Linux Audio

Check our new training course

Loading...
v3.1
   1/*
   2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_types.h"
  21#include "xfs_bit.h"
  22#include "xfs_log.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_sb.h"
  26#include "xfs_ag.h"
  27#include "xfs_mount.h"
  28#include "xfs_trans_priv.h"
  29#include "xfs_bmap_btree.h"
  30#include "xfs_dinode.h"
  31#include "xfs_inode.h"
 
  32#include "xfs_inode_item.h"
  33#include "xfs_error.h"
  34#include "xfs_trace.h"
 
 
  35
  36
  37kmem_zone_t	*xfs_ili_zone;		/* inode log item zone */
  38
  39static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
  40{
  41	return container_of(lip, struct xfs_inode_log_item, ili_item);
  42}
  43
  44
  45/*
  46 * This returns the number of iovecs needed to log the given inode item.
  47 *
  48 * We need one iovec for the inode log format structure, one for the
  49 * inode core, and possibly one for the inode data/extents/b-tree root
  50 * and one for the inode attribute data/extents/b-tree root.
  51 */
  52STATIC uint
  53xfs_inode_item_size(
  54	struct xfs_log_item	*lip)
  55{
  56	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
  57	struct xfs_inode	*ip = iip->ili_inode;
  58	uint			nvecs = 2;
  59
  60	/*
  61	 * Only log the data/extents/b-tree root if there is something
  62	 * left to log.
  63	 */
  64	iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
  65
  66	switch (ip->i_d.di_format) {
  67	case XFS_DINODE_FMT_EXTENTS:
  68		iip->ili_format.ilf_fields &=
  69			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
  70			  XFS_ILOG_DEV | XFS_ILOG_UUID);
  71		if ((iip->ili_format.ilf_fields & XFS_ILOG_DEXT) &&
  72		    (ip->i_d.di_nextents > 0) &&
  73		    (ip->i_df.if_bytes > 0)) {
  74			ASSERT(ip->i_df.if_u1.if_extents != NULL);
  75			nvecs++;
  76		} else {
  77			iip->ili_format.ilf_fields &= ~XFS_ILOG_DEXT;
  78		}
  79		break;
  80
  81	case XFS_DINODE_FMT_BTREE:
  82		ASSERT(ip->i_df.if_ext_max ==
  83		       XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
  84		iip->ili_format.ilf_fields &=
  85			~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
  86			  XFS_ILOG_DEV | XFS_ILOG_UUID);
  87		if ((iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) &&
  88		    (ip->i_df.if_broot_bytes > 0)) {
  89			ASSERT(ip->i_df.if_broot != NULL);
  90			nvecs++;
  91		} else {
  92			ASSERT(!(iip->ili_format.ilf_fields &
  93				 XFS_ILOG_DBROOT));
  94#ifdef XFS_TRANS_DEBUG
  95			if (iip->ili_root_size > 0) {
  96				ASSERT(iip->ili_root_size ==
  97				       ip->i_df.if_broot_bytes);
  98				ASSERT(memcmp(iip->ili_orig_root,
  99					    ip->i_df.if_broot,
 100					    iip->ili_root_size) == 0);
 101			} else {
 102				ASSERT(ip->i_df.if_broot_bytes == 0);
 103			}
 104#endif
 105			iip->ili_format.ilf_fields &= ~XFS_ILOG_DBROOT;
 106		}
 107		break;
 108
 109	case XFS_DINODE_FMT_LOCAL:
 110		iip->ili_format.ilf_fields &=
 111			~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
 112			  XFS_ILOG_DEV | XFS_ILOG_UUID);
 113		if ((iip->ili_format.ilf_fields & XFS_ILOG_DDATA) &&
 114		    (ip->i_df.if_bytes > 0)) {
 115			ASSERT(ip->i_df.if_u1.if_data != NULL);
 116			ASSERT(ip->i_d.di_size > 0);
 117			nvecs++;
 118		} else {
 119			iip->ili_format.ilf_fields &= ~XFS_ILOG_DDATA;
 120		}
 121		break;
 122
 123	case XFS_DINODE_FMT_DEV:
 124		iip->ili_format.ilf_fields &=
 125			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 126			  XFS_ILOG_DEXT | XFS_ILOG_UUID);
 127		break;
 128
 129	case XFS_DINODE_FMT_UUID:
 130		iip->ili_format.ilf_fields &=
 131			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 132			  XFS_ILOG_DEXT | XFS_ILOG_DEV);
 133		break;
 134
 135	default:
 136		ASSERT(0);
 137		break;
 138	}
 
 139
 140	/*
 141	 * If there are no attributes associated with this file,
 142	 * then there cannot be anything more to log.
 143	 * Clear all attribute-related log flags.
 144	 */
 145	if (!XFS_IFORK_Q(ip)) {
 146		iip->ili_format.ilf_fields &=
 147			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
 148		return nvecs;
 149	}
 150
 151	/*
 152	 * Log any necessary attribute data.
 153	 */
 154	switch (ip->i_d.di_aformat) {
 155	case XFS_DINODE_FMT_EXTENTS:
 156		iip->ili_format.ilf_fields &=
 157			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
 158		if ((iip->ili_format.ilf_fields & XFS_ILOG_AEXT) &&
 159		    (ip->i_d.di_anextents > 0) &&
 160		    (ip->i_afp->if_bytes > 0)) {
 161			ASSERT(ip->i_afp->if_u1.if_extents != NULL);
 162			nvecs++;
 163		} else {
 164			iip->ili_format.ilf_fields &= ~XFS_ILOG_AEXT;
 165		}
 166		break;
 167
 168	case XFS_DINODE_FMT_BTREE:
 169		iip->ili_format.ilf_fields &=
 170			~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
 171		if ((iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) &&
 172		    (ip->i_afp->if_broot_bytes > 0)) {
 173			ASSERT(ip->i_afp->if_broot != NULL);
 174			nvecs++;
 175		} else {
 176			iip->ili_format.ilf_fields &= ~XFS_ILOG_ABROOT;
 177		}
 178		break;
 179
 180	case XFS_DINODE_FMT_LOCAL:
 181		iip->ili_format.ilf_fields &=
 182			~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
 183		if ((iip->ili_format.ilf_fields & XFS_ILOG_ADATA) &&
 184		    (ip->i_afp->if_bytes > 0)) {
 185			ASSERT(ip->i_afp->if_u1.if_data != NULL);
 186			nvecs++;
 187		} else {
 188			iip->ili_format.ilf_fields &= ~XFS_ILOG_ADATA;
 189		}
 190		break;
 191
 192	default:
 193		ASSERT(0);
 194		break;
 195	}
 196
 197	return nvecs;
 198}
 199
 200/*
 201 * xfs_inode_item_format_extents - convert in-core extents to on-disk form
 202 *
 203 * For either the data or attr fork in extent format, we need to endian convert
 204 * the in-core extent as we place them into the on-disk inode. In this case, we
 205 * need to do this conversion before we write the extents into the log. Because
 206 * we don't have the disk inode to write into here, we allocate a buffer and
 207 * format the extents into it via xfs_iextents_copy(). We free the buffer in
 208 * the unlock routine after the copy for the log has been made.
 209 *
 210 * In the case of the data fork, the in-core and on-disk fork sizes can be
 211 * different due to delayed allocation extents. We only log on-disk extents
 212 * here, so always use the physical fork size to determine the size of the
 213 * buffer we need to allocate.
 214 */
 215STATIC void
 216xfs_inode_item_format_extents(
 217	struct xfs_inode	*ip,
 218	struct xfs_log_iovec	*vecp,
 219	int			whichfork,
 220	int			type)
 221{
 222	xfs_bmbt_rec_t		*ext_buffer;
 223
 224	ext_buffer = kmem_alloc(XFS_IFORK_SIZE(ip, whichfork), KM_SLEEP);
 225	if (whichfork == XFS_DATA_FORK)
 226		ip->i_itemp->ili_extents_buf = ext_buffer;
 227	else
 228		ip->i_itemp->ili_aextents_buf = ext_buffer;
 229
 230	vecp->i_addr = ext_buffer;
 231	vecp->i_len = xfs_iextents_copy(ip, ext_buffer, whichfork);
 232	vecp->i_type = type;
 233}
 234
 235/*
 236 * This is called to fill in the vector of log iovecs for the
 237 * given inode log item.  It fills the first item with an inode
 238 * log format structure, the second with the on-disk inode structure,
 239 * and a possible third and/or fourth with the inode data/extents/b-tree
 240 * root and inode attributes data/extents/b-tree root.
 241 */
 242STATIC void
 243xfs_inode_item_format(
 244	struct xfs_log_item	*lip,
 245	struct xfs_log_iovec	*vecp)
 
 
 246{
 247	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 248	struct xfs_inode	*ip = iip->ili_inode;
 249	uint			nvecs;
 250	size_t			data_bytes;
 251	xfs_mount_t		*mp;
 252
 253	vecp->i_addr = &iip->ili_format;
 254	vecp->i_len  = sizeof(xfs_inode_log_format_t);
 255	vecp->i_type = XLOG_REG_TYPE_IFORMAT;
 256	vecp++;
 257	nvecs	     = 1;
 258
 259	/*
 260	 * Clear i_update_core if the timestamps (or any other
 261	 * non-transactional modification) need flushing/logging
 262	 * and we're about to log them with the rest of the core.
 263	 *
 264	 * This is the same logic as xfs_iflush() but this code can't
 265	 * run at the same time as xfs_iflush because we're in commit
 266	 * processing here and so we have the inode lock held in
 267	 * exclusive mode.  Although it doesn't really matter
 268	 * for the timestamps if both routines were to grab the
 269	 * timestamps or not.  That would be ok.
 270	 *
 271	 * We clear i_update_core before copying out the data.
 272	 * This is for coordination with our timestamp updates
 273	 * that don't hold the inode lock. They will always
 274	 * update the timestamps BEFORE setting i_update_core,
 275	 * so if we clear i_update_core after they set it we
 276	 * are guaranteed to see their updates to the timestamps
 277	 * either here.  Likewise, if they set it after we clear it
 278	 * here, we'll see it either on the next commit of this
 279	 * inode or the next time the inode gets flushed via
 280	 * xfs_iflush().  This depends on strongly ordered memory
 281	 * semantics, but we have that.  We use the SYNCHRONIZE
 282	 * macro to make sure that the compiler does not reorder
 283	 * the i_update_core access below the data copy below.
 284	 */
 285	if (ip->i_update_core)  {
 286		ip->i_update_core = 0;
 287		SYNCHRONIZE();
 288	}
 289
 290	/*
 291	 * Make sure to get the latest timestamps from the Linux inode.
 292	 */
 293	xfs_synchronize_times(ip);
 294
 295	vecp->i_addr = &ip->i_d;
 296	vecp->i_len  = sizeof(struct xfs_icdinode);
 297	vecp->i_type = XLOG_REG_TYPE_ICORE;
 298	vecp++;
 299	nvecs++;
 300	iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
 301
 302	/*
 303	 * If this is really an old format inode, then we need to
 304	 * log it as such.  This means that we have to copy the link
 305	 * count from the new field to the old.  We don't have to worry
 306	 * about the new fields, because nothing trusts them as long as
 307	 * the old inode version number is there.  If the superblock already
 308	 * has a new version number, then we don't bother converting back.
 309	 */
 310	mp = ip->i_mount;
 311	ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
 312	if (ip->i_d.di_version == 1) {
 313		if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
 314			/*
 315			 * Convert it back.
 316			 */
 317			ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
 318			ip->i_d.di_onlink = ip->i_d.di_nlink;
 319		} else {
 320			/*
 321			 * The superblock version has already been bumped,
 322			 * so just make the conversion to the new inode
 323			 * format permanent.
 324			 */
 325			ip->i_d.di_version = 2;
 326			ip->i_d.di_onlink = 0;
 327			memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 328		}
 329	}
 330
 331	switch (ip->i_d.di_format) {
 332	case XFS_DINODE_FMT_EXTENTS:
 333		ASSERT(!(iip->ili_format.ilf_fields &
 334			 (XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 335			  XFS_ILOG_DEV | XFS_ILOG_UUID)));
 336		if (iip->ili_format.ilf_fields & XFS_ILOG_DEXT) {
 337			ASSERT(ip->i_df.if_bytes > 0);
 338			ASSERT(ip->i_df.if_u1.if_extents != NULL);
 339			ASSERT(ip->i_d.di_nextents > 0);
 340			ASSERT(iip->ili_extents_buf == NULL);
 341			ASSERT((ip->i_df.if_bytes /
 342				(uint)sizeof(xfs_bmbt_rec_t)) > 0);
 343#ifdef XFS_NATIVE_HOST
 344                       if (ip->i_d.di_nextents == ip->i_df.if_bytes /
 345                                               (uint)sizeof(xfs_bmbt_rec_t)) {
 346				/*
 347				 * There are no delayed allocation
 348				 * extents, so just point to the
 349				 * real extents array.
 350				 */
 351				vecp->i_addr = ip->i_df.if_u1.if_extents;
 352				vecp->i_len = ip->i_df.if_bytes;
 353				vecp->i_type = XLOG_REG_TYPE_IEXT;
 354			} else
 355#endif
 356			{
 357				xfs_inode_item_format_extents(ip, vecp,
 358					XFS_DATA_FORK, XLOG_REG_TYPE_IEXT);
 359			}
 360			ASSERT(vecp->i_len <= ip->i_df.if_bytes);
 361			iip->ili_format.ilf_dsize = vecp->i_len;
 362			vecp++;
 363			nvecs++;
 364		}
 365		break;
 366
 367	case XFS_DINODE_FMT_BTREE:
 368		ASSERT(!(iip->ili_format.ilf_fields &
 369			 (XFS_ILOG_DDATA | XFS_ILOG_DEXT |
 370			  XFS_ILOG_DEV | XFS_ILOG_UUID)));
 371		if (iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) {
 372			ASSERT(ip->i_df.if_broot_bytes > 0);
 
 373			ASSERT(ip->i_df.if_broot != NULL);
 374			vecp->i_addr = ip->i_df.if_broot;
 375			vecp->i_len = ip->i_df.if_broot_bytes;
 376			vecp->i_type = XLOG_REG_TYPE_IBROOT;
 377			vecp++;
 378			nvecs++;
 379			iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
 
 
 
 380		}
 381		break;
 382
 383	case XFS_DINODE_FMT_LOCAL:
 384		ASSERT(!(iip->ili_format.ilf_fields &
 385			 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
 386			  XFS_ILOG_DEV | XFS_ILOG_UUID)));
 387		if (iip->ili_format.ilf_fields & XFS_ILOG_DDATA) {
 388			ASSERT(ip->i_df.if_bytes > 0);
 389			ASSERT(ip->i_df.if_u1.if_data != NULL);
 390			ASSERT(ip->i_d.di_size > 0);
 391
 392			vecp->i_addr = ip->i_df.if_u1.if_data;
 393			/*
 394			 * Round i_bytes up to a word boundary.
 395			 * The underlying memory is guaranteed to
 396			 * to be there by xfs_idata_realloc().
 397			 */
 398			data_bytes = roundup(ip->i_df.if_bytes, 4);
 399			ASSERT((ip->i_df.if_real_bytes == 0) ||
 400			       (ip->i_df.if_real_bytes == data_bytes));
 401			vecp->i_len = (int)data_bytes;
 402			vecp->i_type = XLOG_REG_TYPE_ILOCAL;
 403			vecp++;
 404			nvecs++;
 405			iip->ili_format.ilf_dsize = (unsigned)data_bytes;
 
 
 
 406		}
 407		break;
 408
 409	case XFS_DINODE_FMT_DEV:
 410		ASSERT(!(iip->ili_format.ilf_fields &
 411			 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
 412			  XFS_ILOG_DDATA | XFS_ILOG_UUID)));
 413		if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
 414			iip->ili_format.ilf_u.ilfu_rdev =
 415				ip->i_df.if_u2.if_rdev;
 416		}
 417		break;
 418
 419	case XFS_DINODE_FMT_UUID:
 420		ASSERT(!(iip->ili_format.ilf_fields &
 421			 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
 422			  XFS_ILOG_DDATA | XFS_ILOG_DEV)));
 423		if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
 424			iip->ili_format.ilf_u.ilfu_uuid =
 425				ip->i_df.if_u2.if_uuid;
 426		}
 427		break;
 428
 429	default:
 430		ASSERT(0);
 431		break;
 432	}
 
 433
 434	/*
 435	 * If there are no attributes associated with the file,
 436	 * then we're done.
 437	 * Assert that no attribute-related log flags are set.
 438	 */
 439	if (!XFS_IFORK_Q(ip)) {
 440		ASSERT(nvecs == lip->li_desc->lid_size);
 441		iip->ili_format.ilf_size = nvecs;
 442		ASSERT(!(iip->ili_format.ilf_fields &
 443			 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
 444		return;
 445	}
 446
 447	switch (ip->i_d.di_aformat) {
 448	case XFS_DINODE_FMT_EXTENTS:
 449		ASSERT(!(iip->ili_format.ilf_fields &
 450			 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT)));
 451		if (iip->ili_format.ilf_fields & XFS_ILOG_AEXT) {
 452#ifdef DEBUG
 453			int nrecs = ip->i_afp->if_bytes /
 454				(uint)sizeof(xfs_bmbt_rec_t);
 455			ASSERT(nrecs > 0);
 456			ASSERT(nrecs == ip->i_d.di_anextents);
 457			ASSERT(ip->i_afp->if_bytes > 0);
 
 458			ASSERT(ip->i_afp->if_u1.if_extents != NULL);
 459			ASSERT(ip->i_d.di_anextents > 0);
 460#endif
 461#ifdef XFS_NATIVE_HOST
 462			/*
 463			 * There are not delayed allocation extents
 464			 * for attributes, so just point at the array.
 465			 */
 466			vecp->i_addr = ip->i_afp->if_u1.if_extents;
 467			vecp->i_len = ip->i_afp->if_bytes;
 468			vecp->i_type = XLOG_REG_TYPE_IATTR_EXT;
 469#else
 470			ASSERT(iip->ili_aextents_buf == NULL);
 471			xfs_inode_item_format_extents(ip, vecp,
 472					XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT);
 473#endif
 474			iip->ili_format.ilf_asize = vecp->i_len;
 475			vecp++;
 476			nvecs++;
 477		}
 478		break;
 479
 480	case XFS_DINODE_FMT_BTREE:
 481		ASSERT(!(iip->ili_format.ilf_fields &
 482			 (XFS_ILOG_ADATA | XFS_ILOG_AEXT)));
 483		if (iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) {
 484			ASSERT(ip->i_afp->if_broot_bytes > 0);
 
 485			ASSERT(ip->i_afp->if_broot != NULL);
 486			vecp->i_addr = ip->i_afp->if_broot;
 487			vecp->i_len = ip->i_afp->if_broot_bytes;
 488			vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT;
 489			vecp++;
 490			nvecs++;
 491			iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
 
 
 492		}
 493		break;
 494
 495	case XFS_DINODE_FMT_LOCAL:
 496		ASSERT(!(iip->ili_format.ilf_fields &
 497			 (XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
 498		if (iip->ili_format.ilf_fields & XFS_ILOG_ADATA) {
 499			ASSERT(ip->i_afp->if_bytes > 0);
 500			ASSERT(ip->i_afp->if_u1.if_data != NULL);
 501
 502			vecp->i_addr = ip->i_afp->if_u1.if_data;
 
 503			/*
 504			 * Round i_bytes up to a word boundary.
 505			 * The underlying memory is guaranteed to
 506			 * to be there by xfs_idata_realloc().
 507			 */
 508			data_bytes = roundup(ip->i_afp->if_bytes, 4);
 509			ASSERT((ip->i_afp->if_real_bytes == 0) ||
 510			       (ip->i_afp->if_real_bytes == data_bytes));
 511			vecp->i_len = (int)data_bytes;
 512			vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL;
 513			vecp++;
 514			nvecs++;
 515			iip->ili_format.ilf_asize = (unsigned)data_bytes;
 
 
 
 516		}
 517		break;
 518
 519	default:
 520		ASSERT(0);
 521		break;
 522	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 523
 524	ASSERT(nvecs == lip->li_desc->lid_size);
 525	iip->ili_format.ilf_size = nvecs;
 
 526}
 527
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 528
 529/*
 530 * This is called to pin the inode associated with the inode log
 531 * item in memory so it cannot be written out.
 532 */
 533STATIC void
 534xfs_inode_item_pin(
 535	struct xfs_log_item	*lip)
 536{
 537	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
 538
 539	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 540
 541	trace_xfs_inode_pin(ip, _RET_IP_);
 542	atomic_inc(&ip->i_pincount);
 543}
 544
 545
 546/*
 547 * This is called to unpin the inode associated with the inode log
 548 * item which was previously pinned with a call to xfs_inode_item_pin().
 549 *
 550 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
 551 */
 552STATIC void
 553xfs_inode_item_unpin(
 554	struct xfs_log_item	*lip,
 555	int			remove)
 556{
 557	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
 558
 559	trace_xfs_inode_unpin(ip, _RET_IP_);
 560	ASSERT(atomic_read(&ip->i_pincount) > 0);
 561	if (atomic_dec_and_test(&ip->i_pincount))
 562		wake_up(&ip->i_ipin_wait);
 563}
 564
 565/*
 566 * This is called to attempt to lock the inode associated with this
 567 * inode log item, in preparation for the push routine which does the actual
 568 * iflush.  Don't sleep on the inode lock or the flush lock.
 569 *
 570 * If the flush lock is already held, indicating that the inode has
 571 * been or is in the process of being flushed, then (ideally) we'd like to
 572 * see if the inode's buffer is still incore, and if so give it a nudge.
 573 * We delay doing so until the pushbuf routine, though, to avoid holding
 574 * the AIL lock across a call to the blackhole which is the buffer cache.
 575 * Also we don't want to sleep in any device strategy routines, which can happen
 576 * if we do the subsequent bawrite in here.
 577 */
 578STATIC uint
 579xfs_inode_item_trylock(
 580	struct xfs_log_item	*lip)
 
 
 
 581{
 582	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 583	struct xfs_inode	*ip = iip->ili_inode;
 
 
 
 584
 585	if (xfs_ipincount(ip) > 0)
 586		return XFS_ITEM_PINNED;
 587
 588	if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
 589		return XFS_ITEM_LOCKED;
 590
 591	if (!xfs_iflock_nowait(ip)) {
 592		/*
 593		 * inode has already been flushed to the backing buffer,
 594		 * leave it locked in shared mode, pushbuf routine will
 595		 * unlock it.
 596		 */
 597		return XFS_ITEM_PUSHBUF;
 598	}
 599
 600	/* Stale items should force out the iclog */
 
 
 601	if (ip->i_flags & XFS_ISTALE) {
 602		xfs_ifunlock(ip);
 603		/*
 604		 * we hold the AIL lock - notify the unlock routine of this
 605		 * so it doesn't try to get the lock again.
 606		 */
 607		xfs_iunlock(ip, XFS_ILOCK_SHARED|XFS_IUNLOCK_NONOTIFY);
 608		return XFS_ITEM_PINNED;
 
 
 
 
 
 609	}
 610
 611#ifdef DEBUG
 612	if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 613		ASSERT(iip->ili_format.ilf_fields != 0);
 614		ASSERT(iip->ili_logged == 0);
 615		ASSERT(lip->li_flags & XFS_LI_IN_AIL);
 
 
 
 
 
 616	}
 617#endif
 618	return XFS_ITEM_SUCCESS;
 
 
 
 619}
 620
 621/*
 622 * Unlock the inode associated with the inode log item.
 623 * Clear the fields of the inode and inode log item that
 624 * are specific to the current transaction.  If the
 625 * hold flags is set, do not unlock the inode.
 626 */
 627STATIC void
 628xfs_inode_item_unlock(
 629	struct xfs_log_item	*lip)
 630{
 631	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 632	struct xfs_inode	*ip = iip->ili_inode;
 633	unsigned short		lock_flags;
 634
 635	ASSERT(ip->i_itemp != NULL);
 636	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 637
 638	/*
 639	 * If the inode needed a separate buffer with which to log
 640	 * its extents, then free it now.
 641	 */
 642	if (iip->ili_extents_buf != NULL) {
 643		ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
 644		ASSERT(ip->i_d.di_nextents > 0);
 645		ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_DEXT);
 646		ASSERT(ip->i_df.if_bytes > 0);
 647		kmem_free(iip->ili_extents_buf);
 648		iip->ili_extents_buf = NULL;
 649	}
 650	if (iip->ili_aextents_buf != NULL) {
 651		ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
 652		ASSERT(ip->i_d.di_anextents > 0);
 653		ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_AEXT);
 654		ASSERT(ip->i_afp->if_bytes > 0);
 655		kmem_free(iip->ili_aextents_buf);
 656		iip->ili_aextents_buf = NULL;
 657	}
 658
 659	lock_flags = iip->ili_lock_flags;
 660	iip->ili_lock_flags = 0;
 661	if (lock_flags) {
 662		xfs_iunlock(ip, lock_flags);
 663		IRELE(ip);
 664	}
 665}
 666
 667/*
 668 * This is called to find out where the oldest active copy of the inode log
 669 * item in the on disk log resides now that the last log write of it completed
 670 * at the given lsn.  Since we always re-log all dirty data in an inode, the
 671 * latest copy in the on disk log is the only one that matters.  Therefore,
 672 * simply return the given lsn.
 673 *
 674 * If the inode has been marked stale because the cluster is being freed, we
 675 * don't want to (re-)insert this inode into the AIL. There is a race condition
 676 * where the cluster buffer may be unpinned before the inode is inserted into
 677 * the AIL during transaction committed processing. If the buffer is unpinned
 678 * before the inode item has been committed and inserted, then it is possible
 679 * for the buffer to be written and IO completes before the inode is inserted
 680 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
 681 * AIL which will never get removed. It will, however, get reclaimed which
 682 * triggers an assert in xfs_inode_free() complaining about freein an inode
 683 * still in the AIL.
 684 *
 685 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
 686 * transaction committed code knows that it does not need to do any further
 687 * processing on the item.
 688 */
 689STATIC xfs_lsn_t
 690xfs_inode_item_committed(
 691	struct xfs_log_item	*lip,
 692	xfs_lsn_t		lsn)
 693{
 694	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 695	struct xfs_inode	*ip = iip->ili_inode;
 696
 697	if (xfs_iflags_test(ip, XFS_ISTALE)) {
 698		xfs_inode_item_unpin(lip, 0);
 699		return -1;
 700	}
 701	return lsn;
 702}
 703
 704/*
 705 * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
 706 * failed to get the inode flush lock but did get the inode locked SHARED.
 707 * Here we're trying to see if the inode buffer is incore, and if so whether it's
 708 * marked delayed write. If that's the case, we'll promote it and that will
 709 * allow the caller to write the buffer by triggering the xfsbufd to run.
 710 */
 711STATIC bool
 712xfs_inode_item_pushbuf(
 713	struct xfs_log_item	*lip)
 714{
 715	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 716	struct xfs_inode	*ip = iip->ili_inode;
 717	struct xfs_buf		*bp;
 718	bool			ret = true;
 719
 720	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
 721
 722	/*
 723	 * If a flush is not in progress anymore, chances are that the
 724	 * inode was taken off the AIL. So, just get out.
 725	 */
 726	if (completion_done(&ip->i_flush) ||
 727	    !(lip->li_flags & XFS_LI_IN_AIL)) {
 728		xfs_iunlock(ip, XFS_ILOCK_SHARED);
 729		return true;
 730	}
 731
 732	bp = xfs_incore(ip->i_mount->m_ddev_targp, iip->ili_format.ilf_blkno,
 733			iip->ili_format.ilf_len, XBF_TRYLOCK);
 734
 735	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 736	if (!bp)
 737		return true;
 738	if (XFS_BUF_ISDELAYWRITE(bp))
 739		xfs_buf_delwri_promote(bp);
 740	if (xfs_buf_ispinned(bp))
 741		ret = false;
 742	xfs_buf_relse(bp);
 743	return ret;
 744}
 745
 746/*
 747 * This is called to asynchronously write the inode associated with this
 748 * inode log item out to disk. The inode will already have been locked by
 749 * a successful call to xfs_inode_item_trylock().
 750 */
 751STATIC void
 752xfs_inode_item_push(
 753	struct xfs_log_item	*lip)
 754{
 755	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 756	struct xfs_inode	*ip = iip->ili_inode;
 757
 758	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
 759	ASSERT(!completion_done(&ip->i_flush));
 760
 761	/*
 762	 * Since we were able to lock the inode's flush lock and
 763	 * we found it on the AIL, the inode must be dirty.  This
 764	 * is because the inode is removed from the AIL while still
 765	 * holding the flush lock in xfs_iflush_done().  Thus, if
 766	 * we found it in the AIL and were able to obtain the flush
 767	 * lock without sleeping, then there must not have been
 768	 * anyone in the process of flushing the inode.
 769	 */
 770	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
 771	       iip->ili_format.ilf_fields != 0);
 772
 773	/*
 774	 * Push the inode to it's backing buffer. This will not remove the
 775	 * inode from the AIL - a further push will be required to trigger a
 776	 * buffer push. However, this allows all the dirty inodes to be pushed
 777	 * to the buffer before it is pushed to disk. The buffer IO completion
 778	 * will pull the inode from the AIL, mark it clean and unlock the flush
 779	 * lock.
 780	 */
 781	(void) xfs_iflush(ip, SYNC_TRYLOCK);
 782	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 783}
 784
 785/*
 786 * XXX rcc - this one really has to do something.  Probably needs
 787 * to stamp in a new field in the incore inode.
 788 */
 789STATIC void
 790xfs_inode_item_committing(
 791	struct xfs_log_item	*lip,
 792	xfs_lsn_t		lsn)
 793{
 794	INODE_ITEM(lip)->ili_last_lsn = lsn;
 795}
 796
 797/*
 798 * This is the ops vector shared by all buf log items.
 799 */
 800static struct xfs_item_ops xfs_inode_item_ops = {
 801	.iop_size	= xfs_inode_item_size,
 802	.iop_format	= xfs_inode_item_format,
 803	.iop_pin	= xfs_inode_item_pin,
 804	.iop_unpin	= xfs_inode_item_unpin,
 805	.iop_trylock	= xfs_inode_item_trylock,
 806	.iop_unlock	= xfs_inode_item_unlock,
 807	.iop_committed	= xfs_inode_item_committed,
 808	.iop_push	= xfs_inode_item_push,
 809	.iop_pushbuf	= xfs_inode_item_pushbuf,
 810	.iop_committing = xfs_inode_item_committing
 811};
 812
 813
 814/*
 815 * Initialize the inode log item for a newly allocated (in-core) inode.
 816 */
 817void
 818xfs_inode_item_init(
 819	struct xfs_inode	*ip,
 820	struct xfs_mount	*mp)
 821{
 822	struct xfs_inode_log_item *iip;
 823
 824	ASSERT(ip->i_itemp == NULL);
 825	iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
 826
 827	iip->ili_inode = ip;
 828	xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
 829						&xfs_inode_item_ops);
 830	iip->ili_format.ilf_type = XFS_LI_INODE;
 831	iip->ili_format.ilf_ino = ip->i_ino;
 832	iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
 833	iip->ili_format.ilf_len = ip->i_imap.im_len;
 834	iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
 835}
 836
 837/*
 838 * Free the inode log item and any memory hanging off of it.
 839 */
 840void
 841xfs_inode_item_destroy(
 842	xfs_inode_t	*ip)
 843{
 844#ifdef XFS_TRANS_DEBUG
 845	if (ip->i_itemp->ili_root_size != 0) {
 846		kmem_free(ip->i_itemp->ili_orig_root);
 847	}
 848#endif
 849	kmem_zone_free(xfs_ili_zone, ip->i_itemp);
 850}
 851
 852
 853/*
 854 * This is the inode flushing I/O completion routine.  It is called
 855 * from interrupt level when the buffer containing the inode is
 856 * flushed to disk.  It is responsible for removing the inode item
 857 * from the AIL if it has not been re-logged, and unlocking the inode's
 858 * flush lock.
 859 *
 860 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
 861 * list for other inodes that will run this function. We remove them from the
 862 * buffer list so we can process all the inode IO completions in one AIL lock
 863 * traversal.
 864 */
 865void
 866xfs_iflush_done(
 867	struct xfs_buf		*bp,
 868	struct xfs_log_item	*lip)
 869{
 870	struct xfs_inode_log_item *iip;
 871	struct xfs_log_item	*blip;
 872	struct xfs_log_item	*next;
 873	struct xfs_log_item	*prev;
 874	struct xfs_ail		*ailp = lip->li_ailp;
 875	int			need_ail = 0;
 876
 877	/*
 878	 * Scan the buffer IO completions for other inodes being completed and
 879	 * attach them to the current inode log item.
 880	 */
 881	blip = bp->b_fspriv;
 882	prev = NULL;
 883	while (blip != NULL) {
 884		if (lip->li_cb != xfs_iflush_done) {
 885			prev = blip;
 886			blip = blip->li_bio_list;
 887			continue;
 888		}
 889
 890		/* remove from list */
 891		next = blip->li_bio_list;
 892		if (!prev) {
 893			bp->b_fspriv = next;
 894		} else {
 895			prev->li_bio_list = next;
 896		}
 897
 898		/* add to current list */
 899		blip->li_bio_list = lip->li_bio_list;
 900		lip->li_bio_list = blip;
 901
 902		/*
 903		 * while we have the item, do the unlocked check for needing
 904		 * the AIL lock.
 905		 */
 906		iip = INODE_ITEM(blip);
 907		if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
 908			need_ail++;
 909
 910		blip = next;
 911	}
 912
 913	/* make sure we capture the state of the initial inode. */
 914	iip = INODE_ITEM(lip);
 915	if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
 916		need_ail++;
 917
 918	/*
 919	 * We only want to pull the item from the AIL if it is
 920	 * actually there and its location in the log has not
 921	 * changed since we started the flush.  Thus, we only bother
 922	 * if the ili_logged flag is set and the inode's lsn has not
 923	 * changed.  First we check the lsn outside
 924	 * the lock since it's cheaper, and then we recheck while
 925	 * holding the lock before removing the inode from the AIL.
 926	 */
 927	if (need_ail) {
 928		struct xfs_log_item *log_items[need_ail];
 929		int i = 0;
 930		spin_lock(&ailp->xa_lock);
 931		for (blip = lip; blip; blip = blip->li_bio_list) {
 932			iip = INODE_ITEM(blip);
 933			if (iip->ili_logged &&
 934			    blip->li_lsn == iip->ili_flush_lsn) {
 935				log_items[i++] = blip;
 936			}
 937			ASSERT(i <= need_ail);
 938		}
 939		/* xfs_trans_ail_delete_bulk() drops the AIL lock. */
 940		xfs_trans_ail_delete_bulk(ailp, log_items, i);
 
 941	}
 942
 943
 944	/*
 945	 * clean up and unlock the flush lock now we are done. We can clear the
 946	 * ili_last_fields bits now that we know that the data corresponding to
 947	 * them is safely on disk.
 948	 */
 949	for (blip = lip; blip; blip = next) {
 950		next = blip->li_bio_list;
 951		blip->li_bio_list = NULL;
 952
 953		iip = INODE_ITEM(blip);
 954		iip->ili_logged = 0;
 955		iip->ili_last_fields = 0;
 956		xfs_ifunlock(iip->ili_inode);
 957	}
 958}
 959
 960/*
 961 * This is the inode flushing abort routine.  It is called
 962 * from xfs_iflush when the filesystem is shutting down to clean
 963 * up the inode state.
 964 * It is responsible for removing the inode item
 965 * from the AIL if it has not been re-logged, and unlocking the inode's
 966 * flush lock.
 967 */
 968void
 969xfs_iflush_abort(
 970	xfs_inode_t		*ip)
 
 971{
 972	xfs_inode_log_item_t	*iip = ip->i_itemp;
 973
 974	if (iip) {
 975		struct xfs_ail	*ailp = iip->ili_item.li_ailp;
 976		if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 977			spin_lock(&ailp->xa_lock);
 978			if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 979				/* xfs_trans_ail_delete() drops the AIL lock. */
 980				xfs_trans_ail_delete(ailp, (xfs_log_item_t *)iip);
 981			} else
 982				spin_unlock(&ailp->xa_lock);
 983		}
 984		iip->ili_logged = 0;
 985		/*
 986		 * Clear the ili_last_fields bits now that we know that the
 987		 * data corresponding to them is safely on disk.
 988		 */
 989		iip->ili_last_fields = 0;
 990		/*
 991		 * Clear the inode logging fields so no more flushes are
 992		 * attempted.
 993		 */
 994		iip->ili_format.ilf_fields = 0;
 
 995	}
 996	/*
 997	 * Release the inode's flush lock since we're done with it.
 998	 */
 999	xfs_ifunlock(ip);
1000}
1001
1002void
1003xfs_istale_done(
1004	struct xfs_buf		*bp,
1005	struct xfs_log_item	*lip)
1006{
1007	xfs_iflush_abort(INODE_ITEM(lip)->ili_inode);
1008}
1009
1010/*
1011 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
1012 * (which can have different field alignments) to the native version
1013 */
1014int
1015xfs_inode_item_format_convert(
1016	xfs_log_iovec_t		*buf,
1017	xfs_inode_log_format_t	*in_f)
1018{
1019	if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
1020		xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
1021
1022		in_f->ilf_type = in_f32->ilf_type;
1023		in_f->ilf_size = in_f32->ilf_size;
1024		in_f->ilf_fields = in_f32->ilf_fields;
1025		in_f->ilf_asize = in_f32->ilf_asize;
1026		in_f->ilf_dsize = in_f32->ilf_dsize;
1027		in_f->ilf_ino = in_f32->ilf_ino;
1028		/* copy biggest field of ilf_u */
1029		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1030		       in_f32->ilf_u.ilfu_uuid.__u_bits,
1031		       sizeof(uuid_t));
1032		in_f->ilf_blkno = in_f32->ilf_blkno;
1033		in_f->ilf_len = in_f32->ilf_len;
1034		in_f->ilf_boffset = in_f32->ilf_boffset;
1035		return 0;
1036	} else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
1037		xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
1038
1039		in_f->ilf_type = in_f64->ilf_type;
1040		in_f->ilf_size = in_f64->ilf_size;
1041		in_f->ilf_fields = in_f64->ilf_fields;
1042		in_f->ilf_asize = in_f64->ilf_asize;
1043		in_f->ilf_dsize = in_f64->ilf_dsize;
1044		in_f->ilf_ino = in_f64->ilf_ino;
1045		/* copy biggest field of ilf_u */
1046		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1047		       in_f64->ilf_u.ilfu_uuid.__u_bits,
1048		       sizeof(uuid_t));
1049		in_f->ilf_blkno = in_f64->ilf_blkno;
1050		in_f->ilf_len = in_f64->ilf_len;
1051		in_f->ilf_boffset = in_f64->ilf_boffset;
1052		return 0;
1053	}
1054	return EFSCORRUPTED;
1055}
v4.10.11
  1/*
  2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
  3 * All Rights Reserved.
  4 *
  5 * This program is free software; you can redistribute it and/or
  6 * modify it under the terms of the GNU General Public License as
  7 * published by the Free Software Foundation.
  8 *
  9 * This program is distributed in the hope that it would be useful,
 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12 * GNU General Public License for more details.
 13 *
 14 * You should have received a copy of the GNU General Public License
 15 * along with this program; if not, write the Free Software Foundation,
 16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 17 */
 18#include "xfs.h"
 19#include "xfs_fs.h"
 20#include "xfs_format.h"
 21#include "xfs_log_format.h"
 22#include "xfs_trans_resv.h"
 
 
 
 
 23#include "xfs_mount.h"
 
 
 
 24#include "xfs_inode.h"
 25#include "xfs_trans.h"
 26#include "xfs_inode_item.h"
 27#include "xfs_error.h"
 28#include "xfs_trace.h"
 29#include "xfs_trans_priv.h"
 30#include "xfs_log.h"
 31
 32
 33kmem_zone_t	*xfs_ili_zone;		/* inode log item zone */
 34
 35static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
 36{
 37	return container_of(lip, struct xfs_inode_log_item, ili_item);
 38}
 39
 40STATIC void
 41xfs_inode_item_data_fork_size(
 42	struct xfs_inode_log_item *iip,
 43	int			*nvecs,
 44	int			*nbytes)
 
 
 
 
 
 
 45{
 
 46	struct xfs_inode	*ip = iip->ili_inode;
 
 
 
 
 
 
 
 47
 48	switch (ip->i_d.di_format) {
 49	case XFS_DINODE_FMT_EXTENTS:
 50		if ((iip->ili_fields & XFS_ILOG_DEXT) &&
 51		    ip->i_d.di_nextents > 0 &&
 52		    ip->i_df.if_bytes > 0) {
 53			/* worst case, doesn't subtract delalloc extents */
 54			*nbytes += XFS_IFORK_DSIZE(ip);
 55			*nvecs += 1;
 
 
 
 
 56		}
 57		break;
 
 58	case XFS_DINODE_FMT_BTREE:
 59		if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 60		    ip->i_df.if_broot_bytes > 0) {
 61			*nbytes += ip->i_df.if_broot_bytes;
 62			*nvecs += 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 63		}
 64		break;
 
 65	case XFS_DINODE_FMT_LOCAL:
 66		if ((iip->ili_fields & XFS_ILOG_DDATA) &&
 67		    ip->i_df.if_bytes > 0) {
 68			*nbytes += roundup(ip->i_df.if_bytes, 4);
 69			*nvecs += 1;
 
 
 
 
 
 
 70		}
 71		break;
 72
 73	case XFS_DINODE_FMT_DEV:
 
 
 
 
 
 74	case XFS_DINODE_FMT_UUID:
 
 
 
 75		break;
 
 76	default:
 77		ASSERT(0);
 78		break;
 79	}
 80}
 81
 82STATIC void
 83xfs_inode_item_attr_fork_size(
 84	struct xfs_inode_log_item *iip,
 85	int			*nvecs,
 86	int			*nbytes)
 87{
 88	struct xfs_inode	*ip = iip->ili_inode;
 
 
 
 89
 
 
 
 90	switch (ip->i_d.di_aformat) {
 91	case XFS_DINODE_FMT_EXTENTS:
 92		if ((iip->ili_fields & XFS_ILOG_AEXT) &&
 93		    ip->i_d.di_anextents > 0 &&
 94		    ip->i_afp->if_bytes > 0) {
 95			/* worst case, doesn't subtract unused space */
 96			*nbytes += XFS_IFORK_ASIZE(ip);
 97			*nvecs += 1;
 
 
 
 98		}
 99		break;
 
100	case XFS_DINODE_FMT_BTREE:
101		if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
102		    ip->i_afp->if_broot_bytes > 0) {
103			*nbytes += ip->i_afp->if_broot_bytes;
104			*nvecs += 1;
 
 
 
 
105		}
106		break;
 
107	case XFS_DINODE_FMT_LOCAL:
108		if ((iip->ili_fields & XFS_ILOG_ADATA) &&
109		    ip->i_afp->if_bytes > 0) {
110			*nbytes += roundup(ip->i_afp->if_bytes, 4);
111			*nvecs += 1;
 
 
 
 
112		}
113		break;
 
114	default:
115		ASSERT(0);
116		break;
117	}
 
 
118}
119
120/*
121 * This returns the number of iovecs needed to log the given inode item.
 
 
 
 
 
 
 
122 *
123 * We need one iovec for the inode log format structure, one for the
124 * inode core, and possibly one for the inode data/extents/b-tree root
125 * and one for the inode attribute data/extents/b-tree root.
 
126 */
127STATIC void
128xfs_inode_item_size(
129	struct xfs_log_item	*lip,
130	int			*nvecs,
131	int			*nbytes)
132{
133	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
134	struct xfs_inode	*ip = iip->ili_inode;
135
136	*nvecs += 2;
137	*nbytes += sizeof(struct xfs_inode_log_format) +
138		   xfs_log_dinode_size(ip->i_d.di_version);
139
140	xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
141	if (XFS_IFORK_Q(ip))
142		xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
 
 
143}
144
 
 
 
 
 
 
 
145STATIC void
146xfs_inode_item_format_data_fork(
147	struct xfs_inode_log_item *iip,
148	struct xfs_inode_log_format *ilf,
149	struct xfs_log_vec	*lv,
150	struct xfs_log_iovec	**vecp)
151{
 
152	struct xfs_inode	*ip = iip->ili_inode;
 
153	size_t			data_bytes;
 
154
155	switch (ip->i_d.di_format) {
156	case XFS_DINODE_FMT_EXTENTS:
157		iip->ili_fields &=
158			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
159			  XFS_ILOG_DEV | XFS_ILOG_UUID);
160
161		if ((iip->ili_fields & XFS_ILOG_DEXT) &&
162		    ip->i_d.di_nextents > 0 &&
163		    ip->i_df.if_bytes > 0) {
164			struct xfs_bmbt_rec *p;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
165
166			ASSERT(ip->i_df.if_u1.if_extents != NULL);
167			ASSERT(xfs_iext_count(&ip->i_df) > 0);
 
 
168
169			p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
170			data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
171			xlog_finish_iovec(lv, *vecp, data_bytes);
 
 
 
172
173			ASSERT(data_bytes <= ip->i_df.if_bytes);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
174
175			ilf->ilf_dsize = data_bytes;
176			ilf->ilf_size++;
177		} else {
178			iip->ili_fields &= ~XFS_ILOG_DEXT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
179		}
180		break;
 
181	case XFS_DINODE_FMT_BTREE:
182		iip->ili_fields &=
183			~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
184			  XFS_ILOG_DEV | XFS_ILOG_UUID);
185
186		if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
187		    ip->i_df.if_broot_bytes > 0) {
188			ASSERT(ip->i_df.if_broot != NULL);
189			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
190					ip->i_df.if_broot,
191					ip->i_df.if_broot_bytes);
192			ilf->ilf_dsize = ip->i_df.if_broot_bytes;
193			ilf->ilf_size++;
194		} else {
195			ASSERT(!(iip->ili_fields &
196				 XFS_ILOG_DBROOT));
197			iip->ili_fields &= ~XFS_ILOG_DBROOT;
198		}
199		break;
 
200	case XFS_DINODE_FMT_LOCAL:
201		iip->ili_fields &=
202			~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
203			  XFS_ILOG_DEV | XFS_ILOG_UUID);
204		if ((iip->ili_fields & XFS_ILOG_DDATA) &&
205		    ip->i_df.if_bytes > 0) {
 
 
 
 
206			/*
207			 * Round i_bytes up to a word boundary.
208			 * The underlying memory is guaranteed to
209			 * to be there by xfs_idata_realloc().
210			 */
211			data_bytes = roundup(ip->i_df.if_bytes, 4);
212			ASSERT(ip->i_df.if_real_bytes == 0 ||
213			       ip->i_df.if_real_bytes >= data_bytes);
214			ASSERT(ip->i_df.if_u1.if_data != NULL);
215			ASSERT(ip->i_d.di_size > 0);
216			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
217					ip->i_df.if_u1.if_data, data_bytes);
218			ilf->ilf_dsize = (unsigned)data_bytes;
219			ilf->ilf_size++;
220		} else {
221			iip->ili_fields &= ~XFS_ILOG_DDATA;
222		}
223		break;
 
224	case XFS_DINODE_FMT_DEV:
225		iip->ili_fields &=
226			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
227			  XFS_ILOG_DEXT | XFS_ILOG_UUID);
228		if (iip->ili_fields & XFS_ILOG_DEV)
229			ilf->ilf_u.ilfu_rdev = ip->i_df.if_u2.if_rdev;
 
 
230		break;
 
231	case XFS_DINODE_FMT_UUID:
232		iip->ili_fields &=
233			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
234			  XFS_ILOG_DEXT | XFS_ILOG_DEV);
235		if (iip->ili_fields & XFS_ILOG_UUID)
236			ilf->ilf_u.ilfu_uuid = ip->i_df.if_u2.if_uuid;
 
 
237		break;
 
238	default:
239		ASSERT(0);
240		break;
241	}
242}
243
244STATIC void
245xfs_inode_item_format_attr_fork(
246	struct xfs_inode_log_item *iip,
247	struct xfs_inode_log_format *ilf,
248	struct xfs_log_vec	*lv,
249	struct xfs_log_iovec	**vecp)
250{
251	struct xfs_inode	*ip = iip->ili_inode;
252	size_t			data_bytes;
 
 
 
253
254	switch (ip->i_d.di_aformat) {
255	case XFS_DINODE_FMT_EXTENTS:
256		iip->ili_fields &=
257			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
258
259		if ((iip->ili_fields & XFS_ILOG_AEXT) &&
260		    ip->i_d.di_anextents > 0 &&
261		    ip->i_afp->if_bytes > 0) {
262			struct xfs_bmbt_rec *p;
263
264			ASSERT(xfs_iext_count(ip->i_afp) ==
265				ip->i_d.di_anextents);
266			ASSERT(ip->i_afp->if_u1.if_extents != NULL);
267
268			p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
269			data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
270			xlog_finish_iovec(lv, *vecp, data_bytes);
271
272			ilf->ilf_asize = data_bytes;
273			ilf->ilf_size++;
274		} else {
275			iip->ili_fields &= ~XFS_ILOG_AEXT;
 
 
 
 
 
 
 
 
 
276		}
277		break;
 
278	case XFS_DINODE_FMT_BTREE:
279		iip->ili_fields &=
280			~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
281
282		if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
283		    ip->i_afp->if_broot_bytes > 0) {
284			ASSERT(ip->i_afp->if_broot != NULL);
285
286			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
287					ip->i_afp->if_broot,
288					ip->i_afp->if_broot_bytes);
289			ilf->ilf_asize = ip->i_afp->if_broot_bytes;
290			ilf->ilf_size++;
291		} else {
292			iip->ili_fields &= ~XFS_ILOG_ABROOT;
293		}
294		break;
 
295	case XFS_DINODE_FMT_LOCAL:
296		iip->ili_fields &=
297			~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
 
 
 
298
299		if ((iip->ili_fields & XFS_ILOG_ADATA) &&
300		    ip->i_afp->if_bytes > 0) {
301			/*
302			 * Round i_bytes up to a word boundary.
303			 * The underlying memory is guaranteed to
304			 * to be there by xfs_idata_realloc().
305			 */
306			data_bytes = roundup(ip->i_afp->if_bytes, 4);
307			ASSERT(ip->i_afp->if_real_bytes == 0 ||
308			       ip->i_afp->if_real_bytes >= data_bytes);
309			ASSERT(ip->i_afp->if_u1.if_data != NULL);
310			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
311					ip->i_afp->if_u1.if_data,
312					data_bytes);
313			ilf->ilf_asize = (unsigned)data_bytes;
314			ilf->ilf_size++;
315		} else {
316			iip->ili_fields &= ~XFS_ILOG_ADATA;
317		}
318		break;
 
319	default:
320		ASSERT(0);
321		break;
322	}
323}
324
325static void
326xfs_inode_to_log_dinode(
327	struct xfs_inode	*ip,
328	struct xfs_log_dinode	*to,
329	xfs_lsn_t		lsn)
330{
331	struct xfs_icdinode	*from = &ip->i_d;
332	struct inode		*inode = VFS_I(ip);
333
334	to->di_magic = XFS_DINODE_MAGIC;
335
336	to->di_version = from->di_version;
337	to->di_format = from->di_format;
338	to->di_uid = from->di_uid;
339	to->di_gid = from->di_gid;
340	to->di_projid_lo = from->di_projid_lo;
341	to->di_projid_hi = from->di_projid_hi;
342
343	memset(to->di_pad, 0, sizeof(to->di_pad));
344	memset(to->di_pad3, 0, sizeof(to->di_pad3));
345	to->di_atime.t_sec = inode->i_atime.tv_sec;
346	to->di_atime.t_nsec = inode->i_atime.tv_nsec;
347	to->di_mtime.t_sec = inode->i_mtime.tv_sec;
348	to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
349	to->di_ctime.t_sec = inode->i_ctime.tv_sec;
350	to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
351	to->di_nlink = inode->i_nlink;
352	to->di_gen = inode->i_generation;
353	to->di_mode = inode->i_mode;
354
355	to->di_size = from->di_size;
356	to->di_nblocks = from->di_nblocks;
357	to->di_extsize = from->di_extsize;
358	to->di_nextents = from->di_nextents;
359	to->di_anextents = from->di_anextents;
360	to->di_forkoff = from->di_forkoff;
361	to->di_aformat = from->di_aformat;
362	to->di_dmevmask = from->di_dmevmask;
363	to->di_dmstate = from->di_dmstate;
364	to->di_flags = from->di_flags;
365
366	if (from->di_version == 3) {
367		to->di_changecount = inode->i_version;
368		to->di_crtime.t_sec = from->di_crtime.t_sec;
369		to->di_crtime.t_nsec = from->di_crtime.t_nsec;
370		to->di_flags2 = from->di_flags2;
371		to->di_cowextsize = from->di_cowextsize;
372		to->di_ino = ip->i_ino;
373		to->di_lsn = lsn;
374		memset(to->di_pad2, 0, sizeof(to->di_pad2));
375		uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
376		to->di_flushiter = 0;
377	} else {
378		to->di_flushiter = from->di_flushiter;
379	}
380}
381
382/*
383 * Format the inode core. Current timestamp data is only in the VFS inode
384 * fields, so we need to grab them from there. Hence rather than just copying
385 * the XFS inode core structure, format the fields directly into the iovec.
386 */
387static void
388xfs_inode_item_format_core(
389	struct xfs_inode	*ip,
390	struct xfs_log_vec	*lv,
391	struct xfs_log_iovec	**vecp)
392{
393	struct xfs_log_dinode	*dic;
394
395	dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
396	xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
397	xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_d.di_version));
398}
399
400/*
401 * This is called to fill in the vector of log iovecs for the given inode
402 * log item.  It fills the first item with an inode log format structure,
403 * the second with the on-disk inode structure, and a possible third and/or
404 * fourth with the inode data/extents/b-tree root and inode attributes
405 * data/extents/b-tree root.
406 */
407STATIC void
408xfs_inode_item_format(
409	struct xfs_log_item	*lip,
410	struct xfs_log_vec	*lv)
411{
412	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
413	struct xfs_inode	*ip = iip->ili_inode;
414	struct xfs_inode_log_format *ilf;
415	struct xfs_log_iovec	*vecp = NULL;
416
417	ASSERT(ip->i_d.di_version > 1);
418
419	ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
420	ilf->ilf_type = XFS_LI_INODE;
421	ilf->ilf_ino = ip->i_ino;
422	ilf->ilf_blkno = ip->i_imap.im_blkno;
423	ilf->ilf_len = ip->i_imap.im_len;
424	ilf->ilf_boffset = ip->i_imap.im_boffset;
425	ilf->ilf_fields = XFS_ILOG_CORE;
426	ilf->ilf_size = 2; /* format + core */
427	xlog_finish_iovec(lv, vecp, sizeof(struct xfs_inode_log_format));
428
429	xfs_inode_item_format_core(ip, lv, &vecp);
430	xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
431	if (XFS_IFORK_Q(ip)) {
432		xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
433	} else {
434		iip->ili_fields &=
435			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
436	}
437
438	/* update the format with the exact fields we actually logged */
439	ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
440}
441
442/*
443 * This is called to pin the inode associated with the inode log
444 * item in memory so it cannot be written out.
445 */
446STATIC void
447xfs_inode_item_pin(
448	struct xfs_log_item	*lip)
449{
450	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
451
452	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
453
454	trace_xfs_inode_pin(ip, _RET_IP_);
455	atomic_inc(&ip->i_pincount);
456}
457
458
459/*
460 * This is called to unpin the inode associated with the inode log
461 * item which was previously pinned with a call to xfs_inode_item_pin().
462 *
463 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
464 */
465STATIC void
466xfs_inode_item_unpin(
467	struct xfs_log_item	*lip,
468	int			remove)
469{
470	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
471
472	trace_xfs_inode_unpin(ip, _RET_IP_);
473	ASSERT(atomic_read(&ip->i_pincount) > 0);
474	if (atomic_dec_and_test(&ip->i_pincount))
475		wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
476}
477
 
 
 
 
 
 
 
 
 
 
 
 
 
478STATIC uint
479xfs_inode_item_push(
480	struct xfs_log_item	*lip,
481	struct list_head	*buffer_list)
482		__releases(&lip->li_ailp->xa_lock)
483		__acquires(&lip->li_ailp->xa_lock)
484{
485	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
486	struct xfs_inode	*ip = iip->ili_inode;
487	struct xfs_buf		*bp = NULL;
488	uint			rval = XFS_ITEM_SUCCESS;
489	int			error;
490
491	if (xfs_ipincount(ip) > 0)
492		return XFS_ITEM_PINNED;
493
494	if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
495		return XFS_ITEM_LOCKED;
496
497	/*
498	 * Re-check the pincount now that we stabilized the value by
499	 * taking the ilock.
500	 */
501	if (xfs_ipincount(ip) > 0) {
502		rval = XFS_ITEM_PINNED;
503		goto out_unlock;
504	}
505
506	/*
507	 * Stale inode items should force out the iclog.
508	 */
509	if (ip->i_flags & XFS_ISTALE) {
510		rval = XFS_ITEM_PINNED;
511		goto out_unlock;
512	}
513
514	/*
515	 * Someone else is already flushing the inode.  Nothing we can do
516	 * here but wait for the flush to finish and remove the item from
517	 * the AIL.
518	 */
519	if (!xfs_iflock_nowait(ip)) {
520		rval = XFS_ITEM_FLUSHING;
521		goto out_unlock;
522	}
523
524	ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
525	ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
526
527	spin_unlock(&lip->li_ailp->xa_lock);
528
529	error = xfs_iflush(ip, &bp);
530	if (!error) {
531		if (!xfs_buf_delwri_queue(bp, buffer_list))
532			rval = XFS_ITEM_FLUSHING;
533		xfs_buf_relse(bp);
534	}
535
536	spin_lock(&lip->li_ailp->xa_lock);
537out_unlock:
538	xfs_iunlock(ip, XFS_ILOCK_SHARED);
539	return rval;
540}
541
542/*
543 * Unlock the inode associated with the inode log item.
544 * Clear the fields of the inode and inode log item that
545 * are specific to the current transaction.  If the
546 * hold flags is set, do not unlock the inode.
547 */
548STATIC void
549xfs_inode_item_unlock(
550	struct xfs_log_item	*lip)
551{
552	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
553	struct xfs_inode	*ip = iip->ili_inode;
554	unsigned short		lock_flags;
555
556	ASSERT(ip->i_itemp != NULL);
557	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
558
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
559	lock_flags = iip->ili_lock_flags;
560	iip->ili_lock_flags = 0;
561	if (lock_flags)
562		xfs_iunlock(ip, lock_flags);
 
 
563}
564
565/*
566 * This is called to find out where the oldest active copy of the inode log
567 * item in the on disk log resides now that the last log write of it completed
568 * at the given lsn.  Since we always re-log all dirty data in an inode, the
569 * latest copy in the on disk log is the only one that matters.  Therefore,
570 * simply return the given lsn.
571 *
572 * If the inode has been marked stale because the cluster is being freed, we
573 * don't want to (re-)insert this inode into the AIL. There is a race condition
574 * where the cluster buffer may be unpinned before the inode is inserted into
575 * the AIL during transaction committed processing. If the buffer is unpinned
576 * before the inode item has been committed and inserted, then it is possible
577 * for the buffer to be written and IO completes before the inode is inserted
578 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
579 * AIL which will never get removed. It will, however, get reclaimed which
580 * triggers an assert in xfs_inode_free() complaining about freein an inode
581 * still in the AIL.
582 *
583 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
584 * transaction committed code knows that it does not need to do any further
585 * processing on the item.
586 */
587STATIC xfs_lsn_t
588xfs_inode_item_committed(
589	struct xfs_log_item	*lip,
590	xfs_lsn_t		lsn)
591{
592	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
593	struct xfs_inode	*ip = iip->ili_inode;
594
595	if (xfs_iflags_test(ip, XFS_ISTALE)) {
596		xfs_inode_item_unpin(lip, 0);
597		return -1;
598	}
599	return lsn;
600}
601
602/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
603 * XXX rcc - this one really has to do something.  Probably needs
604 * to stamp in a new field in the incore inode.
605 */
606STATIC void
607xfs_inode_item_committing(
608	struct xfs_log_item	*lip,
609	xfs_lsn_t		lsn)
610{
611	INODE_ITEM(lip)->ili_last_lsn = lsn;
612}
613
614/*
615 * This is the ops vector shared by all buf log items.
616 */
617static const struct xfs_item_ops xfs_inode_item_ops = {
618	.iop_size	= xfs_inode_item_size,
619	.iop_format	= xfs_inode_item_format,
620	.iop_pin	= xfs_inode_item_pin,
621	.iop_unpin	= xfs_inode_item_unpin,
 
622	.iop_unlock	= xfs_inode_item_unlock,
623	.iop_committed	= xfs_inode_item_committed,
624	.iop_push	= xfs_inode_item_push,
 
625	.iop_committing = xfs_inode_item_committing
626};
627
628
629/*
630 * Initialize the inode log item for a newly allocated (in-core) inode.
631 */
632void
633xfs_inode_item_init(
634	struct xfs_inode	*ip,
635	struct xfs_mount	*mp)
636{
637	struct xfs_inode_log_item *iip;
638
639	ASSERT(ip->i_itemp == NULL);
640	iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
641
642	iip->ili_inode = ip;
643	xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
644						&xfs_inode_item_ops);
 
 
 
 
 
645}
646
647/*
648 * Free the inode log item and any memory hanging off of it.
649 */
650void
651xfs_inode_item_destroy(
652	xfs_inode_t	*ip)
653{
654	kmem_free(ip->i_itemp->ili_item.li_lv_shadow);
 
 
 
 
655	kmem_zone_free(xfs_ili_zone, ip->i_itemp);
656}
657
658
659/*
660 * This is the inode flushing I/O completion routine.  It is called
661 * from interrupt level when the buffer containing the inode is
662 * flushed to disk.  It is responsible for removing the inode item
663 * from the AIL if it has not been re-logged, and unlocking the inode's
664 * flush lock.
665 *
666 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
667 * list for other inodes that will run this function. We remove them from the
668 * buffer list so we can process all the inode IO completions in one AIL lock
669 * traversal.
670 */
671void
672xfs_iflush_done(
673	struct xfs_buf		*bp,
674	struct xfs_log_item	*lip)
675{
676	struct xfs_inode_log_item *iip;
677	struct xfs_log_item	*blip;
678	struct xfs_log_item	*next;
679	struct xfs_log_item	*prev;
680	struct xfs_ail		*ailp = lip->li_ailp;
681	int			need_ail = 0;
682
683	/*
684	 * Scan the buffer IO completions for other inodes being completed and
685	 * attach them to the current inode log item.
686	 */
687	blip = bp->b_fspriv;
688	prev = NULL;
689	while (blip != NULL) {
690		if (blip->li_cb != xfs_iflush_done) {
691			prev = blip;
692			blip = blip->li_bio_list;
693			continue;
694		}
695
696		/* remove from list */
697		next = blip->li_bio_list;
698		if (!prev) {
699			bp->b_fspriv = next;
700		} else {
701			prev->li_bio_list = next;
702		}
703
704		/* add to current list */
705		blip->li_bio_list = lip->li_bio_list;
706		lip->li_bio_list = blip;
707
708		/*
709		 * while we have the item, do the unlocked check for needing
710		 * the AIL lock.
711		 */
712		iip = INODE_ITEM(blip);
713		if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
714			need_ail++;
715
716		blip = next;
717	}
718
719	/* make sure we capture the state of the initial inode. */
720	iip = INODE_ITEM(lip);
721	if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
722		need_ail++;
723
724	/*
725	 * We only want to pull the item from the AIL if it is
726	 * actually there and its location in the log has not
727	 * changed since we started the flush.  Thus, we only bother
728	 * if the ili_logged flag is set and the inode's lsn has not
729	 * changed.  First we check the lsn outside
730	 * the lock since it's cheaper, and then we recheck while
731	 * holding the lock before removing the inode from the AIL.
732	 */
733	if (need_ail) {
734		struct xfs_log_item *log_items[need_ail];
735		int i = 0;
736		spin_lock(&ailp->xa_lock);
737		for (blip = lip; blip; blip = blip->li_bio_list) {
738			iip = INODE_ITEM(blip);
739			if (iip->ili_logged &&
740			    blip->li_lsn == iip->ili_flush_lsn) {
741				log_items[i++] = blip;
742			}
743			ASSERT(i <= need_ail);
744		}
745		/* xfs_trans_ail_delete_bulk() drops the AIL lock. */
746		xfs_trans_ail_delete_bulk(ailp, log_items, i,
747					  SHUTDOWN_CORRUPT_INCORE);
748	}
749
750
751	/*
752	 * clean up and unlock the flush lock now we are done. We can clear the
753	 * ili_last_fields bits now that we know that the data corresponding to
754	 * them is safely on disk.
755	 */
756	for (blip = lip; blip; blip = next) {
757		next = blip->li_bio_list;
758		blip->li_bio_list = NULL;
759
760		iip = INODE_ITEM(blip);
761		iip->ili_logged = 0;
762		iip->ili_last_fields = 0;
763		xfs_ifunlock(iip->ili_inode);
764	}
765}
766
767/*
768 * This is the inode flushing abort routine.  It is called from xfs_iflush when
769 * the filesystem is shutting down to clean up the inode state.  It is
770 * responsible for removing the inode item from the AIL if it has not been
771 * re-logged, and unlocking the inode's flush lock.
 
 
772 */
773void
774xfs_iflush_abort(
775	xfs_inode_t		*ip,
776	bool			stale)
777{
778	xfs_inode_log_item_t	*iip = ip->i_itemp;
779
780	if (iip) {
 
781		if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
782			xfs_trans_ail_remove(&iip->ili_item,
783					     stale ? SHUTDOWN_LOG_IO_ERROR :
784						     SHUTDOWN_CORRUPT_INCORE);
 
 
 
785		}
786		iip->ili_logged = 0;
787		/*
788		 * Clear the ili_last_fields bits now that we know that the
789		 * data corresponding to them is safely on disk.
790		 */
791		iip->ili_last_fields = 0;
792		/*
793		 * Clear the inode logging fields so no more flushes are
794		 * attempted.
795		 */
796		iip->ili_fields = 0;
797		iip->ili_fsync_fields = 0;
798	}
799	/*
800	 * Release the inode's flush lock since we're done with it.
801	 */
802	xfs_ifunlock(ip);
803}
804
805void
806xfs_istale_done(
807	struct xfs_buf		*bp,
808	struct xfs_log_item	*lip)
809{
810	xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
811}
812
813/*
814 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
815 * (which can have different field alignments) to the native version
816 */
817int
818xfs_inode_item_format_convert(
819	xfs_log_iovec_t		*buf,
820	xfs_inode_log_format_t	*in_f)
821{
822	if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
823		xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
824
825		in_f->ilf_type = in_f32->ilf_type;
826		in_f->ilf_size = in_f32->ilf_size;
827		in_f->ilf_fields = in_f32->ilf_fields;
828		in_f->ilf_asize = in_f32->ilf_asize;
829		in_f->ilf_dsize = in_f32->ilf_dsize;
830		in_f->ilf_ino = in_f32->ilf_ino;
831		/* copy biggest field of ilf_u */
832		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
833		       in_f32->ilf_u.ilfu_uuid.__u_bits,
834		       sizeof(uuid_t));
835		in_f->ilf_blkno = in_f32->ilf_blkno;
836		in_f->ilf_len = in_f32->ilf_len;
837		in_f->ilf_boffset = in_f32->ilf_boffset;
838		return 0;
839	} else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
840		xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
841
842		in_f->ilf_type = in_f64->ilf_type;
843		in_f->ilf_size = in_f64->ilf_size;
844		in_f->ilf_fields = in_f64->ilf_fields;
845		in_f->ilf_asize = in_f64->ilf_asize;
846		in_f->ilf_dsize = in_f64->ilf_dsize;
847		in_f->ilf_ino = in_f64->ilf_ino;
848		/* copy biggest field of ilf_u */
849		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
850		       in_f64->ilf_u.ilfu_uuid.__u_bits,
851		       sizeof(uuid_t));
852		in_f->ilf_blkno = in_f64->ilf_blkno;
853		in_f->ilf_len = in_f64->ilf_len;
854		in_f->ilf_boffset = in_f64->ilf_boffset;
855		return 0;
856	}
857	return -EFSCORRUPTED;
858}