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         xfs_daddr_t     bno;
 129
 130         /* no verification of non-crc buffers */
 131         if (!xfs_sb_version_hascrc(&mp->m_sb))
 132                 return;
 133
 134         ptr = bp->b_addr;
 135         bno = bp->b_bn;
 136         len = BBTOB(bp->b_length);
 137         ASSERT(len >= XFS_LBSIZE(mp));
 138
 139         while (len > 0) {
 140                 if (!xfs_verify_cksum(ptr, XFS_LBSIZE(mp),
 141                                       XFS_ATTR3_RMT_CRC_OFF)) {
 142                         xfs_buf_ioerror(bp, EFSBADCRC);
 143                         break;
 144                 }
 145                 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
 146                         xfs_buf_ioerror(bp, EFSCORRUPTED);
 147                         break;
 148                 }
 149                 len -= XFS_LBSIZE(mp);
 150                 ptr += XFS_LBSIZE(mp);
 151                 bno += mp->m_bsize;
 152         }
 153
 154         if (bp->b_error)
 155                 xfs_verifier_error(bp);
 156         else
 157                 ASSERT(len == 0);
 158 }
 159
 160 static void
 161 xfs_attr3_rmt_write_verify(
 162         struct xfs_buf  *bp)
 163 {
 164         struct xfs_mount *mp = bp->b_target->bt_mount;
 165         struct xfs_buf_log_item *bip = bp->b_fspriv;
 166         char            *ptr;
 167         int             len;
 168         xfs_daddr_t     bno;
 169
 170         /* no verification of non-crc buffers */
 171         if (!xfs_sb_version_hascrc(&mp->m_sb))
 172                 return;
 173
 174         ptr = bp->b_addr;
 175         bno = bp->b_bn;
 176         len = BBTOB(bp->b_length);
 177         ASSERT(len >= XFS_LBSIZE(mp));
 178
 179         while (len > 0) {
 180                 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
 181                         xfs_buf_ioerror(bp, EFSCORRUPTED);
 182                         xfs_verifier_error(bp);
 183                         return;
 184                 }
 185                 if (bip) {
 186                         struct xfs_attr3_rmt_hdr *rmt;
 187
 188                         rmt = (struct xfs_attr3_rmt_hdr *)ptr;
 189                         rmt->rm_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 190                 }
 191                 xfs_update_cksum(ptr, XFS_LBSIZE(mp), XFS_ATTR3_RMT_CRC_OFF);
 192
 193                 len -= XFS_LBSIZE(mp);
 194                 ptr += XFS_LBSIZE(mp);
 195                 bno += mp->m_bsize;
 196         }
 197         ASSERT(len == 0);
 198 }
 199
 200 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
 201         .verify_read = xfs_attr3_rmt_read_verify,
 202         .verify_write = xfs_attr3_rmt_write_verify,
 203 };
 204
 205 STATIC int
 206 xfs_attr3_rmt_hdr_set(
 207         struct xfs_mount        *mp,
 208         void                    *ptr,
 209         xfs_ino_t               ino,
 210         uint32_t                offset,
 211         uint32_t                size,
 212         xfs_daddr_t             bno)
 213 {
 214         struct xfs_attr3_rmt_hdr *rmt = ptr;
 215
 216         if (!xfs_sb_version_hascrc(&mp->m_sb))
 217                 return 0;
 218
 219         rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
 220         rmt->rm_offset = cpu_to_be32(offset);
 221         rmt->rm_bytes = cpu_to_be32(size);
 222         uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_uuid);
 223         rmt->rm_owner = cpu_to_be64(ino);
 224         rmt->rm_blkno = cpu_to_be64(bno);
 225
 226         return sizeof(struct xfs_attr3_rmt_hdr);
 227 }
 228
 229 /*
 230  * Helper functions to copy attribute data in and out of the one disk extents
 231  */
 232 STATIC int
 233 xfs_attr_rmtval_copyout(
 234         struct xfs_mount *mp,
 235         struct xfs_buf  *bp,
 236         xfs_ino_t       ino,
 237         int             *offset,
 238         int             *valuelen,
 239         __uint8_t       **dst)
 240 {
 241         char            *src = bp->b_addr;
 242         xfs_daddr_t     bno = bp->b_bn;
 243         int             len = BBTOB(bp->b_length);
 244
 245         ASSERT(len >= XFS_LBSIZE(mp));
 246
 247         while (len > 0 && *valuelen > 0) {
 248                 int hdr_size = 0;
 249                 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
 250
 251                 byte_cnt = min(*valuelen, byte_cnt);
 252
 253                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 254                         if (!xfs_attr3_rmt_hdr_ok(mp, src, ino, *offset,
 255                                                   byte_cnt, bno)) {
 256                                 xfs_alert(mp,
 257 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
 258                                         bno, *offset, byte_cnt, ino);
 259                                 return EFSCORRUPTED;
 260                         }
 261                         hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
 262                 }
 263
 264                 memcpy(*dst, src + hdr_size, byte_cnt);
 265
 266                 /* roll buffer forwards */
 267                 len -= XFS_LBSIZE(mp);
 268                 src += XFS_LBSIZE(mp);
 269                 bno += mp->m_bsize;
 270
 271                 /* roll attribute data forwards */
 272                 *valuelen -= byte_cnt;
 273                 *dst += byte_cnt;
 274                 *offset += byte_cnt;
 275         }
 276         return 0;
 277 }
 278
 279 STATIC void
 280 xfs_attr_rmtval_copyin(
 281         struct xfs_mount *mp,
 282         struct xfs_buf  *bp,
 283         xfs_ino_t       ino,
 284         int             *offset,
 285         int             *valuelen,
 286         __uint8_t       **src)
 287 {
 288         char            *dst = bp->b_addr;
 289         xfs_daddr_t     bno = bp->b_bn;
 290         int             len = BBTOB(bp->b_length);
 291
 292         ASSERT(len >= XFS_LBSIZE(mp));
 293
 294         while (len > 0 && *valuelen > 0) {
 295                 int hdr_size;
 296                 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
 297
 298                 byte_cnt = min(*valuelen, byte_cnt);
 299                 hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
 300                                                  byte_cnt, bno);
 301
 302                 memcpy(dst + hdr_size, *src, byte_cnt);
 303
 304                 /*
 305                  * If this is the last block, zero the remainder of it.
 306                  * Check that we are actually the last block, too.
 307                  */
 308                 if (byte_cnt + hdr_size < XFS_LBSIZE(mp)) {
 309                         ASSERT(*valuelen - byte_cnt == 0);
 310                         ASSERT(len == XFS_LBSIZE(mp));
 311                         memset(dst + hdr_size + byte_cnt, 0,
 312                                         XFS_LBSIZE(mp) - hdr_size - byte_cnt);
 313                 }
 314
 315                 /* roll buffer forwards */
 316                 len -= XFS_LBSIZE(mp);
 317                 dst += XFS_LBSIZE(mp);
 318                 bno += mp->m_bsize;
 319
 320                 /* roll attribute data forwards */
 321                 *valuelen -= byte_cnt;
 322                 *src += byte_cnt;
 323                 *offset += byte_cnt;
 324         }
 325 }
 326
 327 /*
 328  * Read the value associated with an attribute from the out-of-line buffer
 329  * that we stored it in.
 330  */
 331 int
 332 xfs_attr_rmtval_get(
 333         struct xfs_da_args      *args)
 334 {
 335         struct xfs_bmbt_irec    map[ATTR_RMTVALUE_MAPSIZE];
 336         struct xfs_mount        *mp = args->dp->i_mount;
 337         struct xfs_buf          *bp;
 338         xfs_dablk_t             lblkno = args->rmtblkno;
 339         __uint8_t               *dst = args->value;
 340         int                     valuelen = args->valuelen;
 341         int                     nmap;
 342         int                     error;
 343         int                     blkcnt = args->rmtblkcnt;
 344         int                     i;
 345         int                     offset = 0;
 346
 347         trace_xfs_attr_rmtval_get(args);
 348
 349         ASSERT(!(args->flags & ATTR_KERNOVAL));
 350
 351         while (valuelen > 0) {
 352                 nmap = ATTR_RMTVALUE_MAPSIZE;
 353                 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
 354                                        blkcnt, map, &nmap,
 355                                        XFS_BMAPI_ATTRFORK);
 356                 if (error)
 357                         return error;
 358                 ASSERT(nmap >= 1);
 359
 360                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
 361                         xfs_daddr_t     dblkno;
 362                         int             dblkcnt;
 363
 364                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
 365                                (map[i].br_startblock != HOLESTARTBLOCK));
 366                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
 367                         dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
 368                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 369                                                    dblkno, dblkcnt, 0, &bp,
 370                                                    &xfs_attr3_rmt_buf_ops);
 371                         if (error)
 372                                 return error;
 373
 374                         error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
 375                                                         &offset, &valuelen,
 376                                                         &dst);
 377                         xfs_buf_relse(bp);
 378                         if (error)
 379                                 return error;
 380
 381                         /* roll attribute extent map forwards */
 382                         lblkno += map[i].br_blockcount;
 383                         blkcnt -= map[i].br_blockcount;
 384                 }
 385         }
 386         ASSERT(valuelen == 0);
 387         return 0;
 388 }
 389
 390 /*
 391  * Write the value associated with an attribute into the out-of-line buffer
 392  * that we have defined for it.
 393  */
 394 int
 395 xfs_attr_rmtval_set(
 396         struct xfs_da_args      *args)
 397 {
 398         struct xfs_inode        *dp = args->dp;
 399         struct xfs_mount        *mp = dp->i_mount;
 400         struct xfs_bmbt_irec    map;
 401         xfs_dablk_t             lblkno;
 402         xfs_fileoff_t           lfileoff = 0;
 403         __uint8_t               *src = args->value;
 404         int                     blkcnt;
 405         int                     valuelen;
 406         int                     nmap;
 407         int                     error;
 408         int                     offset = 0;
 409
 410         trace_xfs_attr_rmtval_set(args);
 411
 412         /*
 413          * Find a "hole" in the attribute address space large enough for
 414          * us to drop the new attribute's value into. Because CRC enable
 415          * attributes have headers, we can't just do a straight byte to FSB
 416          * conversion and have to take the header space into account.
 417          */
 418         blkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
 419         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
 420                                                    XFS_ATTR_FORK);
 421         if (error)
 422                 return error;
 423
 424         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
 425         args->rmtblkcnt = blkcnt;
 426
 427         /*
 428          * Roll through the "value", allocating blocks on disk as required.
 429          */
 430         while (blkcnt > 0) {
 431                 int     committed;
 432
 433                 /*
 434                  * Allocate a single extent, up to the size of the value.
 435                  */
 436                 xfs_bmap_init(args->flist, args->firstblock);
 437                 nmap = 1;
 438                 error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
 439                                   blkcnt,
 440                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
 441                                   args->firstblock, args->total, &map, &nmap,
 442                                   args->flist);
 443                 if (!error) {
 444                         error = xfs_bmap_finish(&args->trans, args->flist,
 445                                                 &committed);
 446                 }
 447                 if (error) {
 448                         ASSERT(committed);
 449                         args->trans = NULL;
 450                         xfs_bmap_cancel(args->flist);
 451                         return(error);
 452                 }
 453
 454                 /*
 455                  * bmap_finish() may have committed the last trans and started
 456                  * a new one.  We need the inode to be in all transactions.
 457                  */
 458                 if (committed)
 459                         xfs_trans_ijoin(args->trans, dp, 0);
 460
 461                 ASSERT(nmap == 1);
 462                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 463                        (map.br_startblock != HOLESTARTBLOCK));
 464                 lblkno += map.br_blockcount;
 465                 blkcnt -= map.br_blockcount;
 466
 467                 /*
 468                  * Start the next trans in the chain.
 469                  */
 470                 error = xfs_trans_roll(&args->trans, dp);
 471                 if (error)
 472                         return (error);
 473         }
 474
 475         /*
 476          * Roll through the "value", copying the attribute value to the
 477          * already-allocated blocks.  Blocks are written synchronously
 478          * so that we can know they are all on disk before we turn off
 479          * the INCOMPLETE flag.
 480          */
 481         lblkno = args->rmtblkno;
 482         blkcnt = args->rmtblkcnt;
 483         valuelen = args->valuelen;
 484         while (valuelen > 0) {
 485                 struct xfs_buf  *bp;
 486                 xfs_daddr_t     dblkno;
 487                 int             dblkcnt;
 488
 489                 ASSERT(blkcnt > 0);
 490
 491                 xfs_bmap_init(args->flist, args->firstblock);
 492                 nmap = 1;
 493                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
 494                                        blkcnt, &map, &nmap,
 495                                        XFS_BMAPI_ATTRFORK);
 496                 if (error)
 497                         return(error);
 498                 ASSERT(nmap == 1);
 499                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 500                        (map.br_startblock != HOLESTARTBLOCK));
 501
 502                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
 503                 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
 504
 505                 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
 506                 if (!bp)
 507                         return ENOMEM;
 508                 bp->b_ops = &xfs_attr3_rmt_buf_ops;
 509
 510                 xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
 511                                        &valuelen, &src);
 512
 513                 error = xfs_bwrite(bp); /* GROT: NOTE: synchronous write */
 514                 xfs_buf_relse(bp);
 515                 if (error)
 516                         return error;
 517
 518
 519                 /* roll attribute extent map forwards */
 520                 lblkno += map.br_blockcount;
 521                 blkcnt -= map.br_blockcount;
 522         }
 523         ASSERT(valuelen == 0);
 524         return 0;
 525 }
 526
 527 /*
 528  * Remove the value associated with an attribute by deleting the
 529  * out-of-line buffer that it is stored on.
 530  */
 531 int
 532 xfs_attr_rmtval_remove(
 533         struct xfs_da_args      *args)
 534 {
 535         struct xfs_mount        *mp = args->dp->i_mount;
 536         xfs_dablk_t             lblkno;
 537         int                     blkcnt;
 538         int                     error;
 539         int                     done;
 540
 541         trace_xfs_attr_rmtval_remove(args);
 542
 543         /*
 544          * Roll through the "value", invalidating the attribute value's blocks.
 545          */
 546         lblkno = args->rmtblkno;
 547         blkcnt = args->rmtblkcnt;
 548         while (blkcnt > 0) {
 549                 struct xfs_bmbt_irec    map;
 550                 struct xfs_buf          *bp;
 551                 xfs_daddr_t             dblkno;
 552                 int                     dblkcnt;
 553                 int                     nmap;
 554
 555                 /*
 556                  * Try to remember where we decided to put the value.
 557                  */
 558                 nmap = 1;
 559                 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
 560                                        blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
 561                 if (error)
 562                         return(error);
 563                 ASSERT(nmap == 1);
 564                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 565                        (map.br_startblock != HOLESTARTBLOCK));
 566
 567                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
 568                 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
 569
 570                 /*
 571                  * If the "remote" value is in the cache, remove it.
 572                  */
 573                 bp = xfs_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
 574                 if (bp) {
 575                         xfs_buf_stale(bp);
 576                         xfs_buf_relse(bp);
 577                         bp = NULL;
 578                 }
 579
 580                 lblkno += map.br_blockcount;
 581                 blkcnt -= map.br_blockcount;
 582         }
 583
 584         /*
 585          * Keep de-allocating extents until the remote-value region is gone.
 586          */
 587         lblkno = args->rmtblkno;
 588         blkcnt = args->rmtblkcnt;
 589         done = 0;
 590         while (!done) {
 591                 int committed;
 592
 593                 xfs_bmap_init(args->flist, args->firstblock);
 594                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
 595                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
 596                                     1, args->firstblock, args->flist,
 597                                     &done);
 598                 if (!error) {
 599                         error = xfs_bmap_finish(&args->trans, args->flist,
 600                                                 &committed);
 601                 }
 602                 if (error) {
 603                         ASSERT(committed);
 604                         args->trans = NULL;
 605                         xfs_bmap_cancel(args->flist);
 606                         return error;
 607                 }
 608
 609                 /*
 610                  * bmap_finish() may have committed the last trans and started
 611                  * a new one.  We need the inode to be in all transactions.
 612                  */
 613                 if (committed)
 614                         xfs_trans_ijoin(args->trans, args->dp, 0);
 615
 616                 /*
 617                  * Close out trans and start the next one in the chain.
 618                  */
 619                 error = xfs_trans_roll(&args->trans, args->dp);
 620                 if (error)
 621                         return (error);
 622         }
 623         return(0);
 624 }