fs/jfs/xattr.c

   1 /*
   2  *   Copyright (C) International Business Machines  Corp., 2000-2004
   3  *   Copyright (C) Christoph Hellwig, 2002
   4  *
   5  *   This program is free software;  you can redistribute it and/or modify
   6  *   it under the terms of the GNU General Public License as published by
   7  *   the Free Software Foundation; either version 2 of the License, or
   8  *   (at your option) any later version.
   9  *
  10  *   This program is distributed in the hope that it will be useful,
  11  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
  12  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  13  *   the 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 to the Free Software
  17  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  18  */
  19
  20 #include <linux/capability.h>
  21 #include <linux/fs.h>
  22 #include <linux/xattr.h>
  23 #include <linux/posix_acl_xattr.h>
  24 #include <linux/quotaops.h>
  25 #include <linux/security.h>
  26 #include "jfs_incore.h"
  27 #include "jfs_superblock.h"
  28 #include "jfs_dmap.h"
  29 #include "jfs_debug.h"
  30 #include "jfs_dinode.h"
  31 #include "jfs_extent.h"
  32 #include "jfs_metapage.h"
  33 #include "jfs_xattr.h"
  34 #include "jfs_acl.h"
  35
  36 /*
  37  *      jfs_xattr.c: extended attribute service
  38  *
  39  * Overall design --
  40  *
  41  * Format:
  42  *
  43  *   Extended attribute lists (jfs_ea_list) consist of an overall size (32 bit
  44  *   value) and a variable (0 or more) number of extended attribute
  45  *   entries.  Each extended attribute entry (jfs_ea) is a <name,value> double
  46  *   where <name> is constructed from a null-terminated ascii string
  47  *   (1 ... 255 bytes in the name) and <value> is arbitrary 8 bit data
  48  *   (1 ... 65535 bytes).  The in-memory format is
  49  *
  50  *   0       1        2        4                4 + namelen + 1
  51  *   +-------+--------+--------+----------------+-------------------+
  52  *   | Flags | Name   | Value  | Name String \0 | Data . . . .      |
  53  *   |       | Length | Length |                |                   |
  54  *   +-------+--------+--------+----------------+-------------------+
  55  *
  56  *   A jfs_ea_list then is structured as
  57  *
  58  *   0            4                   4 + EA_SIZE(ea1)
  59  *   +------------+-------------------+--------------------+-----
  60  *   | Overall EA | First FEA Element | Second FEA Element | .....
  61  *   | List Size  |                   |                    |
  62  *   +------------+-------------------+--------------------+-----
  63  *
  64  *   On-disk:
  65  *
  66  *      FEALISTs are stored on disk using blocks allocated by dbAlloc() and
  67  *      written directly. An EA list may be in-lined in the inode if there is
  68  *      sufficient room available.
  69  */
  70
  71 struct ea_buffer {
  72         int flag;               /* Indicates what storage xattr points to */
  73         int max_size;           /* largest xattr that fits in current buffer */
  74         dxd_t new_ea;           /* dxd to replace ea when modifying xattr */
  75         struct metapage *mp;    /* metapage containing ea list */
  76         struct jfs_ea_list *xattr;      /* buffer containing ea list */
  77 };
  78
  79 /*
  80  * ea_buffer.flag values
  81  */
  82 #define EA_INLINE       0x0001
  83 #define EA_EXTENT       0x0002
  84 #define EA_NEW          0x0004
  85 #define EA_MALLOC       0x0008
  86
  87
  88 /*
  89  * These three routines are used to recognize on-disk extended attributes
  90  * that are in a recognized namespace.  If the attribute is not recognized,
  91  * "os2." is prepended to the name
  92  */
  93 static inline int is_os2_xattr(struct jfs_ea *ea)
  94 {
  95         /*
  96          * Check for "system."
  97          */
  98         if ((ea->namelen >= XATTR_SYSTEM_PREFIX_LEN) &&
  99             !strncmp(ea->name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
 100                 return false;
 101         /*
 102          * Check for "user."
 103          */
 104         if ((ea->namelen >= XATTR_USER_PREFIX_LEN) &&
 105             !strncmp(ea->name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN))
 106                 return false;
 107         /*
 108          * Check for "security."
 109          */
 110         if ((ea->namelen >= XATTR_SECURITY_PREFIX_LEN) &&
 111             !strncmp(ea->name, XATTR_SECURITY_PREFIX,
 112                      XATTR_SECURITY_PREFIX_LEN))
 113                 return false;
 114         /*
 115          * Check for "trusted."
 116          */
 117         if ((ea->namelen >= XATTR_TRUSTED_PREFIX_LEN) &&
 118             !strncmp(ea->name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN))
 119                 return false;
 120         /*
 121          * Add any other valid namespace prefixes here
 122          */
 123
 124         /*
 125          * We assume it's OS/2's flat namespace
 126          */
 127         return true;
 128 }
 129
 130 static inline int name_size(struct jfs_ea *ea)
 131 {
 132         if (is_os2_xattr(ea))
 133                 return ea->namelen + XATTR_OS2_PREFIX_LEN;
 134         else
 135                 return ea->namelen;
 136 }
 137
 138 static inline int copy_name(char *buffer, struct jfs_ea *ea)
 139 {
 140         int len = ea->namelen;
 141
 142         if (is_os2_xattr(ea)) {
 143                 memcpy(buffer, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN);
 144                 buffer += XATTR_OS2_PREFIX_LEN;
 145                 len += XATTR_OS2_PREFIX_LEN;
 146         }
 147         memcpy(buffer, ea->name, ea->namelen);
 148         buffer[ea->namelen] = 0;
 149
 150         return len;
 151 }
 152
 153 /* Forward references */
 154 static void ea_release(struct inode *inode, struct ea_buffer *ea_buf);
 155
 156 /*
 157  * NAME: ea_write_inline
 158  *
 159  * FUNCTION: Attempt to write an EA inline if area is available
 160  *
 161  * PRE CONDITIONS:
 162  *      Already verified that the specified EA is small enough to fit inline
 163  *
 164  * PARAMETERS:
 165  *      ip      - Inode pointer
 166  *      ealist  - EA list pointer
 167  *      size    - size of ealist in bytes
 168  *      ea      - dxd_t structure to be filled in with necessary EA information
 169  *                if we successfully copy the EA inline
 170  *
 171  * NOTES:
 172  *      Checks if the inode's inline area is available.  If so, copies EA inline
 173  *      and sets <ea> fields appropriately.  Otherwise, returns failure, EA will
 174  *      have to be put into an extent.
 175  *
 176  * RETURNS: 0 for successful copy to inline area; -1 if area not available
 177  */
 178 static int ea_write_inline(struct inode *ip, struct jfs_ea_list *ealist,
 179                            int size, dxd_t * ea)
 180 {
 181         struct jfs_inode_info *ji = JFS_IP(ip);
 182
 183         /*
 184          * Make sure we have an EA -- the NULL EA list is valid, but you
 185          * can't copy it!
 186          */
 187         if (ealist && size > sizeof (struct jfs_ea_list)) {
 188                 assert(size <= sizeof (ji->i_inline_ea));
 189
 190                 /*
 191                  * See if the space is available or if it is already being
 192                  * used for an inline EA.
 193                  */
 194                 if (!(ji->mode2 & INLINEEA) && !(ji->ea.flag & DXD_INLINE))
 195                         return -EPERM;
 196
 197                 DXDsize(ea, size);
 198                 DXDlength(ea, 0);
 199                 DXDaddress(ea, 0);
 200                 memcpy(ji->i_inline_ea, ealist, size);
 201                 ea->flag = DXD_INLINE;
 202                 ji->mode2 &= ~INLINEEA;
 203         } else {
 204                 ea->flag = 0;
 205                 DXDsize(ea, 0);
 206                 DXDlength(ea, 0);
 207                 DXDaddress(ea, 0);
 208
 209                 /* Free up INLINE area */
 210                 if (ji->ea.flag & DXD_INLINE)
 211                         ji->mode2 |= INLINEEA;
 212         }
 213
 214         return 0;
 215 }
 216
 217 /*
 218  * NAME: ea_write
 219  *
 220  * FUNCTION: Write an EA for an inode
 221  *
 222  * PRE CONDITIONS: EA has been verified
 223  *
 224  * PARAMETERS:
 225  *      ip      - Inode pointer
 226  *      ealist  - EA list pointer
 227  *      size    - size of ealist in bytes
 228  *      ea      - dxd_t structure to be filled in appropriately with where the
 229  *                EA was copied
 230  *
 231  * NOTES: Will write EA inline if able to, otherwise allocates blocks for an
 232  *      extent and synchronously writes it to those blocks.
 233  *
 234  * RETURNS: 0 for success; Anything else indicates failure
 235  */
 236 static int ea_write(struct inode *ip, struct jfs_ea_list *ealist, int size,
 237                        dxd_t * ea)
 238 {
 239         struct super_block *sb = ip->i_sb;
 240         struct jfs_inode_info *ji = JFS_IP(ip);
 241         struct jfs_sb_info *sbi = JFS_SBI(sb);
 242         int nblocks;
 243         s64 blkno;
 244         int rc = 0, i;
 245         char *cp;
 246         s32 nbytes, nb;
 247         s32 bytes_to_write;
 248         struct metapage *mp;
 249
 250         /*
 251          * Quick check to see if this is an in-linable EA.  Short EAs
 252          * and empty EAs are all in-linable, provided the space exists.
 253          */
 254         if (!ealist || size <= sizeof (ji->i_inline_ea)) {
 255                 if (!ea_write_inline(ip, ealist, size, ea))
 256                         return 0;
 257         }
 258
 259         /* figure out how many blocks we need */
 260         nblocks = (size + (sb->s_blocksize - 1)) >> sb->s_blocksize_bits;
 261
 262         /* Allocate new blocks to quota. */
 263         rc = dquot_alloc_block(ip, nblocks);
 264         if (rc)
 265                 return rc;
 266
 267         rc = dbAlloc(ip, INOHINT(ip), nblocks, &blkno);
 268         if (rc) {
 269                 /*Rollback quota allocation. */
 270                 dquot_free_block(ip, nblocks);
 271                 return rc;
 272         }
 273
 274         /*
 275          * Now have nblocks worth of storage to stuff into the FEALIST.
 276          * loop over the FEALIST copying data into the buffer one page at
 277          * a time.
 278          */
 279         cp = (char *) ealist;
 280         nbytes = size;
 281         for (i = 0; i < nblocks; i += sbi->nbperpage) {
 282                 /*
 283                  * Determine how many bytes for this request, and round up to
 284                  * the nearest aggregate block size
 285                  */
 286                 nb = min(PSIZE, nbytes);
 287                 bytes_to_write =
 288                     ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
 289                     << sb->s_blocksize_bits;
 290
 291                 if (!(mp = get_metapage(ip, blkno + i, bytes_to_write, 1))) {
 292                         rc = -EIO;
 293                         goto failed;
 294                 }
 295
 296                 memcpy(mp->data, cp, nb);
 297
 298                 /*
 299                  * We really need a way to propagate errors for
 300                  * forced writes like this one.  --hch
 301                  *
 302                  * (__write_metapage => release_metapage => flush_metapage)
 303                  */
 304 #ifdef _JFS_FIXME
 305                 if ((rc = flush_metapage(mp))) {
 306                         /*
 307                          * the write failed -- this means that the buffer
 308                          * is still assigned and the blocks are not being
 309                          * used.  this seems like the best error recovery
 310                          * we can get ...
 311                          */
 312                         goto failed;
 313                 }
 314 #else
 315                 flush_metapage(mp);
 316 #endif
 317
 318                 cp += PSIZE;
 319                 nbytes -= nb;
 320         }
 321
 322         ea->flag = DXD_EXTENT;
 323         DXDsize(ea, le32_to_cpu(ealist->size));
 324         DXDlength(ea, nblocks);
 325         DXDaddress(ea, blkno);
 326
 327         /* Free up INLINE area */
 328         if (ji->ea.flag & DXD_INLINE)
 329                 ji->mode2 |= INLINEEA;
 330
 331         return 0;
 332
 333       failed:
 334         /* Rollback quota allocation. */
 335         dquot_free_block(ip, nblocks);
 336
 337         dbFree(ip, blkno, nblocks);
 338         return rc;
 339 }
 340
 341 /*
 342  * NAME: ea_read_inline
 343  *
 344  * FUNCTION: Read an inlined EA into user's buffer
 345  *
 346  * PARAMETERS:
 347  *      ip      - Inode pointer
 348  *      ealist  - Pointer to buffer to fill in with EA
 349  *
 350  * RETURNS: 0
 351  */
 352 static int ea_read_inline(struct inode *ip, struct jfs_ea_list *ealist)
 353 {
 354         struct jfs_inode_info *ji = JFS_IP(ip);
 355         int ea_size = sizeDXD(&ji->ea);
 356
 357         if (ea_size == 0) {
 358                 ealist->size = 0;
 359                 return 0;
 360         }
 361
 362         /* Sanity Check */
 363         if ((sizeDXD(&ji->ea) > sizeof (ji->i_inline_ea)))
 364                 return -EIO;
 365         if (le32_to_cpu(((struct jfs_ea_list *) &ji->i_inline_ea)->size)
 366             != ea_size)
 367                 return -EIO;
 368
 369         memcpy(ealist, ji->i_inline_ea, ea_size);
 370         return 0;
 371 }
 372
 373 /*
 374  * NAME: ea_read
 375  *
 376  * FUNCTION: copy EA data into user's buffer
 377  *
 378  * PARAMETERS:
 379  *      ip      - Inode pointer
 380  *      ealist  - Pointer to buffer to fill in with EA
 381  *
 382  * NOTES:  If EA is inline calls ea_read_inline() to copy EA.
 383  *
 384  * RETURNS: 0 for success; other indicates failure
 385  */
 386 static int ea_read(struct inode *ip, struct jfs_ea_list *ealist)
 387 {
 388         struct super_block *sb = ip->i_sb;
 389         struct jfs_inode_info *ji = JFS_IP(ip);
 390         struct jfs_sb_info *sbi = JFS_SBI(sb);
 391         int nblocks;
 392         s64 blkno;
 393         char *cp = (char *) ealist;
 394         int i;
 395         int nbytes, nb;
 396         s32 bytes_to_read;
 397         struct metapage *mp;
 398
 399         /* quick check for in-line EA */
 400         if (ji->ea.flag & DXD_INLINE)
 401                 return ea_read_inline(ip, ealist);
 402
 403         nbytes = sizeDXD(&ji->ea);
 404         if (!nbytes) {
 405                 jfs_error(sb, "ea_read: nbytes is 0");
 406                 return -EIO;
 407         }
 408
 409         /*
 410          * Figure out how many blocks were allocated when this EA list was
 411          * originally written to disk.
 412          */
 413         nblocks = lengthDXD(&ji->ea) << sbi->l2nbperpage;
 414         blkno = addressDXD(&ji->ea) << sbi->l2nbperpage;
 415
 416         /*
 417          * I have found the disk blocks which were originally used to store
 418          * the FEALIST.  now i loop over each contiguous block copying the
 419          * data into the buffer.
 420          */
 421         for (i = 0; i < nblocks; i += sbi->nbperpage) {
 422                 /*
 423                  * Determine how many bytes for this request, and round up to
 424                  * the nearest aggregate block size
 425                  */
 426                 nb = min(PSIZE, nbytes);
 427                 bytes_to_read =
 428                     ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
 429                     << sb->s_blocksize_bits;
 430
 431                 if (!(mp = read_metapage(ip, blkno + i, bytes_to_read, 1)))
 432                         return -EIO;
 433
 434                 memcpy(cp, mp->data, nb);
 435                 release_metapage(mp);
 436
 437                 cp += PSIZE;
 438                 nbytes -= nb;
 439         }
 440
 441         return 0;
 442 }
 443
 444 /*
 445  * NAME: ea_get
 446  *
 447  * FUNCTION: Returns buffer containing existing extended attributes.
 448  *           The size of the buffer will be the larger of the existing
 449  *           attributes size, or min_size.
 450  *
 451  *           The buffer, which may be inlined in the inode or in the
 452  *           page cache must be release by calling ea_release or ea_put
 453  *
 454  * PARAMETERS:
 455  *      inode   - Inode pointer
 456  *      ea_buf  - Structure to be populated with ealist and its metadata
 457  *      min_size- minimum size of buffer to be returned
 458  *
 459  * RETURNS: 0 for success; Other indicates failure
 460  */
 461 static int ea_get(struct inode *inode, struct ea_buffer *ea_buf, int min_size)
 462 {
 463         struct jfs_inode_info *ji = JFS_IP(inode);
 464         struct super_block *sb = inode->i_sb;
 465         int size;
 466         int ea_size = sizeDXD(&ji->ea);
 467         int blocks_needed, current_blocks;
 468         s64 blkno;
 469         int rc;
 470         int quota_allocation = 0;
 471
 472         /* When fsck.jfs clears a bad ea, it doesn't clear the size */
 473         if (ji->ea.flag == 0)
 474                 ea_size = 0;
 475
 476         if (ea_size == 0) {
 477                 if (min_size == 0) {
 478                         ea_buf->flag = 0;
 479                         ea_buf->max_size = 0;
 480                         ea_buf->xattr = NULL;
 481                         return 0;
 482                 }
 483                 if ((min_size <= sizeof (ji->i_inline_ea)) &&
 484                     (ji->mode2 & INLINEEA)) {
 485                         ea_buf->flag = EA_INLINE | EA_NEW;
 486                         ea_buf->max_size = sizeof (ji->i_inline_ea);
 487                         ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
 488                         DXDlength(&ea_buf->new_ea, 0);
 489                         DXDaddress(&ea_buf->new_ea, 0);
 490                         ea_buf->new_ea.flag = DXD_INLINE;
 491                         DXDsize(&ea_buf->new_ea, min_size);
 492                         return 0;
 493                 }
 494                 current_blocks = 0;
 495         } else if (ji->ea.flag & DXD_INLINE) {
 496                 if (min_size <= sizeof (ji->i_inline_ea)) {
 497                         ea_buf->flag = EA_INLINE;
 498                         ea_buf->max_size = sizeof (ji->i_inline_ea);
 499                         ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
 500                         goto size_check;
 501                 }
 502                 current_blocks = 0;
 503         } else {
 504                 if (!(ji->ea.flag & DXD_EXTENT)) {
 505                         jfs_error(sb, "ea_get: invalid ea.flag)");
 506                         return -EIO;
 507                 }
 508                 current_blocks = (ea_size + sb->s_blocksize - 1) >>
 509                     sb->s_blocksize_bits;
 510         }
 511         size = max(min_size, ea_size);
 512
 513         if (size > PSIZE) {
 514                 /*
 515                  * To keep the rest of the code simple.  Allocate a
 516                  * contiguous buffer to work with
 517                  */
 518                 ea_buf->xattr = kmalloc(size, GFP_KERNEL);
 519                 if (ea_buf->xattr == NULL)
 520                         return -ENOMEM;
 521
 522                 ea_buf->flag = EA_MALLOC;
 523                 ea_buf->max_size = (size + sb->s_blocksize - 1) &
 524                     ~(sb->s_blocksize - 1);
 525
 526                 if (ea_size == 0)
 527                         return 0;
 528
 529                 if ((rc = ea_read(inode, ea_buf->xattr))) {
 530                         kfree(ea_buf->xattr);
 531                         ea_buf->xattr = NULL;
 532                         return rc;
 533                 }
 534                 goto size_check;
 535         }
 536         blocks_needed = (min_size + sb->s_blocksize - 1) >>
 537             sb->s_blocksize_bits;
 538
 539         if (blocks_needed > current_blocks) {
 540                 /* Allocate new blocks to quota. */
 541                 rc = dquot_alloc_block(inode, blocks_needed);
 542                 if (rc)
 543                         return -EDQUOT;
 544
 545                 quota_allocation = blocks_needed;
 546
 547                 rc = dbAlloc(inode, INOHINT(inode), (s64) blocks_needed,
 548                              &blkno);
 549                 if (rc)
 550                         goto clean_up;
 551
 552                 DXDlength(&ea_buf->new_ea, blocks_needed);
 553                 DXDaddress(&ea_buf->new_ea, blkno);
 554                 ea_buf->new_ea.flag = DXD_EXTENT;
 555                 DXDsize(&ea_buf->new_ea, min_size);
 556
 557                 ea_buf->flag = EA_EXTENT | EA_NEW;
 558
 559                 ea_buf->mp = get_metapage(inode, blkno,
 560                                           blocks_needed << sb->s_blocksize_bits,
 561                                           1);
 562                 if (ea_buf->mp == NULL) {
 563                         dbFree(inode, blkno, (s64) blocks_needed);
 564                         rc = -EIO;
 565                         goto clean_up;
 566                 }
 567                 ea_buf->xattr = ea_buf->mp->data;
 568                 ea_buf->max_size = (min_size + sb->s_blocksize - 1) &
 569                     ~(sb->s_blocksize - 1);
 570                 if (ea_size == 0)
 571                         return 0;
 572                 if ((rc = ea_read(inode, ea_buf->xattr))) {
 573                         discard_metapage(ea_buf->mp);
 574                         dbFree(inode, blkno, (s64) blocks_needed);
 575                         goto clean_up;
 576                 }
 577                 goto size_check;
 578         }
 579         ea_buf->flag = EA_EXTENT;
 580         ea_buf->mp = read_metapage(inode, addressDXD(&ji->ea),
 581                                    lengthDXD(&ji->ea) << sb->s_blocksize_bits,
 582                                    1);
 583         if (ea_buf->mp == NULL) {
 584                 rc = -EIO;
 585                 goto clean_up;
 586         }
 587         ea_buf->xattr = ea_buf->mp->data;
 588         ea_buf->max_size = (ea_size + sb->s_blocksize - 1) &
 589             ~(sb->s_blocksize - 1);
 590
 591       size_check:
 592         if (EALIST_SIZE(ea_buf->xattr) != ea_size) {
 593                 printk(KERN_ERR "ea_get: invalid extended attribute\n");
 594                 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_ADDRESS, 16, 1,
 595                                      ea_buf->xattr, ea_size, 1);
 596                 ea_release(inode, ea_buf);
 597                 rc = -EIO;
 598                 goto clean_up;
 599         }
 600
 601         return ea_size;
 602
 603       clean_up:
 604         /* Rollback quota allocation */
 605         if (quota_allocation)
 606                 dquot_free_block(inode, quota_allocation);
 607
 608         return (rc);
 609 }
 610
 611 static void ea_release(struct inode *inode, struct ea_buffer *ea_buf)
 612 {
 613         if (ea_buf->flag & EA_MALLOC)
 614                 kfree(ea_buf->xattr);
 615         else if (ea_buf->flag & EA_EXTENT) {
 616                 assert(ea_buf->mp);
 617                 release_metapage(ea_buf->mp);
 618
 619                 if (ea_buf->flag & EA_NEW)
 620                         dbFree(inode, addressDXD(&ea_buf->new_ea),
 621                                lengthDXD(&ea_buf->new_ea));
 622         }
 623 }
 624
 625 static int ea_put(tid_t tid, struct inode *inode, struct ea_buffer *ea_buf,
 626                   int new_size)
 627 {
 628         struct jfs_inode_info *ji = JFS_IP(inode);
 629         unsigned long old_blocks, new_blocks;
 630         int rc = 0;
 631
 632         if (new_size == 0) {
 633                 ea_release(inode, ea_buf);
 634                 ea_buf = NULL;
 635         } else if (ea_buf->flag & EA_INLINE) {
 636                 assert(new_size <= sizeof (ji->i_inline_ea));
 637                 ji->mode2 &= ~INLINEEA;
 638                 ea_buf->new_ea.flag = DXD_INLINE;
 639                 DXDsize(&ea_buf->new_ea, new_size);
 640                 DXDaddress(&ea_buf->new_ea, 0);
 641                 DXDlength(&ea_buf->new_ea, 0);
 642         } else if (ea_buf->flag & EA_MALLOC) {
 643                 rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
 644                 kfree(ea_buf->xattr);
 645         } else if (ea_buf->flag & EA_NEW) {
 646                 /* We have already allocated a new dxd */
 647                 flush_metapage(ea_buf->mp);
 648         } else {
 649                 /* ->xattr must point to original ea's metapage */
 650                 rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
 651                 discard_metapage(ea_buf->mp);
 652         }
 653         if (rc)
 654                 return rc;
 655
 656         old_blocks = new_blocks = 0;
 657
 658         if (ji->ea.flag & DXD_EXTENT) {
 659                 invalidate_dxd_metapages(inode, ji->ea);
 660                 old_blocks = lengthDXD(&ji->ea);
 661         }
 662
 663         if (ea_buf) {
 664                 txEA(tid, inode, &ji->ea, &ea_buf->new_ea);
 665                 if (ea_buf->new_ea.flag & DXD_EXTENT) {
 666                         new_blocks = lengthDXD(&ea_buf->new_ea);
 667                         if (ji->ea.flag & DXD_INLINE)
 668                                 ji->mode2 |= INLINEEA;
 669                 }
 670                 ji->ea = ea_buf->new_ea;
 671         } else {
 672                 txEA(tid, inode, &ji->ea, NULL);
 673                 if (ji->ea.flag & DXD_INLINE)
 674                         ji->mode2 |= INLINEEA;
 675                 ji->ea.flag = 0;
 676                 ji->ea.size = 0;
 677         }
 678
 679         /* If old blocks exist, they must be removed from quota allocation. */
 680         if (old_blocks)
 681                 dquot_free_block(inode, old_blocks);
 682
 683         inode->i_ctime = CURRENT_TIME;
 684
 685         return 0;
 686 }
 687
 688 /*
 689  * can_set_system_xattr
 690  *
 691  * This code is specific to the system.* namespace.  It contains policy
 692  * which doesn't belong in the main xattr codepath.
 693  */
 694 static int can_set_system_xattr(struct inode *inode, const char *name,
 695                                 const void *value, size_t value_len)
 696 {
 697 #ifdef CONFIG_JFS_POSIX_ACL
 698         struct posix_acl *acl;
 699         int rc;
 700
 701         if (!is_owner_or_cap(inode))
 702                 return -EPERM;
 703
 704         /*
 705          * POSIX_ACL_XATTR_ACCESS is tied to i_mode
 706          */
 707         if (strcmp(name, POSIX_ACL_XATTR_ACCESS) == 0) {
 708                 acl = posix_acl_from_xattr(value, value_len);
 709                 if (IS_ERR(acl)) {
 710                         rc = PTR_ERR(acl);
 711                         printk(KERN_ERR "posix_acl_from_xattr returned %d\n",
 712                                rc);
 713                         return rc;
 714                 }
 715                 if (acl) {
 716                         mode_t mode = inode->i_mode;
 717                         rc = posix_acl_equiv_mode(acl, &mode);
 718                         posix_acl_release(acl);
 719                         if (rc < 0) {
 720                                 printk(KERN_ERR
 721                                        "posix_acl_equiv_mode returned %d\n",
 722                                        rc);
 723                                 return rc;
 724                         }
 725                         inode->i_mode = mode;
 726                         mark_inode_dirty(inode);
 727                 }
 728                 /*
 729                  * We're changing the ACL.  Get rid of the cached one
 730                  */
 731                 forget_cached_acl(inode, ACL_TYPE_ACCESS);
 732
 733                 return 0;
 734         } else if (strcmp(name, POSIX_ACL_XATTR_DEFAULT) == 0) {
 735                 acl = posix_acl_from_xattr(value, value_len);
 736                 if (IS_ERR(acl)) {
 737                         rc = PTR_ERR(acl);
 738                         printk(KERN_ERR "posix_acl_from_xattr returned %d\n",
 739                                rc);
 740                         return rc;
 741                 }
 742                 posix_acl_release(acl);
 743
 744                 /*
 745                  * We're changing the default ACL.  Get rid of the cached one
 746                  */
 747                 forget_cached_acl(inode, ACL_TYPE_DEFAULT);
 748
 749                 return 0;
 750         }
 751 #endif                  /* CONFIG_JFS_POSIX_ACL */
 752         return -EOPNOTSUPP;
 753 }
 754
 755 /*
 756  * Most of the permission checking is done by xattr_permission in the vfs.
 757  * The local file system is responsible for handling the system.* namespace.
 758  * We also need to verify that this is a namespace that we recognize.
 759  */
 760 static int can_set_xattr(struct inode *inode, const char *name,
 761                          const void *value, size_t value_len)
 762 {
 763         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
 764                 return can_set_system_xattr(inode, name, value, value_len);
 765
 766         /*
 767          * Don't allow setting an attribute in an unknown namespace.
 768          */
 769         if (strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) &&
 770             strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
 771             strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN) &&
 772             strncmp(name, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN))
 773                 return -EOPNOTSUPP;
 774
 775         return 0;
 776 }
 777
 778 int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
 779                    const void *value, size_t value_len, int flags)
 780 {
 781         struct jfs_ea_list *ealist;
 782         struct jfs_ea *ea, *old_ea = NULL, *next_ea = NULL;
 783         struct ea_buffer ea_buf;
 784         int old_ea_size = 0;
 785         int xattr_size;
 786         int new_size;
 787         int namelen = strlen(name);
 788         char *os2name = NULL;
 789         int found = 0;
 790         int rc;
 791         int length;
 792
 793         if (strncmp(name, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN) == 0) {
 794                 os2name = kmalloc(namelen - XATTR_OS2_PREFIX_LEN + 1,
 795                                   GFP_KERNEL);
 796                 if (!os2name)
 797                         return -ENOMEM;
 798                 strcpy(os2name, name + XATTR_OS2_PREFIX_LEN);
 799                 name = os2name;
 800                 namelen -= XATTR_OS2_PREFIX_LEN;
 801         }
 802
 803         down_write(&JFS_IP(inode)->xattr_sem);
 804
 805         xattr_size = ea_get(inode, &ea_buf, 0);
 806         if (xattr_size < 0) {
 807                 rc = xattr_size;
 808                 goto out;
 809         }
 810
 811       again:
 812         ealist = (struct jfs_ea_list *) ea_buf.xattr;
 813         new_size = sizeof (struct jfs_ea_list);
 814
 815         if (xattr_size) {
 816                 for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
 817                      ea = NEXT_EA(ea)) {
 818                         if ((namelen == ea->namelen) &&
 819                             (memcmp(name, ea->name, namelen) == 0)) {
 820                                 found = 1;
 821                                 if (flags & XATTR_CREATE) {
 822                                         rc = -EEXIST;
 823                                         goto release;
 824                                 }
 825                                 old_ea = ea;
 826                                 old_ea_size = EA_SIZE(ea);
 827                                 next_ea = NEXT_EA(ea);
 828                         } else
 829                                 new_size += EA_SIZE(ea);
 830                 }
 831         }
 832
 833         if (!found) {
 834                 if (flags & XATTR_REPLACE) {
 835                         rc = -ENODATA;
 836                         goto release;
 837                 }
 838                 if (value == NULL) {
 839                         rc = 0;
 840                         goto release;
 841                 }
 842         }
 843         if (value)
 844                 new_size += sizeof (struct jfs_ea) + namelen + 1 + value_len;
 845
 846         if (new_size > ea_buf.max_size) {
 847                 /*
 848                  * We need to allocate more space for merged ea list.
 849                  * We should only have loop to again: once.
 850                  */
 851                 ea_release(inode, &ea_buf);
 852                 xattr_size = ea_get(inode, &ea_buf, new_size);
 853                 if (xattr_size < 0) {
 854                         rc = xattr_size;
 855                         goto out;
 856                 }
 857                 goto again;
 858         }
 859
 860         /* Remove old ea of the same name */
 861         if (found) {
 862                 /* number of bytes following target EA */
 863                 length = (char *) END_EALIST(ealist) - (char *) next_ea;
 864                 if (length > 0)
 865                         memmove(old_ea, next_ea, length);
 866                 xattr_size -= old_ea_size;
 867         }
 868
 869         /* Add new entry to the end */
 870         if (value) {
 871                 if (xattr_size == 0)
 872                         /* Completely new ea list */
 873                         xattr_size = sizeof (struct jfs_ea_list);
 874
 875                 ea = (struct jfs_ea *) ((char *) ealist + xattr_size);
 876                 ea->flag = 0;
 877                 ea->namelen = namelen;
 878                 ea->valuelen = (cpu_to_le16(value_len));
 879                 memcpy(ea->name, name, namelen);
 880                 ea->name[namelen] = 0;
 881                 if (value_len)
 882                         memcpy(&ea->name[namelen + 1], value, value_len);
 883                 xattr_size += EA_SIZE(ea);
 884         }
 885
 886         /* DEBUG - If we did this right, these number match */
 887         if (xattr_size != new_size) {
 888                 printk(KERN_ERR
 889                        "jfs_xsetattr: xattr_size = %d, new_size = %d\n",
 890                        xattr_size, new_size);
 891
 892                 rc = -EINVAL;
 893                 goto release;
 894         }
 895
 896         /*
 897          * If we're left with an empty list, there's no ea
 898          */
 899         if (new_size == sizeof (struct jfs_ea_list))
 900                 new_size = 0;
 901
 902         ealist->size = cpu_to_le32(new_size);
 903
 904         rc = ea_put(tid, inode, &ea_buf, new_size);
 905
 906         goto out;
 907       release:
 908         ea_release(inode, &ea_buf);
 909       out:
 910         up_write(&JFS_IP(inode)->xattr_sem);
 911
 912         kfree(os2name);
 913
 914         return rc;
 915 }
 916
 917 int jfs_setxattr(struct dentry *dentry, const char *name, const void *value,
 918                  size_t value_len, int flags)
 919 {
 920         struct inode *inode = dentry->d_inode;
 921         struct jfs_inode_info *ji = JFS_IP(inode);
 922         int rc;
 923         tid_t tid;
 924
 925         if ((rc = can_set_xattr(inode, name, value, value_len)))
 926                 return rc;
 927
 928         if (value == NULL) {    /* empty EA, do not remove */
 929                 value = "";
 930                 value_len = 0;
 931         }
 932
 933         tid = txBegin(inode->i_sb, 0);
 934         mutex_lock(&ji->commit_mutex);
 935         rc = __jfs_setxattr(tid, dentry->d_inode, name, value, value_len,
 936                             flags);
 937         if (!rc)
 938                 rc = txCommit(tid, 1, &inode, 0);
 939         txEnd(tid);
 940         mutex_unlock(&ji->commit_mutex);
 941
 942         return rc;
 943 }
 944
 945 ssize_t __jfs_getxattr(struct inode *inode, const char *name, void *data,
 946                        size_t buf_size)
 947 {
 948         struct jfs_ea_list *ealist;
 949         struct jfs_ea *ea;
 950         struct ea_buffer ea_buf;
 951         int xattr_size;
 952         ssize_t size;
 953         int namelen = strlen(name);
 954         char *os2name = NULL;
 955         char *value;
 956
 957         if (strncmp(name, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN) == 0) {
 958                 os2name = kmalloc(namelen - XATTR_OS2_PREFIX_LEN + 1,
 959                                   GFP_KERNEL);
 960                 if (!os2name)
 961                         return -ENOMEM;
 962                 strcpy(os2name, name + XATTR_OS2_PREFIX_LEN);
 963                 name = os2name;
 964                 namelen -= XATTR_OS2_PREFIX_LEN;
 965         }
 966
 967         down_read(&JFS_IP(inode)->xattr_sem);
 968
 969         xattr_size = ea_get(inode, &ea_buf, 0);
 970
 971         if (xattr_size < 0) {
 972                 size = xattr_size;
 973                 goto out;
 974         }
 975
 976         if (xattr_size == 0)
 977                 goto not_found;
 978
 979         ealist = (struct jfs_ea_list *) ea_buf.xattr;
 980
 981         /* Find the named attribute */
 982         for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea))
 983                 if ((namelen == ea->namelen) &&
 984                     memcmp(name, ea->name, namelen) == 0) {
 985                         /* Found it */
 986                         size = le16_to_cpu(ea->valuelen);
 987                         if (!data)
 988                                 goto release;
 989                         else if (size > buf_size) {
 990                                 size = -ERANGE;
 991                                 goto release;
 992                         }
 993                         value = ((char *) &ea->name) + ea->namelen + 1;
 994                         memcpy(data, value, size);
 995                         goto release;
 996                 }
 997       not_found:
 998         size = -ENODATA;
 999       release:
1000         ea_release(inode, &ea_buf);
1001       out:
1002         up_read(&JFS_IP(inode)->xattr_sem);
1003
1004         kfree(os2name);
1005
1006         return size;
1007 }
1008
1009 ssize_t jfs_getxattr(struct dentry *dentry, const char *name, void *data,
1010                      size_t buf_size)
1011 {
1012         int err;
1013
1014         err = __jfs_getxattr(dentry->d_inode, name, data, buf_size);
1015
1016         return err;
1017 }
1018
1019 /*
1020  * No special permissions are needed to list attributes except for trusted.*
1021  */
1022 static inline int can_list(struct jfs_ea *ea)
1023 {
1024         return (strncmp(ea->name, XATTR_TRUSTED_PREFIX,
1025                             XATTR_TRUSTED_PREFIX_LEN) ||
1026                 capable(CAP_SYS_ADMIN));
1027 }
1028
1029 ssize_t jfs_listxattr(struct dentry * dentry, char *data, size_t buf_size)
1030 {
1031         struct inode *inode = dentry->d_inode;
1032         char *buffer;
1033         ssize_t size = 0;
1034         int xattr_size;
1035         struct jfs_ea_list *ealist;
1036         struct jfs_ea *ea;
1037         struct ea_buffer ea_buf;
1038
1039         down_read(&JFS_IP(inode)->xattr_sem);
1040
1041         xattr_size = ea_get(inode, &ea_buf, 0);
1042         if (xattr_size < 0) {
1043                 size = xattr_size;
1044                 goto out;
1045         }
1046
1047         if (xattr_size == 0)
1048                 goto release;
1049
1050         ealist = (struct jfs_ea_list *) ea_buf.xattr;
1051
1052         /* compute required size of list */
1053         for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
1054                 if (can_list(ea))
1055                         size += name_size(ea) + 1;
1056         }
1057
1058         if (!data)
1059                 goto release;
1060
1061         if (size > buf_size) {
1062                 size = -ERANGE;
1063                 goto release;
1064         }
1065
1066         /* Copy attribute names to buffer */
1067         buffer = data;
1068         for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
1069                 if (can_list(ea)) {
1070                         int namelen = copy_name(buffer, ea);
1071                         buffer += namelen + 1;
1072                 }
1073         }
1074
1075       release:
1076         ea_release(inode, &ea_buf);
1077       out:
1078         up_read(&JFS_IP(inode)->xattr_sem);
1079         return size;
1080 }
1081
1082 int jfs_removexattr(struct dentry *dentry, const char *name)
1083 {
1084         struct inode *inode = dentry->d_inode;
1085         struct jfs_inode_info *ji = JFS_IP(inode);
1086         int rc;
1087         tid_t tid;
1088
1089         if ((rc = can_set_xattr(inode, name, NULL, 0)))
1090                 return rc;
1091
1092         tid = txBegin(inode->i_sb, 0);
1093         mutex_lock(&ji->commit_mutex);
1094         rc = __jfs_setxattr(tid, dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
1095         if (!rc)
1096                 rc = txCommit(tid, 1, &inode, 0);
1097         txEnd(tid);
1098         mutex_unlock(&ji->commit_mutex);
1099
1100         return rc;
1101 }
1102
1103 #ifdef CONFIG_JFS_SECURITY
1104 int jfs_init_security(tid_t tid, struct inode *inode, struct inode *dir)
1105 {
1106         int rc;
1107         size_t len;
1108         void *value;
1109         char *suffix;
1110         char *name;
1111
1112         rc = security_inode_init_security(inode, dir, &suffix, &value, &len);
1113         if (rc) {
1114                 if (rc == -EOPNOTSUPP)
1115                         return 0;
1116                 return rc;
1117         }
1118         name = kmalloc(XATTR_SECURITY_PREFIX_LEN + 1 + strlen(suffix),
1119                        GFP_NOFS);
1120         if (!name) {
1121                 rc = -ENOMEM;
1122                 goto kmalloc_failed;
1123         }
1124         strcpy(name, XATTR_SECURITY_PREFIX);
1125         strcpy(name + XATTR_SECURITY_PREFIX_LEN, suffix);
1126
1127         rc = __jfs_setxattr(tid, inode, name, value, len, 0);
1128
1129         kfree(name);
1130 kmalloc_failed:
1131         kfree(suffix);
1132         kfree(value);
1133
1134         return rc;
1135 }
1136 #endif