/*
 * This file is part of UBIFS.
 *
 * Copyright (C) 2006-2008 Nokia Corporation.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published by
 * the Free Software Foundation.
 *
 * This program is distributed in the hope that it will 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 to the Free Software Foundation, Inc., 51
 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 *
 * Author: Adrian Hunter
 */

#include "ubifs.h"

/*
 * An orphan is an inode number whose inode node has been committed to the index
 * with a link count of zero. That happens when an open file is deleted
 * (unlinked) and then a commit is run. In the normal course of events the inode
 * would be deleted when the file is closed. However in the case of an unclean
 * unmount, orphans need to be accounted for. After an unclean unmount, the
 * orphans' inodes must be deleted which means either scanning the entire index
 * looking for them, or keeping a list on flash somewhere. This unit implements
 * the latter approach.
 *
 * The orphan area is a fixed number of LEBs situated between the LPT area and
 * the main area. The number of orphan area LEBs is specified when the file
 * system is created. The minimum number is 1. The size of the orphan area
 * should be so that it can hold the maximum number of orphans that are expected
 * to ever exist at one time.
 *
 * The number of orphans that can fit in a LEB is:
 *
 *         (c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64)
 *
 * For example: a 15872 byte LEB can fit 1980 orphans so 1 LEB may be enough.
 *
 * Orphans are accumulated in a rb-tree. When an inode's link count drops to
 * zero, the inode number is added to the rb-tree. It is removed from the tree
 * when the inode is deleted.  Any new orphans that are in the orphan tree when
 * the commit is run, are written to the orphan area in 1 or more orphan nodes.
 * If the orphan area is full, it is consolidated to make space.  There is
 * always enough space because validation prevents the user from creating more
 * than the maximum number of orphans allowed.
 */

/**
 * tot_avail_orphs - calculate total space.
 * @c: UBIFS file-system description object
 *
 * This function returns the number of orphans that can be written in half
 * the total space. That leaves half the space for adding new orphans.
 */
static int tot_avail_orphs(struct ubifs_info *c)
{
        int avail_lebs, avail;

        avail_lebs = c->orph_lebs;
        avail = avail_lebs *
               ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
        return avail / 2;
}

/**
 * ubifs_clear_orphans - erase all LEBs used for orphans.
 * @c: UBIFS file-system description object
 *
 * If recovery is not required, then the orphans from the previous session
 * are not needed. This function locates the LEBs used to record
 * orphans, and un-maps them.
 */
int ubifs_clear_orphans(struct ubifs_info *c)
{
        int lnum, err;

        for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
                err = ubifs_leb_unmap(c, lnum);
                if (err)
                        return err;
        }
        c->ohead_lnum = c->orph_first;
        c->ohead_offs = 0;
        return 0;
}

/**
 * insert_dead_orphan - insert an orphan.
 * @c: UBIFS file-system description object
 * @inum: orphan inode number
 *
 * This function is a helper to the 'do_kill_orphans()' function. The orphan
 * must be kept until the next commit, so it is added to the rb-tree and the
 * deletion list.
 */
static int insert_dead_orphan(struct ubifs_info *c, ino_t inum)
{
        struct ubifs_orphan *orphan, *o;
        struct rb_node **p, *parent = NULL;

        orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_KERNEL);
        if (!orphan)
                return -ENOMEM;
        orphan->inum = inum;

        p = &c->orph_tree.rb_node;
        while (*p) {
                parent = *p;
                o = rb_entry(parent, struct ubifs_orphan, rb);
                if (inum < o->inum)
                        p = &(*p)->rb_left;
                else if (inum > o->inum)
                        p = &(*p)->rb_right;
                else {
                        /* Already added - no problem */
                        kfree(orphan);
                        return 0;
                }
        }
        c->tot_orphans += 1;
        rb_link_node(&orphan->rb, parent, p);
        rb_insert_color(&orphan->rb, &c->orph_tree);
        list_add_tail(&orphan->list, &c->orph_list);
        orphan->dnext = c->orph_dnext;
        c->orph_dnext = orphan;
        dbg_mnt("ino %lu, new %d, tot %d", (unsigned long)inum,
                c->new_orphans, c->tot_orphans);
        return 0;
}

/**
 * do_kill_orphans - remove orphan inodes from the index.
 * @c: UBIFS file-system description object
 * @sleb: scanned LEB
 * @last_cmt_no: cmt_no of last orphan node read is passed and returned here
 * @outofdate: whether the LEB is out of date is returned here
 * @last_flagged: whether the end orphan node is encountered
 *
 * This function is a helper to the 'kill_orphans()' function. It goes through
 * every orphan node in a LEB and for every inode number recorded, removes
 * all keys for that inode from the TNC.
 */
static int do_kill_orphans(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
                           unsigned long long *last_cmt_no, int *outofdate,
                           int *last_flagged)
{
        struct ubifs_scan_node *snod;
        struct ubifs_orph_node *orph;
        unsigned long long cmt_no;
        ino_t inum;
        int i, n, err, first = 1;

        list_for_each_entry(snod, &sleb->nodes, list) {
                if (snod->type != UBIFS_ORPH_NODE) {
                        ubifs_err("invalid node type %d in orphan area at "
                                  "%d:%d", snod->type, sleb->lnum, snod->offs);
                        dbg_dump_node(c, snod->node);
                        return -EINVAL;
                }

                orph = snod->node;

                /* Check commit number */
                cmt_no = le64_to_cpu(orph->cmt_no) & LLONG_MAX;
                /*
                 * The commit number on the master node may be less, because
                 * of a failed commit. If there are several failed commits in a
                 * row, the commit number written on orphan nodes will continue
                 * to increase (because the commit number is adjusted here) even
                 * though the commit number on the master node stays the same
                 * because the master node has not been re-written.
                 */
                if (cmt_no > c->cmt_no)
                        c->cmt_no = cmt_no;
                if (cmt_no < *last_cmt_no && *last_flagged) {
                        /*
                         * The last orphan node had a higher commit number and
                         * was flagged as the last written for that commit
                         * number. That makes this orphan node, out of date.
                         */
                        if (!first) {
                                ubifs_err("out of order commit number %llu in "
                                          "orphan node at %d:%d",
                                          cmt_no, sleb->lnum, snod->offs);
                                dbg_dump_node(c, snod->node);
                                return -EINVAL;
                        }
                        dbg_rcvry("out of date LEB %d", sleb->lnum);
                        *outofdate = 1;
                        return 0;
                }

                if (first)
                        first = 0;

                n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
                for (i = 0; i < n; i++) {
                        inum = le64_to_cpu(orph->inos[i]);
                        dbg_rcvry("deleting orphaned inode %lu",
                                  (unsigned long)inum);
                        err = ubifs_tnc_remove_ino(c, inum);
                        if (err)
                                return err;
                        err = insert_dead_orphan(c, inum);
                        if (err)
                                return err;
                }

                *last_cmt_no = cmt_no;
                if (le64_to_cpu(orph->cmt_no) & (1ULL << 63)) {
                        dbg_rcvry("last orph node for commit %llu at %d:%d",
                                  cmt_no, sleb->lnum, snod->offs);
                        *last_flagged = 1;
                } else
                        *last_flagged = 0;
        }

        return 0;
}

/**
 * kill_orphans - remove all orphan inodes from the index.
 * @c: UBIFS file-system description object
 *
 * If recovery is required, then orphan inodes recorded during the previous
 * session (which ended with an unclean unmount) must be deleted from the index.
 * This is done by updating the TNC, but since the index is not updated until
 * the next commit, the LEBs where the orphan information is recorded are not
 * erased until the next commit.
 */
static int kill_orphans(struct ubifs_info *c)
{
        unsigned long long last_cmt_no = 0;
        int lnum, err = 0, outofdate = 0, last_flagged = 0;

        c->ohead_lnum = c->orph_first;
        c->ohead_offs = 0;
        /* Check no-orphans flag and skip this if no orphans */
        if (c->no_orphs) {
                dbg_rcvry("no orphans");
                return 0;
        }
        /*
         * Orph nodes always start at c->orph_first and are written to each
         * successive LEB in turn. Generally unused LEBs will have been unmapped
         * but may contain out of date orphan nodes if the unmap didn't go
         * through. In addition, the last orphan node written for each commit is
         * marked (top bit of orph->cmt_no is set to 1). It is possible that
         * there are orphan nodes from the next commit (i.e. the commit did not
         * complete successfully). In that case, no orphans will have been lost
         * due to the way that orphans are written, and any orphans added will
         * be valid orphans anyway and so can be deleted.
         */
        for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
                struct ubifs_scan_leb *sleb;

                dbg_rcvry("LEB %d", lnum);
                sleb = ubifs_scan(c, lnum, 0, c->sbuf);
                if (IS_ERR(sleb)) {
                        sleb = ubifs_recover_leb(c, lnum, 0, c->sbuf, 0);
                        if (IS_ERR(sleb)) {
                                err = PTR_ERR(sleb);
                                break;
                        }
                }
                err = do_kill_orphans(c, sleb, &last_cmt_no, &outofdate,
                                      &last_flagged);
                if (err || outofdate) {
                        ubifs_scan_destroy(sleb);
                        break;
                }
                if (sleb->endpt) {
                        c->ohead_lnum = lnum;
                        c->ohead_offs = sleb->endpt;
                }
                ubifs_scan_destroy(sleb);
        }
        return err;
}

/**
 * ubifs_mount_orphans - delete orphan inodes and erase LEBs that recorded them.
 * @c: UBIFS file-system description object
 * @unclean: indicates recovery from unclean unmount
 * @read_only: indicates read only mount
 *
 * This function is called when mounting to erase orphans from the previous
 * session. If UBIFS was not unmounted cleanly, then the inodes recorded as
 * orphans are deleted.
 */
int ubifs_mount_orphans(struct ubifs_info *c, int unclean, int read_only)
{
        int err = 0;

        c->max_orphans = tot_avail_orphs(c);

        if (!read_only) {
                c->orph_buf = vmalloc(c->leb_size);
                if (!c->orph_buf)
                        return -ENOMEM;
        }

        if (unclean)
                err = kill_orphans(c);
        else if (!read_only)
                err = ubifs_clear_orphans(c);

        return err;
}