fs/btrfs/qgroup.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/sched.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/writeback.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/rbtree.h>
  24 #include <linux/slab.h>
  25 #include <linux/workqueue.h>
  26 #include <linux/btrfs.h>
  27
  28 #include "ctree.h"
  29 #include "transaction.h"
  30 #include "disk-io.h"
  31 #include "locking.h"
  32 #include "ulist.h"
  33 #include "backref.h"
  34 #include "extent_io.h"
  35 #include "qgroup.h"
  36
  37
  38 /* TODO XXX FIXME
  39  *  - subvol delete -> delete when ref goes to 0? delete limits also?
  40  *  - reorganize keys
  41  *  - compressed
  42  *  - sync
  43  *  - copy also limits on subvol creation
  44  *  - limit
  45  *  - caches fuer ulists
  46  *  - performance benchmarks
  47  *  - check all ioctl parameters
  48  */
  49
  50 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
  51                                            int mod)
  52 {
  53         if (qg->old_refcnt < seq)
  54                 qg->old_refcnt = seq;
  55         qg->old_refcnt += mod;
  56 }
  57
  58 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
  59                                            int mod)
  60 {
  61         if (qg->new_refcnt < seq)
  62                 qg->new_refcnt = seq;
  63         qg->new_refcnt += mod;
  64 }
  65
  66 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
  67 {
  68         if (qg->old_refcnt < seq)
  69                 return 0;
  70         return qg->old_refcnt - seq;
  71 }
  72
  73 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
  74 {
  75         if (qg->new_refcnt < seq)
  76                 return 0;
  77         return qg->new_refcnt - seq;
  78 }
  79
  80 /*
  81  * glue structure to represent the relations between qgroups.
  82  */
  83 struct btrfs_qgroup_list {
  84         struct list_head next_group;
  85         struct list_head next_member;
  86         struct btrfs_qgroup *group;
  87         struct btrfs_qgroup *member;
  88 };
  89
  90 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
  91 {
  92         return (u64)(uintptr_t)qg;
  93 }
  94
  95 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
  96 {
  97         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
  98 }
  99
 100 static int
 101 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
 102                    int init_flags);
 103 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
 104
 105 /* must be called with qgroup_ioctl_lock held */
 106 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
 107                                            u64 qgroupid)
 108 {
 109         struct rb_node *n = fs_info->qgroup_tree.rb_node;
 110         struct btrfs_qgroup *qgroup;
 111
 112         while (n) {
 113                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 114                 if (qgroup->qgroupid < qgroupid)
 115                         n = n->rb_left;
 116                 else if (qgroup->qgroupid > qgroupid)
 117                         n = n->rb_right;
 118                 else
 119                         return qgroup;
 120         }
 121         return NULL;
 122 }
 123
 124 /* must be called with qgroup_lock held */
 125 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
 126                                           u64 qgroupid)
 127 {
 128         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
 129         struct rb_node *parent = NULL;
 130         struct btrfs_qgroup *qgroup;
 131
 132         while (*p) {
 133                 parent = *p;
 134                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
 135
 136                 if (qgroup->qgroupid < qgroupid)
 137                         p = &(*p)->rb_left;
 138                 else if (qgroup->qgroupid > qgroupid)
 139                         p = &(*p)->rb_right;
 140                 else
 141                         return qgroup;
 142         }
 143
 144         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
 145         if (!qgroup)
 146                 return ERR_PTR(-ENOMEM);
 147
 148         qgroup->qgroupid = qgroupid;
 149         INIT_LIST_HEAD(&qgroup->groups);
 150         INIT_LIST_HEAD(&qgroup->members);
 151         INIT_LIST_HEAD(&qgroup->dirty);
 152
 153         rb_link_node(&qgroup->node, parent, p);
 154         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
 155
 156         return qgroup;
 157 }
 158
 159 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
 160 {
 161         struct btrfs_qgroup_list *list;
 162
 163         list_del(&qgroup->dirty);
 164         while (!list_empty(&qgroup->groups)) {
 165                 list = list_first_entry(&qgroup->groups,
 166                                         struct btrfs_qgroup_list, next_group);
 167                 list_del(&list->next_group);
 168                 list_del(&list->next_member);
 169                 kfree(list);
 170         }
 171
 172         while (!list_empty(&qgroup->members)) {
 173                 list = list_first_entry(&qgroup->members,
 174                                         struct btrfs_qgroup_list, next_member);
 175                 list_del(&list->next_group);
 176                 list_del(&list->next_member);
 177                 kfree(list);
 178         }
 179         kfree(qgroup);
 180 }
 181
 182 /* must be called with qgroup_lock held */
 183 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
 184 {
 185         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
 186
 187         if (!qgroup)
 188                 return -ENOENT;
 189
 190         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
 191         __del_qgroup_rb(qgroup);
 192         return 0;
 193 }
 194
 195 /* must be called with qgroup_lock held */
 196 static int add_relation_rb(struct btrfs_fs_info *fs_info,
 197                            u64 memberid, u64 parentid)
 198 {
 199         struct btrfs_qgroup *member;
 200         struct btrfs_qgroup *parent;
 201         struct btrfs_qgroup_list *list;
 202
 203         member = find_qgroup_rb(fs_info, memberid);
 204         parent = find_qgroup_rb(fs_info, parentid);
 205         if (!member || !parent)
 206                 return -ENOENT;
 207
 208         list = kzalloc(sizeof(*list), GFP_ATOMIC);
 209         if (!list)
 210                 return -ENOMEM;
 211
 212         list->group = parent;
 213         list->member = member;
 214         list_add_tail(&list->next_group, &member->groups);
 215         list_add_tail(&list->next_member, &parent->members);
 216
 217         return 0;
 218 }
 219
 220 /* must be called with qgroup_lock held */
 221 static int del_relation_rb(struct btrfs_fs_info *fs_info,
 222                            u64 memberid, u64 parentid)
 223 {
 224         struct btrfs_qgroup *member;
 225         struct btrfs_qgroup *parent;
 226         struct btrfs_qgroup_list *list;
 227
 228         member = find_qgroup_rb(fs_info, memberid);
 229         parent = find_qgroup_rb(fs_info, parentid);
 230         if (!member || !parent)
 231                 return -ENOENT;
 232
 233         list_for_each_entry(list, &member->groups, next_group) {
 234                 if (list->group == parent) {
 235                         list_del(&list->next_group);
 236                         list_del(&list->next_member);
 237                         kfree(list);
 238                         return 0;
 239                 }
 240         }
 241         return -ENOENT;
 242 }
 243
 244 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 245 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
 246                                u64 rfer, u64 excl)
 247 {
 248         struct btrfs_qgroup *qgroup;
 249
 250         qgroup = find_qgroup_rb(fs_info, qgroupid);
 251         if (!qgroup)
 252                 return -EINVAL;
 253         if (qgroup->rfer != rfer || qgroup->excl != excl)
 254                 return -EINVAL;
 255         return 0;
 256 }
 257 #endif
 258
 259 /*
 260  * The full config is read in one go, only called from open_ctree()
 261  * It doesn't use any locking, as at this point we're still single-threaded
 262  */
 263 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 264 {
 265         struct btrfs_key key;
 266         struct btrfs_key found_key;
 267         struct btrfs_root *quota_root = fs_info->quota_root;
 268         struct btrfs_path *path = NULL;
 269         struct extent_buffer *l;
 270         int slot;
 271         int ret = 0;
 272         u64 flags = 0;
 273         u64 rescan_progress = 0;
 274
 275         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 276                 return 0;
 277
 278         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 279         if (!fs_info->qgroup_ulist) {
 280                 ret = -ENOMEM;
 281                 goto out;
 282         }
 283
 284         path = btrfs_alloc_path();
 285         if (!path) {
 286                 ret = -ENOMEM;
 287                 goto out;
 288         }
 289
 290         /* default this to quota off, in case no status key is found */
 291         fs_info->qgroup_flags = 0;
 292
 293         /*
 294          * pass 1: read status, all qgroup infos and limits
 295          */
 296         key.objectid = 0;
 297         key.type = 0;
 298         key.offset = 0;
 299         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
 300         if (ret)
 301                 goto out;
 302
 303         while (1) {
 304                 struct btrfs_qgroup *qgroup;
 305
 306                 slot = path->slots[0];
 307                 l = path->nodes[0];
 308                 btrfs_item_key_to_cpu(l, &found_key, slot);
 309
 310                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
 311                         struct btrfs_qgroup_status_item *ptr;
 312
 313                         ptr = btrfs_item_ptr(l, slot,
 314                                              struct btrfs_qgroup_status_item);
 315
 316                         if (btrfs_qgroup_status_version(l, ptr) !=
 317                             BTRFS_QGROUP_STATUS_VERSION) {
 318                                 btrfs_err(fs_info,
 319                                  "old qgroup version, quota disabled");
 320                                 goto out;
 321                         }
 322                         if (btrfs_qgroup_status_generation(l, ptr) !=
 323                             fs_info->generation) {
 324                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 325                                 btrfs_err(fs_info,
 326                                         "qgroup generation mismatch, marked as inconsistent");
 327                         }
 328                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
 329                                                                           ptr);
 330                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
 331                         goto next1;
 332                 }
 333
 334                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
 335                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
 336                         goto next1;
 337
 338                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
 339                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
 340                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
 341                         btrfs_err(fs_info, "inconsistent qgroup config");
 342                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 343                 }
 344                 if (!qgroup) {
 345                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 346                         if (IS_ERR(qgroup)) {
 347                                 ret = PTR_ERR(qgroup);
 348                                 goto out;
 349                         }
 350                 }
 351                 switch (found_key.type) {
 352                 case BTRFS_QGROUP_INFO_KEY: {
 353                         struct btrfs_qgroup_info_item *ptr;
 354
 355                         ptr = btrfs_item_ptr(l, slot,
 356                                              struct btrfs_qgroup_info_item);
 357                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
 358                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
 359                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
 360                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
 361                         /* generation currently unused */
 362                         break;
 363                 }
 364                 case BTRFS_QGROUP_LIMIT_KEY: {
 365                         struct btrfs_qgroup_limit_item *ptr;
 366
 367                         ptr = btrfs_item_ptr(l, slot,
 368                                              struct btrfs_qgroup_limit_item);
 369                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
 370                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
 371                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
 372                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
 373                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
 374                         break;
 375                 }
 376                 }
 377 next1:
 378                 ret = btrfs_next_item(quota_root, path);
 379                 if (ret < 0)
 380                         goto out;
 381                 if (ret)
 382                         break;
 383         }
 384         btrfs_release_path(path);
 385
 386         /*
 387          * pass 2: read all qgroup relations
 388          */
 389         key.objectid = 0;
 390         key.type = BTRFS_QGROUP_RELATION_KEY;
 391         key.offset = 0;
 392         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
 393         if (ret)
 394                 goto out;
 395         while (1) {
 396                 slot = path->slots[0];
 397                 l = path->nodes[0];
 398                 btrfs_item_key_to_cpu(l, &found_key, slot);
 399
 400                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
 401                         goto next2;
 402
 403                 if (found_key.objectid > found_key.offset) {
 404                         /* parent <- member, not needed to build config */
 405                         /* FIXME should we omit the key completely? */
 406                         goto next2;
 407                 }
 408
 409                 ret = add_relation_rb(fs_info, found_key.objectid,
 410                                       found_key.offset);
 411                 if (ret == -ENOENT) {
 412                         btrfs_warn(fs_info,
 413                                 "orphan qgroup relation 0x%llx->0x%llx",
 414                                 found_key.objectid, found_key.offset);
 415                         ret = 0;        /* ignore the error */
 416                 }
 417                 if (ret)
 418                         goto out;
 419 next2:
 420                 ret = btrfs_next_item(quota_root, path);
 421                 if (ret < 0)
 422                         goto out;
 423                 if (ret)
 424                         break;
 425         }
 426 out:
 427         fs_info->qgroup_flags |= flags;
 428         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
 429                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 430         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
 431                  ret >= 0)
 432                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
 433         btrfs_free_path(path);
 434
 435         if (ret < 0) {
 436                 ulist_free(fs_info->qgroup_ulist);
 437                 fs_info->qgroup_ulist = NULL;
 438                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
 439         }
 440
 441         return ret < 0 ? ret : 0;
 442 }
 443
 444 /*
 445  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
 446  * first two are in single-threaded paths.And for the third one, we have set
 447  * quota_root to be null with qgroup_lock held before, so it is safe to clean
 448  * up the in-memory structures without qgroup_lock held.
 449  */
 450 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
 451 {
 452         struct rb_node *n;
 453         struct btrfs_qgroup *qgroup;
 454
 455         while ((n = rb_first(&fs_info->qgroup_tree))) {
 456                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
 457                 rb_erase(n, &fs_info->qgroup_tree);
 458                 __del_qgroup_rb(qgroup);
 459         }
 460         /*
 461          * we call btrfs_free_qgroup_config() when umounting
 462          * filesystem and disabling quota, so we set qgroup_ulist
 463          * to be null here to avoid double free.
 464          */
 465         ulist_free(fs_info->qgroup_ulist);
 466         fs_info->qgroup_ulist = NULL;
 467 }
 468
 469 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
 470                                     struct btrfs_root *quota_root,
 471                                     u64 src, u64 dst)
 472 {
 473         int ret;
 474         struct btrfs_path *path;
 475         struct btrfs_key key;
 476
 477         path = btrfs_alloc_path();
 478         if (!path)
 479                 return -ENOMEM;
 480
 481         key.objectid = src;
 482         key.type = BTRFS_QGROUP_RELATION_KEY;
 483         key.offset = dst;
 484
 485         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
 486
 487         btrfs_mark_buffer_dirty(path->nodes[0]);
 488
 489         btrfs_free_path(path);
 490         return ret;
 491 }
 492
 493 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
 494                                     struct btrfs_root *quota_root,
 495                                     u64 src, u64 dst)
 496 {
 497         int ret;
 498         struct btrfs_path *path;
 499         struct btrfs_key key;
 500
 501         path = btrfs_alloc_path();
 502         if (!path)
 503                 return -ENOMEM;
 504
 505         key.objectid = src;
 506         key.type = BTRFS_QGROUP_RELATION_KEY;
 507         key.offset = dst;
 508
 509         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 510         if (ret < 0)
 511                 goto out;
 512
 513         if (ret > 0) {
 514                 ret = -ENOENT;
 515                 goto out;
 516         }
 517
 518         ret = btrfs_del_item(trans, quota_root, path);
 519 out:
 520         btrfs_free_path(path);
 521         return ret;
 522 }
 523
 524 static int add_qgroup_item(struct btrfs_trans_handle *trans,
 525                            struct btrfs_root *quota_root, u64 qgroupid)
 526 {
 527         int ret;
 528         struct btrfs_path *path;
 529         struct btrfs_qgroup_info_item *qgroup_info;
 530         struct btrfs_qgroup_limit_item *qgroup_limit;
 531         struct extent_buffer *leaf;
 532         struct btrfs_key key;
 533
 534         if (btrfs_is_testing(quota_root->fs_info))
 535                 return 0;
 536
 537         path = btrfs_alloc_path();
 538         if (!path)
 539                 return -ENOMEM;
 540
 541         key.objectid = 0;
 542         key.type = BTRFS_QGROUP_INFO_KEY;
 543         key.offset = qgroupid;
 544
 545         /*
 546          * Avoid a transaction abort by catching -EEXIST here. In that
 547          * case, we proceed by re-initializing the existing structure
 548          * on disk.
 549          */
 550
 551         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 552                                       sizeof(*qgroup_info));
 553         if (ret && ret != -EEXIST)
 554                 goto out;
 555
 556         leaf = path->nodes[0];
 557         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
 558                                  struct btrfs_qgroup_info_item);
 559         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
 560         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
 561         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
 562         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
 563         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
 564
 565         btrfs_mark_buffer_dirty(leaf);
 566
 567         btrfs_release_path(path);
 568
 569         key.type = BTRFS_QGROUP_LIMIT_KEY;
 570         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 571                                       sizeof(*qgroup_limit));
 572         if (ret && ret != -EEXIST)
 573                 goto out;
 574
 575         leaf = path->nodes[0];
 576         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
 577                                   struct btrfs_qgroup_limit_item);
 578         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
 579         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
 580         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
 581         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
 582         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
 583
 584         btrfs_mark_buffer_dirty(leaf);
 585
 586         ret = 0;
 587 out:
 588         btrfs_free_path(path);
 589         return ret;
 590 }
 591
 592 static int del_qgroup_item(struct btrfs_trans_handle *trans,
 593                            struct btrfs_root *quota_root, u64 qgroupid)
 594 {
 595         int ret;
 596         struct btrfs_path *path;
 597         struct btrfs_key key;
 598
 599         path = btrfs_alloc_path();
 600         if (!path)
 601                 return -ENOMEM;
 602
 603         key.objectid = 0;
 604         key.type = BTRFS_QGROUP_INFO_KEY;
 605         key.offset = qgroupid;
 606         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
 607         if (ret < 0)
 608                 goto out;
 609
 610         if (ret > 0) {
 611                 ret = -ENOENT;
 612                 goto out;
 613         }
 614
 615         ret = btrfs_del_item(trans, quota_root, path);
 616         if (ret)
 617                 goto out;
 618
 619         btrfs_release_path(path);
 620
 621         key.type = BTRFS_QGROUP_LIMIT_KEY;
 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
 633 out:
 634         btrfs_free_path(path);
 635         return ret;
 636 }
 637
 638 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
 639                                     struct btrfs_root *root,
 640                                     struct btrfs_qgroup *qgroup)
 641 {
 642         struct btrfs_path *path;
 643         struct btrfs_key key;
 644         struct extent_buffer *l;
 645         struct btrfs_qgroup_limit_item *qgroup_limit;
 646         int ret;
 647         int slot;
 648
 649         key.objectid = 0;
 650         key.type = BTRFS_QGROUP_LIMIT_KEY;
 651         key.offset = qgroup->qgroupid;
 652
 653         path = btrfs_alloc_path();
 654         if (!path)
 655                 return -ENOMEM;
 656
 657         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 658         if (ret > 0)
 659                 ret = -ENOENT;
 660
 661         if (ret)
 662                 goto out;
 663
 664         l = path->nodes[0];
 665         slot = path->slots[0];
 666         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
 667         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
 668         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
 669         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
 670         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
 671         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
 672
 673         btrfs_mark_buffer_dirty(l);
 674
 675 out:
 676         btrfs_free_path(path);
 677         return ret;
 678 }
 679
 680 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
 681                                    struct btrfs_root *root,
 682                                    struct btrfs_qgroup *qgroup)
 683 {
 684         struct btrfs_path *path;
 685         struct btrfs_key key;
 686         struct extent_buffer *l;
 687         struct btrfs_qgroup_info_item *qgroup_info;
 688         int ret;
 689         int slot;
 690
 691         if (btrfs_is_testing(root->fs_info))
 692                 return 0;
 693
 694         key.objectid = 0;
 695         key.type = BTRFS_QGROUP_INFO_KEY;
 696         key.offset = qgroup->qgroupid;
 697
 698         path = btrfs_alloc_path();
 699         if (!path)
 700                 return -ENOMEM;
 701
 702         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 703         if (ret > 0)
 704                 ret = -ENOENT;
 705
 706         if (ret)
 707                 goto out;
 708
 709         l = path->nodes[0];
 710         slot = path->slots[0];
 711         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
 712         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
 713         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
 714         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
 715         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
 716         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
 717
 718         btrfs_mark_buffer_dirty(l);
 719
 720 out:
 721         btrfs_free_path(path);
 722         return ret;
 723 }
 724
 725 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
 726                                      struct btrfs_fs_info *fs_info,
 727                                     struct btrfs_root *root)
 728 {
 729         struct btrfs_path *path;
 730         struct btrfs_key key;
 731         struct extent_buffer *l;
 732         struct btrfs_qgroup_status_item *ptr;
 733         int ret;
 734         int slot;
 735
 736         key.objectid = 0;
 737         key.type = BTRFS_QGROUP_STATUS_KEY;
 738         key.offset = 0;
 739
 740         path = btrfs_alloc_path();
 741         if (!path)
 742                 return -ENOMEM;
 743
 744         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 745         if (ret > 0)
 746                 ret = -ENOENT;
 747
 748         if (ret)
 749                 goto out;
 750
 751         l = path->nodes[0];
 752         slot = path->slots[0];
 753         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
 754         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
 755         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
 756         btrfs_set_qgroup_status_rescan(l, ptr,
 757                                 fs_info->qgroup_rescan_progress.objectid);
 758
 759         btrfs_mark_buffer_dirty(l);
 760
 761 out:
 762         btrfs_free_path(path);
 763         return ret;
 764 }
 765
 766 /*
 767  * called with qgroup_lock held
 768  */
 769 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
 770                                   struct btrfs_root *root)
 771 {
 772         struct btrfs_path *path;
 773         struct btrfs_key key;
 774         struct extent_buffer *leaf = NULL;
 775         int ret;
 776         int nr = 0;
 777
 778         path = btrfs_alloc_path();
 779         if (!path)
 780                 return -ENOMEM;
 781
 782         path->leave_spinning = 1;
 783
 784         key.objectid = 0;
 785         key.offset = 0;
 786         key.type = 0;
 787
 788         while (1) {
 789                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 790                 if (ret < 0)
 791                         goto out;
 792                 leaf = path->nodes[0];
 793                 nr = btrfs_header_nritems(leaf);
 794                 if (!nr)
 795                         break;
 796                 /*
 797                  * delete the leaf one by one
 798                  * since the whole tree is going
 799                  * to be deleted.
 800                  */
 801                 path->slots[0] = 0;
 802                 ret = btrfs_del_items(trans, root, path, 0, nr);
 803                 if (ret)
 804                         goto out;
 805
 806                 btrfs_release_path(path);
 807         }
 808         ret = 0;
 809 out:
 810         set_bit(BTRFS_FS_QUOTA_DISABLING, &root->fs_info->flags);
 811         btrfs_free_path(path);
 812         return ret;
 813 }
 814
 815 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 816                        struct btrfs_fs_info *fs_info)
 817 {
 818         struct btrfs_root *quota_root;
 819         struct btrfs_root *tree_root = fs_info->tree_root;
 820         struct btrfs_path *path = NULL;
 821         struct btrfs_qgroup_status_item *ptr;
 822         struct extent_buffer *leaf;
 823         struct btrfs_key key;
 824         struct btrfs_key found_key;
 825         struct btrfs_qgroup *qgroup = NULL;
 826         int ret = 0;
 827         int slot;
 828
 829         mutex_lock(&fs_info->qgroup_ioctl_lock);
 830         if (fs_info->quota_root) {
 831                 set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
 832                 goto out;
 833         }
 834
 835         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 836         if (!fs_info->qgroup_ulist) {
 837                 ret = -ENOMEM;
 838                 goto out;
 839         }
 840
 841         /*
 842          * initially create the quota tree
 843          */
 844         quota_root = btrfs_create_tree(trans, fs_info,
 845                                        BTRFS_QUOTA_TREE_OBJECTID);
 846         if (IS_ERR(quota_root)) {
 847                 ret =  PTR_ERR(quota_root);
 848                 goto out;
 849         }
 850
 851         path = btrfs_alloc_path();
 852         if (!path) {
 853                 ret = -ENOMEM;
 854                 goto out_free_root;
 855         }
 856
 857         key.objectid = 0;
 858         key.type = BTRFS_QGROUP_STATUS_KEY;
 859         key.offset = 0;
 860
 861         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
 862                                       sizeof(*ptr));
 863         if (ret)
 864                 goto out_free_path;
 865
 866         leaf = path->nodes[0];
 867         ptr = btrfs_item_ptr(leaf, path->slots[0],
 868                                  struct btrfs_qgroup_status_item);
 869         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
 870         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
 871         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
 872                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
 873         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
 874         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
 875
 876         btrfs_mark_buffer_dirty(leaf);
 877
 878         key.objectid = 0;
 879         key.type = BTRFS_ROOT_REF_KEY;
 880         key.offset = 0;
 881
 882         btrfs_release_path(path);
 883         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
 884         if (ret > 0)
 885                 goto out_add_root;
 886         if (ret < 0)
 887                 goto out_free_path;
 888
 889
 890         while (1) {
 891                 slot = path->slots[0];
 892                 leaf = path->nodes[0];
 893                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 894
 895                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
 896                         ret = add_qgroup_item(trans, quota_root,
 897                                               found_key.offset);
 898                         if (ret)
 899                                 goto out_free_path;
 900
 901                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
 902                         if (IS_ERR(qgroup)) {
 903                                 ret = PTR_ERR(qgroup);
 904                                 goto out_free_path;
 905                         }
 906                 }
 907                 ret = btrfs_next_item(tree_root, path);
 908                 if (ret < 0)
 909                         goto out_free_path;
 910                 if (ret)
 911                         break;
 912         }
 913
 914 out_add_root:
 915         btrfs_release_path(path);
 916         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
 917         if (ret)
 918                 goto out_free_path;
 919
 920         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
 921         if (IS_ERR(qgroup)) {
 922                 ret = PTR_ERR(qgroup);
 923                 goto out_free_path;
 924         }
 925         spin_lock(&fs_info->qgroup_lock);
 926         fs_info->quota_root = quota_root;
 927         set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
 928         spin_unlock(&fs_info->qgroup_lock);
 929 out_free_path:
 930         btrfs_free_path(path);
 931 out_free_root:
 932         if (ret) {
 933                 free_extent_buffer(quota_root->node);
 934                 free_extent_buffer(quota_root->commit_root);
 935                 kfree(quota_root);
 936         }
 937 out:
 938         if (ret) {
 939                 ulist_free(fs_info->qgroup_ulist);
 940                 fs_info->qgroup_ulist = NULL;
 941         }
 942         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 943         return ret;
 944 }
 945
 946 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
 947                         struct btrfs_fs_info *fs_info)
 948 {
 949         struct btrfs_root *tree_root = fs_info->tree_root;
 950         struct btrfs_root *quota_root;
 951         int ret = 0;
 952
 953         mutex_lock(&fs_info->qgroup_ioctl_lock);
 954         if (!fs_info->quota_root)
 955                 goto out;
 956         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
 957         set_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags);
 958         btrfs_qgroup_wait_for_completion(fs_info, false);
 959         spin_lock(&fs_info->qgroup_lock);
 960         quota_root = fs_info->quota_root;
 961         fs_info->quota_root = NULL;
 962         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
 963         spin_unlock(&fs_info->qgroup_lock);
 964
 965         btrfs_free_qgroup_config(fs_info);
 966
 967         ret = btrfs_clean_quota_tree(trans, quota_root);
 968         if (ret)
 969                 goto out;
 970
 971         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
 972         if (ret)
 973                 goto out;
 974
 975         list_del(&quota_root->dirty_list);
 976
 977         btrfs_tree_lock(quota_root->node);
 978         clean_tree_block(fs_info, quota_root->node);
 979         btrfs_tree_unlock(quota_root->node);
 980         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
 981
 982         free_extent_buffer(quota_root->node);
 983         free_extent_buffer(quota_root->commit_root);
 984         kfree(quota_root);
 985 out:
 986         mutex_unlock(&fs_info->qgroup_ioctl_lock);
 987         return ret;
 988 }
 989
 990 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
 991                          struct btrfs_qgroup *qgroup)
 992 {
 993         if (list_empty(&qgroup->dirty))
 994                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
 995 }
 996
 997 static void report_reserved_underflow(struct btrfs_fs_info *fs_info,
 998                                       struct btrfs_qgroup *qgroup,
 999                                       u64 num_bytes)
1000 {
1001 #ifdef CONFIG_BTRFS_DEBUG
1002         WARN_ON(qgroup->reserved < num_bytes);
1003         btrfs_debug(fs_info,
1004                 "qgroup %llu reserved space underflow, have: %llu, to free: %llu",
1005                 qgroup->qgroupid, qgroup->reserved, num_bytes);
1006 #endif
1007         qgroup->reserved = 0;
1008 }
1009 /*
1010  * The easy accounting, if we are adding/removing the only ref for an extent
1011  * then this qgroup and all of the parent qgroups get their reference and
1012  * exclusive counts adjusted.
1013  *
1014  * Caller should hold fs_info->qgroup_lock.
1015  */
1016 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1017                                     struct ulist *tmp, u64 ref_root,
1018                                     u64 num_bytes, int sign)
1019 {
1020         struct btrfs_qgroup *qgroup;
1021         struct btrfs_qgroup_list *glist;
1022         struct ulist_node *unode;
1023         struct ulist_iterator uiter;
1024         int ret = 0;
1025
1026         qgroup = find_qgroup_rb(fs_info, ref_root);
1027         if (!qgroup)
1028                 goto out;
1029
1030         qgroup->rfer += sign * num_bytes;
1031         qgroup->rfer_cmpr += sign * num_bytes;
1032
1033         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1034         qgroup->excl += sign * num_bytes;
1035         qgroup->excl_cmpr += sign * num_bytes;
1036         if (sign > 0) {
1037                 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes);
1038                 if (qgroup->reserved < num_bytes)
1039                         report_reserved_underflow(fs_info, qgroup, num_bytes);
1040                 else
1041                         qgroup->reserved -= num_bytes;
1042         }
1043
1044         qgroup_dirty(fs_info, qgroup);
1045
1046         /* Get all of the parent groups that contain this qgroup */
1047         list_for_each_entry(glist, &qgroup->groups, next_group) {
1048                 ret = ulist_add(tmp, glist->group->qgroupid,
1049                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1050                 if (ret < 0)
1051                         goto out;
1052         }
1053
1054         /* Iterate all of the parents and adjust their reference counts */
1055         ULIST_ITER_INIT(&uiter);
1056         while ((unode = ulist_next(tmp, &uiter))) {
1057                 qgroup = unode_aux_to_qgroup(unode);
1058                 qgroup->rfer += sign * num_bytes;
1059                 qgroup->rfer_cmpr += sign * num_bytes;
1060                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1061                 qgroup->excl += sign * num_bytes;
1062                 if (sign > 0) {
1063                         trace_qgroup_update_reserve(fs_info, qgroup,
1064                                                     -(s64)num_bytes);
1065                         if (qgroup->reserved < num_bytes)
1066                                 report_reserved_underflow(fs_info, qgroup,
1067                                                           num_bytes);
1068                         else
1069                                 qgroup->reserved -= num_bytes;
1070                 }
1071                 qgroup->excl_cmpr += sign * num_bytes;
1072                 qgroup_dirty(fs_info, qgroup);
1073
1074                 /* Add any parents of the parents */
1075                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1076                         ret = ulist_add(tmp, glist->group->qgroupid,
1077                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1078                         if (ret < 0)
1079                                 goto out;
1080                 }
1081         }
1082         ret = 0;
1083 out:
1084         return ret;
1085 }
1086
1087
1088 /*
1089  * Quick path for updating qgroup with only excl refs.
1090  *
1091  * In that case, just update all parent will be enough.
1092  * Or we needs to do a full rescan.
1093  * Caller should also hold fs_info->qgroup_lock.
1094  *
1095  * Return 0 for quick update, return >0 for need to full rescan
1096  * and mark INCONSISTENT flag.
1097  * Return < 0 for other error.
1098  */
1099 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1100                                    struct ulist *tmp, u64 src, u64 dst,
1101                                    int sign)
1102 {
1103         struct btrfs_qgroup *qgroup;
1104         int ret = 1;
1105         int err = 0;
1106
1107         qgroup = find_qgroup_rb(fs_info, src);
1108         if (!qgroup)
1109                 goto out;
1110         if (qgroup->excl == qgroup->rfer) {
1111                 ret = 0;
1112                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1113                                                qgroup->excl, sign);
1114                 if (err < 0) {
1115                         ret = err;
1116                         goto out;
1117                 }
1118         }
1119 out:
1120         if (ret)
1121                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1122         return ret;
1123 }
1124
1125 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1126                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1127 {
1128         struct btrfs_root *quota_root;
1129         struct btrfs_qgroup *parent;
1130         struct btrfs_qgroup *member;
1131         struct btrfs_qgroup_list *list;
1132         struct ulist *tmp;
1133         int ret = 0;
1134
1135         /* Check the level of src and dst first */
1136         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1137                 return -EINVAL;
1138
1139         tmp = ulist_alloc(GFP_KERNEL);
1140         if (!tmp)
1141                 return -ENOMEM;
1142
1143         mutex_lock(&fs_info->qgroup_ioctl_lock);
1144         quota_root = fs_info->quota_root;
1145         if (!quota_root) {
1146                 ret = -EINVAL;
1147                 goto out;
1148         }
1149         member = find_qgroup_rb(fs_info, src);
1150         parent = find_qgroup_rb(fs_info, dst);
1151         if (!member || !parent) {
1152                 ret = -EINVAL;
1153                 goto out;
1154         }
1155
1156         /* check if such qgroup relation exist firstly */
1157         list_for_each_entry(list, &member->groups, next_group) {
1158                 if (list->group == parent) {
1159                         ret = -EEXIST;
1160                         goto out;
1161                 }
1162         }
1163
1164         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1165         if (ret)
1166                 goto out;
1167
1168         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1169         if (ret) {
1170                 del_qgroup_relation_item(trans, quota_root, src, dst);
1171                 goto out;
1172         }
1173
1174         spin_lock(&fs_info->qgroup_lock);
1175         ret = add_relation_rb(fs_info, src, dst);
1176         if (ret < 0) {
1177                 spin_unlock(&fs_info->qgroup_lock);
1178                 goto out;
1179         }
1180         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1181         spin_unlock(&fs_info->qgroup_lock);
1182 out:
1183         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1184         ulist_free(tmp);
1185         return ret;
1186 }
1187
1188 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1189                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1190 {
1191         struct btrfs_root *quota_root;
1192         struct btrfs_qgroup *parent;
1193         struct btrfs_qgroup *member;
1194         struct btrfs_qgroup_list *list;
1195         struct ulist *tmp;
1196         int ret = 0;
1197         int err;
1198
1199         tmp = ulist_alloc(GFP_KERNEL);
1200         if (!tmp)
1201                 return -ENOMEM;
1202
1203         quota_root = fs_info->quota_root;
1204         if (!quota_root) {
1205                 ret = -EINVAL;
1206                 goto out;
1207         }
1208
1209         member = find_qgroup_rb(fs_info, src);
1210         parent = find_qgroup_rb(fs_info, dst);
1211         if (!member || !parent) {
1212                 ret = -EINVAL;
1213                 goto out;
1214         }
1215
1216         /* check if such qgroup relation exist firstly */
1217         list_for_each_entry(list, &member->groups, next_group) {
1218                 if (list->group == parent)
1219                         goto exist;
1220         }
1221         ret = -ENOENT;
1222         goto out;
1223 exist:
1224         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1225         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1226         if (err && !ret)
1227                 ret = err;
1228
1229         spin_lock(&fs_info->qgroup_lock);
1230         del_relation_rb(fs_info, src, dst);
1231         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1232         spin_unlock(&fs_info->qgroup_lock);
1233 out:
1234         ulist_free(tmp);
1235         return ret;
1236 }
1237
1238 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1239                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1240 {
1241         int ret = 0;
1242
1243         mutex_lock(&fs_info->qgroup_ioctl_lock);
1244         ret = __del_qgroup_relation(trans, fs_info, src, dst);
1245         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1246
1247         return ret;
1248 }
1249
1250 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1251                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1252 {
1253         struct btrfs_root *quota_root;
1254         struct btrfs_qgroup *qgroup;
1255         int ret = 0;
1256
1257         mutex_lock(&fs_info->qgroup_ioctl_lock);
1258         quota_root = fs_info->quota_root;
1259         if (!quota_root) {
1260                 ret = -EINVAL;
1261                 goto out;
1262         }
1263         qgroup = find_qgroup_rb(fs_info, qgroupid);
1264         if (qgroup) {
1265                 ret = -EEXIST;
1266                 goto out;
1267         }
1268
1269         ret = add_qgroup_item(trans, quota_root, qgroupid);
1270         if (ret)
1271                 goto out;
1272
1273         spin_lock(&fs_info->qgroup_lock);
1274         qgroup = add_qgroup_rb(fs_info, qgroupid);
1275         spin_unlock(&fs_info->qgroup_lock);
1276
1277         if (IS_ERR(qgroup))
1278                 ret = PTR_ERR(qgroup);
1279 out:
1280         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1281         return ret;
1282 }
1283
1284 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1285                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1286 {
1287         struct btrfs_root *quota_root;
1288         struct btrfs_qgroup *qgroup;
1289         struct btrfs_qgroup_list *list;
1290         int ret = 0;
1291
1292         mutex_lock(&fs_info->qgroup_ioctl_lock);
1293         quota_root = fs_info->quota_root;
1294         if (!quota_root) {
1295                 ret = -EINVAL;
1296                 goto out;
1297         }
1298
1299         qgroup = find_qgroup_rb(fs_info, qgroupid);
1300         if (!qgroup) {
1301                 ret = -ENOENT;
1302                 goto out;
1303         } else {
1304                 /* check if there are no children of this qgroup */
1305                 if (!list_empty(&qgroup->members)) {
1306                         ret = -EBUSY;
1307                         goto out;
1308                 }
1309         }
1310         ret = del_qgroup_item(trans, quota_root, qgroupid);
1311
1312         while (!list_empty(&qgroup->groups)) {
1313                 list = list_first_entry(&qgroup->groups,
1314                                         struct btrfs_qgroup_list, next_group);
1315                 ret = __del_qgroup_relation(trans, fs_info,
1316                                            qgroupid,
1317                                            list->group->qgroupid);
1318                 if (ret)
1319                         goto out;
1320         }
1321
1322         spin_lock(&fs_info->qgroup_lock);
1323         del_qgroup_rb(fs_info, qgroupid);
1324         spin_unlock(&fs_info->qgroup_lock);
1325 out:
1326         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1327         return ret;
1328 }
1329
1330 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1331                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1332                        struct btrfs_qgroup_limit *limit)
1333 {
1334         struct btrfs_root *quota_root;
1335         struct btrfs_qgroup *qgroup;
1336         int ret = 0;
1337         /* Sometimes we would want to clear the limit on this qgroup.
1338          * To meet this requirement, we treat the -1 as a special value
1339          * which tell kernel to clear the limit on this qgroup.
1340          */
1341         const u64 CLEAR_VALUE = -1;
1342
1343         mutex_lock(&fs_info->qgroup_ioctl_lock);
1344         quota_root = fs_info->quota_root;
1345         if (!quota_root) {
1346                 ret = -EINVAL;
1347                 goto out;
1348         }
1349
1350         qgroup = find_qgroup_rb(fs_info, qgroupid);
1351         if (!qgroup) {
1352                 ret = -ENOENT;
1353                 goto out;
1354         }
1355
1356         spin_lock(&fs_info->qgroup_lock);
1357         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1358                 if (limit->max_rfer == CLEAR_VALUE) {
1359                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1360                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1361                         qgroup->max_rfer = 0;
1362                 } else {
1363                         qgroup->max_rfer = limit->max_rfer;
1364                 }
1365         }
1366         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1367                 if (limit->max_excl == CLEAR_VALUE) {
1368                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1369                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1370                         qgroup->max_excl = 0;
1371                 } else {
1372                         qgroup->max_excl = limit->max_excl;
1373                 }
1374         }
1375         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1376                 if (limit->rsv_rfer == CLEAR_VALUE) {
1377                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1378                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1379                         qgroup->rsv_rfer = 0;
1380                 } else {
1381                         qgroup->rsv_rfer = limit->rsv_rfer;
1382                 }
1383         }
1384         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1385                 if (limit->rsv_excl == CLEAR_VALUE) {
1386                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1387                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1388                         qgroup->rsv_excl = 0;
1389                 } else {
1390                         qgroup->rsv_excl = limit->rsv_excl;
1391                 }
1392         }
1393         qgroup->lim_flags |= limit->flags;
1394
1395         spin_unlock(&fs_info->qgroup_lock);
1396
1397         ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1398         if (ret) {
1399                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1400                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1401                        qgroupid);
1402         }
1403
1404 out:
1405         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1406         return ret;
1407 }
1408
1409 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1410                                 struct btrfs_delayed_ref_root *delayed_refs,
1411                                 struct btrfs_qgroup_extent_record *record)
1412 {
1413         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1414         struct rb_node *parent_node = NULL;
1415         struct btrfs_qgroup_extent_record *entry;
1416         u64 bytenr = record->bytenr;
1417
1418         assert_spin_locked(&delayed_refs->lock);
1419         trace_btrfs_qgroup_trace_extent(fs_info, record);
1420
1421         while (*p) {
1422                 parent_node = *p;
1423                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1424                                  node);
1425                 if (bytenr < entry->bytenr)
1426                         p = &(*p)->rb_left;
1427                 else if (bytenr > entry->bytenr)
1428                         p = &(*p)->rb_right;
1429                 else
1430                         return 1;
1431         }
1432
1433         rb_link_node(&record->node, parent_node, p);
1434         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1435         return 0;
1436 }
1437
1438 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1439                                    struct btrfs_qgroup_extent_record *qrecord)
1440 {
1441         struct ulist *old_root;
1442         u64 bytenr = qrecord->bytenr;
1443         int ret;
1444
1445         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root);
1446         if (ret < 0)
1447                 return ret;
1448
1449         /*
1450          * Here we don't need to get the lock of
1451          * trans->transaction->delayed_refs, since inserted qrecord won't
1452          * be deleted, only qrecord->node may be modified (new qrecord insert)
1453          *
1454          * So modifying qrecord->old_roots is safe here
1455          */
1456         qrecord->old_roots = old_root;
1457         return 0;
1458 }
1459
1460 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1461                 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1462                 gfp_t gfp_flag)
1463 {
1464         struct btrfs_qgroup_extent_record *record;
1465         struct btrfs_delayed_ref_root *delayed_refs;
1466         int ret;
1467
1468         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1469             || bytenr == 0 || num_bytes == 0)
1470                 return 0;
1471         if (WARN_ON(trans == NULL))
1472                 return -EINVAL;
1473         record = kmalloc(sizeof(*record), gfp_flag);
1474         if (!record)
1475                 return -ENOMEM;
1476
1477         delayed_refs = &trans->transaction->delayed_refs;
1478         record->bytenr = bytenr;
1479         record->num_bytes = num_bytes;
1480         record->old_roots = NULL;
1481
1482         spin_lock(&delayed_refs->lock);
1483         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1484         spin_unlock(&delayed_refs->lock);
1485         if (ret > 0) {
1486                 kfree(record);
1487                 return 0;
1488         }
1489         return btrfs_qgroup_trace_extent_post(fs_info, record);
1490 }
1491
1492 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1493                                   struct btrfs_fs_info *fs_info,
1494                                   struct extent_buffer *eb)
1495 {
1496         int nr = btrfs_header_nritems(eb);
1497         int i, extent_type, ret;
1498         struct btrfs_key key;
1499         struct btrfs_file_extent_item *fi;
1500         u64 bytenr, num_bytes;
1501
1502         /* We can be called directly from walk_up_proc() */
1503         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1504                 return 0;
1505
1506         for (i = 0; i < nr; i++) {
1507                 btrfs_item_key_to_cpu(eb, &key, i);
1508
1509                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1510                         continue;
1511
1512                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1513                 /* filter out non qgroup-accountable extents  */
1514                 extent_type = btrfs_file_extent_type(eb, fi);
1515
1516                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1517                         continue;
1518
1519                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1520                 if (!bytenr)
1521                         continue;
1522
1523                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1524
1525                 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1526                                                 num_bytes, GFP_NOFS);
1527                 if (ret)
1528                         return ret;
1529         }
1530         cond_resched();
1531         return 0;
1532 }
1533
1534 /*
1535  * Walk up the tree from the bottom, freeing leaves and any interior
1536  * nodes which have had all slots visited. If a node (leaf or
1537  * interior) is freed, the node above it will have it's slot
1538  * incremented. The root node will never be freed.
1539  *
1540  * At the end of this function, we should have a path which has all
1541  * slots incremented to the next position for a search. If we need to
1542  * read a new node it will be NULL and the node above it will have the
1543  * correct slot selected for a later read.
1544  *
1545  * If we increment the root nodes slot counter past the number of
1546  * elements, 1 is returned to signal completion of the search.
1547  */
1548 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1549 {
1550         int level = 0;
1551         int nr, slot;
1552         struct extent_buffer *eb;
1553
1554         if (root_level == 0)
1555                 return 1;
1556
1557         while (level <= root_level) {
1558                 eb = path->nodes[level];
1559                 nr = btrfs_header_nritems(eb);
1560                 path->slots[level]++;
1561                 slot = path->slots[level];
1562                 if (slot >= nr || level == 0) {
1563                         /*
1564                          * Don't free the root -  we will detect this
1565                          * condition after our loop and return a
1566                          * positive value for caller to stop walking the tree.
1567                          */
1568                         if (level != root_level) {
1569                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1570                                 path->locks[level] = 0;
1571
1572                                 free_extent_buffer(eb);
1573                                 path->nodes[level] = NULL;
1574                                 path->slots[level] = 0;
1575                         }
1576                 } else {
1577                         /*
1578                          * We have a valid slot to walk back down
1579                          * from. Stop here so caller can process these
1580                          * new nodes.
1581                          */
1582                         break;
1583                 }
1584
1585                 level++;
1586         }
1587
1588         eb = path->nodes[root_level];
1589         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1590                 return 1;
1591
1592         return 0;
1593 }
1594
1595 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1596                                struct btrfs_root *root,
1597                                struct extent_buffer *root_eb,
1598                                u64 root_gen, int root_level)
1599 {
1600         struct btrfs_fs_info *fs_info = root->fs_info;
1601         int ret = 0;
1602         int level;
1603         struct extent_buffer *eb = root_eb;
1604         struct btrfs_path *path = NULL;
1605
1606         BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
1607         BUG_ON(root_eb == NULL);
1608
1609         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1610                 return 0;
1611
1612         if (!extent_buffer_uptodate(root_eb)) {
1613                 ret = btrfs_read_buffer(root_eb, root_gen);
1614                 if (ret)
1615                         goto out;
1616         }
1617
1618         if (root_level == 0) {
1619                 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1620                 goto out;
1621         }
1622
1623         path = btrfs_alloc_path();
1624         if (!path)
1625                 return -ENOMEM;
1626
1627         /*
1628          * Walk down the tree.  Missing extent blocks are filled in as
1629          * we go. Metadata is accounted every time we read a new
1630          * extent block.
1631          *
1632          * When we reach a leaf, we account for file extent items in it,
1633          * walk back up the tree (adjusting slot pointers as we go)
1634          * and restart the search process.
1635          */
1636         extent_buffer_get(root_eb); /* For path */
1637         path->nodes[root_level] = root_eb;
1638         path->slots[root_level] = 0;
1639         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1640 walk_down:
1641         level = root_level;
1642         while (level >= 0) {
1643                 if (path->nodes[level] == NULL) {
1644                         int parent_slot;
1645                         u64 child_gen;
1646                         u64 child_bytenr;
1647
1648                         /*
1649                          * We need to get child blockptr/gen from parent before
1650                          * we can read it.
1651                           */
1652                         eb = path->nodes[level + 1];
1653                         parent_slot = path->slots[level + 1];
1654                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1655                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1656
1657                         eb = read_tree_block(fs_info, child_bytenr, child_gen);
1658                         if (IS_ERR(eb)) {
1659                                 ret = PTR_ERR(eb);
1660                                 goto out;
1661                         } else if (!extent_buffer_uptodate(eb)) {
1662                                 free_extent_buffer(eb);
1663                                 ret = -EIO;
1664                                 goto out;
1665                         }
1666
1667                         path->nodes[level] = eb;
1668                         path->slots[level] = 0;
1669
1670                         btrfs_tree_read_lock(eb);
1671                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1672                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1673
1674                         ret = btrfs_qgroup_trace_extent(trans, fs_info,
1675                                                         child_bytenr,
1676                                                         fs_info->nodesize,
1677                                                         GFP_NOFS);
1678                         if (ret)
1679                                 goto out;
1680                 }
1681
1682                 if (level == 0) {
1683                         ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1684                                                            path->nodes[level]);
1685                         if (ret)
1686                                 goto out;
1687
1688                         /* Nonzero return here means we completed our search */
1689                         ret = adjust_slots_upwards(path, root_level);
1690                         if (ret)
1691                                 break;
1692
1693                         /* Restart search with new slots */
1694                         goto walk_down;
1695                 }
1696
1697                 level--;
1698         }
1699
1700         ret = 0;
1701 out:
1702         btrfs_free_path(path);
1703
1704         return ret;
1705 }
1706
1707 #define UPDATE_NEW      0
1708 #define UPDATE_OLD      1
1709 /*
1710  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1711  */
1712 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1713                                 struct ulist *roots, struct ulist *tmp,
1714                                 struct ulist *qgroups, u64 seq, int update_old)
1715 {
1716         struct ulist_node *unode;
1717         struct ulist_iterator uiter;
1718         struct ulist_node *tmp_unode;
1719         struct ulist_iterator tmp_uiter;
1720         struct btrfs_qgroup *qg;
1721         int ret = 0;
1722
1723         if (!roots)
1724                 return 0;
1725         ULIST_ITER_INIT(&uiter);
1726         while ((unode = ulist_next(roots, &uiter))) {
1727                 qg = find_qgroup_rb(fs_info, unode->val);
1728                 if (!qg)
1729                         continue;
1730
1731                 ulist_reinit(tmp);
1732                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1733                                 GFP_ATOMIC);
1734                 if (ret < 0)
1735                         return ret;
1736                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1737                 if (ret < 0)
1738                         return ret;
1739                 ULIST_ITER_INIT(&tmp_uiter);
1740                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1741                         struct btrfs_qgroup_list *glist;
1742
1743                         qg = unode_aux_to_qgroup(tmp_unode);
1744                         if (update_old)
1745                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1746                         else
1747                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1748                         list_for_each_entry(glist, &qg->groups, next_group) {
1749                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1750                                                 qgroup_to_aux(glist->group),
1751                                                 GFP_ATOMIC);
1752                                 if (ret < 0)
1753                                         return ret;
1754                                 ret = ulist_add(tmp, glist->group->qgroupid,
1755                                                 qgroup_to_aux(glist->group),
1756                                                 GFP_ATOMIC);
1757                                 if (ret < 0)
1758                                         return ret;
1759                         }
1760                 }
1761         }
1762         return 0;
1763 }
1764
1765 /*
1766  * Update qgroup rfer/excl counters.
1767  * Rfer update is easy, codes can explain themselves.
1768  *
1769  * Excl update is tricky, the update is split into 2 part.
1770  * Part 1: Possible exclusive <-> sharing detect:
1771  *      |       A       |       !A      |
1772  *  -------------------------------------
1773  *  B   |       *       |       -       |
1774  *  -------------------------------------
1775  *  !B  |       +       |       **      |
1776  *  -------------------------------------
1777  *
1778  * Conditions:
1779  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1780  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1781  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1782  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1783  *
1784  * Results:
1785  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1786  * *: Definitely not changed.           **: Possible unchanged.
1787  *
1788  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1789  *
1790  * To make the logic clear, we first use condition A and B to split
1791  * combination into 4 results.
1792  *
1793  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1794  * only on variant maybe 0.
1795  *
1796  * Lastly, check result **, since there are 2 variants maybe 0, split them
1797  * again(2x2).
1798  * But this time we don't need to consider other things, the codes and logic
1799  * is easy to understand now.
1800  */
1801 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1802                                   struct ulist *qgroups,
1803                                   u64 nr_old_roots,
1804                                   u64 nr_new_roots,
1805                                   u64 num_bytes, u64 seq)
1806 {
1807         struct ulist_node *unode;
1808         struct ulist_iterator uiter;
1809         struct btrfs_qgroup *qg;
1810         u64 cur_new_count, cur_old_count;
1811
1812         ULIST_ITER_INIT(&uiter);
1813         while ((unode = ulist_next(qgroups, &uiter))) {
1814                 bool dirty = false;
1815
1816                 qg = unode_aux_to_qgroup(unode);
1817                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1818                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1819
1820                 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1821                                              cur_old_count, cur_new_count);
1822
1823                 /* Rfer update part */
1824                 if (cur_old_count == 0 && cur_new_count > 0) {
1825                         qg->rfer += num_bytes;
1826                         qg->rfer_cmpr += num_bytes;
1827                         dirty = true;
1828                 }
1829                 if (cur_old_count > 0 && cur_new_count == 0) {
1830                         qg->rfer -= num_bytes;
1831                         qg->rfer_cmpr -= num_bytes;
1832                         dirty = true;
1833                 }
1834
1835                 /* Excl update part */
1836                 /* Exclusive/none -> shared case */
1837                 if (cur_old_count == nr_old_roots &&
1838                     cur_new_count < nr_new_roots) {
1839                         /* Exclusive -> shared */
1840                         if (cur_old_count != 0) {
1841                                 qg->excl -= num_bytes;
1842                                 qg->excl_cmpr -= num_bytes;
1843                                 dirty = true;
1844                         }
1845                 }
1846
1847                 /* Shared -> exclusive/none case */
1848                 if (cur_old_count < nr_old_roots &&
1849                     cur_new_count == nr_new_roots) {
1850                         /* Shared->exclusive */
1851                         if (cur_new_count != 0) {
1852                                 qg->excl += num_bytes;
1853                                 qg->excl_cmpr += num_bytes;
1854                                 dirty = true;
1855                         }
1856                 }
1857
1858                 /* Exclusive/none -> exclusive/none case */
1859                 if (cur_old_count == nr_old_roots &&
1860                     cur_new_count == nr_new_roots) {
1861                         if (cur_old_count == 0) {
1862                                 /* None -> exclusive/none */
1863
1864                                 if (cur_new_count != 0) {
1865                                         /* None -> exclusive */
1866                                         qg->excl += num_bytes;
1867                                         qg->excl_cmpr += num_bytes;
1868                                         dirty = true;
1869                                 }
1870                                 /* None -> none, nothing changed */
1871                         } else {
1872                                 /* Exclusive -> exclusive/none */
1873
1874                                 if (cur_new_count == 0) {
1875                                         /* Exclusive -> none */
1876                                         qg->excl -= num_bytes;
1877                                         qg->excl_cmpr -= num_bytes;
1878                                         dirty = true;
1879                                 }
1880                                 /* Exclusive -> exclusive, nothing changed */
1881                         }
1882                 }
1883
1884                 if (dirty)
1885                         qgroup_dirty(fs_info, qg);
1886         }
1887         return 0;
1888 }
1889
1890 /*
1891  * Check if the @roots potentially is a list of fs tree roots
1892  *
1893  * Return 0 for definitely not a fs/subvol tree roots ulist
1894  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
1895  *          one as well)
1896  */
1897 static int maybe_fs_roots(struct ulist *roots)
1898 {
1899         struct ulist_node *unode;
1900         struct ulist_iterator uiter;
1901
1902         /* Empty one, still possible for fs roots */
1903         if (!roots || roots->nnodes == 0)
1904                 return 1;
1905
1906         ULIST_ITER_INIT(&uiter);
1907         unode = ulist_next(roots, &uiter);
1908         if (!unode)
1909                 return 1;
1910
1911         /*
1912          * If it contains fs tree roots, then it must belong to fs/subvol
1913          * trees.
1914          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
1915          */
1916         return is_fstree(unode->val);
1917 }
1918
1919 int
1920 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1921                             struct btrfs_fs_info *fs_info,
1922                             u64 bytenr, u64 num_bytes,
1923                             struct ulist *old_roots, struct ulist *new_roots)
1924 {
1925         struct ulist *qgroups = NULL;
1926         struct ulist *tmp = NULL;
1927         u64 seq;
1928         u64 nr_new_roots = 0;
1929         u64 nr_old_roots = 0;
1930         int ret = 0;
1931
1932         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1933                 return 0;
1934
1935         if (new_roots) {
1936                 if (!maybe_fs_roots(new_roots))
1937                         goto out_free;
1938                 nr_new_roots = new_roots->nnodes;
1939         }
1940         if (old_roots) {
1941                 if (!maybe_fs_roots(old_roots))
1942                         goto out_free;
1943                 nr_old_roots = old_roots->nnodes;
1944         }
1945
1946         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
1947         if (nr_old_roots == 0 && nr_new_roots == 0)
1948                 goto out_free;
1949
1950         BUG_ON(!fs_info->quota_root);
1951
1952         trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1953                                           nr_old_roots, nr_new_roots);
1954
1955         qgroups = ulist_alloc(GFP_NOFS);
1956         if (!qgroups) {
1957                 ret = -ENOMEM;
1958                 goto out_free;
1959         }
1960         tmp = ulist_alloc(GFP_NOFS);
1961         if (!tmp) {
1962                 ret = -ENOMEM;
1963                 goto out_free;
1964         }
1965
1966         mutex_lock(&fs_info->qgroup_rescan_lock);
1967         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1968                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1969                         mutex_unlock(&fs_info->qgroup_rescan_lock);
1970                         ret = 0;
1971                         goto out_free;
1972                 }
1973         }
1974         mutex_unlock(&fs_info->qgroup_rescan_lock);
1975
1976         spin_lock(&fs_info->qgroup_lock);
1977         seq = fs_info->qgroup_seq;
1978
1979         /* Update old refcnts using old_roots */
1980         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1981                                    UPDATE_OLD);
1982         if (ret < 0)
1983                 goto out;
1984
1985         /* Update new refcnts using new_roots */
1986         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1987                                    UPDATE_NEW);
1988         if (ret < 0)
1989                 goto out;
1990
1991         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1992                                num_bytes, seq);
1993
1994         /*
1995          * Bump qgroup_seq to avoid seq overlap
1996          */
1997         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1998 out:
1999         spin_unlock(&fs_info->qgroup_lock);
2000 out_free:
2001         ulist_free(tmp);
2002         ulist_free(qgroups);
2003         ulist_free(old_roots);
2004         ulist_free(new_roots);
2005         return ret;
2006 }
2007
2008 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
2009                                  struct btrfs_fs_info *fs_info)
2010 {
2011         struct btrfs_qgroup_extent_record *record;
2012         struct btrfs_delayed_ref_root *delayed_refs;
2013         struct ulist *new_roots = NULL;
2014         struct rb_node *node;
2015         u64 qgroup_to_skip;
2016         int ret = 0;
2017
2018         delayed_refs = &trans->transaction->delayed_refs;
2019         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2020         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2021                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2022                                   node);
2023
2024                 trace_btrfs_qgroup_account_extents(fs_info, record);
2025
2026                 if (!ret) {
2027                         /*
2028                          * Old roots should be searched when inserting qgroup
2029                          * extent record
2030                          */
2031                         if (WARN_ON(!record->old_roots)) {
2032                                 /* Search commit root to find old_roots */
2033                                 ret = btrfs_find_all_roots(NULL, fs_info,
2034                                                 record->bytenr, 0,
2035                                                 &record->old_roots);
2036                                 if (ret < 0)
2037                                         goto cleanup;
2038                         }
2039
2040                         /*
2041                          * Use SEQ_LAST as time_seq to do special search, which
2042                          * doesn't lock tree or delayed_refs and search current
2043                          * root. It's safe inside commit_transaction().
2044                          */
2045                         ret = btrfs_find_all_roots(trans, fs_info,
2046                                         record->bytenr, SEQ_LAST, &new_roots);
2047                         if (ret < 0)
2048                                 goto cleanup;
2049                         if (qgroup_to_skip) {
2050                                 ulist_del(new_roots, qgroup_to_skip, 0);
2051                                 ulist_del(record->old_roots, qgroup_to_skip,
2052                                           0);
2053                         }
2054                         ret = btrfs_qgroup_account_extent(trans, fs_info,
2055                                         record->bytenr, record->num_bytes,
2056                                         record->old_roots, new_roots);
2057                         record->old_roots = NULL;
2058                         new_roots = NULL;
2059                 }
2060 cleanup:
2061                 ulist_free(record->old_roots);
2062                 ulist_free(new_roots);
2063                 new_roots = NULL;
2064                 rb_erase(node, &delayed_refs->dirty_extent_root);
2065                 kfree(record);
2066
2067         }
2068         return ret;
2069 }
2070
2071 /*
2072  * called from commit_transaction. Writes all changed qgroups to disk.
2073  */
2074 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2075                       struct btrfs_fs_info *fs_info)
2076 {
2077         struct btrfs_root *quota_root = fs_info->quota_root;
2078         int ret = 0;
2079         int start_rescan_worker = 0;
2080
2081         if (!quota_root)
2082                 goto out;
2083
2084         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
2085             test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2086                 start_rescan_worker = 1;
2087
2088         if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2089                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2090         if (test_and_clear_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags))
2091                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2092
2093         spin_lock(&fs_info->qgroup_lock);
2094         while (!list_empty(&fs_info->dirty_qgroups)) {
2095                 struct btrfs_qgroup *qgroup;
2096                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2097                                           struct btrfs_qgroup, dirty);
2098                 list_del_init(&qgroup->dirty);
2099                 spin_unlock(&fs_info->qgroup_lock);
2100                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2101                 if (ret)
2102                         fs_info->qgroup_flags |=
2103                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2104                 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2105                 if (ret)
2106                         fs_info->qgroup_flags |=
2107                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2108                 spin_lock(&fs_info->qgroup_lock);
2109         }
2110         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2111                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2112         else
2113                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2114         spin_unlock(&fs_info->qgroup_lock);
2115
2116         ret = update_qgroup_status_item(trans, fs_info, quota_root);
2117         if (ret)
2118                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2119
2120         if (!ret && start_rescan_worker) {
2121                 ret = qgroup_rescan_init(fs_info, 0, 1);
2122                 if (!ret) {
2123                         qgroup_rescan_zero_tracking(fs_info);
2124                         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2125                                          &fs_info->qgroup_rescan_work);
2126                 }
2127                 ret = 0;
2128         }
2129
2130 out:
2131
2132         return ret;
2133 }
2134
2135 /*
2136  * Copy the accounting information between qgroups. This is necessary
2137  * when a snapshot or a subvolume is created. Throwing an error will
2138  * cause a transaction abort so we take extra care here to only error
2139  * when a readonly fs is a reasonable outcome.
2140  */
2141 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2142                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2143                          struct btrfs_qgroup_inherit *inherit)
2144 {
2145         int ret = 0;
2146         int i;
2147         u64 *i_qgroups;
2148         struct btrfs_root *quota_root = fs_info->quota_root;
2149         struct btrfs_qgroup *srcgroup;
2150         struct btrfs_qgroup *dstgroup;
2151         u32 level_size = 0;
2152         u64 nums;
2153
2154         mutex_lock(&fs_info->qgroup_ioctl_lock);
2155         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2156                 goto out;
2157
2158         if (!quota_root) {
2159                 ret = -EINVAL;
2160                 goto out;
2161         }
2162
2163         if (inherit) {
2164                 i_qgroups = (u64 *)(inherit + 1);
2165                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2166                        2 * inherit->num_excl_copies;
2167                 for (i = 0; i < nums; ++i) {
2168                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2169
2170                         /*
2171                          * Zero out invalid groups so we can ignore
2172                          * them later.
2173                          */
2174                         if (!srcgroup ||
2175                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2176                                 *i_qgroups = 0ULL;
2177
2178                         ++i_qgroups;
2179                 }
2180         }
2181
2182         /*
2183          * create a tracking group for the subvol itself
2184          */
2185         ret = add_qgroup_item(trans, quota_root, objectid);
2186         if (ret)
2187                 goto out;
2188
2189         if (srcid) {
2190                 struct btrfs_root *srcroot;
2191                 struct btrfs_key srckey;
2192
2193                 srckey.objectid = srcid;
2194                 srckey.type = BTRFS_ROOT_ITEM_KEY;
2195                 srckey.offset = (u64)-1;
2196                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2197                 if (IS_ERR(srcroot)) {
2198                         ret = PTR_ERR(srcroot);
2199                         goto out;
2200                 }
2201
2202                 level_size = fs_info->nodesize;
2203         }
2204
2205         /*
2206          * add qgroup to all inherited groups
2207          */
2208         if (inherit) {
2209                 i_qgroups = (u64 *)(inherit + 1);
2210                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2211                         if (*i_qgroups == 0)
2212                                 continue;
2213                         ret = add_qgroup_relation_item(trans, quota_root,
2214                                                        objectid, *i_qgroups);
2215                         if (ret && ret != -EEXIST)
2216                                 goto out;
2217                         ret = add_qgroup_relation_item(trans, quota_root,
2218                                                        *i_qgroups, objectid);
2219                         if (ret && ret != -EEXIST)
2220                                 goto out;
2221                 }
2222                 ret = 0;
2223         }
2224
2225
2226         spin_lock(&fs_info->qgroup_lock);
2227
2228         dstgroup = add_qgroup_rb(fs_info, objectid);
2229         if (IS_ERR(dstgroup)) {
2230                 ret = PTR_ERR(dstgroup);
2231                 goto unlock;
2232         }
2233
2234         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2235                 dstgroup->lim_flags = inherit->lim.flags;
2236                 dstgroup->max_rfer = inherit->lim.max_rfer;
2237                 dstgroup->max_excl = inherit->lim.max_excl;
2238                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2239                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2240
2241                 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2242                 if (ret) {
2243                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2244                         btrfs_info(fs_info,
2245                                    "unable to update quota limit for %llu",
2246                                    dstgroup->qgroupid);
2247                         goto unlock;
2248                 }
2249         }
2250
2251         if (srcid) {
2252                 srcgroup = find_qgroup_rb(fs_info, srcid);
2253                 if (!srcgroup)
2254                         goto unlock;
2255
2256                 /*
2257                  * We call inherit after we clone the root in order to make sure
2258                  * our counts don't go crazy, so at this point the only
2259                  * difference between the two roots should be the root node.
2260                  */
2261                 dstgroup->rfer = srcgroup->rfer;
2262                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2263                 dstgroup->excl = level_size;
2264                 dstgroup->excl_cmpr = level_size;
2265                 srcgroup->excl = level_size;
2266                 srcgroup->excl_cmpr = level_size;
2267
2268                 /* inherit the limit info */
2269                 dstgroup->lim_flags = srcgroup->lim_flags;
2270                 dstgroup->max_rfer = srcgroup->max_rfer;
2271                 dstgroup->max_excl = srcgroup->max_excl;
2272                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2273                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2274
2275                 qgroup_dirty(fs_info, dstgroup);
2276                 qgroup_dirty(fs_info, srcgroup);
2277         }
2278
2279         if (!inherit)
2280                 goto unlock;
2281
2282         i_qgroups = (u64 *)(inherit + 1);
2283         for (i = 0; i < inherit->num_qgroups; ++i) {
2284                 if (*i_qgroups) {
2285                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2286                         if (ret)
2287                                 goto unlock;
2288                 }
2289                 ++i_qgroups;
2290         }
2291
2292         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2293                 struct btrfs_qgroup *src;
2294                 struct btrfs_qgroup *dst;
2295
2296                 if (!i_qgroups[0] || !i_qgroups[1])
2297                         continue;
2298
2299                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2300                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2301
2302                 if (!src || !dst) {
2303                         ret = -EINVAL;
2304                         goto unlock;
2305                 }
2306
2307                 dst->rfer = src->rfer - level_size;
2308                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2309         }
2310         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2311                 struct btrfs_qgroup *src;
2312                 struct btrfs_qgroup *dst;
2313
2314                 if (!i_qgroups[0] || !i_qgroups[1])
2315                         continue;
2316
2317                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2318                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2319
2320                 if (!src || !dst) {
2321                         ret = -EINVAL;
2322                         goto unlock;
2323                 }
2324
2325                 dst->excl = src->excl + level_size;
2326                 dst->excl_cmpr = src->excl_cmpr + level_size;
2327         }
2328
2329 unlock:
2330         spin_unlock(&fs_info->qgroup_lock);
2331 out:
2332         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2333         return ret;
2334 }
2335
2336 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2337 {
2338         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2339             qg->reserved + (s64)qg->rfer + num_bytes > qg->max_rfer)
2340                 return false;
2341
2342         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2343             qg->reserved + (s64)qg->excl + num_bytes > qg->max_excl)
2344                 return false;
2345
2346         return true;
2347 }
2348
2349 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce)
2350 {
2351         struct btrfs_root *quota_root;
2352         struct btrfs_qgroup *qgroup;
2353         struct btrfs_fs_info *fs_info = root->fs_info;
2354         u64 ref_root = root->root_key.objectid;
2355         int ret = 0;
2356         int retried = 0;
2357         struct ulist_node *unode;
2358         struct ulist_iterator uiter;
2359
2360         if (!is_fstree(ref_root))
2361                 return 0;
2362
2363         if (num_bytes == 0)
2364                 return 0;
2365
2366         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2367             capable(CAP_SYS_RESOURCE))
2368                 enforce = false;
2369
2370 retry:
2371         spin_lock(&fs_info->qgroup_lock);
2372         quota_root = fs_info->quota_root;
2373         if (!quota_root)
2374                 goto out;
2375
2376         qgroup = find_qgroup_rb(fs_info, ref_root);
2377         if (!qgroup)
2378                 goto out;
2379
2380         /*
2381          * in a first step, we check all affected qgroups if any limits would
2382          * be exceeded
2383          */
2384         ulist_reinit(fs_info->qgroup_ulist);
2385         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2386                         (uintptr_t)qgroup, GFP_ATOMIC);
2387         if (ret < 0)
2388                 goto out;
2389         ULIST_ITER_INIT(&uiter);
2390         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2391                 struct btrfs_qgroup *qg;
2392                 struct btrfs_qgroup_list *glist;
2393
2394                 qg = unode_aux_to_qgroup(unode);
2395
2396                 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2397                         /*
2398                          * Commit the tree and retry, since we may have
2399                          * deletions which would free up space.
2400                          */
2401                         if (!retried && qg->reserved > 0) {
2402                                 struct btrfs_trans_handle *trans;
2403
2404                                 spin_unlock(&fs_info->qgroup_lock);
2405                                 ret = btrfs_start_delalloc_inodes(root, 0);
2406                                 if (ret)
2407                                         return ret;
2408                                 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
2409                                 trans = btrfs_join_transaction(root);
2410                                 if (IS_ERR(trans))
2411                                         return PTR_ERR(trans);
2412                                 ret = btrfs_commit_transaction(trans);
2413                                 if (ret)
2414                                         return ret;
2415                                 retried++;
2416                                 goto retry;
2417                         }
2418                         ret = -EDQUOT;
2419                         goto out;
2420                 }
2421
2422                 list_for_each_entry(glist, &qg->groups, next_group) {
2423                         ret = ulist_add(fs_info->qgroup_ulist,
2424                                         glist->group->qgroupid,
2425                                         (uintptr_t)glist->group, GFP_ATOMIC);
2426                         if (ret < 0)
2427                                 goto out;
2428                 }
2429         }
2430         ret = 0;
2431         /*
2432          * no limits exceeded, now record the reservation into all qgroups
2433          */
2434         ULIST_ITER_INIT(&uiter);
2435         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2436                 struct btrfs_qgroup *qg;
2437
2438                 qg = unode_aux_to_qgroup(unode);
2439
2440                 trace_qgroup_update_reserve(fs_info, qg, num_bytes);
2441                 qg->reserved += num_bytes;
2442         }
2443
2444 out:
2445         spin_unlock(&fs_info->qgroup_lock);
2446         return ret;
2447 }
2448
2449 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2450                                u64 ref_root, u64 num_bytes)
2451 {
2452         struct btrfs_root *quota_root;
2453         struct btrfs_qgroup *qgroup;
2454         struct ulist_node *unode;
2455         struct ulist_iterator uiter;
2456         int ret = 0;
2457
2458         if (!is_fstree(ref_root))
2459                 return;
2460
2461         if (num_bytes == 0)
2462                 return;
2463
2464         spin_lock(&fs_info->qgroup_lock);
2465
2466         quota_root = fs_info->quota_root;
2467         if (!quota_root)
2468                 goto out;
2469
2470         qgroup = find_qgroup_rb(fs_info, ref_root);
2471         if (!qgroup)
2472                 goto out;
2473
2474         ulist_reinit(fs_info->qgroup_ulist);
2475         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2476                         (uintptr_t)qgroup, GFP_ATOMIC);
2477         if (ret < 0)
2478                 goto out;
2479         ULIST_ITER_INIT(&uiter);
2480         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2481                 struct btrfs_qgroup *qg;
2482                 struct btrfs_qgroup_list *glist;
2483
2484                 qg = unode_aux_to_qgroup(unode);
2485
2486                 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes);
2487                 if (qg->reserved < num_bytes)
2488                         report_reserved_underflow(fs_info, qg, num_bytes);
2489                 else
2490                         qg->reserved -= num_bytes;
2491
2492                 list_for_each_entry(glist, &qg->groups, next_group) {
2493                         ret = ulist_add(fs_info->qgroup_ulist,
2494                                         glist->group->qgroupid,
2495                                         (uintptr_t)glist->group, GFP_ATOMIC);
2496                         if (ret < 0)
2497                                 goto out;
2498                 }
2499         }
2500
2501 out:
2502         spin_unlock(&fs_info->qgroup_lock);
2503 }
2504
2505 /*
2506  * returns < 0 on error, 0 when more leafs are to be scanned.
2507  * returns 1 when done.
2508  */
2509 static int
2510 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2511                    struct btrfs_trans_handle *trans)
2512 {
2513         struct btrfs_key found;
2514         struct extent_buffer *scratch_leaf = NULL;
2515         struct ulist *roots = NULL;
2516         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2517         u64 num_bytes;
2518         int slot;
2519         int ret;
2520
2521         mutex_lock(&fs_info->qgroup_rescan_lock);
2522         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2523                                          &fs_info->qgroup_rescan_progress,
2524                                          path, 1, 0);
2525
2526         btrfs_debug(fs_info,
2527                 "current progress key (%llu %u %llu), search_slot ret %d",
2528                 fs_info->qgroup_rescan_progress.objectid,
2529                 fs_info->qgroup_rescan_progress.type,
2530                 fs_info->qgroup_rescan_progress.offset, ret);
2531
2532         if (ret) {
2533                 /*
2534                  * The rescan is about to end, we will not be scanning any
2535                  * further blocks. We cannot unset the RESCAN flag here, because
2536                  * we want to commit the transaction if everything went well.
2537                  * To make the live accounting work in this phase, we set our
2538                  * scan progress pointer such that every real extent objectid
2539                  * will be smaller.
2540                  */
2541                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2542                 btrfs_release_path(path);
2543                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2544                 return ret;
2545         }
2546
2547         btrfs_item_key_to_cpu(path->nodes[0], &found,
2548                               btrfs_header_nritems(path->nodes[0]) - 1);
2549         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2550
2551         btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2552         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2553         if (!scratch_leaf) {
2554                 ret = -ENOMEM;
2555                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2556                 goto out;
2557         }
2558         extent_buffer_get(scratch_leaf);
2559         btrfs_tree_read_lock(scratch_leaf);
2560         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2561         slot = path->slots[0];
2562         btrfs_release_path(path);
2563         mutex_unlock(&fs_info->qgroup_rescan_lock);
2564
2565         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2566                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2567                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2568                     found.type != BTRFS_METADATA_ITEM_KEY)
2569                         continue;
2570                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2571                         num_bytes = fs_info->nodesize;
2572                 else
2573                         num_bytes = found.offset;
2574
2575                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2576                                            &roots);
2577                 if (ret < 0)
2578                         goto out;
2579                 /* For rescan, just pass old_roots as NULL */
2580                 ret = btrfs_qgroup_account_extent(trans, fs_info,
2581                                 found.objectid, num_bytes, NULL, roots);
2582                 if (ret < 0)
2583                         goto out;
2584         }
2585 out:
2586         if (scratch_leaf) {
2587                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2588                 free_extent_buffer(scratch_leaf);
2589         }
2590         btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2591
2592         return ret;
2593 }
2594
2595 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2596 {
2597         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2598                                                      qgroup_rescan_work);
2599         struct btrfs_path *path;
2600         struct btrfs_trans_handle *trans = NULL;
2601         int err = -ENOMEM;
2602         int ret = 0;
2603
2604         path = btrfs_alloc_path();
2605         if (!path)
2606                 goto out;
2607
2608         err = 0;
2609         while (!err && !btrfs_fs_closing(fs_info)) {
2610                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2611                 if (IS_ERR(trans)) {
2612                         err = PTR_ERR(trans);
2613                         break;
2614                 }
2615                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2616                         err = -EINTR;
2617                 } else {
2618                         err = qgroup_rescan_leaf(fs_info, path, trans);
2619                 }
2620                 if (err > 0)
2621                         btrfs_commit_transaction(trans);
2622                 else
2623                         btrfs_end_transaction(trans);
2624         }
2625
2626 out:
2627         btrfs_free_path(path);
2628
2629         mutex_lock(&fs_info->qgroup_rescan_lock);
2630         if (!btrfs_fs_closing(fs_info))
2631                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2632
2633         if (err > 0 &&
2634             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2635                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2636         } else if (err < 0) {
2637                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2638         }
2639         mutex_unlock(&fs_info->qgroup_rescan_lock);
2640
2641         /*
2642          * only update status, since the previous part has already updated the
2643          * qgroup info.
2644          */
2645         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2646         if (IS_ERR(trans)) {
2647                 err = PTR_ERR(trans);
2648                 btrfs_err(fs_info,
2649                           "fail to start transaction for status update: %d\n",
2650                           err);
2651                 goto done;
2652         }
2653         ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2654         if (ret < 0) {
2655                 err = ret;
2656                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2657         }
2658         btrfs_end_transaction(trans);
2659
2660         if (btrfs_fs_closing(fs_info)) {
2661                 btrfs_info(fs_info, "qgroup scan paused");
2662         } else if (err >= 0) {
2663                 btrfs_info(fs_info, "qgroup scan completed%s",
2664                         err > 0 ? " (inconsistency flag cleared)" : "");
2665         } else {
2666                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2667         }
2668
2669 done:
2670         mutex_lock(&fs_info->qgroup_rescan_lock);
2671         fs_info->qgroup_rescan_running = false;
2672         mutex_unlock(&fs_info->qgroup_rescan_lock);
2673         complete_all(&fs_info->qgroup_rescan_completion);
2674 }
2675
2676 /*
2677  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2678  * memory required for the rescan context.
2679  */
2680 static int
2681 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2682                    int init_flags)
2683 {
2684         int ret = 0;
2685
2686         if (!init_flags &&
2687             (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2688              !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2689                 ret = -EINVAL;
2690                 goto err;
2691         }
2692
2693         mutex_lock(&fs_info->qgroup_rescan_lock);
2694         spin_lock(&fs_info->qgroup_lock);
2695
2696         if (init_flags) {
2697                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2698                         ret = -EINPROGRESS;
2699                 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2700                         ret = -EINVAL;
2701
2702                 if (ret) {
2703                         spin_unlock(&fs_info->qgroup_lock);
2704                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2705                         goto err;
2706                 }
2707                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2708         }
2709
2710         memset(&fs_info->qgroup_rescan_progress, 0,
2711                 sizeof(fs_info->qgroup_rescan_progress));
2712         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2713         init_completion(&fs_info->qgroup_rescan_completion);
2714         fs_info->qgroup_rescan_running = true;
2715
2716         spin_unlock(&fs_info->qgroup_lock);
2717         mutex_unlock(&fs_info->qgroup_rescan_lock);
2718
2719         memset(&fs_info->qgroup_rescan_work, 0,
2720                sizeof(fs_info->qgroup_rescan_work));
2721         btrfs_init_work(&fs_info->qgroup_rescan_work,
2722                         btrfs_qgroup_rescan_helper,
2723                         btrfs_qgroup_rescan_worker, NULL, NULL);
2724
2725         if (ret) {
2726 err:
2727                 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2728                 return ret;
2729         }
2730
2731         return 0;
2732 }
2733
2734 static void
2735 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2736 {
2737         struct rb_node *n;
2738         struct btrfs_qgroup *qgroup;
2739
2740         spin_lock(&fs_info->qgroup_lock);
2741         /* clear all current qgroup tracking information */
2742         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2743                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2744                 qgroup->rfer = 0;
2745                 qgroup->rfer_cmpr = 0;
2746                 qgroup->excl = 0;
2747                 qgroup->excl_cmpr = 0;
2748         }
2749         spin_unlock(&fs_info->qgroup_lock);
2750 }
2751
2752 int
2753 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2754 {
2755         int ret = 0;
2756         struct btrfs_trans_handle *trans;
2757
2758         ret = qgroup_rescan_init(fs_info, 0, 1);
2759         if (ret)
2760                 return ret;
2761
2762         /*
2763          * We have set the rescan_progress to 0, which means no more
2764          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2765          * However, btrfs_qgroup_account_ref may be right after its call
2766          * to btrfs_find_all_roots, in which case it would still do the
2767          * accounting.
2768          * To solve this, we're committing the transaction, which will
2769          * ensure we run all delayed refs and only after that, we are
2770          * going to clear all tracking information for a clean start.
2771          */
2772
2773         trans = btrfs_join_transaction(fs_info->fs_root);
2774         if (IS_ERR(trans)) {
2775                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2776                 return PTR_ERR(trans);
2777         }
2778         ret = btrfs_commit_transaction(trans);
2779         if (ret) {
2780                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2781                 return ret;
2782         }
2783
2784         qgroup_rescan_zero_tracking(fs_info);
2785
2786         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2787                          &fs_info->qgroup_rescan_work);
2788
2789         return 0;
2790 }
2791
2792 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2793                                      bool interruptible)
2794 {
2795         int running;
2796         int ret = 0;
2797
2798         mutex_lock(&fs_info->qgroup_rescan_lock);
2799         spin_lock(&fs_info->qgroup_lock);
2800         running = fs_info->qgroup_rescan_running;
2801         spin_unlock(&fs_info->qgroup_lock);
2802         mutex_unlock(&fs_info->qgroup_rescan_lock);
2803
2804         if (!running)
2805                 return 0;
2806
2807         if (interruptible)
2808                 ret = wait_for_completion_interruptible(
2809                                         &fs_info->qgroup_rescan_completion);
2810         else
2811                 wait_for_completion(&fs_info->qgroup_rescan_completion);
2812
2813         return ret;
2814 }
2815
2816 /*
2817  * this is only called from open_ctree where we're still single threaded, thus
2818  * locking is omitted here.
2819  */
2820 void
2821 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2822 {
2823         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2824                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2825                                  &fs_info->qgroup_rescan_work);
2826 }
2827
2828 /*
2829  * Reserve qgroup space for range [start, start + len).
2830  *
2831  * This function will either reserve space from related qgroups or doing
2832  * nothing if the range is already reserved.
2833  *
2834  * Return 0 for successful reserve
2835  * Return <0 for error (including -EQUOT)
2836  *
2837  * NOTE: this function may sleep for memory allocation.
2838  *       if btrfs_qgroup_reserve_data() is called multiple times with
2839  *       same @reserved, caller must ensure when error happens it's OK
2840  *       to free *ALL* reserved space.
2841  */
2842 int btrfs_qgroup_reserve_data(struct inode *inode,
2843                         struct extent_changeset **reserved_ret, u64 start,
2844                         u64 len)
2845 {
2846         struct btrfs_root *root = BTRFS_I(inode)->root;
2847         struct ulist_node *unode;
2848         struct ulist_iterator uiter;
2849         struct extent_changeset *reserved;
2850         u64 orig_reserved;
2851         u64 to_reserve;
2852         int ret;
2853
2854         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2855             !is_fstree(root->objectid) || len == 0)
2856                 return 0;
2857
2858         /* @reserved parameter is mandatory for qgroup */
2859         if (WARN_ON(!reserved_ret))
2860                 return -EINVAL;
2861         if (!*reserved_ret) {
2862                 *reserved_ret = extent_changeset_alloc();
2863                 if (!*reserved_ret)
2864                         return -ENOMEM;
2865         }
2866         reserved = *reserved_ret;
2867         /* Record already reserved space */
2868         orig_reserved = reserved->bytes_changed;
2869         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2870                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
2871
2872         /* Newly reserved space */
2873         to_reserve = reserved->bytes_changed - orig_reserved;
2874         trace_btrfs_qgroup_reserve_data(inode, start, len,
2875                                         to_reserve, QGROUP_RESERVE);
2876         if (ret < 0)
2877                 goto cleanup;
2878         ret = qgroup_reserve(root, to_reserve, true);
2879         if (ret < 0)
2880                 goto cleanup;
2881
2882         return ret;
2883
2884 cleanup:
2885         /* cleanup *ALL* already reserved ranges */
2886         ULIST_ITER_INIT(&uiter);
2887         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
2888                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2889                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2890                                  GFP_NOFS);
2891         extent_changeset_release(reserved);
2892         return ret;
2893 }
2894
2895 /* Free ranges specified by @reserved, normally in error path */
2896 static int qgroup_free_reserved_data(struct inode *inode,
2897                         struct extent_changeset *reserved, u64 start, u64 len)
2898 {
2899         struct btrfs_root *root = BTRFS_I(inode)->root;
2900         struct ulist_node *unode;
2901         struct ulist_iterator uiter;
2902         struct extent_changeset changeset;
2903         int freed = 0;
2904         int ret;
2905
2906         extent_changeset_init(&changeset);
2907         len = round_up(start + len, root->fs_info->sectorsize);
2908         start = round_down(start, root->fs_info->sectorsize);
2909
2910         ULIST_ITER_INIT(&uiter);
2911         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
2912                 u64 range_start = unode->val;
2913                 /* unode->aux is the inclusive end */
2914                 u64 range_len = unode->aux - range_start + 1;
2915                 u64 free_start;
2916                 u64 free_len;
2917
2918                 extent_changeset_release(&changeset);
2919
2920                 /* Only free range in range [start, start + len) */
2921                 if (range_start >= start + len ||
2922                     range_start + range_len <= start)
2923                         continue;
2924                 free_start = max(range_start, start);
2925                 free_len = min(start + len, range_start + range_len) -
2926                            free_start;
2927                 /*
2928                  * TODO: To also modify reserved->ranges_reserved to reflect
2929                  * the modification.
2930                  *
2931                  * However as long as we free qgroup reserved according to
2932                  * EXTENT_QGROUP_RESERVED, we won't double free.
2933                  * So not need to rush.
2934                  */
2935                 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
2936                                 free_start, free_start + free_len - 1,
2937                                 EXTENT_QGROUP_RESERVED, &changeset);
2938                 if (ret < 0)
2939                         goto out;
2940                 freed += changeset.bytes_changed;
2941         }
2942         btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed);
2943         ret = freed;
2944 out:
2945         extent_changeset_release(&changeset);
2946         return ret;
2947 }
2948
2949 static int __btrfs_qgroup_release_data(struct inode *inode,
2950                         struct extent_changeset *reserved, u64 start, u64 len,
2951                         int free)
2952 {
2953         struct extent_changeset changeset;
2954         int trace_op = QGROUP_RELEASE;
2955         int ret;
2956
2957         /* In release case, we shouldn't have @reserved */
2958         WARN_ON(!free && reserved);
2959         if (free && reserved)
2960                 return qgroup_free_reserved_data(inode, reserved, start, len);
2961         extent_changeset_init(&changeset);
2962         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2963                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2964         if (ret < 0)
2965                 goto out;
2966
2967         if (free)
2968                 trace_op = QGROUP_FREE;
2969         trace_btrfs_qgroup_release_data(inode, start, len,
2970                                         changeset.bytes_changed, trace_op);
2971         if (free)
2972                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
2973                                 BTRFS_I(inode)->root->objectid,
2974                                 changeset.bytes_changed);
2975         ret = changeset.bytes_changed;
2976 out:
2977         extent_changeset_release(&changeset);
2978         return ret;
2979 }
2980
2981 /*
2982  * Free a reserved space range from io_tree and related qgroups
2983  *
2984  * Should be called when a range of pages get invalidated before reaching disk.
2985  * Or for error cleanup case.
2986  * if @reserved is given, only reserved range in [@start, @start + @len) will
2987  * be freed.
2988  *
2989  * For data written to disk, use btrfs_qgroup_release_data().
2990  *
2991  * NOTE: This function may sleep for memory allocation.
2992  */
2993 int btrfs_qgroup_free_data(struct inode *inode,
2994                         struct extent_changeset *reserved, u64 start, u64 len)
2995 {
2996         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
2997 }
2998
2999 /*
3000  * Release a reserved space range from io_tree only.
3001  *
3002  * Should be called when a range of pages get written to disk and corresponding
3003  * FILE_EXTENT is inserted into corresponding root.
3004  *
3005  * Since new qgroup accounting framework will only update qgroup numbers at
3006  * commit_transaction() time, its reserved space shouldn't be freed from
3007  * related qgroups.
3008  *
3009  * But we should release the range from io_tree, to allow further write to be
3010  * COWed.
3011  *
3012  * NOTE: This function may sleep for memory allocation.
3013  */
3014 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3015 {
3016         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3017 }
3018
3019 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3020                               bool enforce)
3021 {
3022         struct btrfs_fs_info *fs_info = root->fs_info;
3023         int ret;
3024
3025         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3026             !is_fstree(root->objectid) || num_bytes == 0)
3027                 return 0;
3028
3029         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3030         trace_qgroup_meta_reserve(root, (s64)num_bytes);
3031         ret = qgroup_reserve(root, num_bytes, enforce);
3032         if (ret < 0)
3033                 return ret;
3034         atomic64_add(num_bytes, &root->qgroup_meta_rsv);
3035         return ret;
3036 }
3037
3038 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
3039 {
3040         struct btrfs_fs_info *fs_info = root->fs_info;
3041         u64 reserved;
3042
3043         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3044             !is_fstree(root->objectid))
3045                 return;
3046
3047         reserved = atomic64_xchg(&root->qgroup_meta_rsv, 0);
3048         if (reserved == 0)
3049                 return;
3050         trace_qgroup_meta_reserve(root, -(s64)reserved);
3051         btrfs_qgroup_free_refroot(fs_info, root->objectid, reserved);
3052 }
3053
3054 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
3055 {
3056         struct btrfs_fs_info *fs_info = root->fs_info;
3057
3058         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3059             !is_fstree(root->objectid))
3060                 return;
3061
3062         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3063         WARN_ON(atomic64_read(&root->qgroup_meta_rsv) < num_bytes);
3064         atomic64_sub(num_bytes, &root->qgroup_meta_rsv);
3065         trace_qgroup_meta_reserve(root, -(s64)num_bytes);
3066         btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes);
3067 }
3068
3069 /*
3070  * Check qgroup reserved space leaking, normally at destroy inode
3071  * time
3072  */
3073 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3074 {
3075         struct extent_changeset changeset;
3076         struct ulist_node *unode;
3077         struct ulist_iterator iter;
3078         int ret;
3079
3080         extent_changeset_init(&changeset);
3081         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3082                         EXTENT_QGROUP_RESERVED, &changeset);
3083
3084         WARN_ON(ret < 0);
3085         if (WARN_ON(changeset.bytes_changed)) {
3086                 ULIST_ITER_INIT(&iter);
3087                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3088                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3089                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3090                                 inode->i_ino, unode->val, unode->aux);
3091                 }
3092                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3093                                 BTRFS_I(inode)->root->objectid,
3094                                 changeset.bytes_changed);
3095
3096         }
3097         extent_changeset_release(&changeset);
3098 }