fs/xfs/xfs_fsmap.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_mount.h"
  13 #include "xfs_inode.h"
  14 #include "xfs_trans.h"
  15 #include "xfs_btree.h"
  16 #include "xfs_rmap_btree.h"
  17 #include "xfs_trace.h"
  18 #include "xfs_rmap.h"
  19 #include "xfs_alloc.h"
  20 #include "xfs_bit.h"
  21 #include <linux/fsmap.h>
  22 #include "xfs_fsmap.h"
  23 #include "xfs_refcount.h"
  24 #include "xfs_refcount_btree.h"
  25 #include "xfs_alloc_btree.h"
  26 #include "xfs_rtalloc.h"
  27 #include "xfs_ag.h"
  28
  29 /* Convert an xfs_fsmap to an fsmap. */
  30 static void
  31 xfs_fsmap_from_internal(
  32         struct fsmap            *dest,
  33         struct xfs_fsmap        *src)
  34 {
  35         dest->fmr_device = src->fmr_device;
  36         dest->fmr_flags = src->fmr_flags;
  37         dest->fmr_physical = BBTOB(src->fmr_physical);
  38         dest->fmr_owner = src->fmr_owner;
  39         dest->fmr_offset = BBTOB(src->fmr_offset);
  40         dest->fmr_length = BBTOB(src->fmr_length);
  41         dest->fmr_reserved[0] = 0;
  42         dest->fmr_reserved[1] = 0;
  43         dest->fmr_reserved[2] = 0;
  44 }
  45
  46 /* Convert an fsmap to an xfs_fsmap. */
  47 void
  48 xfs_fsmap_to_internal(
  49         struct xfs_fsmap        *dest,
  50         struct fsmap            *src)
  51 {
  52         dest->fmr_device = src->fmr_device;
  53         dest->fmr_flags = src->fmr_flags;
  54         dest->fmr_physical = BTOBBT(src->fmr_physical);
  55         dest->fmr_owner = src->fmr_owner;
  56         dest->fmr_offset = BTOBBT(src->fmr_offset);
  57         dest->fmr_length = BTOBBT(src->fmr_length);
  58 }
  59
  60 /* Convert an fsmap owner into an rmapbt owner. */
  61 static int
  62 xfs_fsmap_owner_to_rmap(
  63         struct xfs_rmap_irec    *dest,
  64         const struct xfs_fsmap  *src)
  65 {
  66         if (!(src->fmr_flags & FMR_OF_SPECIAL_OWNER)) {
  67                 dest->rm_owner = src->fmr_owner;
  68                 return 0;
  69         }
  70
  71         switch (src->fmr_owner) {
  72         case 0:                 /* "lowest owner id possible" */
  73         case -1ULL:             /* "highest owner id possible" */
  74                 dest->rm_owner = 0;
  75                 break;
  76         case XFS_FMR_OWN_FREE:
  77                 dest->rm_owner = XFS_RMAP_OWN_NULL;
  78                 break;
  79         case XFS_FMR_OWN_UNKNOWN:
  80                 dest->rm_owner = XFS_RMAP_OWN_UNKNOWN;
  81                 break;
  82         case XFS_FMR_OWN_FS:
  83                 dest->rm_owner = XFS_RMAP_OWN_FS;
  84                 break;
  85         case XFS_FMR_OWN_LOG:
  86                 dest->rm_owner = XFS_RMAP_OWN_LOG;
  87                 break;
  88         case XFS_FMR_OWN_AG:
  89                 dest->rm_owner = XFS_RMAP_OWN_AG;
  90                 break;
  91         case XFS_FMR_OWN_INOBT:
  92                 dest->rm_owner = XFS_RMAP_OWN_INOBT;
  93                 break;
  94         case XFS_FMR_OWN_INODES:
  95                 dest->rm_owner = XFS_RMAP_OWN_INODES;
  96                 break;
  97         case XFS_FMR_OWN_REFC:
  98                 dest->rm_owner = XFS_RMAP_OWN_REFC;
  99                 break;
 100         case XFS_FMR_OWN_COW:
 101                 dest->rm_owner = XFS_RMAP_OWN_COW;
 102                 break;
 103         case XFS_FMR_OWN_DEFECTIVE:     /* not implemented */
 104                 /* fall through */
 105         default:
 106                 return -EINVAL;
 107         }
 108         return 0;
 109 }
 110
 111 /* Convert an rmapbt owner into an fsmap owner. */
 112 static int
 113 xfs_fsmap_owner_from_rmap(
 114         struct xfs_fsmap                *dest,
 115         const struct xfs_rmap_irec      *src)
 116 {
 117         dest->fmr_flags = 0;
 118         if (!XFS_RMAP_NON_INODE_OWNER(src->rm_owner)) {
 119                 dest->fmr_owner = src->rm_owner;
 120                 return 0;
 121         }
 122         dest->fmr_flags |= FMR_OF_SPECIAL_OWNER;
 123
 124         switch (src->rm_owner) {
 125         case XFS_RMAP_OWN_FS:
 126                 dest->fmr_owner = XFS_FMR_OWN_FS;
 127                 break;
 128         case XFS_RMAP_OWN_LOG:
 129                 dest->fmr_owner = XFS_FMR_OWN_LOG;
 130                 break;
 131         case XFS_RMAP_OWN_AG:
 132                 dest->fmr_owner = XFS_FMR_OWN_AG;
 133                 break;
 134         case XFS_RMAP_OWN_INOBT:
 135                 dest->fmr_owner = XFS_FMR_OWN_INOBT;
 136                 break;
 137         case XFS_RMAP_OWN_INODES:
 138                 dest->fmr_owner = XFS_FMR_OWN_INODES;
 139                 break;
 140         case XFS_RMAP_OWN_REFC:
 141                 dest->fmr_owner = XFS_FMR_OWN_REFC;
 142                 break;
 143         case XFS_RMAP_OWN_COW:
 144                 dest->fmr_owner = XFS_FMR_OWN_COW;
 145                 break;
 146         case XFS_RMAP_OWN_NULL: /* "free" */
 147                 dest->fmr_owner = XFS_FMR_OWN_FREE;
 148                 break;
 149         default:
 150                 ASSERT(0);
 151                 return -EFSCORRUPTED;
 152         }
 153         return 0;
 154 }
 155
 156 /* getfsmap query state */
 157 struct xfs_getfsmap_info {
 158         struct xfs_fsmap_head   *head;
 159         struct fsmap            *fsmap_recs;    /* mapping records */
 160         struct xfs_buf          *agf_bp;        /* AGF, for refcount queries */
 161         struct xfs_perag        *pag;           /* AG info, if applicable */
 162         xfs_daddr_t             next_daddr;     /* next daddr we expect */
 163         /* daddr of low fsmap key when we're using the rtbitmap */
 164         xfs_daddr_t             low_daddr;
 165         u64                     missing_owner;  /* owner of holes */
 166         u32                     dev;            /* device id */
 167         /*
 168          * Low rmap key for the query.  If low.rm_blockcount is nonzero, this
 169          * is the second (or later) call to retrieve the recordset in pieces.
 170          * xfs_getfsmap_rec_before_start will compare all records retrieved
 171          * by the rmapbt query to filter out any records that start before
 172          * the last record.
 173          */
 174         struct xfs_rmap_irec    low;
 175         struct xfs_rmap_irec    high;           /* high rmap key */
 176         bool                    last;           /* last extent? */
 177 };
 178
 179 /* Associate a device with a getfsmap handler. */
 180 struct xfs_getfsmap_dev {
 181         u32                     dev;
 182         int                     (*fn)(struct xfs_trans *tp,
 183                                       const struct xfs_fsmap *keys,
 184                                       struct xfs_getfsmap_info *info);
 185 };
 186
 187 /* Compare two getfsmap device handlers. */
 188 static int
 189 xfs_getfsmap_dev_compare(
 190         const void                      *p1,
 191         const void                      *p2)
 192 {
 193         const struct xfs_getfsmap_dev   *d1 = p1;
 194         const struct xfs_getfsmap_dev   *d2 = p2;
 195
 196         return d1->dev - d2->dev;
 197 }
 198
 199 /* Decide if this mapping is shared. */
 200 STATIC int
 201 xfs_getfsmap_is_shared(
 202         struct xfs_trans                *tp,
 203         struct xfs_getfsmap_info        *info,
 204         const struct xfs_rmap_irec      *rec,
 205         bool                            *stat)
 206 {
 207         struct xfs_mount                *mp = tp->t_mountp;
 208         struct xfs_btree_cur            *cur;
 209         xfs_agblock_t                   fbno;
 210         xfs_extlen_t                    flen;
 211         int                             error;
 212
 213         *stat = false;
 214         if (!xfs_has_reflink(mp))
 215                 return 0;
 216         /* rt files will have no perag structure */
 217         if (!info->pag)
 218                 return 0;
 219
 220         /* Are there any shared blocks here? */
 221         flen = 0;
 222         cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp, info->pag);
 223
 224         error = xfs_refcount_find_shared(cur, rec->rm_startblock,
 225                         rec->rm_blockcount, &fbno, &flen, false);
 226
 227         xfs_btree_del_cursor(cur, error);
 228         if (error)
 229                 return error;
 230
 231         *stat = flen > 0;
 232         return 0;
 233 }
 234
 235 static inline void
 236 xfs_getfsmap_format(
 237         struct xfs_mount                *mp,
 238         struct xfs_fsmap                *xfm,
 239         struct xfs_getfsmap_info        *info)
 240 {
 241         struct fsmap                    *rec;
 242
 243         trace_xfs_getfsmap_mapping(mp, xfm);
 244
 245         rec = &info->fsmap_recs[info->head->fmh_entries++];
 246         xfs_fsmap_from_internal(rec, xfm);
 247 }
 248
 249 static inline bool
 250 xfs_getfsmap_rec_before_start(
 251         struct xfs_getfsmap_info        *info,
 252         const struct xfs_rmap_irec      *rec,
 253         xfs_daddr_t                     rec_daddr)
 254 {
 255         if (info->low_daddr != -1ULL)
 256                 return rec_daddr < info->low_daddr;
 257         if (info->low.rm_blockcount)
 258                 return xfs_rmap_compare(rec, &info->low) < 0;
 259         return false;
 260 }
 261
 262 /*
 263  * Format a reverse mapping for getfsmap, having translated rm_startblock
 264  * into the appropriate daddr units.  Pass in a nonzero @len_daddr if the
 265  * length could be larger than rm_blockcount in struct xfs_rmap_irec.
 266  */
 267 STATIC int
 268 xfs_getfsmap_helper(
 269         struct xfs_trans                *tp,
 270         struct xfs_getfsmap_info        *info,
 271         const struct xfs_rmap_irec      *rec,
 272         xfs_daddr_t                     rec_daddr,
 273         xfs_daddr_t                     len_daddr)
 274 {
 275         struct xfs_fsmap                fmr;
 276         struct xfs_mount                *mp = tp->t_mountp;
 277         bool                            shared;
 278         int                             error;
 279
 280         if (fatal_signal_pending(current))
 281                 return -EINTR;
 282
 283         if (len_daddr == 0)
 284                 len_daddr = XFS_FSB_TO_BB(mp, rec->rm_blockcount);
 285
 286         /*
 287          * Filter out records that start before our startpoint, if the
 288          * caller requested that.
 289          */
 290         if (xfs_getfsmap_rec_before_start(info, rec, rec_daddr)) {
 291                 rec_daddr += len_daddr;
 292                 if (info->next_daddr < rec_daddr)
 293                         info->next_daddr = rec_daddr;
 294                 return 0;
 295         }
 296
 297         /* Are we just counting mappings? */
 298         if (info->head->fmh_count == 0) {
 299                 if (info->head->fmh_entries == UINT_MAX)
 300                         return -ECANCELED;
 301
 302                 if (rec_daddr > info->next_daddr)
 303                         info->head->fmh_entries++;
 304
 305                 if (info->last)
 306                         return 0;
 307
 308                 info->head->fmh_entries++;
 309
 310                 rec_daddr += len_daddr;
 311                 if (info->next_daddr < rec_daddr)
 312                         info->next_daddr = rec_daddr;
 313                 return 0;
 314         }
 315
 316         /*
 317          * If the record starts past the last physical block we saw,
 318          * then we've found a gap.  Report the gap as being owned by
 319          * whatever the caller specified is the missing owner.
 320          */
 321         if (rec_daddr > info->next_daddr) {
 322                 if (info->head->fmh_entries >= info->head->fmh_count)
 323                         return -ECANCELED;
 324
 325                 fmr.fmr_device = info->dev;
 326                 fmr.fmr_physical = info->next_daddr;
 327                 fmr.fmr_owner = info->missing_owner;
 328                 fmr.fmr_offset = 0;
 329                 fmr.fmr_length = rec_daddr - info->next_daddr;
 330                 fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
 331                 xfs_getfsmap_format(mp, &fmr, info);
 332         }
 333
 334         if (info->last)
 335                 goto out;
 336
 337         /* Fill out the extent we found */
 338         if (info->head->fmh_entries >= info->head->fmh_count)
 339                 return -ECANCELED;
 340
 341         trace_xfs_fsmap_mapping(mp, info->dev,
 342                         info->pag ? info->pag->pag_agno : NULLAGNUMBER, rec);
 343
 344         fmr.fmr_device = info->dev;
 345         fmr.fmr_physical = rec_daddr;
 346         error = xfs_fsmap_owner_from_rmap(&fmr, rec);
 347         if (error)
 348                 return error;
 349         fmr.fmr_offset = XFS_FSB_TO_BB(mp, rec->rm_offset);
 350         fmr.fmr_length = len_daddr;
 351         if (rec->rm_flags & XFS_RMAP_UNWRITTEN)
 352                 fmr.fmr_flags |= FMR_OF_PREALLOC;
 353         if (rec->rm_flags & XFS_RMAP_ATTR_FORK)
 354                 fmr.fmr_flags |= FMR_OF_ATTR_FORK;
 355         if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
 356                 fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
 357         if (fmr.fmr_flags == 0) {
 358                 error = xfs_getfsmap_is_shared(tp, info, rec, &shared);
 359                 if (error)
 360                         return error;
 361                 if (shared)
 362                         fmr.fmr_flags |= FMR_OF_SHARED;
 363         }
 364
 365         xfs_getfsmap_format(mp, &fmr, info);
 366 out:
 367         rec_daddr += len_daddr;
 368         if (info->next_daddr < rec_daddr)
 369                 info->next_daddr = rec_daddr;
 370         return 0;
 371 }
 372
 373 /* Transform a rmapbt irec into a fsmap */
 374 STATIC int
 375 xfs_getfsmap_datadev_helper(
 376         struct xfs_btree_cur            *cur,
 377         const struct xfs_rmap_irec      *rec,
 378         void                            *priv)
 379 {
 380         struct xfs_mount                *mp = cur->bc_mp;
 381         struct xfs_getfsmap_info        *info = priv;
 382         xfs_fsblock_t                   fsb;
 383         xfs_daddr_t                     rec_daddr;
 384
 385         fsb = XFS_AGB_TO_FSB(mp, cur->bc_ag.pag->pag_agno, rec->rm_startblock);
 386         rec_daddr = XFS_FSB_TO_DADDR(mp, fsb);
 387
 388         return xfs_getfsmap_helper(cur->bc_tp, info, rec, rec_daddr, 0);
 389 }
 390
 391 /* Transform a bnobt irec into a fsmap */
 392 STATIC int
 393 xfs_getfsmap_datadev_bnobt_helper(
 394         struct xfs_btree_cur            *cur,
 395         const struct xfs_alloc_rec_incore *rec,
 396         void                            *priv)
 397 {
 398         struct xfs_mount                *mp = cur->bc_mp;
 399         struct xfs_getfsmap_info        *info = priv;
 400         struct xfs_rmap_irec            irec;
 401         xfs_daddr_t                     rec_daddr;
 402
 403         rec_daddr = XFS_AGB_TO_DADDR(mp, cur->bc_ag.pag->pag_agno,
 404                         rec->ar_startblock);
 405
 406         irec.rm_startblock = rec->ar_startblock;
 407         irec.rm_blockcount = rec->ar_blockcount;
 408         irec.rm_owner = XFS_RMAP_OWN_NULL;      /* "free" */
 409         irec.rm_offset = 0;
 410         irec.rm_flags = 0;
 411
 412         return xfs_getfsmap_helper(cur->bc_tp, info, &irec, rec_daddr, 0);
 413 }
 414
 415 /* Set rmap flags based on the getfsmap flags */
 416 static void
 417 xfs_getfsmap_set_irec_flags(
 418         struct xfs_rmap_irec    *irec,
 419         const struct xfs_fsmap  *fmr)
 420 {
 421         irec->rm_flags = 0;
 422         if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
 423                 irec->rm_flags |= XFS_RMAP_ATTR_FORK;
 424         if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
 425                 irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
 426         if (fmr->fmr_flags & FMR_OF_PREALLOC)
 427                 irec->rm_flags |= XFS_RMAP_UNWRITTEN;
 428 }
 429
 430 /* Execute a getfsmap query against the log device. */
 431 STATIC int
 432 xfs_getfsmap_logdev(
 433         struct xfs_trans                *tp,
 434         const struct xfs_fsmap          *keys,
 435         struct xfs_getfsmap_info        *info)
 436 {
 437         struct xfs_mount                *mp = tp->t_mountp;
 438         struct xfs_rmap_irec            rmap;
 439         xfs_daddr_t                     rec_daddr, len_daddr;
 440         xfs_fsblock_t                   start_fsb, end_fsb;
 441         uint64_t                        eofs;
 442
 443         eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 444         if (keys[0].fmr_physical >= eofs)
 445                 return 0;
 446         start_fsb = XFS_BB_TO_FSBT(mp,
 447                                 keys[0].fmr_physical + keys[0].fmr_length);
 448         end_fsb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
 449
 450         /* Adjust the low key if we are continuing from where we left off. */
 451         if (keys[0].fmr_length > 0)
 452                 info->low_daddr = XFS_FSB_TO_BB(mp, start_fsb);
 453
 454         trace_xfs_fsmap_low_key_linear(mp, info->dev, start_fsb);
 455         trace_xfs_fsmap_high_key_linear(mp, info->dev, end_fsb);
 456
 457         if (start_fsb > 0)
 458                 return 0;
 459
 460         /* Fabricate an rmap entry for the external log device. */
 461         rmap.rm_startblock = 0;
 462         rmap.rm_blockcount = mp->m_sb.sb_logblocks;
 463         rmap.rm_owner = XFS_RMAP_OWN_LOG;
 464         rmap.rm_offset = 0;
 465         rmap.rm_flags = 0;
 466
 467         rec_daddr = XFS_FSB_TO_BB(mp, rmap.rm_startblock);
 468         len_daddr = XFS_FSB_TO_BB(mp, rmap.rm_blockcount);
 469         return xfs_getfsmap_helper(tp, info, &rmap, rec_daddr, len_daddr);
 470 }
 471
 472 #ifdef CONFIG_XFS_RT
 473 /* Transform a rtbitmap "record" into a fsmap */
 474 STATIC int
 475 xfs_getfsmap_rtdev_rtbitmap_helper(
 476         struct xfs_mount                *mp,
 477         struct xfs_trans                *tp,
 478         const struct xfs_rtalloc_rec    *rec,
 479         void                            *priv)
 480 {
 481         struct xfs_getfsmap_info        *info = priv;
 482         struct xfs_rmap_irec            irec;
 483         xfs_rtblock_t                   rtbno;
 484         xfs_daddr_t                     rec_daddr, len_daddr;
 485
 486         rtbno = rec->ar_startext * mp->m_sb.sb_rextsize;
 487         rec_daddr = XFS_FSB_TO_BB(mp, rtbno);
 488         irec.rm_startblock = rtbno;
 489
 490         rtbno = rec->ar_extcount * mp->m_sb.sb_rextsize;
 491         len_daddr = XFS_FSB_TO_BB(mp, rtbno);
 492         irec.rm_blockcount = rtbno;
 493
 494         irec.rm_owner = XFS_RMAP_OWN_NULL;      /* "free" */
 495         irec.rm_offset = 0;
 496         irec.rm_flags = 0;
 497
 498         return xfs_getfsmap_helper(tp, info, &irec, rec_daddr, len_daddr);
 499 }
 500
 501 /* Execute a getfsmap query against the realtime device rtbitmap. */
 502 STATIC int
 503 xfs_getfsmap_rtdev_rtbitmap(
 504         struct xfs_trans                *tp,
 505         const struct xfs_fsmap          *keys,
 506         struct xfs_getfsmap_info        *info)
 507 {
 508
 509         struct xfs_rtalloc_rec          alow = { 0 };
 510         struct xfs_rtalloc_rec          ahigh = { 0 };
 511         struct xfs_mount                *mp = tp->t_mountp;
 512         xfs_rtblock_t                   start_rtb;
 513         xfs_rtblock_t                   end_rtb;
 514         uint64_t                        eofs;
 515         int                             error;
 516
 517         eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rextents * mp->m_sb.sb_rextsize);
 518         if (keys[0].fmr_physical >= eofs)
 519                 return 0;
 520         start_rtb = XFS_BB_TO_FSBT(mp,
 521                                 keys[0].fmr_physical + keys[0].fmr_length);
 522         end_rtb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
 523
 524         info->missing_owner = XFS_FMR_OWN_UNKNOWN;
 525
 526         /* Adjust the low key if we are continuing from where we left off. */
 527         if (keys[0].fmr_length > 0) {
 528                 info->low_daddr = XFS_FSB_TO_BB(mp, start_rtb);
 529                 if (info->low_daddr >= eofs)
 530                         return 0;
 531         }
 532
 533         trace_xfs_fsmap_low_key_linear(mp, info->dev, start_rtb);
 534         trace_xfs_fsmap_high_key_linear(mp, info->dev, end_rtb);
 535
 536         xfs_ilock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
 537
 538         /*
 539          * Set up query parameters to return free rtextents covering the range
 540          * we want.
 541          */
 542         alow.ar_startext = start_rtb;
 543         ahigh.ar_startext = end_rtb;
 544         do_div(alow.ar_startext, mp->m_sb.sb_rextsize);
 545         if (do_div(ahigh.ar_startext, mp->m_sb.sb_rextsize))
 546                 ahigh.ar_startext++;
 547         error = xfs_rtalloc_query_range(mp, tp, &alow, &ahigh,
 548                         xfs_getfsmap_rtdev_rtbitmap_helper, info);
 549         if (error)
 550                 goto err;
 551
 552         /*
 553          * Report any gaps at the end of the rtbitmap by simulating a null
 554          * rmap starting at the block after the end of the query range.
 555          */
 556         info->last = true;
 557         ahigh.ar_startext = min(mp->m_sb.sb_rextents, ahigh.ar_startext);
 558
 559         error = xfs_getfsmap_rtdev_rtbitmap_helper(mp, tp, &ahigh, info);
 560         if (error)
 561                 goto err;
 562 err:
 563         xfs_iunlock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
 564         return error;
 565 }
 566 #endif /* CONFIG_XFS_RT */
 567
 568 /* Execute a getfsmap query against the regular data device. */
 569 STATIC int
 570 __xfs_getfsmap_datadev(
 571         struct xfs_trans                *tp,
 572         const struct xfs_fsmap          *keys,
 573         struct xfs_getfsmap_info        *info,
 574         int                             (*query_fn)(struct xfs_trans *,
 575                                                     struct xfs_getfsmap_info *,
 576                                                     struct xfs_btree_cur **,
 577                                                     void *),
 578         void                            *priv)
 579 {
 580         struct xfs_mount                *mp = tp->t_mountp;
 581         struct xfs_perag                *pag;
 582         struct xfs_btree_cur            *bt_cur = NULL;
 583         xfs_fsblock_t                   start_fsb;
 584         xfs_fsblock_t                   end_fsb;
 585         xfs_agnumber_t                  start_ag;
 586         xfs_agnumber_t                  end_ag;
 587         uint64_t                        eofs;
 588         int                             error = 0;
 589
 590         eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 591         if (keys[0].fmr_physical >= eofs)
 592                 return 0;
 593         start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
 594         end_fsb = XFS_DADDR_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
 595
 596         /*
 597          * Convert the fsmap low/high keys to AG based keys.  Initialize
 598          * low to the fsmap low key and max out the high key to the end
 599          * of the AG.
 600          */
 601         info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
 602         info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
 603         error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
 604         if (error)
 605                 return error;
 606         info->low.rm_blockcount = XFS_BB_TO_FSBT(mp, keys[0].fmr_length);
 607         xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
 608
 609         /* Adjust the low key if we are continuing from where we left off. */
 610         if (info->low.rm_blockcount == 0) {
 611                 /* empty */
 612         } else if (XFS_RMAP_NON_INODE_OWNER(info->low.rm_owner) ||
 613                    (info->low.rm_flags & (XFS_RMAP_ATTR_FORK |
 614                                           XFS_RMAP_BMBT_BLOCK |
 615                                           XFS_RMAP_UNWRITTEN))) {
 616                 info->low.rm_startblock += info->low.rm_blockcount;
 617                 info->low.rm_owner = 0;
 618                 info->low.rm_offset = 0;
 619
 620                 start_fsb += info->low.rm_blockcount;
 621                 if (XFS_FSB_TO_DADDR(mp, start_fsb) >= eofs)
 622                         return 0;
 623         } else {
 624                 info->low.rm_offset += info->low.rm_blockcount;
 625         }
 626
 627         info->high.rm_startblock = -1U;
 628         info->high.rm_owner = ULLONG_MAX;
 629         info->high.rm_offset = ULLONG_MAX;
 630         info->high.rm_blockcount = 0;
 631         info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
 632
 633         start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
 634         end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
 635
 636         for_each_perag_range(mp, start_ag, end_ag, pag) {
 637                 /*
 638                  * Set the AG high key from the fsmap high key if this
 639                  * is the last AG that we're querying.
 640                  */
 641                 info->pag = pag;
 642                 if (pag->pag_agno == end_ag) {
 643                         info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
 644                                         end_fsb);
 645                         info->high.rm_offset = XFS_BB_TO_FSBT(mp,
 646                                         keys[1].fmr_offset);
 647                         error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
 648                         if (error)
 649                                 break;
 650                         xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
 651                 }
 652
 653                 if (bt_cur) {
 654                         xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
 655                         bt_cur = NULL;
 656                         xfs_trans_brelse(tp, info->agf_bp);
 657                         info->agf_bp = NULL;
 658                 }
 659
 660                 error = xfs_alloc_read_agf(pag, tp, 0, &info->agf_bp);
 661                 if (error)
 662                         break;
 663
 664                 trace_xfs_fsmap_low_key(mp, info->dev, pag->pag_agno,
 665                                 &info->low);
 666                 trace_xfs_fsmap_high_key(mp, info->dev, pag->pag_agno,
 667                                 &info->high);
 668
 669                 error = query_fn(tp, info, &bt_cur, priv);
 670                 if (error)
 671                         break;
 672
 673                 /*
 674                  * Set the AG low key to the start of the AG prior to
 675                  * moving on to the next AG.
 676                  */
 677                 if (pag->pag_agno == start_ag)
 678                         memset(&info->low, 0, sizeof(info->low));
 679
 680                 /*
 681                  * If this is the last AG, report any gap at the end of it
 682                  * before we drop the reference to the perag when the loop
 683                  * terminates.
 684                  */
 685                 if (pag->pag_agno == end_ag) {
 686                         info->last = true;
 687                         error = query_fn(tp, info, &bt_cur, priv);
 688                         if (error)
 689                                 break;
 690                 }
 691                 info->pag = NULL;
 692         }
 693
 694         if (bt_cur)
 695                 xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
 696                                                          XFS_BTREE_NOERROR);
 697         if (info->agf_bp) {
 698                 xfs_trans_brelse(tp, info->agf_bp);
 699                 info->agf_bp = NULL;
 700         }
 701         if (info->pag) {
 702                 xfs_perag_rele(info->pag);
 703                 info->pag = NULL;
 704         } else if (pag) {
 705                 /* loop termination case */
 706                 xfs_perag_rele(pag);
 707         }
 708
 709         return error;
 710 }
 711
 712 /* Actually query the rmap btree. */
 713 STATIC int
 714 xfs_getfsmap_datadev_rmapbt_query(
 715         struct xfs_trans                *tp,
 716         struct xfs_getfsmap_info        *info,
 717         struct xfs_btree_cur            **curpp,
 718         void                            *priv)
 719 {
 720         /* Report any gap at the end of the last AG. */
 721         if (info->last)
 722                 return xfs_getfsmap_datadev_helper(*curpp, &info->high, info);
 723
 724         /* Allocate cursor for this AG and query_range it. */
 725         *curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
 726                         info->pag);
 727         return xfs_rmap_query_range(*curpp, &info->low, &info->high,
 728                         xfs_getfsmap_datadev_helper, info);
 729 }
 730
 731 /* Execute a getfsmap query against the regular data device rmapbt. */
 732 STATIC int
 733 xfs_getfsmap_datadev_rmapbt(
 734         struct xfs_trans                *tp,
 735         const struct xfs_fsmap          *keys,
 736         struct xfs_getfsmap_info        *info)
 737 {
 738         info->missing_owner = XFS_FMR_OWN_FREE;
 739         return __xfs_getfsmap_datadev(tp, keys, info,
 740                         xfs_getfsmap_datadev_rmapbt_query, NULL);
 741 }
 742
 743 /* Actually query the bno btree. */
 744 STATIC int
 745 xfs_getfsmap_datadev_bnobt_query(
 746         struct xfs_trans                *tp,
 747         struct xfs_getfsmap_info        *info,
 748         struct xfs_btree_cur            **curpp,
 749         void                            *priv)
 750 {
 751         struct xfs_alloc_rec_incore     *key = priv;
 752
 753         /* Report any gap at the end of the last AG. */
 754         if (info->last)
 755                 return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
 756
 757         /* Allocate cursor for this AG and query_range it. */
 758         *curpp = xfs_allocbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
 759                         info->pag, XFS_BTNUM_BNO);
 760         key->ar_startblock = info->low.rm_startblock;
 761         key[1].ar_startblock = info->high.rm_startblock;
 762         return xfs_alloc_query_range(*curpp, key, &key[1],
 763                         xfs_getfsmap_datadev_bnobt_helper, info);
 764 }
 765
 766 /* Execute a getfsmap query against the regular data device's bnobt. */
 767 STATIC int
 768 xfs_getfsmap_datadev_bnobt(
 769         struct xfs_trans                *tp,
 770         const struct xfs_fsmap          *keys,
 771         struct xfs_getfsmap_info        *info)
 772 {
 773         struct xfs_alloc_rec_incore     akeys[2];
 774
 775         memset(akeys, 0, sizeof(akeys));
 776         info->missing_owner = XFS_FMR_OWN_UNKNOWN;
 777         return __xfs_getfsmap_datadev(tp, keys, info,
 778                         xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
 779 }
 780
 781 /* Do we recognize the device? */
 782 STATIC bool
 783 xfs_getfsmap_is_valid_device(
 784         struct xfs_mount        *mp,
 785         struct xfs_fsmap        *fm)
 786 {
 787         if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
 788             fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
 789                 return true;
 790         if (mp->m_logdev_targp &&
 791             fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
 792                 return true;
 793         if (mp->m_rtdev_targp &&
 794             fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
 795                 return true;
 796         return false;
 797 }
 798
 799 /* Ensure that the low key is less than the high key. */
 800 STATIC bool
 801 xfs_getfsmap_check_keys(
 802         struct xfs_fsmap                *low_key,
 803         struct xfs_fsmap                *high_key)
 804 {
 805         if (low_key->fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
 806                 if (low_key->fmr_offset)
 807                         return false;
 808         }
 809         if (high_key->fmr_flags != -1U &&
 810             (high_key->fmr_flags & (FMR_OF_SPECIAL_OWNER |
 811                                     FMR_OF_EXTENT_MAP))) {
 812                 if (high_key->fmr_offset && high_key->fmr_offset != -1ULL)
 813                         return false;
 814         }
 815         if (high_key->fmr_length && high_key->fmr_length != -1ULL)
 816                 return false;
 817
 818         if (low_key->fmr_device > high_key->fmr_device)
 819                 return false;
 820         if (low_key->fmr_device < high_key->fmr_device)
 821                 return true;
 822
 823         if (low_key->fmr_physical > high_key->fmr_physical)
 824                 return false;
 825         if (low_key->fmr_physical < high_key->fmr_physical)
 826                 return true;
 827
 828         if (low_key->fmr_owner > high_key->fmr_owner)
 829                 return false;
 830         if (low_key->fmr_owner < high_key->fmr_owner)
 831                 return true;
 832
 833         if (low_key->fmr_offset > high_key->fmr_offset)
 834                 return false;
 835         if (low_key->fmr_offset < high_key->fmr_offset)
 836                 return true;
 837
 838         return false;
 839 }
 840
 841 /*
 842  * There are only two devices if we didn't configure RT devices at build time.
 843  */
 844 #ifdef CONFIG_XFS_RT
 845 #define XFS_GETFSMAP_DEVS       3
 846 #else
 847 #define XFS_GETFSMAP_DEVS       2
 848 #endif /* CONFIG_XFS_RT */
 849
 850 /*
 851  * Get filesystem's extents as described in head, and format for output. Fills
 852  * in the supplied records array until there are no more reverse mappings to
 853  * return or head.fmh_entries == head.fmh_count.  In the second case, this
 854  * function returns -ECANCELED to indicate that more records would have been
 855  * returned.
 856  *
 857  * Key to Confusion
 858  * ----------------
 859  * There are multiple levels of keys and counters at work here:
 860  * xfs_fsmap_head.fmh_keys      -- low and high fsmap keys passed in;
 861  *                                 these reflect fs-wide sector addrs.
 862  * dkeys                        -- fmh_keys used to query each device;
 863  *                                 these are fmh_keys but w/ the low key
 864  *                                 bumped up by fmr_length.
 865  * xfs_getfsmap_info.next_daddr -- next disk addr we expect to see; this
 866  *                                 is how we detect gaps in the fsmap
 867                                    records and report them.
 868  * xfs_getfsmap_info.low/high   -- per-AG low/high keys computed from
 869  *                                 dkeys; used to query the metadata.
 870  */
 871 int
 872 xfs_getfsmap(
 873         struct xfs_mount                *mp,
 874         struct xfs_fsmap_head           *head,
 875         struct fsmap                    *fsmap_recs)
 876 {
 877         struct xfs_trans                *tp = NULL;
 878         struct xfs_fsmap                dkeys[2];       /* per-dev keys */
 879         struct xfs_getfsmap_dev         handlers[XFS_GETFSMAP_DEVS];
 880         struct xfs_getfsmap_info        info = { NULL };
 881         bool                            use_rmap;
 882         int                             i;
 883         int                             error = 0;
 884
 885         if (head->fmh_iflags & ~FMH_IF_VALID)
 886                 return -EINVAL;
 887         if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
 888             !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
 889                 return -EINVAL;
 890         if (!xfs_getfsmap_check_keys(&head->fmh_keys[0], &head->fmh_keys[1]))
 891                 return -EINVAL;
 892
 893         use_rmap = xfs_has_rmapbt(mp) &&
 894                    has_capability_noaudit(current, CAP_SYS_ADMIN);
 895         head->fmh_entries = 0;
 896
 897         /* Set up our device handlers. */
 898         memset(handlers, 0, sizeof(handlers));
 899         handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
 900         if (use_rmap)
 901                 handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
 902         else
 903                 handlers[0].fn = xfs_getfsmap_datadev_bnobt;
 904         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 905                 handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
 906                 handlers[1].fn = xfs_getfsmap_logdev;
 907         }
 908 #ifdef CONFIG_XFS_RT
 909         if (mp->m_rtdev_targp) {
 910                 handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
 911                 handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
 912         }
 913 #endif /* CONFIG_XFS_RT */
 914
 915         xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
 916                         xfs_getfsmap_dev_compare);
 917
 918         /*
 919          * To continue where we left off, we allow userspace to use the
 920          * last mapping from a previous call as the low key of the next.
 921          * This is identified by a non-zero length in the low key. We
 922          * have to increment the low key in this scenario to ensure we
 923          * don't return the same mapping again, and instead return the
 924          * very next mapping.
 925          *
 926          * If the low key mapping refers to file data, the same physical
 927          * blocks could be mapped to several other files/offsets.
 928          * According to rmapbt record ordering, the minimal next
 929          * possible record for the block range is the next starting
 930          * offset in the same inode. Therefore, each fsmap backend bumps
 931          * the file offset to continue the search appropriately.  For
 932          * all other low key mapping types (attr blocks, metadata), each
 933          * fsmap backend bumps the physical offset as there can be no
 934          * other mapping for the same physical block range.
 935          */
 936         dkeys[0] = head->fmh_keys[0];
 937         memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
 938
 939         info.next_daddr = head->fmh_keys[0].fmr_physical +
 940                           head->fmh_keys[0].fmr_length;
 941         info.fsmap_recs = fsmap_recs;
 942         info.head = head;
 943
 944         /* For each device we support... */
 945         for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
 946                 /* Is this device within the range the user asked for? */
 947                 if (!handlers[i].fn)
 948                         continue;
 949                 if (head->fmh_keys[0].fmr_device > handlers[i].dev)
 950                         continue;
 951                 if (head->fmh_keys[1].fmr_device < handlers[i].dev)
 952                         break;
 953
 954                 /*
 955                  * If this device number matches the high key, we have
 956                  * to pass the high key to the handler to limit the
 957                  * query results.  If the device number exceeds the
 958                  * low key, zero out the low key so that we get
 959                  * everything from the beginning.
 960                  */
 961                 if (handlers[i].dev == head->fmh_keys[1].fmr_device)
 962                         dkeys[1] = head->fmh_keys[1];
 963                 if (handlers[i].dev > head->fmh_keys[0].fmr_device)
 964                         memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
 965
 966                 /*
 967                  * Grab an empty transaction so that we can use its recursive
 968                  * buffer locking abilities to detect cycles in the rmapbt
 969                  * without deadlocking.
 970                  */
 971                 error = xfs_trans_alloc_empty(mp, &tp);
 972                 if (error)
 973                         break;
 974
 975                 info.dev = handlers[i].dev;
 976                 info.last = false;
 977                 info.pag = NULL;
 978                 info.low_daddr = -1ULL;
 979                 info.low.rm_blockcount = 0;
 980                 error = handlers[i].fn(tp, dkeys, &info);
 981                 if (error)
 982                         break;
 983                 xfs_trans_cancel(tp);
 984                 tp = NULL;
 985                 info.next_daddr = 0;
 986         }
 987
 988         if (tp)
 989                 xfs_trans_cancel(tp);
 990         head->fmh_oflags = FMH_OF_DEV_T;
 991         return error;
 992 }