Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / xfs / xfs_itable.c
diff --git a/kernel/fs/xfs/xfs_itable.c b/kernel/fs/xfs/xfs_itable.c
new file mode 100644 (file)
index 0000000..8042989
--- /dev/null
@@ -0,0 +1,652 @@
+/*
+ * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
+ * All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_mount.h"
+#include "xfs_inode.h"
+#include "xfs_btree.h"
+#include "xfs_ialloc.h"
+#include "xfs_ialloc_btree.h"
+#include "xfs_itable.h"
+#include "xfs_error.h"
+#include "xfs_trace.h"
+#include "xfs_icache.h"
+
+STATIC int
+xfs_internal_inum(
+       xfs_mount_t     *mp,
+       xfs_ino_t       ino)
+{
+       return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
+               (xfs_sb_version_hasquota(&mp->m_sb) &&
+                xfs_is_quota_inode(&mp->m_sb, ino)));
+}
+
+/*
+ * Return stat information for one inode.
+ * Return 0 if ok, else errno.
+ */
+int
+xfs_bulkstat_one_int(
+       struct xfs_mount        *mp,            /* mount point for filesystem */
+       xfs_ino_t               ino,            /* inode to get data for */
+       void __user             *buffer,        /* buffer to place output in */
+       int                     ubsize,         /* size of buffer */
+       bulkstat_one_fmt_pf     formatter,      /* formatter, copy to user */
+       int                     *ubused,        /* bytes used by me */
+       int                     *stat)          /* BULKSTAT_RV_... */
+{
+       struct xfs_icdinode     *dic;           /* dinode core info pointer */
+       struct xfs_inode        *ip;            /* incore inode pointer */
+       struct xfs_bstat        *buf;           /* return buffer */
+       int                     error = 0;      /* error value */
+
+       *stat = BULKSTAT_RV_NOTHING;
+
+       if (!buffer || xfs_internal_inum(mp, ino))
+               return -EINVAL;
+
+       buf = kmem_alloc(sizeof(*buf), KM_SLEEP | KM_MAYFAIL);
+       if (!buf)
+               return -ENOMEM;
+
+       error = xfs_iget(mp, NULL, ino,
+                        (XFS_IGET_DONTCACHE | XFS_IGET_UNTRUSTED),
+                        XFS_ILOCK_SHARED, &ip);
+       if (error)
+               goto out_free;
+
+       ASSERT(ip != NULL);
+       ASSERT(ip->i_imap.im_blkno != 0);
+
+       dic = &ip->i_d;
+
+       /* xfs_iget returns the following without needing
+        * further change.
+        */
+       buf->bs_nlink = dic->di_nlink;
+       buf->bs_projid_lo = dic->di_projid_lo;
+       buf->bs_projid_hi = dic->di_projid_hi;
+       buf->bs_ino = ino;
+       buf->bs_mode = dic->di_mode;
+       buf->bs_uid = dic->di_uid;
+       buf->bs_gid = dic->di_gid;
+       buf->bs_size = dic->di_size;
+       buf->bs_atime.tv_sec = dic->di_atime.t_sec;
+       buf->bs_atime.tv_nsec = dic->di_atime.t_nsec;
+       buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
+       buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
+       buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
+       buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
+       buf->bs_xflags = xfs_ip2xflags(ip);
+       buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
+       buf->bs_extents = dic->di_nextents;
+       buf->bs_gen = dic->di_gen;
+       memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
+       buf->bs_dmevmask = dic->di_dmevmask;
+       buf->bs_dmstate = dic->di_dmstate;
+       buf->bs_aextents = dic->di_anextents;
+       buf->bs_forkoff = XFS_IFORK_BOFF(ip);
+
+       switch (dic->di_format) {
+       case XFS_DINODE_FMT_DEV:
+               buf->bs_rdev = ip->i_df.if_u2.if_rdev;
+               buf->bs_blksize = BLKDEV_IOSIZE;
+               buf->bs_blocks = 0;
+               break;
+       case XFS_DINODE_FMT_LOCAL:
+       case XFS_DINODE_FMT_UUID:
+               buf->bs_rdev = 0;
+               buf->bs_blksize = mp->m_sb.sb_blocksize;
+               buf->bs_blocks = 0;
+               break;
+       case XFS_DINODE_FMT_EXTENTS:
+       case XFS_DINODE_FMT_BTREE:
+               buf->bs_rdev = 0;
+               buf->bs_blksize = mp->m_sb.sb_blocksize;
+               buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
+               break;
+       }
+       xfs_iunlock(ip, XFS_ILOCK_SHARED);
+       IRELE(ip);
+
+       error = formatter(buffer, ubsize, ubused, buf);
+       if (!error)
+               *stat = BULKSTAT_RV_DIDONE;
+
+ out_free:
+       kmem_free(buf);
+       return error;
+}
+
+/* Return 0 on success or positive error */
+STATIC int
+xfs_bulkstat_one_fmt(
+       void                    __user *ubuffer,
+       int                     ubsize,
+       int                     *ubused,
+       const xfs_bstat_t       *buffer)
+{
+       if (ubsize < sizeof(*buffer))
+               return -ENOMEM;
+       if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
+               return -EFAULT;
+       if (ubused)
+               *ubused = sizeof(*buffer);
+       return 0;
+}
+
+int
+xfs_bulkstat_one(
+       xfs_mount_t     *mp,            /* mount point for filesystem */
+       xfs_ino_t       ino,            /* inode number to get data for */
+       void            __user *buffer, /* buffer to place output in */
+       int             ubsize,         /* size of buffer */
+       int             *ubused,        /* bytes used by me */
+       int             *stat)          /* BULKSTAT_RV_... */
+{
+       return xfs_bulkstat_one_int(mp, ino, buffer, ubsize,
+                                   xfs_bulkstat_one_fmt, ubused, stat);
+}
+
+/*
+ * Loop over all clusters in a chunk for a given incore inode allocation btree
+ * record.  Do a readahead if there are any allocated inodes in that cluster.
+ */
+STATIC void
+xfs_bulkstat_ichunk_ra(
+       struct xfs_mount                *mp,
+       xfs_agnumber_t                  agno,
+       struct xfs_inobt_rec_incore     *irec)
+{
+       xfs_agblock_t                   agbno;
+       struct blk_plug                 plug;
+       int                             blks_per_cluster;
+       int                             inodes_per_cluster;
+       int                             i;      /* inode chunk index */
+
+       agbno = XFS_AGINO_TO_AGBNO(mp, irec->ir_startino);
+       blks_per_cluster = xfs_icluster_size_fsb(mp);
+       inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
+
+       blk_start_plug(&plug);
+       for (i = 0; i < XFS_INODES_PER_CHUNK;
+            i += inodes_per_cluster, agbno += blks_per_cluster) {
+               if (xfs_inobt_maskn(i, inodes_per_cluster) & ~irec->ir_free) {
+                       xfs_btree_reada_bufs(mp, agno, agbno, blks_per_cluster,
+                                            &xfs_inode_buf_ops);
+               }
+       }
+       blk_finish_plug(&plug);
+}
+
+/*
+ * Lookup the inode chunk that the given inode lives in and then get the record
+ * if we found the chunk.  If the inode was not the last in the chunk and there
+ * are some left allocated, update the data for the pointed-to record as well as
+ * return the count of grabbed inodes.
+ */
+STATIC int
+xfs_bulkstat_grab_ichunk(
+       struct xfs_btree_cur            *cur,   /* btree cursor */
+       xfs_agino_t                     agino,  /* starting inode of chunk */
+       int                             *icount,/* return # of inodes grabbed */
+       struct xfs_inobt_rec_incore     *irec)  /* btree record */
+{
+       int                             idx;    /* index into inode chunk */
+       int                             stat;
+       int                             error = 0;
+
+       /* Lookup the inode chunk that this inode lives in */
+       error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &stat);
+       if (error)
+               return error;
+       if (!stat) {
+               *icount = 0;
+               return error;
+       }
+
+       /* Get the record, should always work */
+       error = xfs_inobt_get_rec(cur, irec, &stat);
+       if (error)
+               return error;
+       XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, stat == 1);
+
+       /* Check if the record contains the inode in request */
+       if (irec->ir_startino + XFS_INODES_PER_CHUNK <= agino) {
+               *icount = 0;
+               return 0;
+       }
+
+       idx = agino - irec->ir_startino + 1;
+       if (idx < XFS_INODES_PER_CHUNK &&
+           (xfs_inobt_maskn(idx, XFS_INODES_PER_CHUNK - idx) & ~irec->ir_free)) {
+               int     i;
+
+               /* We got a right chunk with some left inodes allocated at it.
+                * Grab the chunk record.  Mark all the uninteresting inodes
+                * free -- because they're before our start point.
+                */
+               for (i = 0; i < idx; i++) {
+                       if (XFS_INOBT_MASK(i) & ~irec->ir_free)
+                               irec->ir_freecount++;
+               }
+
+               irec->ir_free |= xfs_inobt_maskn(0, idx);
+               *icount = XFS_INODES_PER_CHUNK - irec->ir_freecount;
+       }
+
+       return 0;
+}
+
+#define XFS_BULKSTAT_UBLEFT(ubleft)    ((ubleft) >= statstruct_size)
+
+struct xfs_bulkstat_agichunk {
+       char            __user **ac_ubuffer;/* pointer into user's buffer */
+       int             ac_ubleft;      /* bytes left in user's buffer */
+       int             ac_ubelem;      /* spaces used in user's buffer */
+};
+
+/*
+ * Process inodes in chunk with a pointer to a formatter function
+ * that will iget the inode and fill in the appropriate structure.
+ */
+static int
+xfs_bulkstat_ag_ichunk(
+       struct xfs_mount                *mp,
+       xfs_agnumber_t                  agno,
+       struct xfs_inobt_rec_incore     *irbp,
+       bulkstat_one_pf                 formatter,
+       size_t                          statstruct_size,
+       struct xfs_bulkstat_agichunk    *acp,
+       xfs_agino_t                     *last_agino)
+{
+       char                            __user **ubufp = acp->ac_ubuffer;
+       int                             chunkidx;
+       int                             error = 0;
+       xfs_agino_t                     agino = irbp->ir_startino;
+
+       for (chunkidx = 0; chunkidx < XFS_INODES_PER_CHUNK;
+            chunkidx++, agino++) {
+               int             fmterror;
+               int             ubused;
+
+               /* inode won't fit in buffer, we are done */
+               if (acp->ac_ubleft < statstruct_size)
+                       break;
+
+               /* Skip if this inode is free */
+               if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free)
+                       continue;
+
+               /* Get the inode and fill in a single buffer */
+               ubused = statstruct_size;
+               error = formatter(mp, XFS_AGINO_TO_INO(mp, agno, agino),
+                                 *ubufp, acp->ac_ubleft, &ubused, &fmterror);
+
+               if (fmterror == BULKSTAT_RV_GIVEUP ||
+                   (error && error != -ENOENT && error != -EINVAL)) {
+                       acp->ac_ubleft = 0;
+                       ASSERT(error);
+                       break;
+               }
+
+               /* be careful not to leak error if at end of chunk */
+               if (fmterror == BULKSTAT_RV_NOTHING || error) {
+                       error = 0;
+                       continue;
+               }
+
+               *ubufp += ubused;
+               acp->ac_ubleft -= ubused;
+               acp->ac_ubelem++;
+       }
+
+       /*
+        * Post-update *last_agino. At this point, agino will always point one
+        * inode past the last inode we processed successfully. Hence we
+        * substract that inode when setting the *last_agino cursor so that we
+        * return the correct cookie to userspace. On the next bulkstat call,
+        * the inode under the lastino cookie will be skipped as we have already
+        * processed it here.
+        */
+       *last_agino = agino - 1;
+
+       return error;
+}
+
+/*
+ * Return stat information in bulk (by-inode) for the filesystem.
+ */
+int                                    /* error status */
+xfs_bulkstat(
+       xfs_mount_t             *mp,    /* mount point for filesystem */
+       xfs_ino_t               *lastinop, /* last inode returned */
+       int                     *ubcountp, /* size of buffer/count returned */
+       bulkstat_one_pf         formatter, /* func that'd fill a single buf */
+       size_t                  statstruct_size, /* sizeof struct filling */
+       char                    __user *ubuffer, /* buffer with inode stats */
+       int                     *done)  /* 1 if there are more stats to get */
+{
+       xfs_buf_t               *agbp;  /* agi header buffer */
+       xfs_agino_t             agino;  /* inode # in allocation group */
+       xfs_agnumber_t          agno;   /* allocation group number */
+       xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
+       size_t                  irbsize; /* size of irec buffer in bytes */
+       xfs_inobt_rec_incore_t  *irbuf; /* start of irec buffer */
+       int                     nirbuf; /* size of irbuf */
+       int                     ubcount; /* size of user's buffer */
+       struct xfs_bulkstat_agichunk ac;
+       int                     error = 0;
+
+       /*
+        * Get the last inode value, see if there's nothing to do.
+        */
+       agno = XFS_INO_TO_AGNO(mp, *lastinop);
+       agino = XFS_INO_TO_AGINO(mp, *lastinop);
+       if (agno >= mp->m_sb.sb_agcount ||
+           *lastinop != XFS_AGINO_TO_INO(mp, agno, agino)) {
+               *done = 1;
+               *ubcountp = 0;
+               return 0;
+       }
+
+       ubcount = *ubcountp; /* statstruct's */
+       ac.ac_ubuffer = &ubuffer;
+       ac.ac_ubleft = ubcount * statstruct_size; /* bytes */;
+       ac.ac_ubelem = 0;
+
+       *ubcountp = 0;
+       *done = 0;
+
+       irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4);
+       if (!irbuf)
+               return -ENOMEM;
+
+       nirbuf = irbsize / sizeof(*irbuf);
+
+       /*
+        * Loop over the allocation groups, starting from the last
+        * inode returned; 0 means start of the allocation group.
+        */
+       while (agno < mp->m_sb.sb_agcount) {
+               struct xfs_inobt_rec_incore     *irbp = irbuf;
+               struct xfs_inobt_rec_incore     *irbufend = irbuf + nirbuf;
+               bool                            end_of_ag = false;
+               int                             icount = 0;
+               int                             stat;
+
+               error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
+               if (error)
+                       break;
+               /*
+                * Allocate and initialize a btree cursor for ialloc btree.
+                */
+               cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno,
+                                           XFS_BTNUM_INO);
+               if (agino > 0) {
+                       /*
+                        * In the middle of an allocation group, we need to get
+                        * the remainder of the chunk we're in.
+                        */
+                       struct xfs_inobt_rec_incore     r;
+
+                       error = xfs_bulkstat_grab_ichunk(cur, agino, &icount, &r);
+                       if (error)
+                               goto del_cursor;
+                       if (icount) {
+                               irbp->ir_startino = r.ir_startino;
+                               irbp->ir_freecount = r.ir_freecount;
+                               irbp->ir_free = r.ir_free;
+                               irbp++;
+                       }
+                       /* Increment to the next record */
+                       error = xfs_btree_increment(cur, 0, &stat);
+               } else {
+                       /* Start of ag.  Lookup the first inode chunk */
+                       error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &stat);
+               }
+               if (error || stat == 0) {
+                       end_of_ag = true;
+                       goto del_cursor;
+               }
+
+               /*
+                * Loop through inode btree records in this ag,
+                * until we run out of inodes or space in the buffer.
+                */
+               while (irbp < irbufend && icount < ubcount) {
+                       struct xfs_inobt_rec_incore     r;
+
+                       error = xfs_inobt_get_rec(cur, &r, &stat);
+                       if (error || stat == 0) {
+                               end_of_ag = true;
+                               goto del_cursor;
+                       }
+
+                       /*
+                        * If this chunk has any allocated inodes, save it.
+                        * Also start read-ahead now for this chunk.
+                        */
+                       if (r.ir_freecount < XFS_INODES_PER_CHUNK) {
+                               xfs_bulkstat_ichunk_ra(mp, agno, &r);
+                               irbp->ir_startino = r.ir_startino;
+                               irbp->ir_freecount = r.ir_freecount;
+                               irbp->ir_free = r.ir_free;
+                               irbp++;
+                               icount += XFS_INODES_PER_CHUNK - r.ir_freecount;
+                       }
+                       error = xfs_btree_increment(cur, 0, &stat);
+                       if (error || stat == 0) {
+                               end_of_ag = true;
+                               goto del_cursor;
+                       }
+                       cond_resched();
+               }
+
+               /*
+                * Drop the btree buffers and the agi buffer as we can't hold any
+                * of the locks these represent when calling iget. If there is a
+                * pending error, then we are done.
+                */
+del_cursor:
+               xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
+               xfs_buf_relse(agbp);
+               if (error)
+                       break;
+               /*
+                * Now format all the good inodes into the user's buffer. The
+                * call to xfs_bulkstat_ag_ichunk() sets up the agino pointer
+                * for the next loop iteration.
+                */
+               irbufend = irbp;
+               for (irbp = irbuf;
+                    irbp < irbufend && ac.ac_ubleft >= statstruct_size;
+                    irbp++) {
+                       error = xfs_bulkstat_ag_ichunk(mp, agno, irbp,
+                                       formatter, statstruct_size, &ac,
+                                       &agino);
+                       if (error)
+                               break;
+
+                       cond_resched();
+               }
+
+               /*
+                * If we've run out of space or had a formatting error, we
+                * are now done
+                */
+               if (ac.ac_ubleft < statstruct_size || error)
+                       break;
+
+               if (end_of_ag) {
+                       agno++;
+                       agino = 0;
+               }
+       }
+       /*
+        * Done, we're either out of filesystem or space to put the data.
+        */
+       kmem_free(irbuf);
+       *ubcountp = ac.ac_ubelem;
+
+       /*
+        * We found some inodes, so clear the error status and return them.
+        * The lastino pointer will point directly at the inode that triggered
+        * any error that occurred, so on the next call the error will be
+        * triggered again and propagated to userspace as there will be no
+        * formatted inodes in the buffer.
+        */
+       if (ac.ac_ubelem)
+               error = 0;
+
+       /*
+        * If we ran out of filesystem, lastino will point off the end of
+        * the filesystem so the next call will return immediately.
+        */
+       *lastinop = XFS_AGINO_TO_INO(mp, agno, agino);
+       if (agno >= mp->m_sb.sb_agcount)
+               *done = 1;
+
+       return error;
+}
+
+int
+xfs_inumbers_fmt(
+       void                    __user *ubuffer, /* buffer to write to */
+       const struct xfs_inogrp *buffer,        /* buffer to read from */
+       long                    count,          /* # of elements to read */
+       long                    *written)       /* # of bytes written */
+{
+       if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
+               return -EFAULT;
+       *written = count * sizeof(*buffer);
+       return 0;
+}
+
+/*
+ * Return inode number table for the filesystem.
+ */
+int                                    /* error status */
+xfs_inumbers(
+       struct xfs_mount        *mp,/* mount point for filesystem */
+       xfs_ino_t               *lastino,/* last inode returned */
+       int                     *count,/* size of buffer/count returned */
+       void                    __user *ubuffer,/* buffer with inode descriptions */
+       inumbers_fmt_pf         formatter)
+{
+       xfs_agnumber_t          agno = XFS_INO_TO_AGNO(mp, *lastino);
+       xfs_agino_t             agino = XFS_INO_TO_AGINO(mp, *lastino);
+       struct xfs_btree_cur    *cur = NULL;
+       struct xfs_buf          *agbp = NULL;
+       struct xfs_inogrp       *buffer;
+       int                     bcount;
+       int                     left = *count;
+       int                     bufidx = 0;
+       int                     error = 0;
+
+       *count = 0;
+       if (agno >= mp->m_sb.sb_agcount ||
+           *lastino != XFS_AGINO_TO_INO(mp, agno, agino))
+               return error;
+
+       bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
+       buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
+       do {
+               struct xfs_inobt_rec_incore     r;
+               int                             stat;
+
+               if (!agbp) {
+                       error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
+                       if (error)
+                               break;
+
+                       cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno,
+                                                   XFS_BTNUM_INO);
+                       error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
+                                                &stat);
+                       if (error)
+                               break;
+                       if (!stat)
+                               goto next_ag;
+               }
+
+               error = xfs_inobt_get_rec(cur, &r, &stat);
+               if (error)
+                       break;
+               if (!stat)
+                       goto next_ag;
+
+               agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
+               buffer[bufidx].xi_startino =
+                       XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
+               buffer[bufidx].xi_alloccount =
+                       XFS_INODES_PER_CHUNK - r.ir_freecount;
+               buffer[bufidx].xi_allocmask = ~r.ir_free;
+               if (++bufidx == bcount) {
+                       long    written;
+
+                       error = formatter(ubuffer, buffer, bufidx, &written);
+                       if (error)
+                               break;
+                       ubuffer += written;
+                       *count += bufidx;
+                       bufidx = 0;
+               }
+               if (!--left)
+                       break;
+
+               error = xfs_btree_increment(cur, 0, &stat);
+               if (error)
+                       break;
+               if (stat)
+                       continue;
+
+next_ag:
+               xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
+               cur = NULL;
+               xfs_buf_relse(agbp);
+               agbp = NULL;
+               agino = 0;
+               agno++;
+       } while (agno < mp->m_sb.sb_agcount);
+
+       if (!error) {
+               if (bufidx) {
+                       long    written;
+
+                       error = formatter(ubuffer, buffer, bufidx, &written);
+                       if (!error)
+                               *count += bufidx;
+               }
+               *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
+       }
+
+       kmem_free(buffer);
+       if (cur)
+               xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
+                                          XFS_BTREE_NOERROR));
+       if (agbp)
+               xfs_buf_relse(agbp);
+
+       return error;
+}