fs/xfs/libxfs/xfs_refcount.c

   1 /*
   2  * Copyright (C) 2016 Oracle.  All Rights Reserved.
   3  *
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version 2
   9  * of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it would be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write the Free Software Foundation,
  18  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
  19  */
  20 #include "xfs.h"
  21 #include "xfs_fs.h"
  22 #include "xfs_shared.h"
  23 #include "xfs_format.h"
  24 #include "xfs_log_format.h"
  25 #include "xfs_trans_resv.h"
  26 #include "xfs_sb.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_defer.h"
  29 #include "xfs_btree.h"
  30 #include "xfs_bmap.h"
  31 #include "xfs_refcount_btree.h"
  32 #include "xfs_alloc.h"
  33 #include "xfs_errortag.h"
  34 #include "xfs_error.h"
  35 #include "xfs_trace.h"
  36 #include "xfs_cksum.h"
  37 #include "xfs_trans.h"
  38 #include "xfs_bit.h"
  39 #include "xfs_refcount.h"
  40 #include "xfs_rmap.h"
  41
  42 /* Allowable refcount adjustment amounts. */
  43 enum xfs_refc_adjust_op {
  44         XFS_REFCOUNT_ADJUST_INCREASE    = 1,
  45         XFS_REFCOUNT_ADJUST_DECREASE    = -1,
  46         XFS_REFCOUNT_ADJUST_COW_ALLOC   = 0,
  47         XFS_REFCOUNT_ADJUST_COW_FREE    = -1,
  48 };
  49
  50 STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
  51                 xfs_agblock_t agbno, xfs_extlen_t aglen,
  52                 struct xfs_defer_ops *dfops);
  53 STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
  54                 xfs_agblock_t agbno, xfs_extlen_t aglen,
  55                 struct xfs_defer_ops *dfops);
  56
  57 /*
  58  * Look up the first record less than or equal to [bno, len] in the btree
  59  * given by cur.
  60  */
  61 int
  62 xfs_refcount_lookup_le(
  63         struct xfs_btree_cur    *cur,
  64         xfs_agblock_t           bno,
  65         int                     *stat)
  66 {
  67         trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
  68                         XFS_LOOKUP_LE);
  69         cur->bc_rec.rc.rc_startblock = bno;
  70         cur->bc_rec.rc.rc_blockcount = 0;
  71         return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
  72 }
  73
  74 /*
  75  * Look up the first record greater than or equal to [bno, len] in the btree
  76  * given by cur.
  77  */
  78 int
  79 xfs_refcount_lookup_ge(
  80         struct xfs_btree_cur    *cur,
  81         xfs_agblock_t           bno,
  82         int                     *stat)
  83 {
  84         trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
  85                         XFS_LOOKUP_GE);
  86         cur->bc_rec.rc.rc_startblock = bno;
  87         cur->bc_rec.rc.rc_blockcount = 0;
  88         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
  89 }
  90
  91 /*
  92  * Look up the first record equal to [bno, len] in the btree
  93  * given by cur.
  94  */
  95 int
  96 xfs_refcount_lookup_eq(
  97         struct xfs_btree_cur    *cur,
  98         xfs_agblock_t           bno,
  99         int                     *stat)
 100 {
 101         trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
 102                         XFS_LOOKUP_LE);
 103         cur->bc_rec.rc.rc_startblock = bno;
 104         cur->bc_rec.rc.rc_blockcount = 0;
 105         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 106 }
 107
 108 /* Convert on-disk record to in-core format. */
 109 void
 110 xfs_refcount_btrec_to_irec(
 111         union xfs_btree_rec             *rec,
 112         struct xfs_refcount_irec        *irec)
 113 {
 114         irec->rc_startblock = be32_to_cpu(rec->refc.rc_startblock);
 115         irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
 116         irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
 117 }
 118
 119 /*
 120  * Get the data from the pointed-to record.
 121  */
 122 int
 123 xfs_refcount_get_rec(
 124         struct xfs_btree_cur            *cur,
 125         struct xfs_refcount_irec        *irec,
 126         int                             *stat)
 127 {
 128         union xfs_btree_rec             *rec;
 129         int                             error;
 130
 131         error = xfs_btree_get_rec(cur, &rec, stat);
 132         if (!error && *stat == 1) {
 133                 xfs_refcount_btrec_to_irec(rec, irec);
 134                 trace_xfs_refcount_get(cur->bc_mp, cur->bc_private.a.agno,
 135                                 irec);
 136         }
 137         return error;
 138 }
 139
 140 /*
 141  * Update the record referred to by cur to the value given
 142  * by [bno, len, refcount].
 143  * This either works (return 0) or gets an EFSCORRUPTED error.
 144  */
 145 STATIC int
 146 xfs_refcount_update(
 147         struct xfs_btree_cur            *cur,
 148         struct xfs_refcount_irec        *irec)
 149 {
 150         union xfs_btree_rec     rec;
 151         int                     error;
 152
 153         trace_xfs_refcount_update(cur->bc_mp, cur->bc_private.a.agno, irec);
 154         rec.refc.rc_startblock = cpu_to_be32(irec->rc_startblock);
 155         rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
 156         rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
 157         error = xfs_btree_update(cur, &rec);
 158         if (error)
 159                 trace_xfs_refcount_update_error(cur->bc_mp,
 160                                 cur->bc_private.a.agno, error, _RET_IP_);
 161         return error;
 162 }
 163
 164 /*
 165  * Insert the record referred to by cur to the value given
 166  * by [bno, len, refcount].
 167  * This either works (return 0) or gets an EFSCORRUPTED error.
 168  */
 169 int
 170 xfs_refcount_insert(
 171         struct xfs_btree_cur            *cur,
 172         struct xfs_refcount_irec        *irec,
 173         int                             *i)
 174 {
 175         int                             error;
 176
 177         trace_xfs_refcount_insert(cur->bc_mp, cur->bc_private.a.agno, irec);
 178         cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
 179         cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
 180         cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
 181         error = xfs_btree_insert(cur, i);
 182         if (error)
 183                 goto out_error;
 184         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
 185
 186 out_error:
 187         if (error)
 188                 trace_xfs_refcount_insert_error(cur->bc_mp,
 189                                 cur->bc_private.a.agno, error, _RET_IP_);
 190         return error;
 191 }
 192
 193 /*
 194  * Remove the record referred to by cur, then set the pointer to the spot
 195  * where the record could be re-inserted, in case we want to increment or
 196  * decrement the cursor.
 197  * This either works (return 0) or gets an EFSCORRUPTED error.
 198  */
 199 STATIC int
 200 xfs_refcount_delete(
 201         struct xfs_btree_cur    *cur,
 202         int                     *i)
 203 {
 204         struct xfs_refcount_irec        irec;
 205         int                     found_rec;
 206         int                     error;
 207
 208         error = xfs_refcount_get_rec(cur, &irec, &found_rec);
 209         if (error)
 210                 goto out_error;
 211         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 212         trace_xfs_refcount_delete(cur->bc_mp, cur->bc_private.a.agno, &irec);
 213         error = xfs_btree_delete(cur, i);
 214         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
 215         if (error)
 216                 goto out_error;
 217         error = xfs_refcount_lookup_ge(cur, irec.rc_startblock, &found_rec);
 218 out_error:
 219         if (error)
 220                 trace_xfs_refcount_delete_error(cur->bc_mp,
 221                                 cur->bc_private.a.agno, error, _RET_IP_);
 222         return error;
 223 }
 224
 225 /*
 226  * Adjusting the Reference Count
 227  *
 228  * As stated elsewhere, the reference count btree (refcbt) stores
 229  * >1 reference counts for extents of physical blocks.  In this
 230  * operation, we're either raising or lowering the reference count of
 231  * some subrange stored in the tree:
 232  *
 233  *      <------ adjustment range ------>
 234  * ----+   +---+-----+ +--+--------+---------
 235  *  2  |   | 3 |  4  | |17|   55   |   10
 236  * ----+   +---+-----+ +--+--------+---------
 237  * X axis is physical blocks number;
 238  * reference counts are the numbers inside the rectangles
 239  *
 240  * The first thing we need to do is to ensure that there are no
 241  * refcount extents crossing either boundary of the range to be
 242  * adjusted.  For any extent that does cross a boundary, split it into
 243  * two extents so that we can increment the refcount of one of the
 244  * pieces later:
 245  *
 246  *      <------ adjustment range ------>
 247  * ----+   +---+-----+ +--+--------+----+----
 248  *  2  |   | 3 |  2  | |17|   55   | 10 | 10
 249  * ----+   +---+-----+ +--+--------+----+----
 250  *
 251  * For this next step, let's assume that all the physical blocks in
 252  * the adjustment range are mapped to a file and are therefore in use
 253  * at least once.  Therefore, we can infer that any gap in the
 254  * refcount tree within the adjustment range represents a physical
 255  * extent with refcount == 1:
 256  *
 257  *      <------ adjustment range ------>
 258  * ----+---+---+-----+-+--+--------+----+----
 259  *  2  |"1"| 3 |  2  |1|17|   55   | 10 | 10
 260  * ----+---+---+-----+-+--+--------+----+----
 261  *      ^
 262  *
 263  * For each extent that falls within the interval range, figure out
 264  * which extent is to the left or the right of that extent.  Now we
 265  * have a left, current, and right extent.  If the new reference count
 266  * of the center extent enables us to merge left, center, and right
 267  * into one record covering all three, do so.  If the center extent is
 268  * at the left end of the range, abuts the left extent, and its new
 269  * reference count matches the left extent's record, then merge them.
 270  * If the center extent is at the right end of the range, abuts the
 271  * right extent, and the reference counts match, merge those.  In the
 272  * example, we can left merge (assuming an increment operation):
 273  *
 274  *      <------ adjustment range ------>
 275  * --------+---+-----+-+--+--------+----+----
 276  *    2    | 3 |  2  |1|17|   55   | 10 | 10
 277  * --------+---+-----+-+--+--------+----+----
 278  *          ^
 279  *
 280  * For all other extents within the range, adjust the reference count
 281  * or delete it if the refcount falls below 2.  If we were
 282  * incrementing, the end result looks like this:
 283  *
 284  *      <------ adjustment range ------>
 285  * --------+---+-----+-+--+--------+----+----
 286  *    2    | 4 |  3  |2|18|   56   | 11 | 10
 287  * --------+---+-----+-+--+--------+----+----
 288  *
 289  * The result of a decrement operation looks as such:
 290  *
 291  *      <------ adjustment range ------>
 292  * ----+   +---+       +--+--------+----+----
 293  *  2  |   | 2 |       |16|   54   |  9 | 10
 294  * ----+   +---+       +--+--------+----+----
 295  *      DDDD    111111DD
 296  *
 297  * The blocks marked "D" are freed; the blocks marked "1" are only
 298  * referenced once and therefore the record is removed from the
 299  * refcount btree.
 300  */
 301
 302 /* Next block after this extent. */
 303 static inline xfs_agblock_t
 304 xfs_refc_next(
 305         struct xfs_refcount_irec        *rc)
 306 {
 307         return rc->rc_startblock + rc->rc_blockcount;
 308 }
 309
 310 /*
 311  * Split a refcount extent that crosses agbno.
 312  */
 313 STATIC int
 314 xfs_refcount_split_extent(
 315         struct xfs_btree_cur            *cur,
 316         xfs_agblock_t                   agbno,
 317         bool                            *shape_changed)
 318 {
 319         struct xfs_refcount_irec        rcext, tmp;
 320         int                             found_rec;
 321         int                             error;
 322
 323         *shape_changed = false;
 324         error = xfs_refcount_lookup_le(cur, agbno, &found_rec);
 325         if (error)
 326                 goto out_error;
 327         if (!found_rec)
 328                 return 0;
 329
 330         error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
 331         if (error)
 332                 goto out_error;
 333         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 334         if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
 335                 return 0;
 336
 337         *shape_changed = true;
 338         trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_private.a.agno,
 339                         &rcext, agbno);
 340
 341         /* Establish the right extent. */
 342         tmp = rcext;
 343         tmp.rc_startblock = agbno;
 344         tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
 345         error = xfs_refcount_update(cur, &tmp);
 346         if (error)
 347                 goto out_error;
 348
 349         /* Insert the left extent. */
 350         tmp = rcext;
 351         tmp.rc_blockcount = agbno - rcext.rc_startblock;
 352         error = xfs_refcount_insert(cur, &tmp, &found_rec);
 353         if (error)
 354                 goto out_error;
 355         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 356         return error;
 357
 358 out_error:
 359         trace_xfs_refcount_split_extent_error(cur->bc_mp,
 360                         cur->bc_private.a.agno, error, _RET_IP_);
 361         return error;
 362 }
 363
 364 /*
 365  * Merge the left, center, and right extents.
 366  */
 367 STATIC int
 368 xfs_refcount_merge_center_extents(
 369         struct xfs_btree_cur            *cur,
 370         struct xfs_refcount_irec        *left,
 371         struct xfs_refcount_irec        *center,
 372         struct xfs_refcount_irec        *right,
 373         unsigned long long              extlen,
 374         xfs_extlen_t                    *aglen)
 375 {
 376         int                             error;
 377         int                             found_rec;
 378
 379         trace_xfs_refcount_merge_center_extents(cur->bc_mp,
 380                         cur->bc_private.a.agno, left, center, right);
 381
 382         /*
 383          * Make sure the center and right extents are not in the btree.
 384          * If the center extent was synthesized, the first delete call
 385          * removes the right extent and we skip the second deletion.
 386          * If center and right were in the btree, then the first delete
 387          * call removes the center and the second one removes the right
 388          * extent.
 389          */
 390         error = xfs_refcount_lookup_ge(cur, center->rc_startblock,
 391                         &found_rec);
 392         if (error)
 393                 goto out_error;
 394         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 395
 396         error = xfs_refcount_delete(cur, &found_rec);
 397         if (error)
 398                 goto out_error;
 399         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 400
 401         if (center->rc_refcount > 1) {
 402                 error = xfs_refcount_delete(cur, &found_rec);
 403                 if (error)
 404                         goto out_error;
 405                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 406                                 out_error);
 407         }
 408
 409         /* Enlarge the left extent. */
 410         error = xfs_refcount_lookup_le(cur, left->rc_startblock,
 411                         &found_rec);
 412         if (error)
 413                 goto out_error;
 414         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 415
 416         left->rc_blockcount = extlen;
 417         error = xfs_refcount_update(cur, left);
 418         if (error)
 419                 goto out_error;
 420
 421         *aglen = 0;
 422         return error;
 423
 424 out_error:
 425         trace_xfs_refcount_merge_center_extents_error(cur->bc_mp,
 426                         cur->bc_private.a.agno, error, _RET_IP_);
 427         return error;
 428 }
 429
 430 /*
 431  * Merge with the left extent.
 432  */
 433 STATIC int
 434 xfs_refcount_merge_left_extent(
 435         struct xfs_btree_cur            *cur,
 436         struct xfs_refcount_irec        *left,
 437         struct xfs_refcount_irec        *cleft,
 438         xfs_agblock_t                   *agbno,
 439         xfs_extlen_t                    *aglen)
 440 {
 441         int                             error;
 442         int                             found_rec;
 443
 444         trace_xfs_refcount_merge_left_extent(cur->bc_mp,
 445                         cur->bc_private.a.agno, left, cleft);
 446
 447         /* If the extent at agbno (cleft) wasn't synthesized, remove it. */
 448         if (cleft->rc_refcount > 1) {
 449                 error = xfs_refcount_lookup_le(cur, cleft->rc_startblock,
 450                                 &found_rec);
 451                 if (error)
 452                         goto out_error;
 453                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 454                                 out_error);
 455
 456                 error = xfs_refcount_delete(cur, &found_rec);
 457                 if (error)
 458                         goto out_error;
 459                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 460                                 out_error);
 461         }
 462
 463         /* Enlarge the left extent. */
 464         error = xfs_refcount_lookup_le(cur, left->rc_startblock,
 465                         &found_rec);
 466         if (error)
 467                 goto out_error;
 468         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 469
 470         left->rc_blockcount += cleft->rc_blockcount;
 471         error = xfs_refcount_update(cur, left);
 472         if (error)
 473                 goto out_error;
 474
 475         *agbno += cleft->rc_blockcount;
 476         *aglen -= cleft->rc_blockcount;
 477         return error;
 478
 479 out_error:
 480         trace_xfs_refcount_merge_left_extent_error(cur->bc_mp,
 481                         cur->bc_private.a.agno, error, _RET_IP_);
 482         return error;
 483 }
 484
 485 /*
 486  * Merge with the right extent.
 487  */
 488 STATIC int
 489 xfs_refcount_merge_right_extent(
 490         struct xfs_btree_cur            *cur,
 491         struct xfs_refcount_irec        *right,
 492         struct xfs_refcount_irec        *cright,
 493         xfs_extlen_t                    *aglen)
 494 {
 495         int                             error;
 496         int                             found_rec;
 497
 498         trace_xfs_refcount_merge_right_extent(cur->bc_mp,
 499                         cur->bc_private.a.agno, cright, right);
 500
 501         /*
 502          * If the extent ending at agbno+aglen (cright) wasn't synthesized,
 503          * remove it.
 504          */
 505         if (cright->rc_refcount > 1) {
 506                 error = xfs_refcount_lookup_le(cur, cright->rc_startblock,
 507                         &found_rec);
 508                 if (error)
 509                         goto out_error;
 510                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 511                                 out_error);
 512
 513                 error = xfs_refcount_delete(cur, &found_rec);
 514                 if (error)
 515                         goto out_error;
 516                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 517                                 out_error);
 518         }
 519
 520         /* Enlarge the right extent. */
 521         error = xfs_refcount_lookup_le(cur, right->rc_startblock,
 522                         &found_rec);
 523         if (error)
 524                 goto out_error;
 525         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 526
 527         right->rc_startblock -= cright->rc_blockcount;
 528         right->rc_blockcount += cright->rc_blockcount;
 529         error = xfs_refcount_update(cur, right);
 530         if (error)
 531                 goto out_error;
 532
 533         *aglen -= cright->rc_blockcount;
 534         return error;
 535
 536 out_error:
 537         trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
 538                         cur->bc_private.a.agno, error, _RET_IP_);
 539         return error;
 540 }
 541
 542 #define XFS_FIND_RCEXT_SHARED   1
 543 #define XFS_FIND_RCEXT_COW      2
 544 /*
 545  * Find the left extent and the one after it (cleft).  This function assumes
 546  * that we've already split any extent crossing agbno.
 547  */
 548 STATIC int
 549 xfs_refcount_find_left_extents(
 550         struct xfs_btree_cur            *cur,
 551         struct xfs_refcount_irec        *left,
 552         struct xfs_refcount_irec        *cleft,
 553         xfs_agblock_t                   agbno,
 554         xfs_extlen_t                    aglen,
 555         int                             flags)
 556 {
 557         struct xfs_refcount_irec        tmp;
 558         int                             error;
 559         int                             found_rec;
 560
 561         left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
 562         error = xfs_refcount_lookup_le(cur, agbno - 1, &found_rec);
 563         if (error)
 564                 goto out_error;
 565         if (!found_rec)
 566                 return 0;
 567
 568         error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 569         if (error)
 570                 goto out_error;
 571         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 572
 573         if (xfs_refc_next(&tmp) != agbno)
 574                 return 0;
 575         if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
 576                 return 0;
 577         if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
 578                 return 0;
 579         /* We have a left extent; retrieve (or invent) the next right one */
 580         *left = tmp;
 581
 582         error = xfs_btree_increment(cur, 0, &found_rec);
 583         if (error)
 584                 goto out_error;
 585         if (found_rec) {
 586                 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 587                 if (error)
 588                         goto out_error;
 589                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 590                                 out_error);
 591
 592                 /* if tmp starts at the end of our range, just use that */
 593                 if (tmp.rc_startblock == agbno)
 594                         *cleft = tmp;
 595                 else {
 596                         /*
 597                          * There's a gap in the refcntbt at the start of the
 598                          * range we're interested in (refcount == 1) so
 599                          * synthesize the implied extent and pass it back.
 600                          * We assume here that the agbno/aglen range was
 601                          * passed in from a data fork extent mapping and
 602                          * therefore is allocated to exactly one owner.
 603                          */
 604                         cleft->rc_startblock = agbno;
 605                         cleft->rc_blockcount = min(aglen,
 606                                         tmp.rc_startblock - agbno);
 607                         cleft->rc_refcount = 1;
 608                 }
 609         } else {
 610                 /*
 611                  * No extents, so pretend that there's one covering the whole
 612                  * range.
 613                  */
 614                 cleft->rc_startblock = agbno;
 615                 cleft->rc_blockcount = aglen;
 616                 cleft->rc_refcount = 1;
 617         }
 618         trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_private.a.agno,
 619                         left, cleft, agbno);
 620         return error;
 621
 622 out_error:
 623         trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
 624                         cur->bc_private.a.agno, error, _RET_IP_);
 625         return error;
 626 }
 627
 628 /*
 629  * Find the right extent and the one before it (cright).  This function
 630  * assumes that we've already split any extents crossing agbno + aglen.
 631  */
 632 STATIC int
 633 xfs_refcount_find_right_extents(
 634         struct xfs_btree_cur            *cur,
 635         struct xfs_refcount_irec        *right,
 636         struct xfs_refcount_irec        *cright,
 637         xfs_agblock_t                   agbno,
 638         xfs_extlen_t                    aglen,
 639         int                             flags)
 640 {
 641         struct xfs_refcount_irec        tmp;
 642         int                             error;
 643         int                             found_rec;
 644
 645         right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
 646         error = xfs_refcount_lookup_ge(cur, agbno + aglen, &found_rec);
 647         if (error)
 648                 goto out_error;
 649         if (!found_rec)
 650                 return 0;
 651
 652         error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 653         if (error)
 654                 goto out_error;
 655         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 656
 657         if (tmp.rc_startblock != agbno + aglen)
 658                 return 0;
 659         if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
 660                 return 0;
 661         if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
 662                 return 0;
 663         /* We have a right extent; retrieve (or invent) the next left one */
 664         *right = tmp;
 665
 666         error = xfs_btree_decrement(cur, 0, &found_rec);
 667         if (error)
 668                 goto out_error;
 669         if (found_rec) {
 670                 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 671                 if (error)
 672                         goto out_error;
 673                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 674                                 out_error);
 675
 676                 /* if tmp ends at the end of our range, just use that */
 677                 if (xfs_refc_next(&tmp) == agbno + aglen)
 678                         *cright = tmp;
 679                 else {
 680                         /*
 681                          * There's a gap in the refcntbt at the end of the
 682                          * range we're interested in (refcount == 1) so
 683                          * create the implied extent and pass it back.
 684                          * We assume here that the agbno/aglen range was
 685                          * passed in from a data fork extent mapping and
 686                          * therefore is allocated to exactly one owner.
 687                          */
 688                         cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
 689                         cright->rc_blockcount = right->rc_startblock -
 690                                         cright->rc_startblock;
 691                         cright->rc_refcount = 1;
 692                 }
 693         } else {
 694                 /*
 695                  * No extents, so pretend that there's one covering the whole
 696                  * range.
 697                  */
 698                 cright->rc_startblock = agbno;
 699                 cright->rc_blockcount = aglen;
 700                 cright->rc_refcount = 1;
 701         }
 702         trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_private.a.agno,
 703                         cright, right, agbno + aglen);
 704         return error;
 705
 706 out_error:
 707         trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
 708                         cur->bc_private.a.agno, error, _RET_IP_);
 709         return error;
 710 }
 711
 712 /* Is this extent valid? */
 713 static inline bool
 714 xfs_refc_valid(
 715         struct xfs_refcount_irec        *rc)
 716 {
 717         return rc->rc_startblock != NULLAGBLOCK;
 718 }
 719
 720 /*
 721  * Try to merge with any extents on the boundaries of the adjustment range.
 722  */
 723 STATIC int
 724 xfs_refcount_merge_extents(
 725         struct xfs_btree_cur    *cur,
 726         xfs_agblock_t           *agbno,
 727         xfs_extlen_t            *aglen,
 728         enum xfs_refc_adjust_op adjust,
 729         int                     flags,
 730         bool                    *shape_changed)
 731 {
 732         struct xfs_refcount_irec        left = {0}, cleft = {0};
 733         struct xfs_refcount_irec        cright = {0}, right = {0};
 734         int                             error;
 735         unsigned long long              ulen;
 736         bool                            cequal;
 737
 738         *shape_changed = false;
 739         /*
 740          * Find the extent just below agbno [left], just above agbno [cleft],
 741          * just below (agbno + aglen) [cright], and just above (agbno + aglen)
 742          * [right].
 743          */
 744         error = xfs_refcount_find_left_extents(cur, &left, &cleft, *agbno,
 745                         *aglen, flags);
 746         if (error)
 747                 return error;
 748         error = xfs_refcount_find_right_extents(cur, &right, &cright, *agbno,
 749                         *aglen, flags);
 750         if (error)
 751                 return error;
 752
 753         /* No left or right extent to merge; exit. */
 754         if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
 755                 return 0;
 756
 757         cequal = (cleft.rc_startblock == cright.rc_startblock) &&
 758                  (cleft.rc_blockcount == cright.rc_blockcount);
 759
 760         /* Try to merge left, cleft, and right.  cleft must == cright. */
 761         ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount +
 762                         right.rc_blockcount;
 763         if (xfs_refc_valid(&left) && xfs_refc_valid(&right) &&
 764             xfs_refc_valid(&cleft) && xfs_refc_valid(&cright) && cequal &&
 765             left.rc_refcount == cleft.rc_refcount + adjust &&
 766             right.rc_refcount == cleft.rc_refcount + adjust &&
 767             ulen < MAXREFCEXTLEN) {
 768                 *shape_changed = true;
 769                 return xfs_refcount_merge_center_extents(cur, &left, &cleft,
 770                                 &right, ulen, aglen);
 771         }
 772
 773         /* Try to merge left and cleft. */
 774         ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount;
 775         if (xfs_refc_valid(&left) && xfs_refc_valid(&cleft) &&
 776             left.rc_refcount == cleft.rc_refcount + adjust &&
 777             ulen < MAXREFCEXTLEN) {
 778                 *shape_changed = true;
 779                 error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
 780                                 agbno, aglen);
 781                 if (error)
 782                         return error;
 783
 784                 /*
 785                  * If we just merged left + cleft and cleft == cright,
 786                  * we no longer have a cright to merge with right.  We're done.
 787                  */
 788                 if (cequal)
 789                         return 0;
 790         }
 791
 792         /* Try to merge cright and right. */
 793         ulen = (unsigned long long)right.rc_blockcount + cright.rc_blockcount;
 794         if (xfs_refc_valid(&right) && xfs_refc_valid(&cright) &&
 795             right.rc_refcount == cright.rc_refcount + adjust &&
 796             ulen < MAXREFCEXTLEN) {
 797                 *shape_changed = true;
 798                 return xfs_refcount_merge_right_extent(cur, &right, &cright,
 799                                 aglen);
 800         }
 801
 802         return error;
 803 }
 804
 805 /*
 806  * XXX: This is a pretty hand-wavy estimate.  The penalty for guessing
 807  * true incorrectly is a shutdown FS; the penalty for guessing false
 808  * incorrectly is more transaction rolls than might be necessary.
 809  * Be conservative here.
 810  */
 811 static bool
 812 xfs_refcount_still_have_space(
 813         struct xfs_btree_cur            *cur)
 814 {
 815         unsigned long                   overhead;
 816
 817         overhead = cur->bc_private.a.priv.refc.shape_changes *
 818                         xfs_allocfree_log_count(cur->bc_mp, 1);
 819         overhead *= cur->bc_mp->m_sb.sb_blocksize;
 820
 821         /*
 822          * Only allow 2 refcount extent updates per transaction if the
 823          * refcount continue update "error" has been injected.
 824          */
 825         if (cur->bc_private.a.priv.refc.nr_ops > 2 &&
 826             XFS_TEST_ERROR(false, cur->bc_mp,
 827                         XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
 828                 return false;
 829
 830         if (cur->bc_private.a.priv.refc.nr_ops == 0)
 831                 return true;
 832         else if (overhead > cur->bc_tp->t_log_res)
 833                 return false;
 834         return  cur->bc_tp->t_log_res - overhead >
 835                 cur->bc_private.a.priv.refc.nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
 836 }
 837
 838 /*
 839  * Adjust the refcounts of middle extents.  At this point we should have
 840  * split extents that crossed the adjustment range; merged with adjacent
 841  * extents; and updated agbno/aglen to reflect the merges.  Therefore,
 842  * all we have to do is update the extents inside [agbno, agbno + aglen].
 843  */
 844 STATIC int
 845 xfs_refcount_adjust_extents(
 846         struct xfs_btree_cur    *cur,
 847         xfs_agblock_t           *agbno,
 848         xfs_extlen_t            *aglen,
 849         enum xfs_refc_adjust_op adj,
 850         struct xfs_defer_ops    *dfops,
 851         struct xfs_owner_info   *oinfo)
 852 {
 853         struct xfs_refcount_irec        ext, tmp;
 854         int                             error;
 855         int                             found_rec, found_tmp;
 856         xfs_fsblock_t                   fsbno;
 857
 858         /* Merging did all the work already. */
 859         if (*aglen == 0)
 860                 return 0;
 861
 862         error = xfs_refcount_lookup_ge(cur, *agbno, &found_rec);
 863         if (error)
 864                 goto out_error;
 865
 866         while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
 867                 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
 868                 if (error)
 869                         goto out_error;
 870                 if (!found_rec) {
 871                         ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
 872                         ext.rc_blockcount = 0;
 873                         ext.rc_refcount = 0;
 874                 }
 875
 876                 /*
 877                  * Deal with a hole in the refcount tree; if a file maps to
 878                  * these blocks and there's no refcountbt record, pretend that
 879                  * there is one with refcount == 1.
 880                  */
 881                 if (ext.rc_startblock != *agbno) {
 882                         tmp.rc_startblock = *agbno;
 883                         tmp.rc_blockcount = min(*aglen,
 884                                         ext.rc_startblock - *agbno);
 885                         tmp.rc_refcount = 1 + adj;
 886                         trace_xfs_refcount_modify_extent(cur->bc_mp,
 887                                         cur->bc_private.a.agno, &tmp);
 888
 889                         /*
 890                          * Either cover the hole (increment) or
 891                          * delete the range (decrement).
 892                          */
 893                         if (tmp.rc_refcount) {
 894                                 error = xfs_refcount_insert(cur, &tmp,
 895                                                 &found_tmp);
 896                                 if (error)
 897                                         goto out_error;
 898                                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
 899                                                 found_tmp == 1, out_error);
 900                                 cur->bc_private.a.priv.refc.nr_ops++;
 901                         } else {
 902                                 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
 903                                                 cur->bc_private.a.agno,
 904                                                 tmp.rc_startblock);
 905                                 xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
 906                                                 tmp.rc_blockcount, oinfo);
 907                         }
 908
 909                         (*agbno) += tmp.rc_blockcount;
 910                         (*aglen) -= tmp.rc_blockcount;
 911
 912                         error = xfs_refcount_lookup_ge(cur, *agbno,
 913                                         &found_rec);
 914                         if (error)
 915                                 goto out_error;
 916                 }
 917
 918                 /* Stop if there's nothing left to modify */
 919                 if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
 920                         break;
 921
 922                 /*
 923                  * Adjust the reference count and either update the tree
 924                  * (incr) or free the blocks (decr).
 925                  */
 926                 if (ext.rc_refcount == MAXREFCOUNT)
 927                         goto skip;
 928                 ext.rc_refcount += adj;
 929                 trace_xfs_refcount_modify_extent(cur->bc_mp,
 930                                 cur->bc_private.a.agno, &ext);
 931                 if (ext.rc_refcount > 1) {
 932                         error = xfs_refcount_update(cur, &ext);
 933                         if (error)
 934                                 goto out_error;
 935                         cur->bc_private.a.priv.refc.nr_ops++;
 936                 } else if (ext.rc_refcount == 1) {
 937                         error = xfs_refcount_delete(cur, &found_rec);
 938                         if (error)
 939                                 goto out_error;
 940                         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
 941                                         found_rec == 1, out_error);
 942                         cur->bc_private.a.priv.refc.nr_ops++;
 943                         goto advloop;
 944                 } else {
 945                         fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
 946                                         cur->bc_private.a.agno,
 947                                         ext.rc_startblock);
 948                         xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
 949                                         ext.rc_blockcount, oinfo);
 950                 }
 951
 952 skip:
 953                 error = xfs_btree_increment(cur, 0, &found_rec);
 954                 if (error)
 955                         goto out_error;
 956
 957 advloop:
 958                 (*agbno) += ext.rc_blockcount;
 959                 (*aglen) -= ext.rc_blockcount;
 960         }
 961
 962         return error;
 963 out_error:
 964         trace_xfs_refcount_modify_extent_error(cur->bc_mp,
 965                         cur->bc_private.a.agno, error, _RET_IP_);
 966         return error;
 967 }
 968
 969 /* Adjust the reference count of a range of AG blocks. */
 970 STATIC int
 971 xfs_refcount_adjust(
 972         struct xfs_btree_cur    *cur,
 973         xfs_agblock_t           agbno,
 974         xfs_extlen_t            aglen,
 975         xfs_agblock_t           *new_agbno,
 976         xfs_extlen_t            *new_aglen,
 977         enum xfs_refc_adjust_op adj,
 978         struct xfs_defer_ops    *dfops,
 979         struct xfs_owner_info   *oinfo)
 980 {
 981         bool                    shape_changed;
 982         int                     shape_changes = 0;
 983         int                     error;
 984
 985         *new_agbno = agbno;
 986         *new_aglen = aglen;
 987         if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
 988                 trace_xfs_refcount_increase(cur->bc_mp, cur->bc_private.a.agno,
 989                                 agbno, aglen);
 990         else
 991                 trace_xfs_refcount_decrease(cur->bc_mp, cur->bc_private.a.agno,
 992                                 agbno, aglen);
 993
 994         /*
 995          * Ensure that no rcextents cross the boundary of the adjustment range.
 996          */
 997         error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
 998         if (error)
 999                 goto out_error;
1000         if (shape_changed)
1001                 shape_changes++;
1002
1003         error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1004         if (error)
1005                 goto out_error;
1006         if (shape_changed)
1007                 shape_changes++;
1008
1009         /*
1010          * Try to merge with the left or right extents of the range.
1011          */
1012         error = xfs_refcount_merge_extents(cur, new_agbno, new_aglen, adj,
1013                         XFS_FIND_RCEXT_SHARED, &shape_changed);
1014         if (error)
1015                 goto out_error;
1016         if (shape_changed)
1017                 shape_changes++;
1018         if (shape_changes)
1019                 cur->bc_private.a.priv.refc.shape_changes++;
1020
1021         /* Now that we've taken care of the ends, adjust the middle extents */
1022         error = xfs_refcount_adjust_extents(cur, new_agbno, new_aglen,
1023                         adj, dfops, oinfo);
1024         if (error)
1025                 goto out_error;
1026
1027         return 0;
1028
1029 out_error:
1030         trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_private.a.agno,
1031                         error, _RET_IP_);
1032         return error;
1033 }
1034
1035 /* Clean up after calling xfs_refcount_finish_one. */
1036 void
1037 xfs_refcount_finish_one_cleanup(
1038         struct xfs_trans        *tp,
1039         struct xfs_btree_cur    *rcur,
1040         int                     error)
1041 {
1042         struct xfs_buf          *agbp;
1043
1044         if (rcur == NULL)
1045                 return;
1046         agbp = rcur->bc_private.a.agbp;
1047         xfs_btree_del_cursor(rcur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1048         if (error)
1049                 xfs_trans_brelse(tp, agbp);
1050 }
1051
1052 /*
1053  * Process one of the deferred refcount operations.  We pass back the
1054  * btree cursor to maintain our lock on the btree between calls.
1055  * This saves time and eliminates a buffer deadlock between the
1056  * superblock and the AGF because we'll always grab them in the same
1057  * order.
1058  */
1059 int
1060 xfs_refcount_finish_one(
1061         struct xfs_trans                *tp,
1062         struct xfs_defer_ops            *dfops,
1063         enum xfs_refcount_intent_type   type,
1064         xfs_fsblock_t                   startblock,
1065         xfs_extlen_t                    blockcount,
1066         xfs_fsblock_t                   *new_fsb,
1067         xfs_extlen_t                    *new_len,
1068         struct xfs_btree_cur            **pcur)
1069 {
1070         struct xfs_mount                *mp = tp->t_mountp;
1071         struct xfs_btree_cur            *rcur;
1072         struct xfs_buf                  *agbp = NULL;
1073         int                             error = 0;
1074         xfs_agnumber_t                  agno;
1075         xfs_agblock_t                   bno;
1076         xfs_agblock_t                   new_agbno;
1077         unsigned long                   nr_ops = 0;
1078         int                             shape_changes = 0;
1079
1080         agno = XFS_FSB_TO_AGNO(mp, startblock);
1081         ASSERT(agno != NULLAGNUMBER);
1082         bno = XFS_FSB_TO_AGBNO(mp, startblock);
1083
1084         trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, startblock),
1085                         type, XFS_FSB_TO_AGBNO(mp, startblock),
1086                         blockcount);
1087
1088         if (XFS_TEST_ERROR(false, mp,
1089                         XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1090                 return -EIO;
1091
1092         /*
1093          * If we haven't gotten a cursor or the cursor AG doesn't match
1094          * the startblock, get one now.
1095          */
1096         rcur = *pcur;
1097         if (rcur != NULL && rcur->bc_private.a.agno != agno) {
1098                 nr_ops = rcur->bc_private.a.priv.refc.nr_ops;
1099                 shape_changes = rcur->bc_private.a.priv.refc.shape_changes;
1100                 xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1101                 rcur = NULL;
1102                 *pcur = NULL;
1103         }
1104         if (rcur == NULL) {
1105                 error = xfs_alloc_read_agf(tp->t_mountp, tp, agno,
1106                                 XFS_ALLOC_FLAG_FREEING, &agbp);
1107                 if (error)
1108                         return error;
1109                 if (!agbp)
1110                         return -EFSCORRUPTED;
1111
1112                 rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno, dfops);
1113                 if (!rcur) {
1114                         error = -ENOMEM;
1115                         goto out_cur;
1116                 }
1117                 rcur->bc_private.a.priv.refc.nr_ops = nr_ops;
1118                 rcur->bc_private.a.priv.refc.shape_changes = shape_changes;
1119         }
1120         *pcur = rcur;
1121
1122         switch (type) {
1123         case XFS_REFCOUNT_INCREASE:
1124                 error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1125                         new_len, XFS_REFCOUNT_ADJUST_INCREASE, dfops, NULL);
1126                 *new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1127                 break;
1128         case XFS_REFCOUNT_DECREASE:
1129                 error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1130                         new_len, XFS_REFCOUNT_ADJUST_DECREASE, dfops, NULL);
1131                 *new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1132                 break;
1133         case XFS_REFCOUNT_ALLOC_COW:
1134                 *new_fsb = startblock + blockcount;
1135                 *new_len = 0;
1136                 error = __xfs_refcount_cow_alloc(rcur, bno, blockcount, dfops);
1137                 break;
1138         case XFS_REFCOUNT_FREE_COW:
1139                 *new_fsb = startblock + blockcount;
1140                 *new_len = 0;
1141                 error = __xfs_refcount_cow_free(rcur, bno, blockcount, dfops);
1142                 break;
1143         default:
1144                 ASSERT(0);
1145                 error = -EFSCORRUPTED;
1146         }
1147         if (!error && *new_len > 0)
1148                 trace_xfs_refcount_finish_one_leftover(mp, agno, type,
1149                                 bno, blockcount, new_agbno, *new_len);
1150         return error;
1151
1152 out_cur:
1153         xfs_trans_brelse(tp, agbp);
1154
1155         return error;
1156 }
1157
1158 /*
1159  * Record a refcount intent for later processing.
1160  */
1161 static int
1162 __xfs_refcount_add(
1163         struct xfs_mount                *mp,
1164         struct xfs_defer_ops            *dfops,
1165         enum xfs_refcount_intent_type   type,
1166         xfs_fsblock_t                   startblock,
1167         xfs_extlen_t                    blockcount)
1168 {
1169         struct xfs_refcount_intent      *ri;
1170
1171         trace_xfs_refcount_defer(mp, XFS_FSB_TO_AGNO(mp, startblock),
1172                         type, XFS_FSB_TO_AGBNO(mp, startblock),
1173                         blockcount);
1174
1175         ri = kmem_alloc(sizeof(struct xfs_refcount_intent),
1176                         KM_SLEEP | KM_NOFS);
1177         INIT_LIST_HEAD(&ri->ri_list);
1178         ri->ri_type = type;
1179         ri->ri_startblock = startblock;
1180         ri->ri_blockcount = blockcount;
1181
1182         xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_REFCOUNT, &ri->ri_list);
1183         return 0;
1184 }
1185
1186 /*
1187  * Increase the reference count of the blocks backing a file's extent.
1188  */
1189 int
1190 xfs_refcount_increase_extent(
1191         struct xfs_mount                *mp,
1192         struct xfs_defer_ops            *dfops,
1193         struct xfs_bmbt_irec            *PREV)
1194 {
1195         if (!xfs_sb_version_hasreflink(&mp->m_sb))
1196                 return 0;
1197
1198         return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_INCREASE,
1199                         PREV->br_startblock, PREV->br_blockcount);
1200 }
1201
1202 /*
1203  * Decrease the reference count of the blocks backing a file's extent.
1204  */
1205 int
1206 xfs_refcount_decrease_extent(
1207         struct xfs_mount                *mp,
1208         struct xfs_defer_ops            *dfops,
1209         struct xfs_bmbt_irec            *PREV)
1210 {
1211         if (!xfs_sb_version_hasreflink(&mp->m_sb))
1212                 return 0;
1213
1214         return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_DECREASE,
1215                         PREV->br_startblock, PREV->br_blockcount);
1216 }
1217
1218 /*
1219  * Given an AG extent, find the lowest-numbered run of shared blocks
1220  * within that range and return the range in fbno/flen.  If
1221  * find_end_of_shared is set, return the longest contiguous extent of
1222  * shared blocks; if not, just return the first extent we find.  If no
1223  * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1224  * and 0, respectively.
1225  */
1226 int
1227 xfs_refcount_find_shared(
1228         struct xfs_btree_cur            *cur,
1229         xfs_agblock_t                   agbno,
1230         xfs_extlen_t                    aglen,
1231         xfs_agblock_t                   *fbno,
1232         xfs_extlen_t                    *flen,
1233         bool                            find_end_of_shared)
1234 {
1235         struct xfs_refcount_irec        tmp;
1236         int                             i;
1237         int                             have;
1238         int                             error;
1239
1240         trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_private.a.agno,
1241                         agbno, aglen);
1242
1243         /* By default, skip the whole range */
1244         *fbno = NULLAGBLOCK;
1245         *flen = 0;
1246
1247         /* Try to find a refcount extent that crosses the start */
1248         error = xfs_refcount_lookup_le(cur, agbno, &have);
1249         if (error)
1250                 goto out_error;
1251         if (!have) {
1252                 /* No left extent, look at the next one */
1253                 error = xfs_btree_increment(cur, 0, &have);
1254                 if (error)
1255                         goto out_error;
1256                 if (!have)
1257                         goto done;
1258         }
1259         error = xfs_refcount_get_rec(cur, &tmp, &i);
1260         if (error)
1261                 goto out_error;
1262         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1263
1264         /* If the extent ends before the start, look at the next one */
1265         if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1266                 error = xfs_btree_increment(cur, 0, &have);
1267                 if (error)
1268                         goto out_error;
1269                 if (!have)
1270                         goto done;
1271                 error = xfs_refcount_get_rec(cur, &tmp, &i);
1272                 if (error)
1273                         goto out_error;
1274                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1275         }
1276
1277         /* If the extent starts after the range we want, bail out */
1278         if (tmp.rc_startblock >= agbno + aglen)
1279                 goto done;
1280
1281         /* We found the start of a shared extent! */
1282         if (tmp.rc_startblock < agbno) {
1283                 tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1284                 tmp.rc_startblock = agbno;
1285         }
1286
1287         *fbno = tmp.rc_startblock;
1288         *flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1289         if (!find_end_of_shared)
1290                 goto done;
1291
1292         /* Otherwise, find the end of this shared extent */
1293         while (*fbno + *flen < agbno + aglen) {
1294                 error = xfs_btree_increment(cur, 0, &have);
1295                 if (error)
1296                         goto out_error;
1297                 if (!have)
1298                         break;
1299                 error = xfs_refcount_get_rec(cur, &tmp, &i);
1300                 if (error)
1301                         goto out_error;
1302                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1303                 if (tmp.rc_startblock >= agbno + aglen ||
1304                     tmp.rc_startblock != *fbno + *flen)
1305                         break;
1306                 *flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1307         }
1308
1309 done:
1310         trace_xfs_refcount_find_shared_result(cur->bc_mp,
1311                         cur->bc_private.a.agno, *fbno, *flen);
1312
1313 out_error:
1314         if (error)
1315                 trace_xfs_refcount_find_shared_error(cur->bc_mp,
1316                                 cur->bc_private.a.agno, error, _RET_IP_);
1317         return error;
1318 }
1319
1320 /*
1321  * Recovering CoW Blocks After a Crash
1322  *
1323  * Due to the way that the copy on write mechanism works, there's a window of
1324  * opportunity in which we can lose track of allocated blocks during a crash.
1325  * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1326  * causes blocks to be allocated and stored in the CoW fork.  The blocks are
1327  * no longer in the free space btree but are not otherwise recorded anywhere
1328  * until the write completes and the blocks are mapped into the file.  A crash
1329  * in between allocation and remapping results in the replacement blocks being
1330  * lost.  This situation is exacerbated by the CoW extent size hint because
1331  * allocations can hang around for long time.
1332  *
1333  * However, there is a place where we can record these allocations before they
1334  * become mappings -- the reference count btree.  The btree does not record
1335  * extents with refcount == 1, so we can record allocations with a refcount of
1336  * 1.  Blocks being used for CoW writeout cannot be shared, so there should be
1337  * no conflict with shared block records.  These mappings should be created
1338  * when we allocate blocks to the CoW fork and deleted when they're removed
1339  * from the CoW fork.
1340  *
1341  * Minor nit: records for in-progress CoW allocations and records for shared
1342  * extents must never be merged, to preserve the property that (except for CoW
1343  * allocations) there are no refcount btree entries with refcount == 1.  The
1344  * only time this could potentially happen is when unsharing a block that's
1345  * adjacent to CoW allocations, so we must be careful to avoid this.
1346  *
1347  * At mount time we recover lost CoW allocations by searching the refcount
1348  * btree for these refcount == 1 mappings.  These represent CoW allocations
1349  * that were in progress at the time the filesystem went down, so we can free
1350  * them to get the space back.
1351  *
1352  * This mechanism is superior to creating EFIs for unmapped CoW extents for
1353  * several reasons -- first, EFIs pin the tail of the log and would have to be
1354  * periodically relogged to avoid filling up the log.  Second, CoW completions
1355  * will have to file an EFD and create new EFIs for whatever remains in the
1356  * CoW fork; this partially takes care of (1) but extent-size reservations
1357  * will have to periodically relog even if there's no writeout in progress.
1358  * This can happen if the CoW extent size hint is set, which you really want.
1359  * Third, EFIs cannot currently be automatically relogged into newer
1360  * transactions to advance the log tail.  Fourth, stuffing the log full of
1361  * EFIs places an upper bound on the number of CoW allocations that can be
1362  * held filesystem-wide at any given time.  Recording them in the refcount
1363  * btree doesn't require us to maintain any state in memory and doesn't pin
1364  * the log.
1365  */
1366 /*
1367  * Adjust the refcounts of CoW allocations.  These allocations are "magic"
1368  * in that they're not referenced anywhere else in the filesystem, so we
1369  * stash them in the refcount btree with a refcount of 1 until either file
1370  * remapping (or CoW cancellation) happens.
1371  */
1372 STATIC int
1373 xfs_refcount_adjust_cow_extents(
1374         struct xfs_btree_cur    *cur,
1375         xfs_agblock_t           agbno,
1376         xfs_extlen_t            aglen,
1377         enum xfs_refc_adjust_op adj)
1378 {
1379         struct xfs_refcount_irec        ext, tmp;
1380         int                             error;
1381         int                             found_rec, found_tmp;
1382
1383         if (aglen == 0)
1384                 return 0;
1385
1386         /* Find any overlapping refcount records */
1387         error = xfs_refcount_lookup_ge(cur, agbno, &found_rec);
1388         if (error)
1389                 goto out_error;
1390         error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1391         if (error)
1392                 goto out_error;
1393         if (!found_rec) {
1394                 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks +
1395                                 XFS_REFC_COW_START;
1396                 ext.rc_blockcount = 0;
1397                 ext.rc_refcount = 0;
1398         }
1399
1400         switch (adj) {
1401         case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1402                 /* Adding a CoW reservation, there should be nothing here. */
1403                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1404                                 ext.rc_startblock >= agbno + aglen, out_error);
1405
1406                 tmp.rc_startblock = agbno;
1407                 tmp.rc_blockcount = aglen;
1408                 tmp.rc_refcount = 1;
1409                 trace_xfs_refcount_modify_extent(cur->bc_mp,
1410                                 cur->bc_private.a.agno, &tmp);
1411
1412                 error = xfs_refcount_insert(cur, &tmp,
1413                                 &found_tmp);
1414                 if (error)
1415                         goto out_error;
1416                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1417                                 found_tmp == 1, out_error);
1418                 break;
1419         case XFS_REFCOUNT_ADJUST_COW_FREE:
1420                 /* Removing a CoW reservation, there should be one extent. */
1421                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1422                         ext.rc_startblock == agbno, out_error);
1423                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1424                         ext.rc_blockcount == aglen, out_error);
1425                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1426                         ext.rc_refcount == 1, out_error);
1427
1428                 ext.rc_refcount = 0;
1429                 trace_xfs_refcount_modify_extent(cur->bc_mp,
1430                                 cur->bc_private.a.agno, &ext);
1431                 error = xfs_refcount_delete(cur, &found_rec);
1432                 if (error)
1433                         goto out_error;
1434                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1435                                 found_rec == 1, out_error);
1436                 break;
1437         default:
1438                 ASSERT(0);
1439         }
1440
1441         return error;
1442 out_error:
1443         trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1444                         cur->bc_private.a.agno, error, _RET_IP_);
1445         return error;
1446 }
1447
1448 /*
1449  * Add or remove refcount btree entries for CoW reservations.
1450  */
1451 STATIC int
1452 xfs_refcount_adjust_cow(
1453         struct xfs_btree_cur    *cur,
1454         xfs_agblock_t           agbno,
1455         xfs_extlen_t            aglen,
1456         enum xfs_refc_adjust_op adj)
1457 {
1458         bool                    shape_changed;
1459         int                     error;
1460
1461         agbno += XFS_REFC_COW_START;
1462
1463         /*
1464          * Ensure that no rcextents cross the boundary of the adjustment range.
1465          */
1466         error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
1467         if (error)
1468                 goto out_error;
1469
1470         error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1471         if (error)
1472                 goto out_error;
1473
1474         /*
1475          * Try to merge with the left or right extents of the range.
1476          */
1477         error = xfs_refcount_merge_extents(cur, &agbno, &aglen, adj,
1478                         XFS_FIND_RCEXT_COW, &shape_changed);
1479         if (error)
1480                 goto out_error;
1481
1482         /* Now that we've taken care of the ends, adjust the middle extents */
1483         error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1484         if (error)
1485                 goto out_error;
1486
1487         return 0;
1488
1489 out_error:
1490         trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_private.a.agno,
1491                         error, _RET_IP_);
1492         return error;
1493 }
1494
1495 /*
1496  * Record a CoW allocation in the refcount btree.
1497  */
1498 STATIC int
1499 __xfs_refcount_cow_alloc(
1500         struct xfs_btree_cur    *rcur,
1501         xfs_agblock_t           agbno,
1502         xfs_extlen_t            aglen,
1503         struct xfs_defer_ops    *dfops)
1504 {
1505         trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_private.a.agno,
1506                         agbno, aglen);
1507
1508         /* Add refcount btree reservation */
1509         return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1510                         XFS_REFCOUNT_ADJUST_COW_ALLOC);
1511 }
1512
1513 /*
1514  * Remove a CoW allocation from the refcount btree.
1515  */
1516 STATIC int
1517 __xfs_refcount_cow_free(
1518         struct xfs_btree_cur    *rcur,
1519         xfs_agblock_t           agbno,
1520         xfs_extlen_t            aglen,
1521         struct xfs_defer_ops    *dfops)
1522 {
1523         trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_private.a.agno,
1524                         agbno, aglen);
1525
1526         /* Remove refcount btree reservation */
1527         return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1528                         XFS_REFCOUNT_ADJUST_COW_FREE);
1529 }
1530
1531 /* Record a CoW staging extent in the refcount btree. */
1532 int
1533 xfs_refcount_alloc_cow_extent(
1534         struct xfs_mount                *mp,
1535         struct xfs_defer_ops            *dfops,
1536         xfs_fsblock_t                   fsb,
1537         xfs_extlen_t                    len)
1538 {
1539         int                             error;
1540
1541         if (!xfs_sb_version_hasreflink(&mp->m_sb))
1542                 return 0;
1543
1544         error = __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_ALLOC_COW,
1545                         fsb, len);
1546         if (error)
1547                 return error;
1548
1549         /* Add rmap entry */
1550         return xfs_rmap_alloc_extent(mp, dfops, XFS_FSB_TO_AGNO(mp, fsb),
1551                         XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1552 }
1553
1554 /* Forget a CoW staging event in the refcount btree. */
1555 int
1556 xfs_refcount_free_cow_extent(
1557         struct xfs_mount                *mp,
1558         struct xfs_defer_ops            *dfops,
1559         xfs_fsblock_t                   fsb,
1560         xfs_extlen_t                    len)
1561 {
1562         int                             error;
1563
1564         if (!xfs_sb_version_hasreflink(&mp->m_sb))
1565                 return 0;
1566
1567         /* Remove rmap entry */
1568         error = xfs_rmap_free_extent(mp, dfops, XFS_FSB_TO_AGNO(mp, fsb),
1569                         XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1570         if (error)
1571                 return error;
1572
1573         return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_FREE_COW,
1574                         fsb, len);
1575 }
1576
1577 struct xfs_refcount_recovery {
1578         struct list_head                rr_list;
1579         struct xfs_refcount_irec        rr_rrec;
1580 };
1581
1582 /* Stuff an extent on the recovery list. */
1583 STATIC int
1584 xfs_refcount_recover_extent(
1585         struct xfs_btree_cur            *cur,
1586         union xfs_btree_rec             *rec,
1587         void                            *priv)
1588 {
1589         struct list_head                *debris = priv;
1590         struct xfs_refcount_recovery    *rr;
1591
1592         if (be32_to_cpu(rec->refc.rc_refcount) != 1)
1593                 return -EFSCORRUPTED;
1594
1595         rr = kmem_alloc(sizeof(struct xfs_refcount_recovery), KM_SLEEP);
1596         xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1597         list_add_tail(&rr->rr_list, debris);
1598
1599         return 0;
1600 }
1601
1602 /* Find and remove leftover CoW reservations. */
1603 int
1604 xfs_refcount_recover_cow_leftovers(
1605         struct xfs_mount                *mp,
1606         xfs_agnumber_t                  agno)
1607 {
1608         struct xfs_trans                *tp;
1609         struct xfs_btree_cur            *cur;
1610         struct xfs_buf                  *agbp;
1611         struct xfs_refcount_recovery    *rr, *n;
1612         struct list_head                debris;
1613         union xfs_btree_irec            low;
1614         union xfs_btree_irec            high;
1615         struct xfs_defer_ops            dfops;
1616         xfs_fsblock_t                   fsb;
1617         xfs_agblock_t                   agbno;
1618         int                             error;
1619
1620         if (mp->m_sb.sb_agblocks >= XFS_REFC_COW_START)
1621                 return -EOPNOTSUPP;
1622
1623         INIT_LIST_HEAD(&debris);
1624
1625         /*
1626          * In this first part, we use an empty transaction to gather up
1627          * all the leftover CoW extents so that we can subsequently
1628          * delete them.  The empty transaction is used to avoid
1629          * a buffer lock deadlock if there happens to be a loop in the
1630          * refcountbt because we're allowed to re-grab a buffer that is
1631          * already attached to our transaction.  When we're done
1632          * recording the CoW debris we cancel the (empty) transaction
1633          * and everything goes away cleanly.
1634          */
1635         error = xfs_trans_alloc_empty(mp, &tp);
1636         if (error)
1637                 return error;
1638
1639         error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
1640         if (error)
1641                 goto out_trans;
1642         if (!agbp) {
1643                 error = -ENOMEM;
1644                 goto out_trans;
1645         }
1646         cur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno, NULL);
1647
1648         /* Find all the leftover CoW staging extents. */
1649         memset(&low, 0, sizeof(low));
1650         memset(&high, 0, sizeof(high));
1651         low.rc.rc_startblock = XFS_REFC_COW_START;
1652         high.rc.rc_startblock = -1U;
1653         error = xfs_btree_query_range(cur, &low, &high,
1654                         xfs_refcount_recover_extent, &debris);
1655         if (error)
1656                 goto out_cursor;
1657         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1658         xfs_trans_brelse(tp, agbp);
1659         xfs_trans_cancel(tp);
1660
1661         /* Now iterate the list to free the leftovers */
1662         list_for_each_entry_safe(rr, n, &debris, rr_list) {
1663                 /* Set up transaction. */
1664                 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1665                 if (error)
1666                         goto out_free;
1667
1668                 trace_xfs_refcount_recover_extent(mp, agno, &rr->rr_rrec);
1669
1670                 /* Free the orphan record */
1671                 xfs_defer_init(&dfops, &fsb);
1672                 agbno = rr->rr_rrec.rc_startblock - XFS_REFC_COW_START;
1673                 fsb = XFS_AGB_TO_FSB(mp, agno, agbno);
1674                 error = xfs_refcount_free_cow_extent(mp, &dfops, fsb,
1675                                 rr->rr_rrec.rc_blockcount);
1676                 if (error)
1677                         goto out_defer;
1678
1679                 /* Free the block. */
1680                 xfs_bmap_add_free(mp, &dfops, fsb,
1681                                 rr->rr_rrec.rc_blockcount, NULL);
1682
1683                 error = xfs_defer_finish(&tp, &dfops);
1684                 if (error)
1685                         goto out_defer;
1686
1687                 error = xfs_trans_commit(tp);
1688                 if (error)
1689                         goto out_free;
1690
1691                 list_del(&rr->rr_list);
1692                 kmem_free(rr);
1693         }
1694
1695         return error;
1696 out_defer:
1697         xfs_defer_cancel(&dfops);
1698 out_trans:
1699         xfs_trans_cancel(tp);
1700 out_free:
1701         /* Free the leftover list */
1702         list_for_each_entry_safe(rr, n, &debris, rr_list) {
1703                 list_del(&rr->rr_list);
1704                 kmem_free(rr);
1705         }
1706         return error;
1707
1708 out_cursor:
1709         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1710         xfs_trans_brelse(tp, agbp);
1711         goto out_trans;
1712 }
1713
1714 /* Is there a record covering a given extent? */
1715 int
1716 xfs_refcount_has_record(
1717         struct xfs_btree_cur    *cur,
1718         xfs_agblock_t           bno,
1719         xfs_extlen_t            len,
1720         bool                    *exists)
1721 {
1722         union xfs_btree_irec    low;
1723         union xfs_btree_irec    high;
1724
1725         memset(&low, 0, sizeof(low));
1726         low.rc.rc_startblock = bno;
1727         memset(&high, 0xFF, sizeof(high));
1728         high.rc.rc_startblock = bno + len - 1;
1729
1730         return xfs_btree_has_record(cur, &low, &high, exists);
1731 }