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/slab.h>
  25#include <linux/quotaops.h>
  26#include <linux/security.h>
  27#include "jfs_incore.h"
  28#include "jfs_superblock.h"
  29#include "jfs_dmap.h"
  30#include "jfs_debug.h"
  31#include "jfs_dinode.h"
  32#include "jfs_extent.h"
  33#include "jfs_metapage.h"
  34#include "jfs_xattr.h"
  35#include "jfs_acl.h"
  36
  37/*
  38 *	jfs_xattr.c: extended attribute service
  39 *
  40 * Overall design --
  41 *
  42 * Format:
  43 *
  44 *   Extended attribute lists (jfs_ea_list) consist of an overall size (32 bit
  45 *   value) and a variable (0 or more) number of extended attribute
  46 *   entries.  Each extended attribute entry (jfs_ea) is a <name,value> double
  47 *   where <name> is constructed from a null-terminated ascii string
  48 *   (1 ... 255 bytes in the name) and <value> is arbitrary 8 bit data
  49 *   (1 ... 65535 bytes).  The in-memory format is
  50 *
  51 *   0       1        2        4                4 + namelen + 1
  52 *   +-------+--------+--------+----------------+-------------------+
  53 *   | Flags | Name   | Value  | Name String \0 | Data . . . .      |
  54 *   |       | Length | Length |                |                   |
  55 *   +-------+--------+--------+----------------+-------------------+
  56 *
  57 *   A jfs_ea_list then is structured as
  58 *
  59 *   0            4                   4 + EA_SIZE(ea1)
  60 *   +------------+-------------------+--------------------+-----
  61 *   | Overall EA | First FEA Element | Second FEA Element | .....
  62 *   | List Size  |                   |                    |
  63 *   +------------+-------------------+--------------------+-----
  64 *
  65 *   On-disk:
  66 *
  67 *	FEALISTs are stored on disk using blocks allocated by dbAlloc() and
  68 *	written directly. An EA list may be in-lined in the inode if there is
  69 *	sufficient room available.
  70 */
  71
  72struct ea_buffer {
  73	int flag;		/* Indicates what storage xattr points to */
  74	int max_size;		/* largest xattr that fits in current buffer */
  75	dxd_t new_ea;		/* dxd to replace ea when modifying xattr */
  76	struct metapage *mp;	/* metapage containing ea list */
  77	struct jfs_ea_list *xattr;	/* buffer containing ea list */
  78};
  79
  80/*
  81 * ea_buffer.flag values
  82 */
  83#define EA_INLINE	0x0001
  84#define EA_EXTENT	0x0002
  85#define EA_NEW		0x0004
  86#define EA_MALLOC	0x0008
  87
  88
  89static int is_known_namespace(const char *name)
  90{
  91	if (strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) &&
  92	    strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN) &&
  93	    strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
  94	    strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN))
  95		return false;
  96
  97	return true;
  98}
  99
 100/*
 101 * These three routines are used to recognize on-disk extended attributes
 102 * that are in a recognized namespace.  If the attribute is not recognized,
 103 * "os2." is prepended to the name
 104 */
 105static int is_os2_xattr(struct jfs_ea *ea)
 106{
 107	return !is_known_namespace(ea->name);
 108}
 109
 110static inline int name_size(struct jfs_ea *ea)
 111{
 112	if (is_os2_xattr(ea))
 113		return ea->namelen + XATTR_OS2_PREFIX_LEN;
 114	else
 115		return ea->namelen;
 116}
 117
 118static inline int copy_name(char *buffer, struct jfs_ea *ea)
 119{
 120	int len = ea->namelen;
 121
 122	if (is_os2_xattr(ea)) {
 123		memcpy(buffer, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN);
 124		buffer += XATTR_OS2_PREFIX_LEN;
 125		len += XATTR_OS2_PREFIX_LEN;
 126	}
 127	memcpy(buffer, ea->name, ea->namelen);
 128	buffer[ea->namelen] = 0;
 129
 130	return len;
 131}
 132
 133/* Forward references */
 134static void ea_release(struct inode *inode, struct ea_buffer *ea_buf);
 135
 136/*
 137 * NAME: ea_write_inline
 138 *
 139 * FUNCTION: Attempt to write an EA inline if area is available
 140 *
 141 * PRE CONDITIONS:
 142 *	Already verified that the specified EA is small enough to fit inline
 143 *
 144 * PARAMETERS:
 145 *	ip	- Inode pointer
 146 *	ealist	- EA list pointer
 147 *	size	- size of ealist in bytes
 148 *	ea	- dxd_t structure to be filled in with necessary EA information
 149 *		  if we successfully copy the EA inline
 150 *
 151 * NOTES:
 152 *	Checks if the inode's inline area is available.  If so, copies EA inline
 153 *	and sets <ea> fields appropriately.  Otherwise, returns failure, EA will
 154 *	have to be put into an extent.
 155 *
 156 * RETURNS: 0 for successful copy to inline area; -1 if area not available
 157 */
 158static int ea_write_inline(struct inode *ip, struct jfs_ea_list *ealist,
 159			   int size, dxd_t * ea)
 160{
 161	struct jfs_inode_info *ji = JFS_IP(ip);
 162
 163	/*
 164	 * Make sure we have an EA -- the NULL EA list is valid, but you
 165	 * can't copy it!
 166	 */
 167	if (ealist && size > sizeof (struct jfs_ea_list)) {
 168		assert(size <= sizeof (ji->i_inline_ea));
 169
 170		/*
 171		 * See if the space is available or if it is already being
 172		 * used for an inline EA.
 173		 */
 174		if (!(ji->mode2 & INLINEEA) && !(ji->ea.flag & DXD_INLINE))
 175			return -EPERM;
 176
 177		DXDsize(ea, size);
 178		DXDlength(ea, 0);
 179		DXDaddress(ea, 0);
 180		memcpy(ji->i_inline_ea, ealist, size);
 181		ea->flag = DXD_INLINE;
 182		ji->mode2 &= ~INLINEEA;
 183	} else {
 184		ea->flag = 0;
 185		DXDsize(ea, 0);
 186		DXDlength(ea, 0);
 187		DXDaddress(ea, 0);
 188
 189		/* Free up INLINE area */
 190		if (ji->ea.flag & DXD_INLINE)
 191			ji->mode2 |= INLINEEA;
 192	}
 193
 194	return 0;
 195}
 196
 197/*
 198 * NAME: ea_write
 199 *
 200 * FUNCTION: Write an EA for an inode
 201 *
 202 * PRE CONDITIONS: EA has been verified
 203 *
 204 * PARAMETERS:
 205 *	ip	- Inode pointer
 206 *	ealist	- EA list pointer
 207 *	size	- size of ealist in bytes
 208 *	ea	- dxd_t structure to be filled in appropriately with where the
 209 *		  EA was copied
 210 *
 211 * NOTES: Will write EA inline if able to, otherwise allocates blocks for an
 212 *	extent and synchronously writes it to those blocks.
 213 *
 214 * RETURNS: 0 for success; Anything else indicates failure
 215 */
 216static int ea_write(struct inode *ip, struct jfs_ea_list *ealist, int size,
 217		       dxd_t * ea)
 218{
 219	struct super_block *sb = ip->i_sb;
 220	struct jfs_inode_info *ji = JFS_IP(ip);
 221	struct jfs_sb_info *sbi = JFS_SBI(sb);
 222	int nblocks;
 223	s64 blkno;
 224	int rc = 0, i;
 225	char *cp;
 226	s32 nbytes, nb;
 227	s32 bytes_to_write;
 228	struct metapage *mp;
 229
 230	/*
 231	 * Quick check to see if this is an in-linable EA.  Short EAs
 232	 * and empty EAs are all in-linable, provided the space exists.
 233	 */
 234	if (!ealist || size <= sizeof (ji->i_inline_ea)) {
 235		if (!ea_write_inline(ip, ealist, size, ea))
 236			return 0;
 237	}
 238
 239	/* figure out how many blocks we need */
 240	nblocks = (size + (sb->s_blocksize - 1)) >> sb->s_blocksize_bits;
 241
 242	/* Allocate new blocks to quota. */
 243	rc = dquot_alloc_block(ip, nblocks);
 244	if (rc)
 245		return rc;
 246
 247	rc = dbAlloc(ip, INOHINT(ip), nblocks, &blkno);
 248	if (rc) {
 249		/*Rollback quota allocation. */
 250		dquot_free_block(ip, nblocks);
 251		return rc;
 252	}
 253
 254	/*
 255	 * Now have nblocks worth of storage to stuff into the FEALIST.
 256	 * loop over the FEALIST copying data into the buffer one page at
 257	 * a time.
 258	 */
 259	cp = (char *) ealist;
 260	nbytes = size;
 261	for (i = 0; i < nblocks; i += sbi->nbperpage) {
 262		/*
 263		 * Determine how many bytes for this request, and round up to
 264		 * the nearest aggregate block size
 265		 */
 266		nb = min(PSIZE, nbytes);
 267		bytes_to_write =
 268		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
 269		    << sb->s_blocksize_bits;
 270
 271		if (!(mp = get_metapage(ip, blkno + i, bytes_to_write, 1))) {
 272			rc = -EIO;
 273			goto failed;
 274		}
 275
 276		memcpy(mp->data, cp, nb);
 277
 278		/*
 279		 * We really need a way to propagate errors for
 280		 * forced writes like this one.  --hch
 281		 *
 282		 * (__write_metapage => release_metapage => flush_metapage)
 283		 */
 284#ifdef _JFS_FIXME
 285		if ((rc = flush_metapage(mp))) {
 286			/*
 287			 * the write failed -- this means that the buffer
 288			 * is still assigned and the blocks are not being
 289			 * used.  this seems like the best error recovery
 290			 * we can get ...
 291			 */
 292			goto failed;
 293		}
 294#else
 295		flush_metapage(mp);
 296#endif
 297
 298		cp += PSIZE;
 299		nbytes -= nb;
 300	}
 301
 302	ea->flag = DXD_EXTENT;
 303	DXDsize(ea, le32_to_cpu(ealist->size));
 304	DXDlength(ea, nblocks);
 305	DXDaddress(ea, blkno);
 306
 307	/* Free up INLINE area */
 308	if (ji->ea.flag & DXD_INLINE)
 309		ji->mode2 |= INLINEEA;
 310
 311	return 0;
 312
 313      failed:
 314	/* Rollback quota allocation. */
 315	dquot_free_block(ip, nblocks);
 316
 317	dbFree(ip, blkno, nblocks);
 318	return rc;
 319}
 320
 321/*
 322 * NAME: ea_read_inline
 323 *
 324 * FUNCTION: Read an inlined EA into user's buffer
 325 *
 326 * PARAMETERS:
 327 *	ip	- Inode pointer
 328 *	ealist	- Pointer to buffer to fill in with EA
 329 *
 330 * RETURNS: 0
 331 */
 332static int ea_read_inline(struct inode *ip, struct jfs_ea_list *ealist)
 333{
 334	struct jfs_inode_info *ji = JFS_IP(ip);
 335	int ea_size = sizeDXD(&ji->ea);
 336
 337	if (ea_size == 0) {
 338		ealist->size = 0;
 339		return 0;
 340	}
 341
 342	/* Sanity Check */
 343	if ((sizeDXD(&ji->ea) > sizeof (ji->i_inline_ea)))
 344		return -EIO;
 345	if (le32_to_cpu(((struct jfs_ea_list *) &ji->i_inline_ea)->size)
 346	    != ea_size)
 347		return -EIO;
 348
 349	memcpy(ealist, ji->i_inline_ea, ea_size);
 350	return 0;
 351}
 352
 353/*
 354 * NAME: ea_read
 355 *
 356 * FUNCTION: copy EA data into user's buffer
 357 *
 358 * PARAMETERS:
 359 *	ip	- Inode pointer
 360 *	ealist	- Pointer to buffer to fill in with EA
 361 *
 362 * NOTES:  If EA is inline calls ea_read_inline() to copy EA.
 363 *
 364 * RETURNS: 0 for success; other indicates failure
 365 */
 366static int ea_read(struct inode *ip, struct jfs_ea_list *ealist)
 367{
 368	struct super_block *sb = ip->i_sb;
 369	struct jfs_inode_info *ji = JFS_IP(ip);
 370	struct jfs_sb_info *sbi = JFS_SBI(sb);
 371	int nblocks;
 372	s64 blkno;
 373	char *cp = (char *) ealist;
 374	int i;
 375	int nbytes, nb;
 376	s32 bytes_to_read;
 377	struct metapage *mp;
 378
 379	/* quick check for in-line EA */
 380	if (ji->ea.flag & DXD_INLINE)
 381		return ea_read_inline(ip, ealist);
 382
 383	nbytes = sizeDXD(&ji->ea);
 384	if (!nbytes) {
 385		jfs_error(sb, "nbytes is 0\n");
 386		return -EIO;
 387	}
 388
 389	/*
 390	 * Figure out how many blocks were allocated when this EA list was
 391	 * originally written to disk.
 392	 */
 393	nblocks = lengthDXD(&ji->ea) << sbi->l2nbperpage;
 394	blkno = addressDXD(&ji->ea) << sbi->l2nbperpage;
 395
 396	/*
 397	 * I have found the disk blocks which were originally used to store
 398	 * the FEALIST.  now i loop over each contiguous block copying the
 399	 * data into the buffer.
 400	 */
 401	for (i = 0; i < nblocks; i += sbi->nbperpage) {
 402		/*
 403		 * Determine how many bytes for this request, and round up to
 404		 * the nearest aggregate block size
 405		 */
 406		nb = min(PSIZE, nbytes);
 407		bytes_to_read =
 408		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
 409		    << sb->s_blocksize_bits;
 410
 411		if (!(mp = read_metapage(ip, blkno + i, bytes_to_read, 1)))
 412			return -EIO;
 413
 414		memcpy(cp, mp->data, nb);
 415		release_metapage(mp);
 416
 417		cp += PSIZE;
 418		nbytes -= nb;
 419	}
 420
 421	return 0;
 422}
 423
 424/*
 425 * NAME: ea_get
 426 *
 427 * FUNCTION: Returns buffer containing existing extended attributes.
 428 *	     The size of the buffer will be the larger of the existing
 429 *	     attributes size, or min_size.
 430 *
 431 *	     The buffer, which may be inlined in the inode or in the
 432 *	     page cache must be release by calling ea_release or ea_put
 433 *
 434 * PARAMETERS:
 435 *	inode	- Inode pointer
 436 *	ea_buf	- Structure to be populated with ealist and its metadata
 437 *	min_size- minimum size of buffer to be returned
 438 *
 439 * RETURNS: 0 for success; Other indicates failure
 440 */
 441static int ea_get(struct inode *inode, struct ea_buffer *ea_buf, int min_size)
 442{
 443	struct jfs_inode_info *ji = JFS_IP(inode);
 444	struct super_block *sb = inode->i_sb;
 445	int size;
 446	int ea_size = sizeDXD(&ji->ea);
 447	int blocks_needed, current_blocks;
 448	s64 blkno;
 449	int rc;
 450	int quota_allocation = 0;
 451
 452	/* When fsck.jfs clears a bad ea, it doesn't clear the size */
 453	if (ji->ea.flag == 0)
 454		ea_size = 0;
 455
 456	if (ea_size == 0) {
 457		if (min_size == 0) {
 458			ea_buf->flag = 0;
 459			ea_buf->max_size = 0;
 460			ea_buf->xattr = NULL;
 461			return 0;
 462		}
 463		if ((min_size <= sizeof (ji->i_inline_ea)) &&
 464		    (ji->mode2 & INLINEEA)) {
 465			ea_buf->flag = EA_INLINE | EA_NEW;
 466			ea_buf->max_size = sizeof (ji->i_inline_ea);
 467			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
 468			DXDlength(&ea_buf->new_ea, 0);
 469			DXDaddress(&ea_buf->new_ea, 0);
 470			ea_buf->new_ea.flag = DXD_INLINE;
 471			DXDsize(&ea_buf->new_ea, min_size);
 472			return 0;
 473		}
 474		current_blocks = 0;
 475	} else if (ji->ea.flag & DXD_INLINE) {
 476		if (min_size <= sizeof (ji->i_inline_ea)) {
 477			ea_buf->flag = EA_INLINE;
 478			ea_buf->max_size = sizeof (ji->i_inline_ea);
 479			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
 480			goto size_check;
 481		}
 482		current_blocks = 0;
 483	} else {
 484		if (!(ji->ea.flag & DXD_EXTENT)) {
 485			jfs_error(sb, "invalid ea.flag\n");
 486			return -EIO;
 487		}
 488		current_blocks = (ea_size + sb->s_blocksize - 1) >>
 489		    sb->s_blocksize_bits;
 490	}
 491	size = max(min_size, ea_size);
 492
 493	if (size > PSIZE) {
 494		/*
 495		 * To keep the rest of the code simple.  Allocate a
 496		 * contiguous buffer to work with
 497		 */
 498		ea_buf->xattr = kmalloc(size, GFP_KERNEL);
 499		if (ea_buf->xattr == NULL)
 500			return -ENOMEM;
 501
 502		ea_buf->flag = EA_MALLOC;
 503		ea_buf->max_size = (size + sb->s_blocksize - 1) &
 504		    ~(sb->s_blocksize - 1);
 505
 506		if (ea_size == 0)
 507			return 0;
 508
 509		if ((rc = ea_read(inode, ea_buf->xattr))) {
 510			kfree(ea_buf->xattr);
 511			ea_buf->xattr = NULL;
 512			return rc;
 513		}
 514		goto size_check;
 515	}
 516	blocks_needed = (min_size + sb->s_blocksize - 1) >>
 517	    sb->s_blocksize_bits;
 518
 519	if (blocks_needed > current_blocks) {
 520		/* Allocate new blocks to quota. */
 521		rc = dquot_alloc_block(inode, blocks_needed);
 522		if (rc)
 523			return -EDQUOT;
 524
 525		quota_allocation = blocks_needed;
 526
 527		rc = dbAlloc(inode, INOHINT(inode), (s64) blocks_needed,
 528			     &blkno);
 529		if (rc)
 530			goto clean_up;
 531
 532		DXDlength(&ea_buf->new_ea, blocks_needed);
 533		DXDaddress(&ea_buf->new_ea, blkno);
 534		ea_buf->new_ea.flag = DXD_EXTENT;
 535		DXDsize(&ea_buf->new_ea, min_size);
 536
 537		ea_buf->flag = EA_EXTENT | EA_NEW;
 538
 539		ea_buf->mp = get_metapage(inode, blkno,
 540					  blocks_needed << sb->s_blocksize_bits,
 541					  1);
 542		if (ea_buf->mp == NULL) {
 543			dbFree(inode, blkno, (s64) blocks_needed);
 544			rc = -EIO;
 545			goto clean_up;
 546		}
 547		ea_buf->xattr = ea_buf->mp->data;
 548		ea_buf->max_size = (min_size + sb->s_blocksize - 1) &
 549		    ~(sb->s_blocksize - 1);
 550		if (ea_size == 0)
 551			return 0;
 552		if ((rc = ea_read(inode, ea_buf->xattr))) {
 553			discard_metapage(ea_buf->mp);
 554			dbFree(inode, blkno, (s64) blocks_needed);
 555			goto clean_up;
 556		}
 557		goto size_check;
 558	}
 559	ea_buf->flag = EA_EXTENT;
 560	ea_buf->mp = read_metapage(inode, addressDXD(&ji->ea),
 561				   lengthDXD(&ji->ea) << sb->s_blocksize_bits,
 562				   1);
 563	if (ea_buf->mp == NULL) {
 564		rc = -EIO;
 565		goto clean_up;
 566	}
 567	ea_buf->xattr = ea_buf->mp->data;
 568	ea_buf->max_size = (ea_size + sb->s_blocksize - 1) &
 569	    ~(sb->s_blocksize - 1);
 570
 571      size_check:
 572	if (EALIST_SIZE(ea_buf->xattr) != ea_size) {
 573		printk(KERN_ERR "ea_get: invalid extended attribute\n");
 574		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_ADDRESS, 16, 1,
 575				     ea_buf->xattr, ea_size, 1);
 576		ea_release(inode, ea_buf);
 577		rc = -EIO;
 578		goto clean_up;
 579	}
 580
 581	return ea_size;
 582
 583      clean_up:
 584	/* Rollback quota allocation */
 585	if (quota_allocation)
 586		dquot_free_block(inode, quota_allocation);
 587
 588	return (rc);
 589}
 590
 591static void ea_release(struct inode *inode, struct ea_buffer *ea_buf)
 592{
 593	if (ea_buf->flag & EA_MALLOC)
 594		kfree(ea_buf->xattr);
 595	else if (ea_buf->flag & EA_EXTENT) {
 596		assert(ea_buf->mp);
 597		release_metapage(ea_buf->mp);
 598
 599		if (ea_buf->flag & EA_NEW)
 600			dbFree(inode, addressDXD(&ea_buf->new_ea),
 601			       lengthDXD(&ea_buf->new_ea));
 602	}
 603}
 604
 605static int ea_put(tid_t tid, struct inode *inode, struct ea_buffer *ea_buf,
 606		  int new_size)
 607{
 608	struct jfs_inode_info *ji = JFS_IP(inode);
 609	unsigned long old_blocks, new_blocks;
 610	int rc = 0;
 611
 612	if (new_size == 0) {
 613		ea_release(inode, ea_buf);
 614		ea_buf = NULL;
 615	} else if (ea_buf->flag & EA_INLINE) {
 616		assert(new_size <= sizeof (ji->i_inline_ea));
 617		ji->mode2 &= ~INLINEEA;
 618		ea_buf->new_ea.flag = DXD_INLINE;
 619		DXDsize(&ea_buf->new_ea, new_size);
 620		DXDaddress(&ea_buf->new_ea, 0);
 621		DXDlength(&ea_buf->new_ea, 0);
 622	} else if (ea_buf->flag & EA_MALLOC) {
 623		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
 624		kfree(ea_buf->xattr);
 625	} else if (ea_buf->flag & EA_NEW) {
 626		/* We have already allocated a new dxd */
 627		flush_metapage(ea_buf->mp);
 628	} else {
 629		/* ->xattr must point to original ea's metapage */
 630		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
 631		discard_metapage(ea_buf->mp);
 632	}
 633	if (rc)
 634		return rc;
 635
 636	old_blocks = new_blocks = 0;
 637
 638	if (ji->ea.flag & DXD_EXTENT) {
 639		invalidate_dxd_metapages(inode, ji->ea);
 640		old_blocks = lengthDXD(&ji->ea);
 641	}
 642
 643	if (ea_buf) {
 644		txEA(tid, inode, &ji->ea, &ea_buf->new_ea);
 645		if (ea_buf->new_ea.flag & DXD_EXTENT) {
 646			new_blocks = lengthDXD(&ea_buf->new_ea);
 647			if (ji->ea.flag & DXD_INLINE)
 648				ji->mode2 |= INLINEEA;
 649		}
 650		ji->ea = ea_buf->new_ea;
 651	} else {
 652		txEA(tid, inode, &ji->ea, NULL);
 653		if (ji->ea.flag & DXD_INLINE)
 654			ji->mode2 |= INLINEEA;
 655		ji->ea.flag = 0;
 656		ji->ea.size = 0;
 657	}
 658
 659	/* If old blocks exist, they must be removed from quota allocation. */
 660	if (old_blocks)
 661		dquot_free_block(inode, old_blocks);
 662
 663	inode->i_ctime = CURRENT_TIME;
 664
 665	return 0;
 666}
 667
 668/*
 669 * Most of the permission checking is done by xattr_permission in the vfs.
 670 * We also need to verify that this is a namespace that we recognize.
 671 */
 672static int can_set_xattr(struct inode *inode, const char *name,
 673			 const void *value, size_t value_len)
 674{
 675	if (!strncmp(name, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN)) {
 676		/*
 677		 * This makes sure that we aren't trying to set an
 678		 * attribute in a different namespace by prefixing it
 679		 * with "os2."
 680		 */
 681		if (is_known_namespace(name + XATTR_OS2_PREFIX_LEN))
 682			return -EOPNOTSUPP;
 683		return 0;
 684	}
 685
 686	/*
 687	 * Don't allow setting an attribute in an unknown namespace.
 688	 */
 689	if (strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) &&
 690	    strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
 691	    strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN))
 692		return -EOPNOTSUPP;
 693
 694	return 0;
 695}
 696
 697int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
 698		   const void *value, size_t value_len, int flags)
 699{
 700	struct jfs_ea_list *ealist;
 701	struct jfs_ea *ea, *old_ea = NULL, *next_ea = NULL;
 702	struct ea_buffer ea_buf;
 703	int old_ea_size = 0;
 704	int xattr_size;
 705	int new_size;
 706	int namelen = strlen(name);
 707	char *os2name = NULL;
 708	int found = 0;
 709	int rc;
 710	int length;
 711
 712	if (strncmp(name, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN) == 0) {
 713		os2name = kmalloc(namelen - XATTR_OS2_PREFIX_LEN + 1,
 714				  GFP_KERNEL);
 715		if (!os2name)
 716			return -ENOMEM;
 717		strcpy(os2name, name + XATTR_OS2_PREFIX_LEN);
 718		name = os2name;
 719		namelen -= XATTR_OS2_PREFIX_LEN;
 720	}
 721
 722	down_write(&JFS_IP(inode)->xattr_sem);
 723
 724	xattr_size = ea_get(inode, &ea_buf, 0);
 725	if (xattr_size < 0) {
 726		rc = xattr_size;
 727		goto out;
 728	}
 729
 730      again:
 731	ealist = (struct jfs_ea_list *) ea_buf.xattr;
 732	new_size = sizeof (struct jfs_ea_list);
 733
 734	if (xattr_size) {
 735		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
 736		     ea = NEXT_EA(ea)) {
 737			if ((namelen == ea->namelen) &&
 738			    (memcmp(name, ea->name, namelen) == 0)) {
 739				found = 1;
 740				if (flags & XATTR_CREATE) {
 741					rc = -EEXIST;
 742					goto release;
 743				}
 744				old_ea = ea;
 745				old_ea_size = EA_SIZE(ea);
 746				next_ea = NEXT_EA(ea);
 747			} else
 748				new_size += EA_SIZE(ea);
 749		}
 750	}
 751
 752	if (!found) {
 753		if (flags & XATTR_REPLACE) {
 754			rc = -ENODATA;
 755			goto release;
 756		}
 757		if (value == NULL) {
 758			rc = 0;
 759			goto release;
 760		}
 761	}
 762	if (value)
 763		new_size += sizeof (struct jfs_ea) + namelen + 1 + value_len;
 764
 765	if (new_size > ea_buf.max_size) {
 766		/*
 767		 * We need to allocate more space for merged ea list.
 768		 * We should only have loop to again: once.
 769		 */
 770		ea_release(inode, &ea_buf);
 771		xattr_size = ea_get(inode, &ea_buf, new_size);
 772		if (xattr_size < 0) {
 773			rc = xattr_size;
 774			goto out;
 775		}
 776		goto again;
 777	}
 778
 779	/* Remove old ea of the same name */
 780	if (found) {
 781		/* number of bytes following target EA */
 782		length = (char *) END_EALIST(ealist) - (char *) next_ea;
 783		if (length > 0)
 784			memmove(old_ea, next_ea, length);
 785		xattr_size -= old_ea_size;
 786	}
 787
 788	/* Add new entry to the end */
 789	if (value) {
 790		if (xattr_size == 0)
 791			/* Completely new ea list */
 792			xattr_size = sizeof (struct jfs_ea_list);
 793
 794		/*
 795		 * The size of EA value is limitted by on-disk format up to
 796		 *  __le16, there would be an overflow if the size is equal
 797		 * to XATTR_SIZE_MAX (65536).  In order to avoid this issue,
 798		 * we can pre-checkup the value size against USHRT_MAX, and
 799		 * return -E2BIG in this case, which is consistent with the
 800		 * VFS setxattr interface.
 801		 */
 802		if (value_len >= USHRT_MAX) {
 803			rc = -E2BIG;
 804			goto release;
 805		}
 806
 807		ea = (struct jfs_ea *) ((char *) ealist + xattr_size);
 808		ea->flag = 0;
 809		ea->namelen = namelen;
 810		ea->valuelen = (cpu_to_le16(value_len));
 811		memcpy(ea->name, name, namelen);
 812		ea->name[namelen] = 0;
 813		if (value_len)
 814			memcpy(&ea->name[namelen + 1], value, value_len);
 815		xattr_size += EA_SIZE(ea);
 816	}
 817
 818	/* DEBUG - If we did this right, these number match */
 819	if (xattr_size != new_size) {
 820		printk(KERN_ERR
 821		       "__jfs_setxattr: xattr_size = %d, new_size = %d\n",
 822		       xattr_size, new_size);
 823
 824		rc = -EINVAL;
 825		goto release;
 826	}
 827
 828	/*
 829	 * If we're left with an empty list, there's no ea
 830	 */
 831	if (new_size == sizeof (struct jfs_ea_list))
 832		new_size = 0;
 833
 834	ealist->size = cpu_to_le32(new_size);
 835
 836	rc = ea_put(tid, inode, &ea_buf, new_size);
 837
 838	goto out;
 839      release:
 840	ea_release(inode, &ea_buf);
 841      out:
 842	up_write(&JFS_IP(inode)->xattr_sem);
 843
 844	kfree(os2name);
 845
 846	return rc;
 847}
 848
 849int jfs_setxattr(struct dentry *dentry, const char *name, const void *value,
 850		 size_t value_len, int flags)
 851{
 852	struct inode *inode = dentry->d_inode;
 853	struct jfs_inode_info *ji = JFS_IP(inode);
 854	int rc;
 855	tid_t tid;
 856
 857	/*
 858	 * If this is a request for a synthetic attribute in the system.*
 859	 * namespace use the generic infrastructure to resolve a handler
 860	 * for it via sb->s_xattr.
 861	 */
 862	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
 863		return generic_setxattr(dentry, name, value, value_len, flags);
 864
 865	if ((rc = can_set_xattr(inode, name, value, value_len)))
 866		return rc;
 867
 868	if (value == NULL) {	/* empty EA, do not remove */
 869		value = "";
 870		value_len = 0;
 871	}
 872
 873	tid = txBegin(inode->i_sb, 0);
 874	mutex_lock(&ji->commit_mutex);
 875	rc = __jfs_setxattr(tid, dentry->d_inode, name, value, value_len,
 876			    flags);
 877	if (!rc)
 878		rc = txCommit(tid, 1, &inode, 0);
 879	txEnd(tid);
 880	mutex_unlock(&ji->commit_mutex);
 881
 882	return rc;
 883}
 884
 885ssize_t __jfs_getxattr(struct inode *inode, const char *name, void *data,
 886		       size_t buf_size)
 887{
 888	struct jfs_ea_list *ealist;
 889	struct jfs_ea *ea;
 890	struct ea_buffer ea_buf;
 891	int xattr_size;
 892	ssize_t size;
 893	int namelen = strlen(name);
 894	char *value;
 895
 896	down_read(&JFS_IP(inode)->xattr_sem);
 897
 898	xattr_size = ea_get(inode, &ea_buf, 0);
 899
 900	if (xattr_size < 0) {
 901		size = xattr_size;
 902		goto out;
 903	}
 904
 905	if (xattr_size == 0)
 906		goto not_found;
 907
 908	ealist = (struct jfs_ea_list *) ea_buf.xattr;
 909
 910	/* Find the named attribute */
 911	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea))
 912		if ((namelen == ea->namelen) &&
 913		    memcmp(name, ea->name, namelen) == 0) {
 914			/* Found it */
 915			size = le16_to_cpu(ea->valuelen);
 916			if (!data)
 917				goto release;
 918			else if (size > buf_size) {
 919				size = -ERANGE;
 920				goto release;
 921			}
 922			value = ((char *) &ea->name) + ea->namelen + 1;
 923			memcpy(data, value, size);
 924			goto release;
 925		}
 926      not_found:
 927	size = -ENODATA;
 928      release:
 929	ea_release(inode, &ea_buf);
 930      out:
 931	up_read(&JFS_IP(inode)->xattr_sem);
 932
 933	return size;
 934}
 935
 936ssize_t jfs_getxattr(struct dentry *dentry, const char *name, void *data,
 937		     size_t buf_size)
 938{
 939	int err;
 940
 941	/*
 942	 * If this is a request for a synthetic attribute in the system.*
 943	 * namespace use the generic infrastructure to resolve a handler
 944	 * for it via sb->s_xattr.
 945	 */
 946	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
 947		return generic_getxattr(dentry, name, data, buf_size);
 948
 949	if (strncmp(name, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN) == 0) {
 950		/*
 951		 * skip past "os2." prefix
 952		 */
 953		name += XATTR_OS2_PREFIX_LEN;
 954		/*
 955		 * Don't allow retrieving properly prefixed attributes
 956		 * by prepending them with "os2."
 957		 */
 958		if (is_known_namespace(name))
 959			return -EOPNOTSUPP;
 960	}
 961
 962	err = __jfs_getxattr(dentry->d_inode, name, data, buf_size);
 963
 964	return err;
 965}
 966
 967/*
 968 * No special permissions are needed to list attributes except for trusted.*
 969 */
 970static inline int can_list(struct jfs_ea *ea)
 971{
 972	return (strncmp(ea->name, XATTR_TRUSTED_PREFIX,
 973			    XATTR_TRUSTED_PREFIX_LEN) ||
 974		capable(CAP_SYS_ADMIN));
 975}
 976
 977ssize_t jfs_listxattr(struct dentry * dentry, char *data, size_t buf_size)
 978{
 979	struct inode *inode = dentry->d_inode;
 980	char *buffer;
 981	ssize_t size = 0;
 982	int xattr_size;
 983	struct jfs_ea_list *ealist;
 984	struct jfs_ea *ea;
 985	struct ea_buffer ea_buf;
 986
 987	down_read(&JFS_IP(inode)->xattr_sem);
 988
 989	xattr_size = ea_get(inode, &ea_buf, 0);
 990	if (xattr_size < 0) {
 991		size = xattr_size;
 992		goto out;
 993	}
 994
 995	if (xattr_size == 0)
 996		goto release;
 997
 998	ealist = (struct jfs_ea_list *) ea_buf.xattr;
 999
1000	/* compute required size of list */
1001	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
1002		if (can_list(ea))
1003			size += name_size(ea) + 1;
1004	}
1005
1006	if (!data)
1007		goto release;
1008
1009	if (size > buf_size) {
1010		size = -ERANGE;
1011		goto release;
1012	}
1013
1014	/* Copy attribute names to buffer */
1015	buffer = data;
1016	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
1017		if (can_list(ea)) {
1018			int namelen = copy_name(buffer, ea);
1019			buffer += namelen + 1;
1020		}
1021	}
1022
1023      release:
1024	ea_release(inode, &ea_buf);
1025      out:
1026	up_read(&JFS_IP(inode)->xattr_sem);
1027	return size;
1028}
1029
1030int jfs_removexattr(struct dentry *dentry, const char *name)
1031{
1032	struct inode *inode = dentry->d_inode;
1033	struct jfs_inode_info *ji = JFS_IP(inode);
1034	int rc;
1035	tid_t tid;
1036
1037	/*
1038	 * If this is a request for a synthetic attribute in the system.*
1039	 * namespace use the generic infrastructure to resolve a handler
1040	 * for it via sb->s_xattr.
1041	 */
1042	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1043		return generic_removexattr(dentry, name);
1044
1045	if ((rc = can_set_xattr(inode, name, NULL, 0)))
1046		return rc;
1047
1048	tid = txBegin(inode->i_sb, 0);
1049	mutex_lock(&ji->commit_mutex);
1050	rc = __jfs_setxattr(tid, dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
1051	if (!rc)
1052		rc = txCommit(tid, 1, &inode, 0);
1053	txEnd(tid);
1054	mutex_unlock(&ji->commit_mutex);
1055
1056	return rc;
1057}
1058
1059/*
1060 * List of handlers for synthetic system.* attributes.  All real ondisk
1061 * attributes are handled directly.
1062 */
1063const struct xattr_handler *jfs_xattr_handlers[] = {
1064#ifdef CONFIG_JFS_POSIX_ACL
1065	&posix_acl_access_xattr_handler,
1066	&posix_acl_default_xattr_handler,
1067#endif
1068	NULL,
1069};
1070
1071
1072#ifdef CONFIG_JFS_SECURITY
1073static int jfs_initxattrs(struct inode *inode, const struct xattr *xattr_array,
1074			  void *fs_info)
1075{
1076	const struct xattr *xattr;
1077	tid_t *tid = fs_info;
1078	char *name;
1079	int err = 0;
1080
1081	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
1082		name = kmalloc(XATTR_SECURITY_PREFIX_LEN +
1083			       strlen(xattr->name) + 1, GFP_NOFS);
1084		if (!name) {
1085			err = -ENOMEM;
1086			break;
1087		}
1088		strcpy(name, XATTR_SECURITY_PREFIX);
1089		strcpy(name + XATTR_SECURITY_PREFIX_LEN, xattr->name);
1090
1091		err = __jfs_setxattr(*tid, inode, name,
1092				     xattr->value, xattr->value_len, 0);
1093		kfree(name);
1094		if (err < 0)
1095			break;
1096	}
1097	return err;
1098}
1099
1100int jfs_init_security(tid_t tid, struct inode *inode, struct inode *dir,
1101		      const struct qstr *qstr)
1102{
1103	return security_inode_init_security(inode, dir, qstr,
1104					    &jfs_initxattrs, &tid);
1105}
1106#endif