fs/xfs/xfs_inode_item.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_mount.h"
  13 #include "xfs_inode.h"
  14 #include "xfs_trans.h"
  15 #include "xfs_inode_item.h"
  16 #include "xfs_trace.h"
  17 #include "xfs_trans_priv.h"
  18 #include "xfs_buf_item.h"
  19 #include "xfs_log.h"
  20 #include "xfs_error.h"
  21
  22 #include <linux/iversion.h>
  23
  24 kmem_zone_t     *xfs_ili_zone;          /* inode log item zone */
  25
  26 static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
  27 {
  28         return container_of(lip, struct xfs_inode_log_item, ili_item);
  29 }
  30
  31 STATIC void
  32 xfs_inode_item_data_fork_size(
  33         struct xfs_inode_log_item *iip,
  34         int                     *nvecs,
  35         int                     *nbytes)
  36 {
  37         struct xfs_inode        *ip = iip->ili_inode;
  38
  39         switch (ip->i_df.if_format) {
  40         case XFS_DINODE_FMT_EXTENTS:
  41                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
  42                     ip->i_df.if_nextents > 0 &&
  43                     ip->i_df.if_bytes > 0) {
  44                         /* worst case, doesn't subtract delalloc extents */
  45                         *nbytes += XFS_IFORK_DSIZE(ip);
  46                         *nvecs += 1;
  47                 }
  48                 break;
  49         case XFS_DINODE_FMT_BTREE:
  50                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
  51                     ip->i_df.if_broot_bytes > 0) {
  52                         *nbytes += ip->i_df.if_broot_bytes;
  53                         *nvecs += 1;
  54                 }
  55                 break;
  56         case XFS_DINODE_FMT_LOCAL:
  57                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
  58                     ip->i_df.if_bytes > 0) {
  59                         *nbytes += roundup(ip->i_df.if_bytes, 4);
  60                         *nvecs += 1;
  61                 }
  62                 break;
  63
  64         case XFS_DINODE_FMT_DEV:
  65                 break;
  66         default:
  67                 ASSERT(0);
  68                 break;
  69         }
  70 }
  71
  72 STATIC void
  73 xfs_inode_item_attr_fork_size(
  74         struct xfs_inode_log_item *iip,
  75         int                     *nvecs,
  76         int                     *nbytes)
  77 {
  78         struct xfs_inode        *ip = iip->ili_inode;
  79
  80         switch (ip->i_afp->if_format) {
  81         case XFS_DINODE_FMT_EXTENTS:
  82                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
  83                     ip->i_afp->if_nextents > 0 &&
  84                     ip->i_afp->if_bytes > 0) {
  85                         /* worst case, doesn't subtract unused space */
  86                         *nbytes += XFS_IFORK_ASIZE(ip);
  87                         *nvecs += 1;
  88                 }
  89                 break;
  90         case XFS_DINODE_FMT_BTREE:
  91                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
  92                     ip->i_afp->if_broot_bytes > 0) {
  93                         *nbytes += ip->i_afp->if_broot_bytes;
  94                         *nvecs += 1;
  95                 }
  96                 break;
  97         case XFS_DINODE_FMT_LOCAL:
  98                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
  99                     ip->i_afp->if_bytes > 0) {
 100                         *nbytes += roundup(ip->i_afp->if_bytes, 4);
 101                         *nvecs += 1;
 102                 }
 103                 break;
 104         default:
 105                 ASSERT(0);
 106                 break;
 107         }
 108 }
 109
 110 /*
 111  * This returns the number of iovecs needed to log the given inode item.
 112  *
 113  * We need one iovec for the inode log format structure, one for the
 114  * inode core, and possibly one for the inode data/extents/b-tree root
 115  * and one for the inode attribute data/extents/b-tree root.
 116  */
 117 STATIC void
 118 xfs_inode_item_size(
 119         struct xfs_log_item     *lip,
 120         int                     *nvecs,
 121         int                     *nbytes)
 122 {
 123         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 124         struct xfs_inode        *ip = iip->ili_inode;
 125
 126         *nvecs += 2;
 127         *nbytes += sizeof(struct xfs_inode_log_format) +
 128                    xfs_log_dinode_size(ip->i_mount);
 129
 130         xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
 131         if (XFS_IFORK_Q(ip))
 132                 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
 133 }
 134
 135 STATIC void
 136 xfs_inode_item_format_data_fork(
 137         struct xfs_inode_log_item *iip,
 138         struct xfs_inode_log_format *ilf,
 139         struct xfs_log_vec      *lv,
 140         struct xfs_log_iovec    **vecp)
 141 {
 142         struct xfs_inode        *ip = iip->ili_inode;
 143         size_t                  data_bytes;
 144
 145         switch (ip->i_df.if_format) {
 146         case XFS_DINODE_FMT_EXTENTS:
 147                 iip->ili_fields &=
 148                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
 149
 150                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
 151                     ip->i_df.if_nextents > 0 &&
 152                     ip->i_df.if_bytes > 0) {
 153                         struct xfs_bmbt_rec *p;
 154
 155                         ASSERT(xfs_iext_count(&ip->i_df) > 0);
 156
 157                         p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
 158                         data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
 159                         xlog_finish_iovec(lv, *vecp, data_bytes);
 160
 161                         ASSERT(data_bytes <= ip->i_df.if_bytes);
 162
 163                         ilf->ilf_dsize = data_bytes;
 164                         ilf->ilf_size++;
 165                 } else {
 166                         iip->ili_fields &= ~XFS_ILOG_DEXT;
 167                 }
 168                 break;
 169         case XFS_DINODE_FMT_BTREE:
 170                 iip->ili_fields &=
 171                         ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
 172
 173                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 174                     ip->i_df.if_broot_bytes > 0) {
 175                         ASSERT(ip->i_df.if_broot != NULL);
 176                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
 177                                         ip->i_df.if_broot,
 178                                         ip->i_df.if_broot_bytes);
 179                         ilf->ilf_dsize = ip->i_df.if_broot_bytes;
 180                         ilf->ilf_size++;
 181                 } else {
 182                         ASSERT(!(iip->ili_fields &
 183                                  XFS_ILOG_DBROOT));
 184                         iip->ili_fields &= ~XFS_ILOG_DBROOT;
 185                 }
 186                 break;
 187         case XFS_DINODE_FMT_LOCAL:
 188                 iip->ili_fields &=
 189                         ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
 190                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
 191                     ip->i_df.if_bytes > 0) {
 192                         /*
 193                          * Round i_bytes up to a word boundary.
 194                          * The underlying memory is guaranteed
 195                          * to be there by xfs_idata_realloc().
 196                          */
 197                         data_bytes = roundup(ip->i_df.if_bytes, 4);
 198                         ASSERT(ip->i_df.if_u1.if_data != NULL);
 199                         ASSERT(ip->i_d.di_size > 0);
 200                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
 201                                         ip->i_df.if_u1.if_data, data_bytes);
 202                         ilf->ilf_dsize = (unsigned)data_bytes;
 203                         ilf->ilf_size++;
 204                 } else {
 205                         iip->ili_fields &= ~XFS_ILOG_DDATA;
 206                 }
 207                 break;
 208         case XFS_DINODE_FMT_DEV:
 209                 iip->ili_fields &=
 210                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
 211                 if (iip->ili_fields & XFS_ILOG_DEV)
 212                         ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
 213                 break;
 214         default:
 215                 ASSERT(0);
 216                 break;
 217         }
 218 }
 219
 220 STATIC void
 221 xfs_inode_item_format_attr_fork(
 222         struct xfs_inode_log_item *iip,
 223         struct xfs_inode_log_format *ilf,
 224         struct xfs_log_vec      *lv,
 225         struct xfs_log_iovec    **vecp)
 226 {
 227         struct xfs_inode        *ip = iip->ili_inode;
 228         size_t                  data_bytes;
 229
 230         switch (ip->i_afp->if_format) {
 231         case XFS_DINODE_FMT_EXTENTS:
 232                 iip->ili_fields &=
 233                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
 234
 235                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
 236                     ip->i_afp->if_nextents > 0 &&
 237                     ip->i_afp->if_bytes > 0) {
 238                         struct xfs_bmbt_rec *p;
 239
 240                         ASSERT(xfs_iext_count(ip->i_afp) ==
 241                                 ip->i_afp->if_nextents);
 242
 243                         p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
 244                         data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
 245                         xlog_finish_iovec(lv, *vecp, data_bytes);
 246
 247                         ilf->ilf_asize = data_bytes;
 248                         ilf->ilf_size++;
 249                 } else {
 250                         iip->ili_fields &= ~XFS_ILOG_AEXT;
 251                 }
 252                 break;
 253         case XFS_DINODE_FMT_BTREE:
 254                 iip->ili_fields &=
 255                         ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
 256
 257                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
 258                     ip->i_afp->if_broot_bytes > 0) {
 259                         ASSERT(ip->i_afp->if_broot != NULL);
 260
 261                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
 262                                         ip->i_afp->if_broot,
 263                                         ip->i_afp->if_broot_bytes);
 264                         ilf->ilf_asize = ip->i_afp->if_broot_bytes;
 265                         ilf->ilf_size++;
 266                 } else {
 267                         iip->ili_fields &= ~XFS_ILOG_ABROOT;
 268                 }
 269                 break;
 270         case XFS_DINODE_FMT_LOCAL:
 271                 iip->ili_fields &=
 272                         ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
 273
 274                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
 275                     ip->i_afp->if_bytes > 0) {
 276                         /*
 277                          * Round i_bytes up to a word boundary.
 278                          * The underlying memory is guaranteed
 279                          * to be there by xfs_idata_realloc().
 280                          */
 281                         data_bytes = roundup(ip->i_afp->if_bytes, 4);
 282                         ASSERT(ip->i_afp->if_u1.if_data != NULL);
 283                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
 284                                         ip->i_afp->if_u1.if_data,
 285                                         data_bytes);
 286                         ilf->ilf_asize = (unsigned)data_bytes;
 287                         ilf->ilf_size++;
 288                 } else {
 289                         iip->ili_fields &= ~XFS_ILOG_ADATA;
 290                 }
 291                 break;
 292         default:
 293                 ASSERT(0);
 294                 break;
 295         }
 296 }
 297
 298 /*
 299  * Convert an incore timestamp to a log timestamp.  Note that the log format
 300  * specifies host endian format!
 301  */
 302 static inline xfs_ictimestamp_t
 303 xfs_inode_to_log_dinode_ts(
 304         struct xfs_inode                *ip,
 305         const struct timespec64         tv)
 306 {
 307         struct xfs_legacy_ictimestamp   *lits;
 308         xfs_ictimestamp_t               its;
 309
 310         if (xfs_inode_has_bigtime(ip))
 311                 return xfs_inode_encode_bigtime(tv);
 312
 313         lits = (struct xfs_legacy_ictimestamp *)&its;
 314         lits->t_sec = tv.tv_sec;
 315         lits->t_nsec = tv.tv_nsec;
 316
 317         return its;
 318 }
 319
 320 static void
 321 xfs_inode_to_log_dinode(
 322         struct xfs_inode        *ip,
 323         struct xfs_log_dinode   *to,
 324         xfs_lsn_t               lsn)
 325 {
 326         struct xfs_icdinode     *from = &ip->i_d;
 327         struct inode            *inode = VFS_I(ip);
 328
 329         to->di_magic = XFS_DINODE_MAGIC;
 330         to->di_format = xfs_ifork_format(&ip->i_df);
 331         to->di_uid = i_uid_read(inode);
 332         to->di_gid = i_gid_read(inode);
 333         to->di_projid_lo = from->di_projid & 0xffff;
 334         to->di_projid_hi = from->di_projid >> 16;
 335
 336         memset(to->di_pad, 0, sizeof(to->di_pad));
 337         memset(to->di_pad3, 0, sizeof(to->di_pad3));
 338         to->di_atime = xfs_inode_to_log_dinode_ts(ip, inode->i_atime);
 339         to->di_mtime = xfs_inode_to_log_dinode_ts(ip, inode->i_mtime);
 340         to->di_ctime = xfs_inode_to_log_dinode_ts(ip, inode->i_ctime);
 341         to->di_nlink = inode->i_nlink;
 342         to->di_gen = inode->i_generation;
 343         to->di_mode = inode->i_mode;
 344
 345         to->di_size = from->di_size;
 346         to->di_nblocks = from->di_nblocks;
 347         to->di_extsize = from->di_extsize;
 348         to->di_nextents = xfs_ifork_nextents(&ip->i_df);
 349         to->di_anextents = xfs_ifork_nextents(ip->i_afp);
 350         to->di_forkoff = from->di_forkoff;
 351         to->di_aformat = xfs_ifork_format(ip->i_afp);
 352         to->di_dmevmask = from->di_dmevmask;
 353         to->di_dmstate = from->di_dmstate;
 354         to->di_flags = from->di_flags;
 355
 356         /* log a dummy value to ensure log structure is fully initialised */
 357         to->di_next_unlinked = NULLAGINO;
 358
 359         if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
 360                 to->di_version = 3;
 361                 to->di_changecount = inode_peek_iversion(inode);
 362                 to->di_crtime = xfs_inode_to_log_dinode_ts(ip, from->di_crtime);
 363                 to->di_flags2 = from->di_flags2;
 364                 to->di_cowextsize = from->di_cowextsize;
 365                 to->di_ino = ip->i_ino;
 366                 to->di_lsn = lsn;
 367                 memset(to->di_pad2, 0, sizeof(to->di_pad2));
 368                 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
 369                 to->di_flushiter = 0;
 370         } else {
 371                 to->di_version = 2;
 372                 to->di_flushiter = from->di_flushiter;
 373         }
 374 }
 375
 376 /*
 377  * Format the inode core. Current timestamp data is only in the VFS inode
 378  * fields, so we need to grab them from there. Hence rather than just copying
 379  * the XFS inode core structure, format the fields directly into the iovec.
 380  */
 381 static void
 382 xfs_inode_item_format_core(
 383         struct xfs_inode        *ip,
 384         struct xfs_log_vec      *lv,
 385         struct xfs_log_iovec    **vecp)
 386 {
 387         struct xfs_log_dinode   *dic;
 388
 389         dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
 390         xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
 391         xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));
 392 }
 393
 394 /*
 395  * This is called to fill in the vector of log iovecs for the given inode
 396  * log item.  It fills the first item with an inode log format structure,
 397  * the second with the on-disk inode structure, and a possible third and/or
 398  * fourth with the inode data/extents/b-tree root and inode attributes
 399  * data/extents/b-tree root.
 400  *
 401  * Note: Always use the 64 bit inode log format structure so we don't
 402  * leave an uninitialised hole in the format item on 64 bit systems. Log
 403  * recovery on 32 bit systems handles this just fine, so there's no reason
 404  * for not using an initialising the properly padded structure all the time.
 405  */
 406 STATIC void
 407 xfs_inode_item_format(
 408         struct xfs_log_item     *lip,
 409         struct xfs_log_vec      *lv)
 410 {
 411         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 412         struct xfs_inode        *ip = iip->ili_inode;
 413         struct xfs_log_iovec    *vecp = NULL;
 414         struct xfs_inode_log_format *ilf;
 415
 416         ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
 417         ilf->ilf_type = XFS_LI_INODE;
 418         ilf->ilf_ino = ip->i_ino;
 419         ilf->ilf_blkno = ip->i_imap.im_blkno;
 420         ilf->ilf_len = ip->i_imap.im_len;
 421         ilf->ilf_boffset = ip->i_imap.im_boffset;
 422         ilf->ilf_fields = XFS_ILOG_CORE;
 423         ilf->ilf_size = 2; /* format + core */
 424
 425         /*
 426          * make sure we don't leak uninitialised data into the log in the case
 427          * when we don't log every field in the inode.
 428          */
 429         ilf->ilf_dsize = 0;
 430         ilf->ilf_asize = 0;
 431         ilf->ilf_pad = 0;
 432         memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
 433
 434         xlog_finish_iovec(lv, vecp, sizeof(*ilf));
 435
 436         xfs_inode_item_format_core(ip, lv, &vecp);
 437         xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
 438         if (XFS_IFORK_Q(ip)) {
 439                 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
 440         } else {
 441                 iip->ili_fields &=
 442                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
 443         }
 444
 445         /* update the format with the exact fields we actually logged */
 446         ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
 447 }
 448
 449 /*
 450  * This is called to pin the inode associated with the inode log
 451  * item in memory so it cannot be written out.
 452  */
 453 STATIC void
 454 xfs_inode_item_pin(
 455         struct xfs_log_item     *lip)
 456 {
 457         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 458
 459         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 460         ASSERT(lip->li_buf);
 461
 462         trace_xfs_inode_pin(ip, _RET_IP_);
 463         atomic_inc(&ip->i_pincount);
 464 }
 465
 466
 467 /*
 468  * This is called to unpin the inode associated with the inode log
 469  * item which was previously pinned with a call to xfs_inode_item_pin().
 470  *
 471  * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
 472  *
 473  * Note that unpin can race with inode cluster buffer freeing marking the buffer
 474  * stale. In that case, flush completions are run from the buffer unpin call,
 475  * which may happen before the inode is unpinned. If we lose the race, there
 476  * will be no buffer attached to the log item, but the inode will be marked
 477  * XFS_ISTALE.
 478  */
 479 STATIC void
 480 xfs_inode_item_unpin(
 481         struct xfs_log_item     *lip,
 482         int                     remove)
 483 {
 484         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 485
 486         trace_xfs_inode_unpin(ip, _RET_IP_);
 487         ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));
 488         ASSERT(atomic_read(&ip->i_pincount) > 0);
 489         if (atomic_dec_and_test(&ip->i_pincount))
 490                 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
 491 }
 492
 493 STATIC uint
 494 xfs_inode_item_push(
 495         struct xfs_log_item     *lip,
 496         struct list_head        *buffer_list)
 497                 __releases(&lip->li_ailp->ail_lock)
 498                 __acquires(&lip->li_ailp->ail_lock)
 499 {
 500         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 501         struct xfs_inode        *ip = iip->ili_inode;
 502         struct xfs_buf          *bp = lip->li_buf;
 503         uint                    rval = XFS_ITEM_SUCCESS;
 504         int                     error;
 505
 506         ASSERT(iip->ili_item.li_buf);
 507
 508         if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp) ||
 509             (ip->i_flags & XFS_ISTALE))
 510                 return XFS_ITEM_PINNED;
 511
 512         if (xfs_iflags_test(ip, XFS_IFLUSHING))
 513                 return XFS_ITEM_FLUSHING;
 514
 515         if (!xfs_buf_trylock(bp))
 516                 return XFS_ITEM_LOCKED;
 517
 518         spin_unlock(&lip->li_ailp->ail_lock);
 519
 520         /*
 521          * We need to hold a reference for flushing the cluster buffer as it may
 522          * fail the buffer without IO submission. In which case, we better get a
 523          * reference for that completion because otherwise we don't get a
 524          * reference for IO until we queue the buffer for delwri submission.
 525          */
 526         xfs_buf_hold(bp);
 527         error = xfs_iflush_cluster(bp);
 528         if (!error) {
 529                 if (!xfs_buf_delwri_queue(bp, buffer_list))
 530                         rval = XFS_ITEM_FLUSHING;
 531                 xfs_buf_relse(bp);
 532         } else {
 533                 /*
 534                  * Release the buffer if we were unable to flush anything. On
 535                  * any other error, the buffer has already been released.
 536                  */
 537                 if (error == -EAGAIN)
 538                         xfs_buf_relse(bp);
 539                 rval = XFS_ITEM_LOCKED;
 540         }
 541
 542         spin_lock(&lip->li_ailp->ail_lock);
 543         return rval;
 544 }
 545
 546 /*
 547  * Unlock the inode associated with the inode log item.
 548  */
 549 STATIC void
 550 xfs_inode_item_release(
 551         struct xfs_log_item     *lip)
 552 {
 553         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 554         struct xfs_inode        *ip = iip->ili_inode;
 555         unsigned short          lock_flags;
 556
 557         ASSERT(ip->i_itemp != NULL);
 558         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 559
 560         lock_flags = iip->ili_lock_flags;
 561         iip->ili_lock_flags = 0;
 562         if (lock_flags)
 563                 xfs_iunlock(ip, lock_flags);
 564 }
 565
 566 /*
 567  * This is called to find out where the oldest active copy of the inode log
 568  * item in the on disk log resides now that the last log write of it completed
 569  * at the given lsn.  Since we always re-log all dirty data in an inode, the
 570  * latest copy in the on disk log is the only one that matters.  Therefore,
 571  * simply return the given lsn.
 572  *
 573  * If the inode has been marked stale because the cluster is being freed, we
 574  * don't want to (re-)insert this inode into the AIL. There is a race condition
 575  * where the cluster buffer may be unpinned before the inode is inserted into
 576  * the AIL during transaction committed processing. If the buffer is unpinned
 577  * before the inode item has been committed and inserted, then it is possible
 578  * for the buffer to be written and IO completes before the inode is inserted
 579  * into the AIL. In that case, we'd be inserting a clean, stale inode into the
 580  * AIL which will never get removed. It will, however, get reclaimed which
 581  * triggers an assert in xfs_inode_free() complaining about freein an inode
 582  * still in the AIL.
 583  *
 584  * To avoid this, just unpin the inode directly and return a LSN of -1 so the
 585  * transaction committed code knows that it does not need to do any further
 586  * processing on the item.
 587  */
 588 STATIC xfs_lsn_t
 589 xfs_inode_item_committed(
 590         struct xfs_log_item     *lip,
 591         xfs_lsn_t               lsn)
 592 {
 593         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 594         struct xfs_inode        *ip = iip->ili_inode;
 595
 596         if (xfs_iflags_test(ip, XFS_ISTALE)) {
 597                 xfs_inode_item_unpin(lip, 0);
 598                 return -1;
 599         }
 600         return lsn;
 601 }
 602
 603 STATIC void
 604 xfs_inode_item_committing(
 605         struct xfs_log_item     *lip,
 606         xfs_lsn_t               commit_lsn)
 607 {
 608         INODE_ITEM(lip)->ili_last_lsn = commit_lsn;
 609         return xfs_inode_item_release(lip);
 610 }
 611
 612 static const struct xfs_item_ops xfs_inode_item_ops = {
 613         .iop_size       = xfs_inode_item_size,
 614         .iop_format     = xfs_inode_item_format,
 615         .iop_pin        = xfs_inode_item_pin,
 616         .iop_unpin      = xfs_inode_item_unpin,
 617         .iop_release    = xfs_inode_item_release,
 618         .iop_committed  = xfs_inode_item_committed,
 619         .iop_push       = xfs_inode_item_push,
 620         .iop_committing = xfs_inode_item_committing,
 621 };
 622
 623
 624 /*
 625  * Initialize the inode log item for a newly allocated (in-core) inode.
 626  */
 627 void
 628 xfs_inode_item_init(
 629         struct xfs_inode        *ip,
 630         struct xfs_mount        *mp)
 631 {
 632         struct xfs_inode_log_item *iip;
 633
 634         ASSERT(ip->i_itemp == NULL);
 635         iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_zone,
 636                                               GFP_KERNEL | __GFP_NOFAIL);
 637
 638         iip->ili_inode = ip;
 639         spin_lock_init(&iip->ili_lock);
 640         xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
 641                                                 &xfs_inode_item_ops);
 642 }
 643
 644 /*
 645  * Free the inode log item and any memory hanging off of it.
 646  */
 647 void
 648 xfs_inode_item_destroy(
 649         struct xfs_inode        *ip)
 650 {
 651         struct xfs_inode_log_item *iip = ip->i_itemp;
 652
 653         ASSERT(iip->ili_item.li_buf == NULL);
 654
 655         ip->i_itemp = NULL;
 656         kmem_free(iip->ili_item.li_lv_shadow);
 657         kmem_cache_free(xfs_ili_zone, iip);
 658 }
 659
 660
 661 /*
 662  * We only want to pull the item from the AIL if it is actually there
 663  * and its location in the log has not changed since we started the
 664  * flush.  Thus, we only bother if the inode's lsn has not changed.
 665  */
 666 static void
 667 xfs_iflush_ail_updates(
 668         struct xfs_ail          *ailp,
 669         struct list_head        *list)
 670 {
 671         struct xfs_log_item     *lip;
 672         xfs_lsn_t               tail_lsn = 0;
 673
 674         /* this is an opencoded batch version of xfs_trans_ail_delete */
 675         spin_lock(&ailp->ail_lock);
 676         list_for_each_entry(lip, list, li_bio_list) {
 677                 xfs_lsn_t       lsn;
 678
 679                 clear_bit(XFS_LI_FAILED, &lip->li_flags);
 680                 if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)
 681                         continue;
 682
 683                 lsn = xfs_ail_delete_one(ailp, lip);
 684                 if (!tail_lsn && lsn)
 685                         tail_lsn = lsn;
 686         }
 687         xfs_ail_update_finish(ailp, tail_lsn);
 688 }
 689
 690 /*
 691  * Walk the list of inodes that have completed their IOs. If they are clean
 692  * remove them from the list and dissociate them from the buffer. Buffers that
 693  * are still dirty remain linked to the buffer and on the list. Caller must
 694  * handle them appropriately.
 695  */
 696 static void
 697 xfs_iflush_finish(
 698         struct xfs_buf          *bp,
 699         struct list_head        *list)
 700 {
 701         struct xfs_log_item     *lip, *n;
 702
 703         list_for_each_entry_safe(lip, n, list, li_bio_list) {
 704                 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 705                 bool    drop_buffer = false;
 706
 707                 spin_lock(&iip->ili_lock);
 708
 709                 /*
 710                  * Remove the reference to the cluster buffer if the inode is
 711                  * clean in memory and drop the buffer reference once we've
 712                  * dropped the locks we hold.
 713                  */
 714                 ASSERT(iip->ili_item.li_buf == bp);
 715                 if (!iip->ili_fields) {
 716                         iip->ili_item.li_buf = NULL;
 717                         list_del_init(&lip->li_bio_list);
 718                         drop_buffer = true;
 719                 }
 720                 iip->ili_last_fields = 0;
 721                 iip->ili_flush_lsn = 0;
 722                 spin_unlock(&iip->ili_lock);
 723                 xfs_iflags_clear(iip->ili_inode, XFS_IFLUSHING);
 724                 if (drop_buffer)
 725                         xfs_buf_rele(bp);
 726         }
 727 }
 728
 729 /*
 730  * Inode buffer IO completion routine.  It is responsible for removing inodes
 731  * attached to the buffer from the AIL if they have not been re-logged and
 732  * completing the inode flush.
 733  */
 734 void
 735 xfs_buf_inode_iodone(
 736         struct xfs_buf          *bp)
 737 {
 738         struct xfs_log_item     *lip, *n;
 739         LIST_HEAD(flushed_inodes);
 740         LIST_HEAD(ail_updates);
 741
 742         /*
 743          * Pull the attached inodes from the buffer one at a time and take the
 744          * appropriate action on them.
 745          */
 746         list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
 747                 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 748
 749                 if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {
 750                         xfs_iflush_abort(iip->ili_inode);
 751                         continue;
 752                 }
 753                 if (!iip->ili_last_fields)
 754                         continue;
 755
 756                 /* Do an unlocked check for needing the AIL lock. */
 757                 if (iip->ili_flush_lsn == lip->li_lsn ||
 758                     test_bit(XFS_LI_FAILED, &lip->li_flags))
 759                         list_move_tail(&lip->li_bio_list, &ail_updates);
 760                 else
 761                         list_move_tail(&lip->li_bio_list, &flushed_inodes);
 762         }
 763
 764         if (!list_empty(&ail_updates)) {
 765                 xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);
 766                 list_splice_tail(&ail_updates, &flushed_inodes);
 767         }
 768
 769         xfs_iflush_finish(bp, &flushed_inodes);
 770         if (!list_empty(&flushed_inodes))
 771                 list_splice_tail(&flushed_inodes, &bp->b_li_list);
 772 }
 773
 774 void
 775 xfs_buf_inode_io_fail(
 776         struct xfs_buf          *bp)
 777 {
 778         struct xfs_log_item     *lip;
 779
 780         list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
 781                 set_bit(XFS_LI_FAILED, &lip->li_flags);
 782 }
 783
 784 /*
 785  * This is the inode flushing abort routine.  It is called when
 786  * the filesystem is shutting down to clean up the inode state.  It is
 787  * responsible for removing the inode item from the AIL if it has not been
 788  * re-logged and clearing the inode's flush state.
 789  */
 790 void
 791 xfs_iflush_abort(
 792         struct xfs_inode        *ip)
 793 {
 794         struct xfs_inode_log_item *iip = ip->i_itemp;
 795         struct xfs_buf          *bp = NULL;
 796
 797         if (iip) {
 798                 /*
 799                  * Clear the failed bit before removing the item from the AIL so
 800                  * xfs_trans_ail_delete() doesn't try to clear and release the
 801                  * buffer attached to the log item before we are done with it.
 802                  */
 803                 clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);
 804                 xfs_trans_ail_delete(&iip->ili_item, 0);
 805
 806                 /*
 807                  * Clear the inode logging fields so no more flushes are
 808                  * attempted.
 809                  */
 810                 spin_lock(&iip->ili_lock);
 811                 iip->ili_last_fields = 0;
 812                 iip->ili_fields = 0;
 813                 iip->ili_fsync_fields = 0;
 814                 iip->ili_flush_lsn = 0;
 815                 bp = iip->ili_item.li_buf;
 816                 iip->ili_item.li_buf = NULL;
 817                 list_del_init(&iip->ili_item.li_bio_list);
 818                 spin_unlock(&iip->ili_lock);
 819         }
 820         xfs_iflags_clear(ip, XFS_IFLUSHING);
 821         if (bp)
 822                 xfs_buf_rele(bp);
 823 }
 824
 825 /*
 826  * convert an xfs_inode_log_format struct from the old 32 bit version
 827  * (which can have different field alignments) to the native 64 bit version
 828  */
 829 int
 830 xfs_inode_item_format_convert(
 831         struct xfs_log_iovec            *buf,
 832         struct xfs_inode_log_format     *in_f)
 833 {
 834         struct xfs_inode_log_format_32  *in_f32 = buf->i_addr;
 835
 836         if (buf->i_len != sizeof(*in_f32)) {
 837                 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
 838                 return -EFSCORRUPTED;
 839         }
 840
 841         in_f->ilf_type = in_f32->ilf_type;
 842         in_f->ilf_size = in_f32->ilf_size;
 843         in_f->ilf_fields = in_f32->ilf_fields;
 844         in_f->ilf_asize = in_f32->ilf_asize;
 845         in_f->ilf_dsize = in_f32->ilf_dsize;
 846         in_f->ilf_ino = in_f32->ilf_ino;
 847         memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
 848         in_f->ilf_blkno = in_f32->ilf_blkno;
 849         in_f->ilf_len = in_f32->ilf_len;
 850         in_f->ilf_boffset = in_f32->ilf_boffset;
 851         return 0;
 852 }