xref: /openbmc/linux/fs/jfs/xattr.c (revision a44d924c)
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  
72  struct 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  
89  /*
90   * Mapping of on-disk attribute names: for on-disk attribute names with an
91   * unknown prefix (not "system.", "user.", "security.", or "trusted."), the
92   * prefix "os2." is prepended.  On the way back to disk, "os2." prefixes are
93   * stripped and we make sure that the remaining name does not start with one
94   * of the know prefixes.
95   */
96  
97  static int is_known_namespace(const char *name)
98  {
99  	if (strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) &&
100  	    strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN) &&
101  	    strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
102  	    strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN))
103  		return false;
104  
105  	return true;
106  }
107  
108  static inline int name_size(struct jfs_ea *ea)
109  {
110  	if (is_known_namespace(ea->name))
111  		return ea->namelen;
112  	else
113  		return ea->namelen + XATTR_OS2_PREFIX_LEN;
114  }
115  
116  static inline int copy_name(char *buffer, struct jfs_ea *ea)
117  {
118  	int len = ea->namelen;
119  
120  	if (!is_known_namespace(ea->name)) {
121  		memcpy(buffer, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN);
122  		buffer += XATTR_OS2_PREFIX_LEN;
123  		len += XATTR_OS2_PREFIX_LEN;
124  	}
125  	memcpy(buffer, ea->name, ea->namelen);
126  	buffer[ea->namelen] = 0;
127  
128  	return len;
129  }
130  
131  /* Forward references */
132  static void ea_release(struct inode *inode, struct ea_buffer *ea_buf);
133  
134  /*
135   * NAME: ea_write_inline
136   *
137   * FUNCTION: Attempt to write an EA inline if area is available
138   *
139   * PRE CONDITIONS:
140   *	Already verified that the specified EA is small enough to fit inline
141   *
142   * PARAMETERS:
143   *	ip	- Inode pointer
144   *	ealist	- EA list pointer
145   *	size	- size of ealist in bytes
146   *	ea	- dxd_t structure to be filled in with necessary EA information
147   *		  if we successfully copy the EA inline
148   *
149   * NOTES:
150   *	Checks if the inode's inline area is available.  If so, copies EA inline
151   *	and sets <ea> fields appropriately.  Otherwise, returns failure, EA will
152   *	have to be put into an extent.
153   *
154   * RETURNS: 0 for successful copy to inline area; -1 if area not available
155   */
156  static int ea_write_inline(struct inode *ip, struct jfs_ea_list *ealist,
157  			   int size, dxd_t * ea)
158  {
159  	struct jfs_inode_info *ji = JFS_IP(ip);
160  
161  	/*
162  	 * Make sure we have an EA -- the NULL EA list is valid, but you
163  	 * can't copy it!
164  	 */
165  	if (ealist && size > sizeof (struct jfs_ea_list)) {
166  		assert(size <= sizeof (ji->i_inline_ea));
167  
168  		/*
169  		 * See if the space is available or if it is already being
170  		 * used for an inline EA.
171  		 */
172  		if (!(ji->mode2 & INLINEEA) && !(ji->ea.flag & DXD_INLINE))
173  			return -EPERM;
174  
175  		DXDsize(ea, size);
176  		DXDlength(ea, 0);
177  		DXDaddress(ea, 0);
178  		memcpy(ji->i_inline_ea, ealist, size);
179  		ea->flag = DXD_INLINE;
180  		ji->mode2 &= ~INLINEEA;
181  	} else {
182  		ea->flag = 0;
183  		DXDsize(ea, 0);
184  		DXDlength(ea, 0);
185  		DXDaddress(ea, 0);
186  
187  		/* Free up INLINE area */
188  		if (ji->ea.flag & DXD_INLINE)
189  			ji->mode2 |= INLINEEA;
190  	}
191  
192  	return 0;
193  }
194  
195  /*
196   * NAME: ea_write
197   *
198   * FUNCTION: Write an EA for an inode
199   *
200   * PRE CONDITIONS: EA has been verified
201   *
202   * PARAMETERS:
203   *	ip	- Inode pointer
204   *	ealist	- EA list pointer
205   *	size	- size of ealist in bytes
206   *	ea	- dxd_t structure to be filled in appropriately with where the
207   *		  EA was copied
208   *
209   * NOTES: Will write EA inline if able to, otherwise allocates blocks for an
210   *	extent and synchronously writes it to those blocks.
211   *
212   * RETURNS: 0 for success; Anything else indicates failure
213   */
214  static int ea_write(struct inode *ip, struct jfs_ea_list *ealist, int size,
215  		       dxd_t * ea)
216  {
217  	struct super_block *sb = ip->i_sb;
218  	struct jfs_inode_info *ji = JFS_IP(ip);
219  	struct jfs_sb_info *sbi = JFS_SBI(sb);
220  	int nblocks;
221  	s64 blkno;
222  	int rc = 0, i;
223  	char *cp;
224  	s32 nbytes, nb;
225  	s32 bytes_to_write;
226  	struct metapage *mp;
227  
228  	/*
229  	 * Quick check to see if this is an in-linable EA.  Short EAs
230  	 * and empty EAs are all in-linable, provided the space exists.
231  	 */
232  	if (!ealist || size <= sizeof (ji->i_inline_ea)) {
233  		if (!ea_write_inline(ip, ealist, size, ea))
234  			return 0;
235  	}
236  
237  	/* figure out how many blocks we need */
238  	nblocks = (size + (sb->s_blocksize - 1)) >> sb->s_blocksize_bits;
239  
240  	/* Allocate new blocks to quota. */
241  	rc = dquot_alloc_block(ip, nblocks);
242  	if (rc)
243  		return rc;
244  
245  	rc = dbAlloc(ip, INOHINT(ip), nblocks, &blkno);
246  	if (rc) {
247  		/*Rollback quota allocation. */
248  		dquot_free_block(ip, nblocks);
249  		return rc;
250  	}
251  
252  	/*
253  	 * Now have nblocks worth of storage to stuff into the FEALIST.
254  	 * loop over the FEALIST copying data into the buffer one page at
255  	 * a time.
256  	 */
257  	cp = (char *) ealist;
258  	nbytes = size;
259  	for (i = 0; i < nblocks; i += sbi->nbperpage) {
260  		/*
261  		 * Determine how many bytes for this request, and round up to
262  		 * the nearest aggregate block size
263  		 */
264  		nb = min(PSIZE, nbytes);
265  		bytes_to_write =
266  		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
267  		    << sb->s_blocksize_bits;
268  
269  		if (!(mp = get_metapage(ip, blkno + i, bytes_to_write, 1))) {
270  			rc = -EIO;
271  			goto failed;
272  		}
273  
274  		memcpy(mp->data, cp, nb);
275  
276  		/*
277  		 * We really need a way to propagate errors for
278  		 * forced writes like this one.  --hch
279  		 *
280  		 * (__write_metapage => release_metapage => flush_metapage)
281  		 */
282  #ifdef _JFS_FIXME
283  		if ((rc = flush_metapage(mp))) {
284  			/*
285  			 * the write failed -- this means that the buffer
286  			 * is still assigned and the blocks are not being
287  			 * used.  this seems like the best error recovery
288  			 * we can get ...
289  			 */
290  			goto failed;
291  		}
292  #else
293  		flush_metapage(mp);
294  #endif
295  
296  		cp += PSIZE;
297  		nbytes -= nb;
298  	}
299  
300  	ea->flag = DXD_EXTENT;
301  	DXDsize(ea, le32_to_cpu(ealist->size));
302  	DXDlength(ea, nblocks);
303  	DXDaddress(ea, blkno);
304  
305  	/* Free up INLINE area */
306  	if (ji->ea.flag & DXD_INLINE)
307  		ji->mode2 |= INLINEEA;
308  
309  	return 0;
310  
311        failed:
312  	/* Rollback quota allocation. */
313  	dquot_free_block(ip, nblocks);
314  
315  	dbFree(ip, blkno, nblocks);
316  	return rc;
317  }
318  
319  /*
320   * NAME: ea_read_inline
321   *
322   * FUNCTION: Read an inlined EA into user's buffer
323   *
324   * PARAMETERS:
325   *	ip	- Inode pointer
326   *	ealist	- Pointer to buffer to fill in with EA
327   *
328   * RETURNS: 0
329   */
330  static int ea_read_inline(struct inode *ip, struct jfs_ea_list *ealist)
331  {
332  	struct jfs_inode_info *ji = JFS_IP(ip);
333  	int ea_size = sizeDXD(&ji->ea);
334  
335  	if (ea_size == 0) {
336  		ealist->size = 0;
337  		return 0;
338  	}
339  
340  	/* Sanity Check */
341  	if ((sizeDXD(&ji->ea) > sizeof (ji->i_inline_ea)))
342  		return -EIO;
343  	if (le32_to_cpu(((struct jfs_ea_list *) &ji->i_inline_ea)->size)
344  	    != ea_size)
345  		return -EIO;
346  
347  	memcpy(ealist, ji->i_inline_ea, ea_size);
348  	return 0;
349  }
350  
351  /*
352   * NAME: ea_read
353   *
354   * FUNCTION: copy EA data into user's buffer
355   *
356   * PARAMETERS:
357   *	ip	- Inode pointer
358   *	ealist	- Pointer to buffer to fill in with EA
359   *
360   * NOTES:  If EA is inline calls ea_read_inline() to copy EA.
361   *
362   * RETURNS: 0 for success; other indicates failure
363   */
364  static int ea_read(struct inode *ip, struct jfs_ea_list *ealist)
365  {
366  	struct super_block *sb = ip->i_sb;
367  	struct jfs_inode_info *ji = JFS_IP(ip);
368  	struct jfs_sb_info *sbi = JFS_SBI(sb);
369  	int nblocks;
370  	s64 blkno;
371  	char *cp = (char *) ealist;
372  	int i;
373  	int nbytes, nb;
374  	s32 bytes_to_read;
375  	struct metapage *mp;
376  
377  	/* quick check for in-line EA */
378  	if (ji->ea.flag & DXD_INLINE)
379  		return ea_read_inline(ip, ealist);
380  
381  	nbytes = sizeDXD(&ji->ea);
382  	if (!nbytes) {
383  		jfs_error(sb, "nbytes is 0\n");
384  		return -EIO;
385  	}
386  
387  	/*
388  	 * Figure out how many blocks were allocated when this EA list was
389  	 * originally written to disk.
390  	 */
391  	nblocks = lengthDXD(&ji->ea) << sbi->l2nbperpage;
392  	blkno = addressDXD(&ji->ea) << sbi->l2nbperpage;
393  
394  	/*
395  	 * I have found the disk blocks which were originally used to store
396  	 * the FEALIST.  now i loop over each contiguous block copying the
397  	 * data into the buffer.
398  	 */
399  	for (i = 0; i < nblocks; i += sbi->nbperpage) {
400  		/*
401  		 * Determine how many bytes for this request, and round up to
402  		 * the nearest aggregate block size
403  		 */
404  		nb = min(PSIZE, nbytes);
405  		bytes_to_read =
406  		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
407  		    << sb->s_blocksize_bits;
408  
409  		if (!(mp = read_metapage(ip, blkno + i, bytes_to_read, 1)))
410  			return -EIO;
411  
412  		memcpy(cp, mp->data, nb);
413  		release_metapage(mp);
414  
415  		cp += PSIZE;
416  		nbytes -= nb;
417  	}
418  
419  	return 0;
420  }
421  
422  /*
423   * NAME: ea_get
424   *
425   * FUNCTION: Returns buffer containing existing extended attributes.
426   *	     The size of the buffer will be the larger of the existing
427   *	     attributes size, or min_size.
428   *
429   *	     The buffer, which may be inlined in the inode or in the
430   *	     page cache must be release by calling ea_release or ea_put
431   *
432   * PARAMETERS:
433   *	inode	- Inode pointer
434   *	ea_buf	- Structure to be populated with ealist and its metadata
435   *	min_size- minimum size of buffer to be returned
436   *
437   * RETURNS: 0 for success; Other indicates failure
438   */
439  static int ea_get(struct inode *inode, struct ea_buffer *ea_buf, int min_size)
440  {
441  	struct jfs_inode_info *ji = JFS_IP(inode);
442  	struct super_block *sb = inode->i_sb;
443  	int size;
444  	int ea_size = sizeDXD(&ji->ea);
445  	int blocks_needed, current_blocks;
446  	s64 blkno;
447  	int rc;
448  	int quota_allocation = 0;
449  
450  	/* When fsck.jfs clears a bad ea, it doesn't clear the size */
451  	if (ji->ea.flag == 0)
452  		ea_size = 0;
453  
454  	if (ea_size == 0) {
455  		if (min_size == 0) {
456  			ea_buf->flag = 0;
457  			ea_buf->max_size = 0;
458  			ea_buf->xattr = NULL;
459  			return 0;
460  		}
461  		if ((min_size <= sizeof (ji->i_inline_ea)) &&
462  		    (ji->mode2 & INLINEEA)) {
463  			ea_buf->flag = EA_INLINE | EA_NEW;
464  			ea_buf->max_size = sizeof (ji->i_inline_ea);
465  			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
466  			DXDlength(&ea_buf->new_ea, 0);
467  			DXDaddress(&ea_buf->new_ea, 0);
468  			ea_buf->new_ea.flag = DXD_INLINE;
469  			DXDsize(&ea_buf->new_ea, min_size);
470  			return 0;
471  		}
472  		current_blocks = 0;
473  	} else if (ji->ea.flag & DXD_INLINE) {
474  		if (min_size <= sizeof (ji->i_inline_ea)) {
475  			ea_buf->flag = EA_INLINE;
476  			ea_buf->max_size = sizeof (ji->i_inline_ea);
477  			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
478  			goto size_check;
479  		}
480  		current_blocks = 0;
481  	} else {
482  		if (!(ji->ea.flag & DXD_EXTENT)) {
483  			jfs_error(sb, "invalid ea.flag\n");
484  			return -EIO;
485  		}
486  		current_blocks = (ea_size + sb->s_blocksize - 1) >>
487  		    sb->s_blocksize_bits;
488  	}
489  	size = max(min_size, ea_size);
490  
491  	if (size > PSIZE) {
492  		/*
493  		 * To keep the rest of the code simple.  Allocate a
494  		 * contiguous buffer to work with. Make the buffer large
495  		 * enough to make use of the whole extent.
496  		 */
497  		ea_buf->max_size = (size + sb->s_blocksize - 1) &
498  		    ~(sb->s_blocksize - 1);
499  
500  		ea_buf->xattr = kmalloc(ea_buf->max_size, GFP_KERNEL);
501  		if (ea_buf->xattr == NULL)
502  			return -ENOMEM;
503  
504  		ea_buf->flag = EA_MALLOC;
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  
591  static 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  
605  static 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(inode);
664  
665  	return 0;
666  }
667  
668  int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
669  		   const void *value, size_t value_len, int flags)
670  {
671  	struct jfs_ea_list *ealist;
672  	struct jfs_ea *ea, *old_ea = NULL, *next_ea = NULL;
673  	struct ea_buffer ea_buf;
674  	int old_ea_size = 0;
675  	int xattr_size;
676  	int new_size;
677  	int namelen = strlen(name);
678  	int found = 0;
679  	int rc;
680  	int length;
681  
682  	down_write(&JFS_IP(inode)->xattr_sem);
683  
684  	xattr_size = ea_get(inode, &ea_buf, 0);
685  	if (xattr_size < 0) {
686  		rc = xattr_size;
687  		goto out;
688  	}
689  
690        again:
691  	ealist = (struct jfs_ea_list *) ea_buf.xattr;
692  	new_size = sizeof (struct jfs_ea_list);
693  
694  	if (xattr_size) {
695  		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
696  		     ea = NEXT_EA(ea)) {
697  			if ((namelen == ea->namelen) &&
698  			    (memcmp(name, ea->name, namelen) == 0)) {
699  				found = 1;
700  				if (flags & XATTR_CREATE) {
701  					rc = -EEXIST;
702  					goto release;
703  				}
704  				old_ea = ea;
705  				old_ea_size = EA_SIZE(ea);
706  				next_ea = NEXT_EA(ea);
707  			} else
708  				new_size += EA_SIZE(ea);
709  		}
710  	}
711  
712  	if (!found) {
713  		if (flags & XATTR_REPLACE) {
714  			rc = -ENODATA;
715  			goto release;
716  		}
717  		if (value == NULL) {
718  			rc = 0;
719  			goto release;
720  		}
721  	}
722  	if (value)
723  		new_size += sizeof (struct jfs_ea) + namelen + 1 + value_len;
724  
725  	if (new_size > ea_buf.max_size) {
726  		/*
727  		 * We need to allocate more space for merged ea list.
728  		 * We should only have loop to again: once.
729  		 */
730  		ea_release(inode, &ea_buf);
731  		xattr_size = ea_get(inode, &ea_buf, new_size);
732  		if (xattr_size < 0) {
733  			rc = xattr_size;
734  			goto out;
735  		}
736  		goto again;
737  	}
738  
739  	/* Remove old ea of the same name */
740  	if (found) {
741  		/* number of bytes following target EA */
742  		length = (char *) END_EALIST(ealist) - (char *) next_ea;
743  		if (length > 0)
744  			memmove(old_ea, next_ea, length);
745  		xattr_size -= old_ea_size;
746  	}
747  
748  	/* Add new entry to the end */
749  	if (value) {
750  		if (xattr_size == 0)
751  			/* Completely new ea list */
752  			xattr_size = sizeof (struct jfs_ea_list);
753  
754  		/*
755  		 * The size of EA value is limitted by on-disk format up to
756  		 *  __le16, there would be an overflow if the size is equal
757  		 * to XATTR_SIZE_MAX (65536).  In order to avoid this issue,
758  		 * we can pre-checkup the value size against USHRT_MAX, and
759  		 * return -E2BIG in this case, which is consistent with the
760  		 * VFS setxattr interface.
761  		 */
762  		if (value_len >= USHRT_MAX) {
763  			rc = -E2BIG;
764  			goto release;
765  		}
766  
767  		ea = (struct jfs_ea *) ((char *) ealist + xattr_size);
768  		ea->flag = 0;
769  		ea->namelen = namelen;
770  		ea->valuelen = (cpu_to_le16(value_len));
771  		memcpy(ea->name, name, namelen);
772  		ea->name[namelen] = 0;
773  		if (value_len)
774  			memcpy(&ea->name[namelen + 1], value, value_len);
775  		xattr_size += EA_SIZE(ea);
776  	}
777  
778  	/* DEBUG - If we did this right, these number match */
779  	if (xattr_size != new_size) {
780  		printk(KERN_ERR
781  		       "__jfs_setxattr: xattr_size = %d, new_size = %d\n",
782  		       xattr_size, new_size);
783  
784  		rc = -EINVAL;
785  		goto release;
786  	}
787  
788  	/*
789  	 * If we're left with an empty list, there's no ea
790  	 */
791  	if (new_size == sizeof (struct jfs_ea_list))
792  		new_size = 0;
793  
794  	ealist->size = cpu_to_le32(new_size);
795  
796  	rc = ea_put(tid, inode, &ea_buf, new_size);
797  
798  	goto out;
799        release:
800  	ea_release(inode, &ea_buf);
801        out:
802  	up_write(&JFS_IP(inode)->xattr_sem);
803  
804  	return rc;
805  }
806  
807  ssize_t __jfs_getxattr(struct inode *inode, const char *name, void *data,
808  		       size_t buf_size)
809  {
810  	struct jfs_ea_list *ealist;
811  	struct jfs_ea *ea;
812  	struct ea_buffer ea_buf;
813  	int xattr_size;
814  	ssize_t size;
815  	int namelen = strlen(name);
816  	char *value;
817  
818  	down_read(&JFS_IP(inode)->xattr_sem);
819  
820  	xattr_size = ea_get(inode, &ea_buf, 0);
821  
822  	if (xattr_size < 0) {
823  		size = xattr_size;
824  		goto out;
825  	}
826  
827  	if (xattr_size == 0)
828  		goto not_found;
829  
830  	ealist = (struct jfs_ea_list *) ea_buf.xattr;
831  
832  	/* Find the named attribute */
833  	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea))
834  		if ((namelen == ea->namelen) &&
835  		    memcmp(name, ea->name, namelen) == 0) {
836  			/* Found it */
837  			size = le16_to_cpu(ea->valuelen);
838  			if (!data)
839  				goto release;
840  			else if (size > buf_size) {
841  				size = -ERANGE;
842  				goto release;
843  			}
844  			value = ((char *) &ea->name) + ea->namelen + 1;
845  			memcpy(data, value, size);
846  			goto release;
847  		}
848        not_found:
849  	size = -ENODATA;
850        release:
851  	ea_release(inode, &ea_buf);
852        out:
853  	up_read(&JFS_IP(inode)->xattr_sem);
854  
855  	return size;
856  }
857  
858  /*
859   * No special permissions are needed to list attributes except for trusted.*
860   */
861  static inline int can_list(struct jfs_ea *ea)
862  {
863  	return (strncmp(ea->name, XATTR_TRUSTED_PREFIX,
864  			    XATTR_TRUSTED_PREFIX_LEN) ||
865  		capable(CAP_SYS_ADMIN));
866  }
867  
868  ssize_t jfs_listxattr(struct dentry * dentry, char *data, size_t buf_size)
869  {
870  	struct inode *inode = d_inode(dentry);
871  	char *buffer;
872  	ssize_t size = 0;
873  	int xattr_size;
874  	struct jfs_ea_list *ealist;
875  	struct jfs_ea *ea;
876  	struct ea_buffer ea_buf;
877  
878  	down_read(&JFS_IP(inode)->xattr_sem);
879  
880  	xattr_size = ea_get(inode, &ea_buf, 0);
881  	if (xattr_size < 0) {
882  		size = xattr_size;
883  		goto out;
884  	}
885  
886  	if (xattr_size == 0)
887  		goto release;
888  
889  	ealist = (struct jfs_ea_list *) ea_buf.xattr;
890  
891  	/* compute required size of list */
892  	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
893  		if (can_list(ea))
894  			size += name_size(ea) + 1;
895  	}
896  
897  	if (!data)
898  		goto release;
899  
900  	if (size > buf_size) {
901  		size = -ERANGE;
902  		goto release;
903  	}
904  
905  	/* Copy attribute names to buffer */
906  	buffer = data;
907  	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
908  		if (can_list(ea)) {
909  			int namelen = copy_name(buffer, ea);
910  			buffer += namelen + 1;
911  		}
912  	}
913  
914        release:
915  	ea_release(inode, &ea_buf);
916        out:
917  	up_read(&JFS_IP(inode)->xattr_sem);
918  	return size;
919  }
920  
921  static int __jfs_xattr_set(struct inode *inode, const char *name,
922  			   const void *value, size_t size, int flags)
923  {
924  	struct jfs_inode_info *ji = JFS_IP(inode);
925  	tid_t tid;
926  	int rc;
927  
928  	tid = txBegin(inode->i_sb, 0);
929  	mutex_lock(&ji->commit_mutex);
930  	rc = __jfs_setxattr(tid, inode, name, value, size, flags);
931  	if (!rc)
932  		rc = txCommit(tid, 1, &inode, 0);
933  	txEnd(tid);
934  	mutex_unlock(&ji->commit_mutex);
935  
936  	return rc;
937  }
938  
939  static int jfs_xattr_get(const struct xattr_handler *handler,
940  			 struct dentry *unused, struct inode *inode,
941  			 const char *name, void *value, size_t size)
942  {
943  	name = xattr_full_name(handler, name);
944  	return __jfs_getxattr(inode, name, value, size);
945  }
946  
947  static int jfs_xattr_set(const struct xattr_handler *handler,
948  			 struct dentry *unused, struct inode *inode,
949  			 const char *name, const void *value,
950  			 size_t size, int flags)
951  {
952  	name = xattr_full_name(handler, name);
953  	return __jfs_xattr_set(inode, name, value, size, flags);
954  }
955  
956  static int jfs_xattr_get_os2(const struct xattr_handler *handler,
957  			     struct dentry *unused, struct inode *inode,
958  			     const char *name, void *value, size_t size)
959  {
960  	if (is_known_namespace(name))
961  		return -EOPNOTSUPP;
962  	return __jfs_getxattr(inode, name, value, size);
963  }
964  
965  static int jfs_xattr_set_os2(const struct xattr_handler *handler,
966  			     struct dentry *unused, struct inode *inode,
967  			     const char *name, const void *value,
968  			     size_t size, int flags)
969  {
970  	if (is_known_namespace(name))
971  		return -EOPNOTSUPP;
972  	return __jfs_xattr_set(inode, name, value, size, flags);
973  }
974  
975  static const struct xattr_handler jfs_user_xattr_handler = {
976  	.prefix = XATTR_USER_PREFIX,
977  	.get = jfs_xattr_get,
978  	.set = jfs_xattr_set,
979  };
980  
981  static const struct xattr_handler jfs_os2_xattr_handler = {
982  	.prefix = XATTR_OS2_PREFIX,
983  	.get = jfs_xattr_get_os2,
984  	.set = jfs_xattr_set_os2,
985  };
986  
987  static const struct xattr_handler jfs_security_xattr_handler = {
988  	.prefix = XATTR_SECURITY_PREFIX,
989  	.get = jfs_xattr_get,
990  	.set = jfs_xattr_set,
991  };
992  
993  static const struct xattr_handler jfs_trusted_xattr_handler = {
994  	.prefix = XATTR_TRUSTED_PREFIX,
995  	.get = jfs_xattr_get,
996  	.set = jfs_xattr_set,
997  };
998  
999  const struct xattr_handler *jfs_xattr_handlers[] = {
1000  #ifdef CONFIG_JFS_POSIX_ACL
1001  	&posix_acl_access_xattr_handler,
1002  	&posix_acl_default_xattr_handler,
1003  #endif
1004  	&jfs_os2_xattr_handler,
1005  	&jfs_user_xattr_handler,
1006  	&jfs_security_xattr_handler,
1007  	&jfs_trusted_xattr_handler,
1008  	NULL,
1009  };
1010  
1011  
1012  #ifdef CONFIG_JFS_SECURITY
1013  static int jfs_initxattrs(struct inode *inode, const struct xattr *xattr_array,
1014  			  void *fs_info)
1015  {
1016  	const struct xattr *xattr;
1017  	tid_t *tid = fs_info;
1018  	char *name;
1019  	int err = 0;
1020  
1021  	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
1022  		name = kmalloc(XATTR_SECURITY_PREFIX_LEN +
1023  			       strlen(xattr->name) + 1, GFP_NOFS);
1024  		if (!name) {
1025  			err = -ENOMEM;
1026  			break;
1027  		}
1028  		strcpy(name, XATTR_SECURITY_PREFIX);
1029  		strcpy(name + XATTR_SECURITY_PREFIX_LEN, xattr->name);
1030  
1031  		err = __jfs_setxattr(*tid, inode, name,
1032  				     xattr->value, xattr->value_len, 0);
1033  		kfree(name);
1034  		if (err < 0)
1035  			break;
1036  	}
1037  	return err;
1038  }
1039  
1040  int jfs_init_security(tid_t tid, struct inode *inode, struct inode *dir,
1041  		      const struct qstr *qstr)
1042  {
1043  	return security_inode_init_security(inode, dir, qstr,
1044  					    &jfs_initxattrs, &tid);
1045  }
1046  #endif
1047