1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2008 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/slab.h> 11 #include <linux/spinlock.h> 12 #include <linux/completion.h> 13 #include <linux/buffer_head.h> 14 #include <linux/gfs2_ondisk.h> 15 #include <linux/bio.h> 16 17 #include "gfs2.h" 18 #include "incore.h" 19 #include "bmap.h" 20 #include "glock.h" 21 #include "glops.h" 22 #include "inode.h" 23 #include "log.h" 24 #include "meta_io.h" 25 #include "recovery.h" 26 #include "rgrp.h" 27 #include "util.h" 28 #include "trans.h" 29 30 /** 31 * ail_empty_gl - remove all buffers for a given lock from the AIL 32 * @gl: the glock 33 * 34 * None of the buffers should be dirty, locked, or pinned. 35 */ 36 37 static void gfs2_ail_empty_gl(struct gfs2_glock *gl) 38 { 39 struct gfs2_sbd *sdp = gl->gl_sbd; 40 struct list_head *head = &gl->gl_ail_list; 41 struct gfs2_bufdata *bd; 42 struct buffer_head *bh; 43 struct gfs2_trans tr; 44 45 memset(&tr, 0, sizeof(tr)); 46 tr.tr_revokes = atomic_read(&gl->gl_ail_count); 47 48 if (!tr.tr_revokes) 49 return; 50 51 /* A shortened, inline version of gfs2_trans_begin() */ 52 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64)); 53 tr.tr_ip = (unsigned long)__builtin_return_address(0); 54 INIT_LIST_HEAD(&tr.tr_list_buf); 55 gfs2_log_reserve(sdp, tr.tr_reserved); 56 BUG_ON(current->journal_info); 57 current->journal_info = &tr; 58 59 gfs2_log_lock(sdp); 60 while (!list_empty(head)) { 61 bd = list_entry(head->next, struct gfs2_bufdata, 62 bd_ail_gl_list); 63 bh = bd->bd_bh; 64 gfs2_remove_from_ail(bd); 65 bd->bd_bh = NULL; 66 bh->b_private = NULL; 67 bd->bd_blkno = bh->b_blocknr; 68 gfs2_assert_withdraw(sdp, !buffer_busy(bh)); 69 gfs2_trans_add_revoke(sdp, bd); 70 } 71 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count)); 72 gfs2_log_unlock(sdp); 73 74 gfs2_trans_end(sdp); 75 gfs2_log_flush(sdp, NULL); 76 } 77 78 /** 79 * rgrp_go_sync - sync out the metadata for this glock 80 * @gl: the glock 81 * 82 * Called when demoting or unlocking an EX glock. We must flush 83 * to disk all dirty buffers/pages relating to this glock, and must not 84 * not return to caller to demote/unlock the glock until I/O is complete. 85 */ 86 87 static void rgrp_go_sync(struct gfs2_glock *gl) 88 { 89 struct address_space *metamapping = gl->gl_aspace->i_mapping; 90 int error; 91 92 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) 93 return; 94 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE); 95 96 gfs2_log_flush(gl->gl_sbd, gl); 97 filemap_fdatawrite(metamapping); 98 error = filemap_fdatawait(metamapping); 99 mapping_set_error(metamapping, error); 100 gfs2_ail_empty_gl(gl); 101 } 102 103 /** 104 * rgrp_go_inval - invalidate the metadata for this glock 105 * @gl: the glock 106 * @flags: 107 * 108 * We never used LM_ST_DEFERRED with resource groups, so that we 109 * should always see the metadata flag set here. 110 * 111 */ 112 113 static void rgrp_go_inval(struct gfs2_glock *gl, int flags) 114 { 115 struct address_space *mapping = gl->gl_aspace->i_mapping; 116 117 BUG_ON(!(flags & DIO_METADATA)); 118 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count)); 119 truncate_inode_pages(mapping, 0); 120 121 if (gl->gl_object) { 122 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object; 123 rgd->rd_flags &= ~GFS2_RDF_UPTODATE; 124 } 125 } 126 127 /** 128 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock 129 * @gl: the glock protecting the inode 130 * 131 */ 132 133 static void inode_go_sync(struct gfs2_glock *gl) 134 { 135 struct gfs2_inode *ip = gl->gl_object; 136 struct address_space *metamapping = gl->gl_aspace->i_mapping; 137 int error; 138 139 if (ip && !S_ISREG(ip->i_inode.i_mode)) 140 ip = NULL; 141 if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags)) 142 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0); 143 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) 144 return; 145 146 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE); 147 148 gfs2_log_flush(gl->gl_sbd, gl); 149 filemap_fdatawrite(metamapping); 150 if (ip) { 151 struct address_space *mapping = ip->i_inode.i_mapping; 152 filemap_fdatawrite(mapping); 153 error = filemap_fdatawait(mapping); 154 mapping_set_error(mapping, error); 155 } 156 error = filemap_fdatawait(metamapping); 157 mapping_set_error(metamapping, error); 158 gfs2_ail_empty_gl(gl); 159 /* 160 * Writeback of the data mapping may cause the dirty flag to be set 161 * so we have to clear it again here. 162 */ 163 smp_mb__before_clear_bit(); 164 clear_bit(GLF_DIRTY, &gl->gl_flags); 165 } 166 167 /** 168 * inode_go_inval - prepare a inode glock to be released 169 * @gl: the glock 170 * @flags: 171 * 172 * Normally we invlidate everything, but if we are moving into 173 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we 174 * can keep hold of the metadata, since it won't have changed. 175 * 176 */ 177 178 static void inode_go_inval(struct gfs2_glock *gl, int flags) 179 { 180 struct gfs2_inode *ip = gl->gl_object; 181 182 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count)); 183 184 if (flags & DIO_METADATA) { 185 struct address_space *mapping = gl->gl_aspace->i_mapping; 186 truncate_inode_pages(mapping, 0); 187 if (ip) 188 set_bit(GIF_INVALID, &ip->i_flags); 189 } 190 191 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) 192 gl->gl_sbd->sd_rindex_uptodate = 0; 193 if (ip && S_ISREG(ip->i_inode.i_mode)) 194 truncate_inode_pages(ip->i_inode.i_mapping, 0); 195 } 196 197 /** 198 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock 199 * @gl: the glock 200 * 201 * Returns: 1 if it's ok 202 */ 203 204 static int inode_go_demote_ok(const struct gfs2_glock *gl) 205 { 206 struct gfs2_sbd *sdp = gl->gl_sbd; 207 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object) 208 return 0; 209 return 1; 210 } 211 212 /** 213 * inode_go_lock - operation done after an inode lock is locked by a process 214 * @gl: the glock 215 * @flags: 216 * 217 * Returns: errno 218 */ 219 220 static int inode_go_lock(struct gfs2_holder *gh) 221 { 222 struct gfs2_glock *gl = gh->gh_gl; 223 struct gfs2_sbd *sdp = gl->gl_sbd; 224 struct gfs2_inode *ip = gl->gl_object; 225 int error = 0; 226 227 if (!ip || (gh->gh_flags & GL_SKIP)) 228 return 0; 229 230 if (test_bit(GIF_INVALID, &ip->i_flags)) { 231 error = gfs2_inode_refresh(ip); 232 if (error) 233 return error; 234 } 235 236 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) && 237 (gl->gl_state == LM_ST_EXCLUSIVE) && 238 (gh->gh_state == LM_ST_EXCLUSIVE)) { 239 spin_lock(&sdp->sd_trunc_lock); 240 if (list_empty(&ip->i_trunc_list)) 241 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list); 242 spin_unlock(&sdp->sd_trunc_lock); 243 wake_up(&sdp->sd_quota_wait); 244 return 1; 245 } 246 247 return error; 248 } 249 250 /** 251 * inode_go_dump - print information about an inode 252 * @seq: The iterator 253 * @ip: the inode 254 * 255 * Returns: 0 on success, -ENOBUFS when we run out of space 256 */ 257 258 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl) 259 { 260 const struct gfs2_inode *ip = gl->gl_object; 261 if (ip == NULL) 262 return 0; 263 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu/%llu\n", 264 (unsigned long long)ip->i_no_formal_ino, 265 (unsigned long long)ip->i_no_addr, 266 IF2DT(ip->i_inode.i_mode), ip->i_flags, 267 (unsigned int)ip->i_diskflags, 268 (unsigned long long)ip->i_inode.i_size, 269 (unsigned long long)ip->i_disksize); 270 return 0; 271 } 272 273 /** 274 * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock 275 * @gl: the glock 276 * 277 * Returns: 1 if it's ok 278 */ 279 280 static int rgrp_go_demote_ok(const struct gfs2_glock *gl) 281 { 282 return !gl->gl_aspace->i_mapping->nrpages; 283 } 284 285 /** 286 * rgrp_go_lock - operation done after an rgrp lock is locked by 287 * a first holder on this node. 288 * @gl: the glock 289 * @flags: 290 * 291 * Returns: errno 292 */ 293 294 static int rgrp_go_lock(struct gfs2_holder *gh) 295 { 296 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object); 297 } 298 299 /** 300 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by 301 * a last holder on this node. 302 * @gl: the glock 303 * @flags: 304 * 305 */ 306 307 static void rgrp_go_unlock(struct gfs2_holder *gh) 308 { 309 gfs2_rgrp_bh_put(gh->gh_gl->gl_object); 310 } 311 312 /** 313 * trans_go_sync - promote/demote the transaction glock 314 * @gl: the glock 315 * @state: the requested state 316 * @flags: 317 * 318 */ 319 320 static void trans_go_sync(struct gfs2_glock *gl) 321 { 322 struct gfs2_sbd *sdp = gl->gl_sbd; 323 324 if (gl->gl_state != LM_ST_UNLOCKED && 325 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 326 gfs2_meta_syncfs(sdp); 327 gfs2_log_shutdown(sdp); 328 } 329 } 330 331 /** 332 * trans_go_xmote_bh - After promoting/demoting the transaction glock 333 * @gl: the glock 334 * 335 */ 336 337 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh) 338 { 339 struct gfs2_sbd *sdp = gl->gl_sbd; 340 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); 341 struct gfs2_glock *j_gl = ip->i_gl; 342 struct gfs2_log_header_host head; 343 int error; 344 345 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 346 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); 347 348 error = gfs2_find_jhead(sdp->sd_jdesc, &head); 349 if (error) 350 gfs2_consist(sdp); 351 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) 352 gfs2_consist(sdp); 353 354 /* Initialize some head of the log stuff */ 355 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) { 356 sdp->sd_log_sequence = head.lh_sequence + 1; 357 gfs2_log_pointers_init(sdp, head.lh_blkno); 358 } 359 } 360 return 0; 361 } 362 363 /** 364 * trans_go_demote_ok 365 * @gl: the glock 366 * 367 * Always returns 0 368 */ 369 370 static int trans_go_demote_ok(const struct gfs2_glock *gl) 371 { 372 return 0; 373 } 374 375 const struct gfs2_glock_operations gfs2_meta_glops = { 376 .go_type = LM_TYPE_META, 377 }; 378 379 const struct gfs2_glock_operations gfs2_inode_glops = { 380 .go_xmote_th = inode_go_sync, 381 .go_inval = inode_go_inval, 382 .go_demote_ok = inode_go_demote_ok, 383 .go_lock = inode_go_lock, 384 .go_dump = inode_go_dump, 385 .go_type = LM_TYPE_INODE, 386 .go_min_hold_time = HZ / 5, 387 }; 388 389 const struct gfs2_glock_operations gfs2_rgrp_glops = { 390 .go_xmote_th = rgrp_go_sync, 391 .go_inval = rgrp_go_inval, 392 .go_demote_ok = rgrp_go_demote_ok, 393 .go_lock = rgrp_go_lock, 394 .go_unlock = rgrp_go_unlock, 395 .go_dump = gfs2_rgrp_dump, 396 .go_type = LM_TYPE_RGRP, 397 .go_min_hold_time = HZ / 5, 398 }; 399 400 const struct gfs2_glock_operations gfs2_trans_glops = { 401 .go_xmote_th = trans_go_sync, 402 .go_xmote_bh = trans_go_xmote_bh, 403 .go_demote_ok = trans_go_demote_ok, 404 .go_type = LM_TYPE_NONDISK, 405 }; 406 407 const struct gfs2_glock_operations gfs2_iopen_glops = { 408 .go_type = LM_TYPE_IOPEN, 409 }; 410 411 const struct gfs2_glock_operations gfs2_flock_glops = { 412 .go_type = LM_TYPE_FLOCK, 413 }; 414 415 const struct gfs2_glock_operations gfs2_nondisk_glops = { 416 .go_type = LM_TYPE_NONDISK, 417 }; 418 419 const struct gfs2_glock_operations gfs2_quota_glops = { 420 .go_type = LM_TYPE_QUOTA, 421 }; 422 423 const struct gfs2_glock_operations gfs2_journal_glops = { 424 .go_type = LM_TYPE_JOURNAL, 425 }; 426 427 const struct gfs2_glock_operations *gfs2_glops_list[] = { 428 [LM_TYPE_META] = &gfs2_meta_glops, 429 [LM_TYPE_INODE] = &gfs2_inode_glops, 430 [LM_TYPE_RGRP] = &gfs2_rgrp_glops, 431 [LM_TYPE_NONDISK] = &gfs2_trans_glops, 432 [LM_TYPE_IOPEN] = &gfs2_iopen_glops, 433 [LM_TYPE_FLOCK] = &gfs2_flock_glops, 434 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops, 435 [LM_TYPE_QUOTA] = &gfs2_quota_glops, 436 [LM_TYPE_JOURNAL] = &gfs2_journal_glops, 437 }; 438 439