fs/gfs2/ops_super.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/statfs.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/mount.h>
  18 #include <linux/kthread.h>
  19 #include <linux/delay.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/crc32.h>
  22 #include <linux/lm_interface.h>
  23
  24 #include "gfs2.h"
  25 #include "incore.h"
  26 #include "glock.h"
  27 #include "inode.h"
  28 #include "log.h"
  29 #include "mount.h"
  30 #include "ops_super.h"
  31 #include "quota.h"
  32 #include "recovery.h"
  33 #include "rgrp.h"
  34 #include "super.h"
  35 #include "sys.h"
  36 #include "util.h"
  37 #include "trans.h"
  38 #include "dir.h"
  39 #include "eattr.h"
  40 #include "bmap.h"
  41
  42 /**
  43  * gfs2_write_inode - Make sure the inode is stable on the disk
  44  * @inode: The inode
  45  * @sync: synchronous write flag
  46  *
  47  * Returns: errno
  48  */
  49
  50 static int gfs2_write_inode(struct inode *inode, int sync)
  51 {
  52         struct gfs2_inode *ip = GFS2_I(inode);
  53
  54         /* Check this is a "normal" inode */
  55         if (inode->i_private) {
  56                 if (current->flags & PF_MEMALLOC)
  57                         return 0;
  58                 if (sync)
  59                         gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
  60         }
  61
  62         return 0;
  63 }
  64
  65 /**
  66  * gfs2_put_super - Unmount the filesystem
  67  * @sb: The VFS superblock
  68  *
  69  */
  70
  71 static void gfs2_put_super(struct super_block *sb)
  72 {
  73         struct gfs2_sbd *sdp = sb->s_fs_info;
  74         int error;
  75
  76         if (!sdp)
  77                 return;
  78
  79         if (!strncmp(sb->s_type->name, "gfs2meta", 8))
  80                 return; /* Nothing to do */
  81
  82         /*  Unfreeze the filesystem, if we need to  */
  83
  84         mutex_lock(&sdp->sd_freeze_lock);
  85         if (sdp->sd_freeze_count)
  86                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
  87         mutex_unlock(&sdp->sd_freeze_lock);
  88
  89         kthread_stop(sdp->sd_quotad_process);
  90         kthread_stop(sdp->sd_logd_process);
  91         kthread_stop(sdp->sd_recoverd_process);
  92         while (sdp->sd_glockd_num--)
  93                 kthread_stop(sdp->sd_glockd_process[sdp->sd_glockd_num]);
  94
  95         if (!(sb->s_flags & MS_RDONLY)) {
  96                 error = gfs2_make_fs_ro(sdp);
  97                 if (error)
  98                         gfs2_io_error(sdp);
  99         }
 100         /*  At this point, we're through modifying the disk  */
 101
 102         /*  Release stuff  */
 103
 104         iput(sdp->sd_master_dir);
 105         iput(sdp->sd_jindex);
 106         iput(sdp->sd_inum_inode);
 107         iput(sdp->sd_statfs_inode);
 108         iput(sdp->sd_rindex);
 109         iput(sdp->sd_quota_inode);
 110
 111         gfs2_glock_put(sdp->sd_rename_gl);
 112         gfs2_glock_put(sdp->sd_trans_gl);
 113
 114         if (!sdp->sd_args.ar_spectator) {
 115                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 116                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 117                 gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
 118                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 119                 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 120                 iput(sdp->sd_ir_inode);
 121                 iput(sdp->sd_sc_inode);
 122                 iput(sdp->sd_qc_inode);
 123         }
 124
 125         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 126         gfs2_clear_rgrpd(sdp);
 127         gfs2_jindex_free(sdp);
 128         /*  Take apart glock structures and buffer lists  */
 129         gfs2_gl_hash_clear(sdp, WAIT);
 130         /*  Unmount the locking protocol  */
 131         gfs2_lm_unmount(sdp);
 132
 133         /*  At this point, we're through participating in the lockspace  */
 134         gfs2_sys_fs_del(sdp);
 135         kfree(sdp);
 136 }
 137
 138 /**
 139  * gfs2_write_super
 140  * @sb: the superblock
 141  *
 142  */
 143
 144 static void gfs2_write_super(struct super_block *sb)
 145 {
 146         sb->s_dirt = 0;
 147 }
 148
 149 /**
 150  * gfs2_sync_fs - sync the filesystem
 151  * @sb: the superblock
 152  *
 153  * Flushes the log to disk.
 154  */
 155 static int gfs2_sync_fs(struct super_block *sb, int wait)
 156 {
 157         sb->s_dirt = 0;
 158         if (wait)
 159                 gfs2_log_flush(sb->s_fs_info, NULL);
 160         return 0;
 161 }
 162
 163 /**
 164  * gfs2_write_super_lockfs - prevent further writes to the filesystem
 165  * @sb: the VFS structure for the filesystem
 166  *
 167  */
 168
 169 static void gfs2_write_super_lockfs(struct super_block *sb)
 170 {
 171         struct gfs2_sbd *sdp = sb->s_fs_info;
 172         int error;
 173
 174         if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 175                 return;
 176
 177         for (;;) {
 178                 error = gfs2_freeze_fs(sdp);
 179                 if (!error)
 180                         break;
 181
 182                 switch (error) {
 183                 case -EBUSY:
 184                         fs_err(sdp, "waiting for recovery before freeze\n");
 185                         break;
 186
 187                 default:
 188                         fs_err(sdp, "error freezing FS: %d\n", error);
 189                         break;
 190                 }
 191
 192                 fs_err(sdp, "retrying...\n");
 193                 msleep(1000);
 194         }
 195 }
 196
 197 /**
 198  * gfs2_unlockfs - reallow writes to the filesystem
 199  * @sb: the VFS structure for the filesystem
 200  *
 201  */
 202
 203 static void gfs2_unlockfs(struct super_block *sb)
 204 {
 205         gfs2_unfreeze_fs(sb->s_fs_info);
 206 }
 207
 208 /**
 209  * gfs2_statfs - Gather and return stats about the filesystem
 210  * @sb: The superblock
 211  * @statfsbuf: The buffer
 212  *
 213  * Returns: 0 on success or error code
 214  */
 215
 216 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 217 {
 218         struct super_block *sb = dentry->d_inode->i_sb;
 219         struct gfs2_sbd *sdp = sb->s_fs_info;
 220         struct gfs2_statfs_change_host sc;
 221         int error;
 222
 223         if (gfs2_tune_get(sdp, gt_statfs_slow))
 224                 error = gfs2_statfs_slow(sdp, &sc);
 225         else
 226                 error = gfs2_statfs_i(sdp, &sc);
 227
 228         if (error)
 229                 return error;
 230
 231         buf->f_type = GFS2_MAGIC;
 232         buf->f_bsize = sdp->sd_sb.sb_bsize;
 233         buf->f_blocks = sc.sc_total;
 234         buf->f_bfree = sc.sc_free;
 235         buf->f_bavail = sc.sc_free;
 236         buf->f_files = sc.sc_dinodes + sc.sc_free;
 237         buf->f_ffree = sc.sc_free;
 238         buf->f_namelen = GFS2_FNAMESIZE;
 239
 240         return 0;
 241 }
 242
 243 /**
 244  * gfs2_remount_fs - called when the FS is remounted
 245  * @sb:  the filesystem
 246  * @flags:  the remount flags
 247  * @data:  extra data passed in (not used right now)
 248  *
 249  * Returns: errno
 250  */
 251
 252 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
 253 {
 254         struct gfs2_sbd *sdp = sb->s_fs_info;
 255         int error;
 256
 257         error = gfs2_mount_args(sdp, data, 1);
 258         if (error)
 259                 return error;
 260
 261         if (sdp->sd_args.ar_spectator)
 262                 *flags |= MS_RDONLY;
 263         else {
 264                 if (*flags & MS_RDONLY) {
 265                         if (!(sb->s_flags & MS_RDONLY))
 266                                 error = gfs2_make_fs_ro(sdp);
 267                 } else if (!(*flags & MS_RDONLY) &&
 268                            (sb->s_flags & MS_RDONLY)) {
 269                         error = gfs2_make_fs_rw(sdp);
 270                 }
 271         }
 272
 273         if (*flags & (MS_NOATIME | MS_NODIRATIME))
 274                 set_bit(SDF_NOATIME, &sdp->sd_flags);
 275         else
 276                 clear_bit(SDF_NOATIME, &sdp->sd_flags);
 277
 278         /* Don't let the VFS update atimes.  GFS2 handles this itself. */
 279         *flags |= MS_NOATIME | MS_NODIRATIME;
 280
 281         return error;
 282 }
 283
 284 /**
 285  * gfs2_drop_inode - Drop an inode (test for remote unlink)
 286  * @inode: The inode to drop
 287  *
 288  * If we've received a callback on an iopen lock then its because a
 289  * remote node tried to deallocate the inode but failed due to this node
 290  * still having the inode open. Here we mark the link count zero
 291  * since we know that it must have reached zero if the GLF_DEMOTE flag
 292  * is set on the iopen glock. If we didn't do a disk read since the
 293  * remote node removed the final link then we might otherwise miss
 294  * this event. This check ensures that this node will deallocate the
 295  * inode's blocks, or alternatively pass the baton on to another
 296  * node for later deallocation.
 297  */
 298 static void gfs2_drop_inode(struct inode *inode)
 299 {
 300         if (inode->i_private && inode->i_nlink) {
 301                 struct gfs2_inode *ip = GFS2_I(inode);
 302                 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
 303                 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
 304                         clear_nlink(inode);
 305         }
 306         generic_drop_inode(inode);
 307 }
 308
 309 /**
 310  * gfs2_clear_inode - Deallocate an inode when VFS is done with it
 311  * @inode: The VFS inode
 312  *
 313  */
 314
 315 static void gfs2_clear_inode(struct inode *inode)
 316 {
 317         /* This tells us its a "real" inode and not one which only
 318          * serves to contain an address space (see rgrp.c, meta_io.c)
 319          * which therefore doesn't have its own glocks.
 320          */
 321         if (inode->i_private) {
 322                 struct gfs2_inode *ip = GFS2_I(inode);
 323                 ip->i_gl->gl_object = NULL;
 324                 gfs2_glock_schedule_for_reclaim(ip->i_gl);
 325                 gfs2_glock_put(ip->i_gl);
 326                 ip->i_gl = NULL;
 327                 if (ip->i_iopen_gh.gh_gl) {
 328                         ip->i_iopen_gh.gh_gl->gl_object = NULL;
 329                         gfs2_glock_dq_uninit(&ip->i_iopen_gh);
 330                 }
 331         }
 332 }
 333
 334 /**
 335  * gfs2_show_options - Show mount options for /proc/mounts
 336  * @s: seq_file structure
 337  * @mnt: vfsmount
 338  *
 339  * Returns: 0 on success or error code
 340  */
 341
 342 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
 343 {
 344         struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
 345         struct gfs2_args *args = &sdp->sd_args;
 346
 347         if (args->ar_lockproto[0])
 348                 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
 349         if (args->ar_locktable[0])
 350                 seq_printf(s, ",locktable=%s", args->ar_locktable);
 351         if (args->ar_hostdata[0])
 352                 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
 353         if (args->ar_spectator)
 354                 seq_printf(s, ",spectator");
 355         if (args->ar_ignore_local_fs)
 356                 seq_printf(s, ",ignore_local_fs");
 357         if (args->ar_localflocks)
 358                 seq_printf(s, ",localflocks");
 359         if (args->ar_localcaching)
 360                 seq_printf(s, ",localcaching");
 361         if (args->ar_debug)
 362                 seq_printf(s, ",debug");
 363         if (args->ar_upgrade)
 364                 seq_printf(s, ",upgrade");
 365         if (args->ar_num_glockd != GFS2_GLOCKD_DEFAULT)
 366                 seq_printf(s, ",num_glockd=%u", args->ar_num_glockd);
 367         if (args->ar_posix_acl)
 368                 seq_printf(s, ",acl");
 369         if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
 370                 char *state;
 371                 switch (args->ar_quota) {
 372                 case GFS2_QUOTA_OFF:
 373                         state = "off";
 374                         break;
 375                 case GFS2_QUOTA_ACCOUNT:
 376                         state = "account";
 377                         break;
 378                 case GFS2_QUOTA_ON:
 379                         state = "on";
 380                         break;
 381                 default:
 382                         state = "unknown";
 383                         break;
 384                 }
 385                 seq_printf(s, ",quota=%s", state);
 386         }
 387         if (args->ar_suiddir)
 388                 seq_printf(s, ",suiddir");
 389         if (args->ar_data != GFS2_DATA_DEFAULT) {
 390                 char *state;
 391                 switch (args->ar_data) {
 392                 case GFS2_DATA_WRITEBACK:
 393                         state = "writeback";
 394                         break;
 395                 case GFS2_DATA_ORDERED:
 396                         state = "ordered";
 397                         break;
 398                 default:
 399                         state = "unknown";
 400                         break;
 401                 }
 402                 seq_printf(s, ",data=%s", state);
 403         }
 404
 405         return 0;
 406 }
 407
 408 /*
 409  * We have to (at the moment) hold the inodes main lock to cover
 410  * the gap between unlocking the shared lock on the iopen lock and
 411  * taking the exclusive lock. I'd rather do a shared -> exclusive
 412  * conversion on the iopen lock, but we can change that later. This
 413  * is safe, just less efficient.
 414  */
 415 static void gfs2_delete_inode(struct inode *inode)
 416 {
 417         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 418         struct gfs2_inode *ip = GFS2_I(inode);
 419         struct gfs2_holder gh;
 420         int error;
 421
 422         if (!inode->i_private)
 423                 goto out;
 424
 425         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 426         if (unlikely(error)) {
 427                 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
 428                 goto out;
 429         }
 430
 431         gfs2_glock_dq_wait(&ip->i_iopen_gh);
 432         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
 433         error = gfs2_glock_nq(&ip->i_iopen_gh);
 434         if (error)
 435                 goto out_uninit;
 436
 437         if (S_ISDIR(inode->i_mode) &&
 438             (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
 439                 error = gfs2_dir_exhash_dealloc(ip);
 440                 if (error)
 441                         goto out_unlock;
 442         }
 443
 444         if (ip->i_di.di_eattr) {
 445                 error = gfs2_ea_dealloc(ip);
 446                 if (error)
 447                         goto out_unlock;
 448         }
 449
 450         if (!gfs2_is_stuffed(ip)) {
 451                 error = gfs2_file_dealloc(ip);
 452                 if (error)
 453                         goto out_unlock;
 454         }
 455
 456         error = gfs2_dinode_dealloc(ip);
 457         if (error)
 458                 goto out_unlock;
 459
 460         error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
 461         if (error)
 462                 goto out_unlock;
 463         /* Needs to be done before glock release & also in a transaction */
 464         truncate_inode_pages(&inode->i_data, 0);
 465         gfs2_trans_end(sdp);
 466
 467 out_unlock:
 468         gfs2_glock_dq(&ip->i_iopen_gh);
 469 out_uninit:
 470         gfs2_holder_uninit(&ip->i_iopen_gh);
 471         gfs2_glock_dq_uninit(&gh);
 472         if (error && error != GLR_TRYFAILED)
 473                 fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
 474 out:
 475         truncate_inode_pages(&inode->i_data, 0);
 476         clear_inode(inode);
 477 }
 478
 479
 480
 481 static struct inode *gfs2_alloc_inode(struct super_block *sb)
 482 {
 483         struct gfs2_inode *ip;
 484
 485         ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
 486         if (ip) {
 487                 ip->i_flags = 0;
 488                 ip->i_gl = NULL;
 489         }
 490         return &ip->i_inode;
 491 }
 492
 493 static void gfs2_destroy_inode(struct inode *inode)
 494 {
 495         kmem_cache_free(gfs2_inode_cachep, inode);
 496 }
 497
 498 const struct super_operations gfs2_super_ops = {
 499         .alloc_inode            = gfs2_alloc_inode,
 500         .destroy_inode          = gfs2_destroy_inode,
 501         .write_inode            = gfs2_write_inode,
 502         .delete_inode           = gfs2_delete_inode,
 503         .put_super              = gfs2_put_super,
 504         .write_super            = gfs2_write_super,
 505         .sync_fs                = gfs2_sync_fs,
 506         .write_super_lockfs     = gfs2_write_super_lockfs,
 507         .unlockfs               = gfs2_unlockfs,
 508         .statfs                 = gfs2_statfs,
 509         .remount_fs             = gfs2_remount_fs,
 510         .clear_inode            = gfs2_clear_inode,
 511         .drop_inode             = gfs2_drop_inode,
 512         .show_options           = gfs2_show_options,
 513 };
 514