fs/btrfs/qgroup.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2011 STRATO.  All rights reserved.
   4  */
   5
   6 #include <linux/sched.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/writeback.h>
   9 #include <linux/blkdev.h>
  10 #include <linux/rbtree.h>
  11 #include <linux/slab.h>
  12 #include <linux/workqueue.h>
  13 #include <linux/btrfs.h>
  14 #include <linux/sizes.h>
  15
  16 #include "ctree.h"
  17 #include "transaction.h"
  18 #include "disk-io.h"
  19 #include "locking.h"
  20 #include "ulist.h"
  21 #include "backref.h"
  22 #include "extent_io.h"
  23 #include "qgroup.h"
  24
  25
  26 /* TODO XXX FIXME
  27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  28  *  - reorganize keys
  29  *  - compressed
  30  *  - sync
  31  *  - copy also limits on subvol creation
  32  *  - limit
  33  *  - caches fuer ulists
  34  *  - performance benchmarks
  35  *  - check all ioctl parameters
  36  */
  37
  38 /*
  39  * Helpers to access qgroup reservation
  40  *
  41  * Callers should ensure the lock context and type are valid
  42  */
  43
  44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
  45 {
  46         u64 ret = 0;
  47         int i;
  48
  49         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
  50                 ret += qgroup->rsv.values[i];
  51
  52         return ret;
  53 }
  54
  55 #ifdef CONFIG_BTRFS_DEBUG
  56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
  57 {
  58         if (type == BTRFS_QGROUP_RSV_DATA)
  59                 return "data";
  60         if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
  61                 return "meta_pertrans";
  62         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
  63                 return "meta_prealloc";
  64         return NULL;
  65 }
  66 #endif
  67
  68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
  69                            struct btrfs_qgroup *qgroup, u64 num_bytes,
  70                            enum btrfs_qgroup_rsv_type type)
  71 {
  72         trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
  73         qgroup->rsv.values[type] += num_bytes;
  74 }
  75
  76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
  77                                struct btrfs_qgroup *qgroup, u64 num_bytes,
  78                                enum btrfs_qgroup_rsv_type type)
  79 {
  80         trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
  81         if (qgroup->rsv.values[type] >= num_bytes) {
  82                 qgroup->rsv.values[type] -= num_bytes;
  83                 return;
  84         }
  85 #ifdef CONFIG_BTRFS_DEBUG
  86         WARN_RATELIMIT(1,
  87                 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
  88                 qgroup->qgroupid, qgroup_rsv_type_str(type),
  89                 qgroup->rsv.values[type], num_bytes);
  90 #endif
  91         qgroup->rsv.values[type] = 0;
  92 }
  93
  94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
  95                                      struct btrfs_qgroup *dest,
  96                                      struct btrfs_qgroup *src)
  97 {
  98         int i;
  99
 100         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 101                 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
 102 }
 103
 104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
 105                                          struct btrfs_qgroup *dest,
 106                                           struct btrfs_qgroup *src)
 107 {
 108         int i;
 109
 110         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 111                 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
 112 }
 113
 114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
 115                                            int mod)
 116 {
 117         if (qg->old_refcnt < seq)
 118                 qg->old_refcnt = seq;
 119         qg->old_refcnt += mod;
 120 }
 121
 122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
 123                                            int mod)
 124 {
 125         if (qg->new_refcnt < seq)
 126                 qg->new_refcnt = seq;
 127         qg->new_refcnt += mod;
 128 }
 129
 130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
 131 {
 132         if (qg->old_refcnt < seq)
 133                 return 0;
 134         return qg->old_refcnt - seq;
 135 }
 136
 137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
 138 {
 139         if (qg->new_refcnt < seq)
 140                 return 0;
 141         return qg->new_refcnt - seq;
 142 }
 143
 144 /*
 145  * glue structure to represent the relations between qgroups.
 146  */
 147 struct btrfs_qgroup_list {
 148         struct list_head next_group;
 149         struct list_head next_member;
 150         struct btrfs_qgroup *group;
 151         struct btrfs_qgroup *member;
 152 };
 153
 154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
 155 {
 156         return (u64)(uintptr_t)qg;
 157 }
 158
 159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
 160 {
 161         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
 162 }
 163
 164 static int
 165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 166                    int init_flags);
 167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 168
 169 /* must be called with qgroup_ioctl_lock held */
 170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 171                                            u64 qgroupid)
 172 {
 173         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 174         struct btrfs_qgroup *qgroup;
 175
 176         while (n) {
 177                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 178                 if (qgroup->qgroupid < qgroupid)
 179                         n = n->rb_left;
 180                 else if (qgroup->qgroupid > qgroupid)
 181                         n = n->rb_right;
 182                 else
 183                         return qgroup;
 184         }
 185         return NULL;
 186 }
 187
 188 /* must be called with qgroup_lock held */
 189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 190                                           u64 qgroupid)
 191 {
 192         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 193         struct rb_node *parent = NULL;
 194         struct btrfs_qgroup *qgroup;
 195
 196         while (*p) {
 197                 parent = *p;
 198                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 199
 200                 if (qgroup->qgroupid < qgroupid)
 201                         p = &(*p)->rb_left;
 202                 else if (qgroup->qgroupid > qgroupid)
 203                         p = &(*p)->rb_right;
 204                 else
 205                         return qgroup;
 206         }
 207
 208         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 209         if (!qgroup)
 210                 return ERR_PTR(-ENOMEM);
 211
 212         qgroup->qgroupid = qgroupid;
 213         INIT_LIST_HEAD(&qgroup->groups);
 214         INIT_LIST_HEAD(&qgroup->members);
 215         INIT_LIST_HEAD(&qgroup->dirty);
 216
 217         rb_link_node(&qgroup->node, parent, p);
 218         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 219
 220         return qgroup;
 221 }
 222
 223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
 224 {
 225         struct btrfs_qgroup_list *list;
 226
 227         list_del(&qgroup->dirty);
 228         while (!list_empty(&qgroup->groups)) {
 229                 list = list_first_entry(&qgroup->groups,
 230                                         struct btrfs_qgroup_list, next_group);
 231                 list_del(&list->next_group);
 232                 list_del(&list->next_member);
 233                 kfree(list);
 234         }
 235
 236         while (!list_empty(&qgroup->members)) {
 237                 list = list_first_entry(&qgroup->members,
 238                                         struct btrfs_qgroup_list, next_member);
 239                 list_del(&list->next_group);
 240                 list_del(&list->next_member);
 241                 kfree(list);
 242         }
 243         kfree(qgroup);
 244 }
 245
 246 /* must be called with qgroup_lock held */
 247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 248 {
 249         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 250
 251         if (!qgroup)
 252                 return -ENOENT;
 253
 254         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 255         __del_qgroup_rb(qgroup);
 256         return 0;
 257 }
 258
 259 /* must be called with qgroup_lock held */
 260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 261                            u64 memberid, u64 parentid)
 262 {
 263         struct btrfs_qgroup *member;
 264         struct btrfs_qgroup *parent;
 265         struct btrfs_qgroup_list *list;
 266
 267         member = find_qgroup_rb(fs_info, memberid);
 268         parent = find_qgroup_rb(fs_info, parentid);
 269         if (!member || !parent)
 270                 return -ENOENT;
 271
 272         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 273         if (!list)
 274                 return -ENOMEM;
 275
 276         list->group = parent;
 277         list->member = member;
 278         list_add_tail(&list->next_group, &member->groups);
 279         list_add_tail(&list->next_member, &parent->members);
 280
 281         return 0;
 282 }
 283
 284 /* must be called with qgroup_lock held */
 285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 286                            u64 memberid, u64 parentid)
 287 {
 288         struct btrfs_qgroup *member;
 289         struct btrfs_qgroup *parent;
 290         struct btrfs_qgroup_list *list;
 291
 292         member = find_qgroup_rb(fs_info, memberid);
 293         parent = find_qgroup_rb(fs_info, parentid);
 294         if (!member || !parent)
 295                 return -ENOENT;
 296
 297         list_for_each_entry(list, &member->groups, next_group) {
 298                 if (list->group == parent) {
 299                         list_del(&list->next_group);
 300                         list_del(&list->next_member);
 301                         kfree(list);
 302                         return 0;
 303                 }
 304         }
 305         return -ENOENT;
 306 }
 307
 308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 310                                u64 rfer, u64 excl)
 311 {
 312         struct btrfs_qgroup *qgroup;
 313
 314         qgroup = find_qgroup_rb(fs_info, qgroupid);
 315         if (!qgroup)
 316                 return -EINVAL;
 317         if (qgroup->rfer != rfer || qgroup->excl != excl)
 318                 return -EINVAL;
 319         return 0;
 320 }
 321 #endif
 322
 323 /*
 324  * The full config is read in one go, only called from open_ctree()
 325  * It doesn't use any locking, as at this point we're still single-threaded
 326  */
 327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 328 {
 329         struct btrfs_key key;
 330         struct btrfs_key found_key;
 331         struct btrfs_root *quota_root = fs_info->quota_root;
 332         struct btrfs_path *path = NULL;
 333         struct extent_buffer *l;
 334         int slot;
 335         int ret = 0;
 336         u64 flags = 0;
 337         u64 rescan_progress = 0;
 338
 339         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 340                 return 0;
 341
 342         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 343         if (!fs_info->qgroup_ulist) {
 344                 ret = -ENOMEM;
 345                 goto out;
 346         }
 347
 348         path = btrfs_alloc_path();
 349         if (!path) {
 350                 ret = -ENOMEM;
 351                 goto out;
 352         }
 353
 354         /* default this to quota off, in case no status key is found */
 355         fs_info->qgroup_flags = 0;
 356
 357         /*
 358          * pass 1: read status, all qgroup infos and limits
 359          */
 360         key.objectid = 0;
 361         key.type = 0;
 362         key.offset = 0;
 363         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 364         if (ret)
 365                 goto out;
 366
 367         while (1) {
 368                 struct btrfs_qgroup *qgroup;
 369
 370                 slot = path->slots[0];
 371                 l = path->nodes[0];
 372                 btrfs_item_key_to_cpu(l, &found_key, slot);
 373
 374                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 375                         struct btrfs_qgroup_status_item *ptr;
 376
 377                         ptr = btrfs_item_ptr(l, slot,
 378                                              struct btrfs_qgroup_status_item);
 379
 380                         if (btrfs_qgroup_status_version(l, ptr) !=
 381                             BTRFS_QGROUP_STATUS_VERSION) {
 382                                 btrfs_err(fs_info,
 383                                  "old qgroup version, quota disabled");
 384                                 goto out;
 385                         }
 386                         if (btrfs_qgroup_status_generation(l, ptr) !=
 387                             fs_info->generation) {
 388                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 389                                 btrfs_err(fs_info,
 390                                         "qgroup generation mismatch, marked as inconsistent");
 391                         }
 392                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 393                                                                           ptr);
 394                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 395                         goto next1;
 396                 }
 397
 398                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 399                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 400                         goto next1;
 401
 402                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 403                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 404                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 405                         btrfs_err(fs_info, "inconsistent qgroup config");
 406                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 407                 }
 408                 if (!qgroup) {
 409                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 410                         if (IS_ERR(qgroup)) {
 411                                 ret = PTR_ERR(qgroup);
 412                                 goto out;
 413                         }
 414                 }
 415                 switch (found_key.type) {
 416                 case BTRFS_QGROUP_INFO_KEY: {
 417                         struct btrfs_qgroup_info_item *ptr;
 418
 419                         ptr = btrfs_item_ptr(l, slot,
 420                                              struct btrfs_qgroup_info_item);
 421                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 422                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 423                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 424                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 425                         /* generation currently unused */
 426                         break;
 427                 }
 428                 case BTRFS_QGROUP_LIMIT_KEY: {
 429                         struct btrfs_qgroup_limit_item *ptr;
 430
 431                         ptr = btrfs_item_ptr(l, slot,
 432                                              struct btrfs_qgroup_limit_item);
 433                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 434                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 435                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 436                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 437                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 438                         break;
 439                 }
 440                 }
 441 next1:
 442                 ret = btrfs_next_item(quota_root, path);
 443                 if (ret < 0)
 444                         goto out;
 445                 if (ret)
 446                         break;
 447         }
 448         btrfs_release_path(path);
 449
 450         /*
 451          * pass 2: read all qgroup relations
 452          */
 453         key.objectid = 0;
 454         key.type = BTRFS_QGROUP_RELATION_KEY;
 455         key.offset = 0;
 456         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 457         if (ret)
 458                 goto out;
 459         while (1) {
 460                 slot = path->slots[0];
 461                 l = path->nodes[0];
 462                 btrfs_item_key_to_cpu(l, &found_key, slot);
 463
 464                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 465                         goto next2;
 466
 467                 if (found_key.objectid > found_key.offset) {
 468                         /* parent <- member, not needed to build config */
 469                         /* FIXME should we omit the key completely? */
 470                         goto next2;
 471                 }
 472
 473                 ret = add_relation_rb(fs_info, found_key.objectid,
 474                                       found_key.offset);
 475                 if (ret == -ENOENT) {
 476                         btrfs_warn(fs_info,
 477                                 "orphan qgroup relation 0x%llx->0x%llx",
 478                                 found_key.objectid, found_key.offset);
 479                         ret = 0;        /* ignore the error */
 480                 }
 481                 if (ret)
 482                         goto out;
 483 next2:
 484                 ret = btrfs_next_item(quota_root, path);
 485                 if (ret < 0)
 486                         goto out;
 487                 if (ret)
 488                         break;
 489         }
 490 out:
 491         fs_info->qgroup_flags |= flags;
 492         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
 493                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 494         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 495                  ret >= 0)
 496                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 497         btrfs_free_path(path);
 498
 499         if (ret < 0) {
 500                 ulist_free(fs_info->qgroup_ulist);
 501                 fs_info->qgroup_ulist = NULL;
 502                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 503         }
 504
 505         return ret < 0 ? ret : 0;
 506 }
 507
 508 /*
 509  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 510  * first two are in single-threaded paths.And for the third one, we have set
 511  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 512  * up the in-memory structures without qgroup_lock held.
 513  */
 514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 515 {
 516         struct rb_node *n;
 517         struct btrfs_qgroup *qgroup;
 518
 519         while ((n = rb_first(&fs_info->qgroup_tree))) {
 520                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 521                 rb_erase(n, &fs_info->qgroup_tree);
 522                 __del_qgroup_rb(qgroup);
 523         }
 524         /*
 525          * we call btrfs_free_qgroup_config() when umounting
 526          * filesystem and disabling quota, so we set qgroup_ulist
 527          * to be null here to avoid double free.
 528          */
 529         ulist_free(fs_info->qgroup_ulist);
 530         fs_info->qgroup_ulist = NULL;
 531 }
 532
 533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 534                                     u64 dst)
 535 {
 536         int ret;
 537         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 538         struct btrfs_path *path;
 539         struct btrfs_key key;
 540
 541         path = btrfs_alloc_path();
 542         if (!path)
 543                 return -ENOMEM;
 544
 545         key.objectid = src;
 546         key.type = BTRFS_QGROUP_RELATION_KEY;
 547         key.offset = dst;
 548
 549         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 550
 551         btrfs_mark_buffer_dirty(path->nodes[0]);
 552
 553         btrfs_free_path(path);
 554         return ret;
 555 }
 556
 557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 558                                     u64 dst)
 559 {
 560         int ret;
 561         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 562         struct btrfs_path *path;
 563         struct btrfs_key key;
 564
 565         path = btrfs_alloc_path();
 566         if (!path)
 567                 return -ENOMEM;
 568
 569         key.objectid = src;
 570         key.type = BTRFS_QGROUP_RELATION_KEY;
 571         key.offset = dst;
 572
 573         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 574         if (ret < 0)
 575                 goto out;
 576
 577         if (ret > 0) {
 578                 ret = -ENOENT;
 579                 goto out;
 580         }
 581
 582         ret = btrfs_del_item(trans, quota_root, path);
 583 out:
 584         btrfs_free_path(path);
 585         return ret;
 586 }
 587
 588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 589                            struct btrfs_root *quota_root, u64 qgroupid)
 590 {
 591         int ret;
 592         struct btrfs_path *path;
 593         struct btrfs_qgroup_info_item *qgroup_info;
 594         struct btrfs_qgroup_limit_item *qgroup_limit;
 595         struct extent_buffer *leaf;
 596         struct btrfs_key key;
 597
 598         if (btrfs_is_testing(quota_root->fs_info))
 599                 return 0;
 600
 601         path = btrfs_alloc_path();
 602         if (!path)
 603                 return -ENOMEM;
 604
 605         key.objectid = 0;
 606         key.type = BTRFS_QGROUP_INFO_KEY;
 607         key.offset = qgroupid;
 608
 609         /*
 610          * Avoid a transaction abort by catching -EEXIST here. In that
 611          * case, we proceed by re-initializing the existing structure
 612          * on disk.
 613          */
 614
 615         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 616                                       sizeof(*qgroup_info));
 617         if (ret && ret != -EEXIST)
 618                 goto out;
 619
 620         leaf = path->nodes[0];
 621         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 622                                  struct btrfs_qgroup_info_item);
 623         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 624         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 625         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 626         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 627         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 628
 629         btrfs_mark_buffer_dirty(leaf);
 630
 631         btrfs_release_path(path);
 632
 633         key.type = BTRFS_QGROUP_LIMIT_KEY;
 634         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 635                                       sizeof(*qgroup_limit));
 636         if (ret && ret != -EEXIST)
 637                 goto out;
 638
 639         leaf = path->nodes[0];
 640         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 641                                   struct btrfs_qgroup_limit_item);
 642         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 643         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 644         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 645         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 646         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 647
 648         btrfs_mark_buffer_dirty(leaf);
 649
 650         ret = 0;
 651 out:
 652         btrfs_free_path(path);
 653         return ret;
 654 }
 655
 656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
 657 {
 658         int ret;
 659         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 660         struct btrfs_path *path;
 661         struct btrfs_key key;
 662
 663         path = btrfs_alloc_path();
 664         if (!path)
 665                 return -ENOMEM;
 666
 667         key.objectid = 0;
 668         key.type = BTRFS_QGROUP_INFO_KEY;
 669         key.offset = qgroupid;
 670         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 671         if (ret < 0)
 672                 goto out;
 673
 674         if (ret > 0) {
 675                 ret = -ENOENT;
 676                 goto out;
 677         }
 678
 679         ret = btrfs_del_item(trans, quota_root, path);
 680         if (ret)
 681                 goto out;
 682
 683         btrfs_release_path(path);
 684
 685         key.type = BTRFS_QGROUP_LIMIT_KEY;
 686         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 687         if (ret < 0)
 688                 goto out;
 689
 690         if (ret > 0) {
 691                 ret = -ENOENT;
 692                 goto out;
 693         }
 694
 695         ret = btrfs_del_item(trans, quota_root, path);
 696
 697 out:
 698         btrfs_free_path(path);
 699         return ret;
 700 }
 701
 702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 703                                     struct btrfs_qgroup *qgroup)
 704 {
 705         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 706         struct btrfs_path *path;
 707         struct btrfs_key key;
 708         struct extent_buffer *l;
 709         struct btrfs_qgroup_limit_item *qgroup_limit;
 710         int ret;
 711         int slot;
 712
 713         key.objectid = 0;
 714         key.type = BTRFS_QGROUP_LIMIT_KEY;
 715         key.offset = qgroup->qgroupid;
 716
 717         path = btrfs_alloc_path();
 718         if (!path)
 719                 return -ENOMEM;
 720
 721         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 722         if (ret > 0)
 723                 ret = -ENOENT;
 724
 725         if (ret)
 726                 goto out;
 727
 728         l = path->nodes[0];
 729         slot = path->slots[0];
 730         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 731         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 732         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 733         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 734         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 735         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 736
 737         btrfs_mark_buffer_dirty(l);
 738
 739 out:
 740         btrfs_free_path(path);
 741         return ret;
 742 }
 743
 744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 745                                    struct btrfs_qgroup *qgroup)
 746 {
 747         struct btrfs_fs_info *fs_info = trans->fs_info;
 748         struct btrfs_root *quota_root = fs_info->quota_root;
 749         struct btrfs_path *path;
 750         struct btrfs_key key;
 751         struct extent_buffer *l;
 752         struct btrfs_qgroup_info_item *qgroup_info;
 753         int ret;
 754         int slot;
 755
 756         if (btrfs_is_testing(fs_info))
 757                 return 0;
 758
 759         key.objectid = 0;
 760         key.type = BTRFS_QGROUP_INFO_KEY;
 761         key.offset = qgroup->qgroupid;
 762
 763         path = btrfs_alloc_path();
 764         if (!path)
 765                 return -ENOMEM;
 766
 767         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 768         if (ret > 0)
 769                 ret = -ENOENT;
 770
 771         if (ret)
 772                 goto out;
 773
 774         l = path->nodes[0];
 775         slot = path->slots[0];
 776         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 777         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 778         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 779         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 780         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 781         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 782
 783         btrfs_mark_buffer_dirty(l);
 784
 785 out:
 786         btrfs_free_path(path);
 787         return ret;
 788 }
 789
 790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
 791 {
 792         struct btrfs_fs_info *fs_info = trans->fs_info;
 793         struct btrfs_root *quota_root = fs_info->quota_root;
 794         struct btrfs_path *path;
 795         struct btrfs_key key;
 796         struct extent_buffer *l;
 797         struct btrfs_qgroup_status_item *ptr;
 798         int ret;
 799         int slot;
 800
 801         key.objectid = 0;
 802         key.type = BTRFS_QGROUP_STATUS_KEY;
 803         key.offset = 0;
 804
 805         path = btrfs_alloc_path();
 806         if (!path)
 807                 return -ENOMEM;
 808
 809         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 810         if (ret > 0)
 811                 ret = -ENOENT;
 812
 813         if (ret)
 814                 goto out;
 815
 816         l = path->nodes[0];
 817         slot = path->slots[0];
 818         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 819         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 820         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 821         btrfs_set_qgroup_status_rescan(l, ptr,
 822                                 fs_info->qgroup_rescan_progress.objectid);
 823
 824         btrfs_mark_buffer_dirty(l);
 825
 826 out:
 827         btrfs_free_path(path);
 828         return ret;
 829 }
 830
 831 /*
 832  * called with qgroup_lock held
 833  */
 834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 835                                   struct btrfs_root *root)
 836 {
 837         struct btrfs_path *path;
 838         struct btrfs_key key;
 839         struct extent_buffer *leaf = NULL;
 840         int ret;
 841         int nr = 0;
 842
 843         path = btrfs_alloc_path();
 844         if (!path)
 845                 return -ENOMEM;
 846
 847         path->leave_spinning = 1;
 848
 849         key.objectid = 0;
 850         key.offset = 0;
 851         key.type = 0;
 852
 853         while (1) {
 854                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 855                 if (ret < 0)
 856                         goto out;
 857                 leaf = path->nodes[0];
 858                 nr = btrfs_header_nritems(leaf);
 859                 if (!nr)
 860                         break;
 861                 /*
 862                  * delete the leaf one by one
 863                  * since the whole tree is going
 864                  * to be deleted.
 865                  */
 866                 path->slots[0] = 0;
 867                 ret = btrfs_del_items(trans, root, path, 0, nr);
 868                 if (ret)
 869                         goto out;
 870
 871                 btrfs_release_path(path);
 872         }
 873         ret = 0;
 874 out:
 875         btrfs_free_path(path);
 876         return ret;
 877 }
 878
 879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 880 {
 881         struct btrfs_root *quota_root;
 882         struct btrfs_root *tree_root = fs_info->tree_root;
 883         struct btrfs_path *path = NULL;
 884         struct btrfs_qgroup_status_item *ptr;
 885         struct extent_buffer *leaf;
 886         struct btrfs_key key;
 887         struct btrfs_key found_key;
 888         struct btrfs_qgroup *qgroup = NULL;
 889         struct btrfs_trans_handle *trans = NULL;
 890         int ret = 0;
 891         int slot;
 892
 893         mutex_lock(&fs_info->qgroup_ioctl_lock);
 894         if (fs_info->quota_root)
 895                 goto out;
 896
 897         /*
 898          * 1 for quota root item
 899          * 1 for BTRFS_QGROUP_STATUS item
 900          *
 901          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
 902          * per subvolume. However those are not currently reserved since it
 903          * would be a lot of overkill.
 904          */
 905         trans = btrfs_start_transaction(tree_root, 2);
 906         if (IS_ERR(trans)) {
 907                 ret = PTR_ERR(trans);
 908                 trans = NULL;
 909                 goto out;
 910         }
 911
 912         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 913         if (!fs_info->qgroup_ulist) {
 914                 ret = -ENOMEM;
 915                 btrfs_abort_transaction(trans, ret);
 916                 goto out;
 917         }
 918
 919         /*
 920          * initially create the quota tree
 921          */
 922         quota_root = btrfs_create_tree(trans, fs_info,
 923                                        BTRFS_QUOTA_TREE_OBJECTID);
 924         if (IS_ERR(quota_root)) {
 925                 ret =  PTR_ERR(quota_root);
 926                 btrfs_abort_transaction(trans, ret);
 927                 goto out;
 928         }
 929
 930         path = btrfs_alloc_path();
 931         if (!path) {
 932                 ret = -ENOMEM;
 933                 btrfs_abort_transaction(trans, ret);
 934                 goto out_free_root;
 935         }
 936
 937         key.objectid = 0;
 938         key.type = BTRFS_QGROUP_STATUS_KEY;
 939         key.offset = 0;
 940
 941         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 942                                       sizeof(*ptr));
 943         if (ret) {
 944                 btrfs_abort_transaction(trans, ret);
 945                 goto out_free_path;
 946         }
 947
 948         leaf = path->nodes[0];
 949         ptr = btrfs_item_ptr(leaf, path->slots[0],
 950                                  struct btrfs_qgroup_status_item);
 951         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 952         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 953         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 954                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 955         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 956         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 957
 958         btrfs_mark_buffer_dirty(leaf);
 959
 960         key.objectid = 0;
 961         key.type = BTRFS_ROOT_REF_KEY;
 962         key.offset = 0;
 963
 964         btrfs_release_path(path);
 965         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 966         if (ret > 0)
 967                 goto out_add_root;
 968         if (ret < 0) {
 969                 btrfs_abort_transaction(trans, ret);
 970                 goto out_free_path;
 971         }
 972
 973         while (1) {
 974                 slot = path->slots[0];
 975                 leaf = path->nodes[0];
 976                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 977
 978                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 979                         ret = add_qgroup_item(trans, quota_root,
 980                                               found_key.offset);
 981                         if (ret) {
 982                                 btrfs_abort_transaction(trans, ret);
 983                                 goto out_free_path;
 984                         }
 985
 986                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 987                         if (IS_ERR(qgroup)) {
 988                                 ret = PTR_ERR(qgroup);
 989                                 btrfs_abort_transaction(trans, ret);
 990                                 goto out_free_path;
 991                         }
 992                 }
 993                 ret = btrfs_next_item(tree_root, path);
 994                 if (ret < 0) {
 995                         btrfs_abort_transaction(trans, ret);
 996                         goto out_free_path;
 997                 }
 998                 if (ret)
 999                         break;
1000         }
1001
1002 out_add_root:
1003         btrfs_release_path(path);
1004         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1005         if (ret) {
1006                 btrfs_abort_transaction(trans, ret);
1007                 goto out_free_path;
1008         }
1009
1010         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011         if (IS_ERR(qgroup)) {
1012                 ret = PTR_ERR(qgroup);
1013                 btrfs_abort_transaction(trans, ret);
1014                 goto out_free_path;
1015         }
1016         spin_lock(&fs_info->qgroup_lock);
1017         fs_info->quota_root = quota_root;
1018         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1019         spin_unlock(&fs_info->qgroup_lock);
1020
1021         ret = btrfs_commit_transaction(trans);
1022         if (ret) {
1023                 trans = NULL;
1024                 goto out_free_path;
1025         }
1026
1027         ret = qgroup_rescan_init(fs_info, 0, 1);
1028         if (!ret) {
1029                 qgroup_rescan_zero_tracking(fs_info);
1030                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1031                                  &fs_info->qgroup_rescan_work);
1032         }
1033
1034 out_free_path:
1035         btrfs_free_path(path);
1036 out_free_root:
1037         if (ret) {
1038                 free_extent_buffer(quota_root->node);
1039                 free_extent_buffer(quota_root->commit_root);
1040                 kfree(quota_root);
1041         }
1042 out:
1043         if (ret) {
1044                 ulist_free(fs_info->qgroup_ulist);
1045                 fs_info->qgroup_ulist = NULL;
1046                 if (trans)
1047                         btrfs_end_transaction(trans);
1048         }
1049         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1050         return ret;
1051 }
1052
1053 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1054 {
1055         struct btrfs_root *quota_root;
1056         struct btrfs_trans_handle *trans = NULL;
1057         int ret = 0;
1058
1059         mutex_lock(&fs_info->qgroup_ioctl_lock);
1060         if (!fs_info->quota_root)
1061                 goto out;
1062
1063         /*
1064          * 1 For the root item
1065          *
1066          * We should also reserve enough items for the quota tree deletion in
1067          * btrfs_clean_quota_tree but this is not done.
1068          */
1069         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1070         if (IS_ERR(trans)) {
1071                 ret = PTR_ERR(trans);
1072                 goto out;
1073         }
1074
1075         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1076         btrfs_qgroup_wait_for_completion(fs_info, false);
1077         spin_lock(&fs_info->qgroup_lock);
1078         quota_root = fs_info->quota_root;
1079         fs_info->quota_root = NULL;
1080         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1081         spin_unlock(&fs_info->qgroup_lock);
1082
1083         btrfs_free_qgroup_config(fs_info);
1084
1085         ret = btrfs_clean_quota_tree(trans, quota_root);
1086         if (ret) {
1087                 btrfs_abort_transaction(trans, ret);
1088                 goto end_trans;
1089         }
1090
1091         ret = btrfs_del_root(trans, &quota_root->root_key);
1092         if (ret) {
1093                 btrfs_abort_transaction(trans, ret);
1094                 goto end_trans;
1095         }
1096
1097         list_del(&quota_root->dirty_list);
1098
1099         btrfs_tree_lock(quota_root->node);
1100         clean_tree_block(fs_info, quota_root->node);
1101         btrfs_tree_unlock(quota_root->node);
1102         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1103
1104         free_extent_buffer(quota_root->node);
1105         free_extent_buffer(quota_root->commit_root);
1106         kfree(quota_root);
1107
1108 end_trans:
1109         ret = btrfs_end_transaction(trans);
1110 out:
1111         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1112         return ret;
1113 }
1114
1115 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1116                          struct btrfs_qgroup *qgroup)
1117 {
1118         if (list_empty(&qgroup->dirty))
1119                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1120 }
1121
1122 /*
1123  * The easy accounting, we're updating qgroup relationship whose child qgroup
1124  * only has exclusive extents.
1125  *
1126  * In this case, all exclsuive extents will also be exlusive for parent, so
1127  * excl/rfer just get added/removed.
1128  *
1129  * So is qgroup reservation space, which should also be added/removed to
1130  * parent.
1131  * Or when child tries to release reservation space, parent will underflow its
1132  * reservation (for relationship adding case).
1133  *
1134  * Caller should hold fs_info->qgroup_lock.
1135  */
1136 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1137                                     struct ulist *tmp, u64 ref_root,
1138                                     struct btrfs_qgroup *src, int sign)
1139 {
1140         struct btrfs_qgroup *qgroup;
1141         struct btrfs_qgroup_list *glist;
1142         struct ulist_node *unode;
1143         struct ulist_iterator uiter;
1144         u64 num_bytes = src->excl;
1145         int ret = 0;
1146
1147         qgroup = find_qgroup_rb(fs_info, ref_root);
1148         if (!qgroup)
1149                 goto out;
1150
1151         qgroup->rfer += sign * num_bytes;
1152         qgroup->rfer_cmpr += sign * num_bytes;
1153
1154         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1155         qgroup->excl += sign * num_bytes;
1156         qgroup->excl_cmpr += sign * num_bytes;
1157
1158         if (sign > 0)
1159                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1160         else
1161                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1162
1163         qgroup_dirty(fs_info, qgroup);
1164
1165         /* Get all of the parent groups that contain this qgroup */
1166         list_for_each_entry(glist, &qgroup->groups, next_group) {
1167                 ret = ulist_add(tmp, glist->group->qgroupid,
1168                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1169                 if (ret < 0)
1170                         goto out;
1171         }
1172
1173         /* Iterate all of the parents and adjust their reference counts */
1174         ULIST_ITER_INIT(&uiter);
1175         while ((unode = ulist_next(tmp, &uiter))) {
1176                 qgroup = unode_aux_to_qgroup(unode);
1177                 qgroup->rfer += sign * num_bytes;
1178                 qgroup->rfer_cmpr += sign * num_bytes;
1179                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1180                 qgroup->excl += sign * num_bytes;
1181                 if (sign > 0)
1182                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1183                 else
1184                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1185                 qgroup->excl_cmpr += sign * num_bytes;
1186                 qgroup_dirty(fs_info, qgroup);
1187
1188                 /* Add any parents of the parents */
1189                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1190                         ret = ulist_add(tmp, glist->group->qgroupid,
1191                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1192                         if (ret < 0)
1193                                 goto out;
1194                 }
1195         }
1196         ret = 0;
1197 out:
1198         return ret;
1199 }
1200
1201
1202 /*
1203  * Quick path for updating qgroup with only excl refs.
1204  *
1205  * In that case, just update all parent will be enough.
1206  * Or we needs to do a full rescan.
1207  * Caller should also hold fs_info->qgroup_lock.
1208  *
1209  * Return 0 for quick update, return >0 for need to full rescan
1210  * and mark INCONSISTENT flag.
1211  * Return < 0 for other error.
1212  */
1213 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1214                                    struct ulist *tmp, u64 src, u64 dst,
1215                                    int sign)
1216 {
1217         struct btrfs_qgroup *qgroup;
1218         int ret = 1;
1219         int err = 0;
1220
1221         qgroup = find_qgroup_rb(fs_info, src);
1222         if (!qgroup)
1223                 goto out;
1224         if (qgroup->excl == qgroup->rfer) {
1225                 ret = 0;
1226                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1227                                                qgroup, sign);
1228                 if (err < 0) {
1229                         ret = err;
1230                         goto out;
1231                 }
1232         }
1233 out:
1234         if (ret)
1235                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1236         return ret;
1237 }
1238
1239 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1240                               u64 dst)
1241 {
1242         struct btrfs_fs_info *fs_info = trans->fs_info;
1243         struct btrfs_root *quota_root;
1244         struct btrfs_qgroup *parent;
1245         struct btrfs_qgroup *member;
1246         struct btrfs_qgroup_list *list;
1247         struct ulist *tmp;
1248         int ret = 0;
1249
1250         /* Check the level of src and dst first */
1251         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1252                 return -EINVAL;
1253
1254         tmp = ulist_alloc(GFP_KERNEL);
1255         if (!tmp)
1256                 return -ENOMEM;
1257
1258         mutex_lock(&fs_info->qgroup_ioctl_lock);
1259         quota_root = fs_info->quota_root;
1260         if (!quota_root) {
1261                 ret = -EINVAL;
1262                 goto out;
1263         }
1264         member = find_qgroup_rb(fs_info, src);
1265         parent = find_qgroup_rb(fs_info, dst);
1266         if (!member || !parent) {
1267                 ret = -EINVAL;
1268                 goto out;
1269         }
1270
1271         /* check if such qgroup relation exist firstly */
1272         list_for_each_entry(list, &member->groups, next_group) {
1273                 if (list->group == parent) {
1274                         ret = -EEXIST;
1275                         goto out;
1276                 }
1277         }
1278
1279         ret = add_qgroup_relation_item(trans, src, dst);
1280         if (ret)
1281                 goto out;
1282
1283         ret = add_qgroup_relation_item(trans, dst, src);
1284         if (ret) {
1285                 del_qgroup_relation_item(trans, src, dst);
1286                 goto out;
1287         }
1288
1289         spin_lock(&fs_info->qgroup_lock);
1290         ret = add_relation_rb(fs_info, src, dst);
1291         if (ret < 0) {
1292                 spin_unlock(&fs_info->qgroup_lock);
1293                 goto out;
1294         }
1295         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1296         spin_unlock(&fs_info->qgroup_lock);
1297 out:
1298         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1299         ulist_free(tmp);
1300         return ret;
1301 }
1302
1303 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1304                                  u64 dst)
1305 {
1306         struct btrfs_fs_info *fs_info = trans->fs_info;
1307         struct btrfs_root *quota_root;
1308         struct btrfs_qgroup *parent;
1309         struct btrfs_qgroup *member;
1310         struct btrfs_qgroup_list *list;
1311         struct ulist *tmp;
1312         int ret = 0;
1313         int err;
1314
1315         tmp = ulist_alloc(GFP_KERNEL);
1316         if (!tmp)
1317                 return -ENOMEM;
1318
1319         quota_root = fs_info->quota_root;
1320         if (!quota_root) {
1321                 ret = -EINVAL;
1322                 goto out;
1323         }
1324
1325         member = find_qgroup_rb(fs_info, src);
1326         parent = find_qgroup_rb(fs_info, dst);
1327         if (!member || !parent) {
1328                 ret = -EINVAL;
1329                 goto out;
1330         }
1331
1332         /* check if such qgroup relation exist firstly */
1333         list_for_each_entry(list, &member->groups, next_group) {
1334                 if (list->group == parent)
1335                         goto exist;
1336         }
1337         ret = -ENOENT;
1338         goto out;
1339 exist:
1340         ret = del_qgroup_relation_item(trans, src, dst);
1341         err = del_qgroup_relation_item(trans, dst, src);
1342         if (err && !ret)
1343                 ret = err;
1344
1345         spin_lock(&fs_info->qgroup_lock);
1346         del_relation_rb(fs_info, src, dst);
1347         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1348         spin_unlock(&fs_info->qgroup_lock);
1349 out:
1350         ulist_free(tmp);
1351         return ret;
1352 }
1353
1354 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1355                               u64 dst)
1356 {
1357         struct btrfs_fs_info *fs_info = trans->fs_info;
1358         int ret = 0;
1359
1360         mutex_lock(&fs_info->qgroup_ioctl_lock);
1361         ret = __del_qgroup_relation(trans, src, dst);
1362         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1363
1364         return ret;
1365 }
1366
1367 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1368 {
1369         struct btrfs_fs_info *fs_info = trans->fs_info;
1370         struct btrfs_root *quota_root;
1371         struct btrfs_qgroup *qgroup;
1372         int ret = 0;
1373
1374         mutex_lock(&fs_info->qgroup_ioctl_lock);
1375         quota_root = fs_info->quota_root;
1376         if (!quota_root) {
1377                 ret = -EINVAL;
1378                 goto out;
1379         }
1380         qgroup = find_qgroup_rb(fs_info, qgroupid);
1381         if (qgroup) {
1382                 ret = -EEXIST;
1383                 goto out;
1384         }
1385
1386         ret = add_qgroup_item(trans, quota_root, qgroupid);
1387         if (ret)
1388                 goto out;
1389
1390         spin_lock(&fs_info->qgroup_lock);
1391         qgroup = add_qgroup_rb(fs_info, qgroupid);
1392         spin_unlock(&fs_info->qgroup_lock);
1393
1394         if (IS_ERR(qgroup))
1395                 ret = PTR_ERR(qgroup);
1396 out:
1397         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1398         return ret;
1399 }
1400
1401 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1402 {
1403         struct btrfs_fs_info *fs_info = trans->fs_info;
1404         struct btrfs_root *quota_root;
1405         struct btrfs_qgroup *qgroup;
1406         struct btrfs_qgroup_list *list;
1407         int ret = 0;
1408
1409         mutex_lock(&fs_info->qgroup_ioctl_lock);
1410         quota_root = fs_info->quota_root;
1411         if (!quota_root) {
1412                 ret = -EINVAL;
1413                 goto out;
1414         }
1415
1416         qgroup = find_qgroup_rb(fs_info, qgroupid);
1417         if (!qgroup) {
1418                 ret = -ENOENT;
1419                 goto out;
1420         } else {
1421                 /* check if there are no children of this qgroup */
1422                 if (!list_empty(&qgroup->members)) {
1423                         ret = -EBUSY;
1424                         goto out;
1425                 }
1426         }
1427         ret = del_qgroup_item(trans, qgroupid);
1428         if (ret && ret != -ENOENT)
1429                 goto out;
1430
1431         while (!list_empty(&qgroup->groups)) {
1432                 list = list_first_entry(&qgroup->groups,
1433                                         struct btrfs_qgroup_list, next_group);
1434                 ret = __del_qgroup_relation(trans, qgroupid,
1435                                             list->group->qgroupid);
1436                 if (ret)
1437                         goto out;
1438         }
1439
1440         spin_lock(&fs_info->qgroup_lock);
1441         del_qgroup_rb(fs_info, qgroupid);
1442         spin_unlock(&fs_info->qgroup_lock);
1443 out:
1444         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1445         return ret;
1446 }
1447
1448 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1449                        struct btrfs_qgroup_limit *limit)
1450 {
1451         struct btrfs_fs_info *fs_info = trans->fs_info;
1452         struct btrfs_root *quota_root;
1453         struct btrfs_qgroup *qgroup;
1454         int ret = 0;
1455         /* Sometimes we would want to clear the limit on this qgroup.
1456          * To meet this requirement, we treat the -1 as a special value
1457          * which tell kernel to clear the limit on this qgroup.
1458          */
1459         const u64 CLEAR_VALUE = -1;
1460
1461         mutex_lock(&fs_info->qgroup_ioctl_lock);
1462         quota_root = fs_info->quota_root;
1463         if (!quota_root) {
1464                 ret = -EINVAL;
1465                 goto out;
1466         }
1467
1468         qgroup = find_qgroup_rb(fs_info, qgroupid);
1469         if (!qgroup) {
1470                 ret = -ENOENT;
1471                 goto out;
1472         }
1473
1474         spin_lock(&fs_info->qgroup_lock);
1475         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1476                 if (limit->max_rfer == CLEAR_VALUE) {
1477                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1478                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1479                         qgroup->max_rfer = 0;
1480                 } else {
1481                         qgroup->max_rfer = limit->max_rfer;
1482                 }
1483         }
1484         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1485                 if (limit->max_excl == CLEAR_VALUE) {
1486                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1487                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1488                         qgroup->max_excl = 0;
1489                 } else {
1490                         qgroup->max_excl = limit->max_excl;
1491                 }
1492         }
1493         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1494                 if (limit->rsv_rfer == CLEAR_VALUE) {
1495                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1496                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1497                         qgroup->rsv_rfer = 0;
1498                 } else {
1499                         qgroup->rsv_rfer = limit->rsv_rfer;
1500                 }
1501         }
1502         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1503                 if (limit->rsv_excl == CLEAR_VALUE) {
1504                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1505                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1506                         qgroup->rsv_excl = 0;
1507                 } else {
1508                         qgroup->rsv_excl = limit->rsv_excl;
1509                 }
1510         }
1511         qgroup->lim_flags |= limit->flags;
1512
1513         spin_unlock(&fs_info->qgroup_lock);
1514
1515         ret = update_qgroup_limit_item(trans, qgroup);
1516         if (ret) {
1517                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1518                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1519                        qgroupid);
1520         }
1521
1522 out:
1523         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1524         return ret;
1525 }
1526
1527 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1528                                 struct btrfs_delayed_ref_root *delayed_refs,
1529                                 struct btrfs_qgroup_extent_record *record)
1530 {
1531         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1532         struct rb_node *parent_node = NULL;
1533         struct btrfs_qgroup_extent_record *entry;
1534         u64 bytenr = record->bytenr;
1535
1536         lockdep_assert_held(&delayed_refs->lock);
1537         trace_btrfs_qgroup_trace_extent(fs_info, record);
1538
1539         while (*p) {
1540                 parent_node = *p;
1541                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1542                                  node);
1543                 if (bytenr < entry->bytenr)
1544                         p = &(*p)->rb_left;
1545                 else if (bytenr > entry->bytenr)
1546                         p = &(*p)->rb_right;
1547                 else
1548                         return 1;
1549         }
1550
1551         rb_link_node(&record->node, parent_node, p);
1552         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1553         return 0;
1554 }
1555
1556 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1557                                    struct btrfs_qgroup_extent_record *qrecord)
1558 {
1559         struct ulist *old_root;
1560         u64 bytenr = qrecord->bytenr;
1561         int ret;
1562
1563         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1564         if (ret < 0) {
1565                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1566                 btrfs_warn(fs_info,
1567 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1568                         ret);
1569                 return 0;
1570         }
1571
1572         /*
1573          * Here we don't need to get the lock of
1574          * trans->transaction->delayed_refs, since inserted qrecord won't
1575          * be deleted, only qrecord->node may be modified (new qrecord insert)
1576          *
1577          * So modifying qrecord->old_roots is safe here
1578          */
1579         qrecord->old_roots = old_root;
1580         return 0;
1581 }
1582
1583 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1584                               u64 num_bytes, gfp_t gfp_flag)
1585 {
1586         struct btrfs_fs_info *fs_info = trans->fs_info;
1587         struct btrfs_qgroup_extent_record *record;
1588         struct btrfs_delayed_ref_root *delayed_refs;
1589         int ret;
1590
1591         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1592             || bytenr == 0 || num_bytes == 0)
1593                 return 0;
1594         record = kmalloc(sizeof(*record), gfp_flag);
1595         if (!record)
1596                 return -ENOMEM;
1597
1598         delayed_refs = &trans->transaction->delayed_refs;
1599         record->bytenr = bytenr;
1600         record->num_bytes = num_bytes;
1601         record->old_roots = NULL;
1602
1603         spin_lock(&delayed_refs->lock);
1604         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1605         spin_unlock(&delayed_refs->lock);
1606         if (ret > 0) {
1607                 kfree(record);
1608                 return 0;
1609         }
1610         return btrfs_qgroup_trace_extent_post(fs_info, record);
1611 }
1612
1613 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1614                                   struct extent_buffer *eb)
1615 {
1616         struct btrfs_fs_info *fs_info = trans->fs_info;
1617         int nr = btrfs_header_nritems(eb);
1618         int i, extent_type, ret;
1619         struct btrfs_key key;
1620         struct btrfs_file_extent_item *fi;
1621         u64 bytenr, num_bytes;
1622
1623         /* We can be called directly from walk_up_proc() */
1624         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1625                 return 0;
1626
1627         for (i = 0; i < nr; i++) {
1628                 btrfs_item_key_to_cpu(eb, &key, i);
1629
1630                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1631                         continue;
1632
1633                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1634                 /* filter out non qgroup-accountable extents  */
1635                 extent_type = btrfs_file_extent_type(eb, fi);
1636
1637                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1638                         continue;
1639
1640                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1641                 if (!bytenr)
1642                         continue;
1643
1644                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1645
1646                 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1647                                                 GFP_NOFS);
1648                 if (ret)
1649                         return ret;
1650         }
1651         cond_resched();
1652         return 0;
1653 }
1654
1655 /*
1656  * Walk up the tree from the bottom, freeing leaves and any interior
1657  * nodes which have had all slots visited. If a node (leaf or
1658  * interior) is freed, the node above it will have it's slot
1659  * incremented. The root node will never be freed.
1660  *
1661  * At the end of this function, we should have a path which has all
1662  * slots incremented to the next position for a search. If we need to
1663  * read a new node it will be NULL and the node above it will have the
1664  * correct slot selected for a later read.
1665  *
1666  * If we increment the root nodes slot counter past the number of
1667  * elements, 1 is returned to signal completion of the search.
1668  */
1669 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1670 {
1671         int level = 0;
1672         int nr, slot;
1673         struct extent_buffer *eb;
1674
1675         if (root_level == 0)
1676                 return 1;
1677
1678         while (level <= root_level) {
1679                 eb = path->nodes[level];
1680                 nr = btrfs_header_nritems(eb);
1681                 path->slots[level]++;
1682                 slot = path->slots[level];
1683                 if (slot >= nr || level == 0) {
1684                         /*
1685                          * Don't free the root -  we will detect this
1686                          * condition after our loop and return a
1687                          * positive value for caller to stop walking the tree.
1688                          */
1689                         if (level != root_level) {
1690                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1691                                 path->locks[level] = 0;
1692
1693                                 free_extent_buffer(eb);
1694                                 path->nodes[level] = NULL;
1695                                 path->slots[level] = 0;
1696                         }
1697                 } else {
1698                         /*
1699                          * We have a valid slot to walk back down
1700                          * from. Stop here so caller can process these
1701                          * new nodes.
1702                          */
1703                         break;
1704                 }
1705
1706                 level++;
1707         }
1708
1709         eb = path->nodes[root_level];
1710         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1711                 return 1;
1712
1713         return 0;
1714 }
1715
1716 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1717                                struct extent_buffer *root_eb,
1718                                u64 root_gen, int root_level)
1719 {
1720         struct btrfs_fs_info *fs_info = trans->fs_info;
1721         int ret = 0;
1722         int level;
1723         struct extent_buffer *eb = root_eb;
1724         struct btrfs_path *path = NULL;
1725
1726         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1727         BUG_ON(root_eb == NULL);
1728
1729         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1730                 return 0;
1731
1732         if (!extent_buffer_uptodate(root_eb)) {
1733                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
1734                 if (ret)
1735                         goto out;
1736         }
1737
1738         if (root_level == 0) {
1739                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
1740                 goto out;
1741         }
1742
1743         path = btrfs_alloc_path();
1744         if (!path)
1745                 return -ENOMEM;
1746
1747         /*
1748          * Walk down the tree.  Missing extent blocks are filled in as
1749          * we go. Metadata is accounted every time we read a new
1750          * extent block.
1751          *
1752          * When we reach a leaf, we account for file extent items in it,
1753          * walk back up the tree (adjusting slot pointers as we go)
1754          * and restart the search process.
1755          */
1756         extent_buffer_get(root_eb); /* For path */
1757         path->nodes[root_level] = root_eb;
1758         path->slots[root_level] = 0;
1759         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1760 walk_down:
1761         level = root_level;
1762         while (level >= 0) {
1763                 if (path->nodes[level] == NULL) {
1764                         struct btrfs_key first_key;
1765                         int parent_slot;
1766                         u64 child_gen;
1767                         u64 child_bytenr;
1768
1769                         /*
1770                          * We need to get child blockptr/gen from parent before
1771                          * we can read it.
1772                           */
1773                         eb = path->nodes[level + 1];
1774                         parent_slot = path->slots[level + 1];
1775                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1776                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1777                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1778
1779                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1780                                              level, &first_key);
1781                         if (IS_ERR(eb)) {
1782                                 ret = PTR_ERR(eb);
1783                                 goto out;
1784                         } else if (!extent_buffer_uptodate(eb)) {
1785                                 free_extent_buffer(eb);
1786                                 ret = -EIO;
1787                                 goto out;
1788                         }
1789
1790                         path->nodes[level] = eb;
1791                         path->slots[level] = 0;
1792
1793                         btrfs_tree_read_lock(eb);
1794                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1795                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1796
1797                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
1798                                                         fs_info->nodesize,
1799                                                         GFP_NOFS);
1800                         if (ret)
1801                                 goto out;
1802                 }
1803
1804                 if (level == 0) {
1805                         ret = btrfs_qgroup_trace_leaf_items(trans,
1806                                                             path->nodes[level]);
1807                         if (ret)
1808                                 goto out;
1809
1810                         /* Nonzero return here means we completed our search */
1811                         ret = adjust_slots_upwards(path, root_level);
1812                         if (ret)
1813                                 break;
1814
1815                         /* Restart search with new slots */
1816                         goto walk_down;
1817                 }
1818
1819                 level--;
1820         }
1821
1822         ret = 0;
1823 out:
1824         btrfs_free_path(path);
1825
1826         return ret;
1827 }
1828
1829 #define UPDATE_NEW      0
1830 #define UPDATE_OLD      1
1831 /*
1832  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1833  */
1834 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1835                                 struct ulist *roots, struct ulist *tmp,
1836                                 struct ulist *qgroups, u64 seq, int update_old)
1837 {
1838         struct ulist_node *unode;
1839         struct ulist_iterator uiter;
1840         struct ulist_node *tmp_unode;
1841         struct ulist_iterator tmp_uiter;
1842         struct btrfs_qgroup *qg;
1843         int ret = 0;
1844
1845         if (!roots)
1846                 return 0;
1847         ULIST_ITER_INIT(&uiter);
1848         while ((unode = ulist_next(roots, &uiter))) {
1849                 qg = find_qgroup_rb(fs_info, unode->val);
1850                 if (!qg)
1851                         continue;
1852
1853                 ulist_reinit(tmp);
1854                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1855                                 GFP_ATOMIC);
1856                 if (ret < 0)
1857                         return ret;
1858                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1859                 if (ret < 0)
1860                         return ret;
1861                 ULIST_ITER_INIT(&tmp_uiter);
1862                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1863                         struct btrfs_qgroup_list *glist;
1864
1865                         qg = unode_aux_to_qgroup(tmp_unode);
1866                         if (update_old)
1867                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1868                         else
1869                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1870                         list_for_each_entry(glist, &qg->groups, next_group) {
1871                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1872                                                 qgroup_to_aux(glist->group),
1873                                                 GFP_ATOMIC);
1874                                 if (ret < 0)
1875                                         return ret;
1876                                 ret = ulist_add(tmp, glist->group->qgroupid,
1877                                                 qgroup_to_aux(glist->group),
1878                                                 GFP_ATOMIC);
1879                                 if (ret < 0)
1880                                         return ret;
1881                         }
1882                 }
1883         }
1884         return 0;
1885 }
1886
1887 /*
1888  * Update qgroup rfer/excl counters.
1889  * Rfer update is easy, codes can explain themselves.
1890  *
1891  * Excl update is tricky, the update is split into 2 part.
1892  * Part 1: Possible exclusive <-> sharing detect:
1893  *      |       A       |       !A      |
1894  *  -------------------------------------
1895  *  B   |       *       |       -       |
1896  *  -------------------------------------
1897  *  !B  |       +       |       **      |
1898  *  -------------------------------------
1899  *
1900  * Conditions:
1901  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1902  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1903  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1904  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1905  *
1906  * Results:
1907  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1908  * *: Definitely not changed.           **: Possible unchanged.
1909  *
1910  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1911  *
1912  * To make the logic clear, we first use condition A and B to split
1913  * combination into 4 results.
1914  *
1915  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1916  * only on variant maybe 0.
1917  *
1918  * Lastly, check result **, since there are 2 variants maybe 0, split them
1919  * again(2x2).
1920  * But this time we don't need to consider other things, the codes and logic
1921  * is easy to understand now.
1922  */
1923 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1924                                   struct ulist *qgroups,
1925                                   u64 nr_old_roots,
1926                                   u64 nr_new_roots,
1927                                   u64 num_bytes, u64 seq)
1928 {
1929         struct ulist_node *unode;
1930         struct ulist_iterator uiter;
1931         struct btrfs_qgroup *qg;
1932         u64 cur_new_count, cur_old_count;
1933
1934         ULIST_ITER_INIT(&uiter);
1935         while ((unode = ulist_next(qgroups, &uiter))) {
1936                 bool dirty = false;
1937
1938                 qg = unode_aux_to_qgroup(unode);
1939                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1940                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1941
1942                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
1943                                              cur_new_count);
1944
1945                 /* Rfer update part */
1946                 if (cur_old_count == 0 && cur_new_count > 0) {
1947                         qg->rfer += num_bytes;
1948                         qg->rfer_cmpr += num_bytes;
1949                         dirty = true;
1950                 }
1951                 if (cur_old_count > 0 && cur_new_count == 0) {
1952                         qg->rfer -= num_bytes;
1953                         qg->rfer_cmpr -= num_bytes;
1954                         dirty = true;
1955                 }
1956
1957                 /* Excl update part */
1958                 /* Exclusive/none -> shared case */
1959                 if (cur_old_count == nr_old_roots &&
1960                     cur_new_count < nr_new_roots) {
1961                         /* Exclusive -> shared */
1962                         if (cur_old_count != 0) {
1963                                 qg->excl -= num_bytes;
1964                                 qg->excl_cmpr -= num_bytes;
1965                                 dirty = true;
1966                         }
1967                 }
1968
1969                 /* Shared -> exclusive/none case */
1970                 if (cur_old_count < nr_old_roots &&
1971                     cur_new_count == nr_new_roots) {
1972                         /* Shared->exclusive */
1973                         if (cur_new_count != 0) {
1974                                 qg->excl += num_bytes;
1975                                 qg->excl_cmpr += num_bytes;
1976                                 dirty = true;
1977                         }
1978                 }
1979
1980                 /* Exclusive/none -> exclusive/none case */
1981                 if (cur_old_count == nr_old_roots &&
1982                     cur_new_count == nr_new_roots) {
1983                         if (cur_old_count == 0) {
1984                                 /* None -> exclusive/none */
1985
1986                                 if (cur_new_count != 0) {
1987                                         /* None -> exclusive */
1988                                         qg->excl += num_bytes;
1989                                         qg->excl_cmpr += num_bytes;
1990                                         dirty = true;
1991                                 }
1992                                 /* None -> none, nothing changed */
1993                         } else {
1994                                 /* Exclusive -> exclusive/none */
1995
1996                                 if (cur_new_count == 0) {
1997                                         /* Exclusive -> none */
1998                                         qg->excl -= num_bytes;
1999                                         qg->excl_cmpr -= num_bytes;
2000                                         dirty = true;
2001                                 }
2002                                 /* Exclusive -> exclusive, nothing changed */
2003                         }
2004                 }
2005
2006                 if (dirty)
2007                         qgroup_dirty(fs_info, qg);
2008         }
2009         return 0;
2010 }
2011
2012 /*
2013  * Check if the @roots potentially is a list of fs tree roots
2014  *
2015  * Return 0 for definitely not a fs/subvol tree roots ulist
2016  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2017  *          one as well)
2018  */
2019 static int maybe_fs_roots(struct ulist *roots)
2020 {
2021         struct ulist_node *unode;
2022         struct ulist_iterator uiter;
2023
2024         /* Empty one, still possible for fs roots */
2025         if (!roots || roots->nnodes == 0)
2026                 return 1;
2027
2028         ULIST_ITER_INIT(&uiter);
2029         unode = ulist_next(roots, &uiter);
2030         if (!unode)
2031                 return 1;
2032
2033         /*
2034          * If it contains fs tree roots, then it must belong to fs/subvol
2035          * trees.
2036          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2037          */
2038         return is_fstree(unode->val);
2039 }
2040
2041 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2042                                 u64 num_bytes, struct ulist *old_roots,
2043                                 struct ulist *new_roots)
2044 {
2045         struct btrfs_fs_info *fs_info = trans->fs_info;
2046         struct ulist *qgroups = NULL;
2047         struct ulist *tmp = NULL;
2048         u64 seq;
2049         u64 nr_new_roots = 0;
2050         u64 nr_old_roots = 0;
2051         int ret = 0;
2052
2053         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2054                 return 0;
2055
2056         if (new_roots) {
2057                 if (!maybe_fs_roots(new_roots))
2058                         goto out_free;
2059                 nr_new_roots = new_roots->nnodes;
2060         }
2061         if (old_roots) {
2062                 if (!maybe_fs_roots(old_roots))
2063                         goto out_free;
2064                 nr_old_roots = old_roots->nnodes;
2065         }
2066
2067         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2068         if (nr_old_roots == 0 && nr_new_roots == 0)
2069                 goto out_free;
2070
2071         BUG_ON(!fs_info->quota_root);
2072
2073         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2074                                         num_bytes, nr_old_roots, nr_new_roots);
2075
2076         qgroups = ulist_alloc(GFP_NOFS);
2077         if (!qgroups) {
2078                 ret = -ENOMEM;
2079                 goto out_free;
2080         }
2081         tmp = ulist_alloc(GFP_NOFS);
2082         if (!tmp) {
2083                 ret = -ENOMEM;
2084                 goto out_free;
2085         }
2086
2087         mutex_lock(&fs_info->qgroup_rescan_lock);
2088         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2089                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2090                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2091                         ret = 0;
2092                         goto out_free;
2093                 }
2094         }
2095         mutex_unlock(&fs_info->qgroup_rescan_lock);
2096
2097         spin_lock(&fs_info->qgroup_lock);
2098         seq = fs_info->qgroup_seq;
2099
2100         /* Update old refcnts using old_roots */
2101         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2102                                    UPDATE_OLD);
2103         if (ret < 0)
2104                 goto out;
2105
2106         /* Update new refcnts using new_roots */
2107         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2108                                    UPDATE_NEW);
2109         if (ret < 0)
2110                 goto out;
2111
2112         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2113                                num_bytes, seq);
2114
2115         /*
2116          * Bump qgroup_seq to avoid seq overlap
2117          */
2118         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2119 out:
2120         spin_unlock(&fs_info->qgroup_lock);
2121 out_free:
2122         ulist_free(tmp);
2123         ulist_free(qgroups);
2124         ulist_free(old_roots);
2125         ulist_free(new_roots);
2126         return ret;
2127 }
2128
2129 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2130 {
2131         struct btrfs_fs_info *fs_info = trans->fs_info;
2132         struct btrfs_qgroup_extent_record *record;
2133         struct btrfs_delayed_ref_root *delayed_refs;
2134         struct ulist *new_roots = NULL;
2135         struct rb_node *node;
2136         u64 qgroup_to_skip;
2137         int ret = 0;
2138
2139         delayed_refs = &trans->transaction->delayed_refs;
2140         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2141         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2142                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2143                                   node);
2144
2145                 trace_btrfs_qgroup_account_extents(fs_info, record);
2146
2147                 if (!ret) {
2148                         /*
2149                          * Old roots should be searched when inserting qgroup
2150                          * extent record
2151                          */
2152                         if (WARN_ON(!record->old_roots)) {
2153                                 /* Search commit root to find old_roots */
2154                                 ret = btrfs_find_all_roots(NULL, fs_info,
2155                                                 record->bytenr, 0,
2156                                                 &record->old_roots, false);
2157                                 if (ret < 0)
2158                                         goto cleanup;
2159                         }
2160
2161                         /*
2162                          * Use SEQ_LAST as time_seq to do special search, which
2163                          * doesn't lock tree or delayed_refs and search current
2164                          * root. It's safe inside commit_transaction().
2165                          */
2166                         ret = btrfs_find_all_roots(trans, fs_info,
2167                                 record->bytenr, SEQ_LAST, &new_roots, false);
2168                         if (ret < 0)
2169                                 goto cleanup;
2170                         if (qgroup_to_skip) {
2171                                 ulist_del(new_roots, qgroup_to_skip, 0);
2172                                 ulist_del(record->old_roots, qgroup_to_skip,
2173                                           0);
2174                         }
2175                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2176                                                           record->num_bytes,
2177                                                           record->old_roots,
2178                                                           new_roots);
2179                         record->old_roots = NULL;
2180                         new_roots = NULL;
2181                 }
2182 cleanup:
2183                 ulist_free(record->old_roots);
2184                 ulist_free(new_roots);
2185                 new_roots = NULL;
2186                 rb_erase(node, &delayed_refs->dirty_extent_root);
2187                 kfree(record);
2188
2189         }
2190         return ret;
2191 }
2192
2193 /*
2194  * called from commit_transaction. Writes all changed qgroups to disk.
2195  */
2196 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2197 {
2198         struct btrfs_fs_info *fs_info = trans->fs_info;
2199         struct btrfs_root *quota_root = fs_info->quota_root;
2200         int ret = 0;
2201
2202         if (!quota_root)
2203                 return ret;
2204
2205         spin_lock(&fs_info->qgroup_lock);
2206         while (!list_empty(&fs_info->dirty_qgroups)) {
2207                 struct btrfs_qgroup *qgroup;
2208                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2209                                           struct btrfs_qgroup, dirty);
2210                 list_del_init(&qgroup->dirty);
2211                 spin_unlock(&fs_info->qgroup_lock);
2212                 ret = update_qgroup_info_item(trans, qgroup);
2213                 if (ret)
2214                         fs_info->qgroup_flags |=
2215                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2216                 ret = update_qgroup_limit_item(trans, qgroup);
2217                 if (ret)
2218                         fs_info->qgroup_flags |=
2219                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2220                 spin_lock(&fs_info->qgroup_lock);
2221         }
2222         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2223                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2224         else
2225                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2226         spin_unlock(&fs_info->qgroup_lock);
2227
2228         ret = update_qgroup_status_item(trans);
2229         if (ret)
2230                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2231
2232         return ret;
2233 }
2234
2235 /*
2236  * Copy the accounting information between qgroups. This is necessary
2237  * when a snapshot or a subvolume is created. Throwing an error will
2238  * cause a transaction abort so we take extra care here to only error
2239  * when a readonly fs is a reasonable outcome.
2240  */
2241 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2242                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2243 {
2244         int ret = 0;
2245         int i;
2246         u64 *i_qgroups;
2247         struct btrfs_fs_info *fs_info = trans->fs_info;
2248         struct btrfs_root *quota_root = fs_info->quota_root;
2249         struct btrfs_qgroup *srcgroup;
2250         struct btrfs_qgroup *dstgroup;
2251         u32 level_size = 0;
2252         u64 nums;
2253
2254         mutex_lock(&fs_info->qgroup_ioctl_lock);
2255         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2256                 goto out;
2257
2258         if (!quota_root) {
2259                 ret = -EINVAL;
2260                 goto out;
2261         }
2262
2263         if (inherit) {
2264                 i_qgroups = (u64 *)(inherit + 1);
2265                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2266                        2 * inherit->num_excl_copies;
2267                 for (i = 0; i < nums; ++i) {
2268                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2269
2270                         /*
2271                          * Zero out invalid groups so we can ignore
2272                          * them later.
2273                          */
2274                         if (!srcgroup ||
2275                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2276                                 *i_qgroups = 0ULL;
2277
2278                         ++i_qgroups;
2279                 }
2280         }
2281
2282         /*
2283          * create a tracking group for the subvol itself
2284          */
2285         ret = add_qgroup_item(trans, quota_root, objectid);
2286         if (ret)
2287                 goto out;
2288
2289         /*
2290          * add qgroup to all inherited groups
2291          */
2292         if (inherit) {
2293                 i_qgroups = (u64 *)(inherit + 1);
2294                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2295                         if (*i_qgroups == 0)
2296                                 continue;
2297                         ret = add_qgroup_relation_item(trans, objectid,
2298                                                        *i_qgroups);
2299                         if (ret && ret != -EEXIST)
2300                                 goto out;
2301                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2302                                                        objectid);
2303                         if (ret && ret != -EEXIST)
2304                                 goto out;
2305                 }
2306                 ret = 0;
2307         }
2308
2309
2310         spin_lock(&fs_info->qgroup_lock);
2311
2312         dstgroup = add_qgroup_rb(fs_info, objectid);
2313         if (IS_ERR(dstgroup)) {
2314                 ret = PTR_ERR(dstgroup);
2315                 goto unlock;
2316         }
2317
2318         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2319                 dstgroup->lim_flags = inherit->lim.flags;
2320                 dstgroup->max_rfer = inherit->lim.max_rfer;
2321                 dstgroup->max_excl = inherit->lim.max_excl;
2322                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2323                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2324
2325                 ret = update_qgroup_limit_item(trans, dstgroup);
2326                 if (ret) {
2327                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2328                         btrfs_info(fs_info,
2329                                    "unable to update quota limit for %llu",
2330                                    dstgroup->qgroupid);
2331                         goto unlock;
2332                 }
2333         }
2334
2335         if (srcid) {
2336                 srcgroup = find_qgroup_rb(fs_info, srcid);
2337                 if (!srcgroup)
2338                         goto unlock;
2339
2340                 /*
2341                  * We call inherit after we clone the root in order to make sure
2342                  * our counts don't go crazy, so at this point the only
2343                  * difference between the two roots should be the root node.
2344                  */
2345                 level_size = fs_info->nodesize;
2346                 dstgroup->rfer = srcgroup->rfer;
2347                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2348                 dstgroup->excl = level_size;
2349                 dstgroup->excl_cmpr = level_size;
2350                 srcgroup->excl = level_size;
2351                 srcgroup->excl_cmpr = level_size;
2352
2353                 /* inherit the limit info */
2354                 dstgroup->lim_flags = srcgroup->lim_flags;
2355                 dstgroup->max_rfer = srcgroup->max_rfer;
2356                 dstgroup->max_excl = srcgroup->max_excl;
2357                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2358                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2359
2360                 qgroup_dirty(fs_info, dstgroup);
2361                 qgroup_dirty(fs_info, srcgroup);
2362         }
2363
2364         if (!inherit)
2365                 goto unlock;
2366
2367         i_qgroups = (u64 *)(inherit + 1);
2368         for (i = 0; i < inherit->num_qgroups; ++i) {
2369                 if (*i_qgroups) {
2370                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2371                         if (ret)
2372                                 goto unlock;
2373                 }
2374                 ++i_qgroups;
2375         }
2376
2377         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2378                 struct btrfs_qgroup *src;
2379                 struct btrfs_qgroup *dst;
2380
2381                 if (!i_qgroups[0] || !i_qgroups[1])
2382                         continue;
2383
2384                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2385                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2386
2387                 if (!src || !dst) {
2388                         ret = -EINVAL;
2389                         goto unlock;
2390                 }
2391
2392                 dst->rfer = src->rfer - level_size;
2393                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2394         }
2395         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2396                 struct btrfs_qgroup *src;
2397                 struct btrfs_qgroup *dst;
2398
2399                 if (!i_qgroups[0] || !i_qgroups[1])
2400                         continue;
2401
2402                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2403                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2404
2405                 if (!src || !dst) {
2406                         ret = -EINVAL;
2407                         goto unlock;
2408                 }
2409
2410                 dst->excl = src->excl + level_size;
2411                 dst->excl_cmpr = src->excl_cmpr + level_size;
2412         }
2413
2414 unlock:
2415         spin_unlock(&fs_info->qgroup_lock);
2416 out:
2417         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2418         return ret;
2419 }
2420
2421 /*
2422  * Two limits to commit transaction in advance.
2423  *
2424  * For RATIO, it will be 1/RATIO of the remaining limit
2425  * (excluding data and prealloc meta) as threshold.
2426  * For SIZE, it will be in byte unit as threshold.
2427  */
2428 #define QGROUP_PERTRANS_RATIO           32
2429 #define QGROUP_PERTRANS_SIZE            SZ_32M
2430 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2431                                 const struct btrfs_qgroup *qg, u64 num_bytes)
2432 {
2433         u64 limit;
2434         u64 threshold;
2435
2436         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2437             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2438                 return false;
2439
2440         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2441             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2442                 return false;
2443
2444         /*
2445          * Even if we passed the check, it's better to check if reservation
2446          * for meta_pertrans is pushing us near limit.
2447          * If there is too much pertrans reservation or it's near the limit,
2448          * let's try commit transaction to free some, using transaction_kthread
2449          */
2450         if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2451                               BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2452                 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2453                         limit = qg->max_excl;
2454                 else
2455                         limit = qg->max_rfer;
2456                 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2457                             qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2458                             QGROUP_PERTRANS_RATIO;
2459                 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2460
2461                 /*
2462                  * Use transaction_kthread to commit transaction, so we no
2463                  * longer need to bother nested transaction nor lock context.
2464                  */
2465                 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2466                         btrfs_commit_transaction_locksafe(fs_info);
2467         }
2468
2469         return true;
2470 }
2471
2472 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2473                           enum btrfs_qgroup_rsv_type type)
2474 {
2475         struct btrfs_root *quota_root;
2476         struct btrfs_qgroup *qgroup;
2477         struct btrfs_fs_info *fs_info = root->fs_info;
2478         u64 ref_root = root->root_key.objectid;
2479         int ret = 0;
2480         struct ulist_node *unode;
2481         struct ulist_iterator uiter;
2482
2483         if (!is_fstree(ref_root))
2484                 return 0;
2485
2486         if (num_bytes == 0)
2487                 return 0;
2488
2489         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2490             capable(CAP_SYS_RESOURCE))
2491                 enforce = false;
2492
2493         spin_lock(&fs_info->qgroup_lock);
2494         quota_root = fs_info->quota_root;
2495         if (!quota_root)
2496                 goto out;
2497
2498         qgroup = find_qgroup_rb(fs_info, ref_root);
2499         if (!qgroup)
2500                 goto out;
2501
2502         /*
2503          * in a first step, we check all affected qgroups if any limits would
2504          * be exceeded
2505          */
2506         ulist_reinit(fs_info->qgroup_ulist);
2507         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2508                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2509         if (ret < 0)
2510                 goto out;
2511         ULIST_ITER_INIT(&uiter);
2512         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2513                 struct btrfs_qgroup *qg;
2514                 struct btrfs_qgroup_list *glist;
2515
2516                 qg = unode_aux_to_qgroup(unode);
2517
2518                 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2519                         ret = -EDQUOT;
2520                         goto out;
2521                 }
2522
2523                 list_for_each_entry(glist, &qg->groups, next_group) {
2524                         ret = ulist_add(fs_info->qgroup_ulist,
2525                                         glist->group->qgroupid,
2526                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2527                         if (ret < 0)
2528                                 goto out;
2529                 }
2530         }
2531         ret = 0;
2532         /*
2533          * no limits exceeded, now record the reservation into all qgroups
2534          */
2535         ULIST_ITER_INIT(&uiter);
2536         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2537                 struct btrfs_qgroup *qg;
2538
2539                 qg = unode_aux_to_qgroup(unode);
2540
2541                 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2542                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2543         }
2544
2545 out:
2546         spin_unlock(&fs_info->qgroup_lock);
2547         return ret;
2548 }
2549
2550 /*
2551  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2552  * qgroup).
2553  *
2554  * Will handle all higher level qgroup too.
2555  *
2556  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2557  * This special case is only used for META_PERTRANS type.
2558  */
2559 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2560                                u64 ref_root, u64 num_bytes,
2561                                enum btrfs_qgroup_rsv_type type)
2562 {
2563         struct btrfs_root *quota_root;
2564         struct btrfs_qgroup *qgroup;
2565         struct ulist_node *unode;
2566         struct ulist_iterator uiter;
2567         int ret = 0;
2568
2569         if (!is_fstree(ref_root))
2570                 return;
2571
2572         if (num_bytes == 0)
2573                 return;
2574
2575         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2576                 WARN(1, "%s: Invalid type to free", __func__);
2577                 return;
2578         }
2579         spin_lock(&fs_info->qgroup_lock);
2580
2581         quota_root = fs_info->quota_root;
2582         if (!quota_root)
2583                 goto out;
2584
2585         qgroup = find_qgroup_rb(fs_info, ref_root);
2586         if (!qgroup)
2587                 goto out;
2588
2589         if (num_bytes == (u64)-1)
2590                 /*
2591                  * We're freeing all pertrans rsv, get reserved value from
2592                  * level 0 qgroup as real num_bytes to free.
2593                  */
2594                 num_bytes = qgroup->rsv.values[type];
2595
2596         ulist_reinit(fs_info->qgroup_ulist);
2597         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2598                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2599         if (ret < 0)
2600                 goto out;
2601         ULIST_ITER_INIT(&uiter);
2602         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2603                 struct btrfs_qgroup *qg;
2604                 struct btrfs_qgroup_list *glist;
2605
2606                 qg = unode_aux_to_qgroup(unode);
2607
2608                 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2609                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2610
2611                 list_for_each_entry(glist, &qg->groups, next_group) {
2612                         ret = ulist_add(fs_info->qgroup_ulist,
2613                                         glist->group->qgroupid,
2614                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2615                         if (ret < 0)
2616                                 goto out;
2617                 }
2618         }
2619
2620 out:
2621         spin_unlock(&fs_info->qgroup_lock);
2622 }
2623
2624 /*
2625  * Check if the leaf is the last leaf. Which means all node pointers
2626  * are at their last position.
2627  */
2628 static bool is_last_leaf(struct btrfs_path *path)
2629 {
2630         int i;
2631
2632         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2633                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2634                         return false;
2635         }
2636         return true;
2637 }
2638
2639 /*
2640  * returns < 0 on error, 0 when more leafs are to be scanned.
2641  * returns 1 when done.
2642  */
2643 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
2644                               struct btrfs_path *path)
2645 {
2646         struct btrfs_fs_info *fs_info = trans->fs_info;
2647         struct btrfs_key found;
2648         struct extent_buffer *scratch_leaf = NULL;
2649         struct ulist *roots = NULL;
2650         u64 num_bytes;
2651         bool done;
2652         int slot;
2653         int ret;
2654
2655         mutex_lock(&fs_info->qgroup_rescan_lock);
2656         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2657                                          &fs_info->qgroup_rescan_progress,
2658                                          path, 1, 0);
2659
2660         btrfs_debug(fs_info,
2661                 "current progress key (%llu %u %llu), search_slot ret %d",
2662                 fs_info->qgroup_rescan_progress.objectid,
2663                 fs_info->qgroup_rescan_progress.type,
2664                 fs_info->qgroup_rescan_progress.offset, ret);
2665
2666         if (ret) {
2667                 /*
2668                  * The rescan is about to end, we will not be scanning any
2669                  * further blocks. We cannot unset the RESCAN flag here, because
2670                  * we want to commit the transaction if everything went well.
2671                  * To make the live accounting work in this phase, we set our
2672                  * scan progress pointer such that every real extent objectid
2673                  * will be smaller.
2674                  */
2675                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2676                 btrfs_release_path(path);
2677                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2678                 return ret;
2679         }
2680         done = is_last_leaf(path);
2681
2682         btrfs_item_key_to_cpu(path->nodes[0], &found,
2683                               btrfs_header_nritems(path->nodes[0]) - 1);
2684         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2685
2686         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2687         if (!scratch_leaf) {
2688                 ret = -ENOMEM;
2689                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2690                 goto out;
2691         }
2692         extent_buffer_get(scratch_leaf);
2693         btrfs_tree_read_lock(scratch_leaf);
2694         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2695         slot = path->slots[0];
2696         btrfs_release_path(path);
2697         mutex_unlock(&fs_info->qgroup_rescan_lock);
2698
2699         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2700                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2701                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2702                     found.type != BTRFS_METADATA_ITEM_KEY)
2703                         continue;
2704                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2705                         num_bytes = fs_info->nodesize;
2706                 else
2707                         num_bytes = found.offset;
2708
2709                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2710                                            &roots, false);
2711                 if (ret < 0)
2712                         goto out;
2713                 /* For rescan, just pass old_roots as NULL */
2714                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
2715                                                   num_bytes, NULL, roots);
2716                 if (ret < 0)
2717                         goto out;
2718         }
2719 out:
2720         if (scratch_leaf) {
2721                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2722                 free_extent_buffer(scratch_leaf);
2723         }
2724
2725         if (done && !ret) {
2726                 ret = 1;
2727                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2728         }
2729         return ret;
2730 }
2731
2732 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2733 {
2734         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2735                                                      qgroup_rescan_work);
2736         struct btrfs_path *path;
2737         struct btrfs_trans_handle *trans = NULL;
2738         int err = -ENOMEM;
2739         int ret = 0;
2740
2741         path = btrfs_alloc_path();
2742         if (!path)
2743                 goto out;
2744         /*
2745          * Rescan should only search for commit root, and any later difference
2746          * should be recorded by qgroup
2747          */
2748         path->search_commit_root = 1;
2749         path->skip_locking = 1;
2750
2751         err = 0;
2752         while (!err && !btrfs_fs_closing(fs_info)) {
2753                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2754                 if (IS_ERR(trans)) {
2755                         err = PTR_ERR(trans);
2756                         break;
2757                 }
2758                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2759                         err = -EINTR;
2760                 } else {
2761                         err = qgroup_rescan_leaf(trans, path);
2762                 }
2763                 if (err > 0)
2764                         btrfs_commit_transaction(trans);
2765                 else
2766                         btrfs_end_transaction(trans);
2767         }
2768
2769 out:
2770         btrfs_free_path(path);
2771
2772         mutex_lock(&fs_info->qgroup_rescan_lock);
2773         if (!btrfs_fs_closing(fs_info))
2774                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2775
2776         if (err > 0 &&
2777             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2778                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2779         } else if (err < 0) {
2780                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2781         }
2782         mutex_unlock(&fs_info->qgroup_rescan_lock);
2783
2784         /*
2785          * only update status, since the previous part has already updated the
2786          * qgroup info.
2787          */
2788         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2789         if (IS_ERR(trans)) {
2790                 err = PTR_ERR(trans);
2791                 btrfs_err(fs_info,
2792                           "fail to start transaction for status update: %d",
2793                           err);
2794                 goto done;
2795         }
2796         ret = update_qgroup_status_item(trans);
2797         if (ret < 0) {
2798                 err = ret;
2799                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2800         }
2801         btrfs_end_transaction(trans);
2802
2803         if (btrfs_fs_closing(fs_info)) {
2804                 btrfs_info(fs_info, "qgroup scan paused");
2805         } else if (err >= 0) {
2806                 btrfs_info(fs_info, "qgroup scan completed%s",
2807                         err > 0 ? " (inconsistency flag cleared)" : "");
2808         } else {
2809                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2810         }
2811
2812 done:
2813         mutex_lock(&fs_info->qgroup_rescan_lock);
2814         fs_info->qgroup_rescan_running = false;
2815         mutex_unlock(&fs_info->qgroup_rescan_lock);
2816         complete_all(&fs_info->qgroup_rescan_completion);
2817 }
2818
2819 /*
2820  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2821  * memory required for the rescan context.
2822  */
2823 static int
2824 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2825                    int init_flags)
2826 {
2827         int ret = 0;
2828
2829         if (!init_flags) {
2830                 /* we're resuming qgroup rescan at mount time */
2831                 if (!(fs_info->qgroup_flags &
2832                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
2833                         btrfs_warn(fs_info,
2834                         "qgroup rescan init failed, qgroup is not enabled");
2835                         ret = -EINVAL;
2836                 } else if (!(fs_info->qgroup_flags &
2837                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
2838                         btrfs_warn(fs_info,
2839                         "qgroup rescan init failed, qgroup rescan is not queued");
2840                         ret = -EINVAL;
2841                 }
2842
2843                 if (ret)
2844                         return ret;
2845         }
2846
2847         mutex_lock(&fs_info->qgroup_rescan_lock);
2848         spin_lock(&fs_info->qgroup_lock);
2849
2850         if (init_flags) {
2851                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2852                         btrfs_warn(fs_info,
2853                                    "qgroup rescan is already in progress");
2854                         ret = -EINPROGRESS;
2855                 } else if (!(fs_info->qgroup_flags &
2856                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
2857                         btrfs_warn(fs_info,
2858                         "qgroup rescan init failed, qgroup is not enabled");
2859                         ret = -EINVAL;
2860                 }
2861
2862                 if (ret) {
2863                         spin_unlock(&fs_info->qgroup_lock);
2864                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2865                         return ret;
2866                 }
2867                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2868         }
2869
2870         memset(&fs_info->qgroup_rescan_progress, 0,
2871                 sizeof(fs_info->qgroup_rescan_progress));
2872         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2873         init_completion(&fs_info->qgroup_rescan_completion);
2874         fs_info->qgroup_rescan_running = true;
2875
2876         spin_unlock(&fs_info->qgroup_lock);
2877         mutex_unlock(&fs_info->qgroup_rescan_lock);
2878
2879         memset(&fs_info->qgroup_rescan_work, 0,
2880                sizeof(fs_info->qgroup_rescan_work));
2881         btrfs_init_work(&fs_info->qgroup_rescan_work,
2882                         btrfs_qgroup_rescan_helper,
2883                         btrfs_qgroup_rescan_worker, NULL, NULL);
2884         return 0;
2885 }
2886
2887 static void
2888 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2889 {
2890         struct rb_node *n;
2891         struct btrfs_qgroup *qgroup;
2892
2893         spin_lock(&fs_info->qgroup_lock);
2894         /* clear all current qgroup tracking information */
2895         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2896                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2897                 qgroup->rfer = 0;
2898                 qgroup->rfer_cmpr = 0;
2899                 qgroup->excl = 0;
2900                 qgroup->excl_cmpr = 0;
2901         }
2902         spin_unlock(&fs_info->qgroup_lock);
2903 }
2904
2905 int
2906 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2907 {
2908         int ret = 0;
2909         struct btrfs_trans_handle *trans;
2910
2911         ret = qgroup_rescan_init(fs_info, 0, 1);
2912         if (ret)
2913                 return ret;
2914
2915         /*
2916          * We have set the rescan_progress to 0, which means no more
2917          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2918          * However, btrfs_qgroup_account_ref may be right after its call
2919          * to btrfs_find_all_roots, in which case it would still do the
2920          * accounting.
2921          * To solve this, we're committing the transaction, which will
2922          * ensure we run all delayed refs and only after that, we are
2923          * going to clear all tracking information for a clean start.
2924          */
2925
2926         trans = btrfs_join_transaction(fs_info->fs_root);
2927         if (IS_ERR(trans)) {
2928                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2929                 return PTR_ERR(trans);
2930         }
2931         ret = btrfs_commit_transaction(trans);
2932         if (ret) {
2933                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2934                 return ret;
2935         }
2936
2937         qgroup_rescan_zero_tracking(fs_info);
2938
2939         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2940                          &fs_info->qgroup_rescan_work);
2941
2942         return 0;
2943 }
2944
2945 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2946                                      bool interruptible)
2947 {
2948         int running;
2949         int ret = 0;
2950
2951         mutex_lock(&fs_info->qgroup_rescan_lock);
2952         spin_lock(&fs_info->qgroup_lock);
2953         running = fs_info->qgroup_rescan_running;
2954         spin_unlock(&fs_info->qgroup_lock);
2955         mutex_unlock(&fs_info->qgroup_rescan_lock);
2956
2957         if (!running)
2958                 return 0;
2959
2960         if (interruptible)
2961                 ret = wait_for_completion_interruptible(
2962                                         &fs_info->qgroup_rescan_completion);
2963         else
2964                 wait_for_completion(&fs_info->qgroup_rescan_completion);
2965
2966         return ret;
2967 }
2968
2969 /*
2970  * this is only called from open_ctree where we're still single threaded, thus
2971  * locking is omitted here.
2972  */
2973 void
2974 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2975 {
2976         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2977                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2978                                  &fs_info->qgroup_rescan_work);
2979 }
2980
2981 /*
2982  * Reserve qgroup space for range [start, start + len).
2983  *
2984  * This function will either reserve space from related qgroups or doing
2985  * nothing if the range is already reserved.
2986  *
2987  * Return 0 for successful reserve
2988  * Return <0 for error (including -EQUOT)
2989  *
2990  * NOTE: this function may sleep for memory allocation.
2991  *       if btrfs_qgroup_reserve_data() is called multiple times with
2992  *       same @reserved, caller must ensure when error happens it's OK
2993  *       to free *ALL* reserved space.
2994  */
2995 int btrfs_qgroup_reserve_data(struct inode *inode,
2996                         struct extent_changeset **reserved_ret, u64 start,
2997                         u64 len)
2998 {
2999         struct btrfs_root *root = BTRFS_I(inode)->root;
3000         struct ulist_node *unode;
3001         struct ulist_iterator uiter;
3002         struct extent_changeset *reserved;
3003         u64 orig_reserved;
3004         u64 to_reserve;
3005         int ret;
3006
3007         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3008             !is_fstree(root->objectid) || len == 0)
3009                 return 0;
3010
3011         /* @reserved parameter is mandatory for qgroup */
3012         if (WARN_ON(!reserved_ret))
3013                 return -EINVAL;
3014         if (!*reserved_ret) {
3015                 *reserved_ret = extent_changeset_alloc();
3016                 if (!*reserved_ret)
3017                         return -ENOMEM;
3018         }
3019         reserved = *reserved_ret;
3020         /* Record already reserved space */
3021         orig_reserved = reserved->bytes_changed;
3022         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3023                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3024
3025         /* Newly reserved space */
3026         to_reserve = reserved->bytes_changed - orig_reserved;
3027         trace_btrfs_qgroup_reserve_data(inode, start, len,
3028                                         to_reserve, QGROUP_RESERVE);
3029         if (ret < 0)
3030                 goto cleanup;
3031         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3032         if (ret < 0)
3033                 goto cleanup;
3034
3035         return ret;
3036
3037 cleanup:
3038         /* cleanup *ALL* already reserved ranges */
3039         ULIST_ITER_INIT(&uiter);
3040         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3041                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3042                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3043         extent_changeset_release(reserved);
3044         return ret;
3045 }
3046
3047 /* Free ranges specified by @reserved, normally in error path */
3048 static int qgroup_free_reserved_data(struct inode *inode,
3049                         struct extent_changeset *reserved, u64 start, u64 len)
3050 {
3051         struct btrfs_root *root = BTRFS_I(inode)->root;
3052         struct ulist_node *unode;
3053         struct ulist_iterator uiter;
3054         struct extent_changeset changeset;
3055         int freed = 0;
3056         int ret;
3057
3058         extent_changeset_init(&changeset);
3059         len = round_up(start + len, root->fs_info->sectorsize);
3060         start = round_down(start, root->fs_info->sectorsize);
3061
3062         ULIST_ITER_INIT(&uiter);
3063         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3064                 u64 range_start = unode->val;
3065                 /* unode->aux is the inclusive end */
3066                 u64 range_len = unode->aux - range_start + 1;
3067                 u64 free_start;
3068                 u64 free_len;
3069
3070                 extent_changeset_release(&changeset);
3071
3072                 /* Only free range in range [start, start + len) */
3073                 if (range_start >= start + len ||
3074                     range_start + range_len <= start)
3075                         continue;
3076                 free_start = max(range_start, start);
3077                 free_len = min(start + len, range_start + range_len) -
3078                            free_start;
3079                 /*
3080                  * TODO: To also modify reserved->ranges_reserved to reflect
3081                  * the modification.
3082                  *
3083                  * However as long as we free qgroup reserved according to
3084                  * EXTENT_QGROUP_RESERVED, we won't double free.
3085                  * So not need to rush.
3086                  */
3087                 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3088                                 free_start, free_start + free_len - 1,
3089                                 EXTENT_QGROUP_RESERVED, &changeset);
3090                 if (ret < 0)
3091                         goto out;
3092                 freed += changeset.bytes_changed;
3093         }
3094         btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed,
3095                                   BTRFS_QGROUP_RSV_DATA);
3096         ret = freed;
3097 out:
3098         extent_changeset_release(&changeset);
3099         return ret;
3100 }
3101
3102 static int __btrfs_qgroup_release_data(struct inode *inode,
3103                         struct extent_changeset *reserved, u64 start, u64 len,
3104                         int free)
3105 {
3106         struct extent_changeset changeset;
3107         int trace_op = QGROUP_RELEASE;
3108         int ret;
3109
3110         /* In release case, we shouldn't have @reserved */
3111         WARN_ON(!free && reserved);
3112         if (free && reserved)
3113                 return qgroup_free_reserved_data(inode, reserved, start, len);
3114         extent_changeset_init(&changeset);
3115         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3116                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3117         if (ret < 0)
3118                 goto out;
3119
3120         if (free)
3121                 trace_op = QGROUP_FREE;
3122         trace_btrfs_qgroup_release_data(inode, start, len,
3123                                         changeset.bytes_changed, trace_op);
3124         if (free)
3125                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3126                                 BTRFS_I(inode)->root->objectid,
3127                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3128         ret = changeset.bytes_changed;
3129 out:
3130         extent_changeset_release(&changeset);
3131         return ret;
3132 }
3133
3134 /*
3135  * Free a reserved space range from io_tree and related qgroups
3136  *
3137  * Should be called when a range of pages get invalidated before reaching disk.
3138  * Or for error cleanup case.
3139  * if @reserved is given, only reserved range in [@start, @start + @len) will
3140  * be freed.
3141  *
3142  * For data written to disk, use btrfs_qgroup_release_data().
3143  *
3144  * NOTE: This function may sleep for memory allocation.
3145  */
3146 int btrfs_qgroup_free_data(struct inode *inode,
3147                         struct extent_changeset *reserved, u64 start, u64 len)
3148 {
3149         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3150 }
3151
3152 /*
3153  * Release a reserved space range from io_tree only.
3154  *
3155  * Should be called when a range of pages get written to disk and corresponding
3156  * FILE_EXTENT is inserted into corresponding root.
3157  *
3158  * Since new qgroup accounting framework will only update qgroup numbers at
3159  * commit_transaction() time, its reserved space shouldn't be freed from
3160  * related qgroups.
3161  *
3162  * But we should release the range from io_tree, to allow further write to be
3163  * COWed.
3164  *
3165  * NOTE: This function may sleep for memory allocation.
3166  */
3167 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3168 {
3169         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3170 }
3171
3172 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3173                               enum btrfs_qgroup_rsv_type type)
3174 {
3175         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3176             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3177                 return;
3178         if (num_bytes == 0)
3179                 return;
3180
3181         spin_lock(&root->qgroup_meta_rsv_lock);
3182         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3183                 root->qgroup_meta_rsv_prealloc += num_bytes;
3184         else
3185                 root->qgroup_meta_rsv_pertrans += num_bytes;
3186         spin_unlock(&root->qgroup_meta_rsv_lock);
3187 }
3188
3189 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3190                              enum btrfs_qgroup_rsv_type type)
3191 {
3192         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3193             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3194                 return 0;
3195         if (num_bytes == 0)
3196                 return 0;
3197
3198         spin_lock(&root->qgroup_meta_rsv_lock);
3199         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3200                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3201                                   num_bytes);
3202                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3203         } else {
3204                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3205                                   num_bytes);
3206                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3207         }
3208         spin_unlock(&root->qgroup_meta_rsv_lock);
3209         return num_bytes;
3210 }
3211
3212 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3213                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3214 {
3215         struct btrfs_fs_info *fs_info = root->fs_info;
3216         int ret;
3217
3218         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3219             !is_fstree(root->objectid) || num_bytes == 0)
3220                 return 0;
3221
3222         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3223         trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3224         ret = qgroup_reserve(root, num_bytes, enforce, type);
3225         if (ret < 0)
3226                 return ret;
3227         /*
3228          * Record what we have reserved into root.
3229          *
3230          * To avoid quota disabled->enabled underflow.
3231          * In that case, we may try to free space we haven't reserved
3232          * (since quota was disabled), so record what we reserved into root.
3233          * And ensure later release won't underflow this number.
3234          */
3235         add_root_meta_rsv(root, num_bytes, type);
3236         return ret;
3237 }
3238
3239 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3240 {
3241         struct btrfs_fs_info *fs_info = root->fs_info;
3242
3243         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3244             !is_fstree(root->objectid))
3245                 return;
3246
3247         /* TODO: Update trace point to handle such free */
3248         trace_qgroup_meta_free_all_pertrans(root);
3249         /* Special value -1 means to free all reserved space */
3250         btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1,
3251                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3252 }
3253
3254 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3255                               enum btrfs_qgroup_rsv_type type)
3256 {
3257         struct btrfs_fs_info *fs_info = root->fs_info;
3258
3259         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3260             !is_fstree(root->objectid))
3261                 return;
3262
3263         /*
3264          * reservation for META_PREALLOC can happen before quota is enabled,
3265          * which can lead to underflow.
3266          * Here ensure we will only free what we really have reserved.
3267          */
3268         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3269         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3270         trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3271         btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type);
3272 }
3273
3274 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3275                                 int num_bytes)
3276 {
3277         struct btrfs_root *quota_root = fs_info->quota_root;
3278         struct btrfs_qgroup *qgroup;
3279         struct ulist_node *unode;
3280         struct ulist_iterator uiter;
3281         int ret = 0;
3282
3283         if (num_bytes == 0)
3284                 return;
3285         if (!quota_root)
3286                 return;
3287
3288         spin_lock(&fs_info->qgroup_lock);
3289         qgroup = find_qgroup_rb(fs_info, ref_root);
3290         if (!qgroup)
3291                 goto out;
3292         ulist_reinit(fs_info->qgroup_ulist);
3293         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3294                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3295         if (ret < 0)
3296                 goto out;
3297         ULIST_ITER_INIT(&uiter);
3298         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3299                 struct btrfs_qgroup *qg;
3300                 struct btrfs_qgroup_list *glist;
3301
3302                 qg = unode_aux_to_qgroup(unode);
3303
3304                 qgroup_rsv_release(fs_info, qg, num_bytes,
3305                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3306                 qgroup_rsv_add(fs_info, qg, num_bytes,
3307                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3308                 list_for_each_entry(glist, &qg->groups, next_group) {
3309                         ret = ulist_add(fs_info->qgroup_ulist,
3310                                         glist->group->qgroupid,
3311                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3312                         if (ret < 0)
3313                                 goto out;
3314                 }
3315         }
3316 out:
3317         spin_unlock(&fs_info->qgroup_lock);
3318 }
3319
3320 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3321 {
3322         struct btrfs_fs_info *fs_info = root->fs_info;
3323
3324         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3325             !is_fstree(root->objectid))
3326                 return;
3327         /* Same as btrfs_qgroup_free_meta_prealloc() */
3328         num_bytes = sub_root_meta_rsv(root, num_bytes,
3329                                       BTRFS_QGROUP_RSV_META_PREALLOC);
3330         trace_qgroup_meta_convert(root, num_bytes);
3331         qgroup_convert_meta(fs_info, root->objectid, num_bytes);
3332 }
3333
3334 /*
3335  * Check qgroup reserved space leaking, normally at destroy inode
3336  * time
3337  */
3338 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3339 {
3340         struct extent_changeset changeset;
3341         struct ulist_node *unode;
3342         struct ulist_iterator iter;
3343         int ret;
3344
3345         extent_changeset_init(&changeset);
3346         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3347                         EXTENT_QGROUP_RESERVED, &changeset);
3348
3349         WARN_ON(ret < 0);
3350         if (WARN_ON(changeset.bytes_changed)) {
3351                 ULIST_ITER_INIT(&iter);
3352                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3353                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3354                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3355                                 inode->i_ino, unode->val, unode->aux);
3356                 }
3357                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3358                                 BTRFS_I(inode)->root->objectid,
3359                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3360
3361         }
3362         extent_changeset_release(&changeset);
3363 }