Linux Audio

Check our new training course

Embedded Linux training

Mar 31-Apr 8, 2025
Register
Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *   Copyright (C) International Business Machines Corp., 2000-2004
   4 *   Portions Copyright (C) Christoph Hellwig, 2001-2002
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 */
   6
   7#include <linux/fs.h>
   8#include <linux/namei.h>
   9#include <linux/ctype.h>
  10#include <linux/quotaops.h>
  11#include <linux/exportfs.h>
  12#include "jfs_incore.h"
  13#include "jfs_superblock.h"
  14#include "jfs_inode.h"
  15#include "jfs_dinode.h"
  16#include "jfs_dmap.h"
  17#include "jfs_unicode.h"
  18#include "jfs_metapage.h"
  19#include "jfs_xattr.h"
  20#include "jfs_acl.h"
  21#include "jfs_debug.h"
  22
  23/*
  24 * forward references
  25 */
  26const struct dentry_operations jfs_ci_dentry_operations;
  27
  28static s64 commitZeroLink(tid_t, struct inode *);
  29
  30/*
  31 * NAME:	free_ea_wmap(inode)
  32 *
  33 * FUNCTION:	free uncommitted extended attributes from working map
  34 *
  35 */
  36static inline void free_ea_wmap(struct inode *inode)
  37{
  38	dxd_t *ea = &JFS_IP(inode)->ea;
  39
  40	if (ea->flag & DXD_EXTENT) {
  41		/* free EA pages from cache */
  42		invalidate_dxd_metapages(inode, *ea);
  43		dbFree(inode, addressDXD(ea), lengthDXD(ea));
  44	}
  45	ea->flag = 0;
  46}
  47
  48/*
  49 * NAME:	jfs_create(dip, dentry, mode)
  50 *
  51 * FUNCTION:	create a regular file in the parent directory <dip>
  52 *		with name = <from dentry> and mode = <mode>
  53 *
  54 * PARAMETER:	dip	- parent directory vnode
  55 *		dentry	- dentry of new file
  56 *		mode	- create mode (rwxrwxrwx).
  57 *		nd- nd struct
  58 *
  59 * RETURN:	Errors from subroutines
  60 *
  61 */
  62static int jfs_create(struct user_namespace *mnt_userns, struct inode *dip,
  63		      struct dentry *dentry, umode_t mode, bool excl)
  64{
  65	int rc = 0;
  66	tid_t tid;		/* transaction id */
  67	struct inode *ip = NULL;	/* child directory inode */
  68	ino_t ino;
  69	struct component_name dname;	/* child directory name */
  70	struct btstack btstack;
  71	struct inode *iplist[2];
  72	struct tblock *tblk;
  73
  74	jfs_info("jfs_create: dip:0x%p name:%pd", dip, dentry);
  75
  76	rc = dquot_initialize(dip);
  77	if (rc)
  78		goto out1;
  79
  80	/*
  81	 * search parent directory for entry/freespace
  82	 * (dtSearch() returns parent directory page pinned)
  83	 */
  84	if ((rc = get_UCSname(&dname, dentry)))
  85		goto out1;
  86
  87	/*
  88	 * Either iAlloc() or txBegin() may block.  Deadlock can occur if we
  89	 * block there while holding dtree page, so we allocate the inode &
  90	 * begin the transaction before we search the directory.
  91	 */
  92	ip = ialloc(dip, mode);
  93	if (IS_ERR(ip)) {
  94		rc = PTR_ERR(ip);
  95		goto out2;
  96	}
  97
  98	tid = txBegin(dip->i_sb, 0);
  99
 100	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 101	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 102
 103	rc = jfs_init_acl(tid, ip, dip);
 104	if (rc)
 105		goto out3;
 106
 107	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
 108	if (rc) {
 109		txAbort(tid, 0);
 110		goto out3;
 111	}
 112
 113	if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
 114		jfs_err("jfs_create: dtSearch returned %d", rc);
 115		txAbort(tid, 0);
 116		goto out3;
 117	}
 118
 119	tblk = tid_to_tblock(tid);
 120	tblk->xflag |= COMMIT_CREATE;
 121	tblk->ino = ip->i_ino;
 122	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
 123
 124	iplist[0] = dip;
 125	iplist[1] = ip;
 126
 127	/*
 128	 * initialize the child XAD tree root in-line in inode
 129	 */
 130	xtInitRoot(tid, ip);
 131
 132	/*
 133	 * create entry in parent directory for child directory
 134	 * (dtInsert() releases parent directory page)
 135	 */
 136	ino = ip->i_ino;
 137	if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
 138		if (rc == -EIO) {
 139			jfs_err("jfs_create: dtInsert returned -EIO");
 140			txAbort(tid, 1);	/* Marks Filesystem dirty */
 141		} else
 142			txAbort(tid, 0);	/* Filesystem full */
 143		goto out3;
 144	}
 145
 146	ip->i_op = &jfs_file_inode_operations;
 147	ip->i_fop = &jfs_file_operations;
 148	ip->i_mapping->a_ops = &jfs_aops;
 149
 150	mark_inode_dirty(ip);
 151
 152	dip->i_ctime = dip->i_mtime = current_time(dip);
 153
 154	mark_inode_dirty(dip);
 155
 156	rc = txCommit(tid, 2, &iplist[0], 0);
 157
 158      out3:
 159	txEnd(tid);
 160	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 161	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 162	if (rc) {
 163		free_ea_wmap(ip);
 164		clear_nlink(ip);
 165		discard_new_inode(ip);
 
 166	} else {
 167		d_instantiate_new(dentry, ip);
 
 168	}
 169
 170      out2:
 171	free_UCSname(&dname);
 172
 173      out1:
 174
 175	jfs_info("jfs_create: rc:%d", rc);
 176	return rc;
 177}
 178
 179
 180/*
 181 * NAME:	jfs_mkdir(dip, dentry, mode)
 182 *
 183 * FUNCTION:	create a child directory in the parent directory <dip>
 184 *		with name = <from dentry> and mode = <mode>
 185 *
 186 * PARAMETER:	dip	- parent directory vnode
 187 *		dentry	- dentry of child directory
 188 *		mode	- create mode (rwxrwxrwx).
 189 *
 190 * RETURN:	Errors from subroutines
 191 *
 192 * note:
 193 * EACCES: user needs search+write permission on the parent directory
 194 */
 195static int jfs_mkdir(struct user_namespace *mnt_userns, struct inode *dip,
 196		     struct dentry *dentry, umode_t mode)
 197{
 198	int rc = 0;
 199	tid_t tid;		/* transaction id */
 200	struct inode *ip = NULL;	/* child directory inode */
 201	ino_t ino;
 202	struct component_name dname;	/* child directory name */
 203	struct btstack btstack;
 204	struct inode *iplist[2];
 205	struct tblock *tblk;
 206
 207	jfs_info("jfs_mkdir: dip:0x%p name:%pd", dip, dentry);
 208
 209	rc = dquot_initialize(dip);
 210	if (rc)
 211		goto out1;
 212
 213	/*
 214	 * search parent directory for entry/freespace
 215	 * (dtSearch() returns parent directory page pinned)
 216	 */
 217	if ((rc = get_UCSname(&dname, dentry)))
 218		goto out1;
 219
 220	/*
 221	 * Either iAlloc() or txBegin() may block.  Deadlock can occur if we
 222	 * block there while holding dtree page, so we allocate the inode &
 223	 * begin the transaction before we search the directory.
 224	 */
 225	ip = ialloc(dip, S_IFDIR | mode);
 226	if (IS_ERR(ip)) {
 227		rc = PTR_ERR(ip);
 228		goto out2;
 229	}
 230
 231	tid = txBegin(dip->i_sb, 0);
 232
 233	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 234	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 235
 236	rc = jfs_init_acl(tid, ip, dip);
 237	if (rc)
 238		goto out3;
 239
 240	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
 241	if (rc) {
 242		txAbort(tid, 0);
 243		goto out3;
 244	}
 245
 246	if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
 247		jfs_err("jfs_mkdir: dtSearch returned %d", rc);
 248		txAbort(tid, 0);
 249		goto out3;
 250	}
 251
 252	tblk = tid_to_tblock(tid);
 253	tblk->xflag |= COMMIT_CREATE;
 254	tblk->ino = ip->i_ino;
 255	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
 256
 257	iplist[0] = dip;
 258	iplist[1] = ip;
 259
 260	/*
 261	 * initialize the child directory in-line in inode
 262	 */
 263	dtInitRoot(tid, ip, dip->i_ino);
 264
 265	/*
 266	 * create entry in parent directory for child directory
 267	 * (dtInsert() releases parent directory page)
 268	 */
 269	ino = ip->i_ino;
 270	if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
 271		if (rc == -EIO) {
 272			jfs_err("jfs_mkdir: dtInsert returned -EIO");
 273			txAbort(tid, 1);	/* Marks Filesystem dirty */
 274		} else
 275			txAbort(tid, 0);	/* Filesystem full */
 276		goto out3;
 277	}
 278
 279	set_nlink(ip, 2);	/* for '.' */
 280	ip->i_op = &jfs_dir_inode_operations;
 281	ip->i_fop = &jfs_dir_operations;
 282
 283	mark_inode_dirty(ip);
 284
 285	/* update parent directory inode */
 286	inc_nlink(dip);		/* for '..' from child directory */
 287	dip->i_ctime = dip->i_mtime = current_time(dip);
 288	mark_inode_dirty(dip);
 289
 290	rc = txCommit(tid, 2, &iplist[0], 0);
 291
 292      out3:
 293	txEnd(tid);
 294	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 295	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 296	if (rc) {
 297		free_ea_wmap(ip);
 298		clear_nlink(ip);
 299		discard_new_inode(ip);
 
 300	} else {
 301		d_instantiate_new(dentry, ip);
 
 302	}
 303
 304      out2:
 305	free_UCSname(&dname);
 306
 307
 308      out1:
 309
 310	jfs_info("jfs_mkdir: rc:%d", rc);
 311	return rc;
 312}
 313
 314/*
 315 * NAME:	jfs_rmdir(dip, dentry)
 316 *
 317 * FUNCTION:	remove a link to child directory
 318 *
 319 * PARAMETER:	dip	- parent inode
 320 *		dentry	- child directory dentry
 321 *
 322 * RETURN:	-EINVAL	- if name is . or ..
 323 *		-EINVAL - if . or .. exist but are invalid.
 324 *		errors from subroutines
 325 *
 326 * note:
 327 * if other threads have the directory open when the last link
 328 * is removed, the "." and ".." entries, if present, are removed before
 329 * rmdir() returns and no new entries may be created in the directory,
 330 * but the directory is not removed until the last reference to
 331 * the directory is released (cf.unlink() of regular file).
 332 */
 333static int jfs_rmdir(struct inode *dip, struct dentry *dentry)
 334{
 335	int rc;
 336	tid_t tid;		/* transaction id */
 337	struct inode *ip = d_inode(dentry);
 338	ino_t ino;
 339	struct component_name dname;
 340	struct inode *iplist[2];
 341	struct tblock *tblk;
 342
 343	jfs_info("jfs_rmdir: dip:0x%p name:%pd", dip, dentry);
 344
 345	/* Init inode for quota operations. */
 346	rc = dquot_initialize(dip);
 347	if (rc)
 348		goto out;
 349	rc = dquot_initialize(ip);
 350	if (rc)
 351		goto out;
 352
 353	/* directory must be empty to be removed */
 354	if (!dtEmpty(ip)) {
 355		rc = -ENOTEMPTY;
 356		goto out;
 357	}
 358
 359	if ((rc = get_UCSname(&dname, dentry))) {
 360		goto out;
 361	}
 362
 363	tid = txBegin(dip->i_sb, 0);
 364
 365	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 366	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 367
 368	iplist[0] = dip;
 369	iplist[1] = ip;
 370
 371	tblk = tid_to_tblock(tid);
 372	tblk->xflag |= COMMIT_DELETE;
 373	tblk->u.ip = ip;
 374
 375	/*
 376	 * delete the entry of target directory from parent directory
 377	 */
 378	ino = ip->i_ino;
 379	if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
 380		jfs_err("jfs_rmdir: dtDelete returned %d", rc);
 381		if (rc == -EIO)
 382			txAbort(tid, 1);
 383		txEnd(tid);
 384		mutex_unlock(&JFS_IP(ip)->commit_mutex);
 385		mutex_unlock(&JFS_IP(dip)->commit_mutex);
 386
 387		goto out2;
 388	}
 389
 390	/* update parent directory's link count corresponding
 391	 * to ".." entry of the target directory deleted
 392	 */
 393	dip->i_ctime = dip->i_mtime = current_time(dip);
 394	inode_dec_link_count(dip);
 395
 396	/*
 397	 * OS/2 could have created EA and/or ACL
 398	 */
 399	/* free EA from both persistent and working map */
 400	if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
 401		/* free EA pages */
 402		txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
 403	}
 404	JFS_IP(ip)->ea.flag = 0;
 405
 406	/* free ACL from both persistent and working map */
 407	if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
 408		/* free ACL pages */
 409		txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
 410	}
 411	JFS_IP(ip)->acl.flag = 0;
 412
 413	/* mark the target directory as deleted */
 414	clear_nlink(ip);
 415	mark_inode_dirty(ip);
 416
 417	rc = txCommit(tid, 2, &iplist[0], 0);
 418
 419	txEnd(tid);
 420
 421	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 422	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 423
 424	/*
 425	 * Truncating the directory index table is not guaranteed.  It
 426	 * may need to be done iteratively
 427	 */
 428	if (test_cflag(COMMIT_Stale, dip)) {
 429		if (dip->i_size > 1)
 430			jfs_truncate_nolock(dip, 0);
 431
 432		clear_cflag(COMMIT_Stale, dip);
 433	}
 434
 435      out2:
 436	free_UCSname(&dname);
 437
 438      out:
 439	jfs_info("jfs_rmdir: rc:%d", rc);
 440	return rc;
 441}
 442
 443/*
 444 * NAME:	jfs_unlink(dip, dentry)
 445 *
 446 * FUNCTION:	remove a link to object <vp> named by <name>
 447 *		from parent directory <dvp>
 448 *
 449 * PARAMETER:	dip	- inode of parent directory
 450 *		dentry	- dentry of object to be removed
 451 *
 452 * RETURN:	errors from subroutines
 453 *
 454 * note:
 455 * temporary file: if one or more processes have the file open
 456 * when the last link is removed, the link will be removed before
 457 * unlink() returns, but the removal of the file contents will be
 458 * postponed until all references to the files are closed.
 459 *
 460 * JFS does NOT support unlink() on directories.
 461 *
 462 */
 463static int jfs_unlink(struct inode *dip, struct dentry *dentry)
 464{
 465	int rc;
 466	tid_t tid;		/* transaction id */
 467	struct inode *ip = d_inode(dentry);
 468	ino_t ino;
 469	struct component_name dname;	/* object name */
 470	struct inode *iplist[2];
 471	struct tblock *tblk;
 472	s64 new_size = 0;
 473	int commit_flag;
 474
 475	jfs_info("jfs_unlink: dip:0x%p name:%pd", dip, dentry);
 476
 477	/* Init inode for quota operations. */
 478	rc = dquot_initialize(dip);
 479	if (rc)
 480		goto out;
 481	rc = dquot_initialize(ip);
 482	if (rc)
 483		goto out;
 484
 485	if ((rc = get_UCSname(&dname, dentry)))
 486		goto out;
 487
 488	IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 489
 490	tid = txBegin(dip->i_sb, 0);
 491
 492	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 493	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 494
 495	iplist[0] = dip;
 496	iplist[1] = ip;
 497
 498	/*
 499	 * delete the entry of target file from parent directory
 500	 */
 501	ino = ip->i_ino;
 502	if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
 503		jfs_err("jfs_unlink: dtDelete returned %d", rc);
 504		if (rc == -EIO)
 505			txAbort(tid, 1);	/* Marks FS Dirty */
 506		txEnd(tid);
 507		mutex_unlock(&JFS_IP(ip)->commit_mutex);
 508		mutex_unlock(&JFS_IP(dip)->commit_mutex);
 509		IWRITE_UNLOCK(ip);
 510		goto out1;
 511	}
 512
 513	ASSERT(ip->i_nlink);
 514
 515	ip->i_ctime = dip->i_ctime = dip->i_mtime = current_time(ip);
 516	mark_inode_dirty(dip);
 517
 518	/* update target's inode */
 519	inode_dec_link_count(ip);
 520
 521	/*
 522	 *	commit zero link count object
 523	 */
 524	if (ip->i_nlink == 0) {
 525		assert(!test_cflag(COMMIT_Nolink, ip));
 526		/* free block resources */
 527		if ((new_size = commitZeroLink(tid, ip)) < 0) {
 528			txAbort(tid, 1);	/* Marks FS Dirty */
 529			txEnd(tid);
 530			mutex_unlock(&JFS_IP(ip)->commit_mutex);
 531			mutex_unlock(&JFS_IP(dip)->commit_mutex);
 532			IWRITE_UNLOCK(ip);
 533			rc = new_size;
 534			goto out1;
 535		}
 536		tblk = tid_to_tblock(tid);
 537		tblk->xflag |= COMMIT_DELETE;
 538		tblk->u.ip = ip;
 539	}
 540
 541	/*
 542	 * Incomplete truncate of file data can
 543	 * result in timing problems unless we synchronously commit the
 544	 * transaction.
 545	 */
 546	if (new_size)
 547		commit_flag = COMMIT_SYNC;
 548	else
 549		commit_flag = 0;
 550
 551	/*
 552	 * If xtTruncate was incomplete, commit synchronously to avoid
 553	 * timing complications
 554	 */
 555	rc = txCommit(tid, 2, &iplist[0], commit_flag);
 556
 557	txEnd(tid);
 558
 559	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 560	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 561
 562	while (new_size && (rc == 0)) {
 563		tid = txBegin(dip->i_sb, 0);
 564		mutex_lock(&JFS_IP(ip)->commit_mutex);
 565		new_size = xtTruncate_pmap(tid, ip, new_size);
 566		if (new_size < 0) {
 567			txAbort(tid, 1);	/* Marks FS Dirty */
 568			rc = new_size;
 569		} else
 570			rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
 571		txEnd(tid);
 572		mutex_unlock(&JFS_IP(ip)->commit_mutex);
 573	}
 574
 575	if (ip->i_nlink == 0)
 576		set_cflag(COMMIT_Nolink, ip);
 577
 578	IWRITE_UNLOCK(ip);
 579
 580	/*
 581	 * Truncating the directory index table is not guaranteed.  It
 582	 * may need to be done iteratively
 583	 */
 584	if (test_cflag(COMMIT_Stale, dip)) {
 585		if (dip->i_size > 1)
 586			jfs_truncate_nolock(dip, 0);
 587
 588		clear_cflag(COMMIT_Stale, dip);
 589	}
 590
 591      out1:
 592	free_UCSname(&dname);
 593      out:
 594	jfs_info("jfs_unlink: rc:%d", rc);
 595	return rc;
 596}
 597
 598/*
 599 * NAME:	commitZeroLink()
 600 *
 601 * FUNCTION:	for non-directory, called by jfs_remove(),
 602 *		truncate a regular file, directory or symbolic
 603 *		link to zero length. return 0 if type is not
 604 *		one of these.
 605 *
 606 *		if the file is currently associated with a VM segment
 607 *		only permanent disk and inode map resources are freed,
 608 *		and neither the inode nor indirect blocks are modified
 609 *		so that the resources can be later freed in the work
 610 *		map by ctrunc1.
 611 *		if there is no VM segment on entry, the resources are
 612 *		freed in both work and permanent map.
 613 *		(? for temporary file - memory object is cached even
 614 *		after no reference:
 615 *		reference count > 0 -   )
 616 *
 617 * PARAMETERS:	cd	- pointer to commit data structure.
 618 *			  current inode is the one to truncate.
 619 *
 620 * RETURN:	Errors from subroutines
 621 */
 622static s64 commitZeroLink(tid_t tid, struct inode *ip)
 623{
 624	int filetype;
 625	struct tblock *tblk;
 626
 627	jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
 628
 629	filetype = ip->i_mode & S_IFMT;
 630	switch (filetype) {
 631	case S_IFREG:
 632		break;
 633	case S_IFLNK:
 634		/* fast symbolic link */
 635		if (ip->i_size < IDATASIZE) {
 636			ip->i_size = 0;
 637			return 0;
 638		}
 639		break;
 640	default:
 641		assert(filetype != S_IFDIR);
 642		return 0;
 643	}
 644
 645	set_cflag(COMMIT_Freewmap, ip);
 646
 647	/* mark transaction of block map update type */
 648	tblk = tid_to_tblock(tid);
 649	tblk->xflag |= COMMIT_PMAP;
 650
 651	/*
 652	 * free EA
 653	 */
 654	if (JFS_IP(ip)->ea.flag & DXD_EXTENT)
 655		/* acquire maplock on EA to be freed from block map */
 656		txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
 657
 658	/*
 659	 * free ACL
 660	 */
 661	if (JFS_IP(ip)->acl.flag & DXD_EXTENT)
 662		/* acquire maplock on EA to be freed from block map */
 663		txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
 664
 665	/*
 666	 * free xtree/data (truncate to zero length):
 667	 * free xtree/data pages from cache if COMMIT_PWMAP,
 668	 * free xtree/data blocks from persistent block map, and
 669	 * free xtree/data blocks from working block map if COMMIT_PWMAP;
 670	 */
 671	if (ip->i_size)
 672		return xtTruncate_pmap(tid, ip, 0);
 673
 674	return 0;
 675}
 676
 677
 678/*
 679 * NAME:	jfs_free_zero_link()
 680 *
 681 * FUNCTION:	for non-directory, called by iClose(),
 682 *		free resources of a file from cache and WORKING map
 683 *		for a file previously committed with zero link count
 684 *		while associated with a pager object,
 685 *
 686 * PARAMETER:	ip	- pointer to inode of file.
 687 */
 688void jfs_free_zero_link(struct inode *ip)
 689{
 690	int type;
 691
 692	jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
 693
 694	/* return if not reg or symbolic link or if size is
 695	 * already ok.
 696	 */
 697	type = ip->i_mode & S_IFMT;
 698
 699	switch (type) {
 700	case S_IFREG:
 701		break;
 702	case S_IFLNK:
 703		/* if its contained in inode nothing to do */
 704		if (ip->i_size < IDATASIZE)
 705			return;
 706		break;
 707	default:
 708		return;
 709	}
 710
 711	/*
 712	 * free EA
 713	 */
 714	if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
 715		s64 xaddr = addressDXD(&JFS_IP(ip)->ea);
 716		int xlen = lengthDXD(&JFS_IP(ip)->ea);
 717		struct maplock maplock;	/* maplock for COMMIT_WMAP */
 718		struct pxd_lock *pxdlock;	/* maplock for COMMIT_WMAP */
 719
 720		/* free EA pages from cache */
 721		invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
 722
 723		/* free EA extent from working block map */
 724		maplock.index = 1;
 725		pxdlock = (struct pxd_lock *) & maplock;
 726		pxdlock->flag = mlckFREEPXD;
 727		PXDaddress(&pxdlock->pxd, xaddr);
 728		PXDlength(&pxdlock->pxd, xlen);
 729		txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
 730	}
 731
 732	/*
 733	 * free ACL
 734	 */
 735	if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
 736		s64 xaddr = addressDXD(&JFS_IP(ip)->acl);
 737		int xlen = lengthDXD(&JFS_IP(ip)->acl);
 738		struct maplock maplock;	/* maplock for COMMIT_WMAP */
 739		struct pxd_lock *pxdlock;	/* maplock for COMMIT_WMAP */
 740
 741		invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
 742
 743		/* free ACL extent from working block map */
 744		maplock.index = 1;
 745		pxdlock = (struct pxd_lock *) & maplock;
 746		pxdlock->flag = mlckFREEPXD;
 747		PXDaddress(&pxdlock->pxd, xaddr);
 748		PXDlength(&pxdlock->pxd, xlen);
 749		txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
 750	}
 751
 752	/*
 753	 * free xtree/data (truncate to zero length):
 754	 * free xtree/data pages from cache, and
 755	 * free xtree/data blocks from working block map;
 756	 */
 757	if (ip->i_size)
 758		xtTruncate(0, ip, 0, COMMIT_WMAP);
 759}
 760
 761/*
 762 * NAME:	jfs_link(vp, dvp, name, crp)
 763 *
 764 * FUNCTION:	create a link to <vp> by the name = <name>
 765 *		in the parent directory <dvp>
 766 *
 767 * PARAMETER:	vp	- target object
 768 *		dvp	- parent directory of new link
 769 *		name	- name of new link to target object
 770 *		crp	- credential
 771 *
 772 * RETURN:	Errors from subroutines
 773 *
 774 * note:
 775 * JFS does NOT support link() on directories (to prevent circular
 776 * path in the directory hierarchy);
 777 * EPERM: the target object is a directory, and either the caller
 778 * does not have appropriate privileges or the implementation prohibits
 779 * using link() on directories [XPG4.2].
 780 *
 781 * JFS does NOT support links between file systems:
 782 * EXDEV: target object and new link are on different file systems and
 783 * implementation does not support links between file systems [XPG4.2].
 784 */
 785static int jfs_link(struct dentry *old_dentry,
 786	     struct inode *dir, struct dentry *dentry)
 787{
 788	int rc;
 789	tid_t tid;
 790	struct inode *ip = d_inode(old_dentry);
 791	ino_t ino;
 792	struct component_name dname;
 793	struct btstack btstack;
 794	struct inode *iplist[2];
 795
 796	jfs_info("jfs_link: %pd %pd", old_dentry, dentry);
 797
 798	rc = dquot_initialize(dir);
 799	if (rc)
 800		goto out;
 801
 802	tid = txBegin(ip->i_sb, 0);
 803
 804	mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
 805	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 806
 807	/*
 808	 * scan parent directory for entry/freespace
 809	 */
 810	if ((rc = get_UCSname(&dname, dentry)))
 811		goto out_tx;
 812
 813	if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
 814		goto free_dname;
 815
 816	/*
 817	 * create entry for new link in parent directory
 818	 */
 819	ino = ip->i_ino;
 820	if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
 821		goto free_dname;
 822
 823	/* update object inode */
 824	inc_nlink(ip);		/* for new link */
 825	ip->i_ctime = current_time(ip);
 826	dir->i_ctime = dir->i_mtime = current_time(dir);
 827	mark_inode_dirty(dir);
 828	ihold(ip);
 829
 830	iplist[0] = ip;
 831	iplist[1] = dir;
 832	rc = txCommit(tid, 2, &iplist[0], 0);
 833
 834	if (rc) {
 835		drop_nlink(ip); /* never instantiated */
 836		iput(ip);
 837	} else
 838		d_instantiate(dentry, ip);
 839
 840      free_dname:
 841	free_UCSname(&dname);
 842
 843      out_tx:
 844	txEnd(tid);
 845
 846	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 847	mutex_unlock(&JFS_IP(dir)->commit_mutex);
 848
 849      out:
 850	jfs_info("jfs_link: rc:%d", rc);
 851	return rc;
 852}
 853
 854/*
 855 * NAME:	jfs_symlink(dip, dentry, name)
 856 *
 857 * FUNCTION:	creates a symbolic link to <symlink> by name <name>
 858 *			in directory <dip>
 859 *
 860 * PARAMETER:	dip	- parent directory vnode
 861 *		dentry	- dentry of symbolic link
 862 *		name	- the path name of the existing object
 863 *			  that will be the source of the link
 864 *
 865 * RETURN:	errors from subroutines
 866 *
 867 * note:
 868 * ENAMETOOLONG: pathname resolution of a symbolic link produced
 869 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
 870*/
 871
 872static int jfs_symlink(struct user_namespace *mnt_userns, struct inode *dip,
 873		       struct dentry *dentry, const char *name)
 874{
 875	int rc;
 876	tid_t tid;
 877	ino_t ino = 0;
 878	struct component_name dname;
 879	int ssize;		/* source pathname size */
 880	struct btstack btstack;
 881	struct inode *ip = d_inode(dentry);
 882	s64 xlen = 0;
 883	int bmask = 0, xsize;
 884	s64 xaddr;
 885	struct metapage *mp;
 886	struct super_block *sb;
 887	struct tblock *tblk;
 888
 889	struct inode *iplist[2];
 890
 891	jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
 892
 893	rc = dquot_initialize(dip);
 894	if (rc)
 895		goto out1;
 896
 897	ssize = strlen(name) + 1;
 898
 899	/*
 900	 * search parent directory for entry/freespace
 901	 * (dtSearch() returns parent directory page pinned)
 902	 */
 903
 904	if ((rc = get_UCSname(&dname, dentry)))
 905		goto out1;
 906
 907	/*
 908	 * allocate on-disk/in-memory inode for symbolic link:
 909	 * (iAlloc() returns new, locked inode)
 910	 */
 911	ip = ialloc(dip, S_IFLNK | 0777);
 912	if (IS_ERR(ip)) {
 913		rc = PTR_ERR(ip);
 914		goto out2;
 915	}
 916
 917	tid = txBegin(dip->i_sb, 0);
 918
 919	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 920	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 921
 922	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
 923	if (rc)
 924		goto out3;
 925
 926	tblk = tid_to_tblock(tid);
 927	tblk->xflag |= COMMIT_CREATE;
 928	tblk->ino = ip->i_ino;
 929	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
 930
 931	/* fix symlink access permission
 932	 * (dir_create() ANDs in the u.u_cmask,
 933	 * but symlinks really need to be 777 access)
 934	 */
 935	ip->i_mode |= 0777;
 936
 937	/*
 938	 * write symbolic link target path name
 939	 */
 940	xtInitRoot(tid, ip);
 941
 942	/*
 943	 * write source path name inline in on-disk inode (fast symbolic link)
 944	 */
 945
 946	if (ssize <= IDATASIZE) {
 947		ip->i_op = &jfs_fast_symlink_inode_operations;
 948
 949		ip->i_link = JFS_IP(ip)->i_inline_all;
 950		memcpy(ip->i_link, name, ssize);
 951		ip->i_size = ssize - 1;
 952
 953		/*
 954		 * if symlink is > 128 bytes, we don't have the space to
 955		 * store inline extended attributes
 956		 */
 957		if (ssize > sizeof (JFS_IP(ip)->i_inline))
 958			JFS_IP(ip)->mode2 &= ~INLINEEA;
 959
 960		jfs_info("jfs_symlink: fast symlink added  ssize:%d name:%s ",
 961			 ssize, name);
 962	}
 963	/*
 964	 * write source path name in a single extent
 965	 */
 966	else {
 967		jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
 968
 969		ip->i_op = &jfs_symlink_inode_operations;
 970		inode_nohighmem(ip);
 971		ip->i_mapping->a_ops = &jfs_aops;
 972
 973		/*
 974		 * even though the data of symlink object (source
 975		 * path name) is treated as non-journaled user data,
 976		 * it is read/written thru buffer cache for performance.
 977		 */
 978		sb = ip->i_sb;
 979		bmask = JFS_SBI(sb)->bsize - 1;
 980		xsize = (ssize + bmask) & ~bmask;
 981		xaddr = 0;
 982		xlen = xsize >> JFS_SBI(sb)->l2bsize;
 983		if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
 984			txAbort(tid, 0);
 985			goto out3;
 986		}
 987		ip->i_size = ssize - 1;
 988		while (ssize) {
 989			/* This is kind of silly since PATH_MAX == 4K */
 990			int copy_size = min(ssize, PSIZE);
 991
 992			mp = get_metapage(ip, xaddr, PSIZE, 1);
 993
 994			if (mp == NULL) {
 995				xtTruncate(tid, ip, 0, COMMIT_PWMAP);
 996				rc = -EIO;
 997				txAbort(tid, 0);
 998				goto out3;
 999			}
1000			memcpy(mp->data, name, copy_size);
1001			flush_metapage(mp);
1002			ssize -= copy_size;
1003			name += copy_size;
1004			xaddr += JFS_SBI(sb)->nbperpage;
1005		}
1006	}
1007
1008	/*
1009	 * create entry for symbolic link in parent directory
1010	 */
1011	rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1012	if (rc == 0) {
1013		ino = ip->i_ino;
1014		rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1015	}
1016	if (rc) {
1017		if (xlen)
1018			xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1019		txAbort(tid, 0);
1020		/* discard new inode */
1021		goto out3;
1022	}
1023
1024	mark_inode_dirty(ip);
1025
1026	dip->i_ctime = dip->i_mtime = current_time(dip);
1027	mark_inode_dirty(dip);
1028	/*
1029	 * commit update of parent directory and link object
1030	 */
1031
1032	iplist[0] = dip;
1033	iplist[1] = ip;
1034	rc = txCommit(tid, 2, &iplist[0], 0);
1035
1036      out3:
1037	txEnd(tid);
1038	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1039	mutex_unlock(&JFS_IP(dip)->commit_mutex);
1040	if (rc) {
1041		free_ea_wmap(ip);
1042		clear_nlink(ip);
1043		discard_new_inode(ip);
 
1044	} else {
1045		d_instantiate_new(dentry, ip);
 
1046	}
1047
1048      out2:
1049	free_UCSname(&dname);
1050
1051      out1:
1052	jfs_info("jfs_symlink: rc:%d", rc);
1053	return rc;
1054}
1055
1056
1057/*
1058 * NAME:	jfs_rename
1059 *
1060 * FUNCTION:	rename a file or directory
1061 */
1062static int jfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
1063		      struct dentry *old_dentry, struct inode *new_dir,
1064		      struct dentry *new_dentry, unsigned int flags)
1065{
1066	struct btstack btstack;
1067	ino_t ino;
1068	struct component_name new_dname;
1069	struct inode *new_ip;
1070	struct component_name old_dname;
1071	struct inode *old_ip;
1072	int rc;
1073	tid_t tid;
1074	struct tlock *tlck;
1075	struct dt_lock *dtlck;
1076	struct lv *lv;
1077	int ipcount;
1078	struct inode *iplist[4];
1079	struct tblock *tblk;
1080	s64 new_size = 0;
1081	int commit_flag;
1082
1083	if (flags & ~RENAME_NOREPLACE)
1084		return -EINVAL;
1085
1086	jfs_info("jfs_rename: %pd %pd", old_dentry, new_dentry);
1087
1088	rc = dquot_initialize(old_dir);
1089	if (rc)
1090		goto out1;
1091	rc = dquot_initialize(new_dir);
1092	if (rc)
1093		goto out1;
1094
1095	old_ip = d_inode(old_dentry);
1096	new_ip = d_inode(new_dentry);
1097
1098	if ((rc = get_UCSname(&old_dname, old_dentry)))
1099		goto out1;
1100
1101	if ((rc = get_UCSname(&new_dname, new_dentry)))
1102		goto out2;
1103
1104	/*
1105	 * Make sure source inode number is what we think it is
1106	 */
1107	rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1108	if (rc || (ino != old_ip->i_ino)) {
1109		rc = -ENOENT;
1110		goto out3;
1111	}
1112
1113	/*
1114	 * Make sure dest inode number (if any) is what we think it is
1115	 */
1116	rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1117	if (!rc) {
1118		if ((!new_ip) || (ino != new_ip->i_ino)) {
1119			rc = -ESTALE;
1120			goto out3;
1121		}
1122	} else if (rc != -ENOENT)
1123		goto out3;
1124	else if (new_ip) {
1125		/* no entry exists, but one was expected */
1126		rc = -ESTALE;
1127		goto out3;
1128	}
1129
1130	if (S_ISDIR(old_ip->i_mode)) {
1131		if (new_ip) {
1132			if (!dtEmpty(new_ip)) {
1133				rc = -ENOTEMPTY;
1134				goto out3;
1135			}
1136		}
1137	} else if (new_ip) {
1138		IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1139		/* Init inode for quota operations. */
1140		rc = dquot_initialize(new_ip);
1141		if (rc)
1142			goto out_unlock;
1143	}
1144
1145	/*
1146	 * The real work starts here
1147	 */
1148	tid = txBegin(new_dir->i_sb, 0);
1149
1150	/*
1151	 * How do we know the locking is safe from deadlocks?
1152	 * The vfs does the hard part for us.  Any time we are taking nested
1153	 * commit_mutexes, the vfs already has i_mutex held on the parent.
1154	 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1155	 */
1156	mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1157	mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1158	if (old_dir != new_dir)
1159		mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
1160				  COMMIT_MUTEX_SECOND_PARENT);
1161
1162	if (new_ip) {
1163		mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1164				  COMMIT_MUTEX_VICTIM);
1165		/*
1166		 * Change existing directory entry to new inode number
1167		 */
1168		ino = new_ip->i_ino;
1169		rc = dtModify(tid, new_dir, &new_dname, &ino,
1170			      old_ip->i_ino, JFS_RENAME);
1171		if (rc)
1172			goto out_tx;
1173		drop_nlink(new_ip);
1174		if (S_ISDIR(new_ip->i_mode)) {
1175			drop_nlink(new_ip);
1176			if (new_ip->i_nlink) {
1177				mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1178				if (old_dir != new_dir)
1179					mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1180				mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1181				mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1182				if (!S_ISDIR(old_ip->i_mode) && new_ip)
1183					IWRITE_UNLOCK(new_ip);
1184				jfs_error(new_ip->i_sb,
1185					  "new_ip->i_nlink != 0\n");
1186				return -EIO;
1187			}
1188			tblk = tid_to_tblock(tid);
1189			tblk->xflag |= COMMIT_DELETE;
1190			tblk->u.ip = new_ip;
1191		} else if (new_ip->i_nlink == 0) {
1192			assert(!test_cflag(COMMIT_Nolink, new_ip));
1193			/* free block resources */
1194			if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
1195				txAbort(tid, 1);	/* Marks FS Dirty */
1196				rc = new_size;
1197				goto out_tx;
1198			}
1199			tblk = tid_to_tblock(tid);
1200			tblk->xflag |= COMMIT_DELETE;
1201			tblk->u.ip = new_ip;
1202		} else {
1203			new_ip->i_ctime = current_time(new_ip);
1204			mark_inode_dirty(new_ip);
1205		}
1206	} else {
1207		/*
1208		 * Add new directory entry
1209		 */
1210		rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1211			      JFS_CREATE);
1212		if (rc) {
1213			jfs_err("jfs_rename didn't expect dtSearch to fail w/rc = %d",
1214				rc);
1215			goto out_tx;
1216		}
1217
1218		ino = old_ip->i_ino;
1219		rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1220		if (rc) {
1221			if (rc == -EIO)
1222				jfs_err("jfs_rename: dtInsert returned -EIO");
1223			goto out_tx;
1224		}
1225		if (S_ISDIR(old_ip->i_mode))
1226			inc_nlink(new_dir);
1227	}
1228	/*
1229	 * Remove old directory entry
1230	 */
1231
1232	ino = old_ip->i_ino;
1233	rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1234	if (rc) {
1235		jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1236			rc);
1237		txAbort(tid, 1);	/* Marks Filesystem dirty */
1238		goto out_tx;
1239	}
1240	if (S_ISDIR(old_ip->i_mode)) {
1241		drop_nlink(old_dir);
1242		if (old_dir != new_dir) {
1243			/*
1244			 * Change inode number of parent for moved directory
1245			 */
1246
1247			JFS_IP(old_ip)->i_dtroot.header.idotdot =
1248				cpu_to_le32(new_dir->i_ino);
1249
1250			/* Linelock header of dtree */
1251			tlck = txLock(tid, old_ip,
1252				    (struct metapage *) &JFS_IP(old_ip)->bxflag,
1253				      tlckDTREE | tlckBTROOT | tlckRELINK);
1254			dtlck = (struct dt_lock *) & tlck->lock;
1255			ASSERT(dtlck->index == 0);
1256			lv = & dtlck->lv[0];
1257			lv->offset = 0;
1258			lv->length = 1;
1259			dtlck->index++;
1260		}
1261	}
1262
1263	/*
1264	 * Update ctime on changed/moved inodes & mark dirty
1265	 */
1266	old_ip->i_ctime = current_time(old_ip);
1267	mark_inode_dirty(old_ip);
1268
1269	new_dir->i_ctime = new_dir->i_mtime = current_time(new_dir);
1270	mark_inode_dirty(new_dir);
1271
1272	/* Build list of inodes modified by this transaction */
1273	ipcount = 0;
1274	iplist[ipcount++] = old_ip;
1275	if (new_ip)
1276		iplist[ipcount++] = new_ip;
1277	iplist[ipcount++] = old_dir;
1278
1279	if (old_dir != new_dir) {
1280		iplist[ipcount++] = new_dir;
1281		old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1282		mark_inode_dirty(old_dir);
1283	}
1284
1285	/*
1286	 * Incomplete truncate of file data can
1287	 * result in timing problems unless we synchronously commit the
1288	 * transaction.
1289	 */
1290	if (new_size)
1291		commit_flag = COMMIT_SYNC;
1292	else
1293		commit_flag = 0;
1294
1295	rc = txCommit(tid, ipcount, iplist, commit_flag);
1296
1297      out_tx:
1298	txEnd(tid);
1299	if (new_ip)
1300		mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1301	if (old_dir != new_dir)
1302		mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1303	mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1304	mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1305
1306	while (new_size && (rc == 0)) {
1307		tid = txBegin(new_ip->i_sb, 0);
1308		mutex_lock(&JFS_IP(new_ip)->commit_mutex);
1309		new_size = xtTruncate_pmap(tid, new_ip, new_size);
1310		if (new_size < 0) {
1311			txAbort(tid, 1);
1312			rc = new_size;
1313		} else
1314			rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1315		txEnd(tid);
1316		mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1317	}
1318	if (new_ip && (new_ip->i_nlink == 0))
1319		set_cflag(COMMIT_Nolink, new_ip);
1320	/*
1321	 * Truncating the directory index table is not guaranteed.  It
1322	 * may need to be done iteratively
1323	 */
1324	if (test_cflag(COMMIT_Stale, old_dir)) {
1325		if (old_dir->i_size > 1)
1326			jfs_truncate_nolock(old_dir, 0);
1327
1328		clear_cflag(COMMIT_Stale, old_dir);
1329	}
1330      out_unlock:
1331	if (new_ip && !S_ISDIR(new_ip->i_mode))
1332		IWRITE_UNLOCK(new_ip);
1333      out3:
1334	free_UCSname(&new_dname);
1335      out2:
1336	free_UCSname(&old_dname);
1337      out1:
1338	jfs_info("jfs_rename: returning %d", rc);
1339	return rc;
1340}
1341
1342
1343/*
1344 * NAME:	jfs_mknod
1345 *
1346 * FUNCTION:	Create a special file (device)
1347 */
1348static int jfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1349		     struct dentry *dentry, umode_t mode, dev_t rdev)
1350{
1351	struct jfs_inode_info *jfs_ip;
1352	struct btstack btstack;
1353	struct component_name dname;
1354	ino_t ino;
1355	struct inode *ip;
1356	struct inode *iplist[2];
1357	int rc;
1358	tid_t tid;
1359	struct tblock *tblk;
1360
1361	jfs_info("jfs_mknod: %pd", dentry);
1362
1363	rc = dquot_initialize(dir);
1364	if (rc)
1365		goto out;
1366
1367	if ((rc = get_UCSname(&dname, dentry)))
1368		goto out;
1369
1370	ip = ialloc(dir, mode);
1371	if (IS_ERR(ip)) {
1372		rc = PTR_ERR(ip);
1373		goto out1;
1374	}
1375	jfs_ip = JFS_IP(ip);
1376
1377	tid = txBegin(dir->i_sb, 0);
1378
1379	mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1380	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1381
1382	rc = jfs_init_acl(tid, ip, dir);
1383	if (rc)
1384		goto out3;
1385
1386	rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1387	if (rc) {
1388		txAbort(tid, 0);
1389		goto out3;
1390	}
1391
1392	if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1393		txAbort(tid, 0);
1394		goto out3;
1395	}
1396
1397	tblk = tid_to_tblock(tid);
1398	tblk->xflag |= COMMIT_CREATE;
1399	tblk->ino = ip->i_ino;
1400	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
1401
1402	ino = ip->i_ino;
1403	if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1404		txAbort(tid, 0);
1405		goto out3;
1406	}
1407
1408	ip->i_op = &jfs_file_inode_operations;
1409	jfs_ip->dev = new_encode_dev(rdev);
1410	init_special_inode(ip, ip->i_mode, rdev);
1411
1412	mark_inode_dirty(ip);
1413
1414	dir->i_ctime = dir->i_mtime = current_time(dir);
1415
1416	mark_inode_dirty(dir);
1417
1418	iplist[0] = dir;
1419	iplist[1] = ip;
1420	rc = txCommit(tid, 2, iplist, 0);
1421
1422      out3:
1423	txEnd(tid);
1424	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1425	mutex_unlock(&JFS_IP(dir)->commit_mutex);
1426	if (rc) {
1427		free_ea_wmap(ip);
1428		clear_nlink(ip);
1429		discard_new_inode(ip);
 
1430	} else {
1431		d_instantiate_new(dentry, ip);
 
1432	}
1433
1434      out1:
1435	free_UCSname(&dname);
1436
1437      out:
1438	jfs_info("jfs_mknod: returning %d", rc);
1439	return rc;
1440}
1441
1442static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags)
1443{
1444	struct btstack btstack;
1445	ino_t inum;
1446	struct inode *ip;
1447	struct component_name key;
1448	int rc;
1449
1450	jfs_info("jfs_lookup: name = %pd", dentry);
1451
1452	if ((rc = get_UCSname(&key, dentry)))
1453		return ERR_PTR(rc);
1454	rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1455	free_UCSname(&key);
1456	if (rc == -ENOENT) {
1457		ip = NULL;
1458	} else if (rc) {
1459		jfs_err("jfs_lookup: dtSearch returned %d", rc);
1460		ip = ERR_PTR(rc);
1461	} else {
1462		ip = jfs_iget(dip->i_sb, inum);
1463		if (IS_ERR(ip))
1464			jfs_err("jfs_lookup: iget failed on inum %d", (uint)inum);
1465	}
1466
1467	return d_splice_alias(ip, dentry);
1468}
1469
1470static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1471		u64 ino, u32 generation)
1472{
1473	struct inode *inode;
1474
1475	if (ino == 0)
1476		return ERR_PTR(-ESTALE);
1477	inode = jfs_iget(sb, ino);
1478	if (IS_ERR(inode))
1479		return ERR_CAST(inode);
1480
1481	if (generation && inode->i_generation != generation) {
1482		iput(inode);
1483		return ERR_PTR(-ESTALE);
1484	}
1485
1486	return inode;
1487}
1488
1489struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1490		int fh_len, int fh_type)
1491{
1492	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1493				    jfs_nfs_get_inode);
1494}
1495
1496struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1497		int fh_len, int fh_type)
1498{
1499	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1500				    jfs_nfs_get_inode);
1501}
1502
1503struct dentry *jfs_get_parent(struct dentry *dentry)
1504{
1505	unsigned long parent_ino;
1506
1507	parent_ino =
1508		le32_to_cpu(JFS_IP(d_inode(dentry))->i_dtroot.header.idotdot);
1509
1510	return d_obtain_alias(jfs_iget(dentry->d_sb, parent_ino));
1511}
1512
1513const struct inode_operations jfs_dir_inode_operations = {
1514	.create		= jfs_create,
1515	.lookup		= jfs_lookup,
1516	.link		= jfs_link,
1517	.unlink		= jfs_unlink,
1518	.symlink	= jfs_symlink,
1519	.mkdir		= jfs_mkdir,
1520	.rmdir		= jfs_rmdir,
1521	.mknod		= jfs_mknod,
1522	.rename		= jfs_rename,
1523	.listxattr	= jfs_listxattr,
1524	.setattr	= jfs_setattr,
1525	.fileattr_get	= jfs_fileattr_get,
1526	.fileattr_set	= jfs_fileattr_set,
1527#ifdef CONFIG_JFS_POSIX_ACL
1528	.get_inode_acl	= jfs_get_acl,
1529	.set_acl	= jfs_set_acl,
1530#endif
1531};
1532
1533const struct file_operations jfs_dir_operations = {
1534	.read		= generic_read_dir,
1535	.iterate	= jfs_readdir,
1536	.fsync		= jfs_fsync,
1537	.unlocked_ioctl = jfs_ioctl,
1538	.compat_ioctl	= compat_ptr_ioctl,
 
 
1539	.llseek		= generic_file_llseek,
1540};
1541
1542static int jfs_ci_hash(const struct dentry *dir, struct qstr *this)
1543{
1544	unsigned long hash;
1545	int i;
1546
1547	hash = init_name_hash(dir);
1548	for (i=0; i < this->len; i++)
1549		hash = partial_name_hash(tolower(this->name[i]), hash);
1550	this->hash = end_name_hash(hash);
1551
1552	return 0;
1553}
1554
1555static int jfs_ci_compare(const struct dentry *dentry,
1556		unsigned int len, const char *str, const struct qstr *name)
1557{
1558	int i, result = 1;
1559
1560	if (len != name->len)
1561		goto out;
1562	for (i=0; i < len; i++) {
1563		if (tolower(str[i]) != tolower(name->name[i]))
1564			goto out;
1565	}
1566	result = 0;
1567out:
1568	return result;
1569}
1570
1571static int jfs_ci_revalidate(struct dentry *dentry, unsigned int flags)
1572{
1573	/*
1574	 * This is not negative dentry. Always valid.
1575	 *
1576	 * Note, rename() to existing directory entry will have ->d_inode,
1577	 * and will use existing name which isn't specified name by user.
1578	 *
1579	 * We may be able to drop this positive dentry here. But dropping
1580	 * positive dentry isn't good idea. So it's unsupported like
1581	 * rename("filename", "FILENAME") for now.
1582	 */
1583	if (d_really_is_positive(dentry))
1584		return 1;
1585
1586	/*
1587	 * This may be nfsd (or something), anyway, we can't see the
1588	 * intent of this. So, since this can be for creation, drop it.
1589	 */
1590	if (!flags)
1591		return 0;
1592
1593	/*
1594	 * Drop the negative dentry, in order to make sure to use the
1595	 * case sensitive name which is specified by user if this is
1596	 * for creation.
1597	 */
1598	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1599		return 0;
1600	return 1;
1601}
1602
1603const struct dentry_operations jfs_ci_dentry_operations =
1604{
1605	.d_hash = jfs_ci_hash,
1606	.d_compare = jfs_ci_compare,
1607	.d_revalidate = jfs_ci_revalidate,
1608};
v4.10.11
 
   1/*
   2 *   Copyright (C) International Business Machines Corp., 2000-2004
   3 *   Portions Copyright (C) Christoph Hellwig, 2001-2002
   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/fs.h>
  21#include <linux/namei.h>
  22#include <linux/ctype.h>
  23#include <linux/quotaops.h>
  24#include <linux/exportfs.h>
  25#include "jfs_incore.h"
  26#include "jfs_superblock.h"
  27#include "jfs_inode.h"
  28#include "jfs_dinode.h"
  29#include "jfs_dmap.h"
  30#include "jfs_unicode.h"
  31#include "jfs_metapage.h"
  32#include "jfs_xattr.h"
  33#include "jfs_acl.h"
  34#include "jfs_debug.h"
  35
  36/*
  37 * forward references
  38 */
  39const struct dentry_operations jfs_ci_dentry_operations;
  40
  41static s64 commitZeroLink(tid_t, struct inode *);
  42
  43/*
  44 * NAME:	free_ea_wmap(inode)
  45 *
  46 * FUNCTION:	free uncommitted extended attributes from working map
  47 *
  48 */
  49static inline void free_ea_wmap(struct inode *inode)
  50{
  51	dxd_t *ea = &JFS_IP(inode)->ea;
  52
  53	if (ea->flag & DXD_EXTENT) {
  54		/* free EA pages from cache */
  55		invalidate_dxd_metapages(inode, *ea);
  56		dbFree(inode, addressDXD(ea), lengthDXD(ea));
  57	}
  58	ea->flag = 0;
  59}
  60
  61/*
  62 * NAME:	jfs_create(dip, dentry, mode)
  63 *
  64 * FUNCTION:	create a regular file in the parent directory <dip>
  65 *		with name = <from dentry> and mode = <mode>
  66 *
  67 * PARAMETER:	dip	- parent directory vnode
  68 *		dentry	- dentry of new file
  69 *		mode	- create mode (rwxrwxrwx).
  70 *		nd- nd struct
  71 *
  72 * RETURN:	Errors from subroutines
  73 *
  74 */
  75static int jfs_create(struct inode *dip, struct dentry *dentry, umode_t mode,
  76		bool excl)
  77{
  78	int rc = 0;
  79	tid_t tid;		/* transaction id */
  80	struct inode *ip = NULL;	/* child directory inode */
  81	ino_t ino;
  82	struct component_name dname;	/* child directory name */
  83	struct btstack btstack;
  84	struct inode *iplist[2];
  85	struct tblock *tblk;
  86
  87	jfs_info("jfs_create: dip:0x%p name:%pd", dip, dentry);
  88
  89	rc = dquot_initialize(dip);
  90	if (rc)
  91		goto out1;
  92
  93	/*
  94	 * search parent directory for entry/freespace
  95	 * (dtSearch() returns parent directory page pinned)
  96	 */
  97	if ((rc = get_UCSname(&dname, dentry)))
  98		goto out1;
  99
 100	/*
 101	 * Either iAlloc() or txBegin() may block.  Deadlock can occur if we
 102	 * block there while holding dtree page, so we allocate the inode &
 103	 * begin the transaction before we search the directory.
 104	 */
 105	ip = ialloc(dip, mode);
 106	if (IS_ERR(ip)) {
 107		rc = PTR_ERR(ip);
 108		goto out2;
 109	}
 110
 111	tid = txBegin(dip->i_sb, 0);
 112
 113	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 114	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 115
 116	rc = jfs_init_acl(tid, ip, dip);
 117	if (rc)
 118		goto out3;
 119
 120	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
 121	if (rc) {
 122		txAbort(tid, 0);
 123		goto out3;
 124	}
 125
 126	if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
 127		jfs_err("jfs_create: dtSearch returned %d", rc);
 128		txAbort(tid, 0);
 129		goto out3;
 130	}
 131
 132	tblk = tid_to_tblock(tid);
 133	tblk->xflag |= COMMIT_CREATE;
 134	tblk->ino = ip->i_ino;
 135	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
 136
 137	iplist[0] = dip;
 138	iplist[1] = ip;
 139
 140	/*
 141	 * initialize the child XAD tree root in-line in inode
 142	 */
 143	xtInitRoot(tid, ip);
 144
 145	/*
 146	 * create entry in parent directory for child directory
 147	 * (dtInsert() releases parent directory page)
 148	 */
 149	ino = ip->i_ino;
 150	if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
 151		if (rc == -EIO) {
 152			jfs_err("jfs_create: dtInsert returned -EIO");
 153			txAbort(tid, 1);	/* Marks Filesystem dirty */
 154		} else
 155			txAbort(tid, 0);	/* Filesystem full */
 156		goto out3;
 157	}
 158
 159	ip->i_op = &jfs_file_inode_operations;
 160	ip->i_fop = &jfs_file_operations;
 161	ip->i_mapping->a_ops = &jfs_aops;
 162
 163	mark_inode_dirty(ip);
 164
 165	dip->i_ctime = dip->i_mtime = current_time(dip);
 166
 167	mark_inode_dirty(dip);
 168
 169	rc = txCommit(tid, 2, &iplist[0], 0);
 170
 171      out3:
 172	txEnd(tid);
 173	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 174	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 175	if (rc) {
 176		free_ea_wmap(ip);
 177		clear_nlink(ip);
 178		unlock_new_inode(ip);
 179		iput(ip);
 180	} else {
 181		unlock_new_inode(ip);
 182		d_instantiate(dentry, ip);
 183	}
 184
 185      out2:
 186	free_UCSname(&dname);
 187
 188      out1:
 189
 190	jfs_info("jfs_create: rc:%d", rc);
 191	return rc;
 192}
 193
 194
 195/*
 196 * NAME:	jfs_mkdir(dip, dentry, mode)
 197 *
 198 * FUNCTION:	create a child directory in the parent directory <dip>
 199 *		with name = <from dentry> and mode = <mode>
 200 *
 201 * PARAMETER:	dip	- parent directory vnode
 202 *		dentry	- dentry of child directory
 203 *		mode	- create mode (rwxrwxrwx).
 204 *
 205 * RETURN:	Errors from subroutines
 206 *
 207 * note:
 208 * EACCESS: user needs search+write permission on the parent directory
 209 */
 210static int jfs_mkdir(struct inode *dip, struct dentry *dentry, umode_t mode)
 
 211{
 212	int rc = 0;
 213	tid_t tid;		/* transaction id */
 214	struct inode *ip = NULL;	/* child directory inode */
 215	ino_t ino;
 216	struct component_name dname;	/* child directory name */
 217	struct btstack btstack;
 218	struct inode *iplist[2];
 219	struct tblock *tblk;
 220
 221	jfs_info("jfs_mkdir: dip:0x%p name:%pd", dip, dentry);
 222
 223	rc = dquot_initialize(dip);
 224	if (rc)
 225		goto out1;
 226
 227	/*
 228	 * search parent directory for entry/freespace
 229	 * (dtSearch() returns parent directory page pinned)
 230	 */
 231	if ((rc = get_UCSname(&dname, dentry)))
 232		goto out1;
 233
 234	/*
 235	 * Either iAlloc() or txBegin() may block.  Deadlock can occur if we
 236	 * block there while holding dtree page, so we allocate the inode &
 237	 * begin the transaction before we search the directory.
 238	 */
 239	ip = ialloc(dip, S_IFDIR | mode);
 240	if (IS_ERR(ip)) {
 241		rc = PTR_ERR(ip);
 242		goto out2;
 243	}
 244
 245	tid = txBegin(dip->i_sb, 0);
 246
 247	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 248	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 249
 250	rc = jfs_init_acl(tid, ip, dip);
 251	if (rc)
 252		goto out3;
 253
 254	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
 255	if (rc) {
 256		txAbort(tid, 0);
 257		goto out3;
 258	}
 259
 260	if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
 261		jfs_err("jfs_mkdir: dtSearch returned %d", rc);
 262		txAbort(tid, 0);
 263		goto out3;
 264	}
 265
 266	tblk = tid_to_tblock(tid);
 267	tblk->xflag |= COMMIT_CREATE;
 268	tblk->ino = ip->i_ino;
 269	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
 270
 271	iplist[0] = dip;
 272	iplist[1] = ip;
 273
 274	/*
 275	 * initialize the child directory in-line in inode
 276	 */
 277	dtInitRoot(tid, ip, dip->i_ino);
 278
 279	/*
 280	 * create entry in parent directory for child directory
 281	 * (dtInsert() releases parent directory page)
 282	 */
 283	ino = ip->i_ino;
 284	if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
 285		if (rc == -EIO) {
 286			jfs_err("jfs_mkdir: dtInsert returned -EIO");
 287			txAbort(tid, 1);	/* Marks Filesystem dirty */
 288		} else
 289			txAbort(tid, 0);	/* Filesystem full */
 290		goto out3;
 291	}
 292
 293	set_nlink(ip, 2);	/* for '.' */
 294	ip->i_op = &jfs_dir_inode_operations;
 295	ip->i_fop = &jfs_dir_operations;
 296
 297	mark_inode_dirty(ip);
 298
 299	/* update parent directory inode */
 300	inc_nlink(dip);		/* for '..' from child directory */
 301	dip->i_ctime = dip->i_mtime = current_time(dip);
 302	mark_inode_dirty(dip);
 303
 304	rc = txCommit(tid, 2, &iplist[0], 0);
 305
 306      out3:
 307	txEnd(tid);
 308	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 309	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 310	if (rc) {
 311		free_ea_wmap(ip);
 312		clear_nlink(ip);
 313		unlock_new_inode(ip);
 314		iput(ip);
 315	} else {
 316		unlock_new_inode(ip);
 317		d_instantiate(dentry, ip);
 318	}
 319
 320      out2:
 321	free_UCSname(&dname);
 322
 323
 324      out1:
 325
 326	jfs_info("jfs_mkdir: rc:%d", rc);
 327	return rc;
 328}
 329
 330/*
 331 * NAME:	jfs_rmdir(dip, dentry)
 332 *
 333 * FUNCTION:	remove a link to child directory
 334 *
 335 * PARAMETER:	dip	- parent inode
 336 *		dentry	- child directory dentry
 337 *
 338 * RETURN:	-EINVAL	- if name is . or ..
 339 *		-EINVAL - if . or .. exist but are invalid.
 340 *		errors from subroutines
 341 *
 342 * note:
 343 * if other threads have the directory open when the last link
 344 * is removed, the "." and ".." entries, if present, are removed before
 345 * rmdir() returns and no new entries may be created in the directory,
 346 * but the directory is not removed until the last reference to
 347 * the directory is released (cf.unlink() of regular file).
 348 */
 349static int jfs_rmdir(struct inode *dip, struct dentry *dentry)
 350{
 351	int rc;
 352	tid_t tid;		/* transaction id */
 353	struct inode *ip = d_inode(dentry);
 354	ino_t ino;
 355	struct component_name dname;
 356	struct inode *iplist[2];
 357	struct tblock *tblk;
 358
 359	jfs_info("jfs_rmdir: dip:0x%p name:%pd", dip, dentry);
 360
 361	/* Init inode for quota operations. */
 362	rc = dquot_initialize(dip);
 363	if (rc)
 364		goto out;
 365	rc = dquot_initialize(ip);
 366	if (rc)
 367		goto out;
 368
 369	/* directory must be empty to be removed */
 370	if (!dtEmpty(ip)) {
 371		rc = -ENOTEMPTY;
 372		goto out;
 373	}
 374
 375	if ((rc = get_UCSname(&dname, dentry))) {
 376		goto out;
 377	}
 378
 379	tid = txBegin(dip->i_sb, 0);
 380
 381	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 382	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 383
 384	iplist[0] = dip;
 385	iplist[1] = ip;
 386
 387	tblk = tid_to_tblock(tid);
 388	tblk->xflag |= COMMIT_DELETE;
 389	tblk->u.ip = ip;
 390
 391	/*
 392	 * delete the entry of target directory from parent directory
 393	 */
 394	ino = ip->i_ino;
 395	if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
 396		jfs_err("jfs_rmdir: dtDelete returned %d", rc);
 397		if (rc == -EIO)
 398			txAbort(tid, 1);
 399		txEnd(tid);
 400		mutex_unlock(&JFS_IP(ip)->commit_mutex);
 401		mutex_unlock(&JFS_IP(dip)->commit_mutex);
 402
 403		goto out2;
 404	}
 405
 406	/* update parent directory's link count corresponding
 407	 * to ".." entry of the target directory deleted
 408	 */
 409	dip->i_ctime = dip->i_mtime = current_time(dip);
 410	inode_dec_link_count(dip);
 411
 412	/*
 413	 * OS/2 could have created EA and/or ACL
 414	 */
 415	/* free EA from both persistent and working map */
 416	if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
 417		/* free EA pages */
 418		txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
 419	}
 420	JFS_IP(ip)->ea.flag = 0;
 421
 422	/* free ACL from both persistent and working map */
 423	if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
 424		/* free ACL pages */
 425		txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
 426	}
 427	JFS_IP(ip)->acl.flag = 0;
 428
 429	/* mark the target directory as deleted */
 430	clear_nlink(ip);
 431	mark_inode_dirty(ip);
 432
 433	rc = txCommit(tid, 2, &iplist[0], 0);
 434
 435	txEnd(tid);
 436
 437	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 438	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 439
 440	/*
 441	 * Truncating the directory index table is not guaranteed.  It
 442	 * may need to be done iteratively
 443	 */
 444	if (test_cflag(COMMIT_Stale, dip)) {
 445		if (dip->i_size > 1)
 446			jfs_truncate_nolock(dip, 0);
 447
 448		clear_cflag(COMMIT_Stale, dip);
 449	}
 450
 451      out2:
 452	free_UCSname(&dname);
 453
 454      out:
 455	jfs_info("jfs_rmdir: rc:%d", rc);
 456	return rc;
 457}
 458
 459/*
 460 * NAME:	jfs_unlink(dip, dentry)
 461 *
 462 * FUNCTION:	remove a link to object <vp> named by <name>
 463 *		from parent directory <dvp>
 464 *
 465 * PARAMETER:	dip	- inode of parent directory
 466 *		dentry	- dentry of object to be removed
 467 *
 468 * RETURN:	errors from subroutines
 469 *
 470 * note:
 471 * temporary file: if one or more processes have the file open
 472 * when the last link is removed, the link will be removed before
 473 * unlink() returns, but the removal of the file contents will be
 474 * postponed until all references to the files are closed.
 475 *
 476 * JFS does NOT support unlink() on directories.
 477 *
 478 */
 479static int jfs_unlink(struct inode *dip, struct dentry *dentry)
 480{
 481	int rc;
 482	tid_t tid;		/* transaction id */
 483	struct inode *ip = d_inode(dentry);
 484	ino_t ino;
 485	struct component_name dname;	/* object name */
 486	struct inode *iplist[2];
 487	struct tblock *tblk;
 488	s64 new_size = 0;
 489	int commit_flag;
 490
 491	jfs_info("jfs_unlink: dip:0x%p name:%pd", dip, dentry);
 492
 493	/* Init inode for quota operations. */
 494	rc = dquot_initialize(dip);
 495	if (rc)
 496		goto out;
 497	rc = dquot_initialize(ip);
 498	if (rc)
 499		goto out;
 500
 501	if ((rc = get_UCSname(&dname, dentry)))
 502		goto out;
 503
 504	IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
 505
 506	tid = txBegin(dip->i_sb, 0);
 507
 508	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 509	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 510
 511	iplist[0] = dip;
 512	iplist[1] = ip;
 513
 514	/*
 515	 * delete the entry of target file from parent directory
 516	 */
 517	ino = ip->i_ino;
 518	if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
 519		jfs_err("jfs_unlink: dtDelete returned %d", rc);
 520		if (rc == -EIO)
 521			txAbort(tid, 1);	/* Marks FS Dirty */
 522		txEnd(tid);
 523		mutex_unlock(&JFS_IP(ip)->commit_mutex);
 524		mutex_unlock(&JFS_IP(dip)->commit_mutex);
 525		IWRITE_UNLOCK(ip);
 526		goto out1;
 527	}
 528
 529	ASSERT(ip->i_nlink);
 530
 531	ip->i_ctime = dip->i_ctime = dip->i_mtime = current_time(ip);
 532	mark_inode_dirty(dip);
 533
 534	/* update target's inode */
 535	inode_dec_link_count(ip);
 536
 537	/*
 538	 *	commit zero link count object
 539	 */
 540	if (ip->i_nlink == 0) {
 541		assert(!test_cflag(COMMIT_Nolink, ip));
 542		/* free block resources */
 543		if ((new_size = commitZeroLink(tid, ip)) < 0) {
 544			txAbort(tid, 1);	/* Marks FS Dirty */
 545			txEnd(tid);
 546			mutex_unlock(&JFS_IP(ip)->commit_mutex);
 547			mutex_unlock(&JFS_IP(dip)->commit_mutex);
 548			IWRITE_UNLOCK(ip);
 549			rc = new_size;
 550			goto out1;
 551		}
 552		tblk = tid_to_tblock(tid);
 553		tblk->xflag |= COMMIT_DELETE;
 554		tblk->u.ip = ip;
 555	}
 556
 557	/*
 558	 * Incomplete truncate of file data can
 559	 * result in timing problems unless we synchronously commit the
 560	 * transaction.
 561	 */
 562	if (new_size)
 563		commit_flag = COMMIT_SYNC;
 564	else
 565		commit_flag = 0;
 566
 567	/*
 568	 * If xtTruncate was incomplete, commit synchronously to avoid
 569	 * timing complications
 570	 */
 571	rc = txCommit(tid, 2, &iplist[0], commit_flag);
 572
 573	txEnd(tid);
 574
 575	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 576	mutex_unlock(&JFS_IP(dip)->commit_mutex);
 577
 578	while (new_size && (rc == 0)) {
 579		tid = txBegin(dip->i_sb, 0);
 580		mutex_lock(&JFS_IP(ip)->commit_mutex);
 581		new_size = xtTruncate_pmap(tid, ip, new_size);
 582		if (new_size < 0) {
 583			txAbort(tid, 1);	/* Marks FS Dirty */
 584			rc = new_size;
 585		} else
 586			rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
 587		txEnd(tid);
 588		mutex_unlock(&JFS_IP(ip)->commit_mutex);
 589	}
 590
 591	if (ip->i_nlink == 0)
 592		set_cflag(COMMIT_Nolink, ip);
 593
 594	IWRITE_UNLOCK(ip);
 595
 596	/*
 597	 * Truncating the directory index table is not guaranteed.  It
 598	 * may need to be done iteratively
 599	 */
 600	if (test_cflag(COMMIT_Stale, dip)) {
 601		if (dip->i_size > 1)
 602			jfs_truncate_nolock(dip, 0);
 603
 604		clear_cflag(COMMIT_Stale, dip);
 605	}
 606
 607      out1:
 608	free_UCSname(&dname);
 609      out:
 610	jfs_info("jfs_unlink: rc:%d", rc);
 611	return rc;
 612}
 613
 614/*
 615 * NAME:	commitZeroLink()
 616 *
 617 * FUNCTION:	for non-directory, called by jfs_remove(),
 618 *		truncate a regular file, directory or symbolic
 619 *		link to zero length. return 0 if type is not
 620 *		one of these.
 621 *
 622 *		if the file is currently associated with a VM segment
 623 *		only permanent disk and inode map resources are freed,
 624 *		and neither the inode nor indirect blocks are modified
 625 *		so that the resources can be later freed in the work
 626 *		map by ctrunc1.
 627 *		if there is no VM segment on entry, the resources are
 628 *		freed in both work and permanent map.
 629 *		(? for temporary file - memory object is cached even
 630 *		after no reference:
 631 *		reference count > 0 -   )
 632 *
 633 * PARAMETERS:	cd	- pointer to commit data structure.
 634 *			  current inode is the one to truncate.
 635 *
 636 * RETURN:	Errors from subroutines
 637 */
 638static s64 commitZeroLink(tid_t tid, struct inode *ip)
 639{
 640	int filetype;
 641	struct tblock *tblk;
 642
 643	jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
 644
 645	filetype = ip->i_mode & S_IFMT;
 646	switch (filetype) {
 647	case S_IFREG:
 648		break;
 649	case S_IFLNK:
 650		/* fast symbolic link */
 651		if (ip->i_size < IDATASIZE) {
 652			ip->i_size = 0;
 653			return 0;
 654		}
 655		break;
 656	default:
 657		assert(filetype != S_IFDIR);
 658		return 0;
 659	}
 660
 661	set_cflag(COMMIT_Freewmap, ip);
 662
 663	/* mark transaction of block map update type */
 664	tblk = tid_to_tblock(tid);
 665	tblk->xflag |= COMMIT_PMAP;
 666
 667	/*
 668	 * free EA
 669	 */
 670	if (JFS_IP(ip)->ea.flag & DXD_EXTENT)
 671		/* acquire maplock on EA to be freed from block map */
 672		txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
 673
 674	/*
 675	 * free ACL
 676	 */
 677	if (JFS_IP(ip)->acl.flag & DXD_EXTENT)
 678		/* acquire maplock on EA to be freed from block map */
 679		txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
 680
 681	/*
 682	 * free xtree/data (truncate to zero length):
 683	 * free xtree/data pages from cache if COMMIT_PWMAP,
 684	 * free xtree/data blocks from persistent block map, and
 685	 * free xtree/data blocks from working block map if COMMIT_PWMAP;
 686	 */
 687	if (ip->i_size)
 688		return xtTruncate_pmap(tid, ip, 0);
 689
 690	return 0;
 691}
 692
 693
 694/*
 695 * NAME:	jfs_free_zero_link()
 696 *
 697 * FUNCTION:	for non-directory, called by iClose(),
 698 *		free resources of a file from cache and WORKING map
 699 *		for a file previously committed with zero link count
 700 *		while associated with a pager object,
 701 *
 702 * PARAMETER:	ip	- pointer to inode of file.
 703 */
 704void jfs_free_zero_link(struct inode *ip)
 705{
 706	int type;
 707
 708	jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
 709
 710	/* return if not reg or symbolic link or if size is
 711	 * already ok.
 712	 */
 713	type = ip->i_mode & S_IFMT;
 714
 715	switch (type) {
 716	case S_IFREG:
 717		break;
 718	case S_IFLNK:
 719		/* if its contained in inode nothing to do */
 720		if (ip->i_size < IDATASIZE)
 721			return;
 722		break;
 723	default:
 724		return;
 725	}
 726
 727	/*
 728	 * free EA
 729	 */
 730	if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
 731		s64 xaddr = addressDXD(&JFS_IP(ip)->ea);
 732		int xlen = lengthDXD(&JFS_IP(ip)->ea);
 733		struct maplock maplock;	/* maplock for COMMIT_WMAP */
 734		struct pxd_lock *pxdlock;	/* maplock for COMMIT_WMAP */
 735
 736		/* free EA pages from cache */
 737		invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
 738
 739		/* free EA extent from working block map */
 740		maplock.index = 1;
 741		pxdlock = (struct pxd_lock *) & maplock;
 742		pxdlock->flag = mlckFREEPXD;
 743		PXDaddress(&pxdlock->pxd, xaddr);
 744		PXDlength(&pxdlock->pxd, xlen);
 745		txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
 746	}
 747
 748	/*
 749	 * free ACL
 750	 */
 751	if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
 752		s64 xaddr = addressDXD(&JFS_IP(ip)->acl);
 753		int xlen = lengthDXD(&JFS_IP(ip)->acl);
 754		struct maplock maplock;	/* maplock for COMMIT_WMAP */
 755		struct pxd_lock *pxdlock;	/* maplock for COMMIT_WMAP */
 756
 757		invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
 758
 759		/* free ACL extent from working block map */
 760		maplock.index = 1;
 761		pxdlock = (struct pxd_lock *) & maplock;
 762		pxdlock->flag = mlckFREEPXD;
 763		PXDaddress(&pxdlock->pxd, xaddr);
 764		PXDlength(&pxdlock->pxd, xlen);
 765		txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
 766	}
 767
 768	/*
 769	 * free xtree/data (truncate to zero length):
 770	 * free xtree/data pages from cache, and
 771	 * free xtree/data blocks from working block map;
 772	 */
 773	if (ip->i_size)
 774		xtTruncate(0, ip, 0, COMMIT_WMAP);
 775}
 776
 777/*
 778 * NAME:	jfs_link(vp, dvp, name, crp)
 779 *
 780 * FUNCTION:	create a link to <vp> by the name = <name>
 781 *		in the parent directory <dvp>
 782 *
 783 * PARAMETER:	vp	- target object
 784 *		dvp	- parent directory of new link
 785 *		name	- name of new link to target object
 786 *		crp	- credential
 787 *
 788 * RETURN:	Errors from subroutines
 789 *
 790 * note:
 791 * JFS does NOT support link() on directories (to prevent circular
 792 * path in the directory hierarchy);
 793 * EPERM: the target object is a directory, and either the caller
 794 * does not have appropriate privileges or the implementation prohibits
 795 * using link() on directories [XPG4.2].
 796 *
 797 * JFS does NOT support links between file systems:
 798 * EXDEV: target object and new link are on different file systems and
 799 * implementation does not support links between file systems [XPG4.2].
 800 */
 801static int jfs_link(struct dentry *old_dentry,
 802	     struct inode *dir, struct dentry *dentry)
 803{
 804	int rc;
 805	tid_t tid;
 806	struct inode *ip = d_inode(old_dentry);
 807	ino_t ino;
 808	struct component_name dname;
 809	struct btstack btstack;
 810	struct inode *iplist[2];
 811
 812	jfs_info("jfs_link: %pd %pd", old_dentry, dentry);
 813
 814	rc = dquot_initialize(dir);
 815	if (rc)
 816		goto out;
 817
 818	tid = txBegin(ip->i_sb, 0);
 819
 820	mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
 821	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 822
 823	/*
 824	 * scan parent directory for entry/freespace
 825	 */
 826	if ((rc = get_UCSname(&dname, dentry)))
 827		goto out_tx;
 828
 829	if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
 830		goto free_dname;
 831
 832	/*
 833	 * create entry for new link in parent directory
 834	 */
 835	ino = ip->i_ino;
 836	if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
 837		goto free_dname;
 838
 839	/* update object inode */
 840	inc_nlink(ip);		/* for new link */
 841	ip->i_ctime = current_time(ip);
 842	dir->i_ctime = dir->i_mtime = current_time(dir);
 843	mark_inode_dirty(dir);
 844	ihold(ip);
 845
 846	iplist[0] = ip;
 847	iplist[1] = dir;
 848	rc = txCommit(tid, 2, &iplist[0], 0);
 849
 850	if (rc) {
 851		drop_nlink(ip); /* never instantiated */
 852		iput(ip);
 853	} else
 854		d_instantiate(dentry, ip);
 855
 856      free_dname:
 857	free_UCSname(&dname);
 858
 859      out_tx:
 860	txEnd(tid);
 861
 862	mutex_unlock(&JFS_IP(ip)->commit_mutex);
 863	mutex_unlock(&JFS_IP(dir)->commit_mutex);
 864
 865      out:
 866	jfs_info("jfs_link: rc:%d", rc);
 867	return rc;
 868}
 869
 870/*
 871 * NAME:	jfs_symlink(dip, dentry, name)
 872 *
 873 * FUNCTION:	creates a symbolic link to <symlink> by name <name>
 874 *			in directory <dip>
 875 *
 876 * PARAMETER:	dip	- parent directory vnode
 877 *		dentry	- dentry of symbolic link
 878 *		name	- the path name of the existing object
 879 *			  that will be the source of the link
 880 *
 881 * RETURN:	errors from subroutines
 882 *
 883 * note:
 884 * ENAMETOOLONG: pathname resolution of a symbolic link produced
 885 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
 886*/
 887
 888static int jfs_symlink(struct inode *dip, struct dentry *dentry,
 889		const char *name)
 890{
 891	int rc;
 892	tid_t tid;
 893	ino_t ino = 0;
 894	struct component_name dname;
 895	int ssize;		/* source pathname size */
 896	struct btstack btstack;
 897	struct inode *ip = d_inode(dentry);
 898	s64 xlen = 0;
 899	int bmask = 0, xsize;
 900	s64 xaddr;
 901	struct metapage *mp;
 902	struct super_block *sb;
 903	struct tblock *tblk;
 904
 905	struct inode *iplist[2];
 906
 907	jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
 908
 909	rc = dquot_initialize(dip);
 910	if (rc)
 911		goto out1;
 912
 913	ssize = strlen(name) + 1;
 914
 915	/*
 916	 * search parent directory for entry/freespace
 917	 * (dtSearch() returns parent directory page pinned)
 918	 */
 919
 920	if ((rc = get_UCSname(&dname, dentry)))
 921		goto out1;
 922
 923	/*
 924	 * allocate on-disk/in-memory inode for symbolic link:
 925	 * (iAlloc() returns new, locked inode)
 926	 */
 927	ip = ialloc(dip, S_IFLNK | 0777);
 928	if (IS_ERR(ip)) {
 929		rc = PTR_ERR(ip);
 930		goto out2;
 931	}
 932
 933	tid = txBegin(dip->i_sb, 0);
 934
 935	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
 936	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
 937
 938	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
 939	if (rc)
 940		goto out3;
 941
 942	tblk = tid_to_tblock(tid);
 943	tblk->xflag |= COMMIT_CREATE;
 944	tblk->ino = ip->i_ino;
 945	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
 946
 947	/* fix symlink access permission
 948	 * (dir_create() ANDs in the u.u_cmask,
 949	 * but symlinks really need to be 777 access)
 950	 */
 951	ip->i_mode |= 0777;
 952
 953	/*
 954	 * write symbolic link target path name
 955	 */
 956	xtInitRoot(tid, ip);
 957
 958	/*
 959	 * write source path name inline in on-disk inode (fast symbolic link)
 960	 */
 961
 962	if (ssize <= IDATASIZE) {
 963		ip->i_op = &jfs_fast_symlink_inode_operations;
 964
 965		ip->i_link = JFS_IP(ip)->i_inline;
 966		memcpy(ip->i_link, name, ssize);
 967		ip->i_size = ssize - 1;
 968
 969		/*
 970		 * if symlink is > 128 bytes, we don't have the space to
 971		 * store inline extended attributes
 972		 */
 973		if (ssize > sizeof (JFS_IP(ip)->i_inline))
 974			JFS_IP(ip)->mode2 &= ~INLINEEA;
 975
 976		jfs_info("jfs_symlink: fast symlink added  ssize:%d name:%s ",
 977			 ssize, name);
 978	}
 979	/*
 980	 * write source path name in a single extent
 981	 */
 982	else {
 983		jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
 984
 985		ip->i_op = &jfs_symlink_inode_operations;
 986		inode_nohighmem(ip);
 987		ip->i_mapping->a_ops = &jfs_aops;
 988
 989		/*
 990		 * even though the data of symlink object (source
 991		 * path name) is treated as non-journaled user data,
 992		 * it is read/written thru buffer cache for performance.
 993		 */
 994		sb = ip->i_sb;
 995		bmask = JFS_SBI(sb)->bsize - 1;
 996		xsize = (ssize + bmask) & ~bmask;
 997		xaddr = 0;
 998		xlen = xsize >> JFS_SBI(sb)->l2bsize;
 999		if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
1000			txAbort(tid, 0);
1001			goto out3;
1002		}
1003		ip->i_size = ssize - 1;
1004		while (ssize) {
1005			/* This is kind of silly since PATH_MAX == 4K */
1006			int copy_size = min(ssize, PSIZE);
1007
1008			mp = get_metapage(ip, xaddr, PSIZE, 1);
1009
1010			if (mp == NULL) {
1011				xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1012				rc = -EIO;
1013				txAbort(tid, 0);
1014				goto out3;
1015			}
1016			memcpy(mp->data, name, copy_size);
1017			flush_metapage(mp);
1018			ssize -= copy_size;
1019			name += copy_size;
1020			xaddr += JFS_SBI(sb)->nbperpage;
1021		}
1022	}
1023
1024	/*
1025	 * create entry for symbolic link in parent directory
1026	 */
1027	rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1028	if (rc == 0) {
1029		ino = ip->i_ino;
1030		rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1031	}
1032	if (rc) {
1033		if (xlen)
1034			xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1035		txAbort(tid, 0);
1036		/* discard new inode */
1037		goto out3;
1038	}
1039
1040	mark_inode_dirty(ip);
1041
1042	dip->i_ctime = dip->i_mtime = current_time(dip);
1043	mark_inode_dirty(dip);
1044	/*
1045	 * commit update of parent directory and link object
1046	 */
1047
1048	iplist[0] = dip;
1049	iplist[1] = ip;
1050	rc = txCommit(tid, 2, &iplist[0], 0);
1051
1052      out3:
1053	txEnd(tid);
1054	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1055	mutex_unlock(&JFS_IP(dip)->commit_mutex);
1056	if (rc) {
1057		free_ea_wmap(ip);
1058		clear_nlink(ip);
1059		unlock_new_inode(ip);
1060		iput(ip);
1061	} else {
1062		unlock_new_inode(ip);
1063		d_instantiate(dentry, ip);
1064	}
1065
1066      out2:
1067	free_UCSname(&dname);
1068
1069      out1:
1070	jfs_info("jfs_symlink: rc:%d", rc);
1071	return rc;
1072}
1073
1074
1075/*
1076 * NAME:	jfs_rename
1077 *
1078 * FUNCTION:	rename a file or directory
1079 */
1080static int jfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1081		      struct inode *new_dir, struct dentry *new_dentry,
1082		      unsigned int flags)
1083{
1084	struct btstack btstack;
1085	ino_t ino;
1086	struct component_name new_dname;
1087	struct inode *new_ip;
1088	struct component_name old_dname;
1089	struct inode *old_ip;
1090	int rc;
1091	tid_t tid;
1092	struct tlock *tlck;
1093	struct dt_lock *dtlck;
1094	struct lv *lv;
1095	int ipcount;
1096	struct inode *iplist[4];
1097	struct tblock *tblk;
1098	s64 new_size = 0;
1099	int commit_flag;
1100
1101	if (flags & ~RENAME_NOREPLACE)
1102		return -EINVAL;
1103
1104	jfs_info("jfs_rename: %pd %pd", old_dentry, new_dentry);
1105
1106	rc = dquot_initialize(old_dir);
1107	if (rc)
1108		goto out1;
1109	rc = dquot_initialize(new_dir);
1110	if (rc)
1111		goto out1;
1112
1113	old_ip = d_inode(old_dentry);
1114	new_ip = d_inode(new_dentry);
1115
1116	if ((rc = get_UCSname(&old_dname, old_dentry)))
1117		goto out1;
1118
1119	if ((rc = get_UCSname(&new_dname, new_dentry)))
1120		goto out2;
1121
1122	/*
1123	 * Make sure source inode number is what we think it is
1124	 */
1125	rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1126	if (rc || (ino != old_ip->i_ino)) {
1127		rc = -ENOENT;
1128		goto out3;
1129	}
1130
1131	/*
1132	 * Make sure dest inode number (if any) is what we think it is
1133	 */
1134	rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1135	if (!rc) {
1136		if ((!new_ip) || (ino != new_ip->i_ino)) {
1137			rc = -ESTALE;
1138			goto out3;
1139		}
1140	} else if (rc != -ENOENT)
1141		goto out3;
1142	else if (new_ip) {
1143		/* no entry exists, but one was expected */
1144		rc = -ESTALE;
1145		goto out3;
1146	}
1147
1148	if (S_ISDIR(old_ip->i_mode)) {
1149		if (new_ip) {
1150			if (!dtEmpty(new_ip)) {
1151				rc = -ENOTEMPTY;
1152				goto out3;
1153			}
1154		}
1155	} else if (new_ip) {
1156		IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1157		/* Init inode for quota operations. */
1158		rc = dquot_initialize(new_ip);
1159		if (rc)
1160			goto out_unlock;
1161	}
1162
1163	/*
1164	 * The real work starts here
1165	 */
1166	tid = txBegin(new_dir->i_sb, 0);
1167
1168	/*
1169	 * How do we know the locking is safe from deadlocks?
1170	 * The vfs does the hard part for us.  Any time we are taking nested
1171	 * commit_mutexes, the vfs already has i_mutex held on the parent.
1172	 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1173	 */
1174	mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1175	mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1176	if (old_dir != new_dir)
1177		mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
1178				  COMMIT_MUTEX_SECOND_PARENT);
1179
1180	if (new_ip) {
1181		mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1182				  COMMIT_MUTEX_VICTIM);
1183		/*
1184		 * Change existing directory entry to new inode number
1185		 */
1186		ino = new_ip->i_ino;
1187		rc = dtModify(tid, new_dir, &new_dname, &ino,
1188			      old_ip->i_ino, JFS_RENAME);
1189		if (rc)
1190			goto out_tx;
1191		drop_nlink(new_ip);
1192		if (S_ISDIR(new_ip->i_mode)) {
1193			drop_nlink(new_ip);
1194			if (new_ip->i_nlink) {
1195				mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1196				if (old_dir != new_dir)
1197					mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1198				mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1199				mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1200				if (!S_ISDIR(old_ip->i_mode) && new_ip)
1201					IWRITE_UNLOCK(new_ip);
1202				jfs_error(new_ip->i_sb,
1203					  "new_ip->i_nlink != 0\n");
1204				return -EIO;
1205			}
1206			tblk = tid_to_tblock(tid);
1207			tblk->xflag |= COMMIT_DELETE;
1208			tblk->u.ip = new_ip;
1209		} else if (new_ip->i_nlink == 0) {
1210			assert(!test_cflag(COMMIT_Nolink, new_ip));
1211			/* free block resources */
1212			if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
1213				txAbort(tid, 1);	/* Marks FS Dirty */
1214				rc = new_size;
1215				goto out_tx;
1216			}
1217			tblk = tid_to_tblock(tid);
1218			tblk->xflag |= COMMIT_DELETE;
1219			tblk->u.ip = new_ip;
1220		} else {
1221			new_ip->i_ctime = current_time(new_ip);
1222			mark_inode_dirty(new_ip);
1223		}
1224	} else {
1225		/*
1226		 * Add new directory entry
1227		 */
1228		rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1229			      JFS_CREATE);
1230		if (rc) {
1231			jfs_err("jfs_rename didn't expect dtSearch to fail w/rc = %d",
1232				rc);
1233			goto out_tx;
1234		}
1235
1236		ino = old_ip->i_ino;
1237		rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1238		if (rc) {
1239			if (rc == -EIO)
1240				jfs_err("jfs_rename: dtInsert returned -EIO");
1241			goto out_tx;
1242		}
1243		if (S_ISDIR(old_ip->i_mode))
1244			inc_nlink(new_dir);
1245	}
1246	/*
1247	 * Remove old directory entry
1248	 */
1249
1250	ino = old_ip->i_ino;
1251	rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1252	if (rc) {
1253		jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1254			rc);
1255		txAbort(tid, 1);	/* Marks Filesystem dirty */
1256		goto out_tx;
1257	}
1258	if (S_ISDIR(old_ip->i_mode)) {
1259		drop_nlink(old_dir);
1260		if (old_dir != new_dir) {
1261			/*
1262			 * Change inode number of parent for moved directory
1263			 */
1264
1265			JFS_IP(old_ip)->i_dtroot.header.idotdot =
1266				cpu_to_le32(new_dir->i_ino);
1267
1268			/* Linelock header of dtree */
1269			tlck = txLock(tid, old_ip,
1270				    (struct metapage *) &JFS_IP(old_ip)->bxflag,
1271				      tlckDTREE | tlckBTROOT | tlckRELINK);
1272			dtlck = (struct dt_lock *) & tlck->lock;
1273			ASSERT(dtlck->index == 0);
1274			lv = & dtlck->lv[0];
1275			lv->offset = 0;
1276			lv->length = 1;
1277			dtlck->index++;
1278		}
1279	}
1280
1281	/*
1282	 * Update ctime on changed/moved inodes & mark dirty
1283	 */
1284	old_ip->i_ctime = current_time(old_ip);
1285	mark_inode_dirty(old_ip);
1286
1287	new_dir->i_ctime = new_dir->i_mtime = current_time(new_dir);
1288	mark_inode_dirty(new_dir);
1289
1290	/* Build list of inodes modified by this transaction */
1291	ipcount = 0;
1292	iplist[ipcount++] = old_ip;
1293	if (new_ip)
1294		iplist[ipcount++] = new_ip;
1295	iplist[ipcount++] = old_dir;
1296
1297	if (old_dir != new_dir) {
1298		iplist[ipcount++] = new_dir;
1299		old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1300		mark_inode_dirty(old_dir);
1301	}
1302
1303	/*
1304	 * Incomplete truncate of file data can
1305	 * result in timing problems unless we synchronously commit the
1306	 * transaction.
1307	 */
1308	if (new_size)
1309		commit_flag = COMMIT_SYNC;
1310	else
1311		commit_flag = 0;
1312
1313	rc = txCommit(tid, ipcount, iplist, commit_flag);
1314
1315      out_tx:
1316	txEnd(tid);
1317	if (new_ip)
1318		mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1319	if (old_dir != new_dir)
1320		mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1321	mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1322	mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1323
1324	while (new_size && (rc == 0)) {
1325		tid = txBegin(new_ip->i_sb, 0);
1326		mutex_lock(&JFS_IP(new_ip)->commit_mutex);
1327		new_size = xtTruncate_pmap(tid, new_ip, new_size);
1328		if (new_size < 0) {
1329			txAbort(tid, 1);
1330			rc = new_size;
1331		} else
1332			rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1333		txEnd(tid);
1334		mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1335	}
1336	if (new_ip && (new_ip->i_nlink == 0))
1337		set_cflag(COMMIT_Nolink, new_ip);
1338	/*
1339	 * Truncating the directory index table is not guaranteed.  It
1340	 * may need to be done iteratively
1341	 */
1342	if (test_cflag(COMMIT_Stale, old_dir)) {
1343		if (old_dir->i_size > 1)
1344			jfs_truncate_nolock(old_dir, 0);
1345
1346		clear_cflag(COMMIT_Stale, old_dir);
1347	}
1348      out_unlock:
1349	if (new_ip && !S_ISDIR(new_ip->i_mode))
1350		IWRITE_UNLOCK(new_ip);
1351      out3:
1352	free_UCSname(&new_dname);
1353      out2:
1354	free_UCSname(&old_dname);
1355      out1:
1356	jfs_info("jfs_rename: returning %d", rc);
1357	return rc;
1358}
1359
1360
1361/*
1362 * NAME:	jfs_mknod
1363 *
1364 * FUNCTION:	Create a special file (device)
1365 */
1366static int jfs_mknod(struct inode *dir, struct dentry *dentry,
1367		umode_t mode, dev_t rdev)
1368{
1369	struct jfs_inode_info *jfs_ip;
1370	struct btstack btstack;
1371	struct component_name dname;
1372	ino_t ino;
1373	struct inode *ip;
1374	struct inode *iplist[2];
1375	int rc;
1376	tid_t tid;
1377	struct tblock *tblk;
1378
1379	jfs_info("jfs_mknod: %pd", dentry);
1380
1381	rc = dquot_initialize(dir);
1382	if (rc)
1383		goto out;
1384
1385	if ((rc = get_UCSname(&dname, dentry)))
1386		goto out;
1387
1388	ip = ialloc(dir, mode);
1389	if (IS_ERR(ip)) {
1390		rc = PTR_ERR(ip);
1391		goto out1;
1392	}
1393	jfs_ip = JFS_IP(ip);
1394
1395	tid = txBegin(dir->i_sb, 0);
1396
1397	mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1398	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1399
1400	rc = jfs_init_acl(tid, ip, dir);
1401	if (rc)
1402		goto out3;
1403
1404	rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1405	if (rc) {
1406		txAbort(tid, 0);
1407		goto out3;
1408	}
1409
1410	if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1411		txAbort(tid, 0);
1412		goto out3;
1413	}
1414
1415	tblk = tid_to_tblock(tid);
1416	tblk->xflag |= COMMIT_CREATE;
1417	tblk->ino = ip->i_ino;
1418	tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
1419
1420	ino = ip->i_ino;
1421	if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1422		txAbort(tid, 0);
1423		goto out3;
1424	}
1425
1426	ip->i_op = &jfs_file_inode_operations;
1427	jfs_ip->dev = new_encode_dev(rdev);
1428	init_special_inode(ip, ip->i_mode, rdev);
1429
1430	mark_inode_dirty(ip);
1431
1432	dir->i_ctime = dir->i_mtime = current_time(dir);
1433
1434	mark_inode_dirty(dir);
1435
1436	iplist[0] = dir;
1437	iplist[1] = ip;
1438	rc = txCommit(tid, 2, iplist, 0);
1439
1440      out3:
1441	txEnd(tid);
1442	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1443	mutex_unlock(&JFS_IP(dir)->commit_mutex);
1444	if (rc) {
1445		free_ea_wmap(ip);
1446		clear_nlink(ip);
1447		unlock_new_inode(ip);
1448		iput(ip);
1449	} else {
1450		unlock_new_inode(ip);
1451		d_instantiate(dentry, ip);
1452	}
1453
1454      out1:
1455	free_UCSname(&dname);
1456
1457      out:
1458	jfs_info("jfs_mknod: returning %d", rc);
1459	return rc;
1460}
1461
1462static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags)
1463{
1464	struct btstack btstack;
1465	ino_t inum;
1466	struct inode *ip;
1467	struct component_name key;
1468	int rc;
1469
1470	jfs_info("jfs_lookup: name = %pd", dentry);
1471
1472	if ((rc = get_UCSname(&key, dentry)))
1473		return ERR_PTR(rc);
1474	rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1475	free_UCSname(&key);
1476	if (rc == -ENOENT) {
1477		ip = NULL;
1478	} else if (rc) {
1479		jfs_err("jfs_lookup: dtSearch returned %d", rc);
1480		ip = ERR_PTR(rc);
1481	} else {
1482		ip = jfs_iget(dip->i_sb, inum);
1483		if (IS_ERR(ip))
1484			jfs_err("jfs_lookup: iget failed on inum %d", (uint)inum);
1485	}
1486
1487	return d_splice_alias(ip, dentry);
1488}
1489
1490static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1491		u64 ino, u32 generation)
1492{
1493	struct inode *inode;
1494
1495	if (ino == 0)
1496		return ERR_PTR(-ESTALE);
1497	inode = jfs_iget(sb, ino);
1498	if (IS_ERR(inode))
1499		return ERR_CAST(inode);
1500
1501	if (generation && inode->i_generation != generation) {
1502		iput(inode);
1503		return ERR_PTR(-ESTALE);
1504	}
1505
1506	return inode;
1507}
1508
1509struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1510		int fh_len, int fh_type)
1511{
1512	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1513				    jfs_nfs_get_inode);
1514}
1515
1516struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1517		int fh_len, int fh_type)
1518{
1519	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1520				    jfs_nfs_get_inode);
1521}
1522
1523struct dentry *jfs_get_parent(struct dentry *dentry)
1524{
1525	unsigned long parent_ino;
1526
1527	parent_ino =
1528		le32_to_cpu(JFS_IP(d_inode(dentry))->i_dtroot.header.idotdot);
1529
1530	return d_obtain_alias(jfs_iget(dentry->d_sb, parent_ino));
1531}
1532
1533const struct inode_operations jfs_dir_inode_operations = {
1534	.create		= jfs_create,
1535	.lookup		= jfs_lookup,
1536	.link		= jfs_link,
1537	.unlink		= jfs_unlink,
1538	.symlink	= jfs_symlink,
1539	.mkdir		= jfs_mkdir,
1540	.rmdir		= jfs_rmdir,
1541	.mknod		= jfs_mknod,
1542	.rename		= jfs_rename,
1543	.listxattr	= jfs_listxattr,
1544	.setattr	= jfs_setattr,
 
 
1545#ifdef CONFIG_JFS_POSIX_ACL
1546	.get_acl	= jfs_get_acl,
1547	.set_acl	= jfs_set_acl,
1548#endif
1549};
1550
1551const struct file_operations jfs_dir_operations = {
1552	.read		= generic_read_dir,
1553	.iterate	= jfs_readdir,
1554	.fsync		= jfs_fsync,
1555	.unlocked_ioctl = jfs_ioctl,
1556#ifdef CONFIG_COMPAT
1557	.compat_ioctl	= jfs_compat_ioctl,
1558#endif
1559	.llseek		= generic_file_llseek,
1560};
1561
1562static int jfs_ci_hash(const struct dentry *dir, struct qstr *this)
1563{
1564	unsigned long hash;
1565	int i;
1566
1567	hash = init_name_hash(dir);
1568	for (i=0; i < this->len; i++)
1569		hash = partial_name_hash(tolower(this->name[i]), hash);
1570	this->hash = end_name_hash(hash);
1571
1572	return 0;
1573}
1574
1575static int jfs_ci_compare(const struct dentry *dentry,
1576		unsigned int len, const char *str, const struct qstr *name)
1577{
1578	int i, result = 1;
1579
1580	if (len != name->len)
1581		goto out;
1582	for (i=0; i < len; i++) {
1583		if (tolower(str[i]) != tolower(name->name[i]))
1584			goto out;
1585	}
1586	result = 0;
1587out:
1588	return result;
1589}
1590
1591static int jfs_ci_revalidate(struct dentry *dentry, unsigned int flags)
1592{
1593	/*
1594	 * This is not negative dentry. Always valid.
1595	 *
1596	 * Note, rename() to existing directory entry will have ->d_inode,
1597	 * and will use existing name which isn't specified name by user.
1598	 *
1599	 * We may be able to drop this positive dentry here. But dropping
1600	 * positive dentry isn't good idea. So it's unsupported like
1601	 * rename("filename", "FILENAME") for now.
1602	 */
1603	if (d_really_is_positive(dentry))
1604		return 1;
1605
1606	/*
1607	 * This may be nfsd (or something), anyway, we can't see the
1608	 * intent of this. So, since this can be for creation, drop it.
1609	 */
1610	if (!flags)
1611		return 0;
1612
1613	/*
1614	 * Drop the negative dentry, in order to make sure to use the
1615	 * case sensitive name which is specified by user if this is
1616	 * for creation.
1617	 */
1618	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1619		return 0;
1620	return 1;
1621}
1622
1623const struct dentry_operations jfs_ci_dentry_operations =
1624{
1625	.d_hash = jfs_ci_hash,
1626	.d_compare = jfs_ci_compare,
1627	.d_revalidate = jfs_ci_revalidate,
1628};