[PATCH] cpu hotplug: use hotplug version of cpu notifier in appropriate places
[powerpc.git] / fs / xfs / xfs_mount.c
index 20e8abc..ed7579b 100644 (file)
 #include "xfs_trans.h"
 #include "xfs_sb.h"
 #include "xfs_ag.h"
-#include "xfs_dir.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
 #include "xfs_bmap_btree.h"
 #include "xfs_alloc_btree.h"
 #include "xfs_ialloc_btree.h"
-#include "xfs_dir_sf.h"
 #include "xfs_dir2_sf.h"
 #include "xfs_attr_sf.h"
 #include "xfs_dinode.h"
@@ -196,7 +194,7 @@ xfs_mount_free(
                kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
 
        if (remove_bhv) {
-               struct vfs      *vfsp = XFS_MTOVFS(mp);
+               struct bhv_vfs  *vfsp = XFS_MTOVFS(mp);
 
                bhv_remove_all_vfsops(vfsp, 0);
                VFS_REMOVEBHV(vfsp, &mp->m_bhv);
@@ -213,7 +211,8 @@ xfs_mount_free(
 STATIC int
 xfs_mount_validate_sb(
        xfs_mount_t     *mp,
-       xfs_sb_t        *sbp)
+       xfs_sb_t        *sbp,
+       int             flags)
 {
        /*
         * If the log device and data device have the
@@ -223,33 +222,29 @@ xfs_mount_validate_sb(
         * a volume filesystem in a non-volume manner.
         */
        if (sbp->sb_magicnum != XFS_SB_MAGIC) {
-               cmn_err(CE_WARN, "XFS: bad magic number");
+               xfs_fs_mount_cmn_err(flags, "bad magic number");
                return XFS_ERROR(EWRONGFS);
        }
 
        if (!XFS_SB_GOOD_VERSION(sbp)) {
-               cmn_err(CE_WARN, "XFS: bad version");
+               xfs_fs_mount_cmn_err(flags, "bad version");
                return XFS_ERROR(EWRONGFS);
        }
 
        if (unlikely(
            sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
-               cmn_err(CE_WARN,
-       "XFS: filesystem is marked as having an external log; "
-       "specify logdev on the\nmount command line.");
-               XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(1)",
-                                    XFS_ERRLEVEL_HIGH, mp, sbp);
-               return XFS_ERROR(EFSCORRUPTED);
+               xfs_fs_mount_cmn_err(flags,
+                       "filesystem is marked as having an external log; "
+                       "specify logdev on the\nmount command line.");
+               return XFS_ERROR(EINVAL);
        }
 
        if (unlikely(
            sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
-               cmn_err(CE_WARN,
-       "XFS: filesystem is marked as having an internal log; "
-       "don't specify logdev on\nthe mount command line.");
-               XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(2)",
-                                    XFS_ERRLEVEL_HIGH, mp, sbp);
-               return XFS_ERROR(EFSCORRUPTED);
+               xfs_fs_mount_cmn_err(flags,
+                       "filesystem is marked as having an internal log; "
+                       "do not specify logdev on\nthe mount command line.");
+               return XFS_ERROR(EINVAL);
        }
 
        /*
@@ -273,10 +268,8 @@ xfs_mount_validate_sb(
            (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)   ||
            (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)  ||
            (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)  ||
-           (sbp->sb_imax_pct > 100 || sbp->sb_imax_pct < 1))) {
-               cmn_err(CE_WARN, "XFS: SB sanity check 1 failed");
-               XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(3)",
-                                    XFS_ERRLEVEL_LOW, mp, sbp);
+           (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
+               xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
                return XFS_ERROR(EFSCORRUPTED);
        }
 
@@ -289,9 +282,7 @@ xfs_mount_validate_sb(
             (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
            sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
                              sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
-               cmn_err(CE_WARN, "XFS: SB sanity check 2 failed");
-               XFS_ERROR_REPORT("xfs_mount_validate_sb(4)",
-                                XFS_ERRLEVEL_LOW, mp);
+               xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
                return XFS_ERROR(EFSCORRUPTED);
        }
 
@@ -307,15 +298,13 @@ xfs_mount_validate_sb(
            (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
            (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
 #endif
-               cmn_err(CE_WARN,
-       "XFS: File system is too large to be mounted on this system.");
+               xfs_fs_mount_cmn_err(flags,
+                       "file system too large to be mounted on this system.");
                return XFS_ERROR(E2BIG);
        }
 
        if (unlikely(sbp->sb_inprogress)) {
-               cmn_err(CE_WARN, "XFS: file system busy");
-               XFS_ERROR_REPORT("xfs_mount_validate_sb(5)",
-                                XFS_ERRLEVEL_LOW, mp);
+               xfs_fs_mount_cmn_err(flags, "file system busy");
                return XFS_ERROR(EFSCORRUPTED);
        }
 
@@ -323,8 +312,8 @@ xfs_mount_validate_sb(
         * Version 1 directory format has never worked on Linux.
         */
        if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
-               cmn_err(CE_WARN,
-       "XFS: Attempted to mount file system using version 1 directory format");
+               xfs_fs_mount_cmn_err(flags,
+                       "file system using version 1 directory format");
                return XFS_ERROR(ENOSYS);
        }
 
@@ -332,11 +321,11 @@ xfs_mount_validate_sb(
         * Until this is fixed only page-sized or smaller data blocks work.
         */
        if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
-               cmn_err(CE_WARN,
-               "XFS: Attempted to mount file system with blocksize %d bytes",
+               xfs_fs_mount_cmn_err(flags,
+                       "file system with blocksize %d bytes",
                        sbp->sb_blocksize);
-               cmn_err(CE_WARN,
-               "XFS: Only page-sized (%ld) or less blocksizes currently work.",
+               xfs_fs_mount_cmn_err(flags,
+                       "only pagesize (%ld) or less will currently work.",
                        PAGE_SIZE);
                return XFS_ERROR(ENOSYS);
        }
@@ -346,7 +335,7 @@ xfs_mount_validate_sb(
 
 xfs_agnumber_t
 xfs_initialize_perag(
-       struct vfs      *vfs,
+       bhv_vfs_t       *vfs,
        xfs_mount_t     *mp,
        xfs_agnumber_t  agcount)
 {
@@ -393,7 +382,7 @@ xfs_initialize_perag(
                                break;
                        }
 
-                       /* This ag is prefered for inodes */
+                       /* This ag is preferred for inodes */
                        pag = &mp->m_perag[index];
                        pag->pagi_inodeok = 1;
                        if (index < max_metadata)
@@ -484,7 +473,7 @@ xfs_xlatesb(
  * Does the initial read of the superblock.
  */
 int
-xfs_readsb(xfs_mount_t *mp)
+xfs_readsb(xfs_mount_t *mp, int flags)
 {
        unsigned int    sector_size;
        unsigned int    extra_flags;
@@ -506,7 +495,7 @@ xfs_readsb(xfs_mount_t *mp)
        bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
                                BTOBB(sector_size), extra_flags);
        if (!bp || XFS_BUF_ISERROR(bp)) {
-               cmn_err(CE_WARN, "XFS: SB read failed");
+               xfs_fs_mount_cmn_err(flags, "SB read failed");
                error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
                goto fail;
        }
@@ -520,9 +509,9 @@ xfs_readsb(xfs_mount_t *mp)
        sbp = XFS_BUF_TO_SBP(bp);
        xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS);
 
-       error = xfs_mount_validate_sb(mp, &(mp->m_sb));
+       error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
        if (error) {
-               cmn_err(CE_WARN, "XFS: SB validate failed");
+               xfs_fs_mount_cmn_err(flags, "SB validate failed");
                goto fail;
        }
 
@@ -530,8 +519,8 @@ xfs_readsb(xfs_mount_t *mp)
         * We must be able to do sector-sized and sector-aligned IO.
         */
        if (sector_size > mp->m_sb.sb_sectsize) {
-               cmn_err(CE_WARN,
-                       "XFS: device supports only %u byte sectors (not %u)",
+               xfs_fs_mount_cmn_err(flags,
+                       "device supports only %u byte sectors (not %u)",
                        sector_size, mp->m_sb.sb_sectsize);
                error = ENOSYS;
                goto fail;
@@ -548,7 +537,7 @@ xfs_readsb(xfs_mount_t *mp)
                bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
                                        BTOBB(sector_size), extra_flags);
                if (!bp || XFS_BUF_ISERROR(bp)) {
-                       cmn_err(CE_WARN, "XFS: SB re-read failed");
+                       xfs_fs_mount_cmn_err(flags, "SB re-read failed");
                        error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
                        goto fail;
                }
@@ -660,14 +649,14 @@ xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
  */
 int
 xfs_mountfs(
-       vfs_t           *vfsp,
+       bhv_vfs_t       *vfsp,
        xfs_mount_t     *mp,
        int             mfsi_flags)
 {
        xfs_buf_t       *bp;
        xfs_sb_t        *sbp = &(mp->m_sb);
        xfs_inode_t     *rip;
-       vnode_t         *rvp = NULL;
+       bhv_vnode_t     *rvp = NULL;
        int             readio_log, writeio_log;
        xfs_daddr_t     d;
        __uint64_t      ret64;
@@ -678,7 +667,7 @@ xfs_mountfs(
        int             error = 0;
 
        if (mp->m_sb_bp == NULL) {
-               if ((error = xfs_readsb(mp))) {
+               if ((error = xfs_readsb(mp, mfsi_flags))) {
                        return error;
                }
        }
@@ -943,18 +932,7 @@ xfs_mountfs(
        vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
        mp->m_dmevmask = 0;     /* not persistent; set after each mount */
 
-       /*
-        * Select the right directory manager.
-        */
-       mp->m_dirops =
-               XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ?
-                       xfsv2_dirops :
-                       xfsv1_dirops;
-
-       /*
-        * Initialize directory manager's entries.
-        */
-       XFS_DIR_MOUNT(mp);
+       xfs_dir_mount(mp);
 
        /*
         * Initialize the attribute manager's entries.
@@ -1015,8 +993,9 @@ xfs_mountfs(
 
        if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
                cmn_err(CE_WARN, "XFS: corrupted root inode");
-               prdev("Root inode %llu is not a directory",
-                     mp->m_ddev_targp, (unsigned long long)rip->i_ino);
+               cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
+                       XFS_BUFTARG_NAME(mp->m_ddev_targp),
+                       (unsigned long long)rip->i_ino);
                xfs_iunlock(rip, XFS_ILOCK_EXCL);
                XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
                                 mp);
@@ -1103,7 +1082,7 @@ xfs_mountfs(
 int
 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
 {
-       struct vfs      *vfsp = XFS_MTOVFS(mp);
+       struct bhv_vfs  *vfsp = XFS_MTOVFS(mp);
 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
        int64_t         fsid;
 #endif
@@ -1263,6 +1242,26 @@ xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
 
        xfs_trans_log_buf(tp, bp, first, last);
 }
+
+/*
+ * In order to avoid ENOSPC-related deadlock caused by
+ * out-of-order locking of AGF buffer (PV 947395), we place
+ * constraints on the relationship among actual allocations for
+ * data blocks, freelist blocks, and potential file data bmap
+ * btree blocks. However, these restrictions may result in no
+ * actual space allocated for a delayed extent, for example, a data
+ * block in a certain AG is allocated but there is no additional
+ * block for the additional bmap btree block due to a split of the
+ * bmap btree of the file. The result of this may lead to an
+ * infinite loop in xfssyncd when the file gets flushed to disk and
+ * all delayed extents need to be actually allocated. To get around
+ * this, we explicitly set aside a few blocks which will not be
+ * reserved in delayed allocation. Considering the minimum number of
+ * needed freelist blocks is 4 fsbs, a potential split of file's bmap
+ * btree requires 1 fsb, so we set the number of set-aside blocks to 8.
+*/
+#define SET_ASIDE_BLOCKS 8
+
 /*
  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
  * a delta to a specified field in the in-core superblock.  Simply
@@ -1307,7 +1306,7 @@ xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
                return 0;
        case XFS_SBS_FDBLOCKS:
 
-               lcounter = (long long)mp->m_sb.sb_fdblocks;
+               lcounter = (long long)mp->m_sb.sb_fdblocks - SET_ASIDE_BLOCKS;
                res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
 
                if (delta > 0) {                /* Putting blocks back */
@@ -1341,7 +1340,7 @@ xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
                        }
                }
 
-               mp->m_sb.sb_fdblocks = lcounter;
+               mp->m_sb.sb_fdblocks = lcounter + SET_ASIDE_BLOCKS;
                return 0;
        case XFS_SBS_FREXTENTS:
                lcounter = (long long)mp->m_sb.sb_frextents;
@@ -1722,15 +1721,14 @@ xfs_mount_log_sbunit(
  * is present to prevent thrashing).
  */
 
+#ifdef CONFIG_HOTPLUG_CPU
 /*
  * hot-plug CPU notifier support.
  *
- * We cannot use the hotcpu_register() function because it does
- * not allow notifier instances. We need a notifier per filesystem
- * as we need to be able to identify the filesystem to balance
- * the counters out. This is acheived by having a notifier block
- * embedded in the xfs_mount_t and doing pointer magic to get the
- * mount pointer from the notifier block address.
+ * We need a notifier per filesystem as we need to be able to identify
+ * the filesystem to balance the counters out. This is achieved by
+ * having a notifier block embedded in the xfs_mount_t and doing pointer
+ * magic to get the mount pointer from the notifier block address.
  */
 STATIC int
 xfs_icsb_cpu_notify(
@@ -1780,6 +1778,7 @@ xfs_icsb_cpu_notify(
 
        return NOTIFY_OK;
 }
+#endif /* CONFIG_HOTPLUG_CPU */
 
 int
 xfs_icsb_init_counters(
@@ -1792,9 +1791,11 @@ xfs_icsb_init_counters(
        if (mp->m_sb_cnts == NULL)
                return -ENOMEM;
 
+#ifdef CONFIG_HOTPLUG_CPU
        mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
        mp->m_icsb_notifier.priority = 0;
-       register_cpu_notifier(&mp->m_icsb_notifier);
+       register_hotcpu_notifier(&mp->m_icsb_notifier);
+#endif /* CONFIG_HOTPLUG_CPU */
 
        for_each_online_cpu(i) {
                cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
@@ -1813,7 +1814,7 @@ xfs_icsb_destroy_counters(
        xfs_mount_t     *mp)
 {
        if (mp->m_sb_cnts) {
-               unregister_cpu_notifier(&mp->m_icsb_notifier);
+               unregister_hotcpu_notifier(&mp->m_icsb_notifier);
                free_percpu(mp->m_sb_cnts);
        }
 }