fs/xfs/xfs_fsmap.c

   1 /*
   2  * Copyright (C) 2017 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_inode.h"
  30 #include "xfs_trans.h"
  31 #include "xfs_error.h"
  32 #include "xfs_btree.h"
  33 #include "xfs_rmap_btree.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_log.h"
  36 #include "xfs_rmap.h"
  37 #include "xfs_alloc.h"
  38 #include "xfs_bit.h"
  39 #include <linux/fsmap.h>
  40 #include "xfs_fsmap.h"
  41 #include "xfs_refcount.h"
  42 #include "xfs_refcount_btree.h"
  43 #include "xfs_alloc_btree.h"
  44 #include "xfs_rtalloc.h"
  45
  46 /* Convert an xfs_fsmap to an fsmap. */
  47 void
  48 xfs_fsmap_from_internal(
  49         struct fsmap            *dest,
  50         struct xfs_fsmap        *src)
  51 {
  52         dest->fmr_device = src->fmr_device;
  53         dest->fmr_flags = src->fmr_flags;
  54         dest->fmr_physical = BBTOB(src->fmr_physical);
  55         dest->fmr_owner = src->fmr_owner;
  56         dest->fmr_offset = BBTOB(src->fmr_offset);
  57         dest->fmr_length = BBTOB(src->fmr_length);
  58         dest->fmr_reserved[0] = 0;
  59         dest->fmr_reserved[1] = 0;
  60         dest->fmr_reserved[2] = 0;
  61 }
  62
  63 /* Convert an fsmap to an xfs_fsmap. */
  64 void
  65 xfs_fsmap_to_internal(
  66         struct xfs_fsmap        *dest,
  67         struct fsmap            *src)
  68 {
  69         dest->fmr_device = src->fmr_device;
  70         dest->fmr_flags = src->fmr_flags;
  71         dest->fmr_physical = BTOBBT(src->fmr_physical);
  72         dest->fmr_owner = src->fmr_owner;
  73         dest->fmr_offset = BTOBBT(src->fmr_offset);
  74         dest->fmr_length = BTOBBT(src->fmr_length);
  75 }
  76
  77 /* Convert an fsmap owner into an rmapbt owner. */
  78 static int
  79 xfs_fsmap_owner_to_rmap(
  80         struct xfs_rmap_irec    *dest,
  81         struct xfs_fsmap        *src)
  82 {
  83         if (!(src->fmr_flags & FMR_OF_SPECIAL_OWNER)) {
  84                 dest->rm_owner = src->fmr_owner;
  85                 return 0;
  86         }
  87
  88         switch (src->fmr_owner) {
  89         case 0:                 /* "lowest owner id possible" */
  90         case -1ULL:             /* "highest owner id possible" */
  91                 dest->rm_owner = 0;
  92                 break;
  93         case XFS_FMR_OWN_FREE:
  94                 dest->rm_owner = XFS_RMAP_OWN_NULL;
  95                 break;
  96         case XFS_FMR_OWN_UNKNOWN:
  97                 dest->rm_owner = XFS_RMAP_OWN_UNKNOWN;
  98                 break;
  99         case XFS_FMR_OWN_FS:
 100                 dest->rm_owner = XFS_RMAP_OWN_FS;
 101                 break;
 102         case XFS_FMR_OWN_LOG:
 103                 dest->rm_owner = XFS_RMAP_OWN_LOG;
 104                 break;
 105         case XFS_FMR_OWN_AG:
 106                 dest->rm_owner = XFS_RMAP_OWN_AG;
 107                 break;
 108         case XFS_FMR_OWN_INOBT:
 109                 dest->rm_owner = XFS_RMAP_OWN_INOBT;
 110                 break;
 111         case XFS_FMR_OWN_INODES:
 112                 dest->rm_owner = XFS_RMAP_OWN_INODES;
 113                 break;
 114         case XFS_FMR_OWN_REFC:
 115                 dest->rm_owner = XFS_RMAP_OWN_REFC;
 116                 break;
 117         case XFS_FMR_OWN_COW:
 118                 dest->rm_owner = XFS_RMAP_OWN_COW;
 119                 break;
 120         case XFS_FMR_OWN_DEFECTIVE:     /* not implemented */
 121                 /* fall through */
 122         default:
 123                 return -EINVAL;
 124         }
 125         return 0;
 126 }
 127
 128 /* Convert an rmapbt owner into an fsmap owner. */
 129 static int
 130 xfs_fsmap_owner_from_rmap(
 131         struct xfs_fsmap        *dest,
 132         struct xfs_rmap_irec    *src)
 133 {
 134         dest->fmr_flags = 0;
 135         if (!XFS_RMAP_NON_INODE_OWNER(src->rm_owner)) {
 136                 dest->fmr_owner = src->rm_owner;
 137                 return 0;
 138         }
 139         dest->fmr_flags |= FMR_OF_SPECIAL_OWNER;
 140
 141         switch (src->rm_owner) {
 142         case XFS_RMAP_OWN_FS:
 143                 dest->fmr_owner = XFS_FMR_OWN_FS;
 144                 break;
 145         case XFS_RMAP_OWN_LOG:
 146                 dest->fmr_owner = XFS_FMR_OWN_LOG;
 147                 break;
 148         case XFS_RMAP_OWN_AG:
 149                 dest->fmr_owner = XFS_FMR_OWN_AG;
 150                 break;
 151         case XFS_RMAP_OWN_INOBT:
 152                 dest->fmr_owner = XFS_FMR_OWN_INOBT;
 153                 break;
 154         case XFS_RMAP_OWN_INODES:
 155                 dest->fmr_owner = XFS_FMR_OWN_INODES;
 156                 break;
 157         case XFS_RMAP_OWN_REFC:
 158                 dest->fmr_owner = XFS_FMR_OWN_REFC;
 159                 break;
 160         case XFS_RMAP_OWN_COW:
 161                 dest->fmr_owner = XFS_FMR_OWN_COW;
 162                 break;
 163         case XFS_RMAP_OWN_NULL: /* "free" */
 164                 dest->fmr_owner = XFS_FMR_OWN_FREE;
 165                 break;
 166         default:
 167                 return -EFSCORRUPTED;
 168         }
 169         return 0;
 170 }
 171
 172 /* getfsmap query state */
 173 struct xfs_getfsmap_info {
 174         struct xfs_fsmap_head   *head;
 175         xfs_fsmap_format_t      formatter;      /* formatting fn */
 176         void                    *format_arg;    /* format buffer */
 177         struct xfs_buf          *agf_bp;        /* AGF, for refcount queries */
 178         xfs_daddr_t             next_daddr;     /* next daddr we expect */
 179         u64                     missing_owner;  /* owner of holes */
 180         u32                     dev;            /* device id */
 181         xfs_agnumber_t          agno;           /* AG number, if applicable */
 182         struct xfs_rmap_irec    low;            /* low rmap key */
 183         struct xfs_rmap_irec    high;           /* high rmap key */
 184         bool                    last;           /* last extent? */
 185 };
 186
 187 /* Associate a device with a getfsmap handler. */
 188 struct xfs_getfsmap_dev {
 189         u32                     dev;
 190         int                     (*fn)(struct xfs_trans *tp,
 191                                       struct xfs_fsmap *keys,
 192                                       struct xfs_getfsmap_info *info);
 193 };
 194
 195 /* Compare two getfsmap device handlers. */
 196 static int
 197 xfs_getfsmap_dev_compare(
 198         const void                      *p1,
 199         const void                      *p2)
 200 {
 201         const struct xfs_getfsmap_dev   *d1 = p1;
 202         const struct xfs_getfsmap_dev   *d2 = p2;
 203
 204         return d1->dev - d2->dev;
 205 }
 206
 207 /* Decide if this mapping is shared. */
 208 STATIC int
 209 xfs_getfsmap_is_shared(
 210         struct xfs_trans                *tp,
 211         struct xfs_getfsmap_info        *info,
 212         struct xfs_rmap_irec            *rec,
 213         bool                            *stat)
 214 {
 215         struct xfs_mount                *mp = tp->t_mountp;
 216         struct xfs_btree_cur            *cur;
 217         xfs_agblock_t                   fbno;
 218         xfs_extlen_t                    flen;
 219         int                             error;
 220
 221         *stat = false;
 222         if (!xfs_sb_version_hasreflink(&mp->m_sb))
 223                 return 0;
 224         /* rt files will have agno set to NULLAGNUMBER */
 225         if (info->agno == NULLAGNUMBER)
 226                 return 0;
 227
 228         /* Are there any shared blocks here? */
 229         flen = 0;
 230         cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp,
 231                         info->agno, NULL);
 232
 233         error = xfs_refcount_find_shared(cur, rec->rm_startblock,
 234                         rec->rm_blockcount, &fbno, &flen, false);
 235
 236         xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 237         if (error)
 238                 return error;
 239
 240         *stat = flen > 0;
 241         return 0;
 242 }
 243
 244 /*
 245  * Format a reverse mapping for getfsmap, having translated rm_startblock
 246  * into the appropriate daddr units.
 247  */
 248 STATIC int
 249 xfs_getfsmap_helper(
 250         struct xfs_trans                *tp,
 251         struct xfs_getfsmap_info        *info,
 252         struct xfs_rmap_irec            *rec,
 253         xfs_daddr_t                     rec_daddr)
 254 {
 255         struct xfs_fsmap                fmr;
 256         struct xfs_mount                *mp = tp->t_mountp;
 257         bool                            shared;
 258         int                             error;
 259
 260         if (fatal_signal_pending(current))
 261                 return -EINTR;
 262
 263         /*
 264          * Filter out records that start before our startpoint, if the
 265          * caller requested that.
 266          */
 267         if (xfs_rmap_compare(rec, &info->low) < 0) {
 268                 rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
 269                 if (info->next_daddr < rec_daddr)
 270                         info->next_daddr = rec_daddr;
 271                 return XFS_BTREE_QUERY_RANGE_CONTINUE;
 272         }
 273
 274         /* Are we just counting mappings? */
 275         if (info->head->fmh_count == 0) {
 276                 if (info->head->fmh_entries == UINT_MAX)
 277                         return -ECANCELED;
 278
 279                 if (rec_daddr > info->next_daddr)
 280                         info->head->fmh_entries++;
 281
 282                 if (info->last)
 283                         return XFS_BTREE_QUERY_RANGE_CONTINUE;
 284
 285                 info->head->fmh_entries++;
 286
 287                 rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
 288                 if (info->next_daddr < rec_daddr)
 289                         info->next_daddr = rec_daddr;
 290                 return XFS_BTREE_QUERY_RANGE_CONTINUE;
 291         }
 292
 293         /*
 294          * If the record starts past the last physical block we saw,
 295          * then we've found a gap.  Report the gap as being owned by
 296          * whatever the caller specified is the missing owner.
 297          */
 298         if (rec_daddr > info->next_daddr) {
 299                 if (info->head->fmh_entries >= info->head->fmh_count)
 300                         return XFS_BTREE_QUERY_RANGE_ABORT;
 301
 302                 fmr.fmr_device = info->dev;
 303                 fmr.fmr_physical = info->next_daddr;
 304                 fmr.fmr_owner = info->missing_owner;
 305                 fmr.fmr_offset = 0;
 306                 fmr.fmr_length = rec_daddr - info->next_daddr;
 307                 fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
 308                 error = info->formatter(&fmr, info->format_arg);
 309                 if (error)
 310                         return error;
 311                 info->head->fmh_entries++;
 312         }
 313
 314         if (info->last)
 315                 goto out;
 316
 317         /* Fill out the extent we found */
 318         if (info->head->fmh_entries >= info->head->fmh_count)
 319                 return XFS_BTREE_QUERY_RANGE_ABORT;
 320
 321         trace_xfs_fsmap_mapping(mp, info->dev, info->agno, rec);
 322
 323         fmr.fmr_device = info->dev;
 324         fmr.fmr_physical = rec_daddr;
 325         error = xfs_fsmap_owner_from_rmap(&fmr, rec);
 326         if (error)
 327                 return error;
 328         fmr.fmr_offset = XFS_FSB_TO_BB(mp, rec->rm_offset);
 329         fmr.fmr_length = XFS_FSB_TO_BB(mp, rec->rm_blockcount);
 330         if (rec->rm_flags & XFS_RMAP_UNWRITTEN)
 331                 fmr.fmr_flags |= FMR_OF_PREALLOC;
 332         if (rec->rm_flags & XFS_RMAP_ATTR_FORK)
 333                 fmr.fmr_flags |= FMR_OF_ATTR_FORK;
 334         if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
 335                 fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
 336         if (fmr.fmr_flags == 0) {
 337                 error = xfs_getfsmap_is_shared(tp, info, rec, &shared);
 338                 if (error)
 339                         return error;
 340                 if (shared)
 341                         fmr.fmr_flags |= FMR_OF_SHARED;
 342         }
 343         error = info->formatter(&fmr, info->format_arg);
 344         if (error)
 345                 return error;
 346         info->head->fmh_entries++;
 347
 348 out:
 349         rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
 350         if (info->next_daddr < rec_daddr)
 351                 info->next_daddr = rec_daddr;
 352         return XFS_BTREE_QUERY_RANGE_CONTINUE;
 353 }
 354
 355 /* Transform a rmapbt irec into a fsmap */
 356 STATIC int
 357 xfs_getfsmap_datadev_helper(
 358         struct xfs_btree_cur            *cur,
 359         struct xfs_rmap_irec            *rec,
 360         void                            *priv)
 361 {
 362         struct xfs_mount                *mp = cur->bc_mp;
 363         struct xfs_getfsmap_info        *info = priv;
 364         xfs_fsblock_t                   fsb;
 365         xfs_daddr_t                     rec_daddr;
 366
 367         fsb = XFS_AGB_TO_FSB(mp, cur->bc_private.a.agno, rec->rm_startblock);
 368         rec_daddr = XFS_FSB_TO_DADDR(mp, fsb);
 369
 370         return xfs_getfsmap_helper(cur->bc_tp, info, rec, rec_daddr);
 371 }
 372
 373 /* Transform a bnobt irec into a fsmap */
 374 STATIC int
 375 xfs_getfsmap_datadev_bnobt_helper(
 376         struct xfs_btree_cur            *cur,
 377         struct xfs_alloc_rec_incore     *rec,
 378         void                            *priv)
 379 {
 380         struct xfs_mount                *mp = cur->bc_mp;
 381         struct xfs_getfsmap_info        *info = priv;
 382         struct xfs_rmap_irec            irec;
 383         xfs_daddr_t                     rec_daddr;
 384
 385         rec_daddr = XFS_AGB_TO_DADDR(mp, cur->bc_private.a.agno,
 386                         rec->ar_startblock);
 387
 388         irec.rm_startblock = rec->ar_startblock;
 389         irec.rm_blockcount = rec->ar_blockcount;
 390         irec.rm_owner = XFS_RMAP_OWN_NULL;      /* "free" */
 391         irec.rm_offset = 0;
 392         irec.rm_flags = 0;
 393
 394         return xfs_getfsmap_helper(cur->bc_tp, info, &irec, rec_daddr);
 395 }
 396
 397 /* Set rmap flags based on the getfsmap flags */
 398 static void
 399 xfs_getfsmap_set_irec_flags(
 400         struct xfs_rmap_irec    *irec,
 401         struct xfs_fsmap        *fmr)
 402 {
 403         irec->rm_flags = 0;
 404         if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
 405                 irec->rm_flags |= XFS_RMAP_ATTR_FORK;
 406         if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
 407                 irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
 408         if (fmr->fmr_flags & FMR_OF_PREALLOC)
 409                 irec->rm_flags |= XFS_RMAP_UNWRITTEN;
 410 }
 411
 412 /* Execute a getfsmap query against the log device. */
 413 STATIC int
 414 xfs_getfsmap_logdev(
 415         struct xfs_trans                *tp,
 416         struct xfs_fsmap                *keys,
 417         struct xfs_getfsmap_info        *info)
 418 {
 419         struct xfs_mount                *mp = tp->t_mountp;
 420         struct xfs_rmap_irec            rmap;
 421         int                             error;
 422
 423         /* Set up search keys */
 424         info->low.rm_startblock = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
 425         info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
 426         error = xfs_fsmap_owner_to_rmap(&info->low, keys);
 427         if (error)
 428                 return error;
 429         info->low.rm_blockcount = 0;
 430         xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
 431
 432         error = xfs_fsmap_owner_to_rmap(&info->high, keys + 1);
 433         if (error)
 434                 return error;
 435         info->high.rm_startblock = -1U;
 436         info->high.rm_owner = ULLONG_MAX;
 437         info->high.rm_offset = ULLONG_MAX;
 438         info->high.rm_blockcount = 0;
 439         info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
 440         info->missing_owner = XFS_FMR_OWN_FREE;
 441
 442         trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
 443         trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
 444
 445         if (keys[0].fmr_physical > 0)
 446                 return 0;
 447
 448         /* Fabricate an rmap entry for the external log device. */
 449         rmap.rm_startblock = 0;
 450         rmap.rm_blockcount = mp->m_sb.sb_logblocks;
 451         rmap.rm_owner = XFS_RMAP_OWN_LOG;
 452         rmap.rm_offset = 0;
 453         rmap.rm_flags = 0;
 454
 455         return xfs_getfsmap_helper(tp, info, &rmap, 0);
 456 }
 457
 458 #ifdef CONFIG_XFS_RT
 459 /* Transform a rtbitmap "record" into a fsmap */
 460 STATIC int
 461 xfs_getfsmap_rtdev_rtbitmap_helper(
 462         struct xfs_trans                *tp,
 463         struct xfs_rtalloc_rec          *rec,
 464         void                            *priv)
 465 {
 466         struct xfs_mount                *mp = tp->t_mountp;
 467         struct xfs_getfsmap_info        *info = priv;
 468         struct xfs_rmap_irec            irec;
 469         xfs_daddr_t                     rec_daddr;
 470
 471         rec_daddr = XFS_FSB_TO_BB(mp, rec->ar_startblock);
 472
 473         irec.rm_startblock = rec->ar_startblock;
 474         irec.rm_blockcount = rec->ar_blockcount;
 475         irec.rm_owner = XFS_RMAP_OWN_NULL;      /* "free" */
 476         irec.rm_offset = 0;
 477         irec.rm_flags = 0;
 478
 479         return xfs_getfsmap_helper(tp, info, &irec, rec_daddr);
 480 }
 481
 482 /* Execute a getfsmap query against the realtime device. */
 483 STATIC int
 484 __xfs_getfsmap_rtdev(
 485         struct xfs_trans                *tp,
 486         struct xfs_fsmap                *keys,
 487         int                             (*query_fn)(struct xfs_trans *,
 488                                                     struct xfs_getfsmap_info *),
 489         struct xfs_getfsmap_info        *info)
 490 {
 491         struct xfs_mount                *mp = tp->t_mountp;
 492         xfs_fsblock_t                   start_fsb;
 493         xfs_fsblock_t                   end_fsb;
 494         xfs_daddr_t                     eofs;
 495         int                             error = 0;
 496
 497         eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
 498         if (keys[0].fmr_physical >= eofs)
 499                 return 0;
 500         if (keys[1].fmr_physical >= eofs)
 501                 keys[1].fmr_physical = eofs - 1;
 502         start_fsb = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
 503         end_fsb = XFS_BB_TO_FSB(mp, keys[1].fmr_physical);
 504
 505         /* Set up search keys */
 506         info->low.rm_startblock = start_fsb;
 507         error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
 508         if (error)
 509                 return error;
 510         info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
 511         info->low.rm_blockcount = 0;
 512         xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
 513
 514         info->high.rm_startblock = end_fsb;
 515         error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
 516         if (error)
 517                 return error;
 518         info->high.rm_offset = XFS_BB_TO_FSBT(mp, keys[1].fmr_offset);
 519         info->high.rm_blockcount = 0;
 520         xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
 521
 522         trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
 523         trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
 524
 525         return query_fn(tp, info);
 526 }
 527
 528 /* Actually query the realtime bitmap. */
 529 STATIC int
 530 xfs_getfsmap_rtdev_rtbitmap_query(
 531         struct xfs_trans                *tp,
 532         struct xfs_getfsmap_info        *info)
 533 {
 534         struct xfs_rtalloc_rec          alow;
 535         struct xfs_rtalloc_rec          ahigh;
 536         int                             error;
 537
 538         xfs_ilock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
 539
 540         alow.ar_startblock = info->low.rm_startblock;
 541         ahigh.ar_startblock = info->high.rm_startblock;
 542         error = xfs_rtalloc_query_range(tp, &alow, &ahigh,
 543                         xfs_getfsmap_rtdev_rtbitmap_helper, info);
 544         if (error)
 545                 goto err;
 546
 547         /* Report any gaps at the end of the rtbitmap */
 548         info->last = true;
 549         error = xfs_getfsmap_rtdev_rtbitmap_helper(tp, &ahigh, info);
 550         if (error)
 551                 goto err;
 552 err:
 553         xfs_iunlock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
 554         return error;
 555 }
 556
 557 /* Execute a getfsmap query against the realtime device rtbitmap. */
 558 STATIC int
 559 xfs_getfsmap_rtdev_rtbitmap(
 560         struct xfs_trans                *tp,
 561         struct xfs_fsmap                *keys,
 562         struct xfs_getfsmap_info        *info)
 563 {
 564         info->missing_owner = XFS_FMR_OWN_UNKNOWN;
 565         return __xfs_getfsmap_rtdev(tp, keys, xfs_getfsmap_rtdev_rtbitmap_query,
 566                         info);
 567 }
 568 #endif /* CONFIG_XFS_RT */
 569
 570 /* Execute a getfsmap query against the regular data device. */
 571 STATIC int
 572 __xfs_getfsmap_datadev(
 573         struct xfs_trans                *tp,
 574         struct xfs_fsmap                *keys,
 575         struct xfs_getfsmap_info        *info,
 576         int                             (*query_fn)(struct xfs_trans *,
 577                                                     struct xfs_getfsmap_info *,
 578                                                     struct xfs_btree_cur **,
 579                                                     void *),
 580         void                            *priv)
 581 {
 582         struct xfs_mount                *mp = tp->t_mountp;
 583         struct xfs_btree_cur            *bt_cur = NULL;
 584         xfs_fsblock_t                   start_fsb;
 585         xfs_fsblock_t                   end_fsb;
 586         xfs_agnumber_t                  start_ag;
 587         xfs_agnumber_t                  end_ag;
 588         xfs_daddr_t                     eofs;
 589         int                             error = 0;
 590
 591         eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 592         if (keys[0].fmr_physical >= eofs)
 593                 return 0;
 594         if (keys[1].fmr_physical >= eofs)
 595                 keys[1].fmr_physical = eofs - 1;
 596         start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
 597         end_fsb = XFS_DADDR_TO_FSB(mp, keys[1].fmr_physical);
 598
 599         /*
 600          * Convert the fsmap low/high keys to AG based keys.  Initialize
 601          * low to the fsmap low key and max out the high key to the end
 602          * of the AG.
 603          */
 604         info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
 605         info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
 606         error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
 607         if (error)
 608                 return error;
 609         info->low.rm_blockcount = 0;
 610         xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
 611
 612         info->high.rm_startblock = -1U;
 613         info->high.rm_owner = ULLONG_MAX;
 614         info->high.rm_offset = ULLONG_MAX;
 615         info->high.rm_blockcount = 0;
 616         info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
 617
 618         start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
 619         end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
 620
 621         /* Query each AG */
 622         for (info->agno = start_ag; info->agno <= end_ag; info->agno++) {
 623                 /*
 624                  * Set the AG high key from the fsmap high key if this
 625                  * is the last AG that we're querying.
 626                  */
 627                 if (info->agno == end_ag) {
 628                         info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
 629                                         end_fsb);
 630                         info->high.rm_offset = XFS_BB_TO_FSBT(mp,
 631                                         keys[1].fmr_offset);
 632                         error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
 633                         if (error)
 634                                 goto err;
 635                         xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
 636                 }
 637
 638                 if (bt_cur) {
 639                         xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
 640                         bt_cur = NULL;
 641                         xfs_trans_brelse(tp, info->agf_bp);
 642                         info->agf_bp = NULL;
 643                 }
 644
 645                 error = xfs_alloc_read_agf(mp, tp, info->agno, 0,
 646                                 &info->agf_bp);
 647                 if (error)
 648                         goto err;
 649
 650                 trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
 651                 trace_xfs_fsmap_high_key(mp, info->dev, info->agno,
 652                                 &info->high);
 653
 654                 error = query_fn(tp, info, &bt_cur, priv);
 655                 if (error)
 656                         goto err;
 657
 658                 /*
 659                  * Set the AG low key to the start of the AG prior to
 660                  * moving on to the next AG.
 661                  */
 662                 if (info->agno == start_ag) {
 663                         info->low.rm_startblock = 0;
 664                         info->low.rm_owner = 0;
 665                         info->low.rm_offset = 0;
 666                         info->low.rm_flags = 0;
 667                 }
 668         }
 669
 670         /* Report any gap at the end of the AG */
 671         info->last = true;
 672         error = query_fn(tp, info, &bt_cur, priv);
 673         if (error)
 674                 goto err;
 675
 676 err:
 677         if (bt_cur)
 678                 xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
 679                                                          XFS_BTREE_NOERROR);
 680         if (info->agf_bp) {
 681                 xfs_trans_brelse(tp, info->agf_bp);
 682                 info->agf_bp = NULL;
 683         }
 684
 685         return error;
 686 }
 687
 688 /* Actually query the rmap btree. */
 689 STATIC int
 690 xfs_getfsmap_datadev_rmapbt_query(
 691         struct xfs_trans                *tp,
 692         struct xfs_getfsmap_info        *info,
 693         struct xfs_btree_cur            **curpp,
 694         void                            *priv)
 695 {
 696         /* Report any gap at the end of the last AG. */
 697         if (info->last)
 698                 return xfs_getfsmap_datadev_helper(*curpp, &info->high, info);
 699
 700         /* Allocate cursor for this AG and query_range it. */
 701         *curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
 702                         info->agno);
 703         return xfs_rmap_query_range(*curpp, &info->low, &info->high,
 704                         xfs_getfsmap_datadev_helper, info);
 705 }
 706
 707 /* Execute a getfsmap query against the regular data device rmapbt. */
 708 STATIC int
 709 xfs_getfsmap_datadev_rmapbt(
 710         struct xfs_trans                *tp,
 711         struct xfs_fsmap                *keys,
 712         struct xfs_getfsmap_info        *info)
 713 {
 714         info->missing_owner = XFS_FMR_OWN_FREE;
 715         return __xfs_getfsmap_datadev(tp, keys, info,
 716                         xfs_getfsmap_datadev_rmapbt_query, NULL);
 717 }
 718
 719 /* Actually query the bno btree. */
 720 STATIC int
 721 xfs_getfsmap_datadev_bnobt_query(
 722         struct xfs_trans                *tp,
 723         struct xfs_getfsmap_info        *info,
 724         struct xfs_btree_cur            **curpp,
 725         void                            *priv)
 726 {
 727         struct xfs_alloc_rec_incore     *key = priv;
 728
 729         /* Report any gap at the end of the last AG. */
 730         if (info->last)
 731                 return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
 732
 733         /* Allocate cursor for this AG and query_range it. */
 734         *curpp = xfs_allocbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
 735                         info->agno, XFS_BTNUM_BNO);
 736         key->ar_startblock = info->low.rm_startblock;
 737         key[1].ar_startblock = info->high.rm_startblock;
 738         return xfs_alloc_query_range(*curpp, key, &key[1],
 739                         xfs_getfsmap_datadev_bnobt_helper, info);
 740 }
 741
 742 /* Execute a getfsmap query against the regular data device's bnobt. */
 743 STATIC int
 744 xfs_getfsmap_datadev_bnobt(
 745         struct xfs_trans                *tp,
 746         struct xfs_fsmap                *keys,
 747         struct xfs_getfsmap_info        *info)
 748 {
 749         struct xfs_alloc_rec_incore     akeys[2];
 750
 751         info->missing_owner = XFS_FMR_OWN_UNKNOWN;
 752         return __xfs_getfsmap_datadev(tp, keys, info,
 753                         xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
 754 }
 755
 756 /* Do we recognize the device? */
 757 STATIC bool
 758 xfs_getfsmap_is_valid_device(
 759         struct xfs_mount        *mp,
 760         struct xfs_fsmap        *fm)
 761 {
 762         if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
 763             fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
 764                 return true;
 765         if (mp->m_logdev_targp &&
 766             fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
 767                 return true;
 768         if (mp->m_rtdev_targp &&
 769             fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
 770                 return true;
 771         return false;
 772 }
 773
 774 /* Ensure that the low key is less than the high key. */
 775 STATIC bool
 776 xfs_getfsmap_check_keys(
 777         struct xfs_fsmap                *low_key,
 778         struct xfs_fsmap                *high_key)
 779 {
 780         if (low_key->fmr_device > high_key->fmr_device)
 781                 return false;
 782         if (low_key->fmr_device < high_key->fmr_device)
 783                 return true;
 784
 785         if (low_key->fmr_physical > high_key->fmr_physical)
 786                 return false;
 787         if (low_key->fmr_physical < high_key->fmr_physical)
 788                 return true;
 789
 790         if (low_key->fmr_owner > high_key->fmr_owner)
 791                 return false;
 792         if (low_key->fmr_owner < high_key->fmr_owner)
 793                 return true;
 794
 795         if (low_key->fmr_offset > high_key->fmr_offset)
 796                 return false;
 797         if (low_key->fmr_offset < high_key->fmr_offset)
 798                 return true;
 799
 800         return false;
 801 }
 802
 803 /*
 804  * There are only two devices if we didn't configure RT devices at build time.
 805  */
 806 #ifdef CONFIG_XFS_RT
 807 #define XFS_GETFSMAP_DEVS       3
 808 #else
 809 #define XFS_GETFSMAP_DEVS       2
 810 #endif /* CONFIG_XFS_RT */
 811
 812 /*
 813  * Get filesystem's extents as described in head, and format for
 814  * output.  Calls formatter to fill the user's buffer until all
 815  * extents are mapped, until the passed-in head->fmh_count slots have
 816  * been filled, or until the formatter short-circuits the loop, if it
 817  * is tracking filled-in extents on its own.
 818  *
 819  * Key to Confusion
 820  * ----------------
 821  * There are multiple levels of keys and counters at work here:
 822  * xfs_fsmap_head.fmh_keys      -- low and high fsmap keys passed in;
 823  *                                 these reflect fs-wide sector addrs.
 824  * dkeys                        -- fmh_keys used to query each device;
 825  *                                 these are fmh_keys but w/ the low key
 826  *                                 bumped up by fmr_length.
 827  * xfs_getfsmap_info.next_daddr -- next disk addr we expect to see; this
 828  *                                 is how we detect gaps in the fsmap
 829                                    records and report them.
 830  * xfs_getfsmap_info.low/high   -- per-AG low/high keys computed from
 831  *                                 dkeys; used to query the metadata.
 832  */
 833 int
 834 xfs_getfsmap(
 835         struct xfs_mount                *mp,
 836         struct xfs_fsmap_head           *head,
 837         xfs_fsmap_format_t              formatter,
 838         void                            *arg)
 839 {
 840         struct xfs_trans                *tp = NULL;
 841         struct xfs_fsmap                dkeys[2];       /* per-dev keys */
 842         struct xfs_getfsmap_dev         handlers[XFS_GETFSMAP_DEVS];
 843         struct xfs_getfsmap_info        info = { NULL };
 844         bool                            use_rmap;
 845         int                             i;
 846         int                             error = 0;
 847
 848         if (head->fmh_iflags & ~FMH_IF_VALID)
 849                 return -EINVAL;
 850         if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
 851             !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
 852                 return -EINVAL;
 853
 854         use_rmap = capable(CAP_SYS_ADMIN) &&
 855                    xfs_sb_version_hasrmapbt(&mp->m_sb);
 856         head->fmh_entries = 0;
 857
 858         /* Set up our device handlers. */
 859         memset(handlers, 0, sizeof(handlers));
 860         handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
 861         if (use_rmap)
 862                 handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
 863         else
 864                 handlers[0].fn = xfs_getfsmap_datadev_bnobt;
 865         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 866                 handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
 867                 handlers[1].fn = xfs_getfsmap_logdev;
 868         }
 869 #ifdef CONFIG_XFS_RT
 870         if (mp->m_rtdev_targp) {
 871                 handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
 872                 handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
 873         }
 874 #endif /* CONFIG_XFS_RT */
 875
 876         xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
 877                         xfs_getfsmap_dev_compare);
 878
 879         /*
 880          * To continue where we left off, we allow userspace to use the
 881          * last mapping from a previous call as the low key of the next.
 882          * This is identified by a non-zero length in the low key. We
 883          * have to increment the low key in this scenario to ensure we
 884          * don't return the same mapping again, and instead return the
 885          * very next mapping.
 886          *
 887          * If the low key mapping refers to file data, the same physical
 888          * blocks could be mapped to several other files/offsets.
 889          * According to rmapbt record ordering, the minimal next
 890          * possible record for the block range is the next starting
 891          * offset in the same inode. Therefore, bump the file offset to
 892          * continue the search appropriately.  For all other low key
 893          * mapping types (attr blocks, metadata), bump the physical
 894          * offset as there can be no other mapping for the same physical
 895          * block range.
 896          */
 897         dkeys[0] = head->fmh_keys[0];
 898         if (dkeys[0].fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
 899                 dkeys[0].fmr_physical += dkeys[0].fmr_length;
 900                 dkeys[0].fmr_owner = 0;
 901                 if (dkeys[0].fmr_offset)
 902                         return -EINVAL;
 903         } else
 904                 dkeys[0].fmr_offset += dkeys[0].fmr_length;
 905         dkeys[0].fmr_length = 0;
 906         memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
 907
 908         if (!xfs_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
 909                 return -EINVAL;
 910
 911         info.next_daddr = head->fmh_keys[0].fmr_physical +
 912                           head->fmh_keys[0].fmr_length;
 913         info.formatter = formatter;
 914         info.format_arg = arg;
 915         info.head = head;
 916
 917         /* For each device we support... */
 918         for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
 919                 /* Is this device within the range the user asked for? */
 920                 if (!handlers[i].fn)
 921                         continue;
 922                 if (head->fmh_keys[0].fmr_device > handlers[i].dev)
 923                         continue;
 924                 if (head->fmh_keys[1].fmr_device < handlers[i].dev)
 925                         break;
 926
 927                 /*
 928                  * If this device number matches the high key, we have
 929                  * to pass the high key to the handler to limit the
 930                  * query results.  If the device number exceeds the
 931                  * low key, zero out the low key so that we get
 932                  * everything from the beginning.
 933                  */
 934                 if (handlers[i].dev == head->fmh_keys[1].fmr_device)
 935                         dkeys[1] = head->fmh_keys[1];
 936                 if (handlers[i].dev > head->fmh_keys[0].fmr_device)
 937                         memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
 938
 939                 error = xfs_trans_alloc_empty(mp, &tp);
 940                 if (error)
 941                         break;
 942
 943                 info.dev = handlers[i].dev;
 944                 info.last = false;
 945                 info.agno = NULLAGNUMBER;
 946                 error = handlers[i].fn(tp, dkeys, &info);
 947                 if (error)
 948                         break;
 949                 xfs_trans_cancel(tp);
 950                 tp = NULL;
 951                 info.next_daddr = 0;
 952         }
 953
 954         if (tp)
 955                 xfs_trans_cancel(tp);
 956         head->fmh_oflags = FMH_OF_DEV_T;
 957         return error;
 958 }