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 #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/lm_interface.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 unsigned int blocks; 41 struct list_head *head = &gl->gl_ail_list; 42 struct gfs2_bufdata *bd; 43 struct buffer_head *bh; 44 int error; 45 46 blocks = atomic_read(&gl->gl_ail_count); 47 if (!blocks) 48 return; 49 50 error = gfs2_trans_begin(sdp, 0, blocks); 51 if (gfs2_assert_withdraw(sdp, !error)) 52 return; 53 54 gfs2_log_lock(sdp); 55 while (!list_empty(head)) { 56 bd = list_entry(head->next, struct gfs2_bufdata, 57 bd_ail_gl_list); 58 bh = bd->bd_bh; 59 gfs2_remove_from_ail(NULL, bd); 60 bd->bd_bh = NULL; 61 bh->b_private = NULL; 62 bd->bd_blkno = bh->b_blocknr; 63 gfs2_assert_withdraw(sdp, !buffer_busy(bh)); 64 gfs2_trans_add_revoke(sdp, bd); 65 } 66 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count)); 67 gfs2_log_unlock(sdp); 68 69 gfs2_trans_end(sdp); 70 gfs2_log_flush(sdp, NULL); 71 } 72 73 /** 74 * gfs2_pte_inval - Sync and invalidate all PTEs associated with a glock 75 * @gl: the glock 76 * 77 */ 78 79 static void gfs2_pte_inval(struct gfs2_glock *gl) 80 { 81 struct gfs2_inode *ip; 82 struct inode *inode; 83 84 ip = gl->gl_object; 85 inode = &ip->i_inode; 86 if (!ip || !S_ISREG(inode->i_mode)) 87 return; 88 89 if (!test_bit(GIF_PAGED, &ip->i_flags)) 90 return; 91 92 unmap_shared_mapping_range(inode->i_mapping, 0, 0); 93 94 if (test_bit(GIF_SW_PAGED, &ip->i_flags)) 95 set_bit(GLF_DIRTY, &gl->gl_flags); 96 97 clear_bit(GIF_SW_PAGED, &ip->i_flags); 98 } 99 100 /** 101 * meta_go_sync - sync out the metadata for this glock 102 * @gl: the glock 103 * 104 * Called when demoting or unlocking an EX glock. We must flush 105 * to disk all dirty buffers/pages relating to this glock, and must not 106 * not return to caller to demote/unlock the glock until I/O is complete. 107 */ 108 109 static void meta_go_sync(struct gfs2_glock *gl) 110 { 111 if (gl->gl_state != LM_ST_EXCLUSIVE) 112 return; 113 114 if (test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) { 115 gfs2_log_flush(gl->gl_sbd, gl); 116 gfs2_meta_sync(gl); 117 gfs2_ail_empty_gl(gl); 118 } 119 } 120 121 /** 122 * meta_go_inval - invalidate the metadata for this glock 123 * @gl: the glock 124 * @flags: 125 * 126 */ 127 128 static void meta_go_inval(struct gfs2_glock *gl, int flags) 129 { 130 if (!(flags & DIO_METADATA)) 131 return; 132 133 gfs2_meta_inval(gl); 134 gl->gl_vn++; 135 } 136 137 /** 138 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock 139 * @gl: the glock protecting the inode 140 * 141 */ 142 143 static void inode_go_sync(struct gfs2_glock *gl) 144 { 145 struct gfs2_inode *ip = gl->gl_object; 146 147 if (ip && !S_ISREG(ip->i_inode.i_mode)) 148 ip = NULL; 149 150 if (test_bit(GLF_DIRTY, &gl->gl_flags)) { 151 if (ip && !gfs2_is_jdata(ip)) 152 filemap_fdatawrite(ip->i_inode.i_mapping); 153 gfs2_log_flush(gl->gl_sbd, gl); 154 if (ip && gfs2_is_jdata(ip)) 155 filemap_fdatawrite(ip->i_inode.i_mapping); 156 gfs2_meta_sync(gl); 157 if (ip) { 158 struct address_space *mapping = ip->i_inode.i_mapping; 159 int error = filemap_fdatawait(mapping); 160 mapping_set_error(mapping, error); 161 } 162 clear_bit(GLF_DIRTY, &gl->gl_flags); 163 gfs2_ail_empty_gl(gl); 164 } 165 } 166 167 /** 168 * inode_go_xmote_th - promote/demote a glock 169 * @gl: the glock 170 * @state: the requested state 171 * @flags: 172 * 173 */ 174 175 static void inode_go_xmote_th(struct gfs2_glock *gl) 176 { 177 if (gl->gl_state != LM_ST_UNLOCKED) 178 gfs2_pte_inval(gl); 179 if (gl->gl_state == LM_ST_EXCLUSIVE) 180 inode_go_sync(gl); 181 } 182 183 /** 184 * inode_go_xmote_bh - After promoting/demoting a glock 185 * @gl: the glock 186 * 187 */ 188 189 static void inode_go_xmote_bh(struct gfs2_glock *gl) 190 { 191 struct gfs2_holder *gh = gl->gl_req_gh; 192 struct buffer_head *bh; 193 int error; 194 195 if (gl->gl_state != LM_ST_UNLOCKED && 196 (!gh || !(gh->gh_flags & GL_SKIP))) { 197 error = gfs2_meta_read(gl, gl->gl_name.ln_number, 0, &bh); 198 if (!error) 199 brelse(bh); 200 } 201 } 202 203 /** 204 * inode_go_drop_th - unlock a glock 205 * @gl: the glock 206 * 207 * Invoked from rq_demote(). 208 * Another node needs the lock in EXCLUSIVE mode, or lock (unused for too long) 209 * is being purged from our node's glock cache; we're dropping lock. 210 */ 211 212 static void inode_go_drop_th(struct gfs2_glock *gl) 213 { 214 gfs2_pte_inval(gl); 215 if (gl->gl_state == LM_ST_EXCLUSIVE) 216 inode_go_sync(gl); 217 } 218 219 /** 220 * inode_go_inval - prepare a inode glock to be released 221 * @gl: the glock 222 * @flags: 223 * 224 */ 225 226 static void inode_go_inval(struct gfs2_glock *gl, int flags) 227 { 228 struct gfs2_inode *ip = gl->gl_object; 229 int meta = (flags & DIO_METADATA); 230 231 if (meta) { 232 gfs2_meta_inval(gl); 233 if (ip) 234 set_bit(GIF_INVALID, &ip->i_flags); 235 } 236 237 if (ip && S_ISREG(ip->i_inode.i_mode)) { 238 truncate_inode_pages(ip->i_inode.i_mapping, 0); 239 clear_bit(GIF_PAGED, &ip->i_flags); 240 } 241 } 242 243 /** 244 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock 245 * @gl: the glock 246 * 247 * Returns: 1 if it's ok 248 */ 249 250 static int inode_go_demote_ok(struct gfs2_glock *gl) 251 { 252 struct gfs2_sbd *sdp = gl->gl_sbd; 253 int demote = 0; 254 255 if (!gl->gl_object && !gl->gl_aspace->i_mapping->nrpages) 256 demote = 1; 257 else if (!sdp->sd_args.ar_localcaching && 258 time_after_eq(jiffies, gl->gl_stamp + 259 gfs2_tune_get(sdp, gt_demote_secs) * HZ)) 260 demote = 1; 261 262 return demote; 263 } 264 265 /** 266 * inode_go_lock - operation done after an inode lock is locked by a process 267 * @gl: the glock 268 * @flags: 269 * 270 * Returns: errno 271 */ 272 273 static int inode_go_lock(struct gfs2_holder *gh) 274 { 275 struct gfs2_glock *gl = gh->gh_gl; 276 struct gfs2_inode *ip = gl->gl_object; 277 int error = 0; 278 279 if (!ip) 280 return 0; 281 282 if (test_bit(GIF_INVALID, &ip->i_flags)) { 283 error = gfs2_inode_refresh(ip); 284 if (error) 285 return error; 286 } 287 288 if ((ip->i_di.di_flags & GFS2_DIF_TRUNC_IN_PROG) && 289 (gl->gl_state == LM_ST_EXCLUSIVE) && 290 (gh->gh_state == LM_ST_EXCLUSIVE)) 291 error = gfs2_truncatei_resume(ip); 292 293 return error; 294 } 295 296 /** 297 * inode_go_unlock - operation done before an inode lock is unlocked by a 298 * process 299 * @gl: the glock 300 * @flags: 301 * 302 */ 303 304 static void inode_go_unlock(struct gfs2_holder *gh) 305 { 306 struct gfs2_glock *gl = gh->gh_gl; 307 struct gfs2_inode *ip = gl->gl_object; 308 309 if (ip) 310 gfs2_meta_cache_flush(ip); 311 } 312 313 /** 314 * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock 315 * @gl: the glock 316 * 317 * Returns: 1 if it's ok 318 */ 319 320 static int rgrp_go_demote_ok(struct gfs2_glock *gl) 321 { 322 return !gl->gl_aspace->i_mapping->nrpages; 323 } 324 325 /** 326 * rgrp_go_lock - operation done after an rgrp lock is locked by 327 * a first holder on this node. 328 * @gl: the glock 329 * @flags: 330 * 331 * Returns: errno 332 */ 333 334 static int rgrp_go_lock(struct gfs2_holder *gh) 335 { 336 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object); 337 } 338 339 /** 340 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by 341 * a last holder on this node. 342 * @gl: the glock 343 * @flags: 344 * 345 */ 346 347 static void rgrp_go_unlock(struct gfs2_holder *gh) 348 { 349 gfs2_rgrp_bh_put(gh->gh_gl->gl_object); 350 } 351 352 /** 353 * trans_go_xmote_th - promote/demote the transaction glock 354 * @gl: the glock 355 * @state: the requested state 356 * @flags: 357 * 358 */ 359 360 static void trans_go_xmote_th(struct gfs2_glock *gl) 361 { 362 struct gfs2_sbd *sdp = gl->gl_sbd; 363 364 if (gl->gl_state != LM_ST_UNLOCKED && 365 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 366 gfs2_meta_syncfs(sdp); 367 gfs2_log_shutdown(sdp); 368 } 369 } 370 371 /** 372 * trans_go_xmote_bh - After promoting/demoting the transaction glock 373 * @gl: the glock 374 * 375 */ 376 377 static void trans_go_xmote_bh(struct gfs2_glock *gl) 378 { 379 struct gfs2_sbd *sdp = gl->gl_sbd; 380 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); 381 struct gfs2_glock *j_gl = ip->i_gl; 382 struct gfs2_log_header_host head; 383 int error; 384 385 if (gl->gl_state != LM_ST_UNLOCKED && 386 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 387 gfs2_meta_cache_flush(GFS2_I(sdp->sd_jdesc->jd_inode)); 388 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); 389 390 error = gfs2_find_jhead(sdp->sd_jdesc, &head); 391 if (error) 392 gfs2_consist(sdp); 393 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) 394 gfs2_consist(sdp); 395 396 /* Initialize some head of the log stuff */ 397 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) { 398 sdp->sd_log_sequence = head.lh_sequence + 1; 399 gfs2_log_pointers_init(sdp, head.lh_blkno); 400 } 401 } 402 } 403 404 /** 405 * trans_go_drop_th - unlock the transaction glock 406 * @gl: the glock 407 * 408 * We want to sync the device even with localcaching. Remember 409 * that localcaching journal replay only marks buffers dirty. 410 */ 411 412 static void trans_go_drop_th(struct gfs2_glock *gl) 413 { 414 struct gfs2_sbd *sdp = gl->gl_sbd; 415 416 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { 417 gfs2_meta_syncfs(sdp); 418 gfs2_log_shutdown(sdp); 419 } 420 } 421 422 /** 423 * quota_go_demote_ok - Check to see if it's ok to unlock a quota glock 424 * @gl: the glock 425 * 426 * Returns: 1 if it's ok 427 */ 428 429 static int quota_go_demote_ok(struct gfs2_glock *gl) 430 { 431 return !atomic_read(&gl->gl_lvb_count); 432 } 433 434 const struct gfs2_glock_operations gfs2_meta_glops = { 435 .go_xmote_th = meta_go_sync, 436 .go_drop_th = meta_go_sync, 437 .go_type = LM_TYPE_META, 438 }; 439 440 const struct gfs2_glock_operations gfs2_inode_glops = { 441 .go_xmote_th = inode_go_xmote_th, 442 .go_xmote_bh = inode_go_xmote_bh, 443 .go_drop_th = inode_go_drop_th, 444 .go_inval = inode_go_inval, 445 .go_demote_ok = inode_go_demote_ok, 446 .go_lock = inode_go_lock, 447 .go_unlock = inode_go_unlock, 448 .go_type = LM_TYPE_INODE, 449 .go_min_hold_time = HZ / 10, 450 }; 451 452 const struct gfs2_glock_operations gfs2_rgrp_glops = { 453 .go_xmote_th = meta_go_sync, 454 .go_drop_th = meta_go_sync, 455 .go_inval = meta_go_inval, 456 .go_demote_ok = rgrp_go_demote_ok, 457 .go_lock = rgrp_go_lock, 458 .go_unlock = rgrp_go_unlock, 459 .go_type = LM_TYPE_RGRP, 460 .go_min_hold_time = HZ / 10, 461 }; 462 463 const struct gfs2_glock_operations gfs2_trans_glops = { 464 .go_xmote_th = trans_go_xmote_th, 465 .go_xmote_bh = trans_go_xmote_bh, 466 .go_drop_th = trans_go_drop_th, 467 .go_type = LM_TYPE_NONDISK, 468 }; 469 470 const struct gfs2_glock_operations gfs2_iopen_glops = { 471 .go_type = LM_TYPE_IOPEN, 472 }; 473 474 const struct gfs2_glock_operations gfs2_flock_glops = { 475 .go_type = LM_TYPE_FLOCK, 476 }; 477 478 const struct gfs2_glock_operations gfs2_nondisk_glops = { 479 .go_type = LM_TYPE_NONDISK, 480 }; 481 482 const struct gfs2_glock_operations gfs2_quota_glops = { 483 .go_demote_ok = quota_go_demote_ok, 484 .go_type = LM_TYPE_QUOTA, 485 }; 486 487 const struct gfs2_glock_operations gfs2_journal_glops = { 488 .go_type = LM_TYPE_JOURNAL, 489 }; 490 491