Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
0004  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
0005  */
0006 
0007 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
0008 
0009 #include <linux/bio.h>
0010 #include <linux/sched/signal.h>
0011 #include <linux/slab.h>
0012 #include <linux/spinlock.h>
0013 #include <linux/completion.h>
0014 #include <linux/buffer_head.h>
0015 #include <linux/statfs.h>
0016 #include <linux/seq_file.h>
0017 #include <linux/mount.h>
0018 #include <linux/kthread.h>
0019 #include <linux/delay.h>
0020 #include <linux/gfs2_ondisk.h>
0021 #include <linux/crc32.h>
0022 #include <linux/time.h>
0023 #include <linux/wait.h>
0024 #include <linux/writeback.h>
0025 #include <linux/backing-dev.h>
0026 #include <linux/kernel.h>
0027 
0028 #include "gfs2.h"
0029 #include "incore.h"
0030 #include "bmap.h"
0031 #include "dir.h"
0032 #include "glock.h"
0033 #include "glops.h"
0034 #include "inode.h"
0035 #include "log.h"
0036 #include "meta_io.h"
0037 #include "quota.h"
0038 #include "recovery.h"
0039 #include "rgrp.h"
0040 #include "super.h"
0041 #include "trans.h"
0042 #include "util.h"
0043 #include "sys.h"
0044 #include "xattr.h"
0045 #include "lops.h"
0046 
0047 enum dinode_demise {
0048     SHOULD_DELETE_DINODE,
0049     SHOULD_NOT_DELETE_DINODE,
0050     SHOULD_DEFER_EVICTION,
0051 };
0052 
0053 /**
0054  * gfs2_jindex_free - Clear all the journal index information
0055  * @sdp: The GFS2 superblock
0056  *
0057  */
0058 
0059 void gfs2_jindex_free(struct gfs2_sbd *sdp)
0060 {
0061     struct list_head list;
0062     struct gfs2_jdesc *jd;
0063 
0064     spin_lock(&sdp->sd_jindex_spin);
0065     list_add(&list, &sdp->sd_jindex_list);
0066     list_del_init(&sdp->sd_jindex_list);
0067     sdp->sd_journals = 0;
0068     spin_unlock(&sdp->sd_jindex_spin);
0069 
0070     sdp->sd_jdesc = NULL;
0071     while (!list_empty(&list)) {
0072         jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
0073         gfs2_free_journal_extents(jd);
0074         list_del(&jd->jd_list);
0075         iput(jd->jd_inode);
0076         jd->jd_inode = NULL;
0077         kfree(jd);
0078     }
0079 }
0080 
0081 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
0082 {
0083     struct gfs2_jdesc *jd;
0084 
0085     list_for_each_entry(jd, head, jd_list) {
0086         if (jd->jd_jid == jid)
0087             return jd;
0088     }
0089     return NULL;
0090 }
0091 
0092 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
0093 {
0094     struct gfs2_jdesc *jd;
0095 
0096     spin_lock(&sdp->sd_jindex_spin);
0097     jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
0098     spin_unlock(&sdp->sd_jindex_spin);
0099 
0100     return jd;
0101 }
0102 
0103 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
0104 {
0105     struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
0106     struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
0107     u64 size = i_size_read(jd->jd_inode);
0108 
0109     if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
0110         return -EIO;
0111 
0112     jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
0113 
0114     if (gfs2_write_alloc_required(ip, 0, size)) {
0115         gfs2_consist_inode(ip);
0116         return -EIO;
0117     }
0118 
0119     return 0;
0120 }
0121 
0122 /**
0123  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
0124  * @sdp: the filesystem
0125  *
0126  * Returns: errno
0127  */
0128 
0129 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
0130 {
0131     struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
0132     struct gfs2_glock *j_gl = ip->i_gl;
0133     struct gfs2_log_header_host head;
0134     int error;
0135 
0136     j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
0137     if (gfs2_withdrawn(sdp))
0138         return -EIO;
0139 
0140     error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
0141     if (error || gfs2_withdrawn(sdp))
0142         return error;
0143 
0144     if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
0145         gfs2_consist(sdp);
0146         return -EIO;
0147     }
0148 
0149     /*  Initialize some head of the log stuff  */
0150     sdp->sd_log_sequence = head.lh_sequence + 1;
0151     gfs2_log_pointers_init(sdp, head.lh_blkno);
0152 
0153     error = gfs2_quota_init(sdp);
0154     if (!error && !gfs2_withdrawn(sdp))
0155         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
0156     return error;
0157 }
0158 
0159 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
0160 {
0161     const struct gfs2_statfs_change *str = buf;
0162 
0163     sc->sc_total = be64_to_cpu(str->sc_total);
0164     sc->sc_free = be64_to_cpu(str->sc_free);
0165     sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
0166 }
0167 
0168 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
0169 {
0170     struct gfs2_statfs_change *str = buf;
0171 
0172     str->sc_total = cpu_to_be64(sc->sc_total);
0173     str->sc_free = cpu_to_be64(sc->sc_free);
0174     str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
0175 }
0176 
0177 int gfs2_statfs_init(struct gfs2_sbd *sdp)
0178 {
0179     struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
0180     struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
0181     struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
0182     struct buffer_head *m_bh;
0183     struct gfs2_holder gh;
0184     int error;
0185 
0186     error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
0187                    &gh);
0188     if (error)
0189         return error;
0190 
0191     error = gfs2_meta_inode_buffer(m_ip, &m_bh);
0192     if (error)
0193         goto out;
0194 
0195     if (sdp->sd_args.ar_spectator) {
0196         spin_lock(&sdp->sd_statfs_spin);
0197         gfs2_statfs_change_in(m_sc, m_bh->b_data +
0198                       sizeof(struct gfs2_dinode));
0199         spin_unlock(&sdp->sd_statfs_spin);
0200     } else {
0201         spin_lock(&sdp->sd_statfs_spin);
0202         gfs2_statfs_change_in(m_sc, m_bh->b_data +
0203                       sizeof(struct gfs2_dinode));
0204         gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
0205                       sizeof(struct gfs2_dinode));
0206         spin_unlock(&sdp->sd_statfs_spin);
0207 
0208     }
0209 
0210     brelse(m_bh);
0211 out:
0212     gfs2_glock_dq_uninit(&gh);
0213     return 0;
0214 }
0215 
0216 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
0217             s64 dinodes)
0218 {
0219     struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
0220     struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
0221     struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
0222     s64 x, y;
0223     int need_sync = 0;
0224 
0225     gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
0226 
0227     spin_lock(&sdp->sd_statfs_spin);
0228     l_sc->sc_total += total;
0229     l_sc->sc_free += free;
0230     l_sc->sc_dinodes += dinodes;
0231     gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
0232                    sizeof(struct gfs2_dinode));
0233     if (sdp->sd_args.ar_statfs_percent) {
0234         x = 100 * l_sc->sc_free;
0235         y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
0236         if (x >= y || x <= -y)
0237             need_sync = 1;
0238     }
0239     spin_unlock(&sdp->sd_statfs_spin);
0240 
0241     if (need_sync)
0242         gfs2_wake_up_statfs(sdp);
0243 }
0244 
0245 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
0246 {
0247     struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
0248     struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
0249     struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
0250     struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
0251 
0252     gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
0253     gfs2_trans_add_meta(m_ip->i_gl, m_bh);
0254 
0255     spin_lock(&sdp->sd_statfs_spin);
0256     m_sc->sc_total += l_sc->sc_total;
0257     m_sc->sc_free += l_sc->sc_free;
0258     m_sc->sc_dinodes += l_sc->sc_dinodes;
0259     memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
0260     memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
0261            0, sizeof(struct gfs2_statfs_change));
0262     gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
0263     spin_unlock(&sdp->sd_statfs_spin);
0264 }
0265 
0266 int gfs2_statfs_sync(struct super_block *sb, int type)
0267 {
0268     struct gfs2_sbd *sdp = sb->s_fs_info;
0269     struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
0270     struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
0271     struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
0272     struct gfs2_holder gh;
0273     struct buffer_head *m_bh;
0274     int error;
0275 
0276     error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
0277                    &gh);
0278     if (error)
0279         goto out;
0280 
0281     error = gfs2_meta_inode_buffer(m_ip, &m_bh);
0282     if (error)
0283         goto out_unlock;
0284 
0285     spin_lock(&sdp->sd_statfs_spin);
0286     gfs2_statfs_change_in(m_sc, m_bh->b_data +
0287                   sizeof(struct gfs2_dinode));
0288     if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
0289         spin_unlock(&sdp->sd_statfs_spin);
0290         goto out_bh;
0291     }
0292     spin_unlock(&sdp->sd_statfs_spin);
0293 
0294     error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
0295     if (error)
0296         goto out_bh;
0297 
0298     update_statfs(sdp, m_bh);
0299     sdp->sd_statfs_force_sync = 0;
0300 
0301     gfs2_trans_end(sdp);
0302 
0303 out_bh:
0304     brelse(m_bh);
0305 out_unlock:
0306     gfs2_glock_dq_uninit(&gh);
0307 out:
0308     return error;
0309 }
0310 
0311 struct lfcc {
0312     struct list_head list;
0313     struct gfs2_holder gh;
0314 };
0315 
0316 /**
0317  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
0318  *                            journals are clean
0319  * @sdp: the file system
0320  *
0321  * Returns: errno
0322  */
0323 
0324 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
0325 {
0326     struct gfs2_inode *ip;
0327     struct gfs2_jdesc *jd;
0328     struct lfcc *lfcc;
0329     LIST_HEAD(list);
0330     struct gfs2_log_header_host lh;
0331     int error;
0332 
0333     list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
0334         lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
0335         if (!lfcc) {
0336             error = -ENOMEM;
0337             goto out;
0338         }
0339         ip = GFS2_I(jd->jd_inode);
0340         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
0341         if (error) {
0342             kfree(lfcc);
0343             goto out;
0344         }
0345         list_add(&lfcc->list, &list);
0346     }
0347 
0348     error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
0349                    LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
0350     if (error)
0351         goto out;
0352 
0353     list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
0354         error = gfs2_jdesc_check(jd);
0355         if (error)
0356             break;
0357         error = gfs2_find_jhead(jd, &lh, false);
0358         if (error)
0359             break;
0360         if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
0361             error = -EBUSY;
0362             break;
0363         }
0364     }
0365 
0366     if (error)
0367         gfs2_freeze_unlock(&sdp->sd_freeze_gh);
0368 
0369 out:
0370     while (!list_empty(&list)) {
0371         lfcc = list_first_entry(&list, struct lfcc, list);
0372         list_del(&lfcc->list);
0373         gfs2_glock_dq_uninit(&lfcc->gh);
0374         kfree(lfcc);
0375     }
0376     return error;
0377 }
0378 
0379 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
0380 {
0381     struct gfs2_dinode *str = buf;
0382 
0383     str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
0384     str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
0385     str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
0386     str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
0387     str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
0388     str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
0389     str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
0390     str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
0391     str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
0392     str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
0393     str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
0394     str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
0395     str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
0396     str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
0397 
0398     str->di_goal_meta = cpu_to_be64(ip->i_goal);
0399     str->di_goal_data = cpu_to_be64(ip->i_goal);
0400     str->di_generation = cpu_to_be64(ip->i_generation);
0401 
0402     str->di_flags = cpu_to_be32(ip->i_diskflags);
0403     str->di_height = cpu_to_be16(ip->i_height);
0404     str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
0405                          !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
0406                          GFS2_FORMAT_DE : 0);
0407     str->di_depth = cpu_to_be16(ip->i_depth);
0408     str->di_entries = cpu_to_be32(ip->i_entries);
0409 
0410     str->di_eattr = cpu_to_be64(ip->i_eattr);
0411     str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
0412     str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
0413     str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
0414 }
0415 
0416 /**
0417  * gfs2_write_inode - Make sure the inode is stable on the disk
0418  * @inode: The inode
0419  * @wbc: The writeback control structure
0420  *
0421  * Returns: errno
0422  */
0423 
0424 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
0425 {
0426     struct gfs2_inode *ip = GFS2_I(inode);
0427     struct gfs2_sbd *sdp = GFS2_SB(inode);
0428     struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
0429     struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
0430     int ret = 0;
0431     bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
0432 
0433     if (flush_all)
0434         gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
0435                    GFS2_LOG_HEAD_FLUSH_NORMAL |
0436                    GFS2_LFC_WRITE_INODE);
0437     if (bdi->wb.dirty_exceeded)
0438         gfs2_ail1_flush(sdp, wbc);
0439     else
0440         filemap_fdatawrite(metamapping);
0441     if (flush_all)
0442         ret = filemap_fdatawait(metamapping);
0443     if (ret)
0444         mark_inode_dirty_sync(inode);
0445     else {
0446         spin_lock(&inode->i_lock);
0447         if (!(inode->i_flags & I_DIRTY))
0448             gfs2_ordered_del_inode(ip);
0449         spin_unlock(&inode->i_lock);
0450     }
0451     return ret;
0452 }
0453 
0454 /**
0455  * gfs2_dirty_inode - check for atime updates
0456  * @inode: The inode in question
0457  * @flags: The type of dirty
0458  *
0459  * Unfortunately it can be called under any combination of inode
0460  * glock and transaction lock, so we have to check carefully.
0461  *
0462  * At the moment this deals only with atime - it should be possible
0463  * to expand that role in future, once a review of the locking has
0464  * been carried out.
0465  */
0466 
0467 static void gfs2_dirty_inode(struct inode *inode, int flags)
0468 {
0469     struct gfs2_inode *ip = GFS2_I(inode);
0470     struct gfs2_sbd *sdp = GFS2_SB(inode);
0471     struct buffer_head *bh;
0472     struct gfs2_holder gh;
0473     int need_unlock = 0;
0474     int need_endtrans = 0;
0475     int ret;
0476 
0477     if (unlikely(gfs2_withdrawn(sdp)))
0478         return;
0479     if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
0480         ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
0481         if (ret) {
0482             fs_err(sdp, "dirty_inode: glock %d\n", ret);
0483             gfs2_dump_glock(NULL, ip->i_gl, true);
0484             return;
0485         }
0486         need_unlock = 1;
0487     } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
0488         return;
0489 
0490     if (current->journal_info == NULL) {
0491         ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
0492         if (ret) {
0493             fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
0494             goto out;
0495         }
0496         need_endtrans = 1;
0497     }
0498 
0499     ret = gfs2_meta_inode_buffer(ip, &bh);
0500     if (ret == 0) {
0501         gfs2_trans_add_meta(ip->i_gl, bh);
0502         gfs2_dinode_out(ip, bh->b_data);
0503         brelse(bh);
0504     }
0505 
0506     if (need_endtrans)
0507         gfs2_trans_end(sdp);
0508 out:
0509     if (need_unlock)
0510         gfs2_glock_dq_uninit(&gh);
0511 }
0512 
0513 /**
0514  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
0515  * @sdp: the filesystem
0516  *
0517  * Returns: errno
0518  */
0519 
0520 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
0521 {
0522     int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
0523 
0524     gfs2_flush_delete_work(sdp);
0525     if (!log_write_allowed && current == sdp->sd_quotad_process)
0526         fs_warn(sdp, "The quotad daemon is withdrawing.\n");
0527     else if (sdp->sd_quotad_process)
0528         kthread_stop(sdp->sd_quotad_process);
0529     sdp->sd_quotad_process = NULL;
0530 
0531     if (!log_write_allowed && current == sdp->sd_logd_process)
0532         fs_warn(sdp, "The logd daemon is withdrawing.\n");
0533     else if (sdp->sd_logd_process)
0534         kthread_stop(sdp->sd_logd_process);
0535     sdp->sd_logd_process = NULL;
0536 
0537     if (log_write_allowed) {
0538         gfs2_quota_sync(sdp->sd_vfs, 0);
0539         gfs2_statfs_sync(sdp->sd_vfs, 0);
0540 
0541         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
0542                    GFS2_LFC_MAKE_FS_RO);
0543         wait_event_timeout(sdp->sd_log_waitq,
0544                    gfs2_log_is_empty(sdp),
0545                    HZ * 5);
0546         gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
0547     } else {
0548         wait_event_timeout(sdp->sd_log_waitq,
0549                    gfs2_log_is_empty(sdp),
0550                    HZ * 5);
0551     }
0552     gfs2_quota_cleanup(sdp);
0553 
0554     if (!log_write_allowed)
0555         sdp->sd_vfs->s_flags |= SB_RDONLY;
0556 }
0557 
0558 /**
0559  * gfs2_put_super - Unmount the filesystem
0560  * @sb: The VFS superblock
0561  *
0562  */
0563 
0564 static void gfs2_put_super(struct super_block *sb)
0565 {
0566     struct gfs2_sbd *sdp = sb->s_fs_info;
0567     struct gfs2_jdesc *jd;
0568 
0569     /* No more recovery requests */
0570     set_bit(SDF_NORECOVERY, &sdp->sd_flags);
0571     smp_mb();
0572 
0573     /* Wait on outstanding recovery */
0574 restart:
0575     spin_lock(&sdp->sd_jindex_spin);
0576     list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
0577         if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
0578             continue;
0579         spin_unlock(&sdp->sd_jindex_spin);
0580         wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
0581                 TASK_UNINTERRUPTIBLE);
0582         goto restart;
0583     }
0584     spin_unlock(&sdp->sd_jindex_spin);
0585 
0586     if (!sb_rdonly(sb)) {
0587         gfs2_make_fs_ro(sdp);
0588     }
0589     WARN_ON(gfs2_withdrawing(sdp));
0590 
0591     /*  At this point, we're through modifying the disk  */
0592 
0593     /*  Release stuff  */
0594 
0595     iput(sdp->sd_jindex);
0596     iput(sdp->sd_statfs_inode);
0597     iput(sdp->sd_rindex);
0598     iput(sdp->sd_quota_inode);
0599 
0600     gfs2_glock_put(sdp->sd_rename_gl);
0601     gfs2_glock_put(sdp->sd_freeze_gl);
0602 
0603     if (!sdp->sd_args.ar_spectator) {
0604         if (gfs2_holder_initialized(&sdp->sd_journal_gh))
0605             gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
0606         if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
0607             gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
0608         brelse(sdp->sd_sc_bh);
0609         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
0610         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
0611         free_local_statfs_inodes(sdp);
0612         iput(sdp->sd_qc_inode);
0613     }
0614 
0615     gfs2_glock_dq_uninit(&sdp->sd_live_gh);
0616     gfs2_clear_rgrpd(sdp);
0617     gfs2_jindex_free(sdp);
0618     /*  Take apart glock structures and buffer lists  */
0619     gfs2_gl_hash_clear(sdp);
0620     truncate_inode_pages_final(&sdp->sd_aspace);
0621     gfs2_delete_debugfs_file(sdp);
0622     /*  Unmount the locking protocol  */
0623     gfs2_lm_unmount(sdp);
0624 
0625     /*  At this point, we're through participating in the lockspace  */
0626     gfs2_sys_fs_del(sdp);
0627     free_sbd(sdp);
0628 }
0629 
0630 /**
0631  * gfs2_sync_fs - sync the filesystem
0632  * @sb: the superblock
0633  * @wait: true to wait for completion
0634  *
0635  * Flushes the log to disk.
0636  */
0637 
0638 static int gfs2_sync_fs(struct super_block *sb, int wait)
0639 {
0640     struct gfs2_sbd *sdp = sb->s_fs_info;
0641 
0642     gfs2_quota_sync(sb, -1);
0643     if (wait)
0644         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
0645                    GFS2_LFC_SYNC_FS);
0646     return sdp->sd_log_error;
0647 }
0648 
0649 void gfs2_freeze_func(struct work_struct *work)
0650 {
0651     int error;
0652     struct gfs2_holder freeze_gh;
0653     struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
0654     struct super_block *sb = sdp->sd_vfs;
0655 
0656     atomic_inc(&sb->s_active);
0657     error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
0658     if (error) {
0659         gfs2_assert_withdraw(sdp, 0);
0660     } else {
0661         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
0662         error = thaw_super(sb);
0663         if (error) {
0664             fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
0665                 error);
0666             gfs2_assert_withdraw(sdp, 0);
0667         }
0668         gfs2_freeze_unlock(&freeze_gh);
0669     }
0670     deactivate_super(sb);
0671     clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
0672     wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
0673     return;
0674 }
0675 
0676 /**
0677  * gfs2_freeze - prevent further writes to the filesystem
0678  * @sb: the VFS structure for the filesystem
0679  *
0680  */
0681 
0682 static int gfs2_freeze(struct super_block *sb)
0683 {
0684     struct gfs2_sbd *sdp = sb->s_fs_info;
0685     int error;
0686 
0687     mutex_lock(&sdp->sd_freeze_mutex);
0688     if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
0689         error = -EBUSY;
0690         goto out;
0691     }
0692 
0693     for (;;) {
0694         if (gfs2_withdrawn(sdp)) {
0695             error = -EINVAL;
0696             goto out;
0697         }
0698 
0699         error = gfs2_lock_fs_check_clean(sdp);
0700         if (!error)
0701             break;
0702 
0703         if (error == -EBUSY)
0704             fs_err(sdp, "waiting for recovery before freeze\n");
0705         else if (error == -EIO) {
0706             fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
0707                    "to recovery error.\n");
0708             goto out;
0709         } else {
0710             fs_err(sdp, "error freezing FS: %d\n", error);
0711         }
0712         fs_err(sdp, "retrying...\n");
0713         msleep(1000);
0714     }
0715     set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
0716 out:
0717     mutex_unlock(&sdp->sd_freeze_mutex);
0718     return error;
0719 }
0720 
0721 /**
0722  * gfs2_unfreeze - reallow writes to the filesystem
0723  * @sb: the VFS structure for the filesystem
0724  *
0725  */
0726 
0727 static int gfs2_unfreeze(struct super_block *sb)
0728 {
0729     struct gfs2_sbd *sdp = sb->s_fs_info;
0730 
0731     mutex_lock(&sdp->sd_freeze_mutex);
0732     if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
0733         !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
0734         mutex_unlock(&sdp->sd_freeze_mutex);
0735         return -EINVAL;
0736     }
0737 
0738     gfs2_freeze_unlock(&sdp->sd_freeze_gh);
0739     mutex_unlock(&sdp->sd_freeze_mutex);
0740     return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
0741 }
0742 
0743 /**
0744  * statfs_slow_fill - fill in the sg for a given RG
0745  * @rgd: the RG
0746  * @sc: the sc structure
0747  *
0748  * Returns: 0 on success, -ESTALE if the LVB is invalid
0749  */
0750 
0751 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
0752                 struct gfs2_statfs_change_host *sc)
0753 {
0754     gfs2_rgrp_verify(rgd);
0755     sc->sc_total += rgd->rd_data;
0756     sc->sc_free += rgd->rd_free;
0757     sc->sc_dinodes += rgd->rd_dinodes;
0758     return 0;
0759 }
0760 
0761 /**
0762  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
0763  * @sdp: the filesystem
0764  * @sc: the sc info that will be returned
0765  *
0766  * Any error (other than a signal) will cause this routine to fall back
0767  * to the synchronous version.
0768  *
0769  * FIXME: This really shouldn't busy wait like this.
0770  *
0771  * Returns: errno
0772  */
0773 
0774 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
0775 {
0776     struct gfs2_rgrpd *rgd_next;
0777     struct gfs2_holder *gha, *gh;
0778     unsigned int slots = 64;
0779     unsigned int x;
0780     int done;
0781     int error = 0, err;
0782 
0783     memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
0784     gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
0785     if (!gha)
0786         return -ENOMEM;
0787     for (x = 0; x < slots; x++)
0788         gfs2_holder_mark_uninitialized(gha + x);
0789 
0790     rgd_next = gfs2_rgrpd_get_first(sdp);
0791 
0792     for (;;) {
0793         done = 1;
0794 
0795         for (x = 0; x < slots; x++) {
0796             gh = gha + x;
0797 
0798             if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
0799                 err = gfs2_glock_wait(gh);
0800                 if (err) {
0801                     gfs2_holder_uninit(gh);
0802                     error = err;
0803                 } else {
0804                     if (!error) {
0805                         struct gfs2_rgrpd *rgd =
0806                             gfs2_glock2rgrp(gh->gh_gl);
0807 
0808                         error = statfs_slow_fill(rgd, sc);
0809                     }
0810                     gfs2_glock_dq_uninit(gh);
0811                 }
0812             }
0813 
0814             if (gfs2_holder_initialized(gh))
0815                 done = 0;
0816             else if (rgd_next && !error) {
0817                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
0818                                LM_ST_SHARED,
0819                                GL_ASYNC,
0820                                gh);
0821                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
0822                 done = 0;
0823             }
0824 
0825             if (signal_pending(current))
0826                 error = -ERESTARTSYS;
0827         }
0828 
0829         if (done)
0830             break;
0831 
0832         yield();
0833     }
0834 
0835     kfree(gha);
0836     return error;
0837 }
0838 
0839 /**
0840  * gfs2_statfs_i - Do a statfs
0841  * @sdp: the filesystem
0842  * @sc: the sc structure
0843  *
0844  * Returns: errno
0845  */
0846 
0847 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
0848 {
0849     struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
0850     struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
0851 
0852     spin_lock(&sdp->sd_statfs_spin);
0853 
0854     *sc = *m_sc;
0855     sc->sc_total += l_sc->sc_total;
0856     sc->sc_free += l_sc->sc_free;
0857     sc->sc_dinodes += l_sc->sc_dinodes;
0858 
0859     spin_unlock(&sdp->sd_statfs_spin);
0860 
0861     if (sc->sc_free < 0)
0862         sc->sc_free = 0;
0863     if (sc->sc_free > sc->sc_total)
0864         sc->sc_free = sc->sc_total;
0865     if (sc->sc_dinodes < 0)
0866         sc->sc_dinodes = 0;
0867 
0868     return 0;
0869 }
0870 
0871 /**
0872  * gfs2_statfs - Gather and return stats about the filesystem
0873  * @dentry: The name of the link
0874  * @buf: The buffer
0875  *
0876  * Returns: 0 on success or error code
0877  */
0878 
0879 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
0880 {
0881     struct super_block *sb = dentry->d_sb;
0882     struct gfs2_sbd *sdp = sb->s_fs_info;
0883     struct gfs2_statfs_change_host sc;
0884     int error;
0885 
0886     error = gfs2_rindex_update(sdp);
0887     if (error)
0888         return error;
0889 
0890     if (gfs2_tune_get(sdp, gt_statfs_slow))
0891         error = gfs2_statfs_slow(sdp, &sc);
0892     else
0893         error = gfs2_statfs_i(sdp, &sc);
0894 
0895     if (error)
0896         return error;
0897 
0898     buf->f_type = GFS2_MAGIC;
0899     buf->f_bsize = sdp->sd_sb.sb_bsize;
0900     buf->f_blocks = sc.sc_total;
0901     buf->f_bfree = sc.sc_free;
0902     buf->f_bavail = sc.sc_free;
0903     buf->f_files = sc.sc_dinodes + sc.sc_free;
0904     buf->f_ffree = sc.sc_free;
0905     buf->f_namelen = GFS2_FNAMESIZE;
0906 
0907     return 0;
0908 }
0909 
0910 /**
0911  * gfs2_drop_inode - Drop an inode (test for remote unlink)
0912  * @inode: The inode to drop
0913  *
0914  * If we've received a callback on an iopen lock then it's because a
0915  * remote node tried to deallocate the inode but failed due to this node
0916  * still having the inode open. Here we mark the link count zero
0917  * since we know that it must have reached zero if the GLF_DEMOTE flag
0918  * is set on the iopen glock. If we didn't do a disk read since the
0919  * remote node removed the final link then we might otherwise miss
0920  * this event. This check ensures that this node will deallocate the
0921  * inode's blocks, or alternatively pass the baton on to another
0922  * node for later deallocation.
0923  */
0924 
0925 static int gfs2_drop_inode(struct inode *inode)
0926 {
0927     struct gfs2_inode *ip = GFS2_I(inode);
0928 
0929     if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
0930         inode->i_nlink &&
0931         gfs2_holder_initialized(&ip->i_iopen_gh)) {
0932         struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
0933         if (test_bit(GLF_DEMOTE, &gl->gl_flags))
0934             clear_nlink(inode);
0935     }
0936 
0937     /*
0938      * When under memory pressure when an inode's link count has dropped to
0939      * zero, defer deleting the inode to the delete workqueue.  This avoids
0940      * calling into DLM under memory pressure, which can deadlock.
0941      */
0942     if (!inode->i_nlink &&
0943         unlikely(current->flags & PF_MEMALLOC) &&
0944         gfs2_holder_initialized(&ip->i_iopen_gh)) {
0945         struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
0946 
0947         gfs2_glock_hold(gl);
0948         if (!gfs2_queue_delete_work(gl, 0))
0949             gfs2_glock_queue_put(gl);
0950         return 0;
0951     }
0952 
0953     return generic_drop_inode(inode);
0954 }
0955 
0956 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
0957 {
0958     do {
0959         if (d1 == d2)
0960             return 1;
0961         d1 = d1->d_parent;
0962     } while (!IS_ROOT(d1));
0963     return 0;
0964 }
0965 
0966 /**
0967  * gfs2_show_options - Show mount options for /proc/mounts
0968  * @s: seq_file structure
0969  * @root: root of this (sub)tree
0970  *
0971  * Returns: 0 on success or error code
0972  */
0973 
0974 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
0975 {
0976     struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
0977     struct gfs2_args *args = &sdp->sd_args;
0978     int val;
0979 
0980     if (is_ancestor(root, sdp->sd_master_dir))
0981         seq_puts(s, ",meta");
0982     if (args->ar_lockproto[0])
0983         seq_show_option(s, "lockproto", args->ar_lockproto);
0984     if (args->ar_locktable[0])
0985         seq_show_option(s, "locktable", args->ar_locktable);
0986     if (args->ar_hostdata[0])
0987         seq_show_option(s, "hostdata", args->ar_hostdata);
0988     if (args->ar_spectator)
0989         seq_puts(s, ",spectator");
0990     if (args->ar_localflocks)
0991         seq_puts(s, ",localflocks");
0992     if (args->ar_debug)
0993         seq_puts(s, ",debug");
0994     if (args->ar_posix_acl)
0995         seq_puts(s, ",acl");
0996     if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
0997         char *state;
0998         switch (args->ar_quota) {
0999         case GFS2_QUOTA_OFF:
1000             state = "off";
1001             break;
1002         case GFS2_QUOTA_ACCOUNT:
1003             state = "account";
1004             break;
1005         case GFS2_QUOTA_ON:
1006             state = "on";
1007             break;
1008         default:
1009             state = "unknown";
1010             break;
1011         }
1012         seq_printf(s, ",quota=%s", state);
1013     }
1014     if (args->ar_suiddir)
1015         seq_puts(s, ",suiddir");
1016     if (args->ar_data != GFS2_DATA_DEFAULT) {
1017         char *state;
1018         switch (args->ar_data) {
1019         case GFS2_DATA_WRITEBACK:
1020             state = "writeback";
1021             break;
1022         case GFS2_DATA_ORDERED:
1023             state = "ordered";
1024             break;
1025         default:
1026             state = "unknown";
1027             break;
1028         }
1029         seq_printf(s, ",data=%s", state);
1030     }
1031     if (args->ar_discard)
1032         seq_puts(s, ",discard");
1033     val = sdp->sd_tune.gt_logd_secs;
1034     if (val != 30)
1035         seq_printf(s, ",commit=%d", val);
1036     val = sdp->sd_tune.gt_statfs_quantum;
1037     if (val != 30)
1038         seq_printf(s, ",statfs_quantum=%d", val);
1039     else if (sdp->sd_tune.gt_statfs_slow)
1040         seq_puts(s, ",statfs_quantum=0");
1041     val = sdp->sd_tune.gt_quota_quantum;
1042     if (val != 60)
1043         seq_printf(s, ",quota_quantum=%d", val);
1044     if (args->ar_statfs_percent)
1045         seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1046     if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1047         const char *state;
1048 
1049         switch (args->ar_errors) {
1050         case GFS2_ERRORS_WITHDRAW:
1051             state = "withdraw";
1052             break;
1053         case GFS2_ERRORS_PANIC:
1054             state = "panic";
1055             break;
1056         default:
1057             state = "unknown";
1058             break;
1059         }
1060         seq_printf(s, ",errors=%s", state);
1061     }
1062     if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1063         seq_puts(s, ",nobarrier");
1064     if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1065         seq_puts(s, ",demote_interface_used");
1066     if (args->ar_rgrplvb)
1067         seq_puts(s, ",rgrplvb");
1068     if (args->ar_loccookie)
1069         seq_puts(s, ",loccookie");
1070     return 0;
1071 }
1072 
1073 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1074 {
1075     struct inode *inode = &ip->i_inode;
1076     struct gfs2_glock *gl = ip->i_gl;
1077 
1078     truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1079     truncate_inode_pages(&inode->i_data, 0);
1080 
1081     if (atomic_read(&gl->gl_revokes) == 0) {
1082         clear_bit(GLF_LFLUSH, &gl->gl_flags);
1083         clear_bit(GLF_DIRTY, &gl->gl_flags);
1084     }
1085 }
1086 
1087 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1088 {
1089     struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1090     struct gfs2_rgrpd *rgd;
1091     struct gfs2_holder gh;
1092     int error;
1093 
1094     if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1095         gfs2_consist_inode(ip);
1096         return -EIO;
1097     }
1098 
1099     error = gfs2_rindex_update(sdp);
1100     if (error)
1101         return error;
1102 
1103     error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1104     if (error)
1105         return error;
1106 
1107     rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1108     if (!rgd) {
1109         gfs2_consist_inode(ip);
1110         error = -EIO;
1111         goto out_qs;
1112     }
1113 
1114     error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1115                    LM_FLAG_NODE_SCOPE, &gh);
1116     if (error)
1117         goto out_qs;
1118 
1119     error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1120                  sdp->sd_jdesc->jd_blocks);
1121     if (error)
1122         goto out_rg_gunlock;
1123 
1124     gfs2_free_di(rgd, ip);
1125 
1126     gfs2_final_release_pages(ip);
1127 
1128     gfs2_trans_end(sdp);
1129 
1130 out_rg_gunlock:
1131     gfs2_glock_dq_uninit(&gh);
1132 out_qs:
1133     gfs2_quota_unhold(ip);
1134     return error;
1135 }
1136 
1137 /**
1138  * gfs2_glock_put_eventually
1139  * @gl: The glock to put
1140  *
1141  * When under memory pressure, trigger a deferred glock put to make sure we
1142  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1143  */
1144 
1145 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1146 {
1147     if (current->flags & PF_MEMALLOC)
1148         gfs2_glock_queue_put(gl);
1149     else
1150         gfs2_glock_put(gl);
1151 }
1152 
1153 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1154 {
1155     struct gfs2_inode *ip = GFS2_I(inode);
1156     struct gfs2_sbd *sdp = GFS2_SB(inode);
1157     struct gfs2_holder *gh = &ip->i_iopen_gh;
1158     long timeout = 5 * HZ;
1159     int error;
1160 
1161     gh->gh_flags |= GL_NOCACHE;
1162     gfs2_glock_dq_wait(gh);
1163 
1164     /*
1165      * If there are no other lock holders, we'll get the lock immediately.
1166      * Otherwise, the other nodes holding the lock will be notified about
1167      * our locking request.  If they don't have the inode open, they'll
1168      * evict the cached inode and release the lock.  Otherwise, if they
1169      * poke the inode glock, we'll take this as an indication that they
1170      * still need the iopen glock and that they'll take care of deleting
1171      * the inode when they're done.  As a last resort, if another node
1172      * keeps holding the iopen glock without showing any activity on the
1173      * inode glock, we'll eventually time out.
1174      *
1175      * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1176      * locking request as an optimization to notify lock holders as soon as
1177      * possible.  Without that flag, they'd be notified implicitly by the
1178      * second locking request.
1179      */
1180 
1181     gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1182     error = gfs2_glock_nq(gh);
1183     if (error != GLR_TRYFAILED)
1184         return !error;
1185 
1186     gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1187     error = gfs2_glock_nq(gh);
1188     if (error)
1189         return false;
1190 
1191     timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1192         !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1193         test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1194         timeout);
1195     if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1196         gfs2_glock_dq(gh);
1197         return false;
1198     }
1199     return gfs2_glock_holder_ready(gh) == 0;
1200 }
1201 
1202 /**
1203  * evict_should_delete - determine whether the inode is eligible for deletion
1204  * @inode: The inode to evict
1205  * @gh: The glock holder structure
1206  *
1207  * This function determines whether the evicted inode is eligible to be deleted
1208  * and locks the inode glock.
1209  *
1210  * Returns: the fate of the dinode
1211  */
1212 static enum dinode_demise evict_should_delete(struct inode *inode,
1213                           struct gfs2_holder *gh)
1214 {
1215     struct gfs2_inode *ip = GFS2_I(inode);
1216     struct super_block *sb = inode->i_sb;
1217     struct gfs2_sbd *sdp = sb->s_fs_info;
1218     int ret;
1219 
1220     if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1221         BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1222         goto should_delete;
1223     }
1224 
1225     if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1226         return SHOULD_DEFER_EVICTION;
1227 
1228     /* Deletes should never happen under memory pressure anymore.  */
1229     if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1230         return SHOULD_DEFER_EVICTION;
1231 
1232     /* Must not read inode block until block type has been verified */
1233     ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1234     if (unlikely(ret)) {
1235         glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1236         ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1237         gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1238         return SHOULD_DEFER_EVICTION;
1239     }
1240 
1241     if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1242         return SHOULD_NOT_DELETE_DINODE;
1243     ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1244     if (ret)
1245         return SHOULD_NOT_DELETE_DINODE;
1246 
1247     ret = gfs2_instantiate(gh);
1248     if (ret)
1249         return SHOULD_NOT_DELETE_DINODE;
1250 
1251     /*
1252      * The inode may have been recreated in the meantime.
1253      */
1254     if (inode->i_nlink)
1255         return SHOULD_NOT_DELETE_DINODE;
1256 
1257 should_delete:
1258     if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1259         test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1260         if (!gfs2_upgrade_iopen_glock(inode)) {
1261             gfs2_holder_uninit(&ip->i_iopen_gh);
1262             return SHOULD_NOT_DELETE_DINODE;
1263         }
1264     }
1265     return SHOULD_DELETE_DINODE;
1266 }
1267 
1268 /**
1269  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1270  * @inode: The inode to evict
1271  */
1272 static int evict_unlinked_inode(struct inode *inode)
1273 {
1274     struct gfs2_inode *ip = GFS2_I(inode);
1275     int ret;
1276 
1277     if (S_ISDIR(inode->i_mode) &&
1278         (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1279         ret = gfs2_dir_exhash_dealloc(ip);
1280         if (ret)
1281             goto out;
1282     }
1283 
1284     if (ip->i_eattr) {
1285         ret = gfs2_ea_dealloc(ip);
1286         if (ret)
1287             goto out;
1288     }
1289 
1290     if (!gfs2_is_stuffed(ip)) {
1291         ret = gfs2_file_dealloc(ip);
1292         if (ret)
1293             goto out;
1294     }
1295 
1296     /* We're about to clear the bitmap for the dinode, but as soon as we
1297        do, gfs2_create_inode can create another inode at the same block
1298        location and try to set gl_object again. We clear gl_object here so
1299        that subsequent inode creates don't see an old gl_object. */
1300     glock_clear_object(ip->i_gl, ip);
1301     ret = gfs2_dinode_dealloc(ip);
1302     gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1303 out:
1304     return ret;
1305 }
1306 
1307 /*
1308  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1309  * @inode: The inode to evict
1310  */
1311 static int evict_linked_inode(struct inode *inode)
1312 {
1313     struct super_block *sb = inode->i_sb;
1314     struct gfs2_sbd *sdp = sb->s_fs_info;
1315     struct gfs2_inode *ip = GFS2_I(inode);
1316     struct address_space *metamapping;
1317     int ret;
1318 
1319     gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1320                GFS2_LFC_EVICT_INODE);
1321     metamapping = gfs2_glock2aspace(ip->i_gl);
1322     if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1323         filemap_fdatawrite(metamapping);
1324         filemap_fdatawait(metamapping);
1325     }
1326     write_inode_now(inode, 1);
1327     gfs2_ail_flush(ip->i_gl, 0);
1328 
1329     ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1330     if (ret)
1331         return ret;
1332 
1333     /* Needs to be done before glock release & also in a transaction */
1334     truncate_inode_pages(&inode->i_data, 0);
1335     truncate_inode_pages(metamapping, 0);
1336     gfs2_trans_end(sdp);
1337     return 0;
1338 }
1339 
1340 /**
1341  * gfs2_evict_inode - Remove an inode from cache
1342  * @inode: The inode to evict
1343  *
1344  * There are three cases to consider:
1345  * 1. i_nlink == 0, we are final opener (and must deallocate)
1346  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1347  * 3. i_nlink > 0
1348  *
1349  * If the fs is read only, then we have to treat all cases as per #3
1350  * since we are unable to do any deallocation. The inode will be
1351  * deallocated by the next read/write node to attempt an allocation
1352  * in the same resource group
1353  *
1354  * We have to (at the moment) hold the inodes main lock to cover
1355  * the gap between unlocking the shared lock on the iopen lock and
1356  * taking the exclusive lock. I'd rather do a shared -> exclusive
1357  * conversion on the iopen lock, but we can change that later. This
1358  * is safe, just less efficient.
1359  */
1360 
1361 static void gfs2_evict_inode(struct inode *inode)
1362 {
1363     struct super_block *sb = inode->i_sb;
1364     struct gfs2_sbd *sdp = sb->s_fs_info;
1365     struct gfs2_inode *ip = GFS2_I(inode);
1366     struct gfs2_holder gh;
1367     int ret;
1368 
1369     if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1370         clear_inode(inode);
1371         return;
1372     }
1373 
1374     if (inode->i_nlink || sb_rdonly(sb))
1375         goto out;
1376 
1377     gfs2_holder_mark_uninitialized(&gh);
1378     ret = evict_should_delete(inode, &gh);
1379     if (ret == SHOULD_DEFER_EVICTION)
1380         goto out;
1381     if (ret == SHOULD_DELETE_DINODE)
1382         ret = evict_unlinked_inode(inode);
1383     else
1384         ret = evict_linked_inode(inode);
1385 
1386     if (gfs2_rs_active(&ip->i_res))
1387         gfs2_rs_deltree(&ip->i_res);
1388 
1389     if (gfs2_holder_initialized(&gh)) {
1390         glock_clear_object(ip->i_gl, ip);
1391         gfs2_glock_dq_uninit(&gh);
1392     }
1393     if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1394         fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1395 out:
1396     truncate_inode_pages_final(&inode->i_data);
1397     if (ip->i_qadata)
1398         gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1399     gfs2_rs_deltree(&ip->i_res);
1400     gfs2_ordered_del_inode(ip);
1401     clear_inode(inode);
1402     gfs2_dir_hash_inval(ip);
1403     if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1404         struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1405 
1406         glock_clear_object(gl, ip);
1407         if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1408             ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1409             gfs2_glock_dq(&ip->i_iopen_gh);
1410         }
1411         gfs2_glock_hold(gl);
1412         gfs2_holder_uninit(&ip->i_iopen_gh);
1413         gfs2_glock_put_eventually(gl);
1414     }
1415     if (ip->i_gl) {
1416         glock_clear_object(ip->i_gl, ip);
1417         wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1418         gfs2_glock_add_to_lru(ip->i_gl);
1419         gfs2_glock_put_eventually(ip->i_gl);
1420         ip->i_gl = NULL;
1421     }
1422 }
1423 
1424 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1425 {
1426     struct gfs2_inode *ip;
1427 
1428     ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1429     if (!ip)
1430         return NULL;
1431     ip->i_flags = 0;
1432     ip->i_gl = NULL;
1433     gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1434     memset(&ip->i_res, 0, sizeof(ip->i_res));
1435     RB_CLEAR_NODE(&ip->i_res.rs_node);
1436     ip->i_rahead = 0;
1437     return &ip->i_inode;
1438 }
1439 
1440 static void gfs2_free_inode(struct inode *inode)
1441 {
1442     kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1443 }
1444 
1445 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1446 {
1447     struct local_statfs_inode *lsi, *safe;
1448 
1449     /* Run through the statfs inodes list to iput and free memory */
1450     list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1451         if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1452             sdp->sd_sc_inode = NULL; /* belongs to this node */
1453         if (lsi->si_sc_inode)
1454             iput(lsi->si_sc_inode);
1455         list_del(&lsi->si_list);
1456         kfree(lsi);
1457     }
1458 }
1459 
1460 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1461                          unsigned int index)
1462 {
1463     struct local_statfs_inode *lsi;
1464 
1465     /* Return the local (per node) statfs inode in the
1466      * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1467     list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1468         if (lsi->si_jid == index)
1469             return lsi->si_sc_inode;
1470     }
1471     return NULL;
1472 }
1473 
1474 const struct super_operations gfs2_super_ops = {
1475     .alloc_inode        = gfs2_alloc_inode,
1476     .free_inode     = gfs2_free_inode,
1477     .write_inode        = gfs2_write_inode,
1478     .dirty_inode        = gfs2_dirty_inode,
1479     .evict_inode        = gfs2_evict_inode,
1480     .put_super      = gfs2_put_super,
1481     .sync_fs        = gfs2_sync_fs,
1482     .freeze_super       = gfs2_freeze,
1483     .thaw_super     = gfs2_unfreeze,
1484     .statfs         = gfs2_statfs,
1485     .drop_inode     = gfs2_drop_inode,
1486     .show_options       = gfs2_show_options,
1487 };
1488
[ 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!