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/spinlock.h> 11 #include <linux/completion.h> 12 #include <linux/buffer_head.h> 13 #include <linux/gfs2_ondisk.h> 14 #include <linux/bio.h> 15 #include <linux/posix_acl.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 spin_lock(&sdp->sd_ail_lock); 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 spin_unlock(&sdp->sd_ail_lock); 66 67 bd->bd_bh = NULL; 68 bh->b_private = NULL; 69 bd->bd_blkno = bh->b_blocknr; 70 gfs2_log_lock(sdp); 71 gfs2_assert_withdraw(sdp, !buffer_busy(bh)); 72 gfs2_trans_add_revoke(sdp, bd); 73 gfs2_log_unlock(sdp); 74 75 spin_lock(&sdp->sd_ail_lock); 76 } 77 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count)); 78 spin_unlock(&sdp->sd_ail_lock); 79 80 gfs2_trans_end(sdp); 81 gfs2_log_flush(sdp, NULL); 82 } 83 84 /** 85 * rgrp_go_sync - sync out the metadata for this glock 86 * @gl: the glock 87 * 88 * Called when demoting or unlocking an EX glock. We must flush 89 * to disk all dirty buffers/pages relating to this glock, and must not 90 * not return to caller to demote/unlock the glock until I/O is complete. 91 */ 92 93 static void rgrp_go_sync(struct gfs2_glock *gl) 94 { 95 struct address_space *metamapping = gfs2_glock2aspace(gl); 96 int error; 97 98 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) 99 return; 100 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE); 101 102 gfs2_log_flush(gl->gl_sbd, gl); 103 filemap_fdatawrite(metamapping); 104 error = filemap_fdatawait(metamapping); 105 mapping_set_error(metamapping, error); 106 gfs2_ail_empty_gl(gl); 107 } 108 109 /** 110 * rgrp_go_inval - invalidate the metadata for this glock 111 * @gl: the glock 112 * @flags: 113 * 114 * We never used LM_ST_DEFERRED with resource groups, so that we 115 * should always see the metadata flag set here. 116 * 117 */ 118 119 static void rgrp_go_inval(struct gfs2_glock *gl, int flags) 120 { 121 struct address_space *mapping = gfs2_glock2aspace(gl); 122 123 BUG_ON(!(flags & DIO_METADATA)); 124 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count)); 125 truncate_inode_pages(mapping, 0); 126 127 if (gl->gl_object) { 128 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object; 129 rgd->rd_flags &= ~GFS2_RDF_UPTODATE; 130 } 131 } 132 133 /** 134 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock 135 * @gl: the glock protecting the inode 136 * 137 */ 138 139 static void inode_go_sync(struct gfs2_glock *gl) 140 { 141 struct gfs2_inode *ip = gl->gl_object; 142 struct address_space *metamapping = gfs2_glock2aspace(gl); 143 int error; 144 145 if (ip && !S_ISREG(ip->i_inode.i_mode)) 146 ip = NULL; 147 if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags)) 148 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0); 149 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) 150 return; 151 152 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE); 153 154 gfs2_log_flush(gl->gl_sbd, gl); 155 filemap_fdatawrite(metamapping); 156 if (ip) { 157 struct address_space *mapping = ip->i_inode.i_mapping; 158 filemap_fdatawrite(mapping); 159 error = filemap_fdatawait(mapping); 160 mapping_set_error(mapping, error); 161 } 162 error = filemap_fdatawait(metamapping); 163 mapping_set_error(metamapping, error); 164 gfs2_ail_empty_gl(gl); 165 /* 166 * Writeback of the data mapping may cause the dirty flag to be set 167 * so we have to clear it again here. 168 */ 169 smp_mb__before_clear_bit(); 170 clear_bit(GLF_DIRTY, &gl->gl_flags); 171 } 172 173 /** 174 * inode_go_inval - prepare a inode glock to be released 175 * @gl: the glock 176 * @flags: 177 * 178 * Normally we invlidate everything, but if we are moving into 179 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we 180 * can keep hold of the metadata, since it won't have changed. 181 * 182 */ 183 184 static void inode_go_inval(struct gfs2_glock *gl, int flags) 185 { 186 struct gfs2_inode *ip = gl->gl_object; 187 188 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count)); 189 190 if (flags & DIO_METADATA) { 191 struct address_space *mapping = gfs2_glock2aspace(gl); 192 truncate_inode_pages(mapping, 0); 193 if (ip) { 194 set_bit(GIF_INVALID, &ip->i_flags); 195 forget_all_cached_acls(&ip->i_inode); 196 } 197 } 198 199 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) 200 gl->gl_sbd->sd_rindex_uptodate = 0; 201 if (ip && S_ISREG(ip->i_inode.i_mode)) 202 truncate_inode_pages(ip->i_inode.i_mapping, 0); 203 } 204 205 /** 206 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock 207 * @gl: the glock 208 * 209 * Returns: 1 if it's ok 210 */ 211 212 static int inode_go_demote_ok(const struct gfs2_glock *gl) 213 { 214 struct gfs2_sbd *sdp = gl->gl_sbd; 215 struct gfs2_holder *gh; 216 217 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object) 218 return 0; 219 220 if (!list_empty(&gl->gl_holders)) { 221 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list); 222 if (gh->gh_list.next != &gl->gl_holders) 223 return 0; 224 } 225 226 return 1; 227 } 228 229 /** 230 * inode_go_lock - operation done after an inode lock is locked by a process 231 * @gl: the glock 232 * @flags: 233 * 234 * Returns: errno 235 */ 236 237 static int inode_go_lock(struct gfs2_holder *gh) 238 { 239 struct gfs2_glock *gl = gh->gh_gl; 240 struct gfs2_sbd *sdp = gl->gl_sbd; 241 struct gfs2_inode *ip = gl->gl_object; 242 int error = 0; 243 244 if (!ip || (gh->gh_flags & GL_SKIP)) 245 return 0; 246 247 if (test_bit(GIF_INVALID, &ip->i_flags)) { 248 error = gfs2_inode_refresh(ip); 249 if (error) 250 return error; 251 } 252 253 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) && 254 (gl->gl_state == LM_ST_EXCLUSIVE) && 255 (gh->gh_state == LM_ST_EXCLUSIVE)) { 256 spin_lock(&sdp->sd_trunc_lock); 257 if (list_empty(&ip->i_trunc_list)) 258 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list); 259 spin_unlock(&sdp->sd_trunc_lock); 260 wake_up(&sdp->sd_quota_wait); 261 return 1; 262 } 263 264 return error; 265 } 266 267 /** 268 * inode_go_dump - print information about an inode 269 * @seq: The iterator 270 * @ip: the inode 271 * 272 * Returns: 0 on success, -ENOBUFS when we run out of space 273 */ 274 275 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl) 276 { 277 const struct gfs2_inode *ip = gl->gl_object; 278 if (ip == NULL) 279 return 0; 280 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n", 281 (unsigned long long)ip->i_no_formal_ino, 282 (unsigned long long)ip->i_no_addr, 283 IF2DT(ip->i_inode.i_mode), ip->i_flags, 284 (unsigned int)ip->i_diskflags, 285 (unsigned long long)i_size_read(&ip->i_inode)); 286 return 0; 287 } 288 289 /** 290 * rgrp_go_lock - operation done after an rgrp lock is locked by 291 * a first holder on this node. 292 * @gl: the glock 293 * @flags: 294 * 295 * Returns: errno 296 */ 297 298 static int rgrp_go_lock(struct gfs2_holder *gh) 299 { 300 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object); 301 } 302 303 /** 304 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by 305 * a last holder on this node. 306 * @gl: the glock 307 * @flags: 308 * 309 */ 310 311 static void rgrp_go_unlock(struct gfs2_holder *gh) 312 { 313 gfs2_rgrp_bh_put(gh->gh_gl->gl_object); 314 } 315 316 /** 317 * trans_go_sync - promote/demote the transaction glock 318 * @gl: the glock 319 * @state: the requested state 320 * @flags: 321 * 322 */ 323 324 static void trans_go_sync(struct gfs2_glock *gl) 325 { 326 struct gfs2_sbd *sdp = gl->gl_sbd; 327 328 if (gl->gl_state != LM_ST_UNLOCKED && 329 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 330 gfs2_meta_syncfs(sdp); 331 gfs2_log_shutdown(sdp); 332 } 333 } 334 335 /** 336 * trans_go_xmote_bh - After promoting/demoting the transaction glock 337 * @gl: the glock 338 * 339 */ 340 341 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh) 342 { 343 struct gfs2_sbd *sdp = gl->gl_sbd; 344 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); 345 struct gfs2_glock *j_gl = ip->i_gl; 346 struct gfs2_log_header_host head; 347 int error; 348 349 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 350 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); 351 352 error = gfs2_find_jhead(sdp->sd_jdesc, &head); 353 if (error) 354 gfs2_consist(sdp); 355 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) 356 gfs2_consist(sdp); 357 358 /* Initialize some head of the log stuff */ 359 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) { 360 sdp->sd_log_sequence = head.lh_sequence + 1; 361 gfs2_log_pointers_init(sdp, head.lh_blkno); 362 } 363 } 364 return 0; 365 } 366 367 /** 368 * trans_go_demote_ok 369 * @gl: the glock 370 * 371 * Always returns 0 372 */ 373 374 static int trans_go_demote_ok(const struct gfs2_glock *gl) 375 { 376 return 0; 377 } 378 379 /** 380 * iopen_go_callback - schedule the dcache entry for the inode to be deleted 381 * @gl: the glock 382 * 383 * gl_spin lock is held while calling this 384 */ 385 static void iopen_go_callback(struct gfs2_glock *gl) 386 { 387 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object; 388 389 if (gl->gl_demote_state == LM_ST_UNLOCKED && 390 gl->gl_state == LM_ST_SHARED && ip) { 391 gfs2_glock_hold(gl); 392 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) 393 gfs2_glock_put_nolock(gl); 394 } 395 } 396 397 const struct gfs2_glock_operations gfs2_meta_glops = { 398 .go_type = LM_TYPE_META, 399 }; 400 401 const struct gfs2_glock_operations gfs2_inode_glops = { 402 .go_xmote_th = inode_go_sync, 403 .go_inval = inode_go_inval, 404 .go_demote_ok = inode_go_demote_ok, 405 .go_lock = inode_go_lock, 406 .go_dump = inode_go_dump, 407 .go_type = LM_TYPE_INODE, 408 .go_min_hold_time = HZ / 5, 409 .go_flags = GLOF_ASPACE, 410 }; 411 412 const struct gfs2_glock_operations gfs2_rgrp_glops = { 413 .go_xmote_th = rgrp_go_sync, 414 .go_inval = rgrp_go_inval, 415 .go_lock = rgrp_go_lock, 416 .go_unlock = rgrp_go_unlock, 417 .go_dump = gfs2_rgrp_dump, 418 .go_type = LM_TYPE_RGRP, 419 .go_min_hold_time = HZ / 5, 420 .go_flags = GLOF_ASPACE, 421 }; 422 423 const struct gfs2_glock_operations gfs2_trans_glops = { 424 .go_xmote_th = trans_go_sync, 425 .go_xmote_bh = trans_go_xmote_bh, 426 .go_demote_ok = trans_go_demote_ok, 427 .go_type = LM_TYPE_NONDISK, 428 }; 429 430 const struct gfs2_glock_operations gfs2_iopen_glops = { 431 .go_type = LM_TYPE_IOPEN, 432 .go_callback = iopen_go_callback, 433 }; 434 435 const struct gfs2_glock_operations gfs2_flock_glops = { 436 .go_type = LM_TYPE_FLOCK, 437 }; 438 439 const struct gfs2_glock_operations gfs2_nondisk_glops = { 440 .go_type = LM_TYPE_NONDISK, 441 }; 442 443 const struct gfs2_glock_operations gfs2_quota_glops = { 444 .go_type = LM_TYPE_QUOTA, 445 }; 446 447 const struct gfs2_glock_operations gfs2_journal_glops = { 448 .go_type = LM_TYPE_JOURNAL, 449 }; 450 451 const struct gfs2_glock_operations *gfs2_glops_list[] = { 452 [LM_TYPE_META] = &gfs2_meta_glops, 453 [LM_TYPE_INODE] = &gfs2_inode_glops, 454 [LM_TYPE_RGRP] = &gfs2_rgrp_glops, 455 [LM_TYPE_IOPEN] = &gfs2_iopen_glops, 456 [LM_TYPE_FLOCK] = &gfs2_flock_glops, 457 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops, 458 [LM_TYPE_QUOTA] = &gfs2_quota_glops, 459 [LM_TYPE_JOURNAL] = &gfs2_journal_glops, 460 }; 461 462