Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
0004  * All Rights Reserved.
0005  */
0006 #include "xfs.h"
0007 #include "xfs_fs.h"
0008 #include "xfs_shared.h"
0009 #include "xfs_format.h"
0010 #include "xfs_log_format.h"
0011 #include "xfs_trans_resv.h"
0012 #include "xfs_mount.h"
0013 #include "xfs_inode.h"
0014 #include "xfs_trans.h"
0015 #include "xfs_inode_item.h"
0016 #include "xfs_trace.h"
0017 #include "xfs_trans_priv.h"
0018 #include "xfs_buf_item.h"
0019 #include "xfs_log.h"
0020 #include "xfs_log_priv.h"
0021 #include "xfs_error.h"
0022 
0023 #include <linux/iversion.h>
0024 
0025 struct kmem_cache   *xfs_ili_cache;     /* inode log item */
0026 
0027 static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
0028 {
0029     return container_of(lip, struct xfs_inode_log_item, ili_item);
0030 }
0031 
0032 /*
0033  * The logged size of an inode fork is always the current size of the inode
0034  * fork. This means that when an inode fork is relogged, the size of the logged
0035  * region is determined by the current state, not the combination of the
0036  * previously logged state + the current state. This is different relogging
0037  * behaviour to most other log items which will retain the size of the
0038  * previously logged changes when smaller regions are relogged.
0039  *
0040  * Hence operations that remove data from the inode fork (e.g. shortform
0041  * dir/attr remove, extent form extent removal, etc), the size of the relogged
0042  * inode gets -smaller- rather than stays the same size as the previously logged
0043  * size and this can result in the committing transaction reducing the amount of
0044  * space being consumed by the CIL.
0045  */
0046 STATIC void
0047 xfs_inode_item_data_fork_size(
0048     struct xfs_inode_log_item *iip,
0049     int         *nvecs,
0050     int         *nbytes)
0051 {
0052     struct xfs_inode    *ip = iip->ili_inode;
0053 
0054     switch (ip->i_df.if_format) {
0055     case XFS_DINODE_FMT_EXTENTS:
0056         if ((iip->ili_fields & XFS_ILOG_DEXT) &&
0057             ip->i_df.if_nextents > 0 &&
0058             ip->i_df.if_bytes > 0) {
0059             /* worst case, doesn't subtract delalloc extents */
0060             *nbytes += xfs_inode_data_fork_size(ip);
0061             *nvecs += 1;
0062         }
0063         break;
0064     case XFS_DINODE_FMT_BTREE:
0065         if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
0066             ip->i_df.if_broot_bytes > 0) {
0067             *nbytes += ip->i_df.if_broot_bytes;
0068             *nvecs += 1;
0069         }
0070         break;
0071     case XFS_DINODE_FMT_LOCAL:
0072         if ((iip->ili_fields & XFS_ILOG_DDATA) &&
0073             ip->i_df.if_bytes > 0) {
0074             *nbytes += xlog_calc_iovec_len(ip->i_df.if_bytes);
0075             *nvecs += 1;
0076         }
0077         break;
0078 
0079     case XFS_DINODE_FMT_DEV:
0080         break;
0081     default:
0082         ASSERT(0);
0083         break;
0084     }
0085 }
0086 
0087 STATIC void
0088 xfs_inode_item_attr_fork_size(
0089     struct xfs_inode_log_item *iip,
0090     int         *nvecs,
0091     int         *nbytes)
0092 {
0093     struct xfs_inode    *ip = iip->ili_inode;
0094 
0095     switch (ip->i_af.if_format) {
0096     case XFS_DINODE_FMT_EXTENTS:
0097         if ((iip->ili_fields & XFS_ILOG_AEXT) &&
0098             ip->i_af.if_nextents > 0 &&
0099             ip->i_af.if_bytes > 0) {
0100             /* worst case, doesn't subtract unused space */
0101             *nbytes += xfs_inode_attr_fork_size(ip);
0102             *nvecs += 1;
0103         }
0104         break;
0105     case XFS_DINODE_FMT_BTREE:
0106         if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
0107             ip->i_af.if_broot_bytes > 0) {
0108             *nbytes += ip->i_af.if_broot_bytes;
0109             *nvecs += 1;
0110         }
0111         break;
0112     case XFS_DINODE_FMT_LOCAL:
0113         if ((iip->ili_fields & XFS_ILOG_ADATA) &&
0114             ip->i_af.if_bytes > 0) {
0115             *nbytes += xlog_calc_iovec_len(ip->i_af.if_bytes);
0116             *nvecs += 1;
0117         }
0118         break;
0119     default:
0120         ASSERT(0);
0121         break;
0122     }
0123 }
0124 
0125 /*
0126  * This returns the number of iovecs needed to log the given inode item.
0127  *
0128  * We need one iovec for the inode log format structure, one for the
0129  * inode core, and possibly one for the inode data/extents/b-tree root
0130  * and one for the inode attribute data/extents/b-tree root.
0131  */
0132 STATIC void
0133 xfs_inode_item_size(
0134     struct xfs_log_item *lip,
0135     int         *nvecs,
0136     int         *nbytes)
0137 {
0138     struct xfs_inode_log_item *iip = INODE_ITEM(lip);
0139     struct xfs_inode    *ip = iip->ili_inode;
0140 
0141     *nvecs += 2;
0142     *nbytes += sizeof(struct xfs_inode_log_format) +
0143            xfs_log_dinode_size(ip->i_mount);
0144 
0145     xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
0146     if (xfs_inode_has_attr_fork(ip))
0147         xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
0148 }
0149 
0150 STATIC void
0151 xfs_inode_item_format_data_fork(
0152     struct xfs_inode_log_item *iip,
0153     struct xfs_inode_log_format *ilf,
0154     struct xfs_log_vec  *lv,
0155     struct xfs_log_iovec    **vecp)
0156 {
0157     struct xfs_inode    *ip = iip->ili_inode;
0158     size_t          data_bytes;
0159 
0160     switch (ip->i_df.if_format) {
0161     case XFS_DINODE_FMT_EXTENTS:
0162         iip->ili_fields &=
0163             ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
0164 
0165         if ((iip->ili_fields & XFS_ILOG_DEXT) &&
0166             ip->i_df.if_nextents > 0 &&
0167             ip->i_df.if_bytes > 0) {
0168             struct xfs_bmbt_rec *p;
0169 
0170             ASSERT(xfs_iext_count(&ip->i_df) > 0);
0171 
0172             p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
0173             data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
0174             xlog_finish_iovec(lv, *vecp, data_bytes);
0175 
0176             ASSERT(data_bytes <= ip->i_df.if_bytes);
0177 
0178             ilf->ilf_dsize = data_bytes;
0179             ilf->ilf_size++;
0180         } else {
0181             iip->ili_fields &= ~XFS_ILOG_DEXT;
0182         }
0183         break;
0184     case XFS_DINODE_FMT_BTREE:
0185         iip->ili_fields &=
0186             ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
0187 
0188         if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
0189             ip->i_df.if_broot_bytes > 0) {
0190             ASSERT(ip->i_df.if_broot != NULL);
0191             xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
0192                     ip->i_df.if_broot,
0193                     ip->i_df.if_broot_bytes);
0194             ilf->ilf_dsize = ip->i_df.if_broot_bytes;
0195             ilf->ilf_size++;
0196         } else {
0197             ASSERT(!(iip->ili_fields &
0198                  XFS_ILOG_DBROOT));
0199             iip->ili_fields &= ~XFS_ILOG_DBROOT;
0200         }
0201         break;
0202     case XFS_DINODE_FMT_LOCAL:
0203         iip->ili_fields &=
0204             ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
0205         if ((iip->ili_fields & XFS_ILOG_DDATA) &&
0206             ip->i_df.if_bytes > 0) {
0207             ASSERT(ip->i_df.if_u1.if_data != NULL);
0208             ASSERT(ip->i_disk_size > 0);
0209             xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
0210                     ip->i_df.if_u1.if_data,
0211                     ip->i_df.if_bytes);
0212             ilf->ilf_dsize = (unsigned)ip->i_df.if_bytes;
0213             ilf->ilf_size++;
0214         } else {
0215             iip->ili_fields &= ~XFS_ILOG_DDATA;
0216         }
0217         break;
0218     case XFS_DINODE_FMT_DEV:
0219         iip->ili_fields &=
0220             ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
0221         if (iip->ili_fields & XFS_ILOG_DEV)
0222             ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
0223         break;
0224     default:
0225         ASSERT(0);
0226         break;
0227     }
0228 }
0229 
0230 STATIC void
0231 xfs_inode_item_format_attr_fork(
0232     struct xfs_inode_log_item *iip,
0233     struct xfs_inode_log_format *ilf,
0234     struct xfs_log_vec  *lv,
0235     struct xfs_log_iovec    **vecp)
0236 {
0237     struct xfs_inode    *ip = iip->ili_inode;
0238     size_t          data_bytes;
0239 
0240     switch (ip->i_af.if_format) {
0241     case XFS_DINODE_FMT_EXTENTS:
0242         iip->ili_fields &=
0243             ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
0244 
0245         if ((iip->ili_fields & XFS_ILOG_AEXT) &&
0246             ip->i_af.if_nextents > 0 &&
0247             ip->i_af.if_bytes > 0) {
0248             struct xfs_bmbt_rec *p;
0249 
0250             ASSERT(xfs_iext_count(&ip->i_af) ==
0251                 ip->i_af.if_nextents);
0252 
0253             p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
0254             data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
0255             xlog_finish_iovec(lv, *vecp, data_bytes);
0256 
0257             ilf->ilf_asize = data_bytes;
0258             ilf->ilf_size++;
0259         } else {
0260             iip->ili_fields &= ~XFS_ILOG_AEXT;
0261         }
0262         break;
0263     case XFS_DINODE_FMT_BTREE:
0264         iip->ili_fields &=
0265             ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
0266 
0267         if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
0268             ip->i_af.if_broot_bytes > 0) {
0269             ASSERT(ip->i_af.if_broot != NULL);
0270 
0271             xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
0272                     ip->i_af.if_broot,
0273                     ip->i_af.if_broot_bytes);
0274             ilf->ilf_asize = ip->i_af.if_broot_bytes;
0275             ilf->ilf_size++;
0276         } else {
0277             iip->ili_fields &= ~XFS_ILOG_ABROOT;
0278         }
0279         break;
0280     case XFS_DINODE_FMT_LOCAL:
0281         iip->ili_fields &=
0282             ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
0283 
0284         if ((iip->ili_fields & XFS_ILOG_ADATA) &&
0285             ip->i_af.if_bytes > 0) {
0286             ASSERT(ip->i_af.if_u1.if_data != NULL);
0287             xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
0288                     ip->i_af.if_u1.if_data,
0289                     ip->i_af.if_bytes);
0290             ilf->ilf_asize = (unsigned)ip->i_af.if_bytes;
0291             ilf->ilf_size++;
0292         } else {
0293             iip->ili_fields &= ~XFS_ILOG_ADATA;
0294         }
0295         break;
0296     default:
0297         ASSERT(0);
0298         break;
0299     }
0300 }
0301 
0302 /*
0303  * Convert an incore timestamp to a log timestamp.  Note that the log format
0304  * specifies host endian format!
0305  */
0306 static inline xfs_log_timestamp_t
0307 xfs_inode_to_log_dinode_ts(
0308     struct xfs_inode        *ip,
0309     const struct timespec64     tv)
0310 {
0311     struct xfs_log_legacy_timestamp *lits;
0312     xfs_log_timestamp_t     its;
0313 
0314     if (xfs_inode_has_bigtime(ip))
0315         return xfs_inode_encode_bigtime(tv);
0316 
0317     lits = (struct xfs_log_legacy_timestamp *)&its;
0318     lits->t_sec = tv.tv_sec;
0319     lits->t_nsec = tv.tv_nsec;
0320 
0321     return its;
0322 }
0323 
0324 /*
0325  * The legacy DMAPI fields are only present in the on-disk and in-log inodes,
0326  * but not in the in-memory one.  But we are guaranteed to have an inode buffer
0327  * in memory when logging an inode, so we can just copy it from the on-disk
0328  * inode to the in-log inode here so that recovery of file system with these
0329  * fields set to non-zero values doesn't lose them.  For all other cases we zero
0330  * the fields.
0331  */
0332 static void
0333 xfs_copy_dm_fields_to_log_dinode(
0334     struct xfs_inode    *ip,
0335     struct xfs_log_dinode   *to)
0336 {
0337     struct xfs_dinode   *dip;
0338 
0339     dip = xfs_buf_offset(ip->i_itemp->ili_item.li_buf,
0340                  ip->i_imap.im_boffset);
0341 
0342     if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS)) {
0343         to->di_dmevmask = be32_to_cpu(dip->di_dmevmask);
0344         to->di_dmstate = be16_to_cpu(dip->di_dmstate);
0345     } else {
0346         to->di_dmevmask = 0;
0347         to->di_dmstate = 0;
0348     }
0349 }
0350 
0351 static inline void
0352 xfs_inode_to_log_dinode_iext_counters(
0353     struct xfs_inode    *ip,
0354     struct xfs_log_dinode   *to)
0355 {
0356     if (xfs_inode_has_large_extent_counts(ip)) {
0357         to->di_big_nextents = xfs_ifork_nextents(&ip->i_df);
0358         to->di_big_anextents = xfs_ifork_nextents(&ip->i_af);
0359         to->di_nrext64_pad = 0;
0360     } else {
0361         to->di_nextents = xfs_ifork_nextents(&ip->i_df);
0362         to->di_anextents = xfs_ifork_nextents(&ip->i_af);
0363     }
0364 }
0365 
0366 static void
0367 xfs_inode_to_log_dinode(
0368     struct xfs_inode    *ip,
0369     struct xfs_log_dinode   *to,
0370     xfs_lsn_t       lsn)
0371 {
0372     struct inode        *inode = VFS_I(ip);
0373 
0374     to->di_magic = XFS_DINODE_MAGIC;
0375     to->di_format = xfs_ifork_format(&ip->i_df);
0376     to->di_uid = i_uid_read(inode);
0377     to->di_gid = i_gid_read(inode);
0378     to->di_projid_lo = ip->i_projid & 0xffff;
0379     to->di_projid_hi = ip->i_projid >> 16;
0380 
0381     memset(to->di_pad3, 0, sizeof(to->di_pad3));
0382     to->di_atime = xfs_inode_to_log_dinode_ts(ip, inode->i_atime);
0383     to->di_mtime = xfs_inode_to_log_dinode_ts(ip, inode->i_mtime);
0384     to->di_ctime = xfs_inode_to_log_dinode_ts(ip, inode->i_ctime);
0385     to->di_nlink = inode->i_nlink;
0386     to->di_gen = inode->i_generation;
0387     to->di_mode = inode->i_mode;
0388 
0389     to->di_size = ip->i_disk_size;
0390     to->di_nblocks = ip->i_nblocks;
0391     to->di_extsize = ip->i_extsize;
0392     to->di_forkoff = ip->i_forkoff;
0393     to->di_aformat = xfs_ifork_format(&ip->i_af);
0394     to->di_flags = ip->i_diflags;
0395 
0396     xfs_copy_dm_fields_to_log_dinode(ip, to);
0397 
0398     /* log a dummy value to ensure log structure is fully initialised */
0399     to->di_next_unlinked = NULLAGINO;
0400 
0401     if (xfs_has_v3inodes(ip->i_mount)) {
0402         to->di_version = 3;
0403         to->di_changecount = inode_peek_iversion(inode);
0404         to->di_crtime = xfs_inode_to_log_dinode_ts(ip, ip->i_crtime);
0405         to->di_flags2 = ip->i_diflags2;
0406         to->di_cowextsize = ip->i_cowextsize;
0407         to->di_ino = ip->i_ino;
0408         to->di_lsn = lsn;
0409         memset(to->di_pad2, 0, sizeof(to->di_pad2));
0410         uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
0411         to->di_v3_pad = 0;
0412     } else {
0413         to->di_version = 2;
0414         to->di_flushiter = ip->i_flushiter;
0415         memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
0416     }
0417 
0418     xfs_inode_to_log_dinode_iext_counters(ip, to);
0419 }
0420 
0421 /*
0422  * Format the inode core. Current timestamp data is only in the VFS inode
0423  * fields, so we need to grab them from there. Hence rather than just copying
0424  * the XFS inode core structure, format the fields directly into the iovec.
0425  */
0426 static void
0427 xfs_inode_item_format_core(
0428     struct xfs_inode    *ip,
0429     struct xfs_log_vec  *lv,
0430     struct xfs_log_iovec    **vecp)
0431 {
0432     struct xfs_log_dinode   *dic;
0433 
0434     dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
0435     xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
0436     xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));
0437 }
0438 
0439 /*
0440  * This is called to fill in the vector of log iovecs for the given inode
0441  * log item.  It fills the first item with an inode log format structure,
0442  * the second with the on-disk inode structure, and a possible third and/or
0443  * fourth with the inode data/extents/b-tree root and inode attributes
0444  * data/extents/b-tree root.
0445  *
0446  * Note: Always use the 64 bit inode log format structure so we don't
0447  * leave an uninitialised hole in the format item on 64 bit systems. Log
0448  * recovery on 32 bit systems handles this just fine, so there's no reason
0449  * for not using an initialising the properly padded structure all the time.
0450  */
0451 STATIC void
0452 xfs_inode_item_format(
0453     struct xfs_log_item *lip,
0454     struct xfs_log_vec  *lv)
0455 {
0456     struct xfs_inode_log_item *iip = INODE_ITEM(lip);
0457     struct xfs_inode    *ip = iip->ili_inode;
0458     struct xfs_log_iovec    *vecp = NULL;
0459     struct xfs_inode_log_format *ilf;
0460 
0461     ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
0462     ilf->ilf_type = XFS_LI_INODE;
0463     ilf->ilf_ino = ip->i_ino;
0464     ilf->ilf_blkno = ip->i_imap.im_blkno;
0465     ilf->ilf_len = ip->i_imap.im_len;
0466     ilf->ilf_boffset = ip->i_imap.im_boffset;
0467     ilf->ilf_fields = XFS_ILOG_CORE;
0468     ilf->ilf_size = 2; /* format + core */
0469 
0470     /*
0471      * make sure we don't leak uninitialised data into the log in the case
0472      * when we don't log every field in the inode.
0473      */
0474     ilf->ilf_dsize = 0;
0475     ilf->ilf_asize = 0;
0476     ilf->ilf_pad = 0;
0477     memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
0478 
0479     xlog_finish_iovec(lv, vecp, sizeof(*ilf));
0480 
0481     xfs_inode_item_format_core(ip, lv, &vecp);
0482     xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
0483     if (xfs_inode_has_attr_fork(ip)) {
0484         xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
0485     } else {
0486         iip->ili_fields &=
0487             ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
0488     }
0489 
0490     /* update the format with the exact fields we actually logged */
0491     ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
0492 }
0493 
0494 /*
0495  * This is called to pin the inode associated with the inode log
0496  * item in memory so it cannot be written out.
0497  */
0498 STATIC void
0499 xfs_inode_item_pin(
0500     struct xfs_log_item *lip)
0501 {
0502     struct xfs_inode    *ip = INODE_ITEM(lip)->ili_inode;
0503 
0504     ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
0505     ASSERT(lip->li_buf);
0506 
0507     trace_xfs_inode_pin(ip, _RET_IP_);
0508     atomic_inc(&ip->i_pincount);
0509 }
0510 
0511 
0512 /*
0513  * This is called to unpin the inode associated with the inode log
0514  * item which was previously pinned with a call to xfs_inode_item_pin().
0515  *
0516  * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
0517  *
0518  * Note that unpin can race with inode cluster buffer freeing marking the buffer
0519  * stale. In that case, flush completions are run from the buffer unpin call,
0520  * which may happen before the inode is unpinned. If we lose the race, there
0521  * will be no buffer attached to the log item, but the inode will be marked
0522  * XFS_ISTALE.
0523  */
0524 STATIC void
0525 xfs_inode_item_unpin(
0526     struct xfs_log_item *lip,
0527     int         remove)
0528 {
0529     struct xfs_inode    *ip = INODE_ITEM(lip)->ili_inode;
0530 
0531     trace_xfs_inode_unpin(ip, _RET_IP_);
0532     ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));
0533     ASSERT(atomic_read(&ip->i_pincount) > 0);
0534     if (atomic_dec_and_test(&ip->i_pincount))
0535         wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
0536 }
0537 
0538 STATIC uint
0539 xfs_inode_item_push(
0540     struct xfs_log_item *lip,
0541     struct list_head    *buffer_list)
0542         __releases(&lip->li_ailp->ail_lock)
0543         __acquires(&lip->li_ailp->ail_lock)
0544 {
0545     struct xfs_inode_log_item *iip = INODE_ITEM(lip);
0546     struct xfs_inode    *ip = iip->ili_inode;
0547     struct xfs_buf      *bp = lip->li_buf;
0548     uint            rval = XFS_ITEM_SUCCESS;
0549     int         error;
0550 
0551     if (!bp || (ip->i_flags & XFS_ISTALE)) {
0552         /*
0553          * Inode item/buffer is being being aborted due to cluster
0554          * buffer deletion. Trigger a log force to have that operation
0555          * completed and items removed from the AIL before the next push
0556          * attempt.
0557          */
0558         return XFS_ITEM_PINNED;
0559     }
0560 
0561     if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp))
0562         return XFS_ITEM_PINNED;
0563 
0564     if (xfs_iflags_test(ip, XFS_IFLUSHING))
0565         return XFS_ITEM_FLUSHING;
0566 
0567     if (!xfs_buf_trylock(bp))
0568         return XFS_ITEM_LOCKED;
0569 
0570     spin_unlock(&lip->li_ailp->ail_lock);
0571 
0572     /*
0573      * We need to hold a reference for flushing the cluster buffer as it may
0574      * fail the buffer without IO submission. In which case, we better get a
0575      * reference for that completion because otherwise we don't get a
0576      * reference for IO until we queue the buffer for delwri submission.
0577      */
0578     xfs_buf_hold(bp);
0579     error = xfs_iflush_cluster(bp);
0580     if (!error) {
0581         if (!xfs_buf_delwri_queue(bp, buffer_list))
0582             rval = XFS_ITEM_FLUSHING;
0583         xfs_buf_relse(bp);
0584     } else {
0585         /*
0586          * Release the buffer if we were unable to flush anything. On
0587          * any other error, the buffer has already been released.
0588          */
0589         if (error == -EAGAIN)
0590             xfs_buf_relse(bp);
0591         rval = XFS_ITEM_LOCKED;
0592     }
0593 
0594     spin_lock(&lip->li_ailp->ail_lock);
0595     return rval;
0596 }
0597 
0598 /*
0599  * Unlock the inode associated with the inode log item.
0600  */
0601 STATIC void
0602 xfs_inode_item_release(
0603     struct xfs_log_item *lip)
0604 {
0605     struct xfs_inode_log_item *iip = INODE_ITEM(lip);
0606     struct xfs_inode    *ip = iip->ili_inode;
0607     unsigned short      lock_flags;
0608 
0609     ASSERT(ip->i_itemp != NULL);
0610     ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
0611 
0612     lock_flags = iip->ili_lock_flags;
0613     iip->ili_lock_flags = 0;
0614     if (lock_flags)
0615         xfs_iunlock(ip, lock_flags);
0616 }
0617 
0618 /*
0619  * This is called to find out where the oldest active copy of the inode log
0620  * item in the on disk log resides now that the last log write of it completed
0621  * at the given lsn.  Since we always re-log all dirty data in an inode, the
0622  * latest copy in the on disk log is the only one that matters.  Therefore,
0623  * simply return the given lsn.
0624  *
0625  * If the inode has been marked stale because the cluster is being freed, we
0626  * don't want to (re-)insert this inode into the AIL. There is a race condition
0627  * where the cluster buffer may be unpinned before the inode is inserted into
0628  * the AIL during transaction committed processing. If the buffer is unpinned
0629  * before the inode item has been committed and inserted, then it is possible
0630  * for the buffer to be written and IO completes before the inode is inserted
0631  * into the AIL. In that case, we'd be inserting a clean, stale inode into the
0632  * AIL which will never get removed. It will, however, get reclaimed which
0633  * triggers an assert in xfs_inode_free() complaining about freein an inode
0634  * still in the AIL.
0635  *
0636  * To avoid this, just unpin the inode directly and return a LSN of -1 so the
0637  * transaction committed code knows that it does not need to do any further
0638  * processing on the item.
0639  */
0640 STATIC xfs_lsn_t
0641 xfs_inode_item_committed(
0642     struct xfs_log_item *lip,
0643     xfs_lsn_t       lsn)
0644 {
0645     struct xfs_inode_log_item *iip = INODE_ITEM(lip);
0646     struct xfs_inode    *ip = iip->ili_inode;
0647 
0648     if (xfs_iflags_test(ip, XFS_ISTALE)) {
0649         xfs_inode_item_unpin(lip, 0);
0650         return -1;
0651     }
0652     return lsn;
0653 }
0654 
0655 STATIC void
0656 xfs_inode_item_committing(
0657     struct xfs_log_item *lip,
0658     xfs_csn_t       seq)
0659 {
0660     INODE_ITEM(lip)->ili_commit_seq = seq;
0661     return xfs_inode_item_release(lip);
0662 }
0663 
0664 static const struct xfs_item_ops xfs_inode_item_ops = {
0665     .iop_size   = xfs_inode_item_size,
0666     .iop_format = xfs_inode_item_format,
0667     .iop_pin    = xfs_inode_item_pin,
0668     .iop_unpin  = xfs_inode_item_unpin,
0669     .iop_release    = xfs_inode_item_release,
0670     .iop_committed  = xfs_inode_item_committed,
0671     .iop_push   = xfs_inode_item_push,
0672     .iop_committing = xfs_inode_item_committing,
0673 };
0674 
0675 
0676 /*
0677  * Initialize the inode log item for a newly allocated (in-core) inode.
0678  */
0679 void
0680 xfs_inode_item_init(
0681     struct xfs_inode    *ip,
0682     struct xfs_mount    *mp)
0683 {
0684     struct xfs_inode_log_item *iip;
0685 
0686     ASSERT(ip->i_itemp == NULL);
0687     iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_cache,
0688                           GFP_KERNEL | __GFP_NOFAIL);
0689 
0690     iip->ili_inode = ip;
0691     spin_lock_init(&iip->ili_lock);
0692     xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
0693                         &xfs_inode_item_ops);
0694 }
0695 
0696 /*
0697  * Free the inode log item and any memory hanging off of it.
0698  */
0699 void
0700 xfs_inode_item_destroy(
0701     struct xfs_inode    *ip)
0702 {
0703     struct xfs_inode_log_item *iip = ip->i_itemp;
0704 
0705     ASSERT(iip->ili_item.li_buf == NULL);
0706 
0707     ip->i_itemp = NULL;
0708     kmem_free(iip->ili_item.li_lv_shadow);
0709     kmem_cache_free(xfs_ili_cache, iip);
0710 }
0711 
0712 
0713 /*
0714  * We only want to pull the item from the AIL if it is actually there
0715  * and its location in the log has not changed since we started the
0716  * flush.  Thus, we only bother if the inode's lsn has not changed.
0717  */
0718 static void
0719 xfs_iflush_ail_updates(
0720     struct xfs_ail      *ailp,
0721     struct list_head    *list)
0722 {
0723     struct xfs_log_item *lip;
0724     xfs_lsn_t       tail_lsn = 0;
0725 
0726     /* this is an opencoded batch version of xfs_trans_ail_delete */
0727     spin_lock(&ailp->ail_lock);
0728     list_for_each_entry(lip, list, li_bio_list) {
0729         xfs_lsn_t   lsn;
0730 
0731         clear_bit(XFS_LI_FAILED, &lip->li_flags);
0732         if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)
0733             continue;
0734 
0735         /*
0736          * dgc: Not sure how this happens, but it happens very
0737          * occassionaly via generic/388.  xfs_iflush_abort() also
0738          * silently handles this same "under writeback but not in AIL at
0739          * shutdown" condition via xfs_trans_ail_delete().
0740          */
0741         if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
0742             ASSERT(xlog_is_shutdown(lip->li_log));
0743             continue;
0744         }
0745 
0746         lsn = xfs_ail_delete_one(ailp, lip);
0747         if (!tail_lsn && lsn)
0748             tail_lsn = lsn;
0749     }
0750     xfs_ail_update_finish(ailp, tail_lsn);
0751 }
0752 
0753 /*
0754  * Walk the list of inodes that have completed their IOs. If they are clean
0755  * remove them from the list and dissociate them from the buffer. Buffers that
0756  * are still dirty remain linked to the buffer and on the list. Caller must
0757  * handle them appropriately.
0758  */
0759 static void
0760 xfs_iflush_finish(
0761     struct xfs_buf      *bp,
0762     struct list_head    *list)
0763 {
0764     struct xfs_log_item *lip, *n;
0765 
0766     list_for_each_entry_safe(lip, n, list, li_bio_list) {
0767         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
0768         bool    drop_buffer = false;
0769 
0770         spin_lock(&iip->ili_lock);
0771 
0772         /*
0773          * Remove the reference to the cluster buffer if the inode is
0774          * clean in memory and drop the buffer reference once we've
0775          * dropped the locks we hold.
0776          */
0777         ASSERT(iip->ili_item.li_buf == bp);
0778         if (!iip->ili_fields) {
0779             iip->ili_item.li_buf = NULL;
0780             list_del_init(&lip->li_bio_list);
0781             drop_buffer = true;
0782         }
0783         iip->ili_last_fields = 0;
0784         iip->ili_flush_lsn = 0;
0785         spin_unlock(&iip->ili_lock);
0786         xfs_iflags_clear(iip->ili_inode, XFS_IFLUSHING);
0787         if (drop_buffer)
0788             xfs_buf_rele(bp);
0789     }
0790 }
0791 
0792 /*
0793  * Inode buffer IO completion routine.  It is responsible for removing inodes
0794  * attached to the buffer from the AIL if they have not been re-logged and
0795  * completing the inode flush.
0796  */
0797 void
0798 xfs_buf_inode_iodone(
0799     struct xfs_buf      *bp)
0800 {
0801     struct xfs_log_item *lip, *n;
0802     LIST_HEAD(flushed_inodes);
0803     LIST_HEAD(ail_updates);
0804 
0805     /*
0806      * Pull the attached inodes from the buffer one at a time and take the
0807      * appropriate action on them.
0808      */
0809     list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
0810         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
0811 
0812         if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {
0813             xfs_iflush_abort(iip->ili_inode);
0814             continue;
0815         }
0816         if (!iip->ili_last_fields)
0817             continue;
0818 
0819         /* Do an unlocked check for needing the AIL lock. */
0820         if (iip->ili_flush_lsn == lip->li_lsn ||
0821             test_bit(XFS_LI_FAILED, &lip->li_flags))
0822             list_move_tail(&lip->li_bio_list, &ail_updates);
0823         else
0824             list_move_tail(&lip->li_bio_list, &flushed_inodes);
0825     }
0826 
0827     if (!list_empty(&ail_updates)) {
0828         xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);
0829         list_splice_tail(&ail_updates, &flushed_inodes);
0830     }
0831 
0832     xfs_iflush_finish(bp, &flushed_inodes);
0833     if (!list_empty(&flushed_inodes))
0834         list_splice_tail(&flushed_inodes, &bp->b_li_list);
0835 }
0836 
0837 void
0838 xfs_buf_inode_io_fail(
0839     struct xfs_buf      *bp)
0840 {
0841     struct xfs_log_item *lip;
0842 
0843     list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
0844         set_bit(XFS_LI_FAILED, &lip->li_flags);
0845 }
0846 
0847 /*
0848  * Clear the inode logging fields so no more flushes are attempted.  If we are
0849  * on a buffer list, it is now safe to remove it because the buffer is
0850  * guaranteed to be locked. The caller will drop the reference to the buffer
0851  * the log item held.
0852  */
0853 static void
0854 xfs_iflush_abort_clean(
0855     struct xfs_inode_log_item *iip)
0856 {
0857     iip->ili_last_fields = 0;
0858     iip->ili_fields = 0;
0859     iip->ili_fsync_fields = 0;
0860     iip->ili_flush_lsn = 0;
0861     iip->ili_item.li_buf = NULL;
0862     list_del_init(&iip->ili_item.li_bio_list);
0863 }
0864 
0865 /*
0866  * Abort flushing the inode from a context holding the cluster buffer locked.
0867  *
0868  * This is the normal runtime method of aborting writeback of an inode that is
0869  * attached to a cluster buffer. It occurs when the inode and the backing
0870  * cluster buffer have been freed (i.e. inode is XFS_ISTALE), or when cluster
0871  * flushing or buffer IO completion encounters a log shutdown situation.
0872  *
0873  * If we need to abort inode writeback and we don't already hold the buffer
0874  * locked, call xfs_iflush_shutdown_abort() instead as this should only ever be
0875  * necessary in a shutdown situation.
0876  */
0877 void
0878 xfs_iflush_abort(
0879     struct xfs_inode    *ip)
0880 {
0881     struct xfs_inode_log_item *iip = ip->i_itemp;
0882     struct xfs_buf      *bp;
0883 
0884     if (!iip) {
0885         /* clean inode, nothing to do */
0886         xfs_iflags_clear(ip, XFS_IFLUSHING);
0887         return;
0888     }
0889 
0890     /*
0891      * Remove the inode item from the AIL before we clear its internal
0892      * state. Whilst the inode is in the AIL, it should have a valid buffer
0893      * pointer for push operations to access - it is only safe to remove the
0894      * inode from the buffer once it has been removed from the AIL.
0895      *
0896      * We also clear the failed bit before removing the item from the AIL
0897      * as xfs_trans_ail_delete()->xfs_clear_li_failed() will release buffer
0898      * references the inode item owns and needs to hold until we've fully
0899      * aborted the inode log item and detached it from the buffer.
0900      */
0901     clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);
0902     xfs_trans_ail_delete(&iip->ili_item, 0);
0903 
0904     /*
0905      * Grab the inode buffer so can we release the reference the inode log
0906      * item holds on it.
0907      */
0908     spin_lock(&iip->ili_lock);
0909     bp = iip->ili_item.li_buf;
0910     xfs_iflush_abort_clean(iip);
0911     spin_unlock(&iip->ili_lock);
0912 
0913     xfs_iflags_clear(ip, XFS_IFLUSHING);
0914     if (bp)
0915         xfs_buf_rele(bp);
0916 }
0917 
0918 /*
0919  * Abort an inode flush in the case of a shutdown filesystem. This can be called
0920  * from anywhere with just an inode reference and does not require holding the
0921  * inode cluster buffer locked. If the inode is attached to a cluster buffer,
0922  * it will grab and lock it safely, then abort the inode flush.
0923  */
0924 void
0925 xfs_iflush_shutdown_abort(
0926     struct xfs_inode    *ip)
0927 {
0928     struct xfs_inode_log_item *iip = ip->i_itemp;
0929     struct xfs_buf      *bp;
0930 
0931     if (!iip) {
0932         /* clean inode, nothing to do */
0933         xfs_iflags_clear(ip, XFS_IFLUSHING);
0934         return;
0935     }
0936 
0937     spin_lock(&iip->ili_lock);
0938     bp = iip->ili_item.li_buf;
0939     if (!bp) {
0940         spin_unlock(&iip->ili_lock);
0941         xfs_iflush_abort(ip);
0942         return;
0943     }
0944 
0945     /*
0946      * We have to take a reference to the buffer so that it doesn't get
0947      * freed when we drop the ili_lock and then wait to lock the buffer.
0948      * We'll clean up the extra reference after we pick up the ili_lock
0949      * again.
0950      */
0951     xfs_buf_hold(bp);
0952     spin_unlock(&iip->ili_lock);
0953     xfs_buf_lock(bp);
0954 
0955     spin_lock(&iip->ili_lock);
0956     if (!iip->ili_item.li_buf) {
0957         /*
0958          * Raced with another removal, hold the only reference
0959          * to bp now. Inode should not be in the AIL now, so just clean
0960          * up and return;
0961          */
0962         ASSERT(list_empty(&iip->ili_item.li_bio_list));
0963         ASSERT(!test_bit(XFS_LI_IN_AIL, &iip->ili_item.li_flags));
0964         xfs_iflush_abort_clean(iip);
0965         spin_unlock(&iip->ili_lock);
0966         xfs_iflags_clear(ip, XFS_IFLUSHING);
0967         xfs_buf_relse(bp);
0968         return;
0969     }
0970 
0971     /*
0972      * Got two references to bp. The first will get dropped by
0973      * xfs_iflush_abort() when the item is removed from the buffer list, but
0974      * we can't drop our reference until _abort() returns because we have to
0975      * unlock the buffer as well. Hence we abort and then unlock and release
0976      * our reference to the buffer.
0977      */
0978     ASSERT(iip->ili_item.li_buf == bp);
0979     spin_unlock(&iip->ili_lock);
0980     xfs_iflush_abort(ip);
0981     xfs_buf_relse(bp);
0982 }
0983 
0984 
0985 /*
0986  * convert an xfs_inode_log_format struct from the old 32 bit version
0987  * (which can have different field alignments) to the native 64 bit version
0988  */
0989 int
0990 xfs_inode_item_format_convert(
0991     struct xfs_log_iovec        *buf,
0992     struct xfs_inode_log_format *in_f)
0993 {
0994     struct xfs_inode_log_format_32  *in_f32 = buf->i_addr;
0995 
0996     if (buf->i_len != sizeof(*in_f32)) {
0997         XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
0998         return -EFSCORRUPTED;
0999     }
1000 
1001     in_f->ilf_type = in_f32->ilf_type;
1002     in_f->ilf_size = in_f32->ilf_size;
1003     in_f->ilf_fields = in_f32->ilf_fields;
1004     in_f->ilf_asize = in_f32->ilf_asize;
1005     in_f->ilf_dsize = in_f32->ilf_dsize;
1006     in_f->ilf_ino = in_f32->ilf_ino;
1007     memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
1008     in_f->ilf_blkno = in_f32->ilf_blkno;
1009     in_f->ilf_len = in_f32->ilf_len;
1010     in_f->ilf_boffset = in_f32->ilf_boffset;
1011     return 0;
1012 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!