fs/xfs/xfs_attr_remote.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   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 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_shared.h"
  22 #include "xfs_format.h"
  23 #include "xfs_log_format.h"
  24 #include "xfs_trans_resv.h"
  25 #include "xfs_bit.h"
  26 #include "xfs_sb.h"
  27 #include "xfs_ag.h"
  28 #include "xfs_mount.h"
  29 #include "xfs_da_format.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_alloc.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_inode_item.h"
  35 #include "xfs_bmap.h"
  36 #include "xfs_bmap_util.h"
  37 #include "xfs_attr.h"
  38 #include "xfs_attr_leaf.h"
  39 #include "xfs_attr_remote.h"
  40 #include "xfs_trans_space.h"
  41 #include "xfs_trace.h"
  42 #include "xfs_cksum.h"
  43 #include "xfs_buf_item.h"
  44 #include "xfs_error.h"
  45
  46 #define ATTR_RMTVALUE_MAPSIZE   1       /* # of map entries at once */
  47
  48 /*
  49  * Each contiguous block has a header, so it is not just a simple attribute
  50  * length to FSB conversion.
  51  */
  52 int
  53 xfs_attr3_rmt_blocks(
  54         struct xfs_mount *mp,
  55         int             attrlen)
  56 {
  57         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  58                 int buflen = XFS_ATTR3_RMT_BUF_SPACE(mp, mp->m_sb.sb_blocksize);
  59                 return (attrlen + buflen - 1) / buflen;
  60         }
  61         return XFS_B_TO_FSB(mp, attrlen);
  62 }
  63
  64 /*
  65  * Checking of the remote attribute header is split into two parts. The verifier
  66  * does CRC, location and bounds checking, the unpacking function checks the
  67  * attribute parameters and owner.
  68  */
  69 static bool
  70 xfs_attr3_rmt_hdr_ok(
  71         struct xfs_mount        *mp,
  72         void                    *ptr,
  73         xfs_ino_t               ino,
  74         uint32_t                offset,
  75         uint32_t                size,
  76         xfs_daddr_t             bno)
  77 {
  78         struct xfs_attr3_rmt_hdr *rmt = ptr;
  79
  80         if (bno != be64_to_cpu(rmt->rm_blkno))
  81                 return false;
  82         if (offset != be32_to_cpu(rmt->rm_offset))
  83                 return false;
  84         if (size != be32_to_cpu(rmt->rm_bytes))
  85                 return false;
  86         if (ino != be64_to_cpu(rmt->rm_owner))
  87                 return false;
  88
  89         /* ok */
  90         return true;
  91 }
  92
  93 static bool
  94 xfs_attr3_rmt_verify(
  95         struct xfs_mount        *mp,
  96         void                    *ptr,
  97         int                     fsbsize,
  98         xfs_daddr_t             bno)
  99 {
 100         struct xfs_attr3_rmt_hdr *rmt = ptr;
 101
 102         if (!xfs_sb_version_hascrc(&mp->m_sb))
 103                 return false;
 104         if (rmt->rm_magic != cpu_to_be32(XFS_ATTR3_RMT_MAGIC))
 105                 return false;
 106         if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_uuid))
 107                 return false;
 108         if (be64_to_cpu(rmt->rm_blkno) != bno)
 109                 return false;
 110         if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
 111                 return false;
 112         if (be32_to_cpu(rmt->rm_offset) +
 113                                 be32_to_cpu(rmt->rm_bytes) > XATTR_SIZE_MAX)
 114                 return false;
 115         if (rmt->rm_owner == 0)
 116                 return false;
 117
 118         return true;
 119 }
 120
 121 static void
 122 xfs_attr3_rmt_read_verify(
 123         struct xfs_buf  *bp)
 124 {
 125         struct xfs_mount *mp = bp->b_target->bt_mount;
 126         char            *ptr;
 127         int             len;
 128         bool            corrupt = false;
 129         xfs_daddr_t     bno;
 130
 131         /* no verification of non-crc buffers */
 132         if (!xfs_sb_version_hascrc(&mp->m_sb))
 133                 return;
 134
 135         ptr = bp->b_addr;
 136         bno = bp->b_bn;
 137         len = BBTOB(bp->b_length);
 138         ASSERT(len >= XFS_LBSIZE(mp));
 139
 140         while (len > 0) {
 141                 if (!xfs_verify_cksum(ptr, XFS_LBSIZE(mp),
 142                                       XFS_ATTR3_RMT_CRC_OFF)) {
 143                         corrupt = true;
 144                         break;
 145                 }
 146                 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
 147                         corrupt = true;
 148                         break;
 149                 }
 150                 len -= XFS_LBSIZE(mp);
 151                 ptr += XFS_LBSIZE(mp);
 152                 bno += mp->m_bsize;
 153         }
 154
 155         if (corrupt) {
 156                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 157                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 158         } else
 159                 ASSERT(len == 0);
 160 }
 161
 162 static void
 163 xfs_attr3_rmt_write_verify(
 164         struct xfs_buf  *bp)
 165 {
 166         struct xfs_mount *mp = bp->b_target->bt_mount;
 167         struct xfs_buf_log_item *bip = bp->b_fspriv;
 168         char            *ptr;
 169         int             len;
 170         xfs_daddr_t     bno;
 171
 172         /* no verification of non-crc buffers */
 173         if (!xfs_sb_version_hascrc(&mp->m_sb))
 174                 return;
 175
 176         ptr = bp->b_addr;
 177         bno = bp->b_bn;
 178         len = BBTOB(bp->b_length);
 179         ASSERT(len >= XFS_LBSIZE(mp));
 180
 181         while (len > 0) {
 182                 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
 183                         XFS_CORRUPTION_ERROR(__func__,
 184                                             XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 185                         xfs_buf_ioerror(bp, EFSCORRUPTED);
 186                         return;
 187                 }
 188                 if (bip) {
 189                         struct xfs_attr3_rmt_hdr *rmt;
 190
 191                         rmt = (struct xfs_attr3_rmt_hdr *)ptr;
 192                         rmt->rm_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 193                 }
 194                 xfs_update_cksum(ptr, XFS_LBSIZE(mp), XFS_ATTR3_RMT_CRC_OFF);
 195
 196                 len -= XFS_LBSIZE(mp);
 197                 ptr += XFS_LBSIZE(mp);
 198                 bno += mp->m_bsize;
 199         }
 200         ASSERT(len == 0);
 201 }
 202
 203 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
 204         .verify_read = xfs_attr3_rmt_read_verify,
 205         .verify_write = xfs_attr3_rmt_write_verify,
 206 };
 207
 208 STATIC int
 209 xfs_attr3_rmt_hdr_set(
 210         struct xfs_mount        *mp,
 211         void                    *ptr,
 212         xfs_ino_t               ino,
 213         uint32_t                offset,
 214         uint32_t                size,
 215         xfs_daddr_t             bno)
 216 {
 217         struct xfs_attr3_rmt_hdr *rmt = ptr;
 218
 219         if (!xfs_sb_version_hascrc(&mp->m_sb))
 220                 return 0;
 221
 222         rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
 223         rmt->rm_offset = cpu_to_be32(offset);
 224         rmt->rm_bytes = cpu_to_be32(size);
 225         uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_uuid);
 226         rmt->rm_owner = cpu_to_be64(ino);
 227         rmt->rm_blkno = cpu_to_be64(bno);
 228
 229         return sizeof(struct xfs_attr3_rmt_hdr);
 230 }
 231
 232 /*
 233  * Helper functions to copy attribute data in and out of the one disk extents
 234  */
 235 STATIC int
 236 xfs_attr_rmtval_copyout(
 237         struct xfs_mount *mp,
 238         struct xfs_buf  *bp,
 239         xfs_ino_t       ino,
 240         int             *offset,
 241         int             *valuelen,
 242         __uint8_t       **dst)
 243 {
 244         char            *src = bp->b_addr;
 245         xfs_daddr_t     bno = bp->b_bn;
 246         int             len = BBTOB(bp->b_length);
 247
 248         ASSERT(len >= XFS_LBSIZE(mp));
 249
 250         while (len > 0 && *valuelen > 0) {
 251                 int hdr_size = 0;
 252                 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
 253
 254                 byte_cnt = min(*valuelen, byte_cnt);
 255
 256                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 257                         if (!xfs_attr3_rmt_hdr_ok(mp, src, ino, *offset,
 258                                                   byte_cnt, bno)) {
 259                                 xfs_alert(mp,
 260 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
 261                                         bno, *offset, byte_cnt, ino);
 262                                 return EFSCORRUPTED;
 263                         }
 264                         hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
 265                 }
 266
 267                 memcpy(*dst, src + hdr_size, byte_cnt);
 268
 269                 /* roll buffer forwards */
 270                 len -= XFS_LBSIZE(mp);
 271                 src += XFS_LBSIZE(mp);
 272                 bno += mp->m_bsize;
 273
 274                 /* roll attribute data forwards */
 275                 *valuelen -= byte_cnt;
 276                 *dst += byte_cnt;
 277                 *offset += byte_cnt;
 278         }
 279         return 0;
 280 }
 281
 282 STATIC void
 283 xfs_attr_rmtval_copyin(
 284         struct xfs_mount *mp,
 285         struct xfs_buf  *bp,
 286         xfs_ino_t       ino,
 287         int             *offset,
 288         int             *valuelen,
 289         __uint8_t       **src)
 290 {
 291         char            *dst = bp->b_addr;
 292         xfs_daddr_t     bno = bp->b_bn;
 293         int             len = BBTOB(bp->b_length);
 294
 295         ASSERT(len >= XFS_LBSIZE(mp));
 296
 297         while (len > 0 && *valuelen > 0) {
 298                 int hdr_size;
 299                 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
 300
 301                 byte_cnt = min(*valuelen, byte_cnt);
 302                 hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
 303                                                  byte_cnt, bno);
 304
 305                 memcpy(dst + hdr_size, *src, byte_cnt);
 306
 307                 /*
 308                  * If this is the last block, zero the remainder of it.
 309                  * Check that we are actually the last block, too.
 310                  */
 311                 if (byte_cnt + hdr_size < XFS_LBSIZE(mp)) {
 312                         ASSERT(*valuelen - byte_cnt == 0);
 313                         ASSERT(len == XFS_LBSIZE(mp));
 314                         memset(dst + hdr_size + byte_cnt, 0,
 315                                         XFS_LBSIZE(mp) - hdr_size - byte_cnt);
 316                 }
 317
 318                 /* roll buffer forwards */
 319                 len -= XFS_LBSIZE(mp);
 320                 dst += XFS_LBSIZE(mp);
 321                 bno += mp->m_bsize;
 322
 323                 /* roll attribute data forwards */
 324                 *valuelen -= byte_cnt;
 325                 *src += byte_cnt;
 326                 *offset += byte_cnt;
 327         }
 328 }
 329
 330 /*
 331  * Read the value associated with an attribute from the out-of-line buffer
 332  * that we stored it in.
 333  */
 334 int
 335 xfs_attr_rmtval_get(
 336         struct xfs_da_args      *args)
 337 {
 338         struct xfs_bmbt_irec    map[ATTR_RMTVALUE_MAPSIZE];
 339         struct xfs_mount        *mp = args->dp->i_mount;
 340         struct xfs_buf          *bp;
 341         xfs_dablk_t             lblkno = args->rmtblkno;
 342         __uint8_t               *dst = args->value;
 343         int                     valuelen = args->valuelen;
 344         int                     nmap;
 345         int                     error;
 346         int                     blkcnt = args->rmtblkcnt;
 347         int                     i;
 348         int                     offset = 0;
 349
 350         trace_xfs_attr_rmtval_get(args);
 351
 352         ASSERT(!(args->flags & ATTR_KERNOVAL));
 353
 354         while (valuelen > 0) {
 355                 nmap = ATTR_RMTVALUE_MAPSIZE;
 356                 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
 357                                        blkcnt, map, &nmap,
 358                                        XFS_BMAPI_ATTRFORK);
 359                 if (error)
 360                         return error;
 361                 ASSERT(nmap >= 1);
 362
 363                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
 364                         xfs_daddr_t     dblkno;
 365                         int             dblkcnt;
 366
 367                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
 368                                (map[i].br_startblock != HOLESTARTBLOCK));
 369                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
 370                         dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
 371                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 372                                                    dblkno, dblkcnt, 0, &bp,
 373                                                    &xfs_attr3_rmt_buf_ops);
 374                         if (error)
 375                                 return error;
 376
 377                         error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
 378                                                         &offset, &valuelen,
 379                                                         &dst);
 380                         xfs_buf_relse(bp);
 381                         if (error)
 382                                 return error;
 383
 384                         /* roll attribute extent map forwards */
 385                         lblkno += map[i].br_blockcount;
 386                         blkcnt -= map[i].br_blockcount;
 387                 }
 388         }
 389         ASSERT(valuelen == 0);
 390         return 0;
 391 }
 392
 393 /*
 394  * Write the value associated with an attribute into the out-of-line buffer
 395  * that we have defined for it.
 396  */
 397 int
 398 xfs_attr_rmtval_set(
 399         struct xfs_da_args      *args)
 400 {
 401         struct xfs_inode        *dp = args->dp;
 402         struct xfs_mount        *mp = dp->i_mount;
 403         struct xfs_bmbt_irec    map;
 404         xfs_dablk_t             lblkno;
 405         xfs_fileoff_t           lfileoff = 0;
 406         __uint8_t               *src = args->value;
 407         int                     blkcnt;
 408         int                     valuelen;
 409         int                     nmap;
 410         int                     error;
 411         int                     offset = 0;
 412
 413         trace_xfs_attr_rmtval_set(args);
 414
 415         /*
 416          * Find a "hole" in the attribute address space large enough for
 417          * us to drop the new attribute's value into. Because CRC enable
 418          * attributes have headers, we can't just do a straight byte to FSB
 419          * conversion and have to take the header space into account.
 420          */
 421         blkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
 422         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
 423                                                    XFS_ATTR_FORK);
 424         if (error)
 425                 return error;
 426
 427         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
 428         args->rmtblkcnt = blkcnt;
 429
 430         /*
 431          * Roll through the "value", allocating blocks on disk as required.
 432          */
 433         while (blkcnt > 0) {
 434                 int     committed;
 435
 436                 /*
 437                  * Allocate a single extent, up to the size of the value.
 438                  */
 439                 xfs_bmap_init(args->flist, args->firstblock);
 440                 nmap = 1;
 441                 error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
 442                                   blkcnt,
 443                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
 444                                   args->firstblock, args->total, &map, &nmap,
 445                                   args->flist);
 446                 if (!error) {
 447                         error = xfs_bmap_finish(&args->trans, args->flist,
 448                                                 &committed);
 449                 }
 450                 if (error) {
 451                         ASSERT(committed);
 452                         args->trans = NULL;
 453                         xfs_bmap_cancel(args->flist);
 454                         return(error);
 455                 }
 456
 457                 /*
 458                  * bmap_finish() may have committed the last trans and started
 459                  * a new one.  We need the inode to be in all transactions.
 460                  */
 461                 if (committed)
 462                         xfs_trans_ijoin(args->trans, dp, 0);
 463
 464                 ASSERT(nmap == 1);
 465                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 466                        (map.br_startblock != HOLESTARTBLOCK));
 467                 lblkno += map.br_blockcount;
 468                 blkcnt -= map.br_blockcount;
 469
 470                 /*
 471                  * Start the next trans in the chain.
 472                  */
 473                 error = xfs_trans_roll(&args->trans, dp);
 474                 if (error)
 475                         return (error);
 476         }
 477
 478         /*
 479          * Roll through the "value", copying the attribute value to the
 480          * already-allocated blocks.  Blocks are written synchronously
 481          * so that we can know they are all on disk before we turn off
 482          * the INCOMPLETE flag.
 483          */
 484         lblkno = args->rmtblkno;
 485         blkcnt = args->rmtblkcnt;
 486         valuelen = args->valuelen;
 487         while (valuelen > 0) {
 488                 struct xfs_buf  *bp;
 489                 xfs_daddr_t     dblkno;
 490                 int             dblkcnt;
 491
 492                 ASSERT(blkcnt > 0);
 493
 494                 xfs_bmap_init(args->flist, args->firstblock);
 495                 nmap = 1;
 496                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
 497                                        blkcnt, &map, &nmap,
 498                                        XFS_BMAPI_ATTRFORK);
 499                 if (error)
 500                         return(error);
 501                 ASSERT(nmap == 1);
 502                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 503                        (map.br_startblock != HOLESTARTBLOCK));
 504
 505                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
 506                 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
 507
 508                 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
 509                 if (!bp)
 510                         return ENOMEM;
 511                 bp->b_ops = &xfs_attr3_rmt_buf_ops;
 512
 513                 xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
 514                                        &valuelen, &src);
 515
 516                 error = xfs_bwrite(bp); /* GROT: NOTE: synchronous write */
 517                 xfs_buf_relse(bp);
 518                 if (error)
 519                         return error;
 520
 521
 522                 /* roll attribute extent map forwards */
 523                 lblkno += map.br_blockcount;
 524                 blkcnt -= map.br_blockcount;
 525         }
 526         ASSERT(valuelen == 0);
 527         return 0;
 528 }
 529
 530 /*
 531  * Remove the value associated with an attribute by deleting the
 532  * out-of-line buffer that it is stored on.
 533  */
 534 int
 535 xfs_attr_rmtval_remove(
 536         struct xfs_da_args      *args)
 537 {
 538         struct xfs_mount        *mp = args->dp->i_mount;
 539         xfs_dablk_t             lblkno;
 540         int                     blkcnt;
 541         int                     error;
 542         int                     done;
 543
 544         trace_xfs_attr_rmtval_remove(args);
 545
 546         /*
 547          * Roll through the "value", invalidating the attribute value's blocks.
 548          */
 549         lblkno = args->rmtblkno;
 550         blkcnt = args->rmtblkcnt;
 551         while (blkcnt > 0) {
 552                 struct xfs_bmbt_irec    map;
 553                 struct xfs_buf          *bp;
 554                 xfs_daddr_t             dblkno;
 555                 int                     dblkcnt;
 556                 int                     nmap;
 557
 558                 /*
 559                  * Try to remember where we decided to put the value.
 560                  */
 561                 nmap = 1;
 562                 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
 563                                        blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
 564                 if (error)
 565                         return(error);
 566                 ASSERT(nmap == 1);
 567                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 568                        (map.br_startblock != HOLESTARTBLOCK));
 569
 570                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
 571                 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
 572
 573                 /*
 574                  * If the "remote" value is in the cache, remove it.
 575                  */
 576                 bp = xfs_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
 577                 if (bp) {
 578                         xfs_buf_stale(bp);
 579                         xfs_buf_relse(bp);
 580                         bp = NULL;
 581                 }
 582
 583                 lblkno += map.br_blockcount;
 584                 blkcnt -= map.br_blockcount;
 585         }
 586
 587         /*
 588          * Keep de-allocating extents until the remote-value region is gone.
 589          */
 590         lblkno = args->rmtblkno;
 591         blkcnt = args->rmtblkcnt;
 592         done = 0;
 593         while (!done) {
 594                 int committed;
 595
 596                 xfs_bmap_init(args->flist, args->firstblock);
 597                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
 598                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
 599                                     1, args->firstblock, args->flist,
 600                                     &done);
 601                 if (!error) {
 602                         error = xfs_bmap_finish(&args->trans, args->flist,
 603                                                 &committed);
 604                 }
 605                 if (error) {
 606                         ASSERT(committed);
 607                         args->trans = NULL;
 608                         xfs_bmap_cancel(args->flist);
 609                         return error;
 610                 }
 611
 612                 /*
 613                  * bmap_finish() may have committed the last trans and started
 614                  * a new one.  We need the inode to be in all transactions.
 615                  */
 616                 if (committed)
 617                         xfs_trans_ijoin(args->trans, args->dp, 0);
 618
 619                 /*
 620                  * Close out trans and start the next one in the chain.
 621                  */
 622                 error = xfs_trans_roll(&args->trans, args->dp);
 623                 if (error)
 624                         return (error);
 625         }
 626         return(0);
 627 }