree at bytenr %llu: %d",
4052                              subvol_eb->start, ret);
4053                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4054         }
4055         return ret;
4056 }
4057
4058 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4059 {
4060         struct btrfs_qgroup_extent_record *entry;
4061         struct btrfs_qgroup_extent_record *next;
4062         struct rb_root *root;
4063
4064         root = &trans->delayed_refs.dirty_extent_root;
4065         rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4066                 ulist_free(entry->old_roots);
4067                 kfree(entry);
4068         }
4069 }