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 int alloced = tr->tr_alloced; 93 94 BUG_ON(!tr); 95 current->journal_info = NULL; 96 97 if (!tr->tr_touched) { 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 && !tr->tr_attached) 116 kfree(tr); 117 up_read(&sdp->sd_log_flush_lock); 118 119 if (sdp->sd_vfs->s_flags & MS_SYNCHRONOUS) 120 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH); 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 its 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 gfs2_log_lock(sdp); 173 bd = bh->b_private; 174 if (bd == NULL) { 175 gfs2_log_unlock(sdp); 176 unlock_buffer(bh); 177 if (bh->b_private == NULL) 178 bd = gfs2_alloc_bufdata(gl, bh, &gfs2_databuf_lops); 179 else 180 bd = bh->b_private; 181 lock_buffer(bh); 182 gfs2_log_lock(sdp); 183 } 184 gfs2_assert(sdp, bd->bd_gl == gl); 185 tr->tr_touched = 1; 186 if (list_empty(&bd->bd_list)) { 187 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 188 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); 189 gfs2_pin(sdp, bd->bd_bh); 190 tr->tr_num_databuf_new++; 191 list_add_tail(&bd->bd_list, &tr->tr_databuf); 192 } 193 gfs2_log_unlock(sdp); 194 unlock_buffer(bh); 195 } 196 197 static void meta_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) 198 { 199 struct gfs2_meta_header *mh; 200 struct gfs2_trans *tr; 201 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); 202 203 tr = current->journal_info; 204 tr->tr_touched = 1; 205 if (!list_empty(&bd->bd_list)) 206 return; 207 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 208 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); 209 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data; 210 if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) { 211 pr_err("Attempting to add uninitialised block to journal (inplace block=%lld)\n", 212 (unsigned long long)bd->bd_bh->b_blocknr); 213 BUG(); 214 } 215 if (unlikely(state == SFS_FROZEN)) { 216 printk(KERN_INFO "GFS2:adding buf while frozen\n"); 217 gfs2_assert_withdraw(sdp, 0); 218 } 219 gfs2_pin(sdp, bd->bd_bh); 220 mh->__pad0 = cpu_to_be64(0); 221 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); 222 list_add(&bd->bd_list, &tr->tr_buf); 223 tr->tr_num_buf_new++; 224 } 225 226 void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh) 227 { 228 229 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 230 struct gfs2_bufdata *bd; 231 232 lock_buffer(bh); 233 gfs2_log_lock(sdp); 234 bd = bh->b_private; 235 if (bd == NULL) { 236 gfs2_log_unlock(sdp); 237 unlock_buffer(bh); 238 lock_page(bh->b_page); 239 if (bh->b_private == NULL) 240 bd = gfs2_alloc_bufdata(gl, bh, &gfs2_buf_lops); 241 else 242 bd = bh->b_private; 243 unlock_page(bh->b_page); 244 lock_buffer(bh); 245 gfs2_log_lock(sdp); 246 } 247 gfs2_assert(sdp, bd->bd_gl == gl); 248 meta_lo_add(sdp, bd); 249 gfs2_log_unlock(sdp); 250 unlock_buffer(bh); 251 } 252 253 void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) 254 { 255 struct gfs2_trans *tr = current->journal_info; 256 257 BUG_ON(!list_empty(&bd->bd_list)); 258 gfs2_add_revoke(sdp, bd); 259 tr->tr_touched = 1; 260 tr->tr_num_revoke++; 261 } 262 263 void gfs2_trans_add_unrevoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len) 264 { 265 struct gfs2_bufdata *bd, *tmp; 266 struct gfs2_trans *tr = current->journal_info; 267 unsigned int n = len; 268 269 gfs2_log_lock(sdp); 270 list_for_each_entry_safe(bd, tmp, &sdp->sd_log_le_revoke, bd_list) { 271 if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) { 272 list_del_init(&bd->bd_list); 273 gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke); 274 sdp->sd_log_num_revoke--; 275 kmem_cache_free(gfs2_bufdata_cachep, bd); 276 tr->tr_num_revoke_rm++; 277 if (--n == 0) 278 break; 279 } 280 } 281 gfs2_log_unlock(sdp); 282 } 283 284