fs/ext4/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 "ext4.h"
  21 #include <linux/fsmap.h>
  22 #include "fsmap.h"
  23 #include "mballoc.h"
  24 #include <linux/sort.h>
  25 #include <linux/list_sort.h>
  26 #include <trace/events/ext4.h>
  27
  28 /* Convert an ext4_fsmap to an fsmap. */
  29 void ext4_fsmap_from_internal(struct super_block *sb, struct fsmap *dest,
  30                               struct ext4_fsmap *src)
  31 {
  32         dest->fmr_device = src->fmr_device;
  33         dest->fmr_flags = src->fmr_flags;
  34         dest->fmr_physical = src->fmr_physical << sb->s_blocksize_bits;
  35         dest->fmr_owner = src->fmr_owner;
  36         dest->fmr_offset = 0;
  37         dest->fmr_length = src->fmr_length << sb->s_blocksize_bits;
  38         dest->fmr_reserved[0] = 0;
  39         dest->fmr_reserved[1] = 0;
  40         dest->fmr_reserved[2] = 0;
  41 }
  42
  43 /* Convert an fsmap to an ext4_fsmap. */
  44 void ext4_fsmap_to_internal(struct super_block *sb, struct ext4_fsmap *dest,
  45                             struct fsmap *src)
  46 {
  47         dest->fmr_device = src->fmr_device;
  48         dest->fmr_flags = src->fmr_flags;
  49         dest->fmr_physical = src->fmr_physical >> sb->s_blocksize_bits;
  50         dest->fmr_owner = src->fmr_owner;
  51         dest->fmr_length = src->fmr_length >> sb->s_blocksize_bits;
  52 }
  53
  54 /* getfsmap query state */
  55 struct ext4_getfsmap_info {
  56         struct ext4_fsmap_head  *gfi_head;
  57         ext4_fsmap_format_t     gfi_formatter;  /* formatting fn */
  58         void                    *gfi_format_arg;/* format buffer */
  59         ext4_fsblk_t            gfi_next_fsblk; /* next fsblock we expect */
  60         u32                     gfi_dev;        /* device id */
  61         ext4_group_t            gfi_agno;       /* bg number, if applicable */
  62         struct ext4_fsmap       gfi_low;        /* low rmap key */
  63         struct ext4_fsmap       gfi_high;       /* high rmap key */
  64         struct ext4_fsmap       gfi_lastfree;   /* free ext at end of last bg */
  65         struct list_head        gfi_meta_list;  /* fixed metadata list */
  66         bool                    gfi_last;       /* last extent? */
  67 };
  68
  69 /* Associate a device with a getfsmap handler. */
  70 struct ext4_getfsmap_dev {
  71         int                     (*gfd_fn)(struct super_block *sb,
  72                                       struct ext4_fsmap *keys,
  73                                       struct ext4_getfsmap_info *info);
  74         u32                     gfd_dev;
  75 };
  76
  77 /* Compare two getfsmap device handlers. */
  78 static int ext4_getfsmap_dev_compare(const void *p1, const void *p2)
  79 {
  80         const struct ext4_getfsmap_dev *d1 = p1;
  81         const struct ext4_getfsmap_dev *d2 = p2;
  82
  83         return d1->gfd_dev - d2->gfd_dev;
  84 }
  85
  86 /* Compare a record against our starting point */
  87 static bool ext4_getfsmap_rec_before_low_key(struct ext4_getfsmap_info *info,
  88                                              struct ext4_fsmap *rec)
  89 {
  90         return rec->fmr_physical < info->gfi_low.fmr_physical;
  91 }
  92
  93 /*
  94  * Format a reverse mapping for getfsmap, having translated rm_startblock
  95  * into the appropriate daddr units.
  96  */
  97 static int ext4_getfsmap_helper(struct super_block *sb,
  98                                 struct ext4_getfsmap_info *info,
  99                                 struct ext4_fsmap *rec)
 100 {
 101         struct ext4_fsmap fmr;
 102         struct ext4_sb_info *sbi = EXT4_SB(sb);
 103         ext4_fsblk_t rec_fsblk = rec->fmr_physical;
 104         ext4_group_t agno;
 105         ext4_grpblk_t cno;
 106         int error;
 107
 108         if (fatal_signal_pending(current))
 109                 return -EINTR;
 110
 111         /*
 112          * Filter out records that start before our startpoint, if the
 113          * caller requested that.
 114          */
 115         if (ext4_getfsmap_rec_before_low_key(info, rec)) {
 116                 rec_fsblk += rec->fmr_length;
 117                 if (info->gfi_next_fsblk < rec_fsblk)
 118                         info->gfi_next_fsblk = rec_fsblk;
 119                 return EXT4_QUERY_RANGE_CONTINUE;
 120         }
 121
 122         /* Are we just counting mappings? */
 123         if (info->gfi_head->fmh_count == 0) {
 124                 if (info->gfi_head->fmh_entries == UINT_MAX)
 125                         return EXT4_QUERY_RANGE_ABORT;
 126
 127                 if (rec_fsblk > info->gfi_next_fsblk)
 128                         info->gfi_head->fmh_entries++;
 129
 130                 if (info->gfi_last)
 131                         return EXT4_QUERY_RANGE_CONTINUE;
 132
 133                 info->gfi_head->fmh_entries++;
 134
 135                 rec_fsblk += rec->fmr_length;
 136                 if (info->gfi_next_fsblk < rec_fsblk)
 137                         info->gfi_next_fsblk = rec_fsblk;
 138                 return EXT4_QUERY_RANGE_CONTINUE;
 139         }
 140
 141         /*
 142          * If the record starts past the last physical block we saw,
 143          * then we've found a gap.  Report the gap as being owned by
 144          * whatever the caller specified is the missing owner.
 145          */
 146         if (rec_fsblk > info->gfi_next_fsblk) {
 147                 if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
 148                         return EXT4_QUERY_RANGE_ABORT;
 149
 150                 ext4_get_group_no_and_offset(sb, info->gfi_next_fsblk,
 151                                 &agno, &cno);
 152                 trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno,
 153                                 EXT4_C2B(sbi, cno),
 154                                 rec_fsblk - info->gfi_next_fsblk,
 155                                 EXT4_FMR_OWN_UNKNOWN);
 156
 157                 fmr.fmr_device = info->gfi_dev;
 158                 fmr.fmr_physical = info->gfi_next_fsblk;
 159                 fmr.fmr_owner = EXT4_FMR_OWN_UNKNOWN;
 160                 fmr.fmr_length = rec_fsblk - info->gfi_next_fsblk;
 161                 fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
 162                 error = info->gfi_formatter(&fmr, info->gfi_format_arg);
 163                 if (error)
 164                         return error;
 165                 info->gfi_head->fmh_entries++;
 166         }
 167
 168         if (info->gfi_last)
 169                 goto out;
 170
 171         /* Fill out the extent we found */
 172         if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
 173                 return EXT4_QUERY_RANGE_ABORT;
 174
 175         ext4_get_group_no_and_offset(sb, rec_fsblk, &agno, &cno);
 176         trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno, EXT4_C2B(sbi, cno),
 177                         rec->fmr_length, rec->fmr_owner);
 178
 179         fmr.fmr_device = info->gfi_dev;
 180         fmr.fmr_physical = rec_fsblk;
 181         fmr.fmr_owner = rec->fmr_owner;
 182         fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
 183         fmr.fmr_length = rec->fmr_length;
 184         error = info->gfi_formatter(&fmr, info->gfi_format_arg);
 185         if (error)
 186                 return error;
 187         info->gfi_head->fmh_entries++;
 188
 189 out:
 190         rec_fsblk += rec->fmr_length;
 191         if (info->gfi_next_fsblk < rec_fsblk)
 192                 info->gfi_next_fsblk = rec_fsblk;
 193         return EXT4_QUERY_RANGE_CONTINUE;
 194 }
 195
 196 static inline ext4_fsblk_t ext4_fsmap_next_pblk(struct ext4_fsmap *fmr)
 197 {
 198         return fmr->fmr_physical + fmr->fmr_length;
 199 }
 200
 201 /* Transform a blockgroup's free record into a fsmap */
 202 static int ext4_getfsmap_datadev_helper(struct super_block *sb,
 203                                         ext4_group_t agno, ext4_grpblk_t start,
 204                                         ext4_grpblk_t len, void *priv)
 205 {
 206         struct ext4_fsmap irec;
 207         struct ext4_getfsmap_info *info = priv;
 208         struct ext4_fsmap *p;
 209         struct ext4_fsmap *tmp;
 210         struct ext4_sb_info *sbi = EXT4_SB(sb);
 211         ext4_fsblk_t fsb;
 212         ext4_fsblk_t fslen;
 213         int error;
 214
 215         fsb = (EXT4_C2B(sbi, start) + ext4_group_first_block_no(sb, agno));
 216         fslen = EXT4_C2B(sbi, len);
 217
 218         /* If the retained free extent record is set... */
 219         if (info->gfi_lastfree.fmr_owner) {
 220                 /* ...and abuts this one, lengthen it and return. */
 221                 if (ext4_fsmap_next_pblk(&info->gfi_lastfree) == fsb) {
 222                         info->gfi_lastfree.fmr_length += fslen;
 223                         return 0;
 224                 }
 225
 226                 /*
 227                  * There's a gap between the two free extents; emit the
 228                  * retained extent prior to merging the meta_list.
 229                  */
 230                 error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
 231                 if (error)
 232                         return error;
 233                 info->gfi_lastfree.fmr_owner = 0;
 234         }
 235
 236         /* Merge in any relevant extents from the meta_list */
 237         list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
 238                 if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
 239                         list_del(&p->fmr_list);
 240                         kfree(p);
 241                 } else if (p->fmr_physical < fsb) {
 242                         error = ext4_getfsmap_helper(sb, info, p);
 243                         if (error)
 244                                 return error;
 245
 246                         list_del(&p->fmr_list);
 247                         kfree(p);
 248                 }
 249         }
 250
 251         irec.fmr_device = 0;
 252         irec.fmr_physical = fsb;
 253         irec.fmr_length = fslen;
 254         irec.fmr_owner = EXT4_FMR_OWN_FREE;
 255         irec.fmr_flags = 0;
 256
 257         /* If this is a free extent at the end of a bg, buffer it. */
 258         if (ext4_fsmap_next_pblk(&irec) ==
 259                         ext4_group_first_block_no(sb, agno + 1)) {
 260                 info->gfi_lastfree = irec;
 261                 return 0;
 262         }
 263
 264         /* Otherwise, emit it */
 265         return ext4_getfsmap_helper(sb, info, &irec);
 266 }
 267
 268 /* Execute a getfsmap query against the log device. */
 269 static int ext4_getfsmap_logdev(struct super_block *sb, struct ext4_fsmap *keys,
 270                                 struct ext4_getfsmap_info *info)
 271 {
 272         journal_t *journal = EXT4_SB(sb)->s_journal;
 273         struct ext4_fsmap irec;
 274
 275         /* Set up search keys */
 276         info->gfi_low = keys[0];
 277         info->gfi_low.fmr_length = 0;
 278
 279         memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
 280
 281         trace_ext4_fsmap_low_key(sb, info->gfi_dev, 0,
 282                         info->gfi_low.fmr_physical,
 283                         info->gfi_low.fmr_length,
 284                         info->gfi_low.fmr_owner);
 285
 286         trace_ext4_fsmap_high_key(sb, info->gfi_dev, 0,
 287                         info->gfi_high.fmr_physical,
 288                         info->gfi_high.fmr_length,
 289                         info->gfi_high.fmr_owner);
 290
 291         if (keys[0].fmr_physical > 0)
 292                 return 0;
 293
 294         /* Fabricate an rmap entry for the external log device. */
 295         irec.fmr_physical = journal->j_blk_offset;
 296         irec.fmr_length = journal->j_maxlen;
 297         irec.fmr_owner = EXT4_FMR_OWN_LOG;
 298         irec.fmr_flags = 0;
 299
 300         return ext4_getfsmap_helper(sb, info, &irec);
 301 }
 302
 303 /* Helper to fill out an ext4_fsmap. */
 304 static inline int ext4_getfsmap_fill(struct list_head *meta_list,
 305                                      ext4_fsblk_t fsb, ext4_fsblk_t len,
 306                                      uint64_t owner)
 307 {
 308         struct ext4_fsmap *fsm;
 309
 310         fsm = kmalloc(sizeof(*fsm), GFP_NOFS);
 311         if (!fsm)
 312                 return -ENOMEM;
 313         fsm->fmr_device = 0;
 314         fsm->fmr_flags = 0;
 315         fsm->fmr_physical = fsb;
 316         fsm->fmr_owner = owner;
 317         fsm->fmr_length = len;
 318         list_add_tail(&fsm->fmr_list, meta_list);
 319
 320         return 0;
 321 }
 322
 323 /*
 324  * This function returns the number of file system metadata blocks at
 325  * the beginning of a block group, including the reserved gdt blocks.
 326  */
 327 static unsigned int ext4_getfsmap_find_sb(struct super_block *sb,
 328                                           ext4_group_t agno,
 329                                           struct list_head *meta_list)
 330 {
 331         struct ext4_sb_info *sbi = EXT4_SB(sb);
 332         ext4_fsblk_t fsb = ext4_group_first_block_no(sb, agno);
 333         ext4_fsblk_t len;
 334         unsigned long first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
 335         unsigned long metagroup = agno / EXT4_DESC_PER_BLOCK(sb);
 336         int error;
 337
 338         /* Record the superblock. */
 339         if (ext4_bg_has_super(sb, agno)) {
 340                 error = ext4_getfsmap_fill(meta_list, fsb, 1, EXT4_FMR_OWN_FS);
 341                 if (error)
 342                         return error;
 343                 fsb++;
 344         }
 345
 346         /* Record the group descriptors. */
 347         len = ext4_bg_num_gdb(sb, agno);
 348         if (!len)
 349                 return 0;
 350         error = ext4_getfsmap_fill(meta_list, fsb, len,
 351                                    EXT4_FMR_OWN_GDT);
 352         if (error)
 353                 return error;
 354         fsb += len;
 355
 356         /* Reserved GDT blocks */
 357         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) {
 358                 len = le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 359                 error = ext4_getfsmap_fill(meta_list, fsb, len,
 360                                            EXT4_FMR_OWN_RESV_GDT);
 361                 if (error)
 362                         return error;
 363         }
 364
 365         return 0;
 366 }
 367
 368 /* Compare two fsmap items. */
 369 static int ext4_getfsmap_compare(void *priv,
 370                                  struct list_head *a,
 371                                  struct list_head *b)
 372 {
 373         struct ext4_fsmap *fa;
 374         struct ext4_fsmap *fb;
 375
 376         fa = container_of(a, struct ext4_fsmap, fmr_list);
 377         fb = container_of(b, struct ext4_fsmap, fmr_list);
 378         if (fa->fmr_physical < fb->fmr_physical)
 379                 return -1;
 380         else if (fa->fmr_physical > fb->fmr_physical)
 381                 return 1;
 382         return 0;
 383 }
 384
 385 /* Merge adjacent extents of fixed metadata. */
 386 static void ext4_getfsmap_merge_fixed_metadata(struct list_head *meta_list)
 387 {
 388         struct ext4_fsmap *p;
 389         struct ext4_fsmap *prev = NULL;
 390         struct ext4_fsmap *tmp;
 391
 392         list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
 393                 if (!prev) {
 394                         prev = p;
 395                         continue;
 396                 }
 397
 398                 if (prev->fmr_owner == p->fmr_owner &&
 399                     prev->fmr_physical + prev->fmr_length == p->fmr_physical) {
 400                         prev->fmr_length += p->fmr_length;
 401                         list_del(&p->fmr_list);
 402                         kfree(p);
 403                 } else
 404                         prev = p;
 405         }
 406 }
 407
 408 /* Free a list of fixed metadata. */
 409 static void ext4_getfsmap_free_fixed_metadata(struct list_head *meta_list)
 410 {
 411         struct ext4_fsmap *p;
 412         struct ext4_fsmap *tmp;
 413
 414         list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
 415                 list_del(&p->fmr_list);
 416                 kfree(p);
 417         }
 418 }
 419
 420 /* Find all the fixed metadata in the filesystem. */
 421 int ext4_getfsmap_find_fixed_metadata(struct super_block *sb,
 422                                       struct list_head *meta_list)
 423 {
 424         struct ext4_group_desc *gdp;
 425         ext4_group_t agno;
 426         int error;
 427
 428         INIT_LIST_HEAD(meta_list);
 429
 430         /* Collect everything. */
 431         for (agno = 0; agno < EXT4_SB(sb)->s_groups_count; agno++) {
 432                 gdp = ext4_get_group_desc(sb, agno, NULL);
 433                 if (!gdp) {
 434                         error = -EFSCORRUPTED;
 435                         goto err;
 436                 }
 437
 438                 /* Superblock & GDT */
 439                 error = ext4_getfsmap_find_sb(sb, agno, meta_list);
 440                 if (error)
 441                         goto err;
 442
 443                 /* Block bitmap */
 444                 error = ext4_getfsmap_fill(meta_list,
 445                                            ext4_block_bitmap(sb, gdp), 1,
 446                                            EXT4_FMR_OWN_BLKBM);
 447                 if (error)
 448                         goto err;
 449
 450                 /* Inode bitmap */
 451                 error = ext4_getfsmap_fill(meta_list,
 452                                            ext4_inode_bitmap(sb, gdp), 1,
 453                                            EXT4_FMR_OWN_INOBM);
 454                 if (error)
 455                         goto err;
 456
 457                 /* Inodes */
 458                 error = ext4_getfsmap_fill(meta_list,
 459                                            ext4_inode_table(sb, gdp),
 460                                            EXT4_SB(sb)->s_itb_per_group,
 461                                            EXT4_FMR_OWN_INODES);
 462                 if (error)
 463                         goto err;
 464         }
 465
 466         /* Sort the list */
 467         list_sort(NULL, meta_list, ext4_getfsmap_compare);
 468
 469         /* Merge adjacent extents */
 470         ext4_getfsmap_merge_fixed_metadata(meta_list);
 471
 472         return 0;
 473 err:
 474         ext4_getfsmap_free_fixed_metadata(meta_list);
 475         return error;
 476 }
 477
 478 /* Execute a getfsmap query against the buddy bitmaps */
 479 static int ext4_getfsmap_datadev(struct super_block *sb,
 480                                  struct ext4_fsmap *keys,
 481                                  struct ext4_getfsmap_info *info)
 482 {
 483         struct ext4_sb_info *sbi = EXT4_SB(sb);
 484         ext4_fsblk_t start_fsb;
 485         ext4_fsblk_t end_fsb;
 486         ext4_fsblk_t bofs;
 487         ext4_fsblk_t eofs;
 488         ext4_group_t start_ag;
 489         ext4_group_t end_ag;
 490         ext4_grpblk_t first_cluster;
 491         ext4_grpblk_t last_cluster;
 492         int error = 0;
 493
 494         bofs = le32_to_cpu(sbi->s_es->s_first_data_block);
 495         eofs = ext4_blocks_count(sbi->s_es);
 496         if (keys[0].fmr_physical >= eofs)
 497                 return 0;
 498         else if (keys[0].fmr_physical < bofs)
 499                 keys[0].fmr_physical = bofs;
 500         if (keys[1].fmr_physical >= eofs)
 501                 keys[1].fmr_physical = eofs - 1;
 502         start_fsb = keys[0].fmr_physical;
 503         end_fsb = keys[1].fmr_physical;
 504
 505         /* Determine first and last group to examine based on start and end */
 506         ext4_get_group_no_and_offset(sb, start_fsb, &start_ag, &first_cluster);
 507         ext4_get_group_no_and_offset(sb, end_fsb, &end_ag, &last_cluster);
 508
 509         /*
 510          * Convert the fsmap low/high keys to bg based keys.  Initialize
 511          * low to the fsmap low key and max out the high key to the end
 512          * of the bg.
 513          */
 514         info->gfi_low = keys[0];
 515         info->gfi_low.fmr_physical = EXT4_C2B(sbi, first_cluster);
 516         info->gfi_low.fmr_length = 0;
 517
 518         memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
 519
 520         /* Assemble a list of all the fixed-location metadata. */
 521         error = ext4_getfsmap_find_fixed_metadata(sb, &info->gfi_meta_list);
 522         if (error)
 523                 goto err;
 524
 525         /* Query each bg */
 526         for (info->gfi_agno = start_ag;
 527              info->gfi_agno <= end_ag;
 528              info->gfi_agno++) {
 529                 /*
 530                  * Set the bg high key from the fsmap high key if this
 531                  * is the last bg that we're querying.
 532                  */
 533                 if (info->gfi_agno == end_ag) {
 534                         info->gfi_high = keys[1];
 535                         info->gfi_high.fmr_physical = EXT4_C2B(sbi,
 536                                         last_cluster);
 537                         info->gfi_high.fmr_length = 0;
 538                 }
 539
 540                 trace_ext4_fsmap_low_key(sb, info->gfi_dev, info->gfi_agno,
 541                                 info->gfi_low.fmr_physical,
 542                                 info->gfi_low.fmr_length,
 543                                 info->gfi_low.fmr_owner);
 544
 545                 trace_ext4_fsmap_high_key(sb, info->gfi_dev, info->gfi_agno,
 546                                 info->gfi_high.fmr_physical,
 547                                 info->gfi_high.fmr_length,
 548                                 info->gfi_high.fmr_owner);
 549
 550                 error = ext4_mballoc_query_range(sb, info->gfi_agno,
 551                                 EXT4_B2C(sbi, info->gfi_low.fmr_physical),
 552                                 EXT4_B2C(sbi, info->gfi_high.fmr_physical),
 553                                 ext4_getfsmap_datadev_helper, info);
 554                 if (error)
 555                         goto err;
 556
 557                 /*
 558                  * Set the bg low key to the start of the bg prior to
 559                  * moving on to the next bg.
 560                  */
 561                 if (info->gfi_agno == start_ag)
 562                         memset(&info->gfi_low, 0, sizeof(info->gfi_low));
 563         }
 564
 565         /* Do we have a retained free extent? */
 566         if (info->gfi_lastfree.fmr_owner) {
 567                 error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
 568                 if (error)
 569                         goto err;
 570         }
 571
 572         /* Report any gaps at the end of the bg */
 573         info->gfi_last = true;
 574         error = ext4_getfsmap_datadev_helper(sb, end_ag, last_cluster, 0, info);
 575         if (error)
 576                 goto err;
 577
 578 err:
 579         ext4_getfsmap_free_fixed_metadata(&info->gfi_meta_list);
 580         return error;
 581 }
 582
 583 /* Do we recognize the device? */
 584 static bool ext4_getfsmap_is_valid_device(struct super_block *sb,
 585                                           struct ext4_fsmap *fm)
 586 {
 587         if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
 588             fm->fmr_device == new_encode_dev(sb->s_bdev->bd_dev))
 589                 return true;
 590         if (EXT4_SB(sb)->journal_bdev &&
 591             fm->fmr_device == new_encode_dev(EXT4_SB(sb)->journal_bdev->bd_dev))
 592                 return true;
 593         return false;
 594 }
 595
 596 /* Ensure that the low key is less than the high key. */
 597 static bool ext4_getfsmap_check_keys(struct ext4_fsmap *low_key,
 598                                      struct ext4_fsmap *high_key)
 599 {
 600         if (low_key->fmr_device > high_key->fmr_device)
 601                 return false;
 602         if (low_key->fmr_device < high_key->fmr_device)
 603                 return true;
 604
 605         if (low_key->fmr_physical > high_key->fmr_physical)
 606                 return false;
 607         if (low_key->fmr_physical < high_key->fmr_physical)
 608                 return true;
 609
 610         if (low_key->fmr_owner > high_key->fmr_owner)
 611                 return false;
 612         if (low_key->fmr_owner < high_key->fmr_owner)
 613                 return true;
 614
 615         return false;
 616 }
 617
 618 #define EXT4_GETFSMAP_DEVS      2
 619 /*
 620  * Get filesystem's extents as described in head, and format for
 621  * output.  Calls formatter to fill the user's buffer until all
 622  * extents are mapped, until the passed-in head->fmh_count slots have
 623  * been filled, or until the formatter short-circuits the loop, if it
 624  * is tracking filled-in extents on its own.
 625  *
 626  * Key to Confusion
 627  * ----------------
 628  * There are multiple levels of keys and counters at work here:
 629  * _fsmap_head.fmh_keys         -- low and high fsmap keys passed in;
 630  *                                 these reflect fs-wide block addrs.
 631  * dkeys                        -- fmh_keys used to query each device;
 632  *                                 these are fmh_keys but w/ the low key
 633  *                                 bumped up by fmr_length.
 634  * _getfsmap_info.gfi_next_fsblk-- next fs block we expect to see; this
 635  *                                 is how we detect gaps in the fsmap
 636  *                                 records and report them.
 637  * _getfsmap_info.gfi_low/high  -- per-bg low/high keys computed from
 638  *                                 dkeys; used to query the free space.
 639  */
 640 int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
 641                   ext4_fsmap_format_t formatter, void *arg)
 642 {
 643         struct ext4_fsmap dkeys[2];     /* per-dev keys */
 644         struct ext4_getfsmap_dev handlers[EXT4_GETFSMAP_DEVS];
 645         struct ext4_getfsmap_info info = {0};
 646         int i;
 647         int error = 0;
 648
 649         if (head->fmh_iflags & ~FMH_IF_VALID)
 650                 return -EINVAL;
 651         if (!ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[0]) ||
 652             !ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[1]))
 653                 return -EINVAL;
 654
 655         head->fmh_entries = 0;
 656
 657         /* Set up our device handlers. */
 658         memset(handlers, 0, sizeof(handlers));
 659         handlers[0].gfd_dev = new_encode_dev(sb->s_bdev->bd_dev);
 660         handlers[0].gfd_fn = ext4_getfsmap_datadev;
 661         if (EXT4_SB(sb)->journal_bdev) {
 662                 handlers[1].gfd_dev = new_encode_dev(
 663                                 EXT4_SB(sb)->journal_bdev->bd_dev);
 664                 handlers[1].gfd_fn = ext4_getfsmap_logdev;
 665         }
 666
 667         sort(handlers, EXT4_GETFSMAP_DEVS, sizeof(struct ext4_getfsmap_dev),
 668                         ext4_getfsmap_dev_compare, NULL);
 669
 670         /*
 671          * To continue where we left off, we allow userspace to use the
 672          * last mapping from a previous call as the low key of the next.
 673          * This is identified by a non-zero length in the low key. We
 674          * have to increment the low key in this scenario to ensure we
 675          * don't return the same mapping again, and instead return the
 676          * very next mapping.
 677          *
 678          * Bump the physical offset as there can be no other mapping for
 679          * the same physical block range.
 680          */
 681         dkeys[0] = head->fmh_keys[0];
 682         dkeys[0].fmr_physical += dkeys[0].fmr_length;
 683         dkeys[0].fmr_owner = 0;
 684         dkeys[0].fmr_length = 0;
 685         memset(&dkeys[1], 0xFF, sizeof(struct ext4_fsmap));
 686
 687         if (!ext4_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
 688                 return -EINVAL;
 689
 690         info.gfi_next_fsblk = head->fmh_keys[0].fmr_physical +
 691                           head->fmh_keys[0].fmr_length;
 692         info.gfi_formatter = formatter;
 693         info.gfi_format_arg = arg;
 694         info.gfi_head = head;
 695
 696         /* For each device we support... */
 697         for (i = 0; i < EXT4_GETFSMAP_DEVS; i++) {
 698                 /* Is this device within the range the user asked for? */
 699                 if (!handlers[i].gfd_fn)
 700                         continue;
 701                 if (head->fmh_keys[0].fmr_device > handlers[i].gfd_dev)
 702                         continue;
 703                 if (head->fmh_keys[1].fmr_device < handlers[i].gfd_dev)
 704                         break;
 705
 706                 /*
 707                  * If this device number matches the high key, we have
 708                  * to pass the high key to the handler to limit the
 709                  * query results.  If the device number exceeds the
 710                  * low key, zero out the low key so that we get
 711                  * everything from the beginning.
 712                  */
 713                 if (handlers[i].gfd_dev == head->fmh_keys[1].fmr_device)
 714                         dkeys[1] = head->fmh_keys[1];
 715                 if (handlers[i].gfd_dev > head->fmh_keys[0].fmr_device)
 716                         memset(&dkeys[0], 0, sizeof(struct ext4_fsmap));
 717
 718                 info.gfi_dev = handlers[i].gfd_dev;
 719                 info.gfi_last = false;
 720                 info.gfi_agno = -1;
 721                 error = handlers[i].gfd_fn(sb, dkeys, &info);
 722                 if (error)
 723                         break;
 724                 info.gfi_next_fsblk = 0;
 725         }
 726
 727         head->fmh_oflags = FMH_OF_DEV_T;
 728         return error;
 729 }