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 #include <linux/posix_acl.h> 17 18 #include "gfs2.h" 19 #include "incore.h" 20 #include "bmap.h" 21 #include "glock.h" 22 #include "glops.h" 23 #include "inode.h" 24 #include "log.h" 25 #include "meta_io.h" 26 #include "recovery.h" 27 #include "rgrp.h" 28 #include "util.h" 29 #include "trans.h" 30 31 /** 32 * ail_empty_gl - remove all buffers for a given lock from the AIL 33 * @gl: the glock 34 * 35 * None of the buffers should be dirty, locked, or pinned. 36 */ 37 38 static void gfs2_ail_empty_gl(struct gfs2_glock *gl) 39 { 40 struct gfs2_sbd *sdp = gl->gl_sbd; 41 struct list_head *head = &gl->gl_ail_list; 42 struct gfs2_bufdata *bd; 43 struct buffer_head *bh; 44 struct gfs2_trans tr; 45 46 memset(&tr, 0, sizeof(tr)); 47 tr.tr_revokes = atomic_read(&gl->gl_ail_count); 48 49 if (!tr.tr_revokes) 50 return; 51 52 /* A shortened, inline version of gfs2_trans_begin() */ 53 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64)); 54 tr.tr_ip = (unsigned long)__builtin_return_address(0); 55 INIT_LIST_HEAD(&tr.tr_list_buf); 56 gfs2_log_reserve(sdp, tr.tr_reserved); 57 BUG_ON(current->journal_info); 58 current->journal_info = &tr; 59 60 gfs2_log_lock(sdp); 61 while (!list_empty(head)) { 62 bd = list_entry(head->next, struct gfs2_bufdata, 63 bd_ail_gl_list); 64 bh = bd->bd_bh; 65 gfs2_remove_from_ail(bd); 66 bd->bd_bh = NULL; 67 bh->b_private = NULL; 68 bd->bd_blkno = bh->b_blocknr; 69 gfs2_assert_withdraw(sdp, !buffer_busy(bh)); 70 gfs2_trans_add_revoke(sdp, bd); 71 } 72 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count)); 73 gfs2_log_unlock(sdp); 74 75 gfs2_trans_end(sdp); 76 gfs2_log_flush(sdp, NULL); 77 } 78 79 /** 80 * rgrp_go_sync - sync out the metadata for this glock 81 * @gl: the glock 82 * 83 * Called when demoting or unlocking an EX glock. We must flush 84 * to disk all dirty buffers/pages relating to this glock, and must not 85 * not return to caller to demote/unlock the glock until I/O is complete. 86 */ 87 88 static void rgrp_go_sync(struct gfs2_glock *gl) 89 { 90 struct address_space *metamapping = gl->gl_aspace->i_mapping; 91 int error; 92 93 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) 94 return; 95 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE); 96 97 gfs2_log_flush(gl->gl_sbd, gl); 98 filemap_fdatawrite(metamapping); 99 error = filemap_fdatawait(metamapping); 100 mapping_set_error(metamapping, error); 101 gfs2_ail_empty_gl(gl); 102 } 103 104 /** 105 * rgrp_go_inval - invalidate the metadata for this glock 106 * @gl: the glock 107 * @flags: 108 * 109 * We never used LM_ST_DEFERRED with resource groups, so that we 110 * should always see the metadata flag set here. 111 * 112 */ 113 114 static void rgrp_go_inval(struct gfs2_glock *gl, int flags) 115 { 116 struct address_space *mapping = gl->gl_aspace->i_mapping; 117 118 BUG_ON(!(flags & DIO_METADATA)); 119 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count)); 120 truncate_inode_pages(mapping, 0); 121 122 if (gl->gl_object) { 123 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object; 124 rgd->rd_flags &= ~GFS2_RDF_UPTODATE; 125 } 126 } 127 128 /** 129 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock 130 * @gl: the glock protecting the inode 131 * 132 */ 133 134 static void inode_go_sync(struct gfs2_glock *gl) 135 { 136 struct gfs2_inode *ip = gl->gl_object; 137 struct address_space *metamapping = gl->gl_aspace->i_mapping; 138 int error; 139 140 if (ip && !S_ISREG(ip->i_inode.i_mode)) 141 ip = NULL; 142 if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags)) 143 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0); 144 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) 145 return; 146 147 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE); 148 149 gfs2_log_flush(gl->gl_sbd, gl); 150 filemap_fdatawrite(metamapping); 151 if (ip) { 152 struct address_space *mapping = ip->i_inode.i_mapping; 153 filemap_fdatawrite(mapping); 154 error = filemap_fdatawait(mapping); 155 mapping_set_error(mapping, error); 156 } 157 error = filemap_fdatawait(metamapping); 158 mapping_set_error(metamapping, error); 159 gfs2_ail_empty_gl(gl); 160 /* 161 * Writeback of the data mapping may cause the dirty flag to be set 162 * so we have to clear it again here. 163 */ 164 smp_mb__before_clear_bit(); 165 clear_bit(GLF_DIRTY, &gl->gl_flags); 166 } 167 168 /** 169 * inode_go_inval - prepare a inode glock to be released 170 * @gl: the glock 171 * @flags: 172 * 173 * Normally we invlidate everything, but if we are moving into 174 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we 175 * can keep hold of the metadata, since it won't have changed. 176 * 177 */ 178 179 static void inode_go_inval(struct gfs2_glock *gl, int flags) 180 { 181 struct gfs2_inode *ip = gl->gl_object; 182 183 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count)); 184 185 if (flags & DIO_METADATA) { 186 struct address_space *mapping = gl->gl_aspace->i_mapping; 187 truncate_inode_pages(mapping, 0); 188 if (ip) { 189 set_bit(GIF_INVALID, &ip->i_flags); 190 forget_all_cached_acls(&ip->i_inode); 191 } 192 } 193 194 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) 195 gl->gl_sbd->sd_rindex_uptodate = 0; 196 if (ip && S_ISREG(ip->i_inode.i_mode)) 197 truncate_inode_pages(ip->i_inode.i_mapping, 0); 198 } 199 200 /** 201 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock 202 * @gl: the glock 203 * 204 * Returns: 1 if it's ok 205 */ 206 207 static int inode_go_demote_ok(const struct gfs2_glock *gl) 208 { 209 struct gfs2_sbd *sdp = gl->gl_sbd; 210 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object) 211 return 0; 212 return 1; 213 } 214 215 /** 216 * inode_go_lock - operation done after an inode lock is locked by a process 217 * @gl: the glock 218 * @flags: 219 * 220 * Returns: errno 221 */ 222 223 static int inode_go_lock(struct gfs2_holder *gh) 224 { 225 struct gfs2_glock *gl = gh->gh_gl; 226 struct gfs2_sbd *sdp = gl->gl_sbd; 227 struct gfs2_inode *ip = gl->gl_object; 228 int error = 0; 229 230 if (!ip || (gh->gh_flags & GL_SKIP)) 231 return 0; 232 233 if (test_bit(GIF_INVALID, &ip->i_flags)) { 234 error = gfs2_inode_refresh(ip); 235 if (error) 236 return error; 237 } 238 239 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) && 240 (gl->gl_state == LM_ST_EXCLUSIVE) && 241 (gh->gh_state == LM_ST_EXCLUSIVE)) { 242 spin_lock(&sdp->sd_trunc_lock); 243 if (list_empty(&ip->i_trunc_list)) 244 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list); 245 spin_unlock(&sdp->sd_trunc_lock); 246 wake_up(&sdp->sd_quota_wait); 247 return 1; 248 } 249 250 return error; 251 } 252 253 /** 254 * inode_go_dump - print information about an inode 255 * @seq: The iterator 256 * @ip: the inode 257 * 258 * Returns: 0 on success, -ENOBUFS when we run out of space 259 */ 260 261 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl) 262 { 263 const struct gfs2_inode *ip = gl->gl_object; 264 if (ip == NULL) 265 return 0; 266 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu/%llu\n", 267 (unsigned long long)ip->i_no_formal_ino, 268 (unsigned long long)ip->i_no_addr, 269 IF2DT(ip->i_inode.i_mode), ip->i_flags, 270 (unsigned int)ip->i_diskflags, 271 (unsigned long long)ip->i_inode.i_size, 272 (unsigned long long)ip->i_disksize); 273 return 0; 274 } 275 276 /** 277 * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock 278 * @gl: the glock 279 * 280 * Returns: 1 if it's ok 281 */ 282 283 static int rgrp_go_demote_ok(const struct gfs2_glock *gl) 284 { 285 return !gl->gl_aspace->i_mapping->nrpages; 286 } 287 288 /** 289 * rgrp_go_lock - operation done after an rgrp lock is locked by 290 * a first holder on this node. 291 * @gl: the glock 292 * @flags: 293 * 294 * Returns: errno 295 */ 296 297 static int rgrp_go_lock(struct gfs2_holder *gh) 298 { 299 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object); 300 } 301 302 /** 303 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by 304 * a last holder on this node. 305 * @gl: the glock 306 * @flags: 307 * 308 */ 309 310 static void rgrp_go_unlock(struct gfs2_holder *gh) 311 { 312 gfs2_rgrp_bh_put(gh->gh_gl->gl_object); 313 } 314 315 /** 316 * trans_go_sync - promote/demote the transaction glock 317 * @gl: the glock 318 * @state: the requested state 319 * @flags: 320 * 321 */ 322 323 static void trans_go_sync(struct gfs2_glock *gl) 324 { 325 struct gfs2_sbd *sdp = gl->gl_sbd; 326 327 if (gl->gl_state != LM_ST_UNLOCKED && 328 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 329 flush_workqueue(gfs2_delete_workqueue); 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 && 391 ip && test_bit(GIF_USER, &ip->i_flags)) { 392 gfs2_glock_hold(gl); 393 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) 394 gfs2_glock_put_nolock(gl); 395 } 396 } 397 398 const struct gfs2_glock_operations gfs2_meta_glops = { 399 .go_type = LM_TYPE_META, 400 }; 401 402 const struct gfs2_glock_operations gfs2_inode_glops = { 403 .go_xmote_th = inode_go_sync, 404 .go_inval = inode_go_inval, 405 .go_demote_ok = inode_go_demote_ok, 406 .go_lock = inode_go_lock, 407 .go_dump = inode_go_dump, 408 .go_type = LM_TYPE_INODE, 409 .go_min_hold_time = HZ / 5, 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_demote_ok = rgrp_go_demote_ok, 416 .go_lock = rgrp_go_lock, 417 .go_unlock = rgrp_go_unlock, 418 .go_dump = gfs2_rgrp_dump, 419 .go_type = LM_TYPE_RGRP, 420 .go_min_hold_time = HZ / 5, 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_NONDISK] = &gfs2_trans_glops, 456 [LM_TYPE_IOPEN] = &gfs2_iopen_glops, 457 [LM_TYPE_FLOCK] = &gfs2_flock_glops, 458 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops, 459 [LM_TYPE_QUOTA] = &gfs2_quota_glops, 460 [LM_TYPE_JOURNAL] = &gfs2_journal_glops, 461 }; 462 463