fs/xfs/xfs_icreate_item.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * Copyright (c) 2008-2010, 2013 Dave Chinner
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_trans_priv.h"
0016 #include "xfs_icreate_item.h"
0017 #include "xfs_log.h"
0018 #include "xfs_log_priv.h"
0019 #include "xfs_log_recover.h"
0020 #include "xfs_ialloc.h"
0021 #include "xfs_trace.h"
0022
0023 struct kmem_cache   *xfs_icreate_cache;     /* inode create item */
0024
0025 static inline struct xfs_icreate_item *ICR_ITEM(struct xfs_log_item *lip)
0026 {
0027     return container_of(lip, struct xfs_icreate_item, ic_item);
0028 }
0029
0030 /*
0031  * This returns the number of iovecs needed to log the given inode item.
0032  *
0033  * We only need one iovec for the icreate log structure.
0034  */
0035 STATIC void
0036 xfs_icreate_item_size(
0037     struct xfs_log_item *lip,
0038     int         *nvecs,
0039     int         *nbytes)
0040 {
0041     *nvecs += 1;
0042     *nbytes += sizeof(struct xfs_icreate_log);
0043 }
0044
0045 /*
0046  * This is called to fill in the vector of log iovecs for the
0047  * given inode create log item.
0048  */
0049 STATIC void
0050 xfs_icreate_item_format(
0051     struct xfs_log_item *lip,
0052     struct xfs_log_vec  *lv)
0053 {
0054     struct xfs_icreate_item *icp = ICR_ITEM(lip);
0055     struct xfs_log_iovec    *vecp = NULL;
0056
0057     xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ICREATE,
0058             &icp->ic_format,
0059             sizeof(struct xfs_icreate_log));
0060 }
0061
0062 STATIC void
0063 xfs_icreate_item_release(
0064     struct xfs_log_item *lip)
0065 {
0066     kmem_free(ICR_ITEM(lip)->ic_item.li_lv_shadow);
0067     kmem_cache_free(xfs_icreate_cache, ICR_ITEM(lip));
0068 }
0069
0070 static const struct xfs_item_ops xfs_icreate_item_ops = {
0071     .flags      = XFS_ITEM_RELEASE_WHEN_COMMITTED,
0072     .iop_size   = xfs_icreate_item_size,
0073     .iop_format = xfs_icreate_item_format,
0074     .iop_release    = xfs_icreate_item_release,
0075 };
0076
0077
0078 /*
0079  * Initialize the inode log item for a newly allocated (in-core) inode.
0080  *
0081  * Inode extents can only reside within an AG. Hence specify the starting
0082  * block for the inode chunk by offset within an AG as well as the
0083  * length of the allocated extent.
0084  *
0085  * This joins the item to the transaction and marks it dirty so
0086  * that we don't need a separate call to do this, nor does the
0087  * caller need to know anything about the icreate item.
0088  */
0089 void
0090 xfs_icreate_log(
0091     struct xfs_trans    *tp,
0092     xfs_agnumber_t      agno,
0093     xfs_agblock_t       agbno,
0094     unsigned int        count,
0095     unsigned int        inode_size,
0096     xfs_agblock_t       length,
0097     unsigned int        generation)
0098 {
0099     struct xfs_icreate_item *icp;
0100
0101     icp = kmem_cache_zalloc(xfs_icreate_cache, GFP_KERNEL | __GFP_NOFAIL);
0102
0103     xfs_log_item_init(tp->t_mountp, &icp->ic_item, XFS_LI_ICREATE,
0104               &xfs_icreate_item_ops);
0105
0106     icp->ic_format.icl_type = XFS_LI_ICREATE;
0107     icp->ic_format.icl_size = 1;    /* single vector */
0108     icp->ic_format.icl_ag = cpu_to_be32(agno);
0109     icp->ic_format.icl_agbno = cpu_to_be32(agbno);
0110     icp->ic_format.icl_count = cpu_to_be32(count);
0111     icp->ic_format.icl_isize = cpu_to_be32(inode_size);
0112     icp->ic_format.icl_length = cpu_to_be32(length);
0113     icp->ic_format.icl_gen = cpu_to_be32(generation);
0114
0115     xfs_trans_add_item(tp, &icp->ic_item);
0116     tp->t_flags |= XFS_TRANS_DIRTY;
0117     set_bit(XFS_LI_DIRTY, &icp->ic_item.li_flags);
0118 }
0119
0120 static enum xlog_recover_reorder
0121 xlog_recover_icreate_reorder(
0122         struct xlog_recover_item *item)
0123 {
0124     /*
0125      * Inode allocation buffers must be replayed before subsequent inode
0126      * items try to modify those buffers.  ICREATE items are the logical
0127      * equivalent of logging a newly initialized inode buffer, so recover
0128      * these at the same time that we recover logged buffers.
0129      */
0130     return XLOG_REORDER_BUFFER_LIST;
0131 }
0132
0133 /*
0134  * This routine is called when an inode create format structure is found in a
0135  * committed transaction in the log.  It's purpose is to initialise the inodes
0136  * being allocated on disk. This requires us to get inode cluster buffers that
0137  * match the range to be initialised, stamped with inode templates and written
0138  * by delayed write so that subsequent modifications will hit the cached buffer
0139  * and only need writing out at the end of recovery.
0140  */
0141 STATIC int
0142 xlog_recover_icreate_commit_pass2(
0143     struct xlog         *log,
0144     struct list_head        *buffer_list,
0145     struct xlog_recover_item    *item,
0146     xfs_lsn_t           lsn)
0147 {
0148     struct xfs_mount        *mp = log->l_mp;
0149     struct xfs_icreate_log      *icl;
0150     struct xfs_ino_geometry     *igeo = M_IGEO(mp);
0151     xfs_agnumber_t          agno;
0152     xfs_agblock_t           agbno;
0153     unsigned int            count;
0154     unsigned int            isize;
0155     xfs_agblock_t           length;
0156     int             bb_per_cluster;
0157     int             cancel_count;
0158     int             nbufs;
0159     int             i;
0160
0161     icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
0162     if (icl->icl_type != XFS_LI_ICREATE) {
0163         xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
0164         return -EINVAL;
0165     }
0166
0167     if (icl->icl_size != 1) {
0168         xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
0169         return -EINVAL;
0170     }
0171
0172     agno = be32_to_cpu(icl->icl_ag);
0173     if (agno >= mp->m_sb.sb_agcount) {
0174         xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
0175         return -EINVAL;
0176     }
0177     agbno = be32_to_cpu(icl->icl_agbno);
0178     if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
0179         xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
0180         return -EINVAL;
0181     }
0182     isize = be32_to_cpu(icl->icl_isize);
0183     if (isize != mp->m_sb.sb_inodesize) {
0184         xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
0185         return -EINVAL;
0186     }
0187     count = be32_to_cpu(icl->icl_count);
0188     if (!count) {
0189         xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
0190         return -EINVAL;
0191     }
0192     length = be32_to_cpu(icl->icl_length);
0193     if (!length || length >= mp->m_sb.sb_agblocks) {
0194         xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
0195         return -EINVAL;
0196     }
0197
0198     /*
0199      * The inode chunk is either full or sparse and we only support
0200      * m_ino_geo.ialloc_min_blks sized sparse allocations at this time.
0201      */
0202     if (length != igeo->ialloc_blks &&
0203         length != igeo->ialloc_min_blks) {
0204         xfs_warn(log->l_mp,
0205              "%s: unsupported chunk length", __func__);
0206         return -EINVAL;
0207     }
0208
0209     /* verify inode count is consistent with extent length */
0210     if ((count >> mp->m_sb.sb_inopblog) != length) {
0211         xfs_warn(log->l_mp,
0212              "%s: inconsistent inode count and chunk length",
0213              __func__);
0214         return -EINVAL;
0215     }
0216
0217     /*
0218      * The icreate transaction can cover multiple cluster buffers and these
0219      * buffers could have been freed and reused. Check the individual
0220      * buffers for cancellation so we don't overwrite anything written after
0221      * a cancellation.
0222      */
0223     bb_per_cluster = XFS_FSB_TO_BB(mp, igeo->blocks_per_cluster);
0224     nbufs = length / igeo->blocks_per_cluster;
0225     for (i = 0, cancel_count = 0; i < nbufs; i++) {
0226         xfs_daddr_t daddr;
0227
0228         daddr = XFS_AGB_TO_DADDR(mp, agno,
0229                 agbno + i * igeo->blocks_per_cluster);
0230         if (xlog_is_buffer_cancelled(log, daddr, bb_per_cluster))
0231             cancel_count++;
0232     }
0233
0234     /*
0235      * We currently only use icreate for a single allocation at a time. This
0236      * means we should expect either all or none of the buffers to be
0237      * cancelled. Be conservative and skip replay if at least one buffer is
0238      * cancelled, but warn the user that something is awry if the buffers
0239      * are not consistent.
0240      *
0241      * XXX: This must be refined to only skip cancelled clusters once we use
0242      * icreate for multiple chunk allocations.
0243      */
0244     ASSERT(!cancel_count || cancel_count == nbufs);
0245     if (cancel_count) {
0246         if (cancel_count != nbufs)
0247             xfs_warn(mp,
0248     "WARNING: partial inode chunk cancellation, skipped icreate.");
0249         trace_xfs_log_recover_icreate_cancel(log, icl);
0250         return 0;
0251     }
0252
0253     trace_xfs_log_recover_icreate_recover(log, icl);
0254     return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
0255                      length, be32_to_cpu(icl->icl_gen));
0256 }
0257
0258 const struct xlog_recover_item_ops xlog_icreate_item_ops = {
0259     .item_type      = XFS_LI_ICREATE,
0260     .reorder        = xlog_recover_icreate_reorder,
0261     .commit_pass2       = xlog_recover_icreate_commit_pass2,
0262 };