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
  15 #include "ctree.h"
  16 #include "transaction.h"
  17 #include "disk-io.h"
  18 #include "locking.h"
  19 #include "ulist.h"
  20 #include "backref.h"
  21 #include "extent_io.h"
  22 #include "qgroup.h"
  23 #include "block-group.h"
  24 #include "sysfs.h"
  25
  26 /* TODO XXX FIXME
  27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  28  *  - reorganize keys
  29  *  - compressed
  30  *  - sync
  31  *  - copy also limits on subvol creation
  32  *  - limit
  33  *  - caches for ulists
  34  *  - performance benchmarks
  35  *  - check all ioctl parameters
  36  */
  37
  38 /*
  39  * Helpers to access qgroup reservation
  40  *
  41  * Callers should ensure the lock context and type are valid
  42  */
  43
  44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
  45 {
  46         u64 ret = 0;
  47         int i;
  48
  49         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
  50                 ret += qgroup->rsv.values[i];
  51
  52         return ret;
  53 }
  54
  55 #ifdef CONFIG_BTRFS_DEBUG
  56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
  57 {
  58         if (type == BTRFS_QGROUP_RSV_DATA)
  59                 return "data";
  60         if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
  61                 return "meta_pertrans";
  62         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
  63                 return "meta_prealloc";
  64         return NULL;
  65 }
  66 #endif
  67
  68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
  69                            struct btrfs_qgroup *qgroup, u64 num_bytes,
  70                            enum btrfs_qgroup_rsv_type type)
  71 {
  72         trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
  73         qgroup->rsv.values[type] += num_bytes;
  74 }
  75
  76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
  77                                struct btrfs_qgroup *qgroup, u64 num_bytes,
  78                                enum btrfs_qgroup_rsv_type type)
  79 {
  80         trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
  81         if (qgroup->rsv.values[type] >= num_bytes) {
  82                 qgroup->rsv.values[type] -= num_bytes;
  83                 return;
  84         }
  85 #ifdef CONFIG_BTRFS_DEBUG
  86         WARN_RATELIMIT(1,
  87                 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
  88                 qgroup->qgroupid, qgroup_rsv_type_str(type),
  89                 qgroup->rsv.values[type], num_bytes);
  90 #endif
  91         qgroup->rsv.values[type] = 0;
  92 }
  93
  94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
  95                                      struct btrfs_qgroup *dest,
  96                                      struct btrfs_qgroup *src)
  97 {
  98         int i;
  99
 100         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 101                 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
 102 }
 103
 104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
 105                                          struct btrfs_qgroup *dest,
 106                                           struct btrfs_qgroup *src)
 107 {
 108         int i;
 109
 110         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
 111                 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
 112 }
 113
 114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
 115                                            int mod)
 116 {
 117         if (qg->old_refcnt < seq)
 118                 qg->old_refcnt = seq;
 119         qg->old_refcnt += mod;
 120 }
 121
 122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
 123                                            int mod)
 124 {
 125         if (qg->new_refcnt < seq)
 126                 qg->new_refcnt = seq;
 127         qg->new_refcnt += mod;
 128 }
 129
 130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
 131 {
 132         if (qg->old_refcnt < seq)
 133                 return 0;
 134         return qg->old_refcnt - seq;
 135 }
 136
 137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
 138 {
 139         if (qg->new_refcnt < seq)
 140                 return 0;
 141         return qg->new_refcnt - seq;
 142 }
 143
 144 /*
 145  * glue structure to represent the relations between qgroups.
 146  */
 147 struct btrfs_qgroup_list {
 148         struct list_head next_group;
 149         struct list_head next_member;
 150         struct btrfs_qgroup *group;
 151         struct btrfs_qgroup *member;
 152 };
 153
 154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
 155 {
 156         return (u64)(uintptr_t)qg;
 157 }
 158
 159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
 160 {
 161         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
 162 }
 163
 164 static int
 165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 166                    int init_flags);
 167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 168
 169 /* must be called with qgroup_ioctl_lock held */
 170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 171                                            u64 qgroupid)
 172 {
 173         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 174         struct btrfs_qgroup *qgroup;
 175
 176         while (n) {
 177                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 178                 if (qgroup->qgroupid < qgroupid)
 179                         n = n->rb_left;
 180                 else if (qgroup->qgroupid > qgroupid)
 181                         n = n->rb_right;
 182                 else
 183                         return qgroup;
 184         }
 185         return NULL;
 186 }
 187
 188 /* must be called with qgroup_lock held */
 189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 190                                           u64 qgroupid)
 191 {
 192         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 193         struct rb_node *parent = NULL;
 194         struct btrfs_qgroup *qgroup;
 195
 196         while (*p) {
 197                 parent = *p;
 198                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 199
 200                 if (qgroup->qgroupid < qgroupid)
 201                         p = &(*p)->rb_left;
 202                 else if (qgroup->qgroupid > qgroupid)
 203                         p = &(*p)->rb_right;
 204                 else
 205                         return qgroup;
 206         }
 207
 208         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 209         if (!qgroup)
 210                 return ERR_PTR(-ENOMEM);
 211
 212         qgroup->qgroupid = qgroupid;
 213         INIT_LIST_HEAD(&qgroup->groups);
 214         INIT_LIST_HEAD(&qgroup->members);
 215         INIT_LIST_HEAD(&qgroup->dirty);
 216
 217         rb_link_node(&qgroup->node, parent, p);
 218         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 219
 220         return qgroup;
 221 }
 222
 223 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
 224                             struct btrfs_qgroup *qgroup)
 225 {
 226         struct btrfs_qgroup_list *list;
 227
 228         btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
 229         list_del(&qgroup->dirty);
 230         while (!list_empty(&qgroup->groups)) {
 231                 list = list_first_entry(&qgroup->groups,
 232                                         struct btrfs_qgroup_list, next_group);
 233                 list_del(&list->next_group);
 234                 list_del(&list->next_member);
 235                 kfree(list);
 236         }
 237
 238         while (!list_empty(&qgroup->members)) {
 239                 list = list_first_entry(&qgroup->members,
 240                                         struct btrfs_qgroup_list, next_member);
 241                 list_del(&list->next_group);
 242                 list_del(&list->next_member);
 243                 kfree(list);
 244         }
 245         kfree(qgroup);
 246 }
 247
 248 /* must be called with qgroup_lock held */
 249 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 250 {
 251         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 252
 253         if (!qgroup)
 254                 return -ENOENT;
 255
 256         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 257         __del_qgroup_rb(fs_info, qgroup);
 258         return 0;
 259 }
 260
 261 /* must be called with qgroup_lock held */
 262 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 263                            u64 memberid, u64 parentid)
 264 {
 265         struct btrfs_qgroup *member;
 266         struct btrfs_qgroup *parent;
 267         struct btrfs_qgroup_list *list;
 268
 269         member = find_qgroup_rb(fs_info, memberid);
 270         parent = find_qgroup_rb(fs_info, parentid);
 271         if (!member || !parent)
 272                 return -ENOENT;
 273
 274         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 275         if (!list)
 276                 return -ENOMEM;
 277
 278         list->group = parent;
 279         list->member = member;
 280         list_add_tail(&list->next_group, &member->groups);
 281         list_add_tail(&list->next_member, &parent->members);
 282
 283         return 0;
 284 }
 285
 286 /* must be called with qgroup_lock held */
 287 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 288                            u64 memberid, u64 parentid)
 289 {
 290         struct btrfs_qgroup *member;
 291         struct btrfs_qgroup *parent;
 292         struct btrfs_qgroup_list *list;
 293
 294         member = find_qgroup_rb(fs_info, memberid);
 295         parent = find_qgroup_rb(fs_info, parentid);
 296         if (!member || !parent)
 297                 return -ENOENT;
 298
 299         list_for_each_entry(list, &member->groups, next_group) {
 300                 if (list->group == parent) {
 301                         list_del(&list->next_group);
 302                         list_del(&list->next_member);
 303                         kfree(list);
 304                         return 0;
 305                 }
 306         }
 307         return -ENOENT;
 308 }
 309
 310 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 311 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 312                                u64 rfer, u64 excl)
 313 {
 314         struct btrfs_qgroup *qgroup;
 315
 316         qgroup = find_qgroup_rb(fs_info, qgroupid);
 317         if (!qgroup)
 318                 return -EINVAL;
 319         if (qgroup->rfer != rfer || qgroup->excl != excl)
 320                 return -EINVAL;
 321         return 0;
 322 }
 323 #endif
 324
 325 /*
 326  * The full config is read in one go, only called from open_ctree()
 327  * It doesn't use any locking, as at this point we're still single-threaded
 328  */
 329 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 330 {
 331         struct btrfs_key key;
 332         struct btrfs_key found_key;
 333         struct btrfs_root *quota_root = fs_info->quota_root;
 334         struct btrfs_path *path = NULL;
 335         struct extent_buffer *l;
 336         int slot;
 337         int ret = 0;
 338         u64 flags = 0;
 339         u64 rescan_progress = 0;
 340
 341         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 342                 return 0;
 343
 344         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 345         if (!fs_info->qgroup_ulist) {
 346                 ret = -ENOMEM;
 347                 goto out;
 348         }
 349
 350         path = btrfs_alloc_path();
 351         if (!path) {
 352                 ret = -ENOMEM;
 353                 goto out;
 354         }
 355
 356         ret = btrfs_sysfs_add_qgroups(fs_info);
 357         if (ret < 0)
 358                 goto out;
 359         /* default this to quota off, in case no status key is found */
 360         fs_info->qgroup_flags = 0;
 361
 362         /*
 363          * pass 1: read status, all qgroup infos and limits
 364          */
 365         key.objectid = 0;
 366         key.type = 0;
 367         key.offset = 0;
 368         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 369         if (ret)
 370                 goto out;
 371
 372         while (1) {
 373                 struct btrfs_qgroup *qgroup;
 374
 375                 slot = path->slots[0];
 376                 l = path->nodes[0];
 377                 btrfs_item_key_to_cpu(l, &found_key, slot);
 378
 379                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 380                         struct btrfs_qgroup_status_item *ptr;
 381
 382                         ptr = btrfs_item_ptr(l, slot,
 383                                              struct btrfs_qgroup_status_item);
 384
 385                         if (btrfs_qgroup_status_version(l, ptr) !=
 386                             BTRFS_QGROUP_STATUS_VERSION) {
 387                                 btrfs_err(fs_info,
 388                                  "old qgroup version, quota disabled");
 389                                 goto out;
 390                         }
 391                         if (btrfs_qgroup_status_generation(l, ptr) !=
 392                             fs_info->generation) {
 393                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 394                                 btrfs_err(fs_info,
 395                                         "qgroup generation mismatch, marked as inconsistent");
 396                         }
 397                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 398                                                                           ptr);
 399                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 400                         goto next1;
 401                 }
 402
 403                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 404                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 405                         goto next1;
 406
 407                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 408                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 409                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 410                         btrfs_err(fs_info, "inconsistent qgroup config");
 411                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 412                 }
 413                 if (!qgroup) {
 414                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 415                         if (IS_ERR(qgroup)) {
 416                                 ret = PTR_ERR(qgroup);
 417                                 goto out;
 418                         }
 419                 }
 420                 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
 421                 if (ret < 0)
 422                         goto out;
 423
 424                 switch (found_key.type) {
 425                 case BTRFS_QGROUP_INFO_KEY: {
 426                         struct btrfs_qgroup_info_item *ptr;
 427
 428                         ptr = btrfs_item_ptr(l, slot,
 429                                              struct btrfs_qgroup_info_item);
 430                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 431                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 432                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 433                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 434                         /* generation currently unused */
 435                         break;
 436                 }
 437                 case BTRFS_QGROUP_LIMIT_KEY: {
 438                         struct btrfs_qgroup_limit_item *ptr;
 439
 440                         ptr = btrfs_item_ptr(l, slot,
 441                                              struct btrfs_qgroup_limit_item);
 442                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 443                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 444                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 445                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 446                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 447                         break;
 448                 }
 449                 }
 450 next1:
 451                 ret = btrfs_next_item(quota_root, path);
 452                 if (ret < 0)
 453                         goto out;
 454                 if (ret)
 455                         break;
 456         }
 457         btrfs_release_path(path);
 458
 459         /*
 460          * pass 2: read all qgroup relations
 461          */
 462         key.objectid = 0;
 463         key.type = BTRFS_QGROUP_RELATION_KEY;
 464         key.offset = 0;
 465         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 466         if (ret)
 467                 goto out;
 468         while (1) {
 469                 slot = path->slots[0];
 470                 l = path->nodes[0];
 471                 btrfs_item_key_to_cpu(l, &found_key, slot);
 472
 473                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 474                         goto next2;
 475
 476                 if (found_key.objectid > found_key.offset) {
 477                         /* parent <- member, not needed to build config */
 478                         /* FIXME should we omit the key completely? */
 479                         goto next2;
 480                 }
 481
 482                 ret = add_relation_rb(fs_info, found_key.objectid,
 483                                       found_key.offset);
 484                 if (ret == -ENOENT) {
 485                         btrfs_warn(fs_info,
 486                                 "orphan qgroup relation 0x%llx->0x%llx",
 487                                 found_key.objectid, found_key.offset);
 488                         ret = 0;        /* ignore the error */
 489                 }
 490                 if (ret)
 491                         goto out;
 492 next2:
 493                 ret = btrfs_next_item(quota_root, path);
 494                 if (ret < 0)
 495                         goto out;
 496                 if (ret)
 497                         break;
 498         }
 499 out:
 500         fs_info->qgroup_flags |= flags;
 501         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
 502                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 503         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 504                  ret >= 0)
 505                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 506         btrfs_free_path(path);
 507
 508         if (ret < 0) {
 509                 ulist_free(fs_info->qgroup_ulist);
 510                 fs_info->qgroup_ulist = NULL;
 511                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 512                 btrfs_sysfs_del_qgroups(fs_info);
 513         }
 514
 515         return ret < 0 ? ret : 0;
 516 }
 517
 518 /*
 519  * Called in close_ctree() when quota is still enabled.  This verifies we don't
 520  * leak some reserved space.
 521  *
 522  * Return false if no reserved space is left.
 523  * Return true if some reserved space is leaked.
 524  */
 525 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
 526 {
 527         struct rb_node *node;
 528         bool ret = false;
 529
 530         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 531                 return ret;
 532         /*
 533          * Since we're unmounting, there is no race and no need to grab qgroup
 534          * lock.  And here we don't go post-order to provide a more user
 535          * friendly sorted result.
 536          */
 537         for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
 538                 struct btrfs_qgroup *qgroup;
 539                 int i;
 540
 541                 qgroup = rb_entry(node, struct btrfs_qgroup, node);
 542                 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
 543                         if (qgroup->rsv.values[i]) {
 544                                 ret = true;
 545                                 btrfs_warn(fs_info,
 546                 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
 547                                    btrfs_qgroup_level(qgroup->qgroupid),
 548                                    btrfs_qgroup_subvolid(qgroup->qgroupid),
 549                                    i, qgroup->rsv.values[i]);
 550                         }
 551                 }
 552         }
 553         return ret;
 554 }
 555
 556 /*
 557  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 558  * first two are in single-threaded paths.And for the third one, we have set
 559  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 560  * up the in-memory structures without qgroup_lock held.
 561  */
 562 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 563 {
 564         struct rb_node *n;
 565         struct btrfs_qgroup *qgroup;
 566
 567         while ((n = rb_first(&fs_info->qgroup_tree))) {
 568                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 569                 rb_erase(n, &fs_info->qgroup_tree);
 570                 __del_qgroup_rb(fs_info, qgroup);
 571         }
 572         /*
 573          * We call btrfs_free_qgroup_config() when unmounting
 574          * filesystem and disabling quota, so we set qgroup_ulist
 575          * to be null here to avoid double free.
 576          */
 577         ulist_free(fs_info->qgroup_ulist);
 578         fs_info->qgroup_ulist = NULL;
 579         btrfs_sysfs_del_qgroups(fs_info);
 580 }
 581
 582 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 583                                     u64 dst)
 584 {
 585         int ret;
 586         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 587         struct btrfs_path *path;
 588         struct btrfs_key key;
 589
 590         path = btrfs_alloc_path();
 591         if (!path)
 592                 return -ENOMEM;
 593
 594         key.objectid = src;
 595         key.type = BTRFS_QGROUP_RELATION_KEY;
 596         key.offset = dst;
 597
 598         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 599
 600         btrfs_mark_buffer_dirty(path->nodes[0]);
 601
 602         btrfs_free_path(path);
 603         return ret;
 604 }
 605
 606 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
 607                                     u64 dst)
 608 {
 609         int ret;
 610         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 611         struct btrfs_path *path;
 612         struct btrfs_key key;
 613
 614         path = btrfs_alloc_path();
 615         if (!path)
 616                 return -ENOMEM;
 617
 618         key.objectid = src;
 619         key.type = BTRFS_QGROUP_RELATION_KEY;
 620         key.offset = dst;
 621
 622         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 623         if (ret < 0)
 624                 goto out;
 625
 626         if (ret > 0) {
 627                 ret = -ENOENT;
 628                 goto out;
 629         }
 630
 631         ret = btrfs_del_item(trans, quota_root, path);
 632 out:
 633         btrfs_free_path(path);
 634         return ret;
 635 }
 636
 637 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 638                            struct btrfs_root *quota_root, u64 qgroupid)
 639 {
 640         int ret;
 641         struct btrfs_path *path;
 642         struct btrfs_qgroup_info_item *qgroup_info;
 643         struct btrfs_qgroup_limit_item *qgroup_limit;
 644         struct extent_buffer *leaf;
 645         struct btrfs_key key;
 646
 647         if (btrfs_is_testing(quota_root->fs_info))
 648                 return 0;
 649
 650         path = btrfs_alloc_path();
 651         if (!path)
 652                 return -ENOMEM;
 653
 654         key.objectid = 0;
 655         key.type = BTRFS_QGROUP_INFO_KEY;
 656         key.offset = qgroupid;
 657
 658         /*
 659          * Avoid a transaction abort by catching -EEXIST here. In that
 660          * case, we proceed by re-initializing the existing structure
 661          * on disk.
 662          */
 663
 664         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 665                                       sizeof(*qgroup_info));
 666         if (ret && ret != -EEXIST)
 667                 goto out;
 668
 669         leaf = path->nodes[0];
 670         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 671                                  struct btrfs_qgroup_info_item);
 672         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 673         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 674         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 675         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 676         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 677
 678         btrfs_mark_buffer_dirty(leaf);
 679
 680         btrfs_release_path(path);
 681
 682         key.type = BTRFS_QGROUP_LIMIT_KEY;
 683         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 684                                       sizeof(*qgroup_limit));
 685         if (ret && ret != -EEXIST)
 686                 goto out;
 687
 688         leaf = path->nodes[0];
 689         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 690                                   struct btrfs_qgroup_limit_item);
 691         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 692         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 693         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 694         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 695         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 696
 697         btrfs_mark_buffer_dirty(leaf);
 698
 699         ret = 0;
 700 out:
 701         btrfs_free_path(path);
 702         return ret;
 703 }
 704
 705 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
 706 {
 707         int ret;
 708         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 709         struct btrfs_path *path;
 710         struct btrfs_key key;
 711
 712         path = btrfs_alloc_path();
 713         if (!path)
 714                 return -ENOMEM;
 715
 716         key.objectid = 0;
 717         key.type = BTRFS_QGROUP_INFO_KEY;
 718         key.offset = qgroupid;
 719         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 720         if (ret < 0)
 721                 goto out;
 722
 723         if (ret > 0) {
 724                 ret = -ENOENT;
 725                 goto out;
 726         }
 727
 728         ret = btrfs_del_item(trans, quota_root, path);
 729         if (ret)
 730                 goto out;
 731
 732         btrfs_release_path(path);
 733
 734         key.type = BTRFS_QGROUP_LIMIT_KEY;
 735         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 736         if (ret < 0)
 737                 goto out;
 738
 739         if (ret > 0) {
 740                 ret = -ENOENT;
 741                 goto out;
 742         }
 743
 744         ret = btrfs_del_item(trans, quota_root, path);
 745
 746 out:
 747         btrfs_free_path(path);
 748         return ret;
 749 }
 750
 751 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 752                                     struct btrfs_qgroup *qgroup)
 753 {
 754         struct btrfs_root *quota_root = trans->fs_info->quota_root;
 755         struct btrfs_path *path;
 756         struct btrfs_key key;
 757         struct extent_buffer *l;
 758         struct btrfs_qgroup_limit_item *qgroup_limit;
 759         int ret;
 760         int slot;
 761
 762         key.objectid = 0;
 763         key.type = BTRFS_QGROUP_LIMIT_KEY;
 764         key.offset = qgroup->qgroupid;
 765
 766         path = btrfs_alloc_path();
 767         if (!path)
 768                 return -ENOMEM;
 769
 770         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 771         if (ret > 0)
 772                 ret = -ENOENT;
 773
 774         if (ret)
 775                 goto out;
 776
 777         l = path->nodes[0];
 778         slot = path->slots[0];
 779         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 780         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 781         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 782         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 783         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 784         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 785
 786         btrfs_mark_buffer_dirty(l);
 787
 788 out:
 789         btrfs_free_path(path);
 790         return ret;
 791 }
 792
 793 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 794                                    struct btrfs_qgroup *qgroup)
 795 {
 796         struct btrfs_fs_info *fs_info = trans->fs_info;
 797         struct btrfs_root *quota_root = fs_info->quota_root;
 798         struct btrfs_path *path;
 799         struct btrfs_key key;
 800         struct extent_buffer *l;
 801         struct btrfs_qgroup_info_item *qgroup_info;
 802         int ret;
 803         int slot;
 804
 805         if (btrfs_is_testing(fs_info))
 806                 return 0;
 807
 808         key.objectid = 0;
 809         key.type = BTRFS_QGROUP_INFO_KEY;
 810         key.offset = qgroup->qgroupid;
 811
 812         path = btrfs_alloc_path();
 813         if (!path)
 814                 return -ENOMEM;
 815
 816         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 817         if (ret > 0)
 818                 ret = -ENOENT;
 819
 820         if (ret)
 821                 goto out;
 822
 823         l = path->nodes[0];
 824         slot = path->slots[0];
 825         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 826         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 827         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 828         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 829         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 830         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 831
 832         btrfs_mark_buffer_dirty(l);
 833
 834 out:
 835         btrfs_free_path(path);
 836         return ret;
 837 }
 838
 839 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
 840 {
 841         struct btrfs_fs_info *fs_info = trans->fs_info;
 842         struct btrfs_root *quota_root = fs_info->quota_root;
 843         struct btrfs_path *path;
 844         struct btrfs_key key;
 845         struct extent_buffer *l;
 846         struct btrfs_qgroup_status_item *ptr;
 847         int ret;
 848         int slot;
 849
 850         key.objectid = 0;
 851         key.type = BTRFS_QGROUP_STATUS_KEY;
 852         key.offset = 0;
 853
 854         path = btrfs_alloc_path();
 855         if (!path)
 856                 return -ENOMEM;
 857
 858         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
 859         if (ret > 0)
 860                 ret = -ENOENT;
 861
 862         if (ret)
 863                 goto out;
 864
 865         l = path->nodes[0];
 866         slot = path->slots[0];
 867         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 868         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 869         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 870         btrfs_set_qgroup_status_rescan(l, ptr,
 871                                 fs_info->qgroup_rescan_progress.objectid);
 872
 873         btrfs_mark_buffer_dirty(l);
 874
 875 out:
 876         btrfs_free_path(path);
 877         return ret;
 878 }
 879
 880 /*
 881  * called with qgroup_lock held
 882  */
 883 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 884                                   struct btrfs_root *root)
 885 {
 886         struct btrfs_path *path;
 887         struct btrfs_key key;
 888         struct extent_buffer *leaf = NULL;
 889         int ret;
 890         int nr = 0;
 891
 892         path = btrfs_alloc_path();
 893         if (!path)
 894                 return -ENOMEM;
 895
 896         path->leave_spinning = 1;
 897
 898         key.objectid = 0;
 899         key.offset = 0;
 900         key.type = 0;
 901
 902         while (1) {
 903                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 904                 if (ret < 0)
 905                         goto out;
 906                 leaf = path->nodes[0];
 907                 nr = btrfs_header_nritems(leaf);
 908                 if (!nr)
 909                         break;
 910                 /*
 911                  * delete the leaf one by one
 912                  * since the whole tree is going
 913                  * to be deleted.
 914                  */
 915                 path->slots[0] = 0;
 916                 ret = btrfs_del_items(trans, root, path, 0, nr);
 917                 if (ret)
 918                         goto out;
 919
 920                 btrfs_release_path(path);
 921         }
 922         ret = 0;
 923 out:
 924         btrfs_free_path(path);
 925         return ret;
 926 }
 927
 928 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
 929 {
 930         struct btrfs_root *quota_root;
 931         struct btrfs_root *tree_root = fs_info->tree_root;
 932         struct btrfs_path *path = NULL;
 933         struct btrfs_qgroup_status_item *ptr;
 934         struct extent_buffer *leaf;
 935         struct btrfs_key key;
 936         struct btrfs_key found_key;
 937         struct btrfs_qgroup *qgroup = NULL;
 938         struct btrfs_trans_handle *trans = NULL;
 939         int ret = 0;
 940         int slot;
 941
 942         mutex_lock(&fs_info->qgroup_ioctl_lock);
 943         if (fs_info->quota_root)
 944                 goto out;
 945
 946         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 947         if (!fs_info->qgroup_ulist) {
 948                 ret = -ENOMEM;
 949                 goto out;
 950         }
 951
 952         ret = btrfs_sysfs_add_qgroups(fs_info);
 953         if (ret < 0)
 954                 goto out;
 955         /*
 956          * 1 for quota root item
 957          * 1 for BTRFS_QGROUP_STATUS item
 958          *
 959          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
 960          * per subvolume. However those are not currently reserved since it
 961          * would be a lot of overkill.
 962          */
 963         trans = btrfs_start_transaction(tree_root, 2);
 964         if (IS_ERR(trans)) {
 965                 ret = PTR_ERR(trans);
 966                 trans = NULL;
 967                 goto out;
 968         }
 969
 970         /*
 971          * initially create the quota tree
 972          */
 973         quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
 974         if (IS_ERR(quota_root)) {
 975                 ret =  PTR_ERR(quota_root);
 976                 btrfs_abort_transaction(trans, ret);
 977                 goto out;
 978         }
 979
 980         path = btrfs_alloc_path();
 981         if (!path) {
 982                 ret = -ENOMEM;
 983                 btrfs_abort_transaction(trans, ret);
 984                 goto out_free_root;
 985         }
 986
 987         key.objectid = 0;
 988         key.type = BTRFS_QGROUP_STATUS_KEY;
 989         key.offset = 0;
 990
 991         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 992                                       sizeof(*ptr));
 993         if (ret) {
 994                 btrfs_abort_transaction(trans, ret);
 995                 goto out_free_path;
 996         }
 997
 998         leaf = path->nodes[0];
 999         ptr = btrfs_item_ptr(leaf, path->slots[0],
1000                                  struct btrfs_qgroup_status_item);
1001         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1002         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1003         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1004                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1005         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
1006         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1007
1008         btrfs_mark_buffer_dirty(leaf);
1009
1010         key.objectid = 0;
1011         key.type = BTRFS_ROOT_REF_KEY;
1012         key.offset = 0;
1013
1014         btrfs_release_path(path);
1015         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1016         if (ret > 0)
1017                 goto out_add_root;
1018         if (ret < 0) {
1019                 btrfs_abort_transaction(trans, ret);
1020                 goto out_free_path;
1021         }
1022
1023         while (1) {
1024                 slot = path->slots[0];
1025                 leaf = path->nodes[0];
1026                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1027
1028                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1029
1030                         /* Release locks on tree_root before we access quota_root */
1031                         btrfs_release_path(path);
1032
1033                         ret = add_qgroup_item(trans, quota_root,
1034                                               found_key.offset);
1035                         if (ret) {
1036                                 btrfs_abort_transaction(trans, ret);
1037                                 goto out_free_path;
1038                         }
1039
1040                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
1041                         if (IS_ERR(qgroup)) {
1042                                 ret = PTR_ERR(qgroup);
1043                                 btrfs_abort_transaction(trans, ret);
1044                                 goto out_free_path;
1045                         }
1046                         ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1047                         if (ret < 0) {
1048                                 btrfs_abort_transaction(trans, ret);
1049                                 goto out_free_path;
1050                         }
1051                         ret = btrfs_search_slot_for_read(tree_root, &found_key,
1052                                                          path, 1, 0);
1053                         if (ret < 0) {
1054                                 btrfs_abort_transaction(trans, ret);
1055                                 goto out_free_path;
1056                         }
1057                         if (ret > 0) {
1058                                 /*
1059                                  * Shouldn't happen, but in case it does we
1060                                  * don't need to do the btrfs_next_item, just
1061                                  * continue.
1062                                  */
1063                                 continue;
1064                         }
1065                 }
1066                 ret = btrfs_next_item(tree_root, path);
1067                 if (ret < 0) {
1068                         btrfs_abort_transaction(trans, ret);
1069                         goto out_free_path;
1070                 }
1071                 if (ret)
1072                         break;
1073         }
1074
1075 out_add_root:
1076         btrfs_release_path(path);
1077         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1078         if (ret) {
1079                 btrfs_abort_transaction(trans, ret);
1080                 goto out_free_path;
1081         }
1082
1083         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1084         if (IS_ERR(qgroup)) {
1085                 ret = PTR_ERR(qgroup);
1086                 btrfs_abort_transaction(trans, ret);
1087                 goto out_free_path;
1088         }
1089         ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1090         if (ret < 0) {
1091                 btrfs_abort_transaction(trans, ret);
1092                 goto out_free_path;
1093         }
1094
1095         ret = btrfs_commit_transaction(trans);
1096         trans = NULL;
1097         if (ret)
1098                 goto out_free_path;
1099
1100         /*
1101          * Set quota enabled flag after committing the transaction, to avoid
1102          * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1103          * creation.
1104          */
1105         spin_lock(&fs_info->qgroup_lock);
1106         fs_info->quota_root = quota_root;
1107         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1108         spin_unlock(&fs_info->qgroup_lock);
1109
1110         ret = qgroup_rescan_init(fs_info, 0, 1);
1111         if (!ret) {
1112                 qgroup_rescan_zero_tracking(fs_info);
1113                 fs_info->qgroup_rescan_running = true;
1114                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1115                                  &fs_info->qgroup_rescan_work);
1116         }
1117
1118 out_free_path:
1119         btrfs_free_path(path);
1120 out_free_root:
1121         if (ret)
1122                 btrfs_put_root(quota_root);
1123 out:
1124         if (ret) {
1125                 ulist_free(fs_info->qgroup_ulist);
1126                 fs_info->qgroup_ulist = NULL;
1127                 if (trans)
1128                         btrfs_end_transaction(trans);
1129                 btrfs_sysfs_del_qgroups(fs_info);
1130         }
1131         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1132         return ret;
1133 }
1134
1135 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1136 {
1137         struct btrfs_root *quota_root;
1138         struct btrfs_trans_handle *trans = NULL;
1139         int ret = 0;
1140
1141         mutex_lock(&fs_info->qgroup_ioctl_lock);
1142         if (!fs_info->quota_root)
1143                 goto out;
1144
1145         /*
1146          * 1 For the root item
1147          *
1148          * We should also reserve enough items for the quota tree deletion in
1149          * btrfs_clean_quota_tree but this is not done.
1150          */
1151         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1152         if (IS_ERR(trans)) {
1153                 ret = PTR_ERR(trans);
1154                 goto out;
1155         }
1156
1157         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1158         btrfs_qgroup_wait_for_completion(fs_info, false);
1159         spin_lock(&fs_info->qgroup_lock);
1160         quota_root = fs_info->quota_root;
1161         fs_info->quota_root = NULL;
1162         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1163         spin_unlock(&fs_info->qgroup_lock);
1164
1165         btrfs_free_qgroup_config(fs_info);
1166
1167         ret = btrfs_clean_quota_tree(trans, quota_root);
1168         if (ret) {
1169                 btrfs_abort_transaction(trans, ret);
1170                 goto end_trans;
1171         }
1172
1173         ret = btrfs_del_root(trans, &quota_root->root_key);
1174         if (ret) {
1175                 btrfs_abort_transaction(trans, ret);
1176                 goto end_trans;
1177         }
1178
1179         list_del(&quota_root->dirty_list);
1180
1181         btrfs_tree_lock(quota_root->node);
1182         btrfs_clean_tree_block(quota_root->node);
1183         btrfs_tree_unlock(quota_root->node);
1184         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1185
1186         btrfs_put_root(quota_root);
1187
1188 end_trans:
1189         ret = btrfs_end_transaction(trans);
1190 out:
1191         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1192         return ret;
1193 }
1194
1195 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1196                          struct btrfs_qgroup *qgroup)
1197 {
1198         if (list_empty(&qgroup->dirty))
1199                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1200 }
1201
1202 /*
1203  * The easy accounting, we're updating qgroup relationship whose child qgroup
1204  * only has exclusive extents.
1205  *
1206  * In this case, all exclusive extents will also be exclusive for parent, so
1207  * excl/rfer just get added/removed.
1208  *
1209  * So is qgroup reservation space, which should also be added/removed to
1210  * parent.
1211  * Or when child tries to release reservation space, parent will underflow its
1212  * reservation (for relationship adding case).
1213  *
1214  * Caller should hold fs_info->qgroup_lock.
1215  */
1216 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1217                                     struct ulist *tmp, u64 ref_root,
1218                                     struct btrfs_qgroup *src, int sign)
1219 {
1220         struct btrfs_qgroup *qgroup;
1221         struct btrfs_qgroup_list *glist;
1222         struct ulist_node *unode;
1223         struct ulist_iterator uiter;
1224         u64 num_bytes = src->excl;
1225         int ret = 0;
1226
1227         qgroup = find_qgroup_rb(fs_info, ref_root);
1228         if (!qgroup)
1229                 goto out;
1230
1231         qgroup->rfer += sign * num_bytes;
1232         qgroup->rfer_cmpr += sign * num_bytes;
1233
1234         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1235         qgroup->excl += sign * num_bytes;
1236         qgroup->excl_cmpr += sign * num_bytes;
1237
1238         if (sign > 0)
1239                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1240         else
1241                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1242
1243         qgroup_dirty(fs_info, qgroup);
1244
1245         /* Get all of the parent groups that contain this qgroup */
1246         list_for_each_entry(glist, &qgroup->groups, next_group) {
1247                 ret = ulist_add(tmp, glist->group->qgroupid,
1248                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1249                 if (ret < 0)
1250                         goto out;
1251         }
1252
1253         /* Iterate all of the parents and adjust their reference counts */
1254         ULIST_ITER_INIT(&uiter);
1255         while ((unode = ulist_next(tmp, &uiter))) {
1256                 qgroup = unode_aux_to_qgroup(unode);
1257                 qgroup->rfer += sign * num_bytes;
1258                 qgroup->rfer_cmpr += sign * num_bytes;
1259                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1260                 qgroup->excl += sign * num_bytes;
1261                 if (sign > 0)
1262                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1263                 else
1264                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1265                 qgroup->excl_cmpr += sign * num_bytes;
1266                 qgroup_dirty(fs_info, qgroup);
1267
1268                 /* Add any parents of the parents */
1269                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1270                         ret = ulist_add(tmp, glist->group->qgroupid,
1271                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1272                         if (ret < 0)
1273                                 goto out;
1274                 }
1275         }
1276         ret = 0;
1277 out:
1278         return ret;
1279 }
1280
1281
1282 /*
1283  * Quick path for updating qgroup with only excl refs.
1284  *
1285  * In that case, just update all parent will be enough.
1286  * Or we needs to do a full rescan.
1287  * Caller should also hold fs_info->qgroup_lock.
1288  *
1289  * Return 0 for quick update, return >0 for need to full rescan
1290  * and mark INCONSISTENT flag.
1291  * Return < 0 for other error.
1292  */
1293 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1294                                    struct ulist *tmp, u64 src, u64 dst,
1295                                    int sign)
1296 {
1297         struct btrfs_qgroup *qgroup;
1298         int ret = 1;
1299         int err = 0;
1300
1301         qgroup = find_qgroup_rb(fs_info, src);
1302         if (!qgroup)
1303                 goto out;
1304         if (qgroup->excl == qgroup->rfer) {
1305                 ret = 0;
1306                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1307                                                qgroup, sign);
1308                 if (err < 0) {
1309                         ret = err;
1310                         goto out;
1311                 }
1312         }
1313 out:
1314         if (ret)
1315                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1316         return ret;
1317 }
1318
1319 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1320                               u64 dst)
1321 {
1322         struct btrfs_fs_info *fs_info = trans->fs_info;
1323         struct btrfs_qgroup *parent;
1324         struct btrfs_qgroup *member;
1325         struct btrfs_qgroup_list *list;
1326         struct ulist *tmp;
1327         int ret = 0;
1328
1329         /* Check the level of src and dst first */
1330         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1331                 return -EINVAL;
1332
1333         tmp = ulist_alloc(GFP_KERNEL);
1334         if (!tmp)
1335                 return -ENOMEM;
1336
1337         mutex_lock(&fs_info->qgroup_ioctl_lock);
1338         if (!fs_info->quota_root) {
1339                 ret = -ENOTCONN;
1340                 goto out;
1341         }
1342         member = find_qgroup_rb(fs_info, src);
1343         parent = find_qgroup_rb(fs_info, dst);
1344         if (!member || !parent) {
1345                 ret = -EINVAL;
1346                 goto out;
1347         }
1348
1349         /* check if such qgroup relation exist firstly */
1350         list_for_each_entry(list, &member->groups, next_group) {
1351                 if (list->group == parent) {
1352                         ret = -EEXIST;
1353                         goto out;
1354                 }
1355         }
1356
1357         ret = add_qgroup_relation_item(trans, src, dst);
1358         if (ret)
1359                 goto out;
1360
1361         ret = add_qgroup_relation_item(trans, dst, src);
1362         if (ret) {
1363                 del_qgroup_relation_item(trans, src, dst);
1364                 goto out;
1365         }
1366
1367         spin_lock(&fs_info->qgroup_lock);
1368         ret = add_relation_rb(fs_info, src, dst);
1369         if (ret < 0) {
1370                 spin_unlock(&fs_info->qgroup_lock);
1371                 goto out;
1372         }
1373         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1374         spin_unlock(&fs_info->qgroup_lock);
1375 out:
1376         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1377         ulist_free(tmp);
1378         return ret;
1379 }
1380
1381 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1382                                  u64 dst)
1383 {
1384         struct btrfs_fs_info *fs_info = trans->fs_info;
1385         struct btrfs_qgroup *parent;
1386         struct btrfs_qgroup *member;
1387         struct btrfs_qgroup_list *list;
1388         struct ulist *tmp;
1389         bool found = false;
1390         int ret = 0;
1391         int ret2;
1392
1393         tmp = ulist_alloc(GFP_KERNEL);
1394         if (!tmp)
1395                 return -ENOMEM;
1396
1397         if (!fs_info->quota_root) {
1398                 ret = -ENOTCONN;
1399                 goto out;
1400         }
1401
1402         member = find_qgroup_rb(fs_info, src);
1403         parent = find_qgroup_rb(fs_info, dst);
1404         /*
1405          * The parent/member pair doesn't exist, then try to delete the dead
1406          * relation items only.
1407          */
1408         if (!member || !parent)
1409                 goto delete_item;
1410
1411         /* check if such qgroup relation exist firstly */
1412         list_for_each_entry(list, &member->groups, next_group) {
1413                 if (list->group == parent) {
1414                         found = true;
1415                         break;
1416                 }
1417         }
1418
1419 delete_item:
1420         ret = del_qgroup_relation_item(trans, src, dst);
1421         if (ret < 0 && ret != -ENOENT)
1422                 goto out;
1423         ret2 = del_qgroup_relation_item(trans, dst, src);
1424         if (ret2 < 0 && ret2 != -ENOENT)
1425                 goto out;
1426
1427         /* At least one deletion succeeded, return 0 */
1428         if (!ret || !ret2)
1429                 ret = 0;
1430
1431         if (found) {
1432                 spin_lock(&fs_info->qgroup_lock);
1433                 del_relation_rb(fs_info, src, dst);
1434                 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1435                 spin_unlock(&fs_info->qgroup_lock);
1436         }
1437 out:
1438         ulist_free(tmp);
1439         return ret;
1440 }
1441
1442 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1443                               u64 dst)
1444 {
1445         struct btrfs_fs_info *fs_info = trans->fs_info;
1446         int ret = 0;
1447
1448         mutex_lock(&fs_info->qgroup_ioctl_lock);
1449         ret = __del_qgroup_relation(trans, src, dst);
1450         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1451
1452         return ret;
1453 }
1454
1455 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1456 {
1457         struct btrfs_fs_info *fs_info = trans->fs_info;
1458         struct btrfs_root *quota_root;
1459         struct btrfs_qgroup *qgroup;
1460         int ret = 0;
1461
1462         mutex_lock(&fs_info->qgroup_ioctl_lock);
1463         if (!fs_info->quota_root) {
1464                 ret = -ENOTCONN;
1465                 goto out;
1466         }
1467         quota_root = fs_info->quota_root;
1468         qgroup = find_qgroup_rb(fs_info, qgroupid);
1469         if (qgroup) {
1470                 ret = -EEXIST;
1471                 goto out;
1472         }
1473
1474         ret = add_qgroup_item(trans, quota_root, qgroupid);
1475         if (ret)
1476                 goto out;
1477
1478         spin_lock(&fs_info->qgroup_lock);
1479         qgroup = add_qgroup_rb(fs_info, qgroupid);
1480         spin_unlock(&fs_info->qgroup_lock);
1481
1482         if (IS_ERR(qgroup)) {
1483                 ret = PTR_ERR(qgroup);
1484                 goto out;
1485         }
1486         ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1487 out:
1488         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1489         return ret;
1490 }
1491
1492 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1493 {
1494         struct btrfs_fs_info *fs_info = trans->fs_info;
1495         struct btrfs_qgroup *qgroup;
1496         struct btrfs_qgroup_list *list;
1497         int ret = 0;
1498
1499         mutex_lock(&fs_info->qgroup_ioctl_lock);
1500         if (!fs_info->quota_root) {
1501                 ret = -ENOTCONN;
1502                 goto out;
1503         }
1504
1505         qgroup = find_qgroup_rb(fs_info, qgroupid);
1506         if (!qgroup) {
1507                 ret = -ENOENT;
1508                 goto out;
1509         }
1510
1511         /* Check if there are no children of this qgroup */
1512         if (!list_empty(&qgroup->members)) {
1513                 ret = -EBUSY;
1514                 goto out;
1515         }
1516
1517         ret = del_qgroup_item(trans, qgroupid);
1518         if (ret && ret != -ENOENT)
1519                 goto out;
1520
1521         while (!list_empty(&qgroup->groups)) {
1522                 list = list_first_entry(&qgroup->groups,
1523                                         struct btrfs_qgroup_list, next_group);
1524                 ret = __del_qgroup_relation(trans, qgroupid,
1525                                             list->group->qgroupid);
1526                 if (ret)
1527                         goto out;
1528         }
1529
1530         spin_lock(&fs_info->qgroup_lock);
1531         del_qgroup_rb(fs_info, qgroupid);
1532         spin_unlock(&fs_info->qgroup_lock);
1533 out:
1534         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1535         return ret;
1536 }
1537
1538 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1539                        struct btrfs_qgroup_limit *limit)
1540 {
1541         struct btrfs_fs_info *fs_info = trans->fs_info;
1542         struct btrfs_qgroup *qgroup;
1543         int ret = 0;
1544         /* Sometimes we would want to clear the limit on this qgroup.
1545          * To meet this requirement, we treat the -1 as a special value
1546          * which tell kernel to clear the limit on this qgroup.
1547          */
1548         const u64 CLEAR_VALUE = -1;
1549
1550         mutex_lock(&fs_info->qgroup_ioctl_lock);
1551         if (!fs_info->quota_root) {
1552                 ret = -ENOTCONN;
1553                 goto out;
1554         }
1555
1556         qgroup = find_qgroup_rb(fs_info, qgroupid);
1557         if (!qgroup) {
1558                 ret = -ENOENT;
1559                 goto out;
1560         }
1561
1562         spin_lock(&fs_info->qgroup_lock);
1563         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1564                 if (limit->max_rfer == CLEAR_VALUE) {
1565                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1566                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1567                         qgroup->max_rfer = 0;
1568                 } else {
1569                         qgroup->max_rfer = limit->max_rfer;
1570                 }
1571         }
1572         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1573                 if (limit->max_excl == CLEAR_VALUE) {
1574                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1575                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1576                         qgroup->max_excl = 0;
1577                 } else {
1578                         qgroup->max_excl = limit->max_excl;
1579                 }
1580         }
1581         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1582                 if (limit->rsv_rfer == CLEAR_VALUE) {
1583                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1584                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1585                         qgroup->rsv_rfer = 0;
1586                 } else {
1587                         qgroup->rsv_rfer = limit->rsv_rfer;
1588                 }
1589         }
1590         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1591                 if (limit->rsv_excl == CLEAR_VALUE) {
1592                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1593                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1594                         qgroup->rsv_excl = 0;
1595                 } else {
1596                         qgroup->rsv_excl = limit->rsv_excl;
1597                 }
1598         }
1599         qgroup->lim_flags |= limit->flags;
1600
1601         spin_unlock(&fs_info->qgroup_lock);
1602
1603         ret = update_qgroup_limit_item(trans, qgroup);
1604         if (ret) {
1605                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1606                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1607                        qgroupid);
1608         }
1609
1610 out:
1611         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1612         return ret;
1613 }
1614
1615 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1616                                 struct btrfs_delayed_ref_root *delayed_refs,
1617                                 struct btrfs_qgroup_extent_record *record)
1618 {
1619         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1620         struct rb_node *parent_node = NULL;
1621         struct btrfs_qgroup_extent_record *entry;
1622         u64 bytenr = record->bytenr;
1623
1624         lockdep_assert_held(&delayed_refs->lock);
1625         trace_btrfs_qgroup_trace_extent(fs_info, record);
1626
1627         while (*p) {
1628                 parent_node = *p;
1629                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1630                                  node);
1631                 if (bytenr < entry->bytenr) {
1632                         p = &(*p)->rb_left;
1633                 } else if (bytenr > entry->bytenr) {
1634                         p = &(*p)->rb_right;
1635                 } else {
1636                         if (record->data_rsv && !entry->data_rsv) {
1637                                 entry->data_rsv = record->data_rsv;
1638                                 entry->data_rsv_refroot =
1639                                         record->data_rsv_refroot;
1640                         }
1641                         return 1;
1642                 }
1643         }
1644
1645         rb_link_node(&record->node, parent_node, p);
1646         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1647         return 0;
1648 }
1649
1650 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1651                                    struct btrfs_qgroup_extent_record *qrecord)
1652 {
1653         struct ulist *old_root;
1654         u64 bytenr = qrecord->bytenr;
1655         int ret;
1656
1657         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1658         if (ret < 0) {
1659                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1660                 btrfs_warn(fs_info,
1661 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1662                         ret);
1663                 return 0;
1664         }
1665
1666         /*
1667          * Here we don't need to get the lock of
1668          * trans->transaction->delayed_refs, since inserted qrecord won't
1669          * be deleted, only qrecord->node may be modified (new qrecord insert)
1670          *
1671          * So modifying qrecord->old_roots is safe here
1672          */
1673         qrecord->old_roots = old_root;
1674         return 0;
1675 }
1676
1677 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1678                               u64 num_bytes, gfp_t gfp_flag)
1679 {
1680         struct btrfs_fs_info *fs_info = trans->fs_info;
1681         struct btrfs_qgroup_extent_record *record;
1682         struct btrfs_delayed_ref_root *delayed_refs;
1683         int ret;
1684
1685         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1686             || bytenr == 0 || num_bytes == 0)
1687                 return 0;
1688         record = kzalloc(sizeof(*record), gfp_flag);
1689         if (!record)
1690                 return -ENOMEM;
1691
1692         delayed_refs = &trans->transaction->delayed_refs;
1693         record->bytenr = bytenr;
1694         record->num_bytes = num_bytes;
1695         record->old_roots = NULL;
1696
1697         spin_lock(&delayed_refs->lock);
1698         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1699         spin_unlock(&delayed_refs->lock);
1700         if (ret > 0) {
1701                 kfree(record);
1702                 return 0;
1703         }
1704         return btrfs_qgroup_trace_extent_post(fs_info, record);
1705 }
1706
1707 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1708                                   struct extent_buffer *eb)
1709 {
1710         struct btrfs_fs_info *fs_info = trans->fs_info;
1711         int nr = btrfs_header_nritems(eb);
1712         int i, extent_type, ret;
1713         struct btrfs_key key;
1714         struct btrfs_file_extent_item *fi;
1715         u64 bytenr, num_bytes;
1716
1717         /* We can be called directly from walk_up_proc() */
1718         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1719                 return 0;
1720
1721         for (i = 0; i < nr; i++) {
1722                 btrfs_item_key_to_cpu(eb, &key, i);
1723
1724                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1725                         continue;
1726
1727                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1728                 /* filter out non qgroup-accountable extents  */
1729                 extent_type = btrfs_file_extent_type(eb, fi);
1730
1731                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1732                         continue;
1733
1734                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1735                 if (!bytenr)
1736                         continue;
1737
1738                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1739
1740                 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1741                                                 GFP_NOFS);
1742                 if (ret)
1743                         return ret;
1744         }
1745         cond_resched();
1746         return 0;
1747 }
1748
1749 /*
1750  * Walk up the tree from the bottom, freeing leaves and any interior
1751  * nodes which have had all slots visited. If a node (leaf or
1752  * interior) is freed, the node above it will have it's slot
1753  * incremented. The root node will never be freed.
1754  *
1755  * At the end of this function, we should have a path which has all
1756  * slots incremented to the next position for a search. If we need to
1757  * read a new node it will be NULL and the node above it will have the
1758  * correct slot selected for a later read.
1759  *
1760  * If we increment the root nodes slot counter past the number of
1761  * elements, 1 is returned to signal completion of the search.
1762  */
1763 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1764 {
1765         int level = 0;
1766         int nr, slot;
1767         struct extent_buffer *eb;
1768
1769         if (root_level == 0)
1770                 return 1;
1771
1772         while (level <= root_level) {
1773                 eb = path->nodes[level];
1774                 nr = btrfs_header_nritems(eb);
1775                 path->slots[level]++;
1776                 slot = path->slots[level];
1777                 if (slot >= nr || level == 0) {
1778                         /*
1779                          * Don't free the root -  we will detect this
1780                          * condition after our loop and return a
1781                          * positive value for caller to stop walking the tree.
1782                          */
1783                         if (level != root_level) {
1784                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1785                                 path->locks[level] = 0;
1786
1787                                 free_extent_buffer(eb);
1788                                 path->nodes[level] = NULL;
1789                                 path->slots[level] = 0;
1790                         }
1791                 } else {
1792                         /*
1793                          * We have a valid slot to walk back down
1794                          * from. Stop here so caller can process these
1795                          * new nodes.
1796                          */
1797                         break;
1798                 }
1799
1800                 level++;
1801         }
1802
1803         eb = path->nodes[root_level];
1804         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1805                 return 1;
1806
1807         return 0;
1808 }
1809
1810 /*
1811  * Helper function to trace a subtree tree block swap.
1812  *
1813  * The swap will happen in highest tree block, but there may be a lot of
1814  * tree blocks involved.
1815  *
1816  * For example:
1817  *  OO = Old tree blocks
1818  *  NN = New tree blocks allocated during balance
1819  *
1820  *           File tree (257)                  Reloc tree for 257
1821  * L2              OO                                NN
1822  *               /    \                            /    \
1823  * L1          OO      OO (a)                    OO      NN (a)
1824  *            / \     / \                       / \     / \
1825  * L0       OO   OO OO   OO                   OO   OO NN   NN
1826  *                  (b)  (c)                          (b)  (c)
1827  *
1828  * When calling qgroup_trace_extent_swap(), we will pass:
1829  * @src_eb = OO(a)
1830  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1831  * @dst_level = 0
1832  * @root_level = 1
1833  *
1834  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1835  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1836  *
1837  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1838  *
1839  * 1) Tree search from @src_eb
1840  *    It should acts as a simplified btrfs_search_slot().
1841  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1842  *    (first key).
1843  *
1844  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1845  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1846  *    They should be marked during previous (@dst_level = 1) iteration.
1847  *
1848  * 3) Mark file extents in leaves dirty
1849  *    We don't have good way to pick out new file extents only.
1850  *    So we still follow the old method by scanning all file extents in
1851  *    the leave.
1852  *
1853  * This function can free us from keeping two paths, thus later we only need
1854  * to care about how to iterate all new tree blocks in reloc tree.
1855  */
1856 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1857                                     struct extent_buffer *src_eb,
1858                                     struct btrfs_path *dst_path,
1859                                     int dst_level, int root_level,
1860                                     bool trace_leaf)
1861 {
1862         struct btrfs_key key;
1863         struct btrfs_path *src_path;
1864         struct btrfs_fs_info *fs_info = trans->fs_info;
1865         u32 nodesize = fs_info->nodesize;
1866         int cur_level = root_level;
1867         int ret;
1868
1869         BUG_ON(dst_level > root_level);
1870         /* Level mismatch */
1871         if (btrfs_header_level(src_eb) != root_level)
1872                 return -EINVAL;
1873
1874         src_path = btrfs_alloc_path();
1875         if (!src_path) {
1876                 ret = -ENOMEM;
1877                 goto out;
1878         }
1879
1880         if (dst_level)
1881                 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1882         else
1883                 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1884
1885         /* For src_path */
1886         atomic_inc(&src_eb->refs);
1887         src_path->nodes[root_level] = src_eb;
1888         src_path->slots[root_level] = dst_path->slots[root_level];
1889         src_path->locks[root_level] = 0;
1890
1891         /* A simplified version of btrfs_search_slot() */
1892         while (cur_level >= dst_level) {
1893                 struct btrfs_key src_key;
1894                 struct btrfs_key dst_key;
1895
1896                 if (src_path->nodes[cur_level] == NULL) {
1897                         struct btrfs_key first_key;
1898                         struct extent_buffer *eb;
1899                         int parent_slot;
1900                         u64 child_gen;
1901                         u64 child_bytenr;
1902
1903                         eb = src_path->nodes[cur_level + 1];
1904                         parent_slot = src_path->slots[cur_level + 1];
1905                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1906                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1907                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1908
1909                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1910                                              cur_level, &first_key);
1911                         if (IS_ERR(eb)) {
1912                                 ret = PTR_ERR(eb);
1913                                 goto out;
1914                         } else if (!extent_buffer_uptodate(eb)) {
1915                                 free_extent_buffer(eb);
1916                                 ret = -EIO;
1917                                 goto out;
1918                         }
1919
1920                         src_path->nodes[cur_level] = eb;
1921
1922                         btrfs_tree_read_lock(eb);
1923                         btrfs_set_lock_blocking_read(eb);
1924                         src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1925                 }
1926
1927                 src_path->slots[cur_level] = dst_path->slots[cur_level];
1928                 if (cur_level) {
1929                         btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1930                                         &dst_key, dst_path->slots[cur_level]);
1931                         btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1932                                         &src_key, src_path->slots[cur_level]);
1933                 } else {
1934                         btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1935                                         &dst_key, dst_path->slots[cur_level]);
1936                         btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1937                                         &src_key, src_path->slots[cur_level]);
1938                 }
1939                 /* Content mismatch, something went wrong */
1940                 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1941                         ret = -ENOENT;
1942                         goto out;
1943                 }
1944                 cur_level--;
1945         }
1946
1947         /*
1948          * Now both @dst_path and @src_path have been populated, record the tree
1949          * blocks for qgroup accounting.
1950          */
1951         ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1952                         nodesize, GFP_NOFS);
1953         if (ret < 0)
1954                 goto out;
1955         ret = btrfs_qgroup_trace_extent(trans,
1956                         dst_path->nodes[dst_level]->start,
1957                         nodesize, GFP_NOFS);
1958         if (ret < 0)
1959                 goto out;
1960
1961         /* Record leaf file extents */
1962         if (dst_level == 0 && trace_leaf) {
1963                 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1964                 if (ret < 0)
1965                         goto out;
1966                 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1967         }
1968 out:
1969         btrfs_free_path(src_path);
1970         return ret;
1971 }
1972
1973 /*
1974  * Helper function to do recursive generation-aware depth-first search, to
1975  * locate all new tree blocks in a subtree of reloc tree.
1976  *
1977  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1978  *         reloc tree
1979  * L2         NN (a)
1980  *          /    \
1981  * L1    OO        NN (b)
1982  *      /  \      /  \
1983  * L0  OO  OO    OO  NN
1984  *               (c) (d)
1985  * If we pass:
1986  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1987  * @cur_level = 1
1988  * @root_level = 1
1989  *
1990  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1991  * above tree blocks along with their counter parts in file tree.
1992  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1993  * won't affect OO(c).
1994  */
1995 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1996                                            struct extent_buffer *src_eb,
1997                                            struct btrfs_path *dst_path,
1998                                            int cur_level, int root_level,
1999                                            u64 last_snapshot, bool trace_leaf)
2000 {
2001         struct btrfs_fs_info *fs_info = trans->fs_info;
2002         struct extent_buffer *eb;
2003         bool need_cleanup = false;
2004         int ret = 0;
2005         int i;
2006
2007         /* Level sanity check */
2008         if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2009             root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2010             root_level < cur_level) {
2011                 btrfs_err_rl(fs_info,
2012                         "%s: bad levels, cur_level=%d root_level=%d",
2013                         __func__, cur_level, root_level);
2014                 return -EUCLEAN;
2015         }
2016
2017         /* Read the tree block if needed */
2018         if (dst_path->nodes[cur_level] == NULL) {
2019                 struct btrfs_key first_key;
2020                 int parent_slot;
2021                 u64 child_gen;
2022                 u64 child_bytenr;
2023
2024                 /*
2025                  * dst_path->nodes[root_level] must be initialized before
2026                  * calling this function.
2027                  */
2028                 if (cur_level == root_level) {
2029                         btrfs_err_rl(fs_info,
2030         "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2031                                 __func__, root_level, root_level, cur_level);
2032                         return -EUCLEAN;
2033                 }
2034
2035                 /*
2036                  * We need to get child blockptr/gen from parent before we can
2037                  * read it.
2038                   */
2039                 eb = dst_path->nodes[cur_level + 1];
2040                 parent_slot = dst_path->slots[cur_level + 1];
2041                 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2042                 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2043                 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2044
2045                 /* This node is old, no need to trace */
2046                 if (child_gen < last_snapshot)
2047                         goto out;
2048
2049                 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2050                                      cur_level, &first_key);
2051                 if (IS_ERR(eb)) {
2052                         ret = PTR_ERR(eb);
2053                         goto out;
2054                 } else if (!extent_buffer_uptodate(eb)) {
2055                         free_extent_buffer(eb);
2056                         ret = -EIO;
2057                         goto out;
2058                 }
2059
2060                 dst_path->nodes[cur_level] = eb;
2061                 dst_path->slots[cur_level] = 0;
2062
2063                 btrfs_tree_read_lock(eb);
2064                 btrfs_set_lock_blocking_read(eb);
2065                 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
2066                 need_cleanup = true;
2067         }
2068
2069         /* Now record this tree block and its counter part for qgroups */
2070         ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2071                                        root_level, trace_leaf);
2072         if (ret < 0)
2073                 goto cleanup;
2074
2075         eb = dst_path->nodes[cur_level];
2076
2077         if (cur_level > 0) {
2078                 /* Iterate all child tree blocks */
2079                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2080                         /* Skip old tree blocks as they won't be swapped */
2081                         if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2082                                 continue;
2083                         dst_path->slots[cur_level] = i;
2084
2085                         /* Recursive call (at most 7 times) */
2086                         ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2087                                         dst_path, cur_level - 1, root_level,
2088                                         last_snapshot, trace_leaf);
2089                         if (ret < 0)
2090                                 goto cleanup;
2091                 }
2092         }
2093
2094 cleanup:
2095         if (need_cleanup) {
2096                 /* Clean up */
2097                 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2098                                      dst_path->locks[cur_level]);
2099                 free_extent_buffer(dst_path->nodes[cur_level]);
2100                 dst_path->nodes[cur_level] = NULL;
2101                 dst_path->slots[cur_level] = 0;
2102                 dst_path->locks[cur_level] = 0;
2103         }
2104 out:
2105         return ret;
2106 }
2107
2108 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2109                                 struct extent_buffer *src_eb,
2110                                 struct extent_buffer *dst_eb,
2111                                 u64 last_snapshot, bool trace_leaf)
2112 {
2113         struct btrfs_fs_info *fs_info = trans->fs_info;
2114         struct btrfs_path *dst_path = NULL;
2115         int level;
2116         int ret;
2117
2118         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2119                 return 0;
2120
2121         /* Wrong parameter order */
2122         if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2123                 btrfs_err_rl(fs_info,
2124                 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2125                              btrfs_header_generation(src_eb),
2126                              btrfs_header_generation(dst_eb));
2127                 return -EUCLEAN;
2128         }
2129
2130         if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2131                 ret = -EIO;
2132                 goto out;
2133         }
2134
2135         level = btrfs_header_level(dst_eb);
2136         dst_path = btrfs_alloc_path();
2137         if (!dst_path) {
2138                 ret = -ENOMEM;
2139                 goto out;
2140         }
2141         /* For dst_path */
2142         atomic_inc(&dst_eb->refs);
2143         dst_path->nodes[level] = dst_eb;
2144         dst_path->slots[level] = 0;
2145         dst_path->locks[level] = 0;
2146
2147         /* Do the generation aware breadth-first search */
2148         ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2149                                               level, last_snapshot, trace_leaf);
2150         if (ret < 0)
2151                 goto out;
2152         ret = 0;
2153
2154 out:
2155         btrfs_free_path(dst_path);
2156         if (ret < 0)
2157                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2158         return ret;
2159 }
2160
2161 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2162                                struct extent_buffer *root_eb,
2163                                u64 root_gen, int root_level)
2164 {
2165         struct btrfs_fs_info *fs_info = trans->fs_info;
2166         int ret = 0;
2167         int level;
2168         struct extent_buffer *eb = root_eb;
2169         struct btrfs_path *path = NULL;
2170
2171         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2172         BUG_ON(root_eb == NULL);
2173
2174         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2175                 return 0;
2176
2177         if (!extent_buffer_uptodate(root_eb)) {
2178                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2179                 if (ret)
2180                         goto out;
2181         }
2182
2183         if (root_level == 0) {
2184                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2185                 goto out;
2186         }
2187
2188         path = btrfs_alloc_path();
2189         if (!path)
2190                 return -ENOMEM;
2191
2192         /*
2193          * Walk down the tree.  Missing extent blocks are filled in as
2194          * we go. Metadata is accounted every time we read a new
2195          * extent block.
2196          *
2197          * When we reach a leaf, we account for file extent items in it,
2198          * walk back up the tree (adjusting slot pointers as we go)
2199          * and restart the search process.
2200          */
2201         atomic_inc(&root_eb->refs);     /* For path */
2202         path->nodes[root_level] = root_eb;
2203         path->slots[root_level] = 0;
2204         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2205 walk_down:
2206         level = root_level;
2207         while (level >= 0) {
2208                 if (path->nodes[level] == NULL) {
2209                         struct btrfs_key first_key;
2210                         int parent_slot;
2211                         u64 child_gen;
2212                         u64 child_bytenr;
2213
2214                         /*
2215                          * We need to get child blockptr/gen from parent before
2216                          * we can read it.
2217                           */
2218                         eb = path->nodes[level + 1];
2219                         parent_slot = path->slots[level + 1];
2220                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2221                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2222                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2223
2224                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
2225                                              level, &first_key);
2226                         if (IS_ERR(eb)) {
2227                                 ret = PTR_ERR(eb);
2228                                 goto out;
2229                         } else if (!extent_buffer_uptodate(eb)) {
2230                                 free_extent_buffer(eb);
2231                                 ret = -EIO;
2232                                 goto out;
2233                         }
2234
2235                         path->nodes[level] = eb;
2236                         path->slots[level] = 0;
2237
2238                         btrfs_tree_read_lock(eb);
2239                         btrfs_set_lock_blocking_read(eb);
2240                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2241
2242                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2243                                                         fs_info->nodesize,
2244                                                         GFP_NOFS);
2245                         if (ret)
2246                                 goto out;
2247                 }
2248
2249                 if (level == 0) {
2250                         ret = btrfs_qgroup_trace_leaf_items(trans,
2251                                                             path->nodes[level]);
2252                         if (ret)
2253                                 goto out;
2254
2255                         /* Nonzero return here means we completed our search */
2256                         ret = adjust_slots_upwards(path, root_level);
2257                         if (ret)
2258                                 break;
2259
2260                         /* Restart search with new slots */
2261                         goto walk_down;
2262                 }
2263
2264                 level--;
2265         }
2266
2267         ret = 0;
2268 out:
2269         btrfs_free_path(path);
2270
2271         return ret;
2272 }
2273
2274 #define UPDATE_NEW      0
2275 #define UPDATE_OLD      1
2276 /*
2277  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2278  */
2279 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2280                                 struct ulist *roots, struct ulist *tmp,
2281                                 struct ulist *qgroups, u64 seq, int update_old)
2282 {
2283         struct ulist_node *unode;
2284         struct ulist_iterator uiter;
2285         struct ulist_node *tmp_unode;
2286         struct ulist_iterator tmp_uiter;
2287         struct btrfs_qgroup *qg;
2288         int ret = 0;
2289
2290         if (!roots)
2291                 return 0;
2292         ULIST_ITER_INIT(&uiter);
2293         while ((unode = ulist_next(roots, &uiter))) {
2294                 qg = find_qgroup_rb(fs_info, unode->val);
2295                 if (!qg)
2296                         continue;
2297
2298                 ulist_reinit(tmp);
2299                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2300                                 GFP_ATOMIC);
2301                 if (ret < 0)
2302                         return ret;
2303                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2304                 if (ret < 0)
2305                         return ret;
2306                 ULIST_ITER_INIT(&tmp_uiter);
2307                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2308                         struct btrfs_qgroup_list *glist;
2309
2310                         qg = unode_aux_to_qgroup(tmp_unode);
2311                         if (update_old)
2312                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2313                         else
2314                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2315                         list_for_each_entry(glist, &qg->groups, next_group) {
2316                                 ret = ulist_add(qgroups, glist->group->qgroupid,
2317                                                 qgroup_to_aux(glist->group),
2318                                                 GFP_ATOMIC);
2319                                 if (ret < 0)
2320                                         return ret;
2321                                 ret = ulist_add(tmp, glist->group->qgroupid,
2322                                                 qgroup_to_aux(glist->group),
2323                                                 GFP_ATOMIC);
2324                                 if (ret < 0)
2325                                         return ret;
2326                         }
2327                 }
2328         }
2329         return 0;
2330 }
2331
2332 /*
2333  * Update qgroup rfer/excl counters.
2334  * Rfer update is easy, codes can explain themselves.
2335  *
2336  * Excl update is tricky, the update is split into 2 parts.
2337  * Part 1: Possible exclusive <-> sharing detect:
2338  *      |       A       |       !A      |
2339  *  -------------------------------------
2340  *  B   |       *       |       -       |
2341  *  -------------------------------------
2342  *  !B  |       +       |       **      |
2343  *  -------------------------------------
2344  *
2345  * Conditions:
2346  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
2347  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
2348  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
2349  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
2350  *
2351  * Results:
2352  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
2353  * *: Definitely not changed.           **: Possible unchanged.
2354  *
2355  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2356  *
2357  * To make the logic clear, we first use condition A and B to split
2358  * combination into 4 results.
2359  *
2360  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2361  * only on variant maybe 0.
2362  *
2363  * Lastly, check result **, since there are 2 variants maybe 0, split them
2364  * again(2x2).
2365  * But this time we don't need to consider other things, the codes and logic
2366  * is easy to understand now.
2367  */
2368 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2369                                   struct ulist *qgroups,
2370                                   u64 nr_old_roots,
2371                                   u64 nr_new_roots,
2372                                   u64 num_bytes, u64 seq)
2373 {
2374         struct ulist_node *unode;
2375         struct ulist_iterator uiter;
2376         struct btrfs_qgroup *qg;
2377         u64 cur_new_count, cur_old_count;
2378
2379         ULIST_ITER_INIT(&uiter);
2380         while ((unode = ulist_next(qgroups, &uiter))) {
2381                 bool dirty = false;
2382
2383                 qg = unode_aux_to_qgroup(unode);
2384                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2385                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2386
2387                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2388                                              cur_new_count);
2389
2390                 /* Rfer update part */
2391                 if (cur_old_count == 0 && cur_new_count > 0) {
2392                         qg->rfer += num_bytes;
2393                         qg->rfer_cmpr += num_bytes;
2394                         dirty = true;
2395                 }
2396                 if (cur_old_count > 0 && cur_new_count == 0) {
2397                         qg->rfer -= num_bytes;
2398                         qg->rfer_cmpr -= num_bytes;
2399                         dirty = true;
2400                 }
2401
2402                 /* Excl update part */
2403                 /* Exclusive/none -> shared case */
2404                 if (cur_old_count == nr_old_roots &&
2405                     cur_new_count < nr_new_roots) {
2406                         /* Exclusive -> shared */
2407                         if (cur_old_count != 0) {
2408                                 qg->excl -= num_bytes;
2409                                 qg->excl_cmpr -= num_bytes;
2410                                 dirty = true;
2411                         }
2412                 }
2413
2414                 /* Shared -> exclusive/none case */
2415                 if (cur_old_count < nr_old_roots &&
2416                     cur_new_count == nr_new_roots) {
2417                         /* Shared->exclusive */
2418                         if (cur_new_count != 0) {
2419                                 qg->excl += num_bytes;
2420                                 qg->excl_cmpr += num_bytes;
2421                                 dirty = true;
2422                         }
2423                 }
2424
2425                 /* Exclusive/none -> exclusive/none case */
2426                 if (cur_old_count == nr_old_roots &&
2427                     cur_new_count == nr_new_roots) {
2428                         if (cur_old_count == 0) {
2429                                 /* None -> exclusive/none */
2430
2431                                 if (cur_new_count != 0) {
2432                                         /* None -> exclusive */
2433                                         qg->excl += num_bytes;
2434                                         qg->excl_cmpr += num_bytes;
2435                                         dirty = true;
2436                                 }
2437                                 /* None -> none, nothing changed */
2438                         } else {
2439                                 /* Exclusive -> exclusive/none */
2440
2441                                 if (cur_new_count == 0) {
2442                                         /* Exclusive -> none */
2443                                         qg->excl -= num_bytes;
2444                                         qg->excl_cmpr -= num_bytes;
2445                                         dirty = true;
2446                                 }
2447                                 /* Exclusive -> exclusive, nothing changed */
2448                         }
2449                 }
2450
2451                 if (dirty)
2452                         qgroup_dirty(fs_info, qg);
2453         }
2454         return 0;
2455 }
2456
2457 /*
2458  * Check if the @roots potentially is a list of fs tree roots
2459  *
2460  * Return 0 for definitely not a fs/subvol tree roots ulist
2461  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2462  *          one as well)
2463  */
2464 static int maybe_fs_roots(struct ulist *roots)
2465 {
2466         struct ulist_node *unode;
2467         struct ulist_iterator uiter;
2468
2469         /* Empty one, still possible for fs roots */
2470         if (!roots || roots->nnodes == 0)
2471                 return 1;
2472
2473         ULIST_ITER_INIT(&uiter);
2474         unode = ulist_next(roots, &uiter);
2475         if (!unode)
2476                 return 1;
2477
2478         /*
2479          * If it contains fs tree roots, then it must belong to fs/subvol
2480          * trees.
2481          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2482          */
2483         return is_fstree(unode->val);
2484 }
2485
2486 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2487                                 u64 num_bytes, struct ulist *old_roots,
2488                                 struct ulist *new_roots)
2489 {
2490         struct btrfs_fs_info *fs_info = trans->fs_info;
2491         struct ulist *qgroups = NULL;
2492         struct ulist *tmp = NULL;
2493         u64 seq;
2494         u64 nr_new_roots = 0;
2495         u64 nr_old_roots = 0;
2496         int ret = 0;
2497
2498         /*
2499          * If quotas get disabled meanwhile, the resouces need to be freed and
2500          * we can't just exit here.
2501          */
2502         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2503                 goto out_free;
2504
2505         if (new_roots) {
2506                 if (!maybe_fs_roots(new_roots))
2507                         goto out_free;
2508                 nr_new_roots = new_roots->nnodes;
2509         }
2510         if (old_roots) {
2511                 if (!maybe_fs_roots(old_roots))
2512                         goto out_free;
2513                 nr_old_roots = old_roots->nnodes;
2514         }
2515
2516         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2517         if (nr_old_roots == 0 && nr_new_roots == 0)
2518                 goto out_free;
2519
2520         BUG_ON(!fs_info->quota_root);
2521
2522         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2523                                         num_bytes, nr_old_roots, nr_new_roots);
2524
2525         qgroups = ulist_alloc(GFP_NOFS);
2526         if (!qgroups) {
2527                 ret = -ENOMEM;
2528                 goto out_free;
2529         }
2530         tmp = ulist_alloc(GFP_NOFS);
2531         if (!tmp) {
2532                 ret = -ENOMEM;
2533                 goto out_free;
2534         }
2535
2536         mutex_lock(&fs_info->qgroup_rescan_lock);
2537         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2538                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2539                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2540                         ret = 0;
2541                         goto out_free;
2542                 }
2543         }
2544         mutex_unlock(&fs_info->qgroup_rescan_lock);
2545
2546         spin_lock(&fs_info->qgroup_lock);
2547         seq = fs_info->qgroup_seq;
2548
2549         /* Update old refcnts using old_roots */
2550         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2551                                    UPDATE_OLD);
2552         if (ret < 0)
2553                 goto out;
2554
2555         /* Update new refcnts using new_roots */
2556         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2557                                    UPDATE_NEW);
2558         if (ret < 0)
2559                 goto out;
2560
2561         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2562                                num_bytes, seq);
2563
2564         /*
2565          * Bump qgroup_seq to avoid seq overlap
2566          */
2567         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2568 out:
2569         spin_unlock(&fs_info->qgroup_lock);
2570 out_free:
2571         ulist_free(tmp);
2572         ulist_free(qgroups);
2573         ulist_free(old_roots);
2574         ulist_free(new_roots);
2575         return ret;
2576 }
2577
2578 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2579 {
2580         struct btrfs_fs_info *fs_info = trans->fs_info;
2581         struct btrfs_qgroup_extent_record *record;
2582         struct btrfs_delayed_ref_root *delayed_refs;
2583         struct ulist *new_roots = NULL;
2584         struct rb_node *node;
2585         u64 num_dirty_extents = 0;
2586         u64 qgroup_to_skip;
2587         int ret = 0;
2588
2589         delayed_refs = &trans->transaction->delayed_refs;
2590         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2591         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2592                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2593                                   node);
2594
2595                 num_dirty_extents++;
2596                 trace_btrfs_qgroup_account_extents(fs_info, record);
2597
2598                 if (!ret) {
2599                         /*
2600                          * Old roots should be searched when inserting qgroup
2601                          * extent record
2602                          */
2603                         if (WARN_ON(!record->old_roots)) {
2604                                 /* Search commit root to find old_roots */
2605                                 ret = btrfs_find_all_roots(NULL, fs_info,
2606                                                 record->bytenr, 0,
2607                                                 &record->old_roots, false);
2608                                 if (ret < 0)
2609                                         goto cleanup;
2610                         }
2611
2612                         /* Free the reserved data space */
2613                         btrfs_qgroup_free_refroot(fs_info,
2614                                         record->data_rsv_refroot,
2615                                         record->data_rsv,
2616                                         BTRFS_QGROUP_RSV_DATA);
2617                         /*
2618                          * Use SEQ_LAST as time_seq to do special search, which
2619                          * doesn't lock tree or delayed_refs and search current
2620                          * root. It's safe inside commit_transaction().
2621                          */
2622                         ret = btrfs_find_all_roots(trans, fs_info,
2623                                 record->bytenr, SEQ_LAST, &new_roots, false);
2624                         if (ret < 0)
2625                                 goto cleanup;
2626                         if (qgroup_to_skip) {
2627                                 ulist_del(new_roots, qgroup_to_skip, 0);
2628                                 ulist_del(record->old_roots, qgroup_to_skip,
2629                                           0);
2630                         }
2631                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2632                                                           record->num_bytes,
2633                                                           record->old_roots,
2634                                                           new_roots);
2635                         record->old_roots = NULL;
2636                         new_roots = NULL;
2637                 }
2638 cleanup:
2639                 ulist_free(record->old_roots);
2640                 ulist_free(new_roots);
2641                 new_roots = NULL;
2642                 rb_erase(node, &delayed_refs->dirty_extent_root);
2643                 kfree(record);
2644
2645         }
2646         trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2647                                        num_dirty_extents);
2648         return ret;
2649 }
2650
2651 /*
2652  * called from commit_transaction. Writes all changed qgroups to disk.
2653  */
2654 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2655 {
2656         struct btrfs_fs_info *fs_info = trans->fs_info;
2657         int ret = 0;
2658
2659         if (!fs_info->quota_root)
2660                 return ret;
2661
2662         spin_lock(&fs_info->qgroup_lock);
2663         while (!list_empty(&fs_info->dirty_qgroups)) {
2664                 struct btrfs_qgroup *qgroup;
2665                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2666                                           struct btrfs_qgroup, dirty);
2667                 list_del_init(&qgroup->dirty);
2668                 spin_unlock(&fs_info->qgroup_lock);
2669                 ret = update_qgroup_info_item(trans, qgroup);
2670                 if (ret)
2671                         fs_info->qgroup_flags |=
2672                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2673                 ret = update_qgroup_limit_item(trans, qgroup);
2674                 if (ret)
2675                         fs_info->qgroup_flags |=
2676                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2677                 spin_lock(&fs_info->qgroup_lock);
2678         }
2679         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2680                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2681         else
2682                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2683         spin_unlock(&fs_info->qgroup_lock);
2684
2685         ret = update_qgroup_status_item(trans);
2686         if (ret)
2687                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2688
2689         return ret;
2690 }
2691
2692 /*
2693  * Copy the accounting information between qgroups. This is necessary
2694  * when a snapshot or a subvolume is created. Throwing an error will
2695  * cause a transaction abort so we take extra care here to only error
2696  * when a readonly fs is a reasonable outcome.
2697  */
2698 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2699                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2700 {
2701         int ret = 0;
2702         int i;
2703         u64 *i_qgroups;
2704         bool committing = false;
2705         struct btrfs_fs_info *fs_info = trans->fs_info;
2706         struct btrfs_root *quota_root;
2707         struct btrfs_qgroup *srcgroup;
2708         struct btrfs_qgroup *dstgroup;
2709         bool need_rescan = false;
2710         u32 level_size = 0;
2711         u64 nums;
2712
2713         /*
2714          * There are only two callers of this function.
2715          *
2716          * One in create_subvol() in the ioctl context, which needs to hold
2717          * the qgroup_ioctl_lock.
2718          *
2719          * The other one in create_pending_snapshot() where no other qgroup
2720          * code can modify the fs as they all need to either start a new trans
2721          * or hold a trans handler, thus we don't need to hold
2722          * qgroup_ioctl_lock.
2723          * This would avoid long and complex lock chain and make lockdep happy.
2724          */
2725         spin_lock(&fs_info->trans_lock);
2726         if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2727                 committing = true;
2728         spin_unlock(&fs_info->trans_lock);
2729
2730         if (!committing)
2731                 mutex_lock(&fs_info->qgroup_ioctl_lock);
2732         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2733                 goto out;
2734
2735         quota_root = fs_info->quota_root;
2736         if (!quota_root) {
2737                 ret = -EINVAL;
2738                 goto out;
2739         }
2740
2741         if (inherit) {
2742                 i_qgroups = (u64 *)(inherit + 1);
2743                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2744                        2 * inherit->num_excl_copies;
2745                 for (i = 0; i < nums; ++i) {
2746                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2747
2748                         /*
2749                          * Zero out invalid groups so we can ignore
2750                          * them later.
2751                          */
2752                         if (!srcgroup ||
2753                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2754                                 *i_qgroups = 0ULL;
2755
2756                         ++i_qgroups;
2757                 }
2758         }
2759
2760         /*
2761          * create a tracking group for the subvol itself
2762          */
2763         ret = add_qgroup_item(trans, quota_root, objectid);
2764         if (ret)
2765                 goto out;
2766
2767         /*
2768          * add qgroup to all inherited groups
2769          */
2770         if (inherit) {
2771                 i_qgroups = (u64 *)(inherit + 1);
2772                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2773                         if (*i_qgroups == 0)
2774                                 continue;
2775                         ret = add_qgroup_relation_item(trans, objectid,
2776                                                        *i_qgroups);
2777                         if (ret && ret != -EEXIST)
2778                                 goto out;
2779                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2780                                                        objectid);
2781                         if (ret && ret != -EEXIST)
2782                                 goto out;
2783                 }
2784                 ret = 0;
2785         }
2786
2787
2788         spin_lock(&fs_info->qgroup_lock);
2789
2790         dstgroup = add_qgroup_rb(fs_info, objectid);
2791         if (IS_ERR(dstgroup)) {
2792                 ret = PTR_ERR(dstgroup);
2793                 goto unlock;
2794         }
2795
2796         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2797                 dstgroup->lim_flags = inherit->lim.flags;
2798                 dstgroup->max_rfer = inherit->lim.max_rfer;
2799                 dstgroup->max_excl = inherit->lim.max_excl;
2800                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2801                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2802
2803                 ret = update_qgroup_limit_item(trans, dstgroup);
2804                 if (ret) {
2805                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2806                         btrfs_info(fs_info,
2807                                    "unable to update quota limit for %llu",
2808                                    dstgroup->qgroupid);
2809                         goto unlock;
2810                 }
2811         }
2812
2813         if (srcid) {
2814                 srcgroup = find_qgroup_rb(fs_info, srcid);
2815                 if (!srcgroup)
2816                         goto unlock;
2817
2818                 /*
2819                  * We call inherit after we clone the root in order to make sure
2820                  * our counts don't go crazy, so at this point the only
2821                  * difference between the two roots should be the root node.
2822                  */
2823                 level_size = fs_info->nodesize;
2824                 dstgroup->rfer = srcgroup->rfer;
2825                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2826                 dstgroup->excl = level_size;
2827                 dstgroup->excl_cmpr = level_size;
2828                 srcgroup->excl = level_size;
2829                 srcgroup->excl_cmpr = level_size;
2830
2831                 /* inherit the limit info */
2832                 dstgroup->lim_flags = srcgroup->lim_flags;
2833                 dstgroup->max_rfer = srcgroup->max_rfer;
2834                 dstgroup->max_excl = srcgroup->max_excl;
2835                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2836                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2837
2838                 qgroup_dirty(fs_info, dstgroup);
2839                 qgroup_dirty(fs_info, srcgroup);
2840         }
2841
2842         if (!inherit)
2843                 goto unlock;
2844
2845         i_qgroups = (u64 *)(inherit + 1);
2846         for (i = 0; i < inherit->num_qgroups; ++i) {
2847                 if (*i_qgroups) {
2848                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2849                         if (ret)
2850                                 goto unlock;
2851                 }
2852                 ++i_qgroups;
2853
2854                 /*
2855                  * If we're doing a snapshot, and adding the snapshot to a new
2856                  * qgroup, the numbers are guaranteed to be incorrect.
2857                  */
2858                 if (srcid)
2859                         need_rescan = true;
2860         }
2861
2862         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2863                 struct btrfs_qgroup *src;
2864                 struct btrfs_qgroup *dst;
2865
2866                 if (!i_qgroups[0] || !i_qgroups[1])
2867                         continue;
2868
2869                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2870                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2871
2872                 if (!src || !dst) {
2873                         ret = -EINVAL;
2874                         goto unlock;
2875                 }
2876
2877                 dst->rfer = src->rfer - level_size;
2878                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2879
2880                 /* Manually tweaking numbers certainly needs a rescan */
2881                 need_rescan = true;
2882         }
2883         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2884                 struct btrfs_qgroup *src;
2885                 struct btrfs_qgroup *dst;
2886
2887                 if (!i_qgroups[0] || !i_qgroups[1])
2888                         continue;
2889
2890                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2891                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2892
2893                 if (!src || !dst) {
2894                         ret = -EINVAL;
2895                         goto unlock;
2896                 }
2897
2898                 dst->excl = src->excl + level_size;
2899                 dst->excl_cmpr = src->excl_cmpr + level_size;
2900                 need_rescan = true;
2901         }
2902
2903 unlock:
2904         spin_unlock(&fs_info->qgroup_lock);
2905         if (!ret)
2906                 ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
2907 out:
2908         if (!committing)
2909                 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2910         if (need_rescan)
2911                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2912         return ret;
2913 }
2914
2915 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2916 {
2917         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2918             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2919                 return false;
2920
2921         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2922             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2923                 return false;
2924
2925         return true;
2926 }
2927
2928 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2929                           enum btrfs_qgroup_rsv_type type)
2930 {
2931         struct btrfs_qgroup *qgroup;
2932         struct btrfs_fs_info *fs_info = root->fs_info;
2933         u64 ref_root = root->root_key.objectid;
2934         int ret = 0;
2935         struct ulist_node *unode;
2936         struct ulist_iterator uiter;
2937
2938         if (!is_fstree(ref_root))
2939                 return 0;
2940
2941         if (num_bytes == 0)
2942                 return 0;
2943
2944         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2945             capable(CAP_SYS_RESOURCE))
2946                 enforce = false;
2947
2948         spin_lock(&fs_info->qgroup_lock);
2949         if (!fs_info->quota_root)
2950                 goto out;
2951
2952         qgroup = find_qgroup_rb(fs_info, ref_root);
2953         if (!qgroup)
2954                 goto out;
2955
2956         /*
2957          * in a first step, we check all affected qgroups if any limits would
2958          * be exceeded
2959          */
2960         ulist_reinit(fs_info->qgroup_ulist);
2961         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2962                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2963         if (ret < 0)
2964                 goto out;
2965         ULIST_ITER_INIT(&uiter);
2966         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2967                 struct btrfs_qgroup *qg;
2968                 struct btrfs_qgroup_list *glist;
2969
2970                 qg = unode_aux_to_qgroup(unode);
2971
2972                 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2973                         ret = -EDQUOT;
2974                         goto out;
2975                 }
2976
2977                 list_for_each_entry(glist, &qg->groups, next_group) {
2978                         ret = ulist_add(fs_info->qgroup_ulist,
2979                                         glist->group->qgroupid,
2980                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2981                         if (ret < 0)
2982                                 goto out;
2983                 }
2984         }
2985         ret = 0;
2986         /*
2987          * no limits exceeded, now record the reservation into all qgroups
2988          */
2989         ULIST_ITER_INIT(&uiter);
2990         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2991                 struct btrfs_qgroup *qg;
2992
2993                 qg = unode_aux_to_qgroup(unode);
2994
2995                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2996         }
2997
2998 out:
2999         spin_unlock(&fs_info->qgroup_lock);
3000         return ret;
3001 }
3002
3003 /*
3004  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
3005  * qgroup).
3006  *
3007  * Will handle all higher level qgroup too.
3008  *
3009  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3010  * This special case is only used for META_PERTRANS type.
3011  */
3012 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3013                                u64 ref_root, u64 num_bytes,
3014                                enum btrfs_qgroup_rsv_type type)
3015 {
3016         struct btrfs_qgroup *qgroup;
3017         struct ulist_node *unode;
3018         struct ulist_iterator uiter;
3019         int ret = 0;
3020
3021         if (!is_fstree(ref_root))
3022                 return;
3023
3024         if (num_bytes == 0)
3025                 return;
3026
3027         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3028                 WARN(1, "%s: Invalid type to free", __func__);
3029                 return;
3030         }
3031         spin_lock(&fs_info->qgroup_lock);
3032
3033         if (!fs_info->quota_root)
3034                 goto out;
3035
3036         qgroup = find_qgroup_rb(fs_info, ref_root);
3037         if (!qgroup)
3038                 goto out;
3039
3040         if (num_bytes == (u64)-1)
3041                 /*
3042                  * We're freeing all pertrans rsv, get reserved value from
3043                  * level 0 qgroup as real num_bytes to free.
3044                  */
3045                 num_bytes = qgroup->rsv.values[type];
3046
3047         ulist_reinit(fs_info->qgroup_ulist);
3048         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3049                         qgroup_to_aux(qgroup), GFP_ATOMIC);
3050         if (ret < 0)
3051                 goto out;
3052         ULIST_ITER_INIT(&uiter);
3053         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3054                 struct btrfs_qgroup *qg;
3055                 struct btrfs_qgroup_list *glist;
3056
3057                 qg = unode_aux_to_qgroup(unode);
3058
3059                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3060
3061                 list_for_each_entry(glist, &qg->groups, next_group) {
3062                         ret = ulist_add(fs_info->qgroup_ulist,
3063                                         glist->group->qgroupid,
3064                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3065                         if (ret < 0)
3066                                 goto out;
3067                 }
3068         }
3069
3070 out:
3071         spin_unlock(&fs_info->qgroup_lock);
3072 }
3073
3074 /*
3075  * Check if the leaf is the last leaf. Which means all node pointers
3076  * are at their last position.
3077  */
3078 static bool is_last_leaf(struct btrfs_path *path)
3079 {
3080         int i;
3081
3082         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3083                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3084                         return false;
3085         }
3086         return true;
3087 }
3088
3089 /*
3090  * returns < 0 on error, 0 when more leafs are to be scanned.
3091  * returns 1 when done.
3092  */
3093 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3094                               struct btrfs_path *path)
3095 {
3096         struct btrfs_fs_info *fs_info = trans->fs_info;
3097         struct btrfs_key found;
3098         struct extent_buffer *scratch_leaf = NULL;
3099         struct ulist *roots = NULL;
3100         u64 num_bytes;
3101         bool done;
3102         int slot;
3103         int ret;
3104
3105         mutex_lock(&fs_info->qgroup_rescan_lock);
3106         ret = btrfs_search_slot_for_read(fs_info->extent_root,
3107                                          &fs_info->qgroup_rescan_progress,
3108                                          path, 1, 0);
3109
3110         btrfs_debug(fs_info,
3111                 "current progress key (%llu %u %llu), search_slot ret %d",
3112                 fs_info->qgroup_rescan_progress.objectid,
3113                 fs_info->qgroup_rescan_progress.type,
3114                 fs_info->qgroup_rescan_progress.offset, ret);
3115
3116         if (ret) {
3117                 /*
3118                  * The rescan is about to end, we will not be scanning any
3119                  * further blocks. We cannot unset the RESCAN flag here, because
3120                  * we want to commit the transaction if everything went well.
3121                  * To make the live accounting work in this phase, we set our
3122                  * scan progress pointer such that every real extent objectid
3123                  * will be smaller.
3124                  */
3125                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3126                 btrfs_release_path(path);
3127                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3128                 return ret;
3129         }
3130         done = is_last_leaf(path);
3131
3132         btrfs_item_key_to_cpu(path->nodes[0], &found,
3133                               btrfs_header_nritems(path->nodes[0]) - 1);
3134         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3135
3136         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3137         if (!scratch_leaf) {
3138                 ret = -ENOMEM;
3139                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3140                 goto out;
3141         }
3142         slot = path->slots[0];
3143         btrfs_release_path(path);
3144         mutex_unlock(&fs_info->qgroup_rescan_lock);
3145
3146         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3147                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3148                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3149                     found.type != BTRFS_METADATA_ITEM_KEY)
3150                         continue;
3151                 if (found.type == BTRFS_METADATA_ITEM_KEY)
3152                         num_bytes = fs_info->nodesize;
3153                 else
3154                         num_bytes = found.offset;
3155
3156                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3157                                            &roots, false);
3158                 if (ret < 0)
3159                         goto out;
3160                 /* For rescan, just pass old_roots as NULL */
3161                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3162                                                   num_bytes, NULL, roots);
3163                 if (ret < 0)
3164                         goto out;
3165         }
3166 out:
3167         if (scratch_leaf)
3168                 free_extent_buffer(scratch_leaf);
3169
3170         if (done && !ret) {
3171                 ret = 1;
3172                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3173         }
3174         return ret;
3175 }
3176
3177 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3178 {
3179         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3180                                                      qgroup_rescan_work);
3181         struct btrfs_path *path;
3182         struct btrfs_trans_handle *trans = NULL;
3183         int err = -ENOMEM;
3184         int ret = 0;
3185
3186         path = btrfs_alloc_path();
3187         if (!path)
3188                 goto out;
3189         /*
3190          * Rescan should only search for commit root, and any later difference
3191          * should be recorded by qgroup
3192          */
3193         path->search_commit_root = 1;
3194         path->skip_locking = 1;
3195
3196         err = 0;
3197         while (!err && !btrfs_fs_closing(fs_info)) {
3198                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3199                 if (IS_ERR(trans)) {
3200                         err = PTR_ERR(trans);
3201                         break;
3202                 }
3203                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3204                         err = -EINTR;
3205                 } else {
3206                         err = qgroup_rescan_leaf(trans, path);
3207                 }
3208                 if (err > 0)
3209                         btrfs_commit_transaction(trans);
3210                 else
3211                         btrfs_end_transaction(trans);
3212         }
3213
3214 out:
3215         btrfs_free_path(path);
3216
3217         mutex_lock(&fs_info->qgroup_rescan_lock);
3218         if (err > 0 &&
3219             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3220                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3221         } else if (err < 0) {
3222                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3223         }
3224         mutex_unlock(&fs_info->qgroup_rescan_lock);
3225
3226         /*
3227          * only update status, since the previous part has already updated the
3228          * qgroup info.
3229          */
3230         trans = btrfs_start_transaction(fs_info->quota_root, 1);
3231         if (IS_ERR(trans)) {
3232                 err = PTR_ERR(trans);
3233                 trans = NULL;
3234                 btrfs_err(fs_info,
3235                           "fail to start transaction for status update: %d",
3236                           err);
3237         }
3238
3239         mutex_lock(&fs_info->qgroup_rescan_lock);
3240         if (!btrfs_fs_closing(fs_info))
3241                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3242         if (trans) {
3243                 ret = update_qgroup_status_item(trans);
3244                 if (ret < 0) {
3245                         err = ret;
3246                         btrfs_err(fs_info, "fail to update qgroup status: %d",
3247                                   err);
3248                 }
3249         }
3250         fs_info->qgroup_rescan_running = false;
3251         complete_all(&fs_info->qgroup_rescan_completion);
3252         mutex_unlock(&fs_info->qgroup_rescan_lock);
3253
3254         if (!trans)
3255                 return;
3256
3257         btrfs_end_transaction(trans);
3258
3259         if (btrfs_fs_closing(fs_info)) {
3260                 btrfs_info(fs_info, "qgroup scan paused");
3261         } else if (err >= 0) {
3262                 btrfs_info(fs_info, "qgroup scan completed%s",
3263                         err > 0 ? " (inconsistency flag cleared)" : "");
3264         } else {
3265                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3266         }
3267 }
3268
3269 /*
3270  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3271  * memory required for the rescan context.
3272  */
3273 static int
3274 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3275                    int init_flags)
3276 {
3277         int ret = 0;
3278
3279         if (!init_flags) {
3280                 /* we're resuming qgroup rescan at mount time */
3281                 if (!(fs_info->qgroup_flags &
3282                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3283                         btrfs_warn(fs_info,
3284                         "qgroup rescan init failed, qgroup rescan is not queued");
3285                         ret = -EINVAL;
3286                 } else if (!(fs_info->qgroup_flags &
3287                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3288                         btrfs_warn(fs_info,
3289                         "qgroup rescan init failed, qgroup is not enabled");
3290                         ret = -EINVAL;
3291                 }
3292
3293                 if (ret)
3294                         return ret;
3295         }
3296
3297         mutex_lock(&fs_info->qgroup_rescan_lock);
3298
3299         if (init_flags) {
3300                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3301                         btrfs_warn(fs_info,
3302                                    "qgroup rescan is already in progress");
3303                         ret = -EINPROGRESS;
3304                 } else if (!(fs_info->qgroup_flags &
3305                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3306                         btrfs_warn(fs_info,
3307                         "qgroup rescan init failed, qgroup is not enabled");
3308                         ret = -EINVAL;
3309                 }
3310
3311                 if (ret) {
3312                         mutex_unlock(&fs_info->qgroup_rescan_lock);
3313                         return ret;
3314                 }
3315                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3316         }
3317
3318         memset(&fs_info->qgroup_rescan_progress, 0,
3319                 sizeof(fs_info->qgroup_rescan_progress));
3320         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3321         init_completion(&fs_info->qgroup_rescan_completion);
3322         mutex_unlock(&fs_info->qgroup_rescan_lock);
3323
3324         btrfs_init_work(&fs_info->qgroup_rescan_work,
3325                         btrfs_qgroup_rescan_worker, NULL, NULL);
3326         return 0;
3327 }
3328
3329 static void
3330 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3331 {
3332         struct rb_node *n;
3333         struct btrfs_qgroup *qgroup;
3334
3335         spin_lock(&fs_info->qgroup_lock);
3336         /* clear all current qgroup tracking information */
3337         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3338                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3339                 qgroup->rfer = 0;
3340                 qgroup->rfer_cmpr = 0;
3341                 qgroup->excl = 0;
3342                 qgroup->excl_cmpr = 0;
3343                 qgroup_dirty(fs_info, qgroup);
3344         }
3345         spin_unlock(&fs_info->qgroup_lock);
3346 }
3347
3348 int
3349 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3350 {
3351         int ret = 0;
3352         struct btrfs_trans_handle *trans;
3353
3354         ret = qgroup_rescan_init(fs_info, 0, 1);
3355         if (ret)
3356                 return ret;
3357
3358         /*
3359          * We have set the rescan_progress to 0, which means no more
3360          * delayed refs will be accounted by btrfs_qgroup_account_ref.
3361          * However, btrfs_qgroup_account_ref may be right after its call
3362          * to btrfs_find_all_roots, in which case it would still do the
3363          * accounting.
3364          * To solve this, we're committing the transaction, which will
3365          * ensure we run all delayed refs and only after that, we are
3366          * going to clear all tracking information for a clean start.
3367          */
3368
3369         trans = btrfs_join_transaction(fs_info->fs_root);
3370         if (IS_ERR(trans)) {
3371                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3372                 return PTR_ERR(trans);
3373         }
3374         ret = btrfs_commit_transaction(trans);
3375         if (ret) {
3376                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3377                 return ret;
3378         }
3379
3380         qgroup_rescan_zero_tracking(fs_info);
3381
3382         mutex_lock(&fs_info->qgroup_rescan_lock);
3383         fs_info->qgroup_rescan_running = true;
3384         btrfs_queue_work(fs_info->qgroup_rescan_workers,
3385                          &fs_info->qgroup_rescan_work);
3386         mutex_unlock(&fs_info->qgroup_rescan_lock);
3387
3388         return 0;
3389 }
3390
3391 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3392                                      bool interruptible)
3393 {
3394         int running;
3395         int ret = 0;
3396
3397         mutex_lock(&fs_info->qgroup_rescan_lock);
3398         running = fs_info->qgroup_rescan_running;
3399         mutex_unlock(&fs_info->qgroup_rescan_lock);
3400
3401         if (!running)
3402                 return 0;
3403
3404         if (interruptible)
3405                 ret = wait_for_completion_interruptible(
3406                                         &fs_info->qgroup_rescan_completion);
3407         else
3408                 wait_for_completion(&fs_info->qgroup_rescan_completion);
3409
3410         return ret;
3411 }
3412
3413 /*
3414  * this is only called from open_ctree where we're still single threaded, thus
3415  * locking is omitted here.
3416  */
3417 void
3418 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3419 {
3420         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3421                 mutex_lock(&fs_info->qgroup_rescan_lock);
3422                 fs_info->qgroup_rescan_running = true;
3423                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3424                                  &fs_info->qgroup_rescan_work);
3425                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3426         }
3427 }
3428
3429 #define rbtree_iterate_from_safe(node, next, start)                             \
3430        for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3431
3432 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3433                                   struct extent_changeset *reserved, u64 start,
3434                                   u64 len)
3435 {
3436         struct rb_node *node;
3437         struct rb_node *next;
3438         struct ulist_node *entry = NULL;
3439         int ret = 0;
3440
3441         node = reserved->range_changed.root.rb_node;
3442         while (node) {
3443                 entry = rb_entry(node, struct ulist_node, rb_node);
3444                 if (entry->val < start)
3445                         node = node->rb_right;
3446                 else if (entry)
3447                         node = node->rb_left;
3448                 else
3449                         break;
3450         }
3451
3452         /* Empty changeset */
3453         if (!entry)
3454                 return 0;
3455
3456         if (entry->val > start && rb_prev(&entry->rb_node))
3457                 entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3458                                  rb_node);
3459
3460         rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3461                 u64 entry_start;
3462                 u64 entry_end;
3463                 u64 entry_len;
3464                 int clear_ret;
3465
3466                 entry = rb_entry(node, struct ulist_node, rb_node);
3467                 entry_start = entry->val;
3468                 entry_end = entry->aux;
3469                 entry_len = entry_end - entry_start + 1;
3470
3471                 if (entry_start >= start + len)
3472                         break;
3473                 if (entry_start + entry_len <= start)
3474                         continue;
3475                 /*
3476                  * Now the entry is in [start, start + len), revert the
3477                  * EXTENT_QGROUP_RESERVED bit.
3478                  */
3479                 clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3480                                               entry_end, EXTENT_QGROUP_RESERVED);
3481                 if (!ret && clear_ret < 0)
3482                         ret = clear_ret;
3483
3484                 ulist_del(&reserved->range_changed, entry->val, entry->aux);
3485                 if (likely(reserved->bytes_changed >= entry_len)) {
3486                         reserved->bytes_changed -= entry_len;
3487                 } else {
3488                         WARN_ON(1);
3489                         reserved->bytes_changed = 0;
3490                 }
3491         }
3492
3493         return ret;
3494 }
3495
3496 /*
3497  * Try to free some space for qgroup.
3498  *
3499  * For qgroup, there are only 3 ways to free qgroup space:
3500  * - Flush nodatacow write
3501  *   Any nodatacow write will free its reserved data space at run_delalloc_range().
3502  *   In theory, we should only flush nodatacow inodes, but it's not yet
3503  *   possible, so we need to flush the whole root.
3504  *
3505  * - Wait for ordered extents
3506  *   When ordered extents are finished, their reserved metadata is finally
3507  *   converted to per_trans status, which can be freed by later commit
3508  *   transaction.
3509  *
3510  * - Commit transaction
3511  *   This would free the meta_per_trans space.
3512  *   In theory this shouldn't provide much space, but any more qgroup space
3513  *   is needed.
3514  */
3515 static int try_flush_qgroup(struct btrfs_root *root)
3516 {
3517         struct btrfs_trans_handle *trans;
3518         int ret;
3519
3520         /*
3521          * We don't want to run flush again and again, so if there is a running
3522          * one, we won't try to start a new flush, but exit directly.
3523          */
3524         if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3525                 wait_event(root->qgroup_flush_wait,
3526                         !test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3527                 return 0;
3528         }
3529
3530         ret = btrfs_start_delalloc_snapshot(root);
3531         if (ret < 0)
3532                 goto out;
3533         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3534
3535         trans = btrfs_join_transaction(root);
3536         if (IS_ERR(trans)) {
3537                 ret = PTR_ERR(trans);
3538                 goto out;
3539         }
3540
3541         ret = btrfs_commit_transaction(trans);
3542 out:
3543         clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3544         wake_up(&root->qgroup_flush_wait);
3545         return ret;
3546 }
3547
3548 static int qgroup_reserve_data(struct btrfs_inode *inode,
3549                         struct extent_changeset **reserved_ret, u64 start,
3550                         u64 len)
3551 {
3552         struct btrfs_root *root = inode->root;
3553         struct extent_changeset *reserved;
3554         bool new_reserved = false;
3555         u64 orig_reserved;
3556         u64 to_reserve;
3557         int ret;
3558
3559         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3560             !is_fstree(root->root_key.objectid) || len == 0)
3561                 return 0;
3562
3563         /* @reserved parameter is mandatory for qgroup */
3564         if (WARN_ON(!reserved_ret))
3565                 return -EINVAL;
3566         if (!*reserved_ret) {
3567                 new_reserved = true;
3568                 *reserved_ret = extent_changeset_alloc();
3569                 if (!*reserved_ret)
3570                         return -ENOMEM;
3571         }
3572         reserved = *reserved_ret;
3573         /* Record already reserved space */
3574         orig_reserved = reserved->bytes_changed;
3575         ret = set_record_extent_bits(&inode->io_tree, start,
3576                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3577
3578         /* Newly reserved space */
3579         to_reserve = reserved->bytes_changed - orig_reserved;
3580         trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3581                                         to_reserve, QGROUP_RESERVE);
3582         if (ret < 0)
3583                 goto out;
3584         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3585         if (ret < 0)
3586                 goto cleanup;
3587
3588         return ret;
3589
3590 cleanup:
3591         qgroup_unreserve_range(inode, reserved, start, len);
3592 out:
3593         if (new_reserved) {
3594                 extent_changeset_release(reserved);
3595                 kfree(reserved);
3596                 *reserved_ret = NULL;
3597         }
3598         return ret;
3599 }
3600
3601 /*
3602  * Reserve qgroup space for range [start, start + len).
3603  *
3604  * This function will either reserve space from related qgroups or do nothing
3605  * if the range is already reserved.
3606  *
3607  * Return 0 for successful reservation
3608  * Return <0 for error (including -EQUOT)
3609  *
3610  * NOTE: This function may sleep for memory allocation, dirty page flushing and
3611  *       commit transaction. So caller should not hold any dirty page locked.
3612  */
3613 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3614                         struct extent_changeset **reserved_ret, u64 start,
3615                         u64 len)
3616 {
3617         int ret;
3618
3619         ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3620         if (ret <= 0 && ret != -EDQUOT)
3621                 return ret;
3622
3623         ret = try_flush_qgroup(inode->root);
3624         if (ret < 0)
3625                 return ret;
3626         return qgroup_reserve_data(inode, reserved_ret, start, len);
3627 }
3628
3629 /* Free ranges specified by @reserved, normally in error path */
3630 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3631                         struct extent_changeset *reserved, u64 start, u64 len)
3632 {
3633         struct btrfs_root *root = inode->root;
3634         struct ulist_node *unode;
3635         struct ulist_iterator uiter;
3636         struct extent_changeset changeset;
3637         int freed = 0;
3638         int ret;
3639
3640         extent_changeset_init(&changeset);
3641         len = round_up(start + len, root->fs_info->sectorsize);
3642         start = round_down(start, root->fs_info->sectorsize);
3643
3644         ULIST_ITER_INIT(&uiter);
3645         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3646                 u64 range_start = unode->val;
3647                 /* unode->aux is the inclusive end */
3648                 u64 range_len = unode->aux - range_start + 1;
3649                 u64 free_start;
3650                 u64 free_len;
3651
3652                 extent_changeset_release(&changeset);
3653
3654                 /* Only free range in range [start, start + len) */
3655                 if (range_start >= start + len ||
3656                     range_start + range_len <= start)
3657                         continue;
3658                 free_start = max(range_start, start);
3659                 free_len = min(start + len, range_start + range_len) -
3660                            free_start;
3661                 /*
3662                  * TODO: To also modify reserved->ranges_reserved to reflect
3663                  * the modification.
3664                  *
3665                  * However as long as we free qgroup reserved according to
3666                  * EXTENT_QGROUP_RESERVED, we won't double free.
3667                  * So not need to rush.
3668                  */
3669                 ret = clear_record_extent_bits(&inode->io_tree, free_start,
3670                                 free_start + free_len - 1,
3671                                 EXTENT_QGROUP_RESERVED, &changeset);
3672                 if (ret < 0)
3673                         goto out;
3674                 freed += changeset.bytes_changed;
3675         }
3676         btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3677                                   BTRFS_QGROUP_RSV_DATA);
3678         ret = freed;
3679 out:
3680         extent_changeset_release(&changeset);
3681         return ret;
3682 }
3683
3684 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3685                         struct extent_changeset *reserved, u64 start, u64 len,
3686                         int free)
3687 {
3688         struct extent_changeset changeset;
3689         int trace_op = QGROUP_RELEASE;
3690         int ret;
3691
3692         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3693                 return 0;
3694
3695         /* In release case, we shouldn't have @reserved */
3696         WARN_ON(!free && reserved);
3697         if (free && reserved)
3698                 return qgroup_free_reserved_data(inode, reserved, start, len);
3699         extent_changeset_init(&changeset);
3700         ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3701                                        EXTENT_QGROUP_RESERVED, &changeset);
3702         if (ret < 0)
3703                 goto out;
3704
3705         if (free)
3706                 trace_op = QGROUP_FREE;
3707         trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3708                                         changeset.bytes_changed, trace_op);
3709         if (free)
3710                 btrfs_qgroup_free_refroot(inode->root->fs_info,
3711                                 inode->root->root_key.objectid,
3712                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3713         ret = changeset.bytes_changed;
3714 out:
3715         extent_changeset_release(&changeset);
3716         return ret;
3717 }
3718
3719 /*
3720  * Free a reserved space range from io_tree and related qgroups
3721  *
3722  * Should be called when a range of pages get invalidated before reaching disk.
3723  * Or for error cleanup case.
3724  * if @reserved is given, only reserved range in [@start, @start + @len) will
3725  * be freed.
3726  *
3727  * For data written to disk, use btrfs_qgroup_release_data().
3728  *
3729  * NOTE: This function may sleep for memory allocation.
3730  */
3731 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3732                         struct extent_changeset *reserved, u64 start, u64 len)
3733 {
3734         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3735 }
3736
3737 /*
3738  * Release a reserved space range from io_tree only.
3739  *
3740  * Should be called when a range of pages get written to disk and corresponding
3741  * FILE_EXTENT is inserted into corresponding root.
3742  *
3743  * Since new qgroup accounting framework will only update qgroup numbers at
3744  * commit_transaction() time, its reserved space shouldn't be freed from
3745  * related qgroups.
3746  *
3747  * But we should release the range from io_tree, to allow further write to be
3748  * COWed.
3749  *
3750  * NOTE: This function may sleep for memory allocation.
3751  */
3752 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3753 {
3754         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3755 }
3756
3757 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3758                               enum btrfs_qgroup_rsv_type type)
3759 {
3760         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3761             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3762                 return;
3763         if (num_bytes == 0)
3764                 return;
3765
3766         spin_lock(&root->qgroup_meta_rsv_lock);
3767         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3768                 root->qgroup_meta_rsv_prealloc += num_bytes;
3769         else
3770                 root->qgroup_meta_rsv_pertrans += num_bytes;
3771         spin_unlock(&root->qgroup_meta_rsv_lock);
3772 }
3773
3774 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3775                              enum btrfs_qgroup_rsv_type type)
3776 {
3777         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3778             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3779                 return 0;
3780         if (num_bytes == 0)
3781                 return 0;
3782
3783         spin_lock(&root->qgroup_meta_rsv_lock);
3784         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3785                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3786                                   num_bytes);
3787                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3788         } else {
3789                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3790                                   num_bytes);
3791                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3792         }
3793         spin_unlock(&root->qgroup_meta_rsv_lock);
3794         return num_bytes;
3795 }
3796
3797 static int qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3798                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3799 {
3800         struct btrfs_fs_info *fs_info = root->fs_info;
3801         int ret;
3802
3803         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3804             !is_fstree(root->root_key.objectid) || num_bytes == 0)
3805                 return 0;
3806
3807         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3808         trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3809         ret = qgroup_reserve(root, num_bytes, enforce, type);
3810         if (ret < 0)
3811                 return ret;
3812         /*
3813          * Record what we have reserved into root.
3814          *
3815          * To avoid quota disabled->enabled underflow.
3816          * In that case, we may try to free space we haven't reserved
3817          * (since quota was disabled), so record what we reserved into root.
3818          * And ensure later release won't underflow this number.
3819          */
3820         add_root_meta_rsv(root, num_bytes, type);
3821         return ret;
3822 }
3823
3824 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3825                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3826 {
3827         int ret;
3828
3829         ret = qgroup_reserve_meta(root, num_bytes, type, enforce);
3830         if (ret <= 0 && ret != -EDQUOT)
3831                 return ret;
3832
3833         ret = try_flush_qgroup(root);
3834         if (ret < 0)
3835                 return ret;
3836         return qgroup_reserve_meta(root, num_bytes, type, enforce);
3837 }
3838
3839 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3840 {
3841         struct btrfs_fs_info *fs_info = root->fs_info;
3842
3843         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3844             !is_fstree(root->root_key.objectid))
3845                 return;
3846
3847         /* TODO: Update trace point to handle such free */
3848         trace_qgroup_meta_free_all_pertrans(root);
3849         /* Special value -1 means to free all reserved space */
3850         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3851                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3852 }
3853
3854 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3855                               enum btrfs_qgroup_rsv_type type)
3856 {
3857         struct btrfs_fs_info *fs_info = root->fs_info;
3858
3859         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3860             !is_fstree(root->root_key.objectid))
3861                 return;
3862
3863         /*
3864          * reservation for META_PREALLOC can happen before quota is enabled,
3865          * which can lead to underflow.
3866          * Here ensure we will only free what we really have reserved.
3867          */
3868         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3869         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3870         trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3871         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3872                                   num_bytes, type);
3873 }
3874
3875 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3876                                 int num_bytes)
3877 {
3878         struct btrfs_qgroup *qgroup;
3879         struct ulist_node *unode;
3880         struct ulist_iterator uiter;
3881         int ret = 0;
3882
3883         if (num_bytes == 0)
3884                 return;
3885         if (!fs_info->quota_root)
3886                 return;
3887
3888         spin_lock(&fs_info->qgroup_lock);
3889         qgroup = find_qgroup_rb(fs_info, ref_root);
3890         if (!qgroup)
3891                 goto out;
3892         ulist_reinit(fs_info->qgroup_ulist);
3893         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3894                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3895         if (ret < 0)
3896                 goto out;
3897         ULIST_ITER_INIT(&uiter);
3898         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3899                 struct btrfs_qgroup *qg;
3900                 struct btrfs_qgroup_list *glist;
3901
3902                 qg = unode_aux_to_qgroup(unode);
3903
3904                 qgroup_rsv_release(fs_info, qg, num_bytes,
3905                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3906                 qgroup_rsv_add(fs_info, qg, num_bytes,
3907                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3908                 list_for_each_entry(glist, &qg->groups, next_group) {
3909                         ret = ulist_add(fs_info->qgroup_ulist,
3910                                         glist->group->qgroupid,
3911                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3912                         if (ret < 0)
3913                                 goto out;
3914                 }
3915         }
3916 out:
3917         spin_unlock(&fs_info->qgroup_lock);
3918 }
3919
3920 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3921 {
3922         struct btrfs_fs_info *fs_info = root->fs_info;
3923
3924         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3925             !is_fstree(root->root_key.objectid))
3926                 return;
3927         /* Same as btrfs_qgroup_free_meta_prealloc() */
3928         num_bytes = sub_root_meta_rsv(root, num_bytes,
3929                                       BTRFS_QGROUP_RSV_META_PREALLOC);
3930         trace_qgroup_meta_convert(root, num_bytes);
3931         qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3932 }
3933
3934 /*
3935  * Check qgroup reserved space leaking, normally at destroy inode
3936  * time
3937  */
3938 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
3939 {
3940         struct extent_changeset changeset;
3941         struct ulist_node *unode;
3942         struct ulist_iterator iter;
3943         int ret;
3944
3945         extent_changeset_init(&changeset);
3946         ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
3947                         EXTENT_QGROUP_RESERVED, &changeset);
3948
3949         WARN_ON(ret < 0);
3950         if (WARN_ON(changeset.bytes_changed)) {
3951                 ULIST_ITER_INIT(&iter);
3952                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3953                         btrfs_warn(inode->root->fs_info,
3954                 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
3955                                 btrfs_ino(inode), unode->val, unode->aux);
3956                 }
3957                 btrfs_qgroup_free_refroot(inode->root->fs_info,
3958                                 inode->root->root_key.objectid,
3959                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3960
3961         }
3962         extent_changeset_release(&changeset);
3963 }
3964
3965 void btrfs_qgroup_init_swapped_blocks(
3966         struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3967 {
3968         int i;
3969
3970         spin_lock_init(&swapped_blocks->lock);
3971         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3972                 swapped_blocks->blocks[i] = RB_ROOT;
3973         swapped_blocks->swapped = false;
3974 }
3975
3976 /*
3977  * Delete all swapped blocks record of @root.
3978  * Every record here means we skipped a full subtree scan for qgroup.
3979  *
3980  * Gets called when committing one transaction.
3981  */
3982 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3983 {
3984         struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3985         int i;
3986
3987         swapped_blocks = &root->swapped_blocks;
3988
3989         spin_lock(&swapped_blocks->lock);
3990         if (!swapped_blocks->swapped)
3991                 goto out;
3992         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3993                 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3994                 struct btrfs_qgroup_swapped_block *entry;
3995                 struct btrfs_qgroup_swapped_block *next;
3996
3997                 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3998                                                      node)
3999                         kfree(entry);
4000                 swapped_blocks->blocks[i] = RB_ROOT;
4001         }
4002         swapped_blocks->swapped = false;
4003 out:
4004         spin_unlock(&swapped_blocks->lock);
4005 }
4006
4007 /*
4008  * Add subtree roots record into @subvol_root.
4009  *
4010  * @subvol_root:        tree root of the subvolume tree get swapped
4011  * @bg:                 block group under balance
4012  * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4013  * @reloc_parent/slot:  pointer to the subtree root in reloc tree
4014  *                      BOTH POINTERS ARE BEFORE TREE SWAP
4015  * @last_snapshot:      last snapshot generation of the subvolume tree
4016  */
4017 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4018                 struct btrfs_root *subvol_root,
4019                 struct btrfs_block_group *bg,
4020                 struct extent_buffer *subvol_parent, int subvol_slot,
4021                 struct extent_buffer *reloc_parent, int reloc_slot,
4022                 u64 last_snapshot)
4023 {
4024         struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4025         struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4026         struct btrfs_qgroup_swapped_block *block;
4027         struct rb_node **cur;
4028         struct rb_node *parent = NULL;
4029         int level = btrfs_header_level(subvol_parent) - 1;
4030         int ret = 0;
4031
4032         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4033                 return 0;
4034
4035         if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4036             btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4037                 btrfs_err_rl(fs_info,
4038                 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4039                         __func__,
4040                         btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4041                         btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4042                 return -EUCLEAN;
4043         }
4044
4045         block = kmalloc(sizeof(*block), GFP_NOFS);
4046         if (!block) {
4047                 ret = -ENOMEM;
4048                 goto out;
4049         }
4050
4051         /*
4052          * @reloc_parent/slot is still before swap, while @block is going to
4053          * record the bytenr after swap, so we do the swap here.
4054          */
4055         block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4056         block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4057                                                              reloc_slot);
4058         block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4059         block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4060                                                             subvol_slot);
4061         block->last_snapshot = last_snapshot;
4062         block->level = level;
4063
4064         /*
4065          * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4066          * no one else can modify tree blocks thus we qgroup will not change
4067          * no matter the value of trace_leaf.
4068          */
4069         if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4070                 block->trace_leaf = true;
4071         else
4072                 block->trace_leaf = false;
4073         btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4074
4075         /* Insert @block into @blocks */
4076         spin_lock(&blocks->lock);
4077         cur = &blocks->blocks[level].rb_node;
4078         while (*cur) {
4079                 struct btrfs_qgroup_swapped_block *entry;
4080
4081                 parent = *cur;
4082                 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4083                                  node);
4084
4085                 if (entry->subvol_bytenr < block->subvol_bytenr) {
4086                         cur = &(*cur)->rb_left;
4087                 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
4088                         cur = &(*cur)->rb_right;
4089                 } else {
4090                         if (entry->subvol_generation !=
4091                                         block->subvol_generation ||
4092                             entry->reloc_bytenr != block->reloc_bytenr ||
4093                             entry->reloc_generation !=
4094                                         block->reloc_generation) {
4095                                 /*
4096                                  * Duplicated but mismatch entry found.
4097                                  * Shouldn't happen.
4098                                  *
4099                                  * Marking qgroup inconsistent should be enough
4100                                  * for end users.
4101                                  */
4102                                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4103                                 ret = -EEXIST;
4104                         }
4105                         kfree(block);
4106                         goto out_unlock;
4107                 }
4108         }
4109         rb_link_node(&block->node, parent, cur);
4110         rb_insert_color(&block->node, &blocks->blocks[level]);
4111         blocks->swapped = true;
4112 out_unlock:
4113         spin_unlock(&blocks->lock);
4114 out:
4115         if (ret < 0)
4116                 fs_info->qgroup_flags |=
4117                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4118         return ret;
4119 }
4120
4121 /*
4122  * Check if the tree block is a subtree root, and if so do the needed
4123  * delayed subtree trace for qgroup.
4124  *
4125  * This is called during btrfs_cow_block().
4126  */
4127 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4128                                          struct btrfs_root *root,
4129                                          struct extent_buffer *subvol_eb)
4130 {
4131         struct btrfs_fs_info *fs_info = root->fs_info;
4132         struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4133         struct btrfs_qgroup_swapped_block *block;
4134         struct extent_buffer *reloc_eb = NULL;
4135         struct rb_node *node;
4136         bool found = false;
4137         bool swapped = false;
4138         int level = btrfs_header_level(subvol_eb);
4139         int ret = 0;
4140         int i;
4141
4142         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4143                 return 0;
4144         if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4145                 return 0;
4146
4147         spin_lock(&blocks->lock);
4148         if (!blocks->swapped) {
4149                 spin_unlock(&blocks->lock);
4150                 return 0;
4151         }
4152         node = blocks->blocks[level].rb_node;
4153
4154         while (node) {
4155                 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4156                 if (block->subvol_bytenr < subvol_eb->start) {
4157                         node = node->rb_left;
4158                 } else if (block->subvol_bytenr > subvol_eb->start) {
4159                         node = node->rb_right;
4160                 } else {
4161                         found = true;
4162                         break;
4163                 }
4164         }
4165         if (!found) {
4166                 spin_unlock(&blocks->lock);
4167                 goto out;
4168         }
4169         /* Found one, remove it from @blocks first and update blocks->swapped */
4170         rb_erase(&block->node, &blocks->blocks[level]);
4171         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4172                 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4173                         swapped = true;
4174                         break;
4175                 }
4176         }
4177         blocks->swapped = swapped;
4178         spin_unlock(&blocks->lock);
4179
4180         /* Read out reloc subtree root */
4181         reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
4182                                    block->reloc_generation, block->level,
4183                                    &block->first_key);
4184         if (IS_ERR(reloc_eb)) {
4185                 ret = PTR_ERR(reloc_eb);
4186                 reloc_eb = NULL;
4187                 goto free_out;
4188         }
4189         if (!extent_buffer_uptodate(reloc_eb)) {
4190                 ret = -EIO;
4191                 goto free_out;
4192         }
4193
4194         ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4195                         block->last_snapshot, block->trace_leaf);
4196 free_out:
4197         kfree(block);
4198         free_extent_buffer(reloc_eb);
4199 out:
4200         if (ret < 0) {
4201                 btrfs_err_rl(fs_info,
4202                              "failed to account subtree at bytenr %llu: %d",
4203                              subvol_eb->start, ret);
4204                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4205         }
4206         return ret;
4207 }
4208
4209 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4210 {
4211         struct btrfs_qgroup_extent_record *entry;
4212         struct btrfs_qgroup_extent_record *next;
4213         struct rb_root *root;
4214
4215         root = &trans->delayed_refs.dirty_extent_root;
4216         rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4217                 ulist_free(entry->old_roots);
4218                 kfree(entry);
4219         }
4220 }