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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11 12 #include <linux/sched.h> 13 #include <linux/slab.h> 14 #include <linux/spinlock.h> 15 #include <linux/completion.h> 16 #include <linux/buffer_head.h> 17 #include <linux/kallsyms.h> 18 #include <linux/gfs2_ondisk.h> 19 20 #include "gfs2.h" 21 #include "incore.h" 22 #include "glock.h" 23 #include "inode.h" 24 #include "log.h" 25 #include "lops.h" 26 #include "meta_io.h" 27 #include "trans.h" 28 #include "util.h" 29 #include "trace_gfs2.h" 30 31 int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks, 32 unsigned int revokes) 33 { 34 struct gfs2_trans *tr; 35 int error; 36 37 BUG_ON(current->journal_info); 38 BUG_ON(blocks == 0 && revokes == 0); 39 40 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) 41 return -EROFS; 42 43 tr = kzalloc(sizeof(struct gfs2_trans), GFP_NOFS); 44 if (!tr) 45 return -ENOMEM; 46 47 tr->tr_ip = _RET_IP_; 48 tr->tr_blocks = blocks; 49 tr->tr_revokes = revokes; 50 tr->tr_reserved = 1; 51 tr->tr_alloced = 1; 52 if (blocks) 53 tr->tr_reserved += 6 + blocks; 54 if (revokes) 55 tr->tr_reserved += gfs2_struct2blk(sdp, revokes, 56 sizeof(u64)); 57 INIT_LIST_HEAD(&tr->tr_databuf); 58 INIT_LIST_HEAD(&tr->tr_buf); 59 60 sb_start_intwrite(sdp->sd_vfs); 61 62 error = gfs2_log_reserve(sdp, tr->tr_reserved); 63 if (error) 64 goto fail; 65 66 current->journal_info = tr; 67 68 return 0; 69 70 fail: 71 sb_end_intwrite(sdp->sd_vfs); 72 kfree(tr); 73 74 return error; 75 } 76 77 static void gfs2_print_trans(const struct gfs2_trans *tr) 78 { 79 pr_warn("Transaction created at: %pSR\n", (void *)tr->tr_ip); 80 pr_warn("blocks=%u revokes=%u reserved=%u touched=%u\n", 81 tr->tr_blocks, tr->tr_revokes, tr->tr_reserved, tr->tr_touched); 82 pr_warn("Buf %u/%u Databuf %u/%u Revoke %u/%u\n", 83 tr->tr_num_buf_new, tr->tr_num_buf_rm, 84 tr->tr_num_databuf_new, tr->tr_num_databuf_rm, 85 tr->tr_num_revoke, tr->tr_num_revoke_rm); 86 } 87 88 void gfs2_trans_end(struct gfs2_sbd *sdp) 89 { 90 struct gfs2_trans *tr = current->journal_info; 91 s64 nbuf; 92 BUG_ON(!tr); 93 current->journal_info = NULL; 94 95 if (!tr->tr_touched) { 96 gfs2_log_release(sdp, tr->tr_reserved); 97 if (tr->tr_alloced) 98 kfree(tr); 99 sb_end_intwrite(sdp->sd_vfs); 100 return; 101 } 102 103 nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new; 104 nbuf -= tr->tr_num_buf_rm; 105 nbuf -= tr->tr_num_databuf_rm; 106 107 if (gfs2_assert_withdraw(sdp, (nbuf <= tr->tr_blocks) && 108 (tr->tr_num_revoke <= tr->tr_revokes))) 109 gfs2_print_trans(tr); 110 111 gfs2_log_commit(sdp, tr); 112 if (tr->tr_alloced && !tr->tr_attached) 113 kfree(tr); 114 up_read(&sdp->sd_log_flush_lock); 115 116 if (sdp->sd_vfs->s_flags & MS_SYNCHRONOUS) 117 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH); 118 sb_end_intwrite(sdp->sd_vfs); 119 } 120 121 static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl, 122 struct buffer_head *bh, 123 const struct gfs2_log_operations *lops) 124 { 125 struct gfs2_bufdata *bd; 126 127 bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL); 128 bd->bd_bh = bh; 129 bd->bd_gl = gl; 130 bd->bd_ops = lops; 131 INIT_LIST_HEAD(&bd->bd_list); 132 bh->b_private = bd; 133 return bd; 134 } 135 136 /** 137 * gfs2_trans_add_data - Add a databuf to the transaction. 138 * @gl: The inode glock associated with the buffer 139 * @bh: The buffer to add 140 * 141 * This is used in two distinct cases: 142 * i) In ordered write mode 143 * We put the data buffer on a list so that we can ensure that its 144 * synced to disk at the right time 145 * ii) In journaled data mode 146 * We need to journal the data block in the same way as metadata in 147 * the functions above. The difference is that here we have a tag 148 * which is two __be64's being the block number (as per meta data) 149 * and a flag which says whether the data block needs escaping or 150 * not. This means we need a new log entry for each 251 or so data 151 * blocks, which isn't an enormous overhead but twice as much as 152 * for normal metadata blocks. 153 */ 154 void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh) 155 { 156 struct gfs2_trans *tr = current->journal_info; 157 struct gfs2_sbd *sdp = gl->gl_sbd; 158 struct address_space *mapping = bh->b_page->mapping; 159 struct gfs2_inode *ip = GFS2_I(mapping->host); 160 struct gfs2_bufdata *bd; 161 162 if (!gfs2_is_jdata(ip)) { 163 gfs2_ordered_add_inode(ip); 164 return; 165 } 166 167 lock_buffer(bh); 168 gfs2_log_lock(sdp); 169 bd = bh->b_private; 170 if (bd == NULL) { 171 gfs2_log_unlock(sdp); 172 unlock_buffer(bh); 173 if (bh->b_private == NULL) 174 bd = gfs2_alloc_bufdata(gl, bh, &gfs2_databuf_lops); 175 lock_buffer(bh); 176 gfs2_log_lock(sdp); 177 } 178 gfs2_assert(sdp, bd->bd_gl == gl); 179 tr->tr_touched = 1; 180 if (list_empty(&bd->bd_list)) { 181 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 182 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); 183 gfs2_pin(sdp, bd->bd_bh); 184 tr->tr_num_databuf_new++; 185 list_add_tail(&bd->bd_list, &tr->tr_databuf); 186 } 187 gfs2_log_unlock(sdp); 188 unlock_buffer(bh); 189 } 190 191 static void meta_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) 192 { 193 struct gfs2_meta_header *mh; 194 struct gfs2_trans *tr; 195 196 tr = current->journal_info; 197 tr->tr_touched = 1; 198 if (!list_empty(&bd->bd_list)) 199 return; 200 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 201 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); 202 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data; 203 if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) { 204 pr_err("Attempting to add uninitialised block to journal (inplace block=%lld)\n", 205 (unsigned long long)bd->bd_bh->b_blocknr); 206 BUG(); 207 } 208 gfs2_pin(sdp, bd->bd_bh); 209 mh->__pad0 = cpu_to_be64(0); 210 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); 211 list_add(&bd->bd_list, &tr->tr_buf); 212 tr->tr_num_buf_new++; 213 } 214 215 void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh) 216 { 217 218 struct gfs2_sbd *sdp = gl->gl_sbd; 219 struct gfs2_bufdata *bd; 220 221 lock_buffer(bh); 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, &gfs2_buf_lops); 230 unlock_page(bh->b_page); 231 lock_buffer(bh); 232 gfs2_log_lock(sdp); 233 } 234 gfs2_assert(sdp, bd->bd_gl == gl); 235 meta_lo_add(sdp, bd); 236 gfs2_log_unlock(sdp); 237 unlock_buffer(bh); 238 } 239 240 void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) 241 { 242 struct gfs2_trans *tr = current->journal_info; 243 244 BUG_ON(!list_empty(&bd->bd_list)); 245 gfs2_add_revoke(sdp, bd); 246 tr->tr_touched = 1; 247 tr->tr_num_revoke++; 248 } 249 250 void gfs2_trans_add_unrevoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len) 251 { 252 struct gfs2_bufdata *bd, *tmp; 253 struct gfs2_trans *tr = current->journal_info; 254 unsigned int n = len; 255 256 gfs2_log_lock(sdp); 257 list_for_each_entry_safe(bd, tmp, &sdp->sd_log_le_revoke, bd_list) { 258 if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) { 259 list_del_init(&bd->bd_list); 260 gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke); 261 sdp->sd_log_num_revoke--; 262 kmem_cache_free(gfs2_bufdata_cachep, bd); 263 tr->tr_num_revoke_rm++; 264 if (--n == 0) 265 break; 266 } 267 } 268 gfs2_log_unlock(sdp); 269 } 270 271