fs/gfs2/trans.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/sched.h>
  10 #include <linux/slab.h>
  11 #include <linux/spinlock.h>
  12 #include <linux/completion.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/kallsyms.h>
  15 #include <linux/gfs2_ondisk.h>
  16
  17 #include "gfs2.h"
  18 #include "incore.h"
  19 #include "glock.h"
  20 #include "inode.h"
  21 #include "log.h"
  22 #include "lops.h"
  23 #include "meta_io.h"
  24 #include "trans.h"
  25 #include "util.h"
  26 #include "trace_gfs2.h"
  27
  28 static void gfs2_print_trans(struct gfs2_sbd *sdp, const struct gfs2_trans *tr)
  29 {
  30         fs_warn(sdp, "Transaction created at: %pSR\n", (void *)tr->tr_ip);
  31         fs_warn(sdp, "blocks=%u revokes=%u reserved=%u touched=%u\n",
  32                 tr->tr_blocks, tr->tr_revokes, tr->tr_reserved,
  33                 test_bit(TR_TOUCHED, &tr->tr_flags));
  34         fs_warn(sdp, "Buf %u/%u Databuf %u/%u Revoke %u/%u\n",
  35                 tr->tr_num_buf_new, tr->tr_num_buf_rm,
  36                 tr->tr_num_databuf_new, tr->tr_num_databuf_rm,
  37                 tr->tr_num_revoke, tr->tr_num_revoke_rm);
  38 }
  39
  40 int __gfs2_trans_begin(struct gfs2_trans *tr, struct gfs2_sbd *sdp,
  41                        unsigned int blocks, unsigned int revokes,
  42                        unsigned long ip)
  43 {
  44         if (current->journal_info) {
  45                 gfs2_print_trans(sdp, current->journal_info);
  46                 BUG();
  47         }
  48         BUG_ON(blocks == 0 && revokes == 0);
  49
  50         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
  51                 return -EROFS;
  52
  53         tr->tr_ip = ip;
  54         tr->tr_blocks = blocks;
  55         tr->tr_revokes = revokes;
  56         tr->tr_reserved = 1;
  57         if (blocks) {
  58                 /*
  59                  * The reserved blocks are either used for data or metadata.
  60                  * We can have mixed data and metadata, each with its own log
  61                  * descriptor block; see calc_reserved().
  62                  */
  63                 tr->tr_reserved += blocks + 1 + DIV_ROUND_UP(blocks - 1, databuf_limit(sdp));
  64         }
  65         if (revokes)
  66                 tr->tr_reserved += gfs2_struct2blk(sdp, revokes);
  67         INIT_LIST_HEAD(&tr->tr_databuf);
  68         INIT_LIST_HEAD(&tr->tr_buf);
  69         INIT_LIST_HEAD(&tr->tr_list);
  70         INIT_LIST_HEAD(&tr->tr_ail1_list);
  71         INIT_LIST_HEAD(&tr->tr_ail2_list);
  72
  73         if (gfs2_assert_warn(sdp, tr->tr_reserved <= sdp->sd_jdesc->jd_blocks))
  74                 return -EINVAL;
  75
  76         sb_start_intwrite(sdp->sd_vfs);
  77
  78         gfs2_log_reserve(sdp, tr->tr_reserved);
  79
  80         down_read(&sdp->sd_log_flush_lock);
  81         if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
  82                 up_read(&sdp->sd_log_flush_lock);
  83                 gfs2_log_release(sdp, tr->tr_reserved);
  84                 sb_end_intwrite(sdp->sd_vfs);
  85                 return -EROFS;
  86         }
  87
  88         current->journal_info = tr;
  89
  90         return 0;
  91 }
  92
  93 int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks,
  94                      unsigned int revokes)
  95 {
  96         struct gfs2_trans *tr;
  97         int error;
  98
  99         tr = kmem_cache_zalloc(gfs2_trans_cachep, GFP_NOFS);
 100         if (!tr)
 101                 return -ENOMEM;
 102         error = __gfs2_trans_begin(tr, sdp, blocks, revokes, _RET_IP_);
 103         if (error)
 104                 kmem_cache_free(gfs2_trans_cachep, tr);
 105         return error;
 106 }
 107
 108 void gfs2_trans_end(struct gfs2_sbd *sdp)
 109 {
 110         struct gfs2_trans *tr = current->journal_info;
 111         s64 nbuf;
 112
 113         current->journal_info = NULL;
 114
 115         if (!test_bit(TR_TOUCHED, &tr->tr_flags)) {
 116                 gfs2_log_release(sdp, tr->tr_reserved);
 117                 up_read(&sdp->sd_log_flush_lock);
 118                 if (!test_bit(TR_ONSTACK, &tr->tr_flags))
 119                         gfs2_trans_free(sdp, tr);
 120                 sb_end_intwrite(sdp->sd_vfs);
 121                 return;
 122         }
 123
 124         nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new;
 125         nbuf -= tr->tr_num_buf_rm;
 126         nbuf -= tr->tr_num_databuf_rm;
 127
 128         if (gfs2_assert_withdraw(sdp, nbuf <= tr->tr_blocks) ||
 129             gfs2_assert_withdraw(sdp, tr->tr_num_revoke <= tr->tr_revokes))
 130                 gfs2_print_trans(sdp, tr);
 131
 132         gfs2_log_commit(sdp, tr);
 133         if (!test_bit(TR_ONSTACK, &tr->tr_flags) &&
 134             !test_bit(TR_ATTACHED, &tr->tr_flags))
 135                 gfs2_trans_free(sdp, tr);
 136         up_read(&sdp->sd_log_flush_lock);
 137
 138         if (sdp->sd_vfs->s_flags & SB_SYNCHRONOUS)
 139                 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
 140                                GFS2_LFC_TRANS_END);
 141         sb_end_intwrite(sdp->sd_vfs);
 142 }
 143
 144 static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl,
 145                                                struct buffer_head *bh)
 146 {
 147         struct gfs2_bufdata *bd;
 148
 149         bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
 150         bd->bd_bh = bh;
 151         bd->bd_gl = gl;
 152         INIT_LIST_HEAD(&bd->bd_list);
 153         bh->b_private = bd;
 154         return bd;
 155 }
 156
 157 /**
 158  * gfs2_trans_add_data - Add a databuf to the transaction.
 159  * @gl: The inode glock associated with the buffer
 160  * @bh: The buffer to add
 161  *
 162  * This is used in journaled data mode.
 163  * We need to journal the data block in the same way as metadata in
 164  * the functions above. The difference is that here we have a tag
 165  * which is two __be64's being the block number (as per meta data)
 166  * and a flag which says whether the data block needs escaping or
 167  * not. This means we need a new log entry for each 251 or so data
 168  * blocks, which isn't an enormous overhead but twice as much as
 169  * for normal metadata blocks.
 170  */
 171 void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
 172 {
 173         struct gfs2_trans *tr = current->journal_info;
 174         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 175         struct gfs2_bufdata *bd;
 176
 177         lock_buffer(bh);
 178         if (buffer_pinned(bh)) {
 179                 set_bit(TR_TOUCHED, &tr->tr_flags);
 180                 goto out;
 181         }
 182         gfs2_log_lock(sdp);
 183         bd = bh->b_private;
 184         if (bd == NULL) {
 185                 gfs2_log_unlock(sdp);
 186                 unlock_buffer(bh);
 187                 if (bh->b_private == NULL)
 188                         bd = gfs2_alloc_bufdata(gl, bh);
 189                 else
 190                         bd = bh->b_private;
 191                 lock_buffer(bh);
 192                 gfs2_log_lock(sdp);
 193         }
 194         gfs2_assert(sdp, bd->bd_gl == gl);
 195         set_bit(TR_TOUCHED, &tr->tr_flags);
 196         if (list_empty(&bd->bd_list)) {
 197                 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 198                 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 199                 gfs2_pin(sdp, bd->bd_bh);
 200                 tr->tr_num_databuf_new++;
 201                 list_add_tail(&bd->bd_list, &tr->tr_databuf);
 202         }
 203         gfs2_log_unlock(sdp);
 204 out:
 205         unlock_buffer(bh);
 206 }
 207
 208 void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
 209 {
 210
 211         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 212         struct gfs2_bufdata *bd;
 213         struct gfs2_meta_header *mh;
 214         struct gfs2_trans *tr = current->journal_info;
 215         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 216
 217         lock_buffer(bh);
 218         if (buffer_pinned(bh)) {
 219                 set_bit(TR_TOUCHED, &tr->tr_flags);
 220                 goto out;
 221         }
 222         gfs2_log_lock(sdp);
 223         bd = bh->b_private;
 224         if (bd == NULL) {
 225                 gfs2_log_unlock(sdp);
 226                 unlock_buffer(bh);
 227                 lock_page(bh->b_page);
 228                 if (bh->b_private == NULL)
 229                         bd = gfs2_alloc_bufdata(gl, bh);
 230                 else
 231                         bd = bh->b_private;
 232                 unlock_page(bh->b_page);
 233                 lock_buffer(bh);
 234                 gfs2_log_lock(sdp);
 235         }
 236         gfs2_assert(sdp, bd->bd_gl == gl);
 237         set_bit(TR_TOUCHED, &tr->tr_flags);
 238         if (!list_empty(&bd->bd_list))
 239                 goto out_unlock;
 240         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 241         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 242         mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
 243         if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) {
 244                 fs_err(sdp, "Attempting to add uninitialised block to "
 245                        "journal (inplace block=%lld)\n",
 246                        (unsigned long long)bd->bd_bh->b_blocknr);
 247                 BUG();
 248         }
 249         if (unlikely(state == SFS_FROZEN)) {
 250                 fs_info(sdp, "GFS2:adding buf while frozen\n");
 251                 gfs2_assert_withdraw(sdp, 0);
 252         }
 253         if (unlikely(gfs2_withdrawn(sdp))) {
 254                 fs_info(sdp, "GFS2:adding buf while withdrawn! 0x%llx\n",
 255                         (unsigned long long)bd->bd_bh->b_blocknr);
 256         }
 257         gfs2_pin(sdp, bd->bd_bh);
 258         mh->__pad0 = cpu_to_be64(0);
 259         mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 260         list_add(&bd->bd_list, &tr->tr_buf);
 261         tr->tr_num_buf_new++;
 262 out_unlock:
 263         gfs2_log_unlock(sdp);
 264 out:
 265         unlock_buffer(bh);
 266 }
 267
 268 void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 269 {
 270         struct gfs2_trans *tr = current->journal_info;
 271
 272         BUG_ON(!list_empty(&bd->bd_list));
 273         gfs2_add_revoke(sdp, bd);
 274         set_bit(TR_TOUCHED, &tr->tr_flags);
 275         tr->tr_num_revoke++;
 276 }
 277
 278 void gfs2_trans_remove_revoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len)
 279 {
 280         struct gfs2_bufdata *bd, *tmp;
 281         struct gfs2_trans *tr = current->journal_info;
 282         unsigned int n = len;
 283
 284         gfs2_log_lock(sdp);
 285         list_for_each_entry_safe(bd, tmp, &sdp->sd_log_revokes, bd_list) {
 286                 if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) {
 287                         list_del_init(&bd->bd_list);
 288                         gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke);
 289                         sdp->sd_log_num_revoke--;
 290                         if (bd->bd_gl)
 291                                 gfs2_glock_remove_revoke(bd->bd_gl);
 292                         kmem_cache_free(gfs2_bufdata_cachep, bd);
 293                         tr->tr_num_revoke_rm++;
 294                         if (--n == 0)
 295                                 break;
 296                 }
 297         }
 298         gfs2_log_unlock(sdp);
 299 }
 300
 301 void gfs2_trans_free(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 302 {
 303         if (tr == NULL)
 304                 return;
 305
 306         gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
 307         gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
 308         gfs2_assert_warn(sdp, list_empty(&tr->tr_databuf));
 309         gfs2_assert_warn(sdp, list_empty(&tr->tr_buf));
 310         kmem_cache_free(gfs2_trans_cachep, tr);
 311 }