kernel/fs/xfs/xfs_extent_busy.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2010 David Chinner.
   4  * Copyright (c) 2011 Christoph Hellwig.
   5  * All Rights Reserved.
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License as
   9  * published by the Free Software Foundation.
  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_format.h"
  23 #include "xfs_log_format.h"
  24 #include "xfs_shared.h"
  25 #include "xfs_trans_resv.h"
  26 #include "xfs_sb.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_alloc.h"
  29 #include "xfs_extent_busy.h"
  30 #include "xfs_trace.h"
  31 #include "xfs_trans.h"
  32 #include "xfs_log.h"
  33
  34 void
  35 xfs_extent_busy_insert(
  36         struct xfs_trans        *tp,
  37         xfs_agnumber_t          agno,
  38         xfs_agblock_t           bno,
  39         xfs_extlen_t            len,
  40         unsigned int            flags)
  41 {
  42         struct xfs_extent_busy  *new;
  43         struct xfs_extent_busy  *busyp;
  44         struct xfs_perag        *pag;
  45         struct rb_node          **rbp;
  46         struct rb_node          *parent = NULL;
  47
  48         new = kmem_zalloc(sizeof(struct xfs_extent_busy), KM_MAYFAIL);
  49         if (!new) {
  50                 /*
  51                  * No Memory!  Since it is now not possible to track the free
  52                  * block, make this a synchronous transaction to insure that
  53                  * the block is not reused before this transaction commits.
  54                  */
  55                 trace_xfs_extent_busy_enomem(tp->t_mountp, agno, bno, len);
  56                 xfs_trans_set_sync(tp);
  57                 return;
  58         }
  59
  60         new->agno = agno;
  61         new->bno = bno;
  62         new->length = len;
  63         INIT_LIST_HEAD(&new->list);
  64         new->flags = flags;
  65
  66         /* trace before insert to be able to see failed inserts */
  67         trace_xfs_extent_busy(tp->t_mountp, agno, bno, len);
  68
  69         pag = xfs_perag_get(tp->t_mountp, new->agno);
  70         spin_lock(&pag->pagb_lock);
  71         rbp = &pag->pagb_tree.rb_node;
  72         while (*rbp) {
  73                 parent = *rbp;
  74                 busyp = rb_entry(parent, struct xfs_extent_busy, rb_node);
  75
  76                 if (new->bno < busyp->bno) {
  77                         rbp = &(*rbp)->rb_left;
  78                         ASSERT(new->bno + new->length <= busyp->bno);
  79                 } else if (new->bno > busyp->bno) {
  80                         rbp = &(*rbp)->rb_right;
  81                         ASSERT(bno >= busyp->bno + busyp->length);
  82                 } else {
  83                         ASSERT(0);
  84                 }
  85         }
  86
  87         rb_link_node(&new->rb_node, parent, rbp);
  88         rb_insert_color(&new->rb_node, &pag->pagb_tree);
  89
  90         list_add(&new->list, &tp->t_busy);
  91         spin_unlock(&pag->pagb_lock);
  92         xfs_perag_put(pag);
  93 }
  94
  95 /*
  96  * Search for a busy extent within the range of the extent we are about to
  97  * allocate.  You need to be holding the busy extent tree lock when calling
  98  * xfs_extent_busy_search(). This function returns 0 for no overlapping busy
  99  * extent, -1 for an overlapping but not exact busy extent, and 1 for an exact
 100  * match. This is done so that a non-zero return indicates an overlap that
 101  * will require a synchronous transaction, but it can still be
 102  * used to distinguish between a partial or exact match.
 103  */
 104 int
 105 xfs_extent_busy_search(
 106         struct xfs_mount        *mp,
 107         xfs_agnumber_t          agno,
 108         xfs_agblock_t           bno,
 109         xfs_extlen_t            len)
 110 {
 111         struct xfs_perag        *pag;
 112         struct rb_node          *rbp;
 113         struct xfs_extent_busy  *busyp;
 114         int                     match = 0;
 115
 116         pag = xfs_perag_get(mp, agno);
 117         spin_lock(&pag->pagb_lock);
 118
 119         rbp = pag->pagb_tree.rb_node;
 120
 121         /* find closest start bno overlap */
 122         while (rbp) {
 123                 busyp = rb_entry(rbp, struct xfs_extent_busy, rb_node);
 124                 if (bno < busyp->bno) {
 125                         /* may overlap, but exact start block is lower */
 126                         if (bno + len > busyp->bno)
 127                                 match = -1;
 128                         rbp = rbp->rb_left;
 129                 } else if (bno > busyp->bno) {
 130                         /* may overlap, but exact start block is higher */
 131                         if (bno < busyp->bno + busyp->length)
 132                                 match = -1;
 133                         rbp = rbp->rb_right;
 134                 } else {
 135                         /* bno matches busyp, length determines exact match */
 136                         match = (busyp->length == len) ? 1 : -1;
 137                         break;
 138                 }
 139         }
 140         spin_unlock(&pag->pagb_lock);
 141         xfs_perag_put(pag);
 142         return match;
 143 }
 144
 145 /*
 146  * The found free extent [fbno, fend] overlaps part or all of the given busy
 147  * extent.  If the overlap covers the beginning, the end, or all of the busy
 148  * extent, the overlapping portion can be made unbusy and used for the
 149  * allocation.  We can't split a busy extent because we can't modify a
 150  * transaction/CIL context busy list, but we can update an entry's block
 151  * number or length.
 152  *
 153  * Returns true if the extent can safely be reused, or false if the search
 154  * needs to be restarted.
 155  */
 156 STATIC bool
 157 xfs_extent_busy_update_extent(
 158         struct xfs_mount        *mp,
 159         struct xfs_perag        *pag,
 160         struct xfs_extent_busy  *busyp,
 161         xfs_agblock_t           fbno,
 162         xfs_extlen_t            flen,
 163         bool                    userdata) __releases(&pag->pagb_lock)
 164                                           __acquires(&pag->pagb_lock)
 165 {
 166         xfs_agblock_t           fend = fbno + flen;
 167         xfs_agblock_t           bbno = busyp->bno;
 168         xfs_agblock_t           bend = bbno + busyp->length;
 169
 170         /*
 171          * This extent is currently being discarded.  Give the thread
 172          * performing the discard a chance to mark the extent unbusy
 173          * and retry.
 174          */
 175         if (busyp->flags & XFS_EXTENT_BUSY_DISCARDED) {
 176                 spin_unlock(&pag->pagb_lock);
 177                 delay(1);
 178                 spin_lock(&pag->pagb_lock);
 179                 return false;
 180         }
 181
 182         /*
 183          * If there is a busy extent overlapping a user allocation, we have
 184          * no choice but to force the log and retry the search.
 185          *
 186          * Fortunately this does not happen during normal operation, but
 187          * only if the filesystem is very low on space and has to dip into
 188          * the AGFL for normal allocations.
 189          */
 190         if (userdata)
 191                 goto out_force_log;
 192
 193         if (bbno < fbno && bend > fend) {
 194                 /*
 195                  * Case 1:
 196                  *    bbno           bend
 197                  *    +BBBBBBBBBBBBBBBBB+
 198                  *        +---------+
 199                  *        fbno   fend
 200                  */
 201
 202                 /*
 203                  * We would have to split the busy extent to be able to track
 204                  * it correct, which we cannot do because we would have to
 205                  * modify the list of busy extents attached to the transaction
 206                  * or CIL context, which is immutable.
 207                  *
 208                  * Force out the log to clear the busy extent and retry the
 209                  * search.
 210                  */
 211                 goto out_force_log;
 212         } else if (bbno >= fbno && bend <= fend) {
 213                 /*
 214                  * Case 2:
 215                  *    bbno           bend
 216                  *    +BBBBBBBBBBBBBBBBB+
 217                  *    +-----------------+
 218                  *    fbno           fend
 219                  *
 220                  * Case 3:
 221                  *    bbno           bend
 222                  *    +BBBBBBBBBBBBBBBBB+
 223                  *    +--------------------------+
 224                  *    fbno                    fend
 225                  *
 226                  * Case 4:
 227                  *             bbno           bend
 228                  *             +BBBBBBBBBBBBBBBBB+
 229                  *    +--------------------------+
 230                  *    fbno                    fend
 231                  *
 232                  * Case 5:
 233                  *             bbno           bend
 234                  *             +BBBBBBBBBBBBBBBBB+
 235                  *    +-----------------------------------+
 236                  *    fbno                             fend
 237                  *
 238                  */
 239
 240                 /*
 241                  * The busy extent is fully covered by the extent we are
 242                  * allocating, and can simply be removed from the rbtree.
 243                  * However we cannot remove it from the immutable list
 244                  * tracking busy extents in the transaction or CIL context,
 245                  * so set the length to zero to mark it invalid.
 246                  *
 247                  * We also need to restart the busy extent search from the
 248                  * tree root, because erasing the node can rearrange the
 249                  * tree topology.
 250                  */
 251                 rb_erase(&busyp->rb_node, &pag->pagb_tree);
 252                 busyp->length = 0;
 253                 return false;
 254         } else if (fend < bend) {
 255                 /*
 256                  * Case 6:
 257                  *              bbno           bend
 258                  *             +BBBBBBBBBBBBBBBBB+
 259                  *             +---------+
 260                  *             fbno   fend
 261                  *
 262                  * Case 7:
 263                  *             bbno           bend
 264                  *             +BBBBBBBBBBBBBBBBB+
 265                  *    +------------------+
 266                  *    fbno            fend
 267                  *
 268                  */
 269                 busyp->bno = fend;
 270         } else if (bbno < fbno) {
 271                 /*
 272                  * Case 8:
 273                  *    bbno           bend
 274                  *    +BBBBBBBBBBBBBBBBB+
 275                  *        +-------------+
 276                  *        fbno       fend
 277                  *
 278                  * Case 9:
 279                  *    bbno           bend
 280                  *    +BBBBBBBBBBBBBBBBB+
 281                  *        +----------------------+
 282                  *        fbno                fend
 283                  */
 284                 busyp->length = fbno - busyp->bno;
 285         } else {
 286                 ASSERT(0);
 287         }
 288
 289         trace_xfs_extent_busy_reuse(mp, pag->pag_agno, fbno, flen);
 290         return true;
 291
 292 out_force_log:
 293         spin_unlock(&pag->pagb_lock);
 294         xfs_log_force(mp, XFS_LOG_SYNC);
 295         trace_xfs_extent_busy_force(mp, pag->pag_agno, fbno, flen);
 296         spin_lock(&pag->pagb_lock);
 297         return false;
 298 }
 299
 300
 301 /*
 302  * For a given extent [fbno, flen], make sure we can reuse it safely.
 303  */
 304 void
 305 xfs_extent_busy_reuse(
 306         struct xfs_mount        *mp,
 307         xfs_agnumber_t          agno,
 308         xfs_agblock_t           fbno,
 309         xfs_extlen_t            flen,
 310         bool                    userdata)
 311 {
 312         struct xfs_perag        *pag;
 313         struct rb_node          *rbp;
 314
 315         ASSERT(flen > 0);
 316
 317         pag = xfs_perag_get(mp, agno);
 318         spin_lock(&pag->pagb_lock);
 319 restart:
 320         rbp = pag->pagb_tree.rb_node;
 321         while (rbp) {
 322                 struct xfs_extent_busy *busyp =
 323                         rb_entry(rbp, struct xfs_extent_busy, rb_node);
 324                 xfs_agblock_t   bbno = busyp->bno;
 325                 xfs_agblock_t   bend = bbno + busyp->length;
 326
 327                 if (fbno + flen <= bbno) {
 328                         rbp = rbp->rb_left;
 329                         continue;
 330                 } else if (fbno >= bend) {
 331                         rbp = rbp->rb_right;
 332                         continue;
 333                 }
 334
 335                 if (!xfs_extent_busy_update_extent(mp, pag, busyp, fbno, flen,
 336                                                   userdata))
 337                         goto restart;
 338         }
 339         spin_unlock(&pag->pagb_lock);
 340         xfs_perag_put(pag);
 341 }
 342
 343 /*
 344  * For a given extent [fbno, flen], search the busy extent list to find a
 345  * subset of the extent that is not busy.  If *rlen is smaller than
 346  * args->minlen no suitable extent could be found, and the higher level
 347  * code needs to force out the log and retry the allocation.
 348  */
 349 void
 350 xfs_extent_busy_trim(
 351         struct xfs_alloc_arg    *args,
 352         xfs_agblock_t           bno,
 353         xfs_extlen_t            len,
 354         xfs_agblock_t           *rbno,
 355         xfs_extlen_t            *rlen)
 356 {
 357         xfs_agblock_t           fbno;
 358         xfs_extlen_t            flen;
 359         struct rb_node          *rbp;
 360
 361         ASSERT(len > 0);
 362
 363         spin_lock(&args->pag->pagb_lock);
 364 restart:
 365         fbno = bno;
 366         flen = len;
 367         rbp = args->pag->pagb_tree.rb_node;
 368         while (rbp && flen >= args->minlen) {
 369                 struct xfs_extent_busy *busyp =
 370                         rb_entry(rbp, struct xfs_extent_busy, rb_node);
 371                 xfs_agblock_t   fend = fbno + flen;
 372                 xfs_agblock_t   bbno = busyp->bno;
 373                 xfs_agblock_t   bend = bbno + busyp->length;
 374
 375                 if (fend <= bbno) {
 376                         rbp = rbp->rb_left;
 377                         continue;
 378                 } else if (fbno >= bend) {
 379                         rbp = rbp->rb_right;
 380                         continue;
 381                 }
 382
 383                 /*
 384                  * If this is a metadata allocation, try to reuse the busy
 385                  * extent instead of trimming the allocation.
 386                  */
 387                 if (!args->userdata &&
 388                     !(busyp->flags & XFS_EXTENT_BUSY_DISCARDED)) {
 389                         if (!xfs_extent_busy_update_extent(args->mp, args->pag,
 390                                                           busyp, fbno, flen,
 391                                                           false))
 392                                 goto restart;
 393                         continue;
 394                 }
 395
 396                 if (bbno <= fbno) {
 397                         /* start overlap */
 398
 399                         /*
 400                          * Case 1:
 401                          *    bbno           bend
 402                          *    +BBBBBBBBBBBBBBBBB+
 403                          *        +---------+
 404                          *        fbno   fend
 405                          *
 406                          * Case 2:
 407                          *    bbno           bend
 408                          *    +BBBBBBBBBBBBBBBBB+
 409                          *    +-------------+
 410                          *    fbno       fend
 411                          *
 412                          * Case 3:
 413                          *    bbno           bend
 414                          *    +BBBBBBBBBBBBBBBBB+
 415                          *        +-------------+
 416                          *        fbno       fend
 417                          *
 418                          * Case 4:
 419                          *    bbno           bend
 420                          *    +BBBBBBBBBBBBBBBBB+
 421                          *    +-----------------+
 422                          *    fbno           fend
 423                          *
 424                          * No unbusy region in extent, return failure.
 425                          */
 426                         if (fend <= bend)
 427                                 goto fail;
 428
 429                         /*
 430                          * Case 5:
 431                          *    bbno           bend
 432                          *    +BBBBBBBBBBBBBBBBB+
 433                          *        +----------------------+
 434                          *        fbno                fend
 435                          *
 436                          * Case 6:
 437                          *    bbno           bend
 438                          *    +BBBBBBBBBBBBBBBBB+
 439                          *    +--------------------------+
 440                          *    fbno                    fend
 441                          *
 442                          * Needs to be trimmed to:
 443                          *                       +-------+
 444                          *                       fbno fend
 445                          */
 446                         fbno = bend;
 447                 } else if (bend >= fend) {
 448                         /* end overlap */
 449
 450                         /*
 451                          * Case 7:
 452                          *             bbno           bend
 453                          *             +BBBBBBBBBBBBBBBBB+
 454                          *    +------------------+
 455                          *    fbno            fend
 456                          *
 457                          * Case 8:
 458                          *             bbno           bend
 459                          *             +BBBBBBBBBBBBBBBBB+
 460                          *    +--------------------------+
 461                          *    fbno                    fend
 462                          *
 463                          * Needs to be trimmed to:
 464                          *    +-------+
 465                          *    fbno fend
 466                          */
 467                         fend = bbno;
 468                 } else {
 469                         /* middle overlap */
 470
 471                         /*
 472                          * Case 9:
 473                          *             bbno           bend
 474                          *             +BBBBBBBBBBBBBBBBB+
 475                          *    +-----------------------------------+
 476                          *    fbno                             fend
 477                          *
 478                          * Can be trimmed to:
 479                          *    +-------+        OR         +-------+
 480                          *    fbno fend                   fbno fend
 481                          *
 482                          * Backward allocation leads to significant
 483                          * fragmentation of directories, which degrades
 484                          * directory performance, therefore we always want to
 485                          * choose the option that produces forward allocation
 486                          * patterns.
 487                          * Preferring the lower bno extent will make the next
 488                          * request use "fend" as the start of the next
 489                          * allocation;  if the segment is no longer busy at
 490                          * that point, we'll get a contiguous allocation, but
 491                          * even if it is still busy, we will get a forward
 492                          * allocation.
 493                          * We try to avoid choosing the segment at "bend",
 494                          * because that can lead to the next allocation
 495                          * taking the segment at "fbno", which would be a
 496                          * backward allocation.  We only use the segment at
 497                          * "fbno" if it is much larger than the current
 498                          * requested size, because in that case there's a
 499                          * good chance subsequent allocations will be
 500                          * contiguous.
 501                          */
 502                         if (bbno - fbno >= args->maxlen) {
 503                                 /* left candidate fits perfect */
 504                                 fend = bbno;
 505                         } else if (fend - bend >= args->maxlen * 4) {
 506                                 /* right candidate has enough free space */
 507                                 fbno = bend;
 508                         } else if (bbno - fbno >= args->minlen) {
 509                                 /* left candidate fits minimum requirement */
 510                                 fend = bbno;
 511                         } else {
 512                                 goto fail;
 513                         }
 514                 }
 515
 516                 flen = fend - fbno;
 517         }
 518         spin_unlock(&args->pag->pagb_lock);
 519
 520         if (fbno != bno || flen != len) {
 521                 trace_xfs_extent_busy_trim(args->mp, args->agno, bno, len,
 522                                           fbno, flen);
 523         }
 524         *rbno = fbno;
 525         *rlen = flen;
 526         return;
 527 fail:
 528         /*
 529          * Return a zero extent length as failure indications.  All callers
 530          * re-check if the trimmed extent satisfies the minlen requirement.
 531          */
 532         spin_unlock(&args->pag->pagb_lock);
 533         trace_xfs_extent_busy_trim(args->mp, args->agno, bno, len, fbno, 0);
 534         *rbno = fbno;
 535         *rlen = 0;
 536 }
 537
 538 STATIC void
 539 xfs_extent_busy_clear_one(
 540         struct xfs_mount        *mp,
 541         struct xfs_perag        *pag,
 542         struct xfs_extent_busy  *busyp)
 543 {
 544         if (busyp->length) {
 545                 trace_xfs_extent_busy_clear(mp, busyp->agno, busyp->bno,
 546                                                 busyp->length);
 547                 rb_erase(&busyp->rb_node, &pag->pagb_tree);
 548         }
 549
 550         list_del_init(&busyp->list);
 551         kmem_free(busyp);
 552 }
 553
 554 /*
 555  * Remove all extents on the passed in list from the busy extents tree.
 556  * If do_discard is set skip extents that need to be discarded, and mark
 557  * these as undergoing a discard operation instead.
 558  */
 559 void
 560 xfs_extent_busy_clear(
 561         struct xfs_mount        *mp,
 562         struct list_head        *list,
 563         bool                    do_discard)
 564 {
 565         struct xfs_extent_busy  *busyp, *n;
 566         struct xfs_perag        *pag = NULL;
 567         xfs_agnumber_t          agno = NULLAGNUMBER;
 568
 569         list_for_each_entry_safe(busyp, n, list, list) {
 570                 if (busyp->agno != agno) {
 571                         if (pag) {
 572                                 spin_unlock(&pag->pagb_lock);
 573                                 xfs_perag_put(pag);
 574                         }
 575                         pag = xfs_perag_get(mp, busyp->agno);
 576                         spin_lock(&pag->pagb_lock);
 577                         agno = busyp->agno;
 578                 }
 579
 580                 if (do_discard && busyp->length &&
 581                     !(busyp->flags & XFS_EXTENT_BUSY_SKIP_DISCARD))
 582                         busyp->flags = XFS_EXTENT_BUSY_DISCARDED;
 583                 else
 584                         xfs_extent_busy_clear_one(mp, pag, busyp);
 585         }
 586
 587         if (pag) {
 588                 spin_unlock(&pag->pagb_lock);
 589                 xfs_perag_put(pag);
 590         }
 591 }
 592
 593 /*
 594  * Callback for list_sort to sort busy extents by the AG they reside in.
 595  */
 596 int
 597 xfs_extent_busy_ag_cmp(
 598         void                    *priv,
 599         struct list_head        *a,
 600         struct list_head        *b)
 601 {
 602         return container_of(a, struct xfs_extent_busy, list)->agno -
 603                 container_of(b, struct xfs_extent_busy, list)->agno;
 604 }