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/sched/mm.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 #include "sysfs.h"
  26 #include "tree-mod-log.h"
  27
  28 /*
  29  * Helpers to access qgroup reservation
  30  *
  31  * Callers should ensure the lock context and type are valid
  32  */
  33
  34 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
  35 {
  36         u64 ret = 0;
  37         int i;
  38
  39         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
  40                 ret += qgroup->rsv.values[i];
  41
  42         return ret;
  43 }
  44
  45 #ifdef CONFIG_BTRFS_DEBUG
  46 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
  47 {
  48         if (type == BTRFS_QGROUP_RSV_DATA)
  49                 return "data";
  50         if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
  51                 return "meta_pertrans";
  52         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
  53                 return "meta_prealloc";
  54         return NULL;
  55 }
  56 #endif
  57
  58 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
  59                            struct btrfs_qgroup *qgroup, u64 num_bytes,
  60                            enum btrfs_qgroup_rsv_type type)
  61 {
  62         trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
  63         qgroup->rsv.values[type] += num_bytes;
  64 }
  65
  66 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
  67                                struct btrfs_qgroup *qgroup, u64 num_bytes,
  68                                enum btrfs_qgroup_rsv_type type)
  69 {
  70         trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
  71         if (qgroup->rsv.values[type] >= num_bytes) {
  72                 qgroup->rsv.values[type] -= num_bytes;
  73                 return;
  74         }
  75 #ifdef CONFIG_BTRFS_DEBUG
  76         WARN_RATELIMIT(1,
  77                 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
  78                 qgroup->qgroupid, qgroup_rsv_type_str(type),
  79                 qgroup->rsv.values[type], num_bytes);
  80 #endif
  81         qgroup->rsv.values[type] = 0;
  82 }
  83
  84 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
  85                                      struct btrfs_qgroup *dest,
  86                                      struct btrfs_qgroup *src)
  87 {
  88         int i;
  89
  90         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
  91                 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
  92 }
  93
  94 static void qgroup_rsv_release_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_release(fs_info, dest, src->rsv.values[i], i);
 102 }
 103
 104 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
 105                                            int mod)
 106 {
 107         if (qg->old_refcnt < seq)
 108                 qg->old_refcnt = seq;
 109         qg->old_refcnt += mod;
 110 }
 111
 112 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
 113                                            int mod)
 114 {
 115         if (qg->new_refcnt < seq)
 116                 qg->new_refcnt = seq;
 117         qg->new_refcnt += mod;
 118 }
 119
 120 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
 121 {
 122         if (qg->old_refcnt < seq)
 123                 return 0;
 124         return qg->old_refcnt - seq;
 125 }
 126
 127 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
 128 {
 129         if (qg->new_refcnt < seq)
 130                 return 0;
 131         return qg->new_refcnt - seq;
 132 }
 133
 134 /*
 135  * glue structure to represent the relations between qgroups.
 136  */
 137 struct btrfs_qgroup_list {
 138         struct list_head next_group;
 139         struct list_head next_member;
 140         struct btrfs_qgroup *group;
 141         struct btrfs_qgroup *member;
 142 };
 143
 144 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
 145 {
 146         return (u64)(uintptr_t)qg;
 147 }
 148
 149 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
 150 {
 151         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
 152 }
 153
 154 static int
 155 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 156                    int init_flags);
 157 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 158
 159 /* must be called with qgroup_ioctl_lock held */
 160 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 161                                            u64 qgroupid)
 162 {
 163         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 164         struct btrfs_qgroup *qgroup;
 165
 166         while (n) {
 167                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 168                 if (qgroup->qgroupid < qgroupid)
 169                         n = n->rb_left;
 170                 else if (qgroup->qgroupid > qgroupid)
 171                         n = n->rb_right;
 172                 else
 173                         return qgroup;
 174         }
 175         return NULL;
 176 }
 177
 178 /* must be called with qgroup_lock held */
 179 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 180                                           u64 qgroupid)
 181 {
 182         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 183         struct rb_node *parent = NULL;
 184         struct btrfs_qgroup *qgroup;
 185
 186         while (*p) {
 187                 parent = *p;
 188                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 189
 190                 if (qgroup->qgroupid < qgroupid)
 191                         p = &(*p)->rb_left;
 192                 else if (qgroup->qgroupid > qgroupid)
 193                         p = &(*p)->rb_right;
 194                 else
 195                         return qgroup;
 196         }
 197
 198         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 199         if (!qgroup)
 200                 return ERR_PTR(-ENOMEM);
 201
 202         qgroup->qgroupid = qgroupid;
 203         INIT_LIST_HEAD(&qgroup->groups);
 204         INIT_LIST_HEAD(&qgroup->members);
 205         INIT_LIST_HEAD(&qgroup->dirty);
 206
 207         rb_link_node(&qgroup->node, parent, p);
 208         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 209
 210         return qgroup;
 211 }
 212
 213 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
 214                             struct btrfs_qgroup *qgroup)
 215 {
 216         struct btrfs_qgroup_list *list;
 217
 218         list_del(&qgroup->dirty);
 219         while (!list_empty(&qgroup->groups)) {
 220                 list = list_first_entry(&qgroup->groups,
 221                                         struct btrfs_qgroup_list, next_group);
 222                 list_del(&list->next_group);
 223                 list_del(&list->next_member);
 224                 kfree(list);
 225         }
 226
 227         while (!list_empty(&qgroup->members)) {
 228                 list = list_first_entry(&qgroup->members,
 229                                         struct btrfs_qgroup_list, next_member);
 230                 list_del(&list->next_group);
 231                 list_del(&list->next_member);
 232                 kfree(list);
 233         }
 234 }
 235
 236 /* must be called with qgroup_lock held */
 237 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 238 {
 239         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 240
 241         if (!qgroup)
 242                 return -ENOENT;
 243
 244         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 245         __del_qgroup_rb(fs_info, qgroup);
 246         return 0;
 247 }
 248
 249 /*
 250  * Add relation specified by two qgroups.
 251  *
 252  * Must be called with qgroup_lock held.
 253  *
 254  * Return: 0        on success
 255  *         -ENOENT  if one of the qgroups is NULL
 256  *         <0       other errors
 257  */
 258 static int __add_relation_rb(struct btrfs_qgroup *member, struct btrfs_qgroup *parent)
 259 {
 260         struct btrfs_qgroup_list *list;
 261
 262         if (!member || !parent)
 263                 return -ENOENT;
 264
 265         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 266         if (!list)
 267                 return -ENOMEM;
 268
 269         list->group = parent;
 270         list->member = member;
 271         list_add_tail(&list->next_group, &member->groups);
 272         list_add_tail(&list->next_member, &parent->members);
 273
 274         return 0;
 275 }
 276
 277 /*
 278  * Add relation specified by two qgoup ids.
 279  *
 280  * Must be called with qgroup_lock held.
 281  *
 282  * Return: 0        on success
 283  *         -ENOENT  if one of the ids does not exist
 284  *         <0       other errors
 285  */
 286 static int add_relation_rb(struct btrfs_fs_info *fs_info, u64 memberid, u64 parentid)
 287 {
 288         struct btrfs_qgroup *member;
 289         struct btrfs_qgroup *parent;
 290
 291         member = find_qgroup_rb(fs_info, memberid);
 292         parent = find_qgroup_rb(fs_info, parentid);
 293
 294         return __add_relation_rb(member, parent);
 295 }
 296
 297 /* Must be called with qgroup_lock held */
 298 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 299                            u64 memberid, u64 parentid)
 300 {
 301         struct btrfs_qgroup *member;
 302         struct btrfs_qgroup *parent;
 303         struct btrfs_qgroup_list *list;
 304
 305         member = find_qgroup_rb(fs_info, memberid);
 306         parent = find_qgroup_rb(fs_info, parentid);
 307         if (!member || !parent)
 308                 return -ENOENT;
 309
 310         list_for_each_entry(list, &member->groups, next_group) {
 311                 if (list->group == parent) {
 312                         list_del(&list->next_group);
 313                         list_del(&list->next_member);
 314                         kfree(list);
 315                         return 0;
 316                 }
 317         }
 318         return -ENOENT;
 319 }
 320
 321 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 322 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 323                                u64 rfer, u64 excl)
 324 {
 325         struct btrfs_qgroup *qgroup;
 326
 327         qgroup = find_qgroup_rb(fs_info, qgroupid);
 328         if (!qgroup)
 329                 return -EINVAL;
 330         if (qgroup->rfer != rfer || qgroup->excl != excl)
 331                 return -EINVAL;
 332         return 0;
 333 }
 334 #endif
 335
 336 /*
 337  * The full config is read in one go, only called from open_ctree()
 338  * It doesn't use any locking, as at this point we're still single-threaded
 339  */
 340 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 341 {
 342         struct btrfs_key key;
 343         struct btrfs_key found_key;
 344         struct btrfs_root *quota_root = fs_info->quota_root;
 345         struct btrfs_path *path = NULL;
 346         struct extent_buffer *l;
 347         int slot;
 348         int ret = 0;
 349         u64 flags = 0;
 350         u64 rescan_progress = 0;
 351
 352         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 353                 return 0;
 354
 355         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 356         if (!fs_info->qgroup_ulist) {
 357                 ret = -ENOMEM;
 358                 goto out;
 359         }
 360
 361         path = btrfs_alloc_path();
 362         if (!path) {
 363                 ret = -ENOMEM;
 364                 goto out;
 365         }
 366
 367         ret = btrfs_sysfs_add_qgroups(fs_info);
 368         if (ret < 0)
 369                 goto out;
 370         /* default this to quota off, in case no status key is found */
 371         fs_info->qgroup_flags = 0;
 372
 373         /*
 374          * pass 1: read status, all qgroup infos and limits
 375          */
 376         key.objectid = 0;
 377         key.type = 0;
 378         key.offset = 0;
 379         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 380         if (ret)
 381                 goto out;
 382
 383         while (1) {
 384                 struct btrfs_qgroup *qgroup;
 385
 386                 slot = path->slots[0];
 387                 l = path->nodes[0];
 388                 btrfs_item_key_to_cpu(l, &found_key, slot);
 389
 390                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 391                         struct btrfs_qgroup_status_item *ptr;
 392
 393                         ptr = btrfs_item_ptr(l, slot,
 394                                              struct btrfs_qgroup_status_item);
 395
 396                         if (btrfs_qgroup_status_version(l, ptr) !=
 397                             BTRFS_QGROUP_STATUS_VERSION) {
 398                                 btrfs_err(fs_info,
 399                                  "old qgroup version, quota disabled");
 400                                 goto out;
 401                         }
 402                         if (btrfs_qgroup_status_generation(l, ptr) !=
 403                             fs_info->generation) {
 404                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 405                                 btrfs_err(fs_info,
 406                                         "qgroup generation mismatch, marked as inconsistent");
 407                         }
 408                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 409                                                                           ptr);
 410                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 411                         goto next1;
 412                 }
 413
 414                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 415                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 416                         goto next1;
 417
 418                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 419                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 420                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 421                         btrfs_err(fs_info, "inconsistent qgroup config");
 422                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 423                 }
 424                 if (!qgroup) {
 425                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 426                         if (IS_ERR(qgroup)) {
 427                                 ret = PTR_ERR(qgroup);
 428                                 goto out;
 429                         }
 430                 }
 431                 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
 432                 if (ret < 0)
 433                         goto out;
 434
 435                 switch (found_key.type) {
 436                 case BTRFS_QGROUP_INFO_KEY: {
 437                         struct btrfs_qgroup_info_item *ptr;
 438
 439                         ptr = btrfs_item_ptr(l, slot,
 440                                              struct btrfs_qgroup_info_item);
 441                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 442                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 443                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 444                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 445                         /* generation currently unused */
 446                         break;
 447                 }
 448                 case BTRFS_QGROUP_LIMIT_KEY: {
 449                         struct btrfs_qgroup_limit_item *ptr;
 450
 451                         ptr = btrfs_item_ptr(l, slot,
 452                                              struct btrfs_qgroup_limit_item);
 453                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 454                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 455                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 456                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 457                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 458                         break;
 459                 }
 460                 }
 461 next1:
 462                 ret = btrfs_next_item(quota_root, path);
 463                 if (ret < 0)
 464                         goto out;
 465                 if (ret)
 466                         break;
 467         }
 468         btrfs_release_path(path);
 469
 470         /*
 471          * pass 2: read all qgroup relations
 472          */
 473         key.objectid = 0;
 474         key.type = BTRFS_QGROUP_RELATION_KEY;
 475         key.offset = 0;
 476         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 477         if (ret)
 478                 goto out;
 479         while (1) {
 480                 slot = path->slots[0];
 481                 l = path->nodes[0];
 482                 btrfs_item_key_to_cpu(l, &found_key, slot);
 483
 484                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 485                         goto next2;
 486
 487                 if (found_key.objectid > found_key.offset) {
 488                         /* parent <- member, not needed to build config */
 489                         /* FIXME should we omit the key completely? */
 490                         goto next2;
 491                 }
 492
 493                 ret = add_relation_rb(fs_info, found_key.objectid,
 494                                       found_key.offset);
 495                 if (ret == -ENOENT) {
 496                         btrfs_warn(fs_info,
 497                                 "orphan qgroup relation 0x%llx->0x%llx",
 498                                 found_key.objectid, found_key.offset);
 499                         ret = 0;        /* ignore the error */
 500                 }
 501                 if (ret)
 502                         goto out;
 503 next2:
 504                 ret = btrfs_next_item(quota_root, path);
 505                 if (ret < 0)
 506                         goto out;
 507                 if (ret)
 508                         break;
 509         }
 510 out:
 511         btrfs_free_path(path);
 512         fs_info->qgroup_flags |= flags;
 513         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
 514                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 515         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 516                  ret >= 0)
 517                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 518
 519         if (ret < 0) {
 520                 ulist_free(fs_info->qgroup_ulist);
 521                 fs_info->qgroup_ulist = NULL;
 522                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 523                 btrfs_sysfs_del_qgroups(fs_info);
 524         }
 525
 526         return ret < 0 ? ret : 0;
 527 }
 528
 529 /*
 530  * Called in close_ctree() when quota is still enabled.  This verifies we don't
 531  * leak some reserved space.
 532  *
 533  * Return false if no reserved space is left.
 534  * Return true if some reserved space is leaked.
 535  */
 536 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
 537 {
 538         struct rb_node *node;
 539         bool ret = false;
 540
 541         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 542                 return ret;
 543         /*
 544          * Since we're unmounting, there is no race and no need to grab qgroup
 545          * lock.  And here we don't go post-order to provide a more user
 546          * friendly sorted result.
 547          */
 548         for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
 549                 struct btrfs_qgroup *qgroup;
 550                 int i;
 551
 552                 qgroup = rb_entry(node, struct btrfs_qgroup, node);
 553                 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
 554                         if (qgroup->rsv.values[i]) {
 555                                 ret = true;
 556                                 btrfs_warn(fs_info,
 557                 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
 558                                    btrfs_qgroup_level(qgroup->qgroupid),
 559                                    btrfs_qgroup_subvolid(qgroup->qgroupid),
 560                                    i, qgroup->rsv.values[i]);
 561                         }
 562                 }
 563         }
 564         return ret;
 565 }
 566
 567 /*
 568  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 569  * first two are in single-threaded paths.And for the third one, we have set
 570  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 571  * up the in-memory structures without qgroup_lock held.
 572  */
 573 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 574 {
 575         struct rb_node *n;
 576         struct btrfs_qgroup *qgroup;
 577
 578         while ((n = rb_first(&fs_info->qgroup_tree))) {
 579                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 580                 rb_erase(n, &fs_info->qgroup_tree);
 581                 __del_qgroup_rb(fs_info, qgroup);
 582                 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
 583                 kfree(qgroup);
 584         }
 585         /*
 586          * We call btrfs_free_qgroup_config() when unmounting
 587          * filesystem and disabling quota, so we set qgroup_ulist
 588          * to be null here to avoid double free.
 589          */
 590         ulist_free(fs_info->qgroup_ulist);
 591         fs_info->qgroup_ulist = NULL;
 592         btrfs_sysfs_del_qgroups(fs_info);
 593 }
 594
 595 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 596                                     u64 dst)
 597 {
 598         int ret;
 599         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 600         struct btrfs_path *path;
 601         struct btrfs_key key;
 602
 603         path = btrfs_alloc_path();
 604         if (!path)
 605                 return -ENOMEM;
 606
 607         key.objectid = src;
 608         key.type = BTRFS_QGROUP_RELATION_KEY;
 609         key.offset = dst;
 610
 611         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 612
 613         btrfs_mark_buffer_dirty(path->nodes[0]);
 614
 615         btrfs_free_path(path);
 616         return ret;
 617 }
 618
 619 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 620                                     u64 dst)
 621 {
 622         int ret;
 623         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 624         struct btrfs_path *path;
 625         struct btrfs_key key;
 626
 627         path = btrfs_alloc_path();
 628         if (!path)
 629                 return -ENOMEM;
 630
 631         key.objectid = src;
 632         key.type = BTRFS_QGROUP_RELATION_KEY;
 633         key.offset = dst;
 634
 635         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 636         if (ret < 0)
 637                 goto out;
 638
 639         if (ret > 0) {
 640                 ret = -ENOENT;
 641                 goto out;
 642         }
 643
 644         ret = btrfs_del_item(trans, quota_root, path);
 645 out:
 646         btrfs_free_path(path);
 647         return ret;
 648 }
 649
 650 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 651                            struct btrfs_root *quota_root, u64 qgroupid)
 652 {
 653         int ret;
 654         struct btrfs_path *path;
 655         struct btrfs_qgroup_info_item *qgroup_info;
 656         struct btrfs_qgroup_limit_item *qgroup_limit;
 657         struct extent_buffer *leaf;
 658         struct btrfs_key key;
 659
 660         if (btrfs_is_testing(quota_root->fs_info))
 661                 return 0;
 662
 663         path = btrfs_alloc_path();
 664         if (!path)
 665                 return -ENOMEM;
 666
 667         key.objectid = 0;
 668         key.type = BTRFS_QGROUP_INFO_KEY;
 669         key.offset = qgroupid;
 670
 671         /*
 672          * Avoid a transaction abort by catching -EEXIST here. In that
 673          * case, we proceed by re-initializing the existing structure
 674          * on disk.
 675          */
 676
 677         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 678                                       sizeof(*qgroup_info));
 679         if (ret && ret != -EEXIST)
 680                 goto out;
 681
 682         leaf = path->nodes[0];
 683         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 684                                  struct btrfs_qgroup_info_item);
 685         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 686         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 687         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 688         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 689         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 690
 691         btrfs_mark_buffer_dirty(leaf);
 692
 693         btrfs_release_path(path);
 694
 695         key.type = BTRFS_QGROUP_LIMIT_KEY;
 696         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 697                                       sizeof(*qgroup_limit));
 698         if (ret && ret != -EEXIST)
 699                 goto out;
 700
 701         leaf = path->nodes[0];
 702         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 703                                   struct btrfs_qgroup_limit_item);
 704         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 705         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 706         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 707         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 708         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 709
 710         btrfs_mark_buffer_dirty(leaf);
 711
 712         ret = 0;
 713 out:
 714         btrfs_free_path(path);
 715         return ret;
 716 }
 717
 718 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
 719 {
 720         int ret;
 721         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 722         struct btrfs_path *path;
 723         struct btrfs_key key;
 724
 725         path = btrfs_alloc_path();
 726         if (!path)
 727                 return -ENOMEM;
 728
 729         key.objectid = 0;
 730         key.type = BTRFS_QGROUP_INFO_KEY;
 731         key.offset = qgroupid;
 732         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 733         if (ret < 0)
 734                 goto out;
 735
 736         if (ret > 0) {
 737                 ret = -ENOENT;
 738                 goto out;
 739         }
 740
 741         ret = btrfs_del_item(trans, quota_root, path);
 742         if (ret)
 743                 goto out;
 744
 745         btrfs_release_path(path);
 746
 747         key.type = BTRFS_QGROUP_LIMIT_KEY;
 748         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 749         if (ret < 0)
 750                 goto out;
 751
 752         if (ret > 0) {
 753                 ret = -ENOENT;
 754                 goto out;
 755         }
 756
 757         ret = btrfs_del_item(trans, quota_root, path);
 758
 759 out:
 760         btrfs_free_path(path);
 761         return ret;
 762 }
 763
 764 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 765                                     struct btrfs_qgroup *qgroup)
 766 {
 767         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 768         struct btrfs_path *path;
 769         struct btrfs_key key;
 770         struct extent_buffer *l;
 771         struct btrfs_qgroup_limit_item *qgroup_limit;
 772         int ret;
 773         int slot;
 774
 775         key.objectid = 0;
 776         key.type = BTRFS_QGROUP_LIMIT_KEY;
 777         key.offset = qgroup->qgroupid;
 778
 779         path = btrfs_alloc_path();
 780         if (!path)
 781                 return -ENOMEM;
 782
 783         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 784         if (ret > 0)
 785                 ret = -ENOENT;
 786
 787         if (ret)
 788                 goto out;
 789
 790         l = path->nodes[0];
 791         slot = path->slots[0];
 792         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 793         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 794         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 795         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 796         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 797         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 798
 799         btrfs_mark_buffer_dirty(l);
 800
 801 out:
 802         btrfs_free_path(path);
 803         return ret;
 804 }
 805
 806 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 807                                    struct btrfs_qgroup *qgroup)
 808 {
 809         struct btrfs_fs_info *fs_info = trans->fs_info;
 810         struct btrfs_root *quota_root = fs_info->quota_root;
 811         struct btrfs_path *path;
 812         struct btrfs_key key;
 813         struct extent_buffer *l;
 814         struct btrfs_qgroup_info_item *qgroup_info;
 815         int ret;
 816         int slot;
 817
 818         if (btrfs_is_testing(fs_info))
 819                 return 0;
 820
 821         key.objectid = 0;
 822         key.type = BTRFS_QGROUP_INFO_KEY;
 823         key.offset = qgroup->qgroupid;
 824
 825         path = btrfs_alloc_path();
 826         if (!path)
 827                 return -ENOMEM;
 828
 829         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 830         if (ret > 0)
 831                 ret = -ENOENT;
 832
 833         if (ret)
 834                 goto out;
 835
 836         l = path->nodes[0];
 837         slot = path->slots[0];
 838         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 839         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 840         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 841         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 842         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 843         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 844
 845         btrfs_mark_buffer_dirty(l);
 846
 847 out:
 848         btrfs_free_path(path);
 849         return ret;
 850 }
 851
 852 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
 853 {
 854         struct btrfs_fs_info *fs_info = trans->fs_info;
 855         struct btrfs_root *quota_root = fs_info->quota_root;
 856         struct btrfs_path *path;
 857         struct btrfs_key key;
 858         struct extent_buffer *l;
 859         struct btrfs_qgroup_status_item *ptr;
 860         int ret;
 861         int slot;
 862
 863         key.objectid = 0;
 864         key.type = BTRFS_QGROUP_STATUS_KEY;
 865         key.offset = 0;
 866
 867         path = btrfs_alloc_path();
 868         if (!path)
 869                 return -ENOMEM;
 870
 871         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 872         if (ret > 0)
 873                 ret = -ENOENT;
 874
 875         if (ret)
 876                 goto out;
 877
 878         l = path->nodes[0];
 879         slot = path->slots[0];
 880         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 881         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 882         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 883         btrfs_set_qgroup_status_rescan(l, ptr,
 884                                 fs_info->qgroup_rescan_progress.objectid);
 885
 886         btrfs_mark_buffer_dirty(l);
 887
 888 out:
 889         btrfs_free_path(path);
 890         return ret;
 891 }
 892
 893 /*
 894  * called with qgroup_lock held
 895  */
 896 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 897                                   struct btrfs_root *root)
 898 {
 899         struct btrfs_path *path;
 900         struct btrfs_key key;
 901         struct extent_buffer *leaf = NULL;
 902         int ret;
 903         int nr = 0;
 904
 905         path = btrfs_alloc_path();
 906         if (!path)
 907                 return -ENOMEM;
 908
 909         key.objectid = 0;
 910         key.offset = 0;
 911         key.type = 0;
 912
 913         while (1) {
 914                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 915                 if (ret < 0)
 916                         goto out;
 917                 leaf = path->nodes[0];
 918                 nr = btrfs_header_nritems(leaf);
 919                 if (!nr)
 920                         break;
 921                 /*
 922                  * delete the leaf one by one
 923                  * since the whole tree is going
 924                  * to be deleted.
 925                  */
 926                 path->slots[0] = 0;
 927                 ret = btrfs_del_items(trans, root, path, 0, nr);
 928                 if (ret)
 929                         goto out;
 930
 931                 btrfs_release_path(path);
 932         }
 933         ret = 0;
 934 out:
 935         btrfs_free_path(path);
 936         return ret;
 937 }
 938
 939 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 940 {
 941         struct btrfs_root *quota_root;
 942         struct btrfs_root *tree_root = fs_info->tree_root;
 943         struct btrfs_path *path = NULL;
 944         struct btrfs_qgroup_status_item *ptr;
 945         struct extent_buffer *leaf;
 946         struct btrfs_key key;
 947         struct btrfs_key found_key;
 948         struct btrfs_qgroup *qgroup = NULL;
 949         struct btrfs_trans_handle *trans = NULL;
 950         struct ulist *ulist = NULL;
 951         int ret = 0;
 952         int slot;
 953
 954         /*
 955          * We need to have subvol_sem write locked, to prevent races between
 956          * concurrent tasks trying to enable quotas, because we will unlock
 957          * and relock qgroup_ioctl_lock before setting fs_info->quota_root
 958          * and before setting BTRFS_FS_QUOTA_ENABLED.
 959          */
 960         lockdep_assert_held_write(&fs_info->subvol_sem);
 961
 962         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 963                 btrfs_err(fs_info,
 964                           "qgroups are currently unsupported in extent tree v2");
 965                 return -EINVAL;
 966         }
 967
 968         mutex_lock(&fs_info->qgroup_ioctl_lock);
 969         if (fs_info->quota_root)
 970                 goto out;
 971
 972         ulist = ulist_alloc(GFP_KERNEL);
 973         if (!ulist) {
 974                 ret = -ENOMEM;
 975                 goto out;
 976         }
 977
 978         ret = btrfs_sysfs_add_qgroups(fs_info);
 979         if (ret < 0)
 980                 goto out;
 981
 982         /*
 983          * Unlock qgroup_ioctl_lock before starting the transaction. This is to
 984          * avoid lock acquisition inversion problems (reported by lockdep) between
 985          * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
 986          * start a transaction.
 987          * After we started the transaction lock qgroup_ioctl_lock again and
 988          * check if someone else created the quota root in the meanwhile. If so,
 989          * just return success and release the transaction handle.
 990          *
 991          * Also we don't need to worry about someone else calling
 992          * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
 993          * that function returns 0 (success) when the sysfs entries already exist.
 994          */
 995         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 996
 997         /*
 998          * 1 for quota root item
 999          * 1 for BTRFS_QGROUP_STATUS item
1000          *
1001          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1002          * per subvolume. However those are not currently reserved since it
1003          * would be a lot of overkill.
1004          */
1005         trans = btrfs_start_transaction(tree_root, 2);
1006
1007         mutex_lock(&fs_info->qgroup_ioctl_lock);
1008         if (IS_ERR(trans)) {
1009                 ret = PTR_ERR(trans);
1010                 trans = NULL;
1011                 goto out;
1012         }
1013
1014         if (fs_info->quota_root)
1015                 goto out;
1016
1017         fs_info->qgroup_ulist = ulist;
1018         ulist = NULL;
1019
1020         /*
1021          * initially create the quota tree
1022          */
1023         quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1024         if (IS_ERR(quota_root)) {
1025                 ret =  PTR_ERR(quota_root);
1026                 btrfs_abort_transaction(trans, ret);
1027                 goto out;
1028         }
1029
1030         path = btrfs_alloc_path();
1031         if (!path) {
1032                 ret = -ENOMEM;
1033                 btrfs_abort_transaction(trans, ret);
1034                 goto out_free_root;
1035         }
1036
1037         key.objectid = 0;
1038         key.type = BTRFS_QGROUP_STATUS_KEY;
1039         key.offset = 0;
1040
1041         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1042                                       sizeof(*ptr));
1043         if (ret) {
1044                 btrfs_abort_transaction(trans, ret);
1045                 goto out_free_path;
1046         }
1047
1048         leaf = path->nodes[0];
1049         ptr = btrfs_item_ptr(leaf, path->slots[0],
1050                                  struct btrfs_qgroup_status_item);
1051         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1052         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1053         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1054                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1055         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
1056         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1057
1058         btrfs_mark_buffer_dirty(leaf);
1059
1060         key.objectid = 0;
1061         key.type = BTRFS_ROOT_REF_KEY;
1062         key.offset = 0;
1063
1064         btrfs_release_path(path);
1065         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1066         if (ret > 0)
1067                 goto out_add_root;
1068         if (ret < 0) {
1069                 btrfs_abort_transaction(trans, ret);
1070                 goto out_free_path;
1071         }
1072
1073         while (1) {
1074                 slot = path->slots[0];
1075                 leaf = path->nodes[0];
1076                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1077
1078                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1079
1080                         /* Release locks on tree_root before we access quota_root */
1081                         btrfs_release_path(path);
1082
1083                         ret = add_qgroup_item(trans, quota_root,
1084                                               found_key.offset);
1085                         if (ret) {
1086                                 btrfs_abort_transaction(trans, ret);
1087                                 goto out_free_path;
1088                         }
1089
1090                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
1091                         if (IS_ERR(qgroup)) {
1092                                 ret = PTR_ERR(qgroup);
1093                                 btrfs_abort_transaction(trans, ret);
1094                                 goto out_free_path;
1095                         }
1096                         ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1097                         if (ret < 0) {
1098                                 btrfs_abort_transaction(trans, ret);
1099                                 goto out_free_path;
1100                         }
1101                         ret = btrfs_search_slot_for_read(tree_root, &found_key,
1102                                                          path, 1, 0);
1103                         if (ret < 0) {
1104                                 btrfs_abort_transaction(trans, ret);
1105                                 goto out_free_path;
1106                         }
1107                         if (ret > 0) {
1108                                 /*
1109                                  * Shouldn't happen, but in case it does we
1110                                  * don't need to do the btrfs_next_item, just
1111                                  * continue.
1112                                  */
1113                                 continue;
1114                         }
1115                 }
1116                 ret = btrfs_next_item(tree_root, path);
1117                 if (ret < 0) {
1118                         btrfs_abort_transaction(trans, ret);
1119                         goto out_free_path;
1120                 }
1121                 if (ret)
1122                         break;
1123         }
1124
1125 out_add_root:
1126         btrfs_release_path(path);
1127         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1128         if (ret) {
1129                 btrfs_abort_transaction(trans, ret);
1130                 goto out_free_path;
1131         }
1132
1133         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1134         if (IS_ERR(qgroup)) {
1135                 ret = PTR_ERR(qgroup);
1136                 btrfs_abort_transaction(trans, ret);
1137                 goto out_free_path;
1138         }
1139         ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1140         if (ret < 0) {
1141                 btrfs_abort_transaction(trans, ret);
1142                 goto out_free_path;
1143         }
1144
1145         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1146         /*
1147          * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1148          * a deadlock with tasks concurrently doing other qgroup operations, such
1149          * adding/removing qgroups or adding/deleting qgroup relations for example,
1150          * because all qgroup operations first start or join a transaction and then
1151          * lock the qgroup_ioctl_lock mutex.
1152          * We are safe from a concurrent task trying to enable quotas, by calling
1153          * this function, since we are serialized by fs_info->subvol_sem.
1154          */
1155         ret = btrfs_commit_transaction(trans);
1156         trans = NULL;
1157         mutex_lock(&fs_info->qgroup_ioctl_lock);
1158         if (ret)
1159                 goto out_free_path;
1160
1161         /*
1162          * Set quota enabled flag after committing the transaction, to avoid
1163          * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1164          * creation.
1165          */
1166         spin_lock(&fs_info->qgroup_lock);
1167         fs_info->quota_root = quota_root;
1168         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1169         spin_unlock(&fs_info->qgroup_lock);
1170
1171         ret = qgroup_rescan_init(fs_info, 0, 1);
1172         if (!ret) {
1173                 qgroup_rescan_zero_tracking(fs_info);
1174                 fs_info->qgroup_rescan_running = true;
1175                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1176                                  &fs_info->qgroup_rescan_work);
1177         }
1178
1179 out_free_path:
1180         btrfs_free_path(path);
1181 out_free_root:
1182         if (ret)
1183                 btrfs_put_root(quota_root);
1184 out:
1185         if (ret) {
1186                 ulist_free(fs_info->qgroup_ulist);
1187                 fs_info->qgroup_ulist = NULL;
1188                 btrfs_sysfs_del_qgroups(fs_info);
1189         }
1190         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1191         if (ret && trans)
1192                 btrfs_end_transaction(trans);
1193         else if (trans)
1194                 ret = btrfs_end_transaction(trans);
1195         ulist_free(ulist);
1196         return ret;
1197 }
1198
1199 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1200 {
1201         struct btrfs_root *quota_root;
1202         struct btrfs_trans_handle *trans = NULL;
1203         int ret = 0;
1204
1205         /*
1206          * We need to have subvol_sem write locked, to prevent races between
1207          * concurrent tasks trying to disable quotas, because we will unlock
1208          * and relock qgroup_ioctl_lock across BTRFS_FS_QUOTA_ENABLED changes.
1209          */
1210         lockdep_assert_held_write(&fs_info->subvol_sem);
1211
1212         mutex_lock(&fs_info->qgroup_ioctl_lock);
1213         if (!fs_info->quota_root)
1214                 goto out;
1215
1216         /*
1217          * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1218          * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1219          * to lock that mutex while holding a transaction handle and the rescan
1220          * worker needs to commit a transaction.
1221          */
1222         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1223
1224         /*
1225          * Request qgroup rescan worker to complete and wait for it. This wait
1226          * must be done before transaction start for quota disable since it may
1227          * deadlock with transaction by the qgroup rescan worker.
1228          */
1229         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1230         btrfs_qgroup_wait_for_completion(fs_info, false);
1231
1232         /*
1233          * 1 For the root item
1234          *
1235          * We should also reserve enough items for the quota tree deletion in
1236          * btrfs_clean_quota_tree but this is not done.
1237          *
1238          * Also, we must always start a transaction without holding the mutex
1239          * qgroup_ioctl_lock, see btrfs_quota_enable().
1240          */
1241         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1242
1243         mutex_lock(&fs_info->qgroup_ioctl_lock);
1244         if (IS_ERR(trans)) {
1245                 ret = PTR_ERR(trans);
1246                 trans = NULL;
1247                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1248                 goto out;
1249         }
1250
1251         if (!fs_info->quota_root)
1252                 goto out;
1253
1254         spin_lock(&fs_info->qgroup_lock);
1255         quota_root = fs_info->quota_root;
1256         fs_info->quota_root = NULL;
1257         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1258         spin_unlock(&fs_info->qgroup_lock);
1259
1260         btrfs_free_qgroup_config(fs_info);
1261
1262         ret = btrfs_clean_quota_tree(trans, quota_root);
1263         if (ret) {
1264                 btrfs_abort_transaction(trans, ret);
1265                 goto out;
1266         }
1267
1268         ret = btrfs_del_root(trans, &quota_root->root_key);
1269         if (ret) {
1270                 btrfs_abort_transaction(trans, ret);
1271                 goto out;
1272         }
1273
1274         list_del(&quota_root->dirty_list);
1275
1276         btrfs_tree_lock(quota_root->node);
1277         btrfs_clean_tree_block(quota_root->node);
1278         btrfs_tree_unlock(quota_root->node);
1279         btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1280                               quota_root->node, 0, 1);
1281
1282         btrfs_put_root(quota_root);
1283
1284 out:
1285         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1286         if (ret && trans)
1287                 btrfs_end_transaction(trans);
1288         else if (trans)
1289                 ret = btrfs_end_transaction(trans);
1290
1291         return ret;
1292 }
1293
1294 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1295                          struct btrfs_qgroup *qgroup)
1296 {
1297         if (list_empty(&qgroup->dirty))
1298                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1299 }
1300
1301 /*
1302  * The easy accounting, we're updating qgroup relationship whose child qgroup
1303  * only has exclusive extents.
1304  *
1305  * In this case, all exclusive extents will also be exclusive for parent, so
1306  * excl/rfer just get added/removed.
1307  *
1308  * So is qgroup reservation space, which should also be added/removed to
1309  * parent.
1310  * Or when child tries to release reservation space, parent will underflow its
1311  * reservation (for relationship adding case).
1312  *
1313  * Caller should hold fs_info->qgroup_lock.
1314  */
1315 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1316                                     struct ulist *tmp, u64 ref_root,
1317                                     struct btrfs_qgroup *src, int sign)
1318 {
1319         struct btrfs_qgroup *qgroup;
1320         struct btrfs_qgroup_list *glist;
1321         struct ulist_node *unode;
1322         struct ulist_iterator uiter;
1323         u64 num_bytes = src->excl;
1324         int ret = 0;
1325
1326         qgroup = find_qgroup_rb(fs_info, ref_root);
1327         if (!qgroup)
1328                 goto out;
1329
1330         qgroup->rfer += sign * num_bytes;
1331         qgroup->rfer_cmpr += sign * num_bytes;
1332
1333         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1334         qgroup->excl += sign * num_bytes;
1335         qgroup->excl_cmpr += sign * num_bytes;
1336
1337         if (sign > 0)
1338                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1339         else
1340                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1341
1342         qgroup_dirty(fs_info, qgroup);
1343
1344         /* Get all of the parent groups that contain this qgroup */
1345         list_for_each_entry(glist, &qgroup->groups, next_group) {
1346                 ret = ulist_add(tmp, glist->group->qgroupid,
1347                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1348                 if (ret < 0)
1349                         goto out;
1350         }
1351
1352         /* Iterate all of the parents and adjust their reference counts */
1353         ULIST_ITER_INIT(&uiter);
1354         while ((unode = ulist_next(tmp, &uiter))) {
1355                 qgroup = unode_aux_to_qgroup(unode);
1356                 qgroup->rfer += sign * num_bytes;
1357                 qgroup->rfer_cmpr += sign * num_bytes;
1358                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1359                 qgroup->excl += sign * num_bytes;
1360                 if (sign > 0)
1361                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1362                 else
1363                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1364                 qgroup->excl_cmpr += sign * num_bytes;
1365                 qgroup_dirty(fs_info, qgroup);
1366
1367                 /* Add any parents of the parents */
1368                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1369                         ret = ulist_add(tmp, glist->group->qgroupid,
1370                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1371                         if (ret < 0)
1372                                 goto out;
1373                 }
1374         }
1375         ret = 0;
1376 out:
1377         return ret;
1378 }
1379
1380
1381 /*
1382  * Quick path for updating qgroup with only excl refs.
1383  *
1384  * In that case, just update all parent will be enough.
1385  * Or we needs to do a full rescan.
1386  * Caller should also hold fs_info->qgroup_lock.
1387  *
1388  * Return 0 for quick update, return >0 for need to full rescan
1389  * and mark INCONSISTENT flag.
1390  * Return < 0 for other error.
1391  */
1392 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1393                                    struct ulist *tmp, u64 src, u64 dst,
1394                                    int sign)
1395 {
1396         struct btrfs_qgroup *qgroup;
1397         int ret = 1;
1398         int err = 0;
1399
1400         qgroup = find_qgroup_rb(fs_info, src);
1401         if (!qgroup)
1402                 goto out;
1403         if (qgroup->excl == qgroup->rfer) {
1404                 ret = 0;
1405                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1406                                                qgroup, sign);
1407                 if (err < 0) {
1408                         ret = err;
1409                         goto out;
1410                 }
1411         }
1412 out:
1413         if (ret)
1414                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1415         return ret;
1416 }
1417
1418 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1419                               u64 dst)
1420 {
1421         struct btrfs_fs_info *fs_info = trans->fs_info;
1422         struct btrfs_qgroup *parent;
1423         struct btrfs_qgroup *member;
1424         struct btrfs_qgroup_list *list;
1425         struct ulist *tmp;
1426         unsigned int nofs_flag;
1427         int ret = 0;
1428
1429         /* Check the level of src and dst first */
1430         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1431                 return -EINVAL;
1432
1433         /* We hold a transaction handle open, must do a NOFS allocation. */
1434         nofs_flag = memalloc_nofs_save();
1435         tmp = ulist_alloc(GFP_KERNEL);
1436         memalloc_nofs_restore(nofs_flag);
1437         if (!tmp)
1438                 return -ENOMEM;
1439
1440         mutex_lock(&fs_info->qgroup_ioctl_lock);
1441         if (!fs_info->quota_root) {
1442                 ret = -ENOTCONN;
1443                 goto out;
1444         }
1445         member = find_qgroup_rb(fs_info, src);
1446         parent = find_qgroup_rb(fs_info, dst);
1447         if (!member || !parent) {
1448                 ret = -EINVAL;
1449                 goto out;
1450         }
1451
1452         /* check if such qgroup relation exist firstly */
1453         list_for_each_entry(list, &member->groups, next_group) {
1454                 if (list->group == parent) {
1455                         ret = -EEXIST;
1456                         goto out;
1457                 }
1458         }
1459
1460         ret = add_qgroup_relation_item(trans, src, dst);
1461         if (ret)
1462                 goto out;
1463
1464         ret = add_qgroup_relation_item(trans, dst, src);
1465         if (ret) {
1466                 del_qgroup_relation_item(trans, src, dst);
1467                 goto out;
1468         }
1469
1470         spin_lock(&fs_info->qgroup_lock);
1471         ret = __add_relation_rb(member, parent);
1472         if (ret < 0) {
1473                 spin_unlock(&fs_info->qgroup_lock);
1474                 goto out;
1475         }
1476         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1477         spin_unlock(&fs_info->qgroup_lock);
1478 out:
1479         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1480         ulist_free(tmp);
1481         return ret;
1482 }
1483
1484 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1485                                  u64 dst)
1486 {
1487         struct btrfs_fs_info *fs_info = trans->fs_info;
1488         struct btrfs_qgroup *parent;
1489         struct btrfs_qgroup *member;
1490         struct btrfs_qgroup_list *list;
1491         struct ulist *tmp;
1492         bool found = false;
1493         unsigned int nofs_flag;
1494         int ret = 0;
1495         int ret2;
1496
1497         /* We hold a transaction handle open, must do a NOFS allocation. */
1498         nofs_flag = memalloc_nofs_save();
1499         tmp = ulist_alloc(GFP_KERNEL);
1500         memalloc_nofs_restore(nofs_flag);
1501         if (!tmp)
1502                 return -ENOMEM;
1503
1504         if (!fs_info->quota_root) {
1505                 ret = -ENOTCONN;
1506                 goto out;
1507         }
1508
1509         member = find_qgroup_rb(fs_info, src);
1510         parent = find_qgroup_rb(fs_info, dst);
1511         /*
1512          * The parent/member pair doesn't exist, then try to delete the dead
1513          * relation items only.
1514          */
1515         if (!member || !parent)
1516                 goto delete_item;
1517
1518         /* check if such qgroup relation exist firstly */
1519         list_for_each_entry(list, &member->groups, next_group) {
1520                 if (list->group == parent) {
1521                         found = true;
1522                         break;
1523                 }
1524         }
1525
1526 delete_item:
1527         ret = del_qgroup_relation_item(trans, src, dst);
1528         if (ret < 0 && ret != -ENOENT)
1529                 goto out;
1530         ret2 = del_qgroup_relation_item(trans, dst, src);
1531         if (ret2 < 0 && ret2 != -ENOENT)
1532                 goto out;
1533
1534         /* At least one deletion succeeded, return 0 */
1535         if (!ret || !ret2)
1536                 ret = 0;
1537
1538         if (found) {
1539                 spin_lock(&fs_info->qgroup_lock);
1540                 del_relation_rb(fs_info, src, dst);
1541                 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1542                 spin_unlock(&fs_info->qgroup_lock);
1543         }
1544 out:
1545         ulist_free(tmp);
1546         return ret;
1547 }
1548
1549 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1550                               u64 dst)
1551 {
1552         struct btrfs_fs_info *fs_info = trans->fs_info;
1553         int ret = 0;
1554
1555         mutex_lock(&fs_info->qgroup_ioctl_lock);
1556         ret = __del_qgroup_relation(trans, src, dst);
1557         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1558
1559         return ret;
1560 }
1561
1562 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1563 {
1564         struct btrfs_fs_info *fs_info = trans->fs_info;
1565         struct btrfs_root *quota_root;
1566         struct btrfs_qgroup *qgroup;
1567         int ret = 0;
1568
1569         mutex_lock(&fs_info->qgroup_ioctl_lock);
1570         if (!fs_info->quota_root) {
1571                 ret = -ENOTCONN;
1572                 goto out;
1573         }
1574         quota_root = fs_info->quota_root;
1575         qgroup = find_qgroup_rb(fs_info, qgroupid);
1576         if (qgroup) {
1577                 ret = -EEXIST;
1578                 goto out;
1579         }
1580
1581         ret = add_qgroup_item(trans, quota_root, qgroupid);
1582         if (ret)
1583                 goto out;
1584
1585         spin_lock(&fs_info->qgroup_lock);
1586         qgroup = add_qgroup_rb(fs_info, qgroupid);
1587         spin_unlock(&fs_info->qgroup_lock);
1588
1589         if (IS_ERR(qgroup)) {
1590                 ret = PTR_ERR(qgroup);
1591                 goto out;
1592         }
1593         ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1594 out:
1595         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1596         return ret;
1597 }
1598
1599 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1600 {
1601         struct btrfs_fs_info *fs_info = trans->fs_info;
1602         struct btrfs_qgroup *qgroup;
1603         struct btrfs_qgroup_list *list;
1604         int ret = 0;
1605
1606         mutex_lock(&fs_info->qgroup_ioctl_lock);
1607         if (!fs_info->quota_root) {
1608                 ret = -ENOTCONN;
1609                 goto out;
1610         }
1611
1612         qgroup = find_qgroup_rb(fs_info, qgroupid);
1613         if (!qgroup) {
1614                 ret = -ENOENT;
1615                 goto out;
1616         }
1617
1618         /* Check if there are no children of this qgroup */
1619         if (!list_empty(&qgroup->members)) {
1620                 ret = -EBUSY;
1621                 goto out;
1622         }
1623
1624         ret = del_qgroup_item(trans, qgroupid);
1625         if (ret && ret != -ENOENT)
1626                 goto out;
1627
1628         while (!list_empty(&qgroup->groups)) {
1629                 list = list_first_entry(&qgroup->groups,
1630                                         struct btrfs_qgroup_list, next_group);
1631                 ret = __del_qgroup_relation(trans, qgroupid,
1632                                             list->group->qgroupid);
1633                 if (ret)
1634                         goto out;
1635         }
1636
1637         spin_lock(&fs_info->qgroup_lock);
1638         del_qgroup_rb(fs_info, qgroupid);
1639         spin_unlock(&fs_info->qgroup_lock);
1640
1641         /*
1642          * Remove the qgroup from sysfs now without holding the qgroup_lock
1643          * spinlock, since the sysfs_remove_group() function needs to take
1644          * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1645          */
1646         btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1647         kfree(qgroup);
1648 out:
1649         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1650         return ret;
1651 }
1652
1653 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1654                        struct btrfs_qgroup_limit *limit)
1655 {
1656         struct btrfs_fs_info *fs_info = trans->fs_info;
1657         struct btrfs_qgroup *qgroup;
1658         int ret = 0;
1659         /* Sometimes we would want to clear the limit on this qgroup.
1660          * To meet this requirement, we treat the -1 as a special value
1661          * which tell kernel to clear the limit on this qgroup.
1662          */
1663         const u64 CLEAR_VALUE = -1;
1664
1665         mutex_lock(&fs_info->qgroup_ioctl_lock);
1666         if (!fs_info->quota_root) {
1667                 ret = -ENOTCONN;
1668                 goto out;
1669         }
1670
1671         qgroup = find_qgroup_rb(fs_info, qgroupid);
1672         if (!qgroup) {
1673                 ret = -ENOENT;
1674                 goto out;
1675         }
1676
1677         spin_lock(&fs_info->qgroup_lock);
1678         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1679                 if (limit->max_rfer == CLEAR_VALUE) {
1680                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1681                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1682                         qgroup->max_rfer = 0;
1683                 } else {
1684                         qgroup->max_rfer = limit->max_rfer;
1685                 }
1686         }
1687         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1688                 if (limit->max_excl == CLEAR_VALUE) {
1689                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1690                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1691                         qgroup->max_excl = 0;
1692                 } else {
1693                         qgroup->max_excl = limit->max_excl;
1694                 }
1695         }
1696         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1697                 if (limit->rsv_rfer == CLEAR_VALUE) {
1698                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1699                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1700                         qgroup->rsv_rfer = 0;
1701                 } else {
1702                         qgroup->rsv_rfer = limit->rsv_rfer;
1703                 }
1704         }
1705         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1706                 if (limit->rsv_excl == CLEAR_VALUE) {
1707                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1708                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1709                         qgroup->rsv_excl = 0;
1710                 } else {
1711                         qgroup->rsv_excl = limit->rsv_excl;
1712                 }
1713         }
1714         qgroup->lim_flags |= limit->flags;
1715
1716         spin_unlock(&fs_info->qgroup_lock);
1717
1718         ret = update_qgroup_limit_item(trans, qgroup);
1719         if (ret) {
1720                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1721                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1722                        qgroupid);
1723         }
1724
1725 out:
1726         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1727         return ret;
1728 }
1729
1730 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1731                                 struct btrfs_delayed_ref_root *delayed_refs,
1732                                 struct btrfs_qgroup_extent_record *record)
1733 {
1734         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1735         struct rb_node *parent_node = NULL;
1736         struct btrfs_qgroup_extent_record *entry;
1737         u64 bytenr = record->bytenr;
1738
1739         lockdep_assert_held(&delayed_refs->lock);
1740         trace_btrfs_qgroup_trace_extent(fs_info, record);
1741
1742         while (*p) {
1743                 parent_node = *p;
1744                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1745                                  node);
1746                 if (bytenr < entry->bytenr) {
1747                         p = &(*p)->rb_left;
1748                 } else if (bytenr > entry->bytenr) {
1749                         p = &(*p)->rb_right;
1750                 } else {
1751                         if (record->data_rsv && !entry->data_rsv) {
1752                                 entry->data_rsv = record->data_rsv;
1753                                 entry->data_rsv_refroot =
1754                                         record->data_rsv_refroot;
1755                         }
1756                         return 1;
1757                 }
1758         }
1759
1760         rb_link_node(&record->node, parent_node, p);
1761         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1762         return 0;
1763 }
1764
1765 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1766                                    struct btrfs_qgroup_extent_record *qrecord)
1767 {
1768         struct ulist *old_root;
1769         u64 bytenr = qrecord->bytenr;
1770         int ret;
1771
1772         /*
1773          * We are always called in a context where we are already holding a
1774          * transaction handle. Often we are called when adding a data delayed
1775          * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1776          * in which case we will be holding a write lock on extent buffer from a
1777          * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1778          * acquire fs_info->commit_root_sem, because that is a higher level lock
1779          * that must be acquired before locking any extent buffers.
1780          *
1781          * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1782          * but we can't pass it a non-NULL transaction handle, because otherwise
1783          * it would not use commit roots and would lock extent buffers, causing
1784          * a deadlock if it ends up trying to read lock the same extent buffer
1785          * that was previously write locked at btrfs_truncate_inode_items().
1786          *
1787          * So pass a NULL transaction handle to btrfs_find_all_roots() and
1788          * explicitly tell it to not acquire the commit_root_sem - if we are
1789          * holding a transaction handle we don't need its protection.
1790          */
1791         ASSERT(trans != NULL);
1792
1793         ret = btrfs_find_all_roots(NULL, trans->fs_info, bytenr, 0, &old_root,
1794                                    true);
1795         if (ret < 0) {
1796                 trans->fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1797                 btrfs_warn(trans->fs_info,
1798 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1799                         ret);
1800                 return 0;
1801         }
1802
1803         /*
1804          * Here we don't need to get the lock of
1805          * trans->transaction->delayed_refs, since inserted qrecord won't
1806          * be deleted, only qrecord->node may be modified (new qrecord insert)
1807          *
1808          * So modifying qrecord->old_roots is safe here
1809          */
1810         qrecord->old_roots = old_root;
1811         return 0;
1812 }
1813
1814 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1815                               u64 num_bytes, gfp_t gfp_flag)
1816 {
1817         struct btrfs_fs_info *fs_info = trans->fs_info;
1818         struct btrfs_qgroup_extent_record *record;
1819         struct btrfs_delayed_ref_root *delayed_refs;
1820         int ret;
1821
1822         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1823             || bytenr == 0 || num_bytes == 0)
1824                 return 0;
1825         record = kzalloc(sizeof(*record), gfp_flag);
1826         if (!record)
1827                 return -ENOMEM;
1828
1829         delayed_refs = &trans->transaction->delayed_refs;
1830         record->bytenr = bytenr;
1831         record->num_bytes = num_bytes;
1832         record->old_roots = NULL;
1833
1834         spin_lock(&delayed_refs->lock);
1835         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1836         spin_unlock(&delayed_refs->lock);
1837         if (ret > 0) {
1838                 kfree(record);
1839                 return 0;
1840         }
1841         return btrfs_qgroup_trace_extent_post(trans, record);
1842 }
1843
1844 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1845                                   struct extent_buffer *eb)
1846 {
1847         struct btrfs_fs_info *fs_info = trans->fs_info;
1848         int nr = btrfs_header_nritems(eb);
1849         int i, extent_type, ret;
1850         struct btrfs_key key;
1851         struct btrfs_file_extent_item *fi;
1852         u64 bytenr, num_bytes;
1853
1854         /* We can be called directly from walk_up_proc() */
1855         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1856                 return 0;
1857
1858         for (i = 0; i < nr; i++) {
1859                 btrfs_item_key_to_cpu(eb, &key, i);
1860
1861                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1862                         continue;
1863
1864                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1865                 /* filter out non qgroup-accountable extents  */
1866                 extent_type = btrfs_file_extent_type(eb, fi);
1867
1868                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1869                         continue;
1870
1871                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1872                 if (!bytenr)
1873                         continue;
1874
1875                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1876
1877                 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1878                                                 GFP_NOFS);
1879                 if (ret)
1880                         return ret;
1881         }
1882         cond_resched();
1883         return 0;
1884 }
1885
1886 /*
1887  * Walk up the tree from the bottom, freeing leaves and any interior
1888  * nodes which have had all slots visited. If a node (leaf or
1889  * interior) is freed, the node above it will have it's slot
1890  * incremented. The root node will never be freed.
1891  *
1892  * At the end of this function, we should have a path which has all
1893  * slots incremented to the next position for a search. If we need to
1894  * read a new node it will be NULL and the node above it will have the
1895  * correct slot selected for a later read.
1896  *
1897  * If we increment the root nodes slot counter past the number of
1898  * elements, 1 is returned to signal completion of the search.
1899  */
1900 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1901 {
1902         int level = 0;
1903         int nr, slot;
1904         struct extent_buffer *eb;
1905
1906         if (root_level == 0)
1907                 return 1;
1908
1909         while (level <= root_level) {
1910                 eb = path->nodes[level];
1911                 nr = btrfs_header_nritems(eb);
1912                 path->slots[level]++;
1913                 slot = path->slots[level];
1914                 if (slot >= nr || level == 0) {
1915                         /*
1916                          * Don't free the root -  we will detect this
1917                          * condition after our loop and return a
1918                          * positive value for caller to stop walking the tree.
1919                          */
1920                         if (level != root_level) {
1921                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1922                                 path->locks[level] = 0;
1923
1924                                 free_extent_buffer(eb);
1925                                 path->nodes[level] = NULL;
1926                                 path->slots[level] = 0;
1927                         }
1928                 } else {
1929                         /*
1930                          * We have a valid slot to walk back down
1931                          * from. Stop here so caller can process these
1932                          * new nodes.
1933                          */
1934                         break;
1935                 }
1936
1937                 level++;
1938         }
1939
1940         eb = path->nodes[root_level];
1941         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1942                 return 1;
1943
1944         return 0;
1945 }
1946
1947 /*
1948  * Helper function to trace a subtree tree block swap.
1949  *
1950  * The swap will happen in highest tree block, but there may be a lot of
1951  * tree blocks involved.
1952  *
1953  * For example:
1954  *  OO = Old tree blocks
1955  *  NN = New tree blocks allocated during balance
1956  *
1957  *           File tree (257)                  Reloc tree for 257
1958  * L2              OO                                NN
1959  *               /    \                            /    \
1960  * L1          OO      OO (a)                    OO      NN (a)
1961  *            / \     / \                       / \     / \
1962  * L0       OO   OO OO   OO                   OO   OO NN   NN
1963  *                  (b)  (c)                          (b)  (c)
1964  *
1965  * When calling qgroup_trace_extent_swap(), we will pass:
1966  * @src_eb = OO(a)
1967  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1968  * @dst_level = 0
1969  * @root_level = 1
1970  *
1971  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1972  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1973  *
1974  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1975  *
1976  * 1) Tree search from @src_eb
1977  *    It should acts as a simplified btrfs_search_slot().
1978  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1979  *    (first key).
1980  *
1981  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1982  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1983  *    They should be marked during previous (@dst_level = 1) iteration.
1984  *
1985  * 3) Mark file extents in leaves dirty
1986  *    We don't have good way to pick out new file extents only.
1987  *    So we still follow the old method by scanning all file extents in
1988  *    the leave.
1989  *
1990  * This function can free us from keeping two paths, thus later we only need
1991  * to care about how to iterate all new tree blocks in reloc tree.
1992  */
1993 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1994                                     struct extent_buffer *src_eb,
1995                                     struct btrfs_path *dst_path,
1996                                     int dst_level, int root_level,
1997                                     bool trace_leaf)
1998 {
1999         struct btrfs_key key;
2000         struct btrfs_path *src_path;
2001         struct btrfs_fs_info *fs_info = trans->fs_info;
2002         u32 nodesize = fs_info->nodesize;
2003         int cur_level = root_level;
2004         int ret;
2005
2006         BUG_ON(dst_level > root_level);
2007         /* Level mismatch */
2008         if (btrfs_header_level(src_eb) != root_level)
2009                 return -EINVAL;
2010
2011         src_path = btrfs_alloc_path();
2012         if (!src_path) {
2013                 ret = -ENOMEM;
2014                 goto out;
2015         }
2016
2017         if (dst_level)
2018                 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2019         else
2020                 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2021
2022         /* For src_path */
2023         atomic_inc(&src_eb->refs);
2024         src_path->nodes[root_level] = src_eb;
2025         src_path->slots[root_level] = dst_path->slots[root_level];
2026         src_path->locks[root_level] = 0;
2027
2028         /* A simplified version of btrfs_search_slot() */
2029         while (cur_level >= dst_level) {
2030                 struct btrfs_key src_key;
2031                 struct btrfs_key dst_key;
2032
2033                 if (src_path->nodes[cur_level] == NULL) {
2034                         struct extent_buffer *eb;
2035                         int parent_slot;
2036
2037                         eb = src_path->nodes[cur_level + 1];
2038                         parent_slot = src_path->slots[cur_level + 1];
2039
2040                         eb = btrfs_read_node_slot(eb, parent_slot);
2041                         if (IS_ERR(eb)) {
2042                                 ret = PTR_ERR(eb);
2043                                 goto out;
2044                         }
2045
2046                         src_path->nodes[cur_level] = eb;
2047
2048                         btrfs_tree_read_lock(eb);
2049                         src_path->locks[cur_level] = BTRFS_READ_LOCK;
2050                 }
2051
2052                 src_path->slots[cur_level] = dst_path->slots[cur_level];
2053                 if (cur_level) {
2054                         btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2055                                         &dst_key, dst_path->slots[cur_level]);
2056                         btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2057                                         &src_key, src_path->slots[cur_level]);
2058                 } else {
2059                         btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2060                                         &dst_key, dst_path->slots[cur_level]);
2061                         btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2062                                         &src_key, src_path->slots[cur_level]);
2063                 }
2064                 /* Content mismatch, something went wrong */
2065                 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2066                         ret = -ENOENT;
2067                         goto out;
2068                 }
2069                 cur_level--;
2070         }
2071
2072         /*
2073          * Now both @dst_path and @src_path have been populated, record the tree
2074          * blocks for qgroup accounting.
2075          */
2076         ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2077                         nodesize, GFP_NOFS);
2078         if (ret < 0)
2079                 goto out;
2080         ret = btrfs_qgroup_trace_extent(trans,
2081                         dst_path->nodes[dst_level]->start,
2082                         nodesize, GFP_NOFS);
2083         if (ret < 0)
2084                 goto out;
2085
2086         /* Record leaf file extents */
2087         if (dst_level == 0 && trace_leaf) {
2088                 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2089                 if (ret < 0)
2090                         goto out;
2091                 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2092         }
2093 out:
2094         btrfs_free_path(src_path);
2095         return ret;
2096 }
2097
2098 /*
2099  * Helper function to do recursive generation-aware depth-first search, to
2100  * locate all new tree blocks in a subtree of reloc tree.
2101  *
2102  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2103  *         reloc tree
2104  * L2         NN (a)
2105  *          /    \
2106  * L1    OO        NN (b)
2107  *      /  \      /  \
2108  * L0  OO  OO    OO  NN
2109  *               (c) (d)
2110  * If we pass:
2111  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2112  * @cur_level = 1
2113  * @root_level = 1
2114  *
2115  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2116  * above tree blocks along with their counter parts in file tree.
2117  * While during search, old tree blocks OO(c) will be skipped as tree block swap
2118  * won't affect OO(c).
2119  */
2120 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2121                                            struct extent_buffer *src_eb,
2122                                            struct btrfs_path *dst_path,
2123                                            int cur_level, int root_level,
2124                                            u64 last_snapshot, bool trace_leaf)
2125 {
2126         struct btrfs_fs_info *fs_info = trans->fs_info;
2127         struct extent_buffer *eb;
2128         bool need_cleanup = false;
2129         int ret = 0;
2130         int i;
2131
2132         /* Level sanity check */
2133         if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2134             root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2135             root_level < cur_level) {
2136                 btrfs_err_rl(fs_info,
2137                         "%s: bad levels, cur_level=%d root_level=%d",
2138                         __func__, cur_level, root_level);
2139                 return -EUCLEAN;
2140         }
2141
2142         /* Read the tree block if needed */
2143         if (dst_path->nodes[cur_level] == NULL) {
2144                 int parent_slot;
2145                 u64 child_gen;
2146
2147                 /*
2148                  * dst_path->nodes[root_level] must be initialized before
2149                  * calling this function.
2150                  */
2151                 if (cur_level == root_level) {
2152                         btrfs_err_rl(fs_info,
2153         "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2154                                 __func__, root_level, root_level, cur_level);
2155                         return -EUCLEAN;
2156                 }
2157
2158                 /*
2159                  * We need to get child blockptr/gen from parent before we can
2160                  * read it.
2161                   */
2162                 eb = dst_path->nodes[cur_level + 1];
2163                 parent_slot = dst_path->slots[cur_level + 1];
2164                 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2165
2166                 /* This node is old, no need to trace */
2167                 if (child_gen < last_snapshot)
2168                         goto out;
2169
2170                 eb = btrfs_read_node_slot(eb, parent_slot);
2171                 if (IS_ERR(eb)) {
2172                         ret = PTR_ERR(eb);
2173                         goto out;
2174                 }
2175
2176                 dst_path->nodes[cur_level] = eb;
2177                 dst_path->slots[cur_level] = 0;
2178
2179                 btrfs_tree_read_lock(eb);
2180                 dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2181                 need_cleanup = true;
2182         }
2183
2184         /* Now record this tree block and its counter part for qgroups */
2185         ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2186                                        root_level, trace_leaf);
2187         if (ret < 0)
2188                 goto cleanup;
2189
2190         eb = dst_path->nodes[cur_level];
2191
2192         if (cur_level > 0) {
2193                 /* Iterate all child tree blocks */
2194                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2195                         /* Skip old tree blocks as they won't be swapped */
2196                         if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2197                                 continue;
2198                         dst_path->slots[cur_level] = i;
2199
2200                         /* Recursive call (at most 7 times) */
2201                         ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2202                                         dst_path, cur_level - 1, root_level,
2203                                         last_snapshot, trace_leaf);
2204                         if (ret < 0)
2205                                 goto cleanup;
2206                 }
2207         }
2208
2209 cleanup:
2210         if (need_cleanup) {
2211                 /* Clean up */
2212                 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2213                                      dst_path->locks[cur_level]);
2214                 free_extent_buffer(dst_path->nodes[cur_level]);
2215                 dst_path->nodes[cur_level] = NULL;
2216                 dst_path->slots[cur_level] = 0;
2217                 dst_path->locks[cur_level] = 0;
2218         }
2219 out:
2220         return ret;
2221 }
2222
2223 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2224                                 struct extent_buffer *src_eb,
2225                                 struct extent_buffer *dst_eb,
2226                                 u64 last_snapshot, bool trace_leaf)
2227 {
2228         struct btrfs_fs_info *fs_info = trans->fs_info;
2229         struct btrfs_path *dst_path = NULL;
2230         int level;
2231         int ret;
2232
2233         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2234                 return 0;
2235
2236         /* Wrong parameter order */
2237         if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2238                 btrfs_err_rl(fs_info,
2239                 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2240                              btrfs_header_generation(src_eb),
2241                              btrfs_header_generation(dst_eb));
2242                 return -EUCLEAN;
2243         }
2244
2245         if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2246                 ret = -EIO;
2247                 goto out;
2248         }
2249
2250         level = btrfs_header_level(dst_eb);
2251         dst_path = btrfs_alloc_path();
2252         if (!dst_path) {
2253                 ret = -ENOMEM;
2254                 goto out;
2255         }
2256         /* For dst_path */
2257         atomic_inc(&dst_eb->refs);
2258         dst_path->nodes[level] = dst_eb;
2259         dst_path->slots[level] = 0;
2260         dst_path->locks[level] = 0;
2261
2262         /* Do the generation aware breadth-first search */
2263         ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2264                                               level, last_snapshot, trace_leaf);
2265         if (ret < 0)
2266                 goto out;
2267         ret = 0;
2268
2269 out:
2270         btrfs_free_path(dst_path);
2271         if (ret < 0)
2272                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2273         return ret;
2274 }
2275
2276 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2277                                struct extent_buffer *root_eb,
2278                                u64 root_gen, int root_level)
2279 {
2280         struct btrfs_fs_info *fs_info = trans->fs_info;
2281         int ret = 0;
2282         int level;
2283         struct extent_buffer *eb = root_eb;
2284         struct btrfs_path *path = NULL;
2285
2286         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2287         BUG_ON(root_eb == NULL);
2288
2289         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2290                 return 0;
2291
2292         if (!extent_buffer_uptodate(root_eb)) {
2293                 ret = btrfs_read_extent_buffer(root_eb, root_gen, root_level, NULL);
2294                 if (ret)
2295                         goto out;
2296         }
2297
2298         if (root_level == 0) {
2299                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2300                 goto out;
2301         }
2302
2303         path = btrfs_alloc_path();
2304         if (!path)
2305                 return -ENOMEM;
2306
2307         /*
2308          * Walk down the tree.  Missing extent blocks are filled in as
2309          * we go. Metadata is accounted every time we read a new
2310          * extent block.
2311          *
2312          * When we reach a leaf, we account for file extent items in it,
2313          * walk back up the tree (adjusting slot pointers as we go)
2314          * and restart the search process.
2315          */
2316         atomic_inc(&root_eb->refs);     /* For path */
2317         path->nodes[root_level] = root_eb;
2318         path->slots[root_level] = 0;
2319         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2320 walk_down:
2321         level = root_level;
2322         while (level >= 0) {
2323                 if (path->nodes[level] == NULL) {
2324                         int parent_slot;
2325                         u64 child_bytenr;
2326
2327                         /*
2328                          * We need to get child blockptr from parent before we
2329                          * can read it.
2330                           */
2331                         eb = path->nodes[level + 1];
2332                         parent_slot = path->slots[level + 1];
2333                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2334
2335                         eb = btrfs_read_node_slot(eb, parent_slot);
2336                         if (IS_ERR(eb)) {
2337                                 ret = PTR_ERR(eb);
2338                                 goto out;
2339                         }
2340
2341                         path->nodes[level] = eb;
2342                         path->slots[level] = 0;
2343
2344                         btrfs_tree_read_lock(eb);
2345                         path->locks[level] = BTRFS_READ_LOCK;
2346
2347                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2348                                                         fs_info->nodesize,
2349                                                         GFP_NOFS);
2350                         if (ret)
2351                                 goto out;
2352                 }
2353
2354                 if (level == 0) {
2355                         ret = btrfs_qgroup_trace_leaf_items(trans,
2356                                                             path->nodes[level]);
2357                         if (ret)
2358                                 goto out;
2359
2360                         /* Nonzero return here means we completed our search */
2361                         ret = adjust_slots_upwards(path, root_level);
2362                         if (ret)
2363                                 break;
2364
2365                         /* Restart search with new slots */
2366                         goto walk_down;
2367                 }
2368
2369                 level--;
2370         }
2371
2372         ret = 0;
2373 out:
2374         btrfs_free_path(path);
2375
2376         return ret;
2377 }
2378
2379 #define UPDATE_NEW      0
2380 #define UPDATE_OLD      1
2381 /*
2382  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2383  */
2384 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2385                                 struct ulist *roots, struct ulist *tmp,
2386                                 struct ulist *qgroups, u64 seq, int update_old)
2387 {
2388         struct ulist_node *unode;
2389         struct ulist_iterator uiter;
2390         struct ulist_node *tmp_unode;
2391         struct ulist_iterator tmp_uiter;
2392         struct btrfs_qgroup *qg;
2393         int ret = 0;
2394
2395         if (!roots)
2396                 return 0;
2397         ULIST_ITER_INIT(&uiter);
2398         while ((unode = ulist_next(roots, &uiter))) {
2399                 qg = find_qgroup_rb(fs_info, unode->val);
2400                 if (!qg)
2401                         continue;
2402
2403                 ulist_reinit(tmp);
2404                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2405                                 GFP_ATOMIC);
2406                 if (ret < 0)
2407                         return ret;
2408                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2409                 if (ret < 0)
2410                         return ret;
2411                 ULIST_ITER_INIT(&tmp_uiter);
2412                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2413                         struct btrfs_qgroup_list *glist;
2414
2415                         qg = unode_aux_to_qgroup(tmp_unode);
2416                         if (update_old)
2417                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2418                         else
2419                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2420                         list_for_each_entry(glist, &qg->groups, next_group) {
2421                                 ret = ulist_add(qgroups, glist->group->qgroupid,
2422                                                 qgroup_to_aux(glist->group),
2423                                                 GFP_ATOMIC);
2424                                 if (ret < 0)
2425                                         return ret;
2426                                 ret = ulist_add(tmp, glist->group->qgroupid,
2427                                                 qgroup_to_aux(glist->group),
2428                                                 GFP_ATOMIC);
2429                                 if (ret < 0)
2430                                         return ret;
2431                         }
2432                 }
2433         }
2434         return 0;
2435 }
2436
2437 /*
2438  * Update qgroup rfer/excl counters.
2439  * Rfer update is easy, codes can explain themselves.
2440  *
2441  * Excl update is tricky, the update is split into 2 parts.
2442  * Part 1: Possible exclusive <-> sharing detect:
2443  *      |       A       |       !A      |
2444  *  -------------------------------------
2445  *  B   |       *       |       -       |
2446  *  -------------------------------------
2447  *  !B  |       +       |       **      |
2448  *  -------------------------------------
2449  *
2450  * Conditions:
2451  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
2452  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
2453  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
2454  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
2455  *
2456  * Results:
2457  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
2458  * *: Definitely not changed.           **: Possible unchanged.
2459  *
2460  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2461  *
2462  * To make the logic clear, we first use condition A and B to split
2463  * combination into 4 results.
2464  *
2465  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2466  * only on variant maybe 0.
2467  *
2468  * Lastly, check result **, since there are 2 variants maybe 0, split them
2469  * again(2x2).
2470  * But this time we don't need to consider other things, the codes and logic
2471  * is easy to understand now.
2472  */
2473 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2474                                   struct ulist *qgroups,
2475                                   u64 nr_old_roots,
2476                                   u64 nr_new_roots,
2477                                   u64 num_bytes, u64 seq)
2478 {
2479         struct ulist_node *unode;
2480         struct ulist_iterator uiter;
2481         struct btrfs_qgroup *qg;
2482         u64 cur_new_count, cur_old_count;
2483
2484         ULIST_ITER_INIT(&uiter);
2485         while ((unode = ulist_next(qgroups, &uiter))) {
2486                 bool dirty = false;
2487
2488                 qg = unode_aux_to_qgroup(unode);
2489                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2490                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2491
2492                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2493                                              cur_new_count);
2494
2495                 /* Rfer update part */
2496                 if (cur_old_count == 0 && cur_new_count > 0) {
2497                         qg->rfer += num_bytes;
2498                         qg->rfer_cmpr += num_bytes;
2499                         dirty = true;
2500                 }
2501                 if (cur_old_count > 0 && cur_new_count == 0) {
2502                         qg->rfer -= num_bytes;
2503                         qg->rfer_cmpr -= num_bytes;
2504                         dirty = true;
2505                 }
2506
2507                 /* Excl update part */
2508                 /* Exclusive/none -> shared case */
2509                 if (cur_old_count == nr_old_roots &&
2510                     cur_new_count < nr_new_roots) {
2511                         /* Exclusive -> shared */
2512                         if (cur_old_count != 0) {
2513                                 qg->excl -= num_bytes;
2514                                 qg->excl_cmpr -= num_bytes;
2515                                 dirty = true;
2516                         }
2517                 }
2518
2519                 /* Shared -> exclusive/none case */
2520                 if (cur_old_count < nr_old_roots &&
2521                     cur_new_count == nr_new_roots) {
2522                         /* Shared->exclusive */
2523                         if (cur_new_count != 0) {
2524                                 qg->excl += num_bytes;
2525                                 qg->excl_cmpr += num_bytes;
2526                                 dirty = true;
2527                         }
2528                 }
2529
2530                 /* Exclusive/none -> exclusive/none case */
2531                 if (cur_old_count == nr_old_roots &&
2532                     cur_new_count == nr_new_roots) {
2533                         if (cur_old_count == 0) {
2534                                 /* None -> exclusive/none */
2535
2536                                 if (cur_new_count != 0) {
2537                                         /* None -> exclusive */
2538                                         qg->excl += num_bytes;
2539                                         qg->excl_cmpr += num_bytes;
2540                                         dirty = true;
2541                                 }
2542                                 /* None -> none, nothing changed */
2543                         } else {
2544                                 /* Exclusive -> exclusive/none */
2545
2546                                 if (cur_new_count == 0) {
2547                                         /* Exclusive -> none */
2548                                         qg->excl -= num_bytes;
2549                                         qg->excl_cmpr -= num_bytes;
2550                                         dirty = true;
2551                                 }
2552                                 /* Exclusive -> exclusive, nothing changed */
2553                         }
2554                 }
2555
2556                 if (dirty)
2557                         qgroup_dirty(fs_info, qg);
2558         }
2559         return 0;
2560 }
2561
2562 /*
2563  * Check if the @roots potentially is a list of fs tree roots
2564  *
2565  * Return 0 for definitely not a fs/subvol tree roots ulist
2566  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2567  *          one as well)
2568  */
2569 static int maybe_fs_roots(struct ulist *roots)
2570 {
2571         struct ulist_node *unode;
2572         struct ulist_iterator uiter;
2573
2574         /* Empty one, still possible for fs roots */
2575         if (!roots || roots->nnodes == 0)
2576                 return 1;
2577
2578         ULIST_ITER_INIT(&uiter);
2579         unode = ulist_next(roots, &uiter);
2580         if (!unode)
2581                 return 1;
2582
2583         /*
2584          * If it contains fs tree roots, then it must belong to fs/subvol
2585          * trees.
2586          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2587          */
2588         return is_fstree(unode->val);
2589 }
2590
2591 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2592                                 u64 num_bytes, struct ulist *old_roots,
2593                                 struct ulist *new_roots)
2594 {
2595         struct btrfs_fs_info *fs_info = trans->fs_info;
2596         struct ulist *qgroups = NULL;
2597         struct ulist *tmp = NULL;
2598         u64 seq;
2599         u64 nr_new_roots = 0;
2600         u64 nr_old_roots = 0;
2601         int ret = 0;
2602
2603         /*
2604          * If quotas get disabled meanwhile, the resources need to be freed and
2605          * we can't just exit here.
2606          */
2607         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2608                 goto out_free;
2609
2610         if (new_roots) {
2611                 if (!maybe_fs_roots(new_roots))
2612                         goto out_free;
2613                 nr_new_roots = new_roots->nnodes;
2614         }
2615         if (old_roots) {
2616                 if (!maybe_fs_roots(old_roots))
2617                         goto out_free;
2618                 nr_old_roots = old_roots->nnodes;
2619         }
2620
2621         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2622         if (nr_old_roots == 0 && nr_new_roots == 0)
2623                 goto out_free;
2624
2625         BUG_ON(!fs_info->quota_root);
2626
2627         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2628                                         num_bytes, nr_old_roots, nr_new_roots);
2629
2630         qgroups = ulist_alloc(GFP_NOFS);
2631         if (!qgroups) {
2632                 ret = -ENOMEM;
2633                 goto out_free;
2634         }
2635         tmp = ulist_alloc(GFP_NOFS);
2636         if (!tmp) {
2637                 ret = -ENOMEM;
2638                 goto out_free;
2639         }
2640
2641         mutex_lock(&fs_info->qgroup_rescan_lock);
2642         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2643                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2644                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2645                         ret = 0;
2646                         goto out_free;
2647                 }
2648         }
2649         mutex_unlock(&fs_info->qgroup_rescan_lock);
2650
2651         spin_lock(&fs_info->qgroup_lock);
2652         seq = fs_info->qgroup_seq;
2653
2654         /* Update old refcnts using old_roots */
2655         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2656                                    UPDATE_OLD);
2657         if (ret < 0)
2658                 goto out;
2659
2660         /* Update new refcnts using new_roots */
2661         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2662                                    UPDATE_NEW);
2663         if (ret < 0)
2664                 goto out;
2665
2666         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2667                                num_bytes, seq);
2668
2669         /*
2670          * Bump qgroup_seq to avoid seq overlap
2671          */
2672         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2673 out:
2674         spin_unlock(&fs_info->qgroup_lock);
2675 out_free:
2676         ulist_free(tmp);
2677         ulist_free(qgroups);
2678         ulist_free(old_roots);
2679         ulist_free(new_roots);
2680         return ret;
2681 }
2682
2683 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2684 {
2685         struct btrfs_fs_info *fs_info = trans->fs_info;
2686         struct btrfs_qgroup_extent_record *record;
2687         struct btrfs_delayed_ref_root *delayed_refs;
2688         struct ulist *new_roots = NULL;
2689         struct rb_node *node;
2690         u64 num_dirty_extents = 0;
2691         u64 qgroup_to_skip;
2692         int ret = 0;
2693
2694         delayed_refs = &trans->transaction->delayed_refs;
2695         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2696         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2697                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2698                                   node);
2699
2700                 num_dirty_extents++;
2701                 trace_btrfs_qgroup_account_extents(fs_info, record);
2702
2703                 if (!ret) {
2704                         /*
2705                          * Old roots should be searched when inserting qgroup
2706                          * extent record
2707                          */
2708                         if (WARN_ON(!record->old_roots)) {
2709                                 /* Search commit root to find old_roots */
2710                                 ret = btrfs_find_all_roots(NULL, fs_info,
2711                                                 record->bytenr, 0,
2712                                                 &record->old_roots, false);
2713                                 if (ret < 0)
2714                                         goto cleanup;
2715                         }
2716
2717                         /* Free the reserved data space */
2718                         btrfs_qgroup_free_refroot(fs_info,
2719                                         record->data_rsv_refroot,
2720                                         record->data_rsv,
2721                                         BTRFS_QGROUP_RSV_DATA);
2722                         /*
2723                          * Use BTRFS_SEQ_LAST as time_seq to do special search,
2724                          * which doesn't lock tree or delayed_refs and search
2725                          * current root. It's safe inside commit_transaction().
2726                          */
2727                         ret = btrfs_find_all_roots(trans, fs_info,
2728                            record->bytenr, BTRFS_SEQ_LAST, &new_roots, false);
2729                         if (ret < 0)
2730                                 goto cleanup;
2731                         if (qgroup_to_skip) {
2732                                 ulist_del(new_roots, qgroup_to_skip, 0);
2733                                 ulist_del(record->old_roots, qgroup_to_skip,
2734                                           0);
2735                         }
2736                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2737                                                           record->num_bytes,
2738                                                           record->old_roots,
2739                                                           new_roots);
2740                         record->old_roots = NULL;
2741                         new_roots = NULL;
2742                 }
2743 cleanup:
2744                 ulist_free(record->old_roots);
2745                 ulist_free(new_roots);
2746                 new_roots = NULL;
2747                 rb_erase(node, &delayed_refs->dirty_extent_root);
2748                 kfree(record);
2749
2750         }
2751         trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2752                                        num_dirty_extents);
2753         return ret;
2754 }
2755
2756 /*
2757  * called from commit_transaction. Writes all changed qgroups to disk.
2758  */
2759 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2760 {
2761         struct btrfs_fs_info *fs_info = trans->fs_info;
2762         int ret = 0;
2763
2764         if (!fs_info->quota_root)
2765                 return ret;
2766
2767         spin_lock(&fs_info->qgroup_lock);
2768         while (!list_empty(&fs_info->dirty_qgroups)) {
2769                 struct btrfs_qgroup *qgroup;
2770                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2771                                           struct btrfs_qgroup, dirty);
2772                 list_del_init(&qgroup->dirty);
2773                 spin_unlock(&fs_info->qgroup_lock);
2774                 ret = update_qgroup_info_item(trans, qgroup);
2775                 if (ret)
2776                         fs_info->qgroup_flags |=
2777                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2778                 ret = update_qgroup_limit_item(trans, qgroup);
2779                 if (ret)
2780                         fs_info->qgroup_flags |=
2781                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2782                 spin_lock(&fs_info->qgroup_lock);
2783         }
2784         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2785                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2786         else
2787                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2788         spin_unlock(&fs_info->qgroup_lock);
2789
2790         ret = update_qgroup_status_item(trans);
2791         if (ret)
2792                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2793
2794         return ret;
2795 }
2796
2797 /*
2798  * Copy the accounting information between qgroups. This is necessary
2799  * when a snapshot or a subvolume is created. Throwing an error will
2800  * cause a transaction abort so we take extra care here to only error
2801  * when a readonly fs is a reasonable outcome.
2802  */
2803 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2804                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2805 {
2806         int ret = 0;
2807         int i;
2808         u64 *i_qgroups;
2809         bool committing = false;
2810         struct btrfs_fs_info *fs_info = trans->fs_info;
2811         struct btrfs_root *quota_root;
2812         struct btrfs_qgroup *srcgroup;
2813         struct btrfs_qgroup *dstgroup;
2814         bool need_rescan = false;
2815         u32 level_size = 0;
2816         u64 nums;
2817
2818         /*
2819          * There are only two callers of this function.
2820          *
2821          * One in create_subvol() in the ioctl context, which needs to hold
2822          * the qgroup_ioctl_lock.
2823          *
2824          * The other one in create_pending_snapshot() where no other qgroup
2825          * code can modify the fs as they all need to either start a new trans
2826          * or hold a trans handler, thus we don't need to hold
2827          * qgroup_ioctl_lock.
2828          * This would avoid long and complex lock chain and make lockdep happy.
2829          */
2830         spin_lock(&fs_info->trans_lock);
2831         if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2832                 committing = true;
2833         spin_unlock(&fs_info->trans_lock);
2834
2835         if (!committing)
2836                 mutex_lock(&fs_info->qgroup_ioctl_lock);
2837         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2838                 goto out;
2839
2840         quota_root = fs_info->quota_root;
2841         if (!quota_root) {
2842                 ret = -EINVAL;
2843                 goto out;
2844         }
2845
2846         if (inherit) {
2847                 i_qgroups = (u64 *)(inherit + 1);
2848                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2849                        2 * inherit->num_excl_copies;
2850                 for (i = 0; i < nums; ++i) {
2851                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2852
2853                         /*
2854                          * Zero out invalid groups so we can ignore
2855                          * them later.
2856                          */
2857                         if (!srcgroup ||
2858                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2859                                 *i_qgroups = 0ULL;
2860
2861                         ++i_qgroups;
2862                 }
2863         }
2864
2865         /*
2866          * create a tracking group for the subvol itself
2867          */
2868         ret = add_qgroup_item(trans, quota_root, objectid);
2869         if (ret)
2870                 goto out;
2871
2872         /*
2873          * add qgroup to all inherited groups
2874          */
2875         if (inherit) {
2876                 i_qgroups = (u64 *)(inherit + 1);
2877                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2878                         if (*i_qgroups == 0)
2879                                 continue;
2880                         ret = add_qgroup_relation_item(trans, objectid,
2881                                                        *i_qgroups);
2882                         if (ret && ret != -EEXIST)
2883                                 goto out;
2884                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2885                                                        objectid);
2886                         if (ret && ret != -EEXIST)
2887                                 goto out;
2888                 }
2889                 ret = 0;
2890         }
2891
2892
2893         spin_lock(&fs_info->qgroup_lock);
2894
2895         dstgroup = add_qgroup_rb(fs_info, objectid);
2896         if (IS_ERR(dstgroup)) {
2897                 ret = PTR_ERR(dstgroup);
2898                 goto unlock;
2899         }
2900
2901         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2902                 dstgroup->lim_flags = inherit->lim.flags;
2903                 dstgroup->max_rfer = inherit->lim.max_rfer;
2904                 dstgroup->max_excl = inherit->lim.max_excl;
2905                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2906                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2907
2908                 ret = update_qgroup_limit_item(trans, dstgroup);
2909                 if (ret) {
2910                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2911                         btrfs_info(fs_info,
2912                                    "unable to update quota limit for %llu",
2913                                    dstgroup->qgroupid);
2914                         goto unlock;
2915                 }
2916         }
2917
2918         if (srcid) {
2919                 srcgroup = find_qgroup_rb(fs_info, srcid);
2920                 if (!srcgroup)
2921                         goto unlock;
2922
2923                 /*
2924                  * We call inherit after we clone the root in order to make sure
2925                  * our counts don't go crazy, so at this point the only
2926                  * difference between the two roots should be the root node.
2927                  */
2928                 level_size = fs_info->nodesize;
2929                 dstgroup->rfer = srcgroup->rfer;
2930                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2931                 dstgroup->excl = level_size;
2932                 dstgroup->excl_cmpr = level_size;
2933                 srcgroup->excl = level_size;
2934                 srcgroup->excl_cmpr = level_size;
2935
2936                 /* inherit the limit info */
2937                 dstgroup->lim_flags = srcgroup->lim_flags;
2938                 dstgroup->max_rfer = srcgroup->max_rfer;
2939                 dstgroup->max_excl = srcgroup->max_excl;
2940                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2941                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2942
2943                 qgroup_dirty(fs_info, dstgroup);
2944                 qgroup_dirty(fs_info, srcgroup);
2945         }
2946
2947         if (!inherit)
2948                 goto unlock;
2949
2950         i_qgroups = (u64 *)(inherit + 1);
2951         for (i = 0; i < inherit->num_qgroups; ++i) {
2952                 if (*i_qgroups) {
2953                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2954                         if (ret)
2955                                 goto unlock;
2956                 }
2957                 ++i_qgroups;
2958
2959                 /*
2960                  * If we're doing a snapshot, and adding the snapshot to a new
2961                  * qgroup, the numbers are guaranteed to be incorrect.
2962                  */
2963                 if (srcid)
2964                         need_rescan = true;
2965         }
2966
2967         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2968                 struct btrfs_qgroup *src;
2969                 struct btrfs_qgroup *dst;
2970
2971                 if (!i_qgroups[0] || !i_qgroups[1])
2972                         continue;
2973
2974                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2975                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2976
2977                 if (!src || !dst) {
2978                         ret = -EINVAL;
2979                         goto unlock;
2980                 }
2981
2982                 dst->rfer = src->rfer - level_size;
2983                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2984
2985                 /* Manually tweaking numbers certainly needs a rescan */
2986                 need_rescan = true;
2987         }
2988         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2989                 struct btrfs_qgroup *src;
2990                 struct btrfs_qgroup *dst;
2991
2992                 if (!i_qgroups[0] || !i_qgroups[1])
2993                         continue;
2994
2995                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2996                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2997
2998                 if (!src || !dst) {
2999                         ret = -EINVAL;
3000                         goto unlock;
3001                 }
3002
3003                 dst->excl = src->excl + level_size;
3004                 dst->excl_cmpr = src->excl_cmpr + level_size;
3005                 need_rescan = true;
3006         }
3007
3008 unlock:
3009         spin_unlock(&fs_info->qgroup_lock);
3010         if (!ret)
3011                 ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3012 out:
3013         if (!committing)
3014                 mutex_unlock(&fs_info->qgroup_ioctl_lock);
3015         if (need_rescan)
3016                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3017         return ret;
3018 }
3019
3020 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3021 {
3022         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3023             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3024                 return false;
3025
3026         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3027             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3028                 return false;
3029
3030         return true;
3031 }
3032
3033 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3034                           enum btrfs_qgroup_rsv_type type)
3035 {
3036         struct btrfs_qgroup *qgroup;
3037         struct btrfs_fs_info *fs_info = root->fs_info;
3038         u64 ref_root = root->root_key.objectid;
3039         int ret = 0;
3040         struct ulist_node *unode;
3041         struct ulist_iterator uiter;
3042
3043         if (!is_fstree(ref_root))
3044                 return 0;
3045
3046         if (num_bytes == 0)
3047                 return 0;
3048
3049         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3050             capable(CAP_SYS_RESOURCE))
3051                 enforce = false;
3052
3053         spin_lock(&fs_info->qgroup_lock);
3054         if (!fs_info->quota_root)
3055                 goto out;
3056
3057         qgroup = find_qgroup_rb(fs_info, ref_root);
3058         if (!qgroup)
3059                 goto out;
3060
3061         /*
3062          * in a first step, we check all affected qgroups if any limits would
3063          * be exceeded
3064          */
3065         ulist_reinit(fs_info->qgroup_ulist);
3066         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3067                         qgroup_to_aux(qgroup), GFP_ATOMIC);
3068         if (ret < 0)
3069                 goto out;
3070         ULIST_ITER_INIT(&uiter);
3071         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3072                 struct btrfs_qgroup *qg;
3073                 struct btrfs_qgroup_list *glist;
3074
3075                 qg = unode_aux_to_qgroup(unode);
3076
3077                 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
3078                         ret = -EDQUOT;
3079                         goto out;
3080                 }
3081
3082                 list_for_each_entry(glist, &qg->groups, next_group) {
3083                         ret = ulist_add(fs_info->qgroup_ulist,
3084                                         glist->group->qgroupid,
3085                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3086                         if (ret < 0)
3087                                 goto out;
3088                 }
3089         }
3090         ret = 0;
3091         /*
3092          * no limits exceeded, now record the reservation into all qgroups
3093          */
3094         ULIST_ITER_INIT(&uiter);
3095         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3096                 struct btrfs_qgroup *qg;
3097
3098                 qg = unode_aux_to_qgroup(unode);
3099
3100                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
3101         }
3102
3103 out:
3104         spin_unlock(&fs_info->qgroup_lock);
3105         return ret;
3106 }
3107
3108 /*
3109  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
3110  * qgroup).
3111  *
3112  * Will handle all higher level qgroup too.
3113  *
3114  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3115  * This special case is only used for META_PERTRANS type.
3116  */
3117 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3118                                u64 ref_root, u64 num_bytes,
3119                                enum btrfs_qgroup_rsv_type type)
3120 {
3121         struct btrfs_qgroup *qgroup;
3122         struct ulist_node *unode;
3123         struct ulist_iterator uiter;
3124         int ret = 0;
3125
3126         if (!is_fstree(ref_root))
3127                 return;
3128
3129         if (num_bytes == 0)
3130                 return;
3131
3132         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3133                 WARN(1, "%s: Invalid type to free", __func__);
3134                 return;
3135         }
3136         spin_lock(&fs_info->qgroup_lock);
3137
3138         if (!fs_info->quota_root)
3139                 goto out;
3140
3141         qgroup = find_qgroup_rb(fs_info, ref_root);
3142         if (!qgroup)
3143                 goto out;
3144
3145         if (num_bytes == (u64)-1)
3146                 /*
3147                  * We're freeing all pertrans rsv, get reserved value from
3148                  * level 0 qgroup as real num_bytes to free.
3149                  */
3150                 num_bytes = qgroup->rsv.values[type];
3151
3152         ulist_reinit(fs_info->qgroup_ulist);
3153         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3154                         qgroup_to_aux(qgroup), GFP_ATOMIC);
3155         if (ret < 0)
3156                 goto out;
3157         ULIST_ITER_INIT(&uiter);
3158         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3159                 struct btrfs_qgroup *qg;
3160                 struct btrfs_qgroup_list *glist;
3161
3162                 qg = unode_aux_to_qgroup(unode);
3163
3164                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3165
3166                 list_for_each_entry(glist, &qg->groups, next_group) {
3167                         ret = ulist_add(fs_info->qgroup_ulist,
3168                                         glist->group->qgroupid,
3169                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3170                         if (ret < 0)
3171                                 goto out;
3172                 }
3173         }
3174
3175 out:
3176         spin_unlock(&fs_info->qgroup_lock);
3177 }
3178
3179 /*
3180  * Check if the leaf is the last leaf. Which means all node pointers
3181  * are at their last position.
3182  */
3183 static bool is_last_leaf(struct btrfs_path *path)
3184 {
3185         int i;
3186
3187         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3188                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3189                         return false;
3190         }
3191         return true;
3192 }
3193
3194 /*
3195  * returns < 0 on error, 0 when more leafs are to be scanned.
3196  * returns 1 when done.
3197  */
3198 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3199                               struct btrfs_path *path)
3200 {
3201         struct btrfs_fs_info *fs_info = trans->fs_info;
3202         struct btrfs_root *extent_root;
3203         struct btrfs_key found;
3204         struct extent_buffer *scratch_leaf = NULL;
3205         struct ulist *roots = NULL;
3206         u64 num_bytes;
3207         bool done;
3208         int slot;
3209         int ret;
3210
3211         mutex_lock(&fs_info->qgroup_rescan_lock);
3212         extent_root = btrfs_extent_root(fs_info,
3213                                 fs_info->qgroup_rescan_progress.objectid);
3214         ret = btrfs_search_slot_for_read(extent_root,
3215                                          &fs_info->qgroup_rescan_progress,
3216                                          path, 1, 0);
3217
3218         btrfs_debug(fs_info,
3219                 "current progress key (%llu %u %llu), search_slot ret %d",
3220                 fs_info->qgroup_rescan_progress.objectid,
3221                 fs_info->qgroup_rescan_progress.type,
3222                 fs_info->qgroup_rescan_progress.offset, ret);
3223
3224         if (ret) {
3225                 /*
3226                  * The rescan is about to end, we will not be scanning any
3227                  * further blocks. We cannot unset the RESCAN flag here, because
3228                  * we want to commit the transaction if everything went well.
3229                  * To make the live accounting work in this phase, we set our
3230                  * scan progress pointer such that every real extent objectid
3231                  * will be smaller.
3232                  */
3233                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3234                 btrfs_release_path(path);
3235                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3236                 return ret;
3237         }
3238         done = is_last_leaf(path);
3239
3240         btrfs_item_key_to_cpu(path->nodes[0], &found,
3241                               btrfs_header_nritems(path->nodes[0]) - 1);
3242         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3243
3244         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3245         if (!scratch_leaf) {
3246                 ret = -ENOMEM;
3247                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3248                 goto out;
3249         }
3250         slot = path->slots[0];
3251         btrfs_release_path(path);
3252         mutex_unlock(&fs_info->qgroup_rescan_lock);
3253
3254         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3255                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3256                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3257                     found.type != BTRFS_METADATA_ITEM_KEY)
3258                         continue;
3259                 if (found.type == BTRFS_METADATA_ITEM_KEY)
3260                         num_bytes = fs_info->nodesize;
3261                 else
3262                         num_bytes = found.offset;
3263
3264                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3265                                            &roots, false);
3266                 if (ret < 0)
3267                         goto out;
3268                 /* For rescan, just pass old_roots as NULL */
3269                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3270                                                   num_bytes, NULL, roots);
3271                 if (ret < 0)
3272                         goto out;
3273         }
3274 out:
3275         if (scratch_leaf)
3276                 free_extent_buffer(scratch_leaf);
3277
3278         if (done && !ret) {
3279                 ret = 1;
3280                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3281         }
3282         return ret;
3283 }
3284
3285 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3286 {
3287         return btrfs_fs_closing(fs_info) ||
3288                 test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
3289                 !test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
3290 }
3291
3292 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3293 {
3294         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3295                                                      qgroup_rescan_work);
3296         struct btrfs_path *path;
3297         struct btrfs_trans_handle *trans = NULL;
3298         int err = -ENOMEM;
3299         int ret = 0;
3300         bool stopped = false;
3301
3302         path = btrfs_alloc_path();
3303         if (!path)
3304                 goto out;
3305         /*
3306          * Rescan should only search for commit root, and any later difference
3307          * should be recorded by qgroup
3308          */
3309         path->search_commit_root = 1;
3310         path->skip_locking = 1;
3311
3312         err = 0;
3313         while (!err && !(stopped = rescan_should_stop(fs_info))) {
3314                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3315                 if (IS_ERR(trans)) {
3316                         err = PTR_ERR(trans);
3317                         break;
3318                 }
3319
3320                 err = qgroup_rescan_leaf(trans, path);
3321
3322                 if (err > 0)
3323                         btrfs_commit_transaction(trans);
3324                 else
3325                         btrfs_end_transaction(trans);
3326         }
3327
3328 out:
3329         btrfs_free_path(path);
3330
3331         mutex_lock(&fs_info->qgroup_rescan_lock);
3332         if (err > 0 &&
3333             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3334                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3335         } else if (err < 0 || stopped) {
3336                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3337         }
3338         mutex_unlock(&fs_info->qgroup_rescan_lock);
3339
3340         /*
3341          * only update status, since the previous part has already updated the
3342          * qgroup info.
3343          */
3344         trans = btrfs_start_transaction(fs_info->quota_root, 1);
3345         if (IS_ERR(trans)) {
3346                 err = PTR_ERR(trans);
3347                 trans = NULL;
3348                 btrfs_err(fs_info,
3349                           "fail to start transaction for status update: %d",
3350                           err);
3351         }
3352
3353         mutex_lock(&fs_info->qgroup_rescan_lock);
3354         if (!stopped)
3355                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3356         if (trans) {
3357                 ret = update_qgroup_status_item(trans);
3358                 if (ret < 0) {
3359                         err = ret;
3360                         btrfs_err(fs_info, "fail to update qgroup status: %d",
3361                                   err);
3362                 }
3363         }
3364         fs_info->qgroup_rescan_running = false;
3365         complete_all(&fs_info->qgroup_rescan_completion);
3366         mutex_unlock(&fs_info->qgroup_rescan_lock);
3367
3368         if (!trans)
3369                 return;
3370
3371         btrfs_end_transaction(trans);
3372
3373         if (stopped) {
3374                 btrfs_info(fs_info, "qgroup scan paused");
3375         } else if (err >= 0) {
3376                 btrfs_info(fs_info, "qgroup scan completed%s",
3377                         err > 0 ? " (inconsistency flag cleared)" : "");
3378         } else {
3379                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3380         }
3381 }
3382
3383 /*
3384  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3385  * memory required for the rescan context.
3386  */
3387 static int
3388 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3389                    int init_flags)
3390 {
3391         int ret = 0;
3392
3393         if (!init_flags) {
3394                 /* we're resuming qgroup rescan at mount time */
3395                 if (!(fs_info->qgroup_flags &
3396                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3397                         btrfs_warn(fs_info,
3398                         "qgroup rescan init failed, qgroup rescan is not queued");
3399                         ret = -EINVAL;
3400                 } else if (!(fs_info->qgroup_flags &
3401                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3402                         btrfs_warn(fs_info,
3403                         "qgroup rescan init failed, qgroup is not enabled");
3404                         ret = -EINVAL;
3405                 }
3406
3407                 if (ret)
3408                         return ret;
3409         }
3410
3411         mutex_lock(&fs_info->qgroup_rescan_lock);
3412
3413         if (init_flags) {
3414                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3415                         btrfs_warn(fs_info,
3416                                    "qgroup rescan is already in progress");
3417                         ret = -EINPROGRESS;
3418                 } else if (!(fs_info->qgroup_flags &
3419                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3420                         btrfs_warn(fs_info,
3421                         "qgroup rescan init failed, qgroup is not enabled");
3422                         ret = -EINVAL;
3423                 } else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3424                         /* Quota disable is in progress */
3425                         ret = -EBUSY;
3426                 }
3427
3428                 if (ret) {
3429                         mutex_unlock(&fs_info->qgroup_rescan_lock);
3430                         return ret;
3431                 }
3432                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3433         }
3434
3435         memset(&fs_info->qgroup_rescan_progress, 0,
3436                 sizeof(fs_info->qgroup_rescan_progress));
3437         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3438         init_completion(&fs_info->qgroup_rescan_completion);
3439         mutex_unlock(&fs_info->qgroup_rescan_lock);
3440
3441         btrfs_init_work(&fs_info->qgroup_rescan_work,
3442                         btrfs_qgroup_rescan_worker, NULL, NULL);
3443         return 0;
3444 }
3445
3446 static void
3447 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3448 {
3449         struct rb_node *n;
3450         struct btrfs_qgroup *qgroup;
3451
3452         spin_lock(&fs_info->qgroup_lock);
3453         /* clear all current qgroup tracking information */
3454         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3455                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3456                 qgroup->rfer = 0;
3457                 qgroup->rfer_cmpr = 0;
3458                 qgroup->excl = 0;
3459                 qgroup->excl_cmpr = 0;
3460                 qgroup_dirty(fs_info, qgroup);
3461         }
3462         spin_unlock(&fs_info->qgroup_lock);
3463 }
3464
3465 int
3466 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3467 {
3468         int ret = 0;
3469         struct btrfs_trans_handle *trans;
3470
3471         ret = qgroup_rescan_init(fs_info, 0, 1);
3472         if (ret)
3473                 return ret;
3474
3475         /*
3476          * We have set the rescan_progress to 0, which means no more
3477          * delayed refs will be accounted by btrfs_qgroup_account_ref.
3478          * However, btrfs_qgroup_account_ref may be right after its call
3479          * to btrfs_find_all_roots, in which case it would still do the
3480          * accounting.
3481          * To solve this, we're committing the transaction, which will
3482          * ensure we run all delayed refs and only after that, we are
3483          * going to clear all tracking information for a clean start.
3484          */
3485
3486         trans = btrfs_join_transaction(fs_info->fs_root);
3487         if (IS_ERR(trans)) {
3488                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3489                 return PTR_ERR(trans);
3490         }
3491         ret = btrfs_commit_transaction(trans);
3492         if (ret) {
3493                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3494                 return ret;
3495         }
3496
3497         qgroup_rescan_zero_tracking(fs_info);
3498
3499         mutex_lock(&fs_info->qgroup_rescan_lock);
3500         fs_info->qgroup_rescan_running = true;
3501         btrfs_queue_work(fs_info->qgroup_rescan_workers,
3502                          &fs_info->qgroup_rescan_work);
3503         mutex_unlock(&fs_info->qgroup_rescan_lock);
3504
3505         return 0;
3506 }
3507
3508 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3509                                      bool interruptible)
3510 {
3511         int running;
3512         int ret = 0;
3513
3514         mutex_lock(&fs_info->qgroup_rescan_lock);
3515         running = fs_info->qgroup_rescan_running;
3516         mutex_unlock(&fs_info->qgroup_rescan_lock);
3517
3518         if (!running)
3519                 return 0;
3520
3521         if (interruptible)
3522                 ret = wait_for_completion_interruptible(
3523                                         &fs_info->qgroup_rescan_completion);
3524         else
3525                 wait_for_completion(&fs_info->qgroup_rescan_completion);
3526
3527         return ret;
3528 }
3529
3530 /*
3531  * this is only called from open_ctree where we're still single threaded, thus
3532  * locking is omitted here.
3533  */
3534 void
3535 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3536 {
3537         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3538                 mutex_lock(&fs_info->qgroup_rescan_lock);
3539                 fs_info->qgroup_rescan_running = true;
3540                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3541                                  &fs_info->qgroup_rescan_work);
3542                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3543         }
3544 }
3545
3546 #define rbtree_iterate_from_safe(node, next, start)                             \
3547        for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3548
3549 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3550                                   struct extent_changeset *reserved, u64 start,
3551                                   u64 len)
3552 {
3553         struct rb_node *node;
3554         struct rb_node *next;
3555         struct ulist_node *entry;
3556         int ret = 0;
3557
3558         node = reserved->range_changed.root.rb_node;
3559         if (!node)
3560                 return 0;
3561         while (node) {
3562                 entry = rb_entry(node, struct ulist_node, rb_node);
3563                 if (entry->val < start)
3564                         node = node->rb_right;
3565                 else
3566                         node = node->rb_left;
3567         }
3568
3569         if (entry->val > start && rb_prev(&entry->rb_node))
3570                 entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3571                                  rb_node);
3572
3573         rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3574                 u64 entry_start;
3575                 u64 entry_end;
3576                 u64 entry_len;
3577                 int clear_ret;
3578
3579                 entry = rb_entry(node, struct ulist_node, rb_node);
3580                 entry_start = entry->val;
3581                 entry_end = entry->aux;
3582                 entry_len = entry_end - entry_start + 1;
3583
3584                 if (entry_start >= start + len)
3585                         break;
3586                 if (entry_start + entry_len <= start)
3587                         continue;
3588                 /*
3589                  * Now the entry is in [start, start + len), revert the
3590                  * EXTENT_QGROUP_RESERVED bit.
3591                  */
3592                 clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3593                                               entry_end, EXTENT_QGROUP_RESERVED);
3594                 if (!ret && clear_ret < 0)
3595                         ret = clear_ret;
3596
3597                 ulist_del(&reserved->range_changed, entry->val, entry->aux);
3598                 if (likely(reserved->bytes_changed >= entry_len)) {
3599                         reserved->bytes_changed -= entry_len;
3600                 } else {
3601                         WARN_ON(1);
3602                         reserved->bytes_changed = 0;
3603                 }
3604         }
3605
3606         return ret;
3607 }
3608
3609 /*
3610  * Try to free some space for qgroup.
3611  *
3612  * For qgroup, there are only 3 ways to free qgroup space:
3613  * - Flush nodatacow write
3614  *   Any nodatacow write will free its reserved data space at run_delalloc_range().
3615  *   In theory, we should only flush nodatacow inodes, but it's not yet
3616  *   possible, so we need to flush the whole root.
3617  *
3618  * - Wait for ordered extents
3619  *   When ordered extents are finished, their reserved metadata is finally
3620  *   converted to per_trans status, which can be freed by later commit
3621  *   transaction.
3622  *
3623  * - Commit transaction
3624  *   This would free the meta_per_trans space.
3625  *   In theory this shouldn't provide much space, but any more qgroup space
3626  *   is needed.
3627  */
3628 static int try_flush_qgroup(struct btrfs_root *root)
3629 {
3630         struct btrfs_trans_handle *trans;
3631         int ret;
3632
3633         /* Can't hold an open transaction or we run the risk of deadlocking. */
3634         ASSERT(current->journal_info == NULL);
3635         if (WARN_ON(current->journal_info))
3636                 return 0;
3637
3638         /*
3639          * We don't want to run flush again and again, so if there is a running
3640          * one, we won't try to start a new flush, but exit directly.
3641          */
3642         if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3643                 wait_event(root->qgroup_flush_wait,
3644                         !test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3645                 return 0;
3646         }
3647
3648         ret = btrfs_start_delalloc_snapshot(root, true);
3649         if (ret < 0)
3650                 goto out;
3651         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3652
3653         trans = btrfs_join_transaction(root);
3654         if (IS_ERR(trans)) {
3655                 ret = PTR_ERR(trans);
3656                 goto out;
3657         }
3658
3659         ret = btrfs_commit_transaction(trans);
3660 out:
3661         clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3662         wake_up(&root->qgroup_flush_wait);
3663         return ret;
3664 }
3665
3666 static int qgroup_reserve_data(struct btrfs_inode *inode,
3667                         struct extent_changeset **reserved_ret, u64 start,
3668                         u64 len)
3669 {
3670         struct btrfs_root *root = inode->root;
3671         struct extent_changeset *reserved;
3672         bool new_reserved = false;
3673         u64 orig_reserved;
3674         u64 to_reserve;
3675         int ret;
3676
3677         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3678             !is_fstree(root->root_key.objectid) || len == 0)
3679                 return 0;
3680
3681         /* @reserved parameter is mandatory for qgroup */
3682         if (WARN_ON(!reserved_ret))
3683                 return -EINVAL;
3684         if (!*reserved_ret) {
3685                 new_reserved = true;
3686                 *reserved_ret = extent_changeset_alloc();
3687                 if (!*reserved_ret)
3688                         return -ENOMEM;
3689         }
3690         reserved = *reserved_ret;
3691         /* Record already reserved space */
3692         orig_reserved = reserved->bytes_changed;
3693         ret = set_record_extent_bits(&inode->io_tree, start,
3694                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3695
3696         /* Newly reserved space */
3697         to_reserve = reserved->bytes_changed - orig_reserved;
3698         trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3699                                         to_reserve, QGROUP_RESERVE);
3700         if (ret < 0)
3701                 goto out;
3702         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3703         if (ret < 0)
3704                 goto cleanup;
3705
3706         return ret;
3707
3708 cleanup:
3709         qgroup_unreserve_range(inode, reserved, start, len);
3710 out:
3711         if (new_reserved) {
3712                 extent_changeset_free(reserved);
3713                 *reserved_ret = NULL;
3714         }
3715         return ret;
3716 }
3717
3718 /*
3719  * Reserve qgroup space for range [start, start + len).
3720  *
3721  * This function will either reserve space from related qgroups or do nothing
3722  * if the range is already reserved.
3723  *
3724  * Return 0 for successful reservation
3725  * Return <0 for error (including -EQUOT)
3726  *
3727  * NOTE: This function may sleep for memory allocation, dirty page flushing and
3728  *       commit transaction. So caller should not hold any dirty page locked.
3729  */
3730 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3731                         struct extent_changeset **reserved_ret, u64 start,
3732                         u64 len)
3733 {
3734         int ret;
3735
3736         ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3737         if (ret <= 0 && ret != -EDQUOT)
3738                 return ret;
3739
3740         ret = try_flush_qgroup(inode->root);
3741         if (ret < 0)
3742                 return ret;
3743         return qgroup_reserve_data(inode, reserved_ret, start, len);
3744 }
3745
3746 /* Free ranges specified by @reserved, normally in error path */
3747 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3748                         struct extent_changeset *reserved, u64 start, u64 len)
3749 {
3750         struct btrfs_root *root = inode->root;
3751         struct ulist_node *unode;
3752         struct ulist_iterator uiter;
3753         struct extent_changeset changeset;
3754         int freed = 0;
3755         int ret;
3756
3757         extent_changeset_init(&changeset);
3758         len = round_up(start + len, root->fs_info->sectorsize);
3759         start = round_down(start, root->fs_info->sectorsize);
3760
3761         ULIST_ITER_INIT(&uiter);
3762         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3763                 u64 range_start = unode->val;
3764                 /* unode->aux is the inclusive end */
3765                 u64 range_len = unode->aux - range_start + 1;
3766                 u64 free_start;
3767                 u64 free_len;
3768
3769                 extent_changeset_release(&changeset);
3770
3771                 /* Only free range in range [start, start + len) */
3772                 if (range_start >= start + len ||
3773                     range_start + range_len <= start)
3774                         continue;
3775                 free_start = max(range_start, start);
3776                 free_len = min(start + len, range_start + range_len) -
3777                            free_start;
3778                 /*
3779                  * TODO: To also modify reserved->ranges_reserved to reflect
3780                  * the modification.
3781                  *
3782                  * However as long as we free qgroup reserved according to
3783                  * EXTENT_QGROUP_RESERVED, we won't double free.
3784                  * So not need to rush.
3785                  */
3786                 ret = clear_record_extent_bits(&inode->io_tree, free_start,
3787                                 free_start + free_len - 1,
3788                                 EXTENT_QGROUP_RESERVED, &changeset);
3789                 if (ret < 0)
3790                         goto out;
3791                 freed += changeset.bytes_changed;
3792         }
3793         btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3794                                   BTRFS_QGROUP_RSV_DATA);
3795         ret = freed;
3796 out:
3797         extent_changeset_release(&changeset);
3798         return ret;
3799 }
3800
3801 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3802                         struct extent_changeset *reserved, u64 start, u64 len,
3803                         int free)
3804 {
3805         struct extent_changeset changeset;
3806         int trace_op = QGROUP_RELEASE;
3807         int ret;
3808
3809         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3810                 return 0;
3811
3812         /* In release case, we shouldn't have @reserved */
3813         WARN_ON(!free && reserved);
3814         if (free && reserved)
3815                 return qgroup_free_reserved_data(inode, reserved, start, len);
3816         extent_changeset_init(&changeset);
3817         ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3818                                        EXTENT_QGROUP_RESERVED, &changeset);
3819         if (ret < 0)
3820                 goto out;
3821
3822         if (free)
3823                 trace_op = QGROUP_FREE;
3824         trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3825                                         changeset.bytes_changed, trace_op);
3826         if (free)
3827                 btrfs_qgroup_free_refroot(inode->root->fs_info,
3828                                 inode->root->root_key.objectid,
3829                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3830         ret = changeset.bytes_changed;
3831 out:
3832         extent_changeset_release(&changeset);
3833         return ret;
3834 }
3835
3836 /*
3837  * Free a reserved space range from io_tree and related qgroups
3838  *
3839  * Should be called when a range of pages get invalidated before reaching disk.
3840  * Or for error cleanup case.
3841  * if @reserved is given, only reserved range in [@start, @start + @len) will
3842  * be freed.
3843  *
3844  * For data written to disk, use btrfs_qgroup_release_data().
3845  *
3846  * NOTE: This function may sleep for memory allocation.
3847  */
3848 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3849                         struct extent_changeset *reserved, u64 start, u64 len)
3850 {
3851         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3852 }
3853
3854 /*
3855  * Release a reserved space range from io_tree only.
3856  *
3857  * Should be called when a range of pages get written to disk and corresponding
3858  * FILE_EXTENT is inserted into corresponding root.
3859  *
3860  * Since new qgroup accounting framework will only update qgroup numbers at
3861  * commit_transaction() time, its reserved space shouldn't be freed from
3862  * related qgroups.
3863  *
3864  * But we should release the range from io_tree, to allow further write to be
3865  * COWed.
3866  *
3867  * NOTE: This function may sleep for memory allocation.
3868  */
3869 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3870 {
3871         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3872 }
3873
3874 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3875                               enum btrfs_qgroup_rsv_type type)
3876 {
3877         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3878             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3879                 return;
3880         if (num_bytes == 0)
3881                 return;
3882
3883         spin_lock(&root->qgroup_meta_rsv_lock);
3884         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3885                 root->qgroup_meta_rsv_prealloc += num_bytes;
3886         else
3887                 root->qgroup_meta_rsv_pertrans += num_bytes;
3888         spin_unlock(&root->qgroup_meta_rsv_lock);
3889 }
3890
3891 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3892                              enum btrfs_qgroup_rsv_type type)
3893 {
3894         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3895             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3896                 return 0;
3897         if (num_bytes == 0)
3898                 return 0;
3899
3900         spin_lock(&root->qgroup_meta_rsv_lock);
3901         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3902                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3903                                   num_bytes);
3904                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3905         } else {
3906                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3907                                   num_bytes);
3908                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3909         }
3910         spin_unlock(&root->qgroup_meta_rsv_lock);
3911         return num_bytes;
3912 }
3913
3914 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3915                               enum btrfs_qgroup_rsv_type type, bool enforce)
3916 {
3917         struct btrfs_fs_info *fs_info = root->fs_info;
3918         int ret;
3919
3920         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3921             !is_fstree(root->root_key.objectid) || num_bytes == 0)
3922                 return 0;
3923
3924         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3925         trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3926         ret = qgroup_reserve(root, num_bytes, enforce, type);
3927         if (ret < 0)
3928                 return ret;
3929         /*
3930          * Record what we have reserved into root.
3931          *
3932          * To avoid quota disabled->enabled underflow.
3933          * In that case, we may try to free space we haven't reserved
3934          * (since quota was disabled), so record what we reserved into root.
3935          * And ensure later release won't underflow this number.
3936          */
3937         add_root_meta_rsv(root, num_bytes, type);
3938         return ret;
3939 }
3940
3941 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3942                                 enum btrfs_qgroup_rsv_type type, bool enforce,
3943                                 bool noflush)
3944 {
3945         int ret;
3946
3947         ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3948         if ((ret <= 0 && ret != -EDQUOT) || noflush)
3949                 return ret;
3950
3951         ret = try_flush_qgroup(root);
3952         if (ret < 0)
3953                 return ret;
3954         return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3955 }
3956
3957 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3958 {
3959         struct btrfs_fs_info *fs_info = root->fs_info;
3960
3961         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3962             !is_fstree(root->root_key.objectid))
3963                 return;
3964
3965         /* TODO: Update trace point to handle such free */
3966         trace_qgroup_meta_free_all_pertrans(root);
3967         /* Special value -1 means to free all reserved space */
3968         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3969                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3970 }
3971
3972 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3973                               enum btrfs_qgroup_rsv_type type)
3974 {
3975         struct btrfs_fs_info *fs_info = root->fs_info;
3976
3977         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3978             !is_fstree(root->root_key.objectid))
3979                 return;
3980
3981         /*
3982          * reservation for META_PREALLOC can happen before quota is enabled,
3983          * which can lead to underflow.
3984          * Here ensure we will only free what we really have reserved.
3985          */
3986         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3987         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3988         trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3989         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3990                                   num_bytes, type);
3991 }
3992
3993 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3994                                 int num_bytes)
3995 {
3996         struct btrfs_qgroup *qgroup;
3997         struct ulist_node *unode;
3998         struct ulist_iterator uiter;
3999         int ret = 0;
4000
4001         if (num_bytes == 0)
4002                 return;
4003         if (!fs_info->quota_root)
4004                 return;
4005
4006         spin_lock(&fs_info->qgroup_lock);
4007         qgroup = find_qgroup_rb(fs_info, ref_root);
4008         if (!qgroup)
4009                 goto out;
4010         ulist_reinit(fs_info->qgroup_ulist);
4011         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
4012                        qgroup_to_aux(qgroup), GFP_ATOMIC);
4013         if (ret < 0)
4014                 goto out;
4015         ULIST_ITER_INIT(&uiter);
4016         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
4017                 struct btrfs_qgroup *qg;
4018                 struct btrfs_qgroup_list *glist;
4019
4020                 qg = unode_aux_to_qgroup(unode);
4021
4022                 qgroup_rsv_release(fs_info, qg, num_bytes,
4023                                 BTRFS_QGROUP_RSV_META_PREALLOC);
4024                 qgroup_rsv_add(fs_info, qg, num_bytes,
4025                                 BTRFS_QGROUP_RSV_META_PERTRANS);
4026                 list_for_each_entry(glist, &qg->groups, next_group) {
4027                         ret = ulist_add(fs_info->qgroup_ulist,
4028                                         glist->group->qgroupid,
4029                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
4030                         if (ret < 0)
4031                                 goto out;
4032                 }
4033         }
4034 out:
4035         spin_unlock(&fs_info->qgroup_lock);
4036 }
4037
4038 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4039 {
4040         struct btrfs_fs_info *fs_info = root->fs_info;
4041
4042         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4043             !is_fstree(root->root_key.objectid))
4044                 return;
4045         /* Same as btrfs_qgroup_free_meta_prealloc() */
4046         num_bytes = sub_root_meta_rsv(root, num_bytes,
4047                                       BTRFS_QGROUP_RSV_META_PREALLOC);
4048         trace_qgroup_meta_convert(root, num_bytes);
4049         qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4050 }
4051
4052 /*
4053  * Check qgroup reserved space leaking, normally at destroy inode
4054  * time
4055  */
4056 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4057 {
4058         struct extent_changeset changeset;
4059         struct ulist_node *unode;
4060         struct ulist_iterator iter;
4061         int ret;
4062
4063         extent_changeset_init(&changeset);
4064         ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4065                         EXTENT_QGROUP_RESERVED, &changeset);
4066
4067         WARN_ON(ret < 0);
4068         if (WARN_ON(changeset.bytes_changed)) {
4069                 ULIST_ITER_INIT(&iter);
4070                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4071                         btrfs_warn(inode->root->fs_info,
4072                 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4073                                 btrfs_ino(inode), unode->val, unode->aux);
4074                 }
4075                 btrfs_qgroup_free_refroot(inode->root->fs_info,
4076                                 inode->root->root_key.objectid,
4077                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4078
4079         }
4080         extent_changeset_release(&changeset);
4081 }
4082
4083 void btrfs_qgroup_init_swapped_blocks(
4084         struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4085 {
4086         int i;
4087
4088         spin_lock_init(&swapped_blocks->lock);
4089         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4090                 swapped_blocks->blocks[i] = RB_ROOT;
4091         swapped_blocks->swapped = false;
4092 }
4093
4094 /*
4095  * Delete all swapped blocks record of @root.
4096  * Every record here means we skipped a full subtree scan for qgroup.
4097  *
4098  * Gets called when committing one transaction.
4099  */
4100 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4101 {
4102         struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4103         int i;
4104
4105         swapped_blocks = &root->swapped_blocks;
4106
4107         spin_lock(&swapped_blocks->lock);
4108         if (!swapped_blocks->swapped)
4109                 goto out;
4110         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4111                 struct rb_root *cur_root = &swapped_blocks->blocks[i];
4112                 struct btrfs_qgroup_swapped_block *entry;
4113                 struct btrfs_qgroup_swapped_block *next;
4114
4115                 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4116                                                      node)
4117                         kfree(entry);
4118                 swapped_blocks->blocks[i] = RB_ROOT;
4119         }
4120         swapped_blocks->swapped = false;
4121 out:
4122         spin_unlock(&swapped_blocks->lock);
4123 }
4124
4125 /*
4126  * Add subtree roots record into @subvol_root.
4127  *
4128  * @subvol_root:        tree root of the subvolume tree get swapped
4129  * @bg:                 block group under balance
4130  * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4131  * @reloc_parent/slot:  pointer to the subtree root in reloc tree
4132  *                      BOTH POINTERS ARE BEFORE TREE SWAP
4133  * @last_snapshot:      last snapshot generation of the subvolume tree
4134  */
4135 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4136                 struct btrfs_root *subvol_root,
4137                 struct btrfs_block_group *bg,
4138                 struct extent_buffer *subvol_parent, int subvol_slot,
4139                 struct extent_buffer *reloc_parent, int reloc_slot,
4140                 u64 last_snapshot)
4141 {
4142         struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4143         struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4144         struct btrfs_qgroup_swapped_block *block;
4145         struct rb_node **cur;
4146         struct rb_node *parent = NULL;
4147         int level = btrfs_header_level(subvol_parent) - 1;
4148         int ret = 0;
4149
4150         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4151                 return 0;
4152
4153         if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4154             btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4155                 btrfs_err_rl(fs_info,
4156                 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4157                         __func__,
4158                         btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4159                         btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4160                 return -EUCLEAN;
4161         }
4162
4163         block = kmalloc(sizeof(*block), GFP_NOFS);
4164         if (!block) {
4165                 ret = -ENOMEM;
4166                 goto out;
4167         }
4168
4169         /*
4170          * @reloc_parent/slot is still before swap, while @block is going to
4171          * record the bytenr after swap, so we do the swap here.
4172          */
4173         block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4174         block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4175                                                              reloc_slot);
4176         block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4177         block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4178                                                             subvol_slot);
4179         block->last_snapshot = last_snapshot;
4180         block->level = level;
4181
4182         /*
4183          * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4184          * no one else can modify tree blocks thus we qgroup will not change
4185          * no matter the value of trace_leaf.
4186          */
4187         if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4188                 block->trace_leaf = true;
4189         else
4190                 block->trace_leaf = false;
4191         btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4192
4193         /* Insert @block into @blocks */
4194         spin_lock(&blocks->lock);
4195         cur = &blocks->blocks[level].rb_node;
4196         while (*cur) {
4197                 struct btrfs_qgroup_swapped_block *entry;
4198
4199                 parent = *cur;
4200                 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4201                                  node);
4202
4203                 if (entry->subvol_bytenr < block->subvol_bytenr) {
4204                         cur = &(*cur)->rb_left;
4205                 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
4206                         cur = &(*cur)->rb_right;
4207                 } else {
4208                         if (entry->subvol_generation !=
4209                                         block->subvol_generation ||
4210                             entry->reloc_bytenr != block->reloc_bytenr ||
4211                             entry->reloc_generation !=
4212                                         block->reloc_generation) {
4213                                 /*
4214                                  * Duplicated but mismatch entry found.
4215                                  * Shouldn't happen.
4216                                  *
4217                                  * Marking qgroup inconsistent should be enough
4218                                  * for end users.
4219                                  */
4220                                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4221                                 ret = -EEXIST;
4222                         }
4223                         kfree(block);
4224                         goto out_unlock;
4225                 }
4226         }
4227         rb_link_node(&block->node, parent, cur);
4228         rb_insert_color(&block->node, &blocks->blocks[level]);
4229         blocks->swapped = true;
4230 out_unlock:
4231         spin_unlock(&blocks->lock);
4232 out:
4233         if (ret < 0)
4234                 fs_info->qgroup_flags |=
4235                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4236         return ret;
4237 }
4238
4239 /*
4240  * Check if the tree block is a subtree root, and if so do the needed
4241  * delayed subtree trace for qgroup.
4242  *
4243  * This is called during btrfs_cow_block().
4244  */
4245 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4246                                          struct btrfs_root *root,
4247                                          struct extent_buffer *subvol_eb)
4248 {
4249         struct btrfs_fs_info *fs_info = root->fs_info;
4250         struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4251         struct btrfs_qgroup_swapped_block *block;
4252         struct extent_buffer *reloc_eb = NULL;
4253         struct rb_node *node;
4254         bool found = false;
4255         bool swapped = false;
4256         int level = btrfs_header_level(subvol_eb);
4257         int ret = 0;
4258         int i;
4259
4260         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4261                 return 0;
4262         if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4263                 return 0;
4264
4265         spin_lock(&blocks->lock);
4266         if (!blocks->swapped) {
4267                 spin_unlock(&blocks->lock);
4268                 return 0;
4269         }
4270         node = blocks->blocks[level].rb_node;
4271
4272         while (node) {
4273                 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4274                 if (block->subvol_bytenr < subvol_eb->start) {
4275                         node = node->rb_left;
4276                 } else if (block->subvol_bytenr > subvol_eb->start) {
4277                         node = node->rb_right;
4278                 } else {
4279                         found = true;
4280                         break;
4281                 }
4282         }
4283         if (!found) {
4284                 spin_unlock(&blocks->lock);
4285                 goto out;
4286         }
4287         /* Found one, remove it from @blocks first and update blocks->swapped */
4288         rb_erase(&block->node, &blocks->blocks[level]);
4289         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4290                 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4291                         swapped = true;
4292                         break;
4293                 }
4294         }
4295         blocks->swapped = swapped;
4296         spin_unlock(&blocks->lock);
4297
4298         /* Read out reloc subtree root */
4299         reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, 0,
4300                                    block->reloc_generation, block->level,
4301                                    &block->first_key);
4302         if (IS_ERR(reloc_eb)) {
4303                 ret = PTR_ERR(reloc_eb);
4304                 reloc_eb = NULL;
4305                 goto free_out;
4306         }
4307         if (!extent_buffer_uptodate(reloc_eb)) {
4308                 ret = -EIO;
4309                 goto free_out;
4310         }
4311
4312         ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4313                         block->last_snapshot, block->trace_leaf);
4314 free_out:
4315         kfree(block);
4316         free_extent_buffer(reloc_eb);
4317 out:
4318         if (ret < 0) {
4319                 btrfs_err_rl(fs_info,
4320                              "failed to account subtree at bytenr %llu: %d",
4321                              subvol_eb->start, ret);
4322                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4323         }
4324         return ret;
4325 }
4326
4327 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4328 {
4329         struct btrfs_qgroup_extent_record *entry;
4330         struct btrfs_qgroup_extent_record *next;
4331         struct rb_root *root;
4332
4333         root = &trans->delayed_refs.dirty_extent_root;
4334         rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4335                 ulist_free(entry->old_roots);
4336                 kfree(entry);
4337         }
4338 }