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 set_bit(TR_ALLOCED, &tr->tr_flags); 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, 82 test_bit(TR_TOUCHED, &tr->tr_flags)); 83 pr_warn("Buf %u/%u Databuf %u/%u Revoke %u/%u\n", 84 tr->tr_num_buf_new, tr->tr_num_buf_rm, 85 tr->tr_num_databuf_new, tr->tr_num_databuf_rm, 86 tr->tr_num_revoke, tr->tr_num_revoke_rm); 87 } 88 89 void gfs2_trans_end(struct gfs2_sbd *sdp) 90 { 91 struct gfs2_trans *tr = current->journal_info; 92 s64 nbuf; 93 int alloced = test_bit(TR_ALLOCED, &tr->tr_flags); 94 95 current->journal_info = NULL; 96 97 if (!test_bit(TR_TOUCHED, &tr->tr_flags)) { 98 gfs2_log_release(sdp, tr->tr_reserved); 99 if (alloced) { 100 kfree(tr); 101 sb_end_intwrite(sdp->sd_vfs); 102 } 103 return; 104 } 105 106 nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new; 107 nbuf -= tr->tr_num_buf_rm; 108 nbuf -= tr->tr_num_databuf_rm; 109 110 if (gfs2_assert_withdraw(sdp, (nbuf <= tr->tr_blocks) && 111 (tr->tr_num_revoke <= tr->tr_revokes))) 112 gfs2_print_trans(tr); 113 114 gfs2_log_commit(sdp, tr); 115 if (alloced && !test_bit(TR_ATTACHED, &tr->tr_flags)) 116 kfree(tr); 117 up_read(&sdp->sd_log_flush_lock); 118 119 if (sdp->sd_vfs->s_flags & SB_SYNCHRONOUS) 120 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL); 121 if (alloced) 122 sb_end_intwrite(sdp->sd_vfs); 123 } 124 125 static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl, 126 struct buffer_head *bh, 127 const struct gfs2_log_operations *lops) 128 { 129 struct gfs2_bufdata *bd; 130 131 bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL); 132 bd->bd_bh = bh; 133 bd->bd_gl = gl; 134 bd->bd_ops = lops; 135 INIT_LIST_HEAD(&bd->bd_list); 136 bh->b_private = bd; 137 return bd; 138 } 139 140 /** 141 * gfs2_trans_add_data - Add a databuf to the transaction. 142 * @gl: The inode glock associated with the buffer 143 * @bh: The buffer to add 144 * 145 * This is used in two distinct cases: 146 * i) In ordered write mode 147 * We put the data buffer on a list so that we can ensure that it's 148 * synced to disk at the right time 149 * ii) In journaled data mode 150 * We need to journal the data block in the same way as metadata in 151 * the functions above. The difference is that here we have a tag 152 * which is two __be64's being the block number (as per meta data) 153 * and a flag which says whether the data block needs escaping or 154 * not. This means we need a new log entry for each 251 or so data 155 * blocks, which isn't an enormous overhead but twice as much as 156 * for normal metadata blocks. 157 */ 158 void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh) 159 { 160 struct gfs2_trans *tr = current->journal_info; 161 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 162 struct address_space *mapping = bh->b_page->mapping; 163 struct gfs2_inode *ip = GFS2_I(mapping->host); 164 struct gfs2_bufdata *bd; 165 166 if (!gfs2_is_jdata(ip)) { 167 gfs2_ordered_add_inode(ip); 168 return; 169 } 170 171 lock_buffer(bh); 172 if (buffer_pinned(bh)) { 173 set_bit(TR_TOUCHED, &tr->tr_flags); 174 goto out; 175 } 176 gfs2_log_lock(sdp); 177 bd = bh->b_private; 178 if (bd == NULL) { 179 gfs2_log_unlock(sdp); 180 unlock_buffer(bh); 181 if (bh->b_private == NULL) 182 bd = gfs2_alloc_bufdata(gl, bh, &gfs2_databuf_lops); 183 else 184 bd = bh->b_private; 185 lock_buffer(bh); 186 gfs2_log_lock(sdp); 187 } 188 gfs2_assert(sdp, bd->bd_gl == gl); 189 set_bit(TR_TOUCHED, &tr->tr_flags); 190 if (list_empty(&bd->bd_list)) { 191 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 192 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); 193 gfs2_pin(sdp, bd->bd_bh); 194 tr->tr_num_databuf_new++; 195 list_add_tail(&bd->bd_list, &tr->tr_databuf); 196 } 197 gfs2_log_unlock(sdp); 198 out: 199 unlock_buffer(bh); 200 } 201 202 void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh) 203 { 204 205 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 206 struct gfs2_bufdata *bd; 207 struct gfs2_meta_header *mh; 208 struct gfs2_trans *tr = current->journal_info; 209 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); 210 211 lock_buffer(bh); 212 if (buffer_pinned(bh)) { 213 set_bit(TR_TOUCHED, &tr->tr_flags); 214 goto out; 215 } 216 gfs2_log_lock(sdp); 217 bd = bh->b_private; 218 if (bd == NULL) { 219 gfs2_log_unlock(sdp); 220 unlock_buffer(bh); 221 lock_page(bh->b_page); 222 if (bh->b_private == NULL) 223 bd = gfs2_alloc_bufdata(gl, bh, &gfs2_buf_lops); 224 else 225 bd = bh->b_private; 226 unlock_page(bh->b_page); 227 lock_buffer(bh); 228 gfs2_log_lock(sdp); 229 } 230 gfs2_assert(sdp, bd->bd_gl == gl); 231 set_bit(TR_TOUCHED, &tr->tr_flags); 232 if (!list_empty(&bd->bd_list)) 233 goto out_unlock; 234 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 235 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); 236 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data; 237 if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) { 238 pr_err("Attempting to add uninitialised block to journal (inplace block=%lld)\n", 239 (unsigned long long)bd->bd_bh->b_blocknr); 240 BUG(); 241 } 242 if (unlikely(state == SFS_FROZEN)) { 243 printk(KERN_INFO "GFS2:adding buf while frozen\n"); 244 gfs2_assert_withdraw(sdp, 0); 245 } 246 gfs2_pin(sdp, bd->bd_bh); 247 mh->__pad0 = cpu_to_be64(0); 248 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); 249 list_add(&bd->bd_list, &tr->tr_buf); 250 tr->tr_num_buf_new++; 251 out_unlock: 252 gfs2_log_unlock(sdp); 253 out: 254 unlock_buffer(bh); 255 } 256 257 void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) 258 { 259 struct gfs2_trans *tr = current->journal_info; 260 261 BUG_ON(!list_empty(&bd->bd_list)); 262 gfs2_add_revoke(sdp, bd); 263 set_bit(TR_TOUCHED, &tr->tr_flags); 264 tr->tr_num_revoke++; 265 } 266 267 void gfs2_trans_add_unrevoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len) 268 { 269 struct gfs2_bufdata *bd, *tmp; 270 struct gfs2_trans *tr = current->journal_info; 271 unsigned int n = len; 272 273 gfs2_log_lock(sdp); 274 list_for_each_entry_safe(bd, tmp, &sdp->sd_log_le_revoke, bd_list) { 275 if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) { 276 list_del_init(&bd->bd_list); 277 gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke); 278 sdp->sd_log_num_revoke--; 279 kmem_cache_free(gfs2_bufdata_cachep, bd); 280 tr->tr_num_revoke_rm++; 281 if (--n == 0) 282 break; 283 } 284 } 285 gfs2_log_unlock(sdp); 286 } 287 288