1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. 5 */ 6 7 /* 8 * Quota change tags are associated with each transaction that allocates or 9 * deallocates space. Those changes are accumulated locally to each node (in a 10 * per-node file) and then are periodically synced to the quota file. This 11 * avoids the bottleneck of constantly touching the quota file, but introduces 12 * fuzziness in the current usage value of IDs that are being used on different 13 * nodes in the cluster simultaneously. So, it is possible for a user on 14 * multiple nodes to overrun their quota, but that overrun is controlable. 15 * Since quota tags are part of transactions, there is no need for a quota check 16 * program to be run on node crashes or anything like that. 17 * 18 * There are couple of knobs that let the administrator manage the quota 19 * fuzziness. "quota_quantum" sets the maximum time a quota change can be 20 * sitting on one node before being synced to the quota file. (The default is 21 * 60 seconds.) Another knob, "quota_scale" controls how quickly the frequency 22 * of quota file syncs increases as the user moves closer to their limit. The 23 * more frequent the syncs, the more accurate the quota enforcement, but that 24 * means that there is more contention between the nodes for the quota file. 25 * The default value is one. This sets the maximum theoretical quota overrun 26 * (with infinite node with infinite bandwidth) to twice the user's limit. (In 27 * practice, the maximum overrun you see should be much less.) A "quota_scale" 28 * number greater than one makes quota syncs more frequent and reduces the 29 * maximum overrun. Numbers less than one (but greater than zero) make quota 30 * syncs less frequent. 31 * 32 * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of 33 * the quota file, so it is not being constantly read. 34 */ 35 36 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 37 38 #include <linux/sched.h> 39 #include <linux/slab.h> 40 #include <linux/mm.h> 41 #include <linux/spinlock.h> 42 #include <linux/completion.h> 43 #include <linux/buffer_head.h> 44 #include <linux/sort.h> 45 #include <linux/fs.h> 46 #include <linux/bio.h> 47 #include <linux/gfs2_ondisk.h> 48 #include <linux/kthread.h> 49 #include <linux/freezer.h> 50 #include <linux/quota.h> 51 #include <linux/dqblk_xfs.h> 52 #include <linux/lockref.h> 53 #include <linux/list_lru.h> 54 #include <linux/rcupdate.h> 55 #include <linux/rculist_bl.h> 56 #include <linux/bit_spinlock.h> 57 #include <linux/jhash.h> 58 #include <linux/vmalloc.h> 59 60 #include "gfs2.h" 61 #include "incore.h" 62 #include "bmap.h" 63 #include "glock.h" 64 #include "glops.h" 65 #include "log.h" 66 #include "meta_io.h" 67 #include "quota.h" 68 #include "rgrp.h" 69 #include "super.h" 70 #include "trans.h" 71 #include "inode.h" 72 #include "util.h" 73 74 #define GFS2_QD_HASH_SHIFT 12 75 #define GFS2_QD_HASH_SIZE BIT(GFS2_QD_HASH_SHIFT) 76 #define GFS2_QD_HASH_MASK (GFS2_QD_HASH_SIZE - 1) 77 78 /* Lock order: qd_lock -> bucket lock -> qd->lockref.lock -> lru lock */ 79 /* -> sd_bitmap_lock */ 80 static DEFINE_SPINLOCK(qd_lock); 81 struct list_lru gfs2_qd_lru; 82 83 static struct hlist_bl_head qd_hash_table[GFS2_QD_HASH_SIZE]; 84 85 static unsigned int gfs2_qd_hash(const struct gfs2_sbd *sdp, 86 const struct kqid qid) 87 { 88 unsigned int h; 89 90 h = jhash(&sdp, sizeof(struct gfs2_sbd *), 0); 91 h = jhash(&qid, sizeof(struct kqid), h); 92 93 return h & GFS2_QD_HASH_MASK; 94 } 95 96 static inline void spin_lock_bucket(unsigned int hash) 97 { 98 hlist_bl_lock(&qd_hash_table[hash]); 99 } 100 101 static inline void spin_unlock_bucket(unsigned int hash) 102 { 103 hlist_bl_unlock(&qd_hash_table[hash]); 104 } 105 106 static void gfs2_qd_dealloc(struct rcu_head *rcu) 107 { 108 struct gfs2_quota_data *qd = container_of(rcu, struct gfs2_quota_data, qd_rcu); 109 struct gfs2_sbd *sdp = qd->qd_sbd; 110 111 kmem_cache_free(gfs2_quotad_cachep, qd); 112 if (atomic_dec_and_test(&sdp->sd_quota_count)) 113 wake_up(&sdp->sd_kill_wait); 114 } 115 116 static void gfs2_qd_dispose(struct gfs2_quota_data *qd) 117 { 118 struct gfs2_sbd *sdp = qd->qd_sbd; 119 120 spin_lock(&qd_lock); 121 list_del(&qd->qd_list); 122 spin_unlock(&qd_lock); 123 124 spin_lock_bucket(qd->qd_hash); 125 hlist_bl_del_rcu(&qd->qd_hlist); 126 spin_unlock_bucket(qd->qd_hash); 127 128 if (!gfs2_withdrawing_or_withdrawn(sdp)) { 129 gfs2_assert_warn(sdp, !qd->qd_change); 130 gfs2_assert_warn(sdp, !qd->qd_slot_ref); 131 gfs2_assert_warn(sdp, !qd->qd_bh_count); 132 } 133 134 gfs2_glock_put(qd->qd_gl); 135 call_rcu(&qd->qd_rcu, gfs2_qd_dealloc); 136 } 137 138 static void gfs2_qd_list_dispose(struct list_head *list) 139 { 140 struct gfs2_quota_data *qd; 141 142 while (!list_empty(list)) { 143 qd = list_first_entry(list, struct gfs2_quota_data, qd_lru); 144 list_del(&qd->qd_lru); 145 146 gfs2_qd_dispose(qd); 147 } 148 } 149 150 151 static enum lru_status gfs2_qd_isolate(struct list_head *item, 152 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) 153 { 154 struct list_head *dispose = arg; 155 struct gfs2_quota_data *qd = 156 list_entry(item, struct gfs2_quota_data, qd_lru); 157 enum lru_status status; 158 159 if (!spin_trylock(&qd->qd_lockref.lock)) 160 return LRU_SKIP; 161 162 status = LRU_SKIP; 163 if (qd->qd_lockref.count == 0) { 164 lockref_mark_dead(&qd->qd_lockref); 165 list_lru_isolate_move(lru, &qd->qd_lru, dispose); 166 status = LRU_REMOVED; 167 } 168 169 spin_unlock(&qd->qd_lockref.lock); 170 return status; 171 } 172 173 static unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink, 174 struct shrink_control *sc) 175 { 176 LIST_HEAD(dispose); 177 unsigned long freed; 178 179 if (!(sc->gfp_mask & __GFP_FS)) 180 return SHRINK_STOP; 181 182 freed = list_lru_shrink_walk(&gfs2_qd_lru, sc, 183 gfs2_qd_isolate, &dispose); 184 185 gfs2_qd_list_dispose(&dispose); 186 187 return freed; 188 } 189 190 static unsigned long gfs2_qd_shrink_count(struct shrinker *shrink, 191 struct shrink_control *sc) 192 { 193 return vfs_pressure_ratio(list_lru_shrink_count(&gfs2_qd_lru, sc)); 194 } 195 196 static struct shrinker *gfs2_qd_shrinker; 197 198 int __init gfs2_qd_shrinker_init(void) 199 { 200 gfs2_qd_shrinker = shrinker_alloc(SHRINKER_NUMA_AWARE, "gfs2-qd"); 201 if (!gfs2_qd_shrinker) 202 return -ENOMEM; 203 204 gfs2_qd_shrinker->count_objects = gfs2_qd_shrink_count; 205 gfs2_qd_shrinker->scan_objects = gfs2_qd_shrink_scan; 206 207 shrinker_register(gfs2_qd_shrinker); 208 209 return 0; 210 } 211 212 void gfs2_qd_shrinker_exit(void) 213 { 214 shrinker_free(gfs2_qd_shrinker); 215 } 216 217 static u64 qd2index(struct gfs2_quota_data *qd) 218 { 219 struct kqid qid = qd->qd_id; 220 return (2 * (u64)from_kqid(&init_user_ns, qid)) + 221 ((qid.type == USRQUOTA) ? 0 : 1); 222 } 223 224 static u64 qd2offset(struct gfs2_quota_data *qd) 225 { 226 return qd2index(qd) * sizeof(struct gfs2_quota); 227 } 228 229 static struct gfs2_quota_data *qd_alloc(unsigned hash, struct gfs2_sbd *sdp, struct kqid qid) 230 { 231 struct gfs2_quota_data *qd; 232 int error; 233 234 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS); 235 if (!qd) 236 return NULL; 237 238 qd->qd_sbd = sdp; 239 qd->qd_lockref.count = 0; 240 spin_lock_init(&qd->qd_lockref.lock); 241 qd->qd_id = qid; 242 qd->qd_slot = -1; 243 INIT_LIST_HEAD(&qd->qd_lru); 244 qd->qd_hash = hash; 245 246 error = gfs2_glock_get(sdp, qd2index(qd), 247 &gfs2_quota_glops, CREATE, &qd->qd_gl); 248 if (error) 249 goto fail; 250 251 return qd; 252 253 fail: 254 kmem_cache_free(gfs2_quotad_cachep, qd); 255 return NULL; 256 } 257 258 static struct gfs2_quota_data *gfs2_qd_search_bucket(unsigned int hash, 259 const struct gfs2_sbd *sdp, 260 struct kqid qid) 261 { 262 struct gfs2_quota_data *qd; 263 struct hlist_bl_node *h; 264 265 hlist_bl_for_each_entry_rcu(qd, h, &qd_hash_table[hash], qd_hlist) { 266 if (!qid_eq(qd->qd_id, qid)) 267 continue; 268 if (qd->qd_sbd != sdp) 269 continue; 270 if (lockref_get_not_dead(&qd->qd_lockref)) { 271 list_lru_del_obj(&gfs2_qd_lru, &qd->qd_lru); 272 return qd; 273 } 274 } 275 276 return NULL; 277 } 278 279 280 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid, 281 struct gfs2_quota_data **qdp) 282 { 283 struct gfs2_quota_data *qd, *new_qd; 284 unsigned int hash = gfs2_qd_hash(sdp, qid); 285 286 rcu_read_lock(); 287 *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid); 288 rcu_read_unlock(); 289 290 if (qd) 291 return 0; 292 293 new_qd = qd_alloc(hash, sdp, qid); 294 if (!new_qd) 295 return -ENOMEM; 296 297 spin_lock(&qd_lock); 298 spin_lock_bucket(hash); 299 *qdp = qd = gfs2_qd_search_bucket(hash, sdp, qid); 300 if (qd == NULL) { 301 new_qd->qd_lockref.count++; 302 *qdp = new_qd; 303 list_add(&new_qd->qd_list, &sdp->sd_quota_list); 304 hlist_bl_add_head_rcu(&new_qd->qd_hlist, &qd_hash_table[hash]); 305 atomic_inc(&sdp->sd_quota_count); 306 } 307 spin_unlock_bucket(hash); 308 spin_unlock(&qd_lock); 309 310 if (qd) { 311 gfs2_glock_put(new_qd->qd_gl); 312 kmem_cache_free(gfs2_quotad_cachep, new_qd); 313 } 314 315 return 0; 316 } 317 318 319 static void __qd_hold(struct gfs2_quota_data *qd) 320 { 321 struct gfs2_sbd *sdp = qd->qd_sbd; 322 gfs2_assert(sdp, qd->qd_lockref.count > 0); 323 qd->qd_lockref.count++; 324 } 325 326 static void qd_put(struct gfs2_quota_data *qd) 327 { 328 struct gfs2_sbd *sdp; 329 330 if (lockref_put_or_lock(&qd->qd_lockref)) 331 return; 332 333 BUG_ON(__lockref_is_dead(&qd->qd_lockref)); 334 sdp = qd->qd_sbd; 335 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) { 336 lockref_mark_dead(&qd->qd_lockref); 337 spin_unlock(&qd->qd_lockref.lock); 338 339 gfs2_qd_dispose(qd); 340 return; 341 } 342 343 qd->qd_lockref.count = 0; 344 list_lru_add_obj(&gfs2_qd_lru, &qd->qd_lru); 345 spin_unlock(&qd->qd_lockref.lock); 346 } 347 348 static int slot_get(struct gfs2_quota_data *qd) 349 { 350 struct gfs2_sbd *sdp = qd->qd_sbd; 351 unsigned int bit; 352 int error = 0; 353 354 spin_lock(&sdp->sd_bitmap_lock); 355 if (qd->qd_slot_ref == 0) { 356 bit = find_first_zero_bit(sdp->sd_quota_bitmap, 357 sdp->sd_quota_slots); 358 if (bit >= sdp->sd_quota_slots) { 359 error = -ENOSPC; 360 goto out; 361 } 362 set_bit(bit, sdp->sd_quota_bitmap); 363 qd->qd_slot = bit; 364 } 365 qd->qd_slot_ref++; 366 out: 367 spin_unlock(&sdp->sd_bitmap_lock); 368 return error; 369 } 370 371 static void slot_hold(struct gfs2_quota_data *qd) 372 { 373 struct gfs2_sbd *sdp = qd->qd_sbd; 374 375 spin_lock(&sdp->sd_bitmap_lock); 376 gfs2_assert(sdp, qd->qd_slot_ref); 377 qd->qd_slot_ref++; 378 spin_unlock(&sdp->sd_bitmap_lock); 379 } 380 381 static void slot_put(struct gfs2_quota_data *qd) 382 { 383 struct gfs2_sbd *sdp = qd->qd_sbd; 384 385 spin_lock(&sdp->sd_bitmap_lock); 386 gfs2_assert(sdp, qd->qd_slot_ref); 387 if (!--qd->qd_slot_ref) { 388 BUG_ON(!test_and_clear_bit(qd->qd_slot, sdp->sd_quota_bitmap)); 389 qd->qd_slot = -1; 390 } 391 spin_unlock(&sdp->sd_bitmap_lock); 392 } 393 394 static int bh_get(struct gfs2_quota_data *qd) 395 { 396 struct gfs2_sbd *sdp = qd->qd_sbd; 397 struct inode *inode = sdp->sd_qc_inode; 398 struct gfs2_inode *ip = GFS2_I(inode); 399 unsigned int block, offset; 400 struct buffer_head *bh = NULL; 401 struct iomap iomap = { }; 402 int error; 403 404 spin_lock(&qd->qd_lockref.lock); 405 if (qd->qd_bh_count) { 406 qd->qd_bh_count++; 407 spin_unlock(&qd->qd_lockref.lock); 408 return 0; 409 } 410 spin_unlock(&qd->qd_lockref.lock); 411 412 block = qd->qd_slot / sdp->sd_qc_per_block; 413 offset = qd->qd_slot % sdp->sd_qc_per_block; 414 415 error = gfs2_iomap_get(inode, 416 (loff_t)block << inode->i_blkbits, 417 i_blocksize(inode), &iomap); 418 if (error) 419 return error; 420 error = -ENOENT; 421 if (iomap.type != IOMAP_MAPPED) 422 return error; 423 424 error = gfs2_meta_read(ip->i_gl, iomap.addr >> inode->i_blkbits, 425 DIO_WAIT, 0, &bh); 426 if (error) 427 return error; 428 error = -EIO; 429 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) 430 goto out; 431 432 spin_lock(&qd->qd_lockref.lock); 433 if (qd->qd_bh == NULL) { 434 qd->qd_bh = bh; 435 qd->qd_bh_qc = (struct gfs2_quota_change *) 436 (bh->b_data + sizeof(struct gfs2_meta_header) + 437 offset * sizeof(struct gfs2_quota_change)); 438 bh = NULL; 439 } 440 qd->qd_bh_count++; 441 spin_unlock(&qd->qd_lockref.lock); 442 error = 0; 443 444 out: 445 brelse(bh); 446 return error; 447 } 448 449 static void bh_put(struct gfs2_quota_data *qd) 450 { 451 struct gfs2_sbd *sdp = qd->qd_sbd; 452 struct buffer_head *bh = NULL; 453 454 spin_lock(&qd->qd_lockref.lock); 455 gfs2_assert(sdp, qd->qd_bh_count); 456 if (!--qd->qd_bh_count) { 457 bh = qd->qd_bh; 458 qd->qd_bh = NULL; 459 qd->qd_bh_qc = NULL; 460 } 461 spin_unlock(&qd->qd_lockref.lock); 462 brelse(bh); 463 } 464 465 static bool qd_grab_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd, 466 u64 sync_gen) 467 { 468 bool ret = false; 469 470 spin_lock(&qd->qd_lockref.lock); 471 if (test_bit(QDF_LOCKED, &qd->qd_flags) || 472 !test_bit(QDF_CHANGE, &qd->qd_flags) || 473 qd->qd_sync_gen >= sync_gen) 474 goto out; 475 476 if (__lockref_is_dead(&qd->qd_lockref)) 477 goto out; 478 qd->qd_lockref.count++; 479 480 list_move_tail(&qd->qd_list, &sdp->sd_quota_list); 481 set_bit(QDF_LOCKED, &qd->qd_flags); 482 qd->qd_change_sync = qd->qd_change; 483 slot_hold(qd); 484 ret = true; 485 486 out: 487 spin_unlock(&qd->qd_lockref.lock); 488 return ret; 489 } 490 491 static void qd_ungrab_sync(struct gfs2_quota_data *qd) 492 { 493 clear_bit(QDF_LOCKED, &qd->qd_flags); 494 slot_put(qd); 495 qd_put(qd); 496 } 497 498 static void qdsb_put(struct gfs2_quota_data *qd) 499 { 500 bh_put(qd); 501 slot_put(qd); 502 qd_put(qd); 503 } 504 505 static void qd_unlock(struct gfs2_quota_data *qd) 506 { 507 spin_lock(&qd->qd_lockref.lock); 508 gfs2_assert_warn(qd->qd_sbd, test_bit(QDF_LOCKED, &qd->qd_flags)); 509 clear_bit(QDF_LOCKED, &qd->qd_flags); 510 spin_unlock(&qd->qd_lockref.lock); 511 qdsb_put(qd); 512 } 513 514 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid, 515 struct gfs2_quota_data **qdp) 516 { 517 int error; 518 519 error = qd_get(sdp, qid, qdp); 520 if (error) 521 return error; 522 523 error = slot_get(*qdp); 524 if (error) 525 goto fail; 526 527 error = bh_get(*qdp); 528 if (error) 529 goto fail_slot; 530 531 return 0; 532 533 fail_slot: 534 slot_put(*qdp); 535 fail: 536 qd_put(*qdp); 537 return error; 538 } 539 540 /** 541 * gfs2_qa_get - make sure we have a quota allocations data structure, 542 * if necessary 543 * @ip: the inode for this reservation 544 */ 545 int gfs2_qa_get(struct gfs2_inode *ip) 546 { 547 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 548 struct inode *inode = &ip->i_inode; 549 550 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF) 551 return 0; 552 553 spin_lock(&inode->i_lock); 554 if (ip->i_qadata == NULL) { 555 struct gfs2_qadata *tmp; 556 557 spin_unlock(&inode->i_lock); 558 tmp = kmem_cache_zalloc(gfs2_qadata_cachep, GFP_NOFS); 559 if (!tmp) 560 return -ENOMEM; 561 562 spin_lock(&inode->i_lock); 563 if (ip->i_qadata == NULL) 564 ip->i_qadata = tmp; 565 else 566 kmem_cache_free(gfs2_qadata_cachep, tmp); 567 } 568 ip->i_qadata->qa_ref++; 569 spin_unlock(&inode->i_lock); 570 return 0; 571 } 572 573 void gfs2_qa_put(struct gfs2_inode *ip) 574 { 575 struct inode *inode = &ip->i_inode; 576 577 spin_lock(&inode->i_lock); 578 if (ip->i_qadata && --ip->i_qadata->qa_ref == 0) { 579 kmem_cache_free(gfs2_qadata_cachep, ip->i_qadata); 580 ip->i_qadata = NULL; 581 } 582 spin_unlock(&inode->i_lock); 583 } 584 585 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid) 586 { 587 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 588 struct gfs2_quota_data **qd; 589 int error; 590 591 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF) 592 return 0; 593 594 error = gfs2_qa_get(ip); 595 if (error) 596 return error; 597 598 qd = ip->i_qadata->qa_qd; 599 600 if (gfs2_assert_warn(sdp, !ip->i_qadata->qa_qd_num) || 601 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags))) { 602 error = -EIO; 603 gfs2_qa_put(ip); 604 goto out; 605 } 606 607 error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd); 608 if (error) 609 goto out_unhold; 610 ip->i_qadata->qa_qd_num++; 611 qd++; 612 613 error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd); 614 if (error) 615 goto out_unhold; 616 ip->i_qadata->qa_qd_num++; 617 qd++; 618 619 if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) && 620 !uid_eq(uid, ip->i_inode.i_uid)) { 621 error = qdsb_get(sdp, make_kqid_uid(uid), qd); 622 if (error) 623 goto out_unhold; 624 ip->i_qadata->qa_qd_num++; 625 qd++; 626 } 627 628 if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) && 629 !gid_eq(gid, ip->i_inode.i_gid)) { 630 error = qdsb_get(sdp, make_kqid_gid(gid), qd); 631 if (error) 632 goto out_unhold; 633 ip->i_qadata->qa_qd_num++; 634 qd++; 635 } 636 637 out_unhold: 638 if (error) 639 gfs2_quota_unhold(ip); 640 out: 641 return error; 642 } 643 644 void gfs2_quota_unhold(struct gfs2_inode *ip) 645 { 646 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 647 u32 x; 648 649 if (ip->i_qadata == NULL) 650 return; 651 652 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)); 653 654 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) { 655 qdsb_put(ip->i_qadata->qa_qd[x]); 656 ip->i_qadata->qa_qd[x] = NULL; 657 } 658 ip->i_qadata->qa_qd_num = 0; 659 gfs2_qa_put(ip); 660 } 661 662 static int sort_qd(const void *a, const void *b) 663 { 664 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a; 665 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b; 666 667 if (qid_lt(qd_a->qd_id, qd_b->qd_id)) 668 return -1; 669 if (qid_lt(qd_b->qd_id, qd_a->qd_id)) 670 return 1; 671 return 0; 672 } 673 674 static void do_qc(struct gfs2_quota_data *qd, s64 change) 675 { 676 struct gfs2_sbd *sdp = qd->qd_sbd; 677 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode); 678 struct gfs2_quota_change *qc = qd->qd_bh_qc; 679 bool needs_put = false; 680 s64 x; 681 682 gfs2_trans_add_meta(ip->i_gl, qd->qd_bh); 683 684 /* 685 * The QDF_CHANGE flag indicates that the slot in the quota change file 686 * is used. Here, we use the value of qc->qc_change when the slot is 687 * used, and we assume a value of 0 otherwise. 688 */ 689 690 spin_lock(&qd->qd_lockref.lock); 691 692 x = 0; 693 if (test_bit(QDF_CHANGE, &qd->qd_flags)) 694 x = be64_to_cpu(qc->qc_change); 695 x += change; 696 qd->qd_change += change; 697 698 if (!x && test_bit(QDF_CHANGE, &qd->qd_flags)) { 699 /* The slot in the quota change file becomes unused. */ 700 clear_bit(QDF_CHANGE, &qd->qd_flags); 701 qc->qc_flags = 0; 702 qc->qc_id = 0; 703 needs_put = true; 704 } else if (x && !test_bit(QDF_CHANGE, &qd->qd_flags)) { 705 /* The slot in the quota change file becomes used. */ 706 set_bit(QDF_CHANGE, &qd->qd_flags); 707 __qd_hold(qd); 708 slot_hold(qd); 709 710 qc->qc_flags = 0; 711 if (qd->qd_id.type == USRQUOTA) 712 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER); 713 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id)); 714 } 715 qc->qc_change = cpu_to_be64(x); 716 717 spin_unlock(&qd->qd_lockref.lock); 718 719 if (needs_put) { 720 slot_put(qd); 721 qd_put(qd); 722 } 723 if (change < 0) /* Reset quiet flag if we freed some blocks */ 724 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags); 725 } 726 727 static int gfs2_write_buf_to_page(struct gfs2_sbd *sdp, unsigned long index, 728 unsigned off, void *buf, unsigned bytes) 729 { 730 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode); 731 struct inode *inode = &ip->i_inode; 732 struct address_space *mapping = inode->i_mapping; 733 struct folio *folio; 734 struct buffer_head *bh; 735 u64 blk; 736 unsigned bsize = sdp->sd_sb.sb_bsize, bnum = 0, boff = 0; 737 unsigned to_write = bytes, pg_off = off; 738 739 blk = index << (PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift); 740 boff = off % bsize; 741 742 folio = filemap_grab_folio(mapping, index); 743 if (IS_ERR(folio)) 744 return PTR_ERR(folio); 745 bh = folio_buffers(folio); 746 if (!bh) 747 bh = create_empty_buffers(folio, bsize, 0); 748 749 for (;;) { 750 /* Find the beginning block within the folio */ 751 if (pg_off >= ((bnum * bsize) + bsize)) { 752 bh = bh->b_this_page; 753 bnum++; 754 blk++; 755 continue; 756 } 757 if (!buffer_mapped(bh)) { 758 gfs2_block_map(inode, blk, bh, 1); 759 if (!buffer_mapped(bh)) 760 goto unlock_out; 761 /* If it's a newly allocated disk block, zero it */ 762 if (buffer_new(bh)) 763 folio_zero_range(folio, bnum * bsize, 764 bh->b_size); 765 } 766 if (folio_test_uptodate(folio)) 767 set_buffer_uptodate(bh); 768 if (bh_read(bh, REQ_META | REQ_PRIO) < 0) 769 goto unlock_out; 770 gfs2_trans_add_data(ip->i_gl, bh); 771 772 /* If we need to write to the next block as well */ 773 if (to_write > (bsize - boff)) { 774 pg_off += (bsize - boff); 775 to_write -= (bsize - boff); 776 boff = pg_off % bsize; 777 continue; 778 } 779 break; 780 } 781 782 /* Write to the folio, now that we have setup the buffer(s) */ 783 memcpy_to_folio(folio, off, buf, bytes); 784 flush_dcache_folio(folio); 785 folio_unlock(folio); 786 folio_put(folio); 787 788 return 0; 789 790 unlock_out: 791 folio_unlock(folio); 792 folio_put(folio); 793 return -EIO; 794 } 795 796 static int gfs2_write_disk_quota(struct gfs2_sbd *sdp, struct gfs2_quota *qp, 797 loff_t loc) 798 { 799 unsigned long pg_beg; 800 unsigned pg_off, nbytes, overflow = 0; 801 int error; 802 void *ptr; 803 804 nbytes = sizeof(struct gfs2_quota); 805 806 pg_beg = loc >> PAGE_SHIFT; 807 pg_off = offset_in_page(loc); 808 809 /* If the quota straddles a page boundary, split the write in two */ 810 if ((pg_off + nbytes) > PAGE_SIZE) 811 overflow = (pg_off + nbytes) - PAGE_SIZE; 812 813 ptr = qp; 814 error = gfs2_write_buf_to_page(sdp, pg_beg, pg_off, ptr, 815 nbytes - overflow); 816 /* If there's an overflow, write the remaining bytes to the next page */ 817 if (!error && overflow) 818 error = gfs2_write_buf_to_page(sdp, pg_beg + 1, 0, 819 ptr + nbytes - overflow, 820 overflow); 821 return error; 822 } 823 824 /** 825 * gfs2_adjust_quota - adjust record of current block usage 826 * @sdp: The superblock 827 * @loc: Offset of the entry in the quota file 828 * @change: The amount of usage change to record 829 * @qd: The quota data 830 * @fdq: The updated limits to record 831 * 832 * This function was mostly borrowed from gfs2_block_truncate_page which was 833 * in turn mostly borrowed from ext3 834 * 835 * Returns: 0 or -ve on error 836 */ 837 838 static int gfs2_adjust_quota(struct gfs2_sbd *sdp, loff_t loc, 839 s64 change, struct gfs2_quota_data *qd, 840 struct qc_dqblk *fdq) 841 { 842 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode); 843 struct inode *inode = &ip->i_inode; 844 struct gfs2_quota q; 845 int err; 846 u64 size; 847 848 if (gfs2_is_stuffed(ip)) { 849 err = gfs2_unstuff_dinode(ip); 850 if (err) 851 return err; 852 } 853 854 memset(&q, 0, sizeof(struct gfs2_quota)); 855 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q)); 856 if (err < 0) 857 return err; 858 859 loc -= sizeof(q); /* gfs2_internal_read would've advanced the loc ptr */ 860 be64_add_cpu(&q.qu_value, change); 861 if (((s64)be64_to_cpu(q.qu_value)) < 0) 862 q.qu_value = 0; /* Never go negative on quota usage */ 863 spin_lock(&qd->qd_lockref.lock); 864 qd->qd_qb.qb_value = q.qu_value; 865 if (fdq) { 866 if (fdq->d_fieldmask & QC_SPC_SOFT) { 867 q.qu_warn = cpu_to_be64(fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift); 868 qd->qd_qb.qb_warn = q.qu_warn; 869 } 870 if (fdq->d_fieldmask & QC_SPC_HARD) { 871 q.qu_limit = cpu_to_be64(fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift); 872 qd->qd_qb.qb_limit = q.qu_limit; 873 } 874 if (fdq->d_fieldmask & QC_SPACE) { 875 q.qu_value = cpu_to_be64(fdq->d_space >> sdp->sd_sb.sb_bsize_shift); 876 qd->qd_qb.qb_value = q.qu_value; 877 } 878 } 879 spin_unlock(&qd->qd_lockref.lock); 880 881 err = gfs2_write_disk_quota(sdp, &q, loc); 882 if (!err) { 883 size = loc + sizeof(struct gfs2_quota); 884 if (size > inode->i_size) 885 i_size_write(inode, size); 886 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 887 mark_inode_dirty(inode); 888 set_bit(QDF_REFRESH, &qd->qd_flags); 889 } 890 891 return err; 892 } 893 894 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda, 895 u64 sync_gen) 896 { 897 struct gfs2_sbd *sdp = (*qda)->qd_sbd; 898 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode); 899 struct gfs2_alloc_parms ap = {}; 900 unsigned int data_blocks, ind_blocks; 901 struct gfs2_holder *ghs, i_gh; 902 unsigned int qx, x; 903 struct gfs2_quota_data *qd; 904 unsigned reserved; 905 loff_t offset; 906 unsigned int nalloc = 0, blocks; 907 int error; 908 909 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota), 910 &data_blocks, &ind_blocks); 911 912 ghs = kmalloc_array(num_qd, sizeof(struct gfs2_holder), GFP_NOFS); 913 if (!ghs) 914 return -ENOMEM; 915 916 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL); 917 inode_lock(&ip->i_inode); 918 for (qx = 0; qx < num_qd; qx++) { 919 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE, 920 GL_NOCACHE, &ghs[qx]); 921 if (error) 922 goto out_dq; 923 } 924 925 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh); 926 if (error) 927 goto out_dq; 928 929 for (x = 0; x < num_qd; x++) { 930 offset = qd2offset(qda[x]); 931 if (gfs2_write_alloc_required(ip, offset, 932 sizeof(struct gfs2_quota))) 933 nalloc++; 934 } 935 936 /* 937 * 1 blk for unstuffing inode if stuffed. We add this extra 938 * block to the reservation unconditionally. If the inode 939 * doesn't need unstuffing, the block will be released to the 940 * rgrp since it won't be allocated during the transaction 941 */ 942 /* +3 in the end for unstuffing block, inode size update block 943 * and another block in case quota straddles page boundary and 944 * two blocks need to be updated instead of 1 */ 945 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3; 946 947 reserved = 1 + (nalloc * (data_blocks + ind_blocks)); 948 ap.target = reserved; 949 error = gfs2_inplace_reserve(ip, &ap); 950 if (error) 951 goto out_alloc; 952 953 if (nalloc) 954 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS; 955 956 error = gfs2_trans_begin(sdp, blocks, 0); 957 if (error) 958 goto out_ipres; 959 960 for (x = 0; x < num_qd; x++) { 961 qd = qda[x]; 962 offset = qd2offset(qd); 963 error = gfs2_adjust_quota(sdp, offset, qd->qd_change_sync, qd, 964 NULL); 965 if (error) 966 goto out_end_trans; 967 968 do_qc(qd, -qd->qd_change_sync); 969 set_bit(QDF_REFRESH, &qd->qd_flags); 970 } 971 972 out_end_trans: 973 gfs2_trans_end(sdp); 974 out_ipres: 975 gfs2_inplace_release(ip); 976 out_alloc: 977 gfs2_glock_dq_uninit(&i_gh); 978 out_dq: 979 while (qx--) 980 gfs2_glock_dq_uninit(&ghs[qx]); 981 inode_unlock(&ip->i_inode); 982 kfree(ghs); 983 gfs2_log_flush(ip->i_gl->gl_name.ln_sbd, ip->i_gl, 984 GFS2_LOG_HEAD_FLUSH_NORMAL | GFS2_LFC_DO_SYNC); 985 if (!error) { 986 for (x = 0; x < num_qd; x++) { 987 qd = qda[x]; 988 spin_lock(&qd->qd_lockref.lock); 989 if (qd->qd_sync_gen < sync_gen) 990 qd->qd_sync_gen = sync_gen; 991 spin_unlock(&qd->qd_lockref.lock); 992 } 993 } 994 return error; 995 } 996 997 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd) 998 { 999 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode); 1000 struct gfs2_quota q; 1001 struct gfs2_quota_lvb *qlvb; 1002 loff_t pos; 1003 int error; 1004 1005 memset(&q, 0, sizeof(struct gfs2_quota)); 1006 pos = qd2offset(qd); 1007 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q)); 1008 if (error < 0) 1009 return error; 1010 1011 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr; 1012 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC); 1013 qlvb->__pad = 0; 1014 qlvb->qb_limit = q.qu_limit; 1015 qlvb->qb_warn = q.qu_warn; 1016 qlvb->qb_value = q.qu_value; 1017 spin_lock(&qd->qd_lockref.lock); 1018 qd->qd_qb = *qlvb; 1019 spin_unlock(&qd->qd_lockref.lock); 1020 1021 return 0; 1022 } 1023 1024 static int do_glock(struct gfs2_quota_data *qd, int force_refresh, 1025 struct gfs2_holder *q_gh) 1026 { 1027 struct gfs2_sbd *sdp = qd->qd_sbd; 1028 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode); 1029 struct gfs2_holder i_gh; 1030 int error; 1031 1032 gfs2_assert_warn(sdp, sdp == qd->qd_gl->gl_name.ln_sbd); 1033 restart: 1034 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh); 1035 if (error) 1036 return error; 1037 1038 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags)) 1039 force_refresh = FORCE; 1040 1041 spin_lock(&qd->qd_lockref.lock); 1042 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr; 1043 spin_unlock(&qd->qd_lockref.lock); 1044 1045 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) { 1046 gfs2_glock_dq_uninit(q_gh); 1047 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 1048 GL_NOCACHE, q_gh); 1049 if (error) 1050 return error; 1051 1052 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh); 1053 if (error) 1054 goto fail; 1055 1056 error = update_qd(sdp, qd); 1057 if (error) 1058 goto fail_gunlock; 1059 1060 gfs2_glock_dq_uninit(&i_gh); 1061 gfs2_glock_dq_uninit(q_gh); 1062 force_refresh = 0; 1063 goto restart; 1064 } 1065 1066 return 0; 1067 1068 fail_gunlock: 1069 gfs2_glock_dq_uninit(&i_gh); 1070 fail: 1071 gfs2_glock_dq_uninit(q_gh); 1072 return error; 1073 } 1074 1075 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid) 1076 { 1077 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1078 struct gfs2_quota_data *qd; 1079 u32 x; 1080 int error; 1081 1082 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF) 1083 return 0; 1084 1085 error = gfs2_quota_hold(ip, uid, gid); 1086 if (error) 1087 return error; 1088 1089 sort(ip->i_qadata->qa_qd, ip->i_qadata->qa_qd_num, 1090 sizeof(struct gfs2_quota_data *), sort_qd, NULL); 1091 1092 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) { 1093 qd = ip->i_qadata->qa_qd[x]; 1094 error = do_glock(qd, NO_FORCE, &ip->i_qadata->qa_qd_ghs[x]); 1095 if (error) 1096 break; 1097 } 1098 1099 if (!error) 1100 set_bit(GIF_QD_LOCKED, &ip->i_flags); 1101 else { 1102 while (x--) 1103 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]); 1104 gfs2_quota_unhold(ip); 1105 } 1106 1107 return error; 1108 } 1109 1110 static bool need_sync(struct gfs2_quota_data *qd) 1111 { 1112 struct gfs2_sbd *sdp = qd->qd_sbd; 1113 struct gfs2_tune *gt = &sdp->sd_tune; 1114 s64 value, change, limit; 1115 unsigned int num, den; 1116 int ret = false; 1117 1118 spin_lock(&qd->qd_lockref.lock); 1119 if (!qd->qd_qb.qb_limit) 1120 goto out; 1121 1122 change = qd->qd_change; 1123 if (change <= 0) 1124 goto out; 1125 value = (s64)be64_to_cpu(qd->qd_qb.qb_value); 1126 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit); 1127 if (value >= limit) 1128 goto out; 1129 1130 spin_lock(>->gt_spin); 1131 num = gt->gt_quota_scale_num; 1132 den = gt->gt_quota_scale_den; 1133 spin_unlock(>->gt_spin); 1134 1135 change *= gfs2_jindex_size(sdp) * num; 1136 change = div_s64(change, den); 1137 if (value + change < limit) 1138 goto out; 1139 1140 ret = true; 1141 out: 1142 spin_unlock(&qd->qd_lockref.lock); 1143 return ret; 1144 } 1145 1146 void gfs2_quota_unlock(struct gfs2_inode *ip) 1147 { 1148 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1149 struct gfs2_quota_data *qda[2 * GFS2_MAXQUOTAS]; 1150 unsigned int count = 0; 1151 u32 x; 1152 1153 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags)) 1154 return; 1155 1156 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) { 1157 struct gfs2_quota_data *qd; 1158 bool sync; 1159 int error; 1160 1161 qd = ip->i_qadata->qa_qd[x]; 1162 sync = need_sync(qd); 1163 1164 gfs2_glock_dq_uninit(&ip->i_qadata->qa_qd_ghs[x]); 1165 if (!sync) 1166 continue; 1167 1168 spin_lock(&qd_lock); 1169 sync = qd_grab_sync(sdp, qd, U64_MAX); 1170 spin_unlock(&qd_lock); 1171 1172 if (!sync) 1173 continue; 1174 1175 gfs2_assert_warn(sdp, qd->qd_change_sync); 1176 error = bh_get(qd); 1177 if (error) { 1178 qd_ungrab_sync(qd); 1179 continue; 1180 } 1181 1182 qda[count++] = qd; 1183 } 1184 1185 if (count) { 1186 u64 sync_gen = READ_ONCE(sdp->sd_quota_sync_gen); 1187 1188 do_sync(count, qda, sync_gen); 1189 for (x = 0; x < count; x++) 1190 qd_unlock(qda[x]); 1191 } 1192 1193 gfs2_quota_unhold(ip); 1194 } 1195 1196 #define MAX_LINE 256 1197 1198 static void print_message(struct gfs2_quota_data *qd, char *type) 1199 { 1200 struct gfs2_sbd *sdp = qd->qd_sbd; 1201 1202 if (sdp->sd_args.ar_quota != GFS2_QUOTA_QUIET) { 1203 fs_info(sdp, "quota %s for %s %u\n", 1204 type, 1205 (qd->qd_id.type == USRQUOTA) ? "user" : "group", 1206 from_kqid(&init_user_ns, qd->qd_id)); 1207 } 1208 } 1209 1210 /** 1211 * gfs2_quota_check - check if allocating new blocks will exceed quota 1212 * @ip: The inode for which this check is being performed 1213 * @uid: The uid to check against 1214 * @gid: The gid to check against 1215 * @ap: The allocation parameters. ap->target contains the requested 1216 * blocks. ap->min_target, if set, contains the minimum blks 1217 * requested. 1218 * 1219 * Returns: 0 on success. 1220 * min_req = ap->min_target ? ap->min_target : ap->target; 1221 * quota must allow at least min_req blks for success and 1222 * ap->allowed is set to the number of blocks allowed 1223 * 1224 * -EDQUOT otherwise, quota violation. ap->allowed is set to number 1225 * of blocks available. 1226 */ 1227 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid, 1228 struct gfs2_alloc_parms *ap) 1229 { 1230 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1231 struct gfs2_quota_data *qd; 1232 s64 value, warn, limit; 1233 u32 x; 1234 int error = 0; 1235 1236 ap->allowed = UINT_MAX; /* Assume we are permitted a whole lot */ 1237 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags)) 1238 return 0; 1239 1240 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) { 1241 qd = ip->i_qadata->qa_qd[x]; 1242 1243 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) || 1244 qid_eq(qd->qd_id, make_kqid_gid(gid)))) 1245 continue; 1246 1247 spin_lock(&qd->qd_lockref.lock); 1248 warn = (s64)be64_to_cpu(qd->qd_qb.qb_warn); 1249 limit = (s64)be64_to_cpu(qd->qd_qb.qb_limit); 1250 value = (s64)be64_to_cpu(qd->qd_qb.qb_value); 1251 value += qd->qd_change; 1252 spin_unlock(&qd->qd_lockref.lock); 1253 1254 if (limit > 0 && (limit - value) < ap->allowed) 1255 ap->allowed = limit - value; 1256 /* If we can't meet the target */ 1257 if (limit && limit < (value + (s64)ap->target)) { 1258 /* If no min_target specified or we don't meet 1259 * min_target, return -EDQUOT */ 1260 if (!ap->min_target || ap->min_target > ap->allowed) { 1261 if (!test_and_set_bit(QDF_QMSG_QUIET, 1262 &qd->qd_flags)) { 1263 print_message(qd, "exceeded"); 1264 quota_send_warning(qd->qd_id, 1265 sdp->sd_vfs->s_dev, 1266 QUOTA_NL_BHARDWARN); 1267 } 1268 error = -EDQUOT; 1269 break; 1270 } 1271 } else if (warn && warn < value && 1272 time_after_eq(jiffies, qd->qd_last_warn + 1273 gfs2_tune_get(sdp, gt_quota_warn_period) 1274 * HZ)) { 1275 quota_send_warning(qd->qd_id, 1276 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN); 1277 print_message(qd, "warning"); 1278 error = 0; 1279 qd->qd_last_warn = jiffies; 1280 } 1281 } 1282 return error; 1283 } 1284 1285 void gfs2_quota_change(struct gfs2_inode *ip, s64 change, 1286 kuid_t uid, kgid_t gid) 1287 { 1288 struct gfs2_quota_data *qd; 1289 u32 x; 1290 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1291 1292 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF || 1293 gfs2_assert_warn(sdp, change)) 1294 return; 1295 if (ip->i_diskflags & GFS2_DIF_SYSTEM) 1296 return; 1297 1298 if (gfs2_assert_withdraw(sdp, ip->i_qadata && 1299 ip->i_qadata->qa_ref > 0)) 1300 return; 1301 for (x = 0; x < ip->i_qadata->qa_qd_num; x++) { 1302 qd = ip->i_qadata->qa_qd[x]; 1303 1304 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) || 1305 qid_eq(qd->qd_id, make_kqid_gid(gid))) { 1306 do_qc(qd, change); 1307 } 1308 } 1309 } 1310 1311 int gfs2_quota_sync(struct super_block *sb, int type) 1312 { 1313 struct gfs2_sbd *sdp = sb->s_fs_info; 1314 struct gfs2_quota_data **qda; 1315 unsigned int max_qd = PAGE_SIZE / sizeof(struct gfs2_holder); 1316 u64 sync_gen; 1317 int error = 0; 1318 1319 if (sb_rdonly(sdp->sd_vfs)) 1320 return 0; 1321 1322 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL); 1323 if (!qda) 1324 return -ENOMEM; 1325 1326 mutex_lock(&sdp->sd_quota_sync_mutex); 1327 sync_gen = sdp->sd_quota_sync_gen + 1; 1328 1329 do { 1330 struct gfs2_quota_data *iter; 1331 unsigned int num_qd = 0; 1332 unsigned int x; 1333 1334 spin_lock(&qd_lock); 1335 list_for_each_entry(iter, &sdp->sd_quota_list, qd_list) { 1336 if (qd_grab_sync(sdp, iter, sync_gen)) { 1337 qda[num_qd++] = iter; 1338 if (num_qd == max_qd) 1339 break; 1340 } 1341 } 1342 spin_unlock(&qd_lock); 1343 1344 if (!num_qd) 1345 break; 1346 1347 for (x = 0; x < num_qd; x++) { 1348 error = bh_get(qda[x]); 1349 if (!error) 1350 continue; 1351 1352 while (x < num_qd) 1353 qd_ungrab_sync(qda[--num_qd]); 1354 break; 1355 } 1356 1357 if (!error) { 1358 WRITE_ONCE(sdp->sd_quota_sync_gen, sync_gen); 1359 error = do_sync(num_qd, qda, sync_gen); 1360 } 1361 1362 for (x = 0; x < num_qd; x++) 1363 qd_unlock(qda[x]); 1364 } while (!error); 1365 1366 mutex_unlock(&sdp->sd_quota_sync_mutex); 1367 kfree(qda); 1368 1369 return error; 1370 } 1371 1372 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid) 1373 { 1374 struct gfs2_quota_data *qd; 1375 struct gfs2_holder q_gh; 1376 int error; 1377 1378 error = qd_get(sdp, qid, &qd); 1379 if (error) 1380 return error; 1381 1382 error = do_glock(qd, FORCE, &q_gh); 1383 if (!error) 1384 gfs2_glock_dq_uninit(&q_gh); 1385 1386 qd_put(qd); 1387 return error; 1388 } 1389 1390 int gfs2_quota_init(struct gfs2_sbd *sdp) 1391 { 1392 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode); 1393 u64 size = i_size_read(sdp->sd_qc_inode); 1394 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift; 1395 unsigned int x, slot = 0; 1396 unsigned int found = 0; 1397 unsigned int hash; 1398 unsigned int bm_size; 1399 struct buffer_head *bh; 1400 u64 dblock; 1401 u32 extlen = 0; 1402 int error; 1403 1404 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20)) 1405 return -EIO; 1406 1407 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block; 1408 bm_size = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * sizeof(unsigned long)); 1409 bm_size *= sizeof(unsigned long); 1410 error = -ENOMEM; 1411 sdp->sd_quota_bitmap = kzalloc(bm_size, GFP_NOFS | __GFP_NOWARN); 1412 if (sdp->sd_quota_bitmap == NULL) 1413 sdp->sd_quota_bitmap = __vmalloc(bm_size, GFP_NOFS | 1414 __GFP_ZERO); 1415 if (!sdp->sd_quota_bitmap) 1416 return error; 1417 1418 for (x = 0; x < blocks; x++) { 1419 struct gfs2_quota_change *qc; 1420 unsigned int y; 1421 1422 if (!extlen) { 1423 extlen = 32; 1424 error = gfs2_get_extent(&ip->i_inode, x, &dblock, &extlen); 1425 if (error) 1426 goto fail; 1427 } 1428 error = -EIO; 1429 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); 1430 if (!bh) 1431 goto fail; 1432 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) 1433 goto fail_brelse; 1434 1435 qc = (struct gfs2_quota_change *)(bh->b_data + sizeof(struct gfs2_meta_header)); 1436 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots; 1437 y++, slot++) { 1438 struct gfs2_quota_data *old_qd, *qd; 1439 s64 qc_change = be64_to_cpu(qc->qc_change); 1440 u32 qc_flags = be32_to_cpu(qc->qc_flags); 1441 enum quota_type qtype = (qc_flags & GFS2_QCF_USER) ? 1442 USRQUOTA : GRPQUOTA; 1443 struct kqid qc_id = make_kqid(&init_user_ns, qtype, 1444 be32_to_cpu(qc->qc_id)); 1445 qc++; 1446 if (!qc_change) 1447 continue; 1448 1449 hash = gfs2_qd_hash(sdp, qc_id); 1450 qd = qd_alloc(hash, sdp, qc_id); 1451 if (qd == NULL) 1452 goto fail_brelse; 1453 1454 set_bit(QDF_CHANGE, &qd->qd_flags); 1455 qd->qd_change = qc_change; 1456 qd->qd_slot = slot; 1457 qd->qd_slot_ref = 1; 1458 1459 spin_lock(&qd_lock); 1460 spin_lock_bucket(hash); 1461 old_qd = gfs2_qd_search_bucket(hash, sdp, qc_id); 1462 if (old_qd) { 1463 fs_err(sdp, "Corruption found in quota_change%u" 1464 "file: duplicate identifier in " 1465 "slot %u\n", 1466 sdp->sd_jdesc->jd_jid, slot); 1467 1468 spin_unlock_bucket(hash); 1469 spin_unlock(&qd_lock); 1470 qd_put(old_qd); 1471 1472 gfs2_glock_put(qd->qd_gl); 1473 kmem_cache_free(gfs2_quotad_cachep, qd); 1474 1475 /* zero out the duplicate slot */ 1476 lock_buffer(bh); 1477 memset(qc, 0, sizeof(*qc)); 1478 mark_buffer_dirty(bh); 1479 unlock_buffer(bh); 1480 1481 continue; 1482 } 1483 BUG_ON(test_and_set_bit(slot, sdp->sd_quota_bitmap)); 1484 list_add(&qd->qd_list, &sdp->sd_quota_list); 1485 atomic_inc(&sdp->sd_quota_count); 1486 hlist_bl_add_head_rcu(&qd->qd_hlist, &qd_hash_table[hash]); 1487 spin_unlock_bucket(hash); 1488 spin_unlock(&qd_lock); 1489 1490 found++; 1491 } 1492 1493 if (buffer_dirty(bh)) 1494 sync_dirty_buffer(bh); 1495 brelse(bh); 1496 dblock++; 1497 extlen--; 1498 } 1499 1500 if (found) 1501 fs_info(sdp, "found %u quota changes\n", found); 1502 1503 return 0; 1504 1505 fail_brelse: 1506 if (buffer_dirty(bh)) 1507 sync_dirty_buffer(bh); 1508 brelse(bh); 1509 fail: 1510 gfs2_quota_cleanup(sdp); 1511 return error; 1512 } 1513 1514 void gfs2_quota_cleanup(struct gfs2_sbd *sdp) 1515 { 1516 struct gfs2_quota_data *qd; 1517 LIST_HEAD(dispose); 1518 int count; 1519 1520 BUG_ON(!test_bit(SDF_NORECOVERY, &sdp->sd_flags) && 1521 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)); 1522 1523 spin_lock(&qd_lock); 1524 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) { 1525 spin_lock(&qd->qd_lockref.lock); 1526 if (qd->qd_lockref.count != 0) { 1527 spin_unlock(&qd->qd_lockref.lock); 1528 continue; 1529 } 1530 lockref_mark_dead(&qd->qd_lockref); 1531 spin_unlock(&qd->qd_lockref.lock); 1532 1533 list_lru_del_obj(&gfs2_qd_lru, &qd->qd_lru); 1534 list_add(&qd->qd_lru, &dispose); 1535 } 1536 spin_unlock(&qd_lock); 1537 1538 gfs2_qd_list_dispose(&dispose); 1539 1540 wait_event_timeout(sdp->sd_kill_wait, 1541 (count = atomic_read(&sdp->sd_quota_count)) == 0, 1542 HZ * 60); 1543 1544 if (count != 0) 1545 fs_err(sdp, "%d left-over quota data objects\n", count); 1546 1547 kvfree(sdp->sd_quota_bitmap); 1548 sdp->sd_quota_bitmap = NULL; 1549 } 1550 1551 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error) 1552 { 1553 if (error == 0 || error == -EROFS) 1554 return; 1555 if (!gfs2_withdrawing_or_withdrawn(sdp)) { 1556 if (!cmpxchg(&sdp->sd_log_error, 0, error)) 1557 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error); 1558 wake_up(&sdp->sd_logd_waitq); 1559 } 1560 } 1561 1562 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg, 1563 int (*fxn)(struct super_block *sb, int type), 1564 unsigned long t, unsigned long *timeo, 1565 unsigned int *new_timeo) 1566 { 1567 if (t >= *timeo) { 1568 int error = fxn(sdp->sd_vfs, 0); 1569 quotad_error(sdp, msg, error); 1570 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ; 1571 } else { 1572 *timeo -= t; 1573 } 1574 } 1575 1576 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) { 1577 if (!sdp->sd_statfs_force_sync) { 1578 sdp->sd_statfs_force_sync = 1; 1579 wake_up(&sdp->sd_quota_wait); 1580 } 1581 } 1582 1583 1584 /** 1585 * gfs2_quotad - Write cached quota changes into the quota file 1586 * @data: Pointer to GFS2 superblock 1587 * 1588 */ 1589 1590 int gfs2_quotad(void *data) 1591 { 1592 struct gfs2_sbd *sdp = data; 1593 struct gfs2_tune *tune = &sdp->sd_tune; 1594 unsigned long statfs_timeo = 0; 1595 unsigned long quotad_timeo = 0; 1596 unsigned long t = 0; 1597 1598 set_freezable(); 1599 while (!kthread_should_stop()) { 1600 if (gfs2_withdrawing_or_withdrawn(sdp)) 1601 break; 1602 1603 /* Update the master statfs file */ 1604 if (sdp->sd_statfs_force_sync) { 1605 int error = gfs2_statfs_sync(sdp->sd_vfs, 0); 1606 quotad_error(sdp, "statfs", error); 1607 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ; 1608 } 1609 else 1610 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t, 1611 &statfs_timeo, 1612 &tune->gt_statfs_quantum); 1613 1614 /* Update quota file */ 1615 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t, 1616 "ad_timeo, &tune->gt_quota_quantum); 1617 1618 t = min(quotad_timeo, statfs_timeo); 1619 1620 t = wait_event_freezable_timeout(sdp->sd_quota_wait, 1621 sdp->sd_statfs_force_sync || 1622 gfs2_withdrawing_or_withdrawn(sdp) || 1623 kthread_should_stop(), 1624 t); 1625 1626 if (sdp->sd_statfs_force_sync) 1627 t = 0; 1628 } 1629 1630 return 0; 1631 } 1632 1633 static int gfs2_quota_get_state(struct super_block *sb, struct qc_state *state) 1634 { 1635 struct gfs2_sbd *sdp = sb->s_fs_info; 1636 1637 memset(state, 0, sizeof(*state)); 1638 1639 switch (sdp->sd_args.ar_quota) { 1640 case GFS2_QUOTA_QUIET: 1641 fallthrough; 1642 case GFS2_QUOTA_ON: 1643 state->s_state[USRQUOTA].flags |= QCI_LIMITS_ENFORCED; 1644 state->s_state[GRPQUOTA].flags |= QCI_LIMITS_ENFORCED; 1645 fallthrough; 1646 case GFS2_QUOTA_ACCOUNT: 1647 state->s_state[USRQUOTA].flags |= QCI_ACCT_ENABLED | 1648 QCI_SYSFILE; 1649 state->s_state[GRPQUOTA].flags |= QCI_ACCT_ENABLED | 1650 QCI_SYSFILE; 1651 break; 1652 case GFS2_QUOTA_OFF: 1653 break; 1654 } 1655 if (sdp->sd_quota_inode) { 1656 state->s_state[USRQUOTA].ino = 1657 GFS2_I(sdp->sd_quota_inode)->i_no_addr; 1658 state->s_state[USRQUOTA].blocks = sdp->sd_quota_inode->i_blocks; 1659 } 1660 state->s_state[USRQUOTA].nextents = 1; /* unsupported */ 1661 state->s_state[GRPQUOTA] = state->s_state[USRQUOTA]; 1662 state->s_incoredqs = list_lru_count(&gfs2_qd_lru); 1663 return 0; 1664 } 1665 1666 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid, 1667 struct qc_dqblk *fdq) 1668 { 1669 struct gfs2_sbd *sdp = sb->s_fs_info; 1670 struct gfs2_quota_lvb *qlvb; 1671 struct gfs2_quota_data *qd; 1672 struct gfs2_holder q_gh; 1673 int error; 1674 1675 memset(fdq, 0, sizeof(*fdq)); 1676 1677 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF) 1678 return -ESRCH; /* Crazy XFS error code */ 1679 1680 if ((qid.type != USRQUOTA) && 1681 (qid.type != GRPQUOTA)) 1682 return -EINVAL; 1683 1684 error = qd_get(sdp, qid, &qd); 1685 if (error) 1686 return error; 1687 error = do_glock(qd, FORCE, &q_gh); 1688 if (error) 1689 goto out; 1690 1691 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr; 1692 fdq->d_spc_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_sb.sb_bsize_shift; 1693 fdq->d_spc_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_sb.sb_bsize_shift; 1694 fdq->d_space = be64_to_cpu(qlvb->qb_value) << sdp->sd_sb.sb_bsize_shift; 1695 1696 gfs2_glock_dq_uninit(&q_gh); 1697 out: 1698 qd_put(qd); 1699 return error; 1700 } 1701 1702 /* GFS2 only supports a subset of the XFS fields */ 1703 #define GFS2_FIELDMASK (QC_SPC_SOFT|QC_SPC_HARD|QC_SPACE) 1704 1705 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid, 1706 struct qc_dqblk *fdq) 1707 { 1708 struct gfs2_sbd *sdp = sb->s_fs_info; 1709 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode); 1710 struct gfs2_quota_data *qd; 1711 struct gfs2_holder q_gh, i_gh; 1712 unsigned int data_blocks, ind_blocks; 1713 unsigned int blocks = 0; 1714 int alloc_required; 1715 loff_t offset; 1716 int error; 1717 1718 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF) 1719 return -ESRCH; /* Crazy XFS error code */ 1720 1721 if ((qid.type != USRQUOTA) && 1722 (qid.type != GRPQUOTA)) 1723 return -EINVAL; 1724 1725 if (fdq->d_fieldmask & ~GFS2_FIELDMASK) 1726 return -EINVAL; 1727 1728 error = qd_get(sdp, qid, &qd); 1729 if (error) 1730 return error; 1731 1732 error = gfs2_qa_get(ip); 1733 if (error) 1734 goto out_put; 1735 1736 inode_lock(&ip->i_inode); 1737 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh); 1738 if (error) 1739 goto out_unlockput; 1740 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh); 1741 if (error) 1742 goto out_q; 1743 1744 /* Check for existing entry, if none then alloc new blocks */ 1745 error = update_qd(sdp, qd); 1746 if (error) 1747 goto out_i; 1748 1749 /* If nothing has changed, this is a no-op */ 1750 if ((fdq->d_fieldmask & QC_SPC_SOFT) && 1751 ((fdq->d_spc_softlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_warn))) 1752 fdq->d_fieldmask ^= QC_SPC_SOFT; 1753 1754 if ((fdq->d_fieldmask & QC_SPC_HARD) && 1755 ((fdq->d_spc_hardlimit >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_limit))) 1756 fdq->d_fieldmask ^= QC_SPC_HARD; 1757 1758 if ((fdq->d_fieldmask & QC_SPACE) && 1759 ((fdq->d_space >> sdp->sd_sb.sb_bsize_shift) == be64_to_cpu(qd->qd_qb.qb_value))) 1760 fdq->d_fieldmask ^= QC_SPACE; 1761 1762 if (fdq->d_fieldmask == 0) 1763 goto out_i; 1764 1765 offset = qd2offset(qd); 1766 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota)); 1767 if (gfs2_is_stuffed(ip)) 1768 alloc_required = 1; 1769 if (alloc_required) { 1770 struct gfs2_alloc_parms ap = {}; 1771 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota), 1772 &data_blocks, &ind_blocks); 1773 blocks = 1 + data_blocks + ind_blocks; 1774 ap.target = blocks; 1775 error = gfs2_inplace_reserve(ip, &ap); 1776 if (error) 1777 goto out_i; 1778 blocks += gfs2_rg_blocks(ip, blocks); 1779 } 1780 1781 /* Some quotas span block boundaries and can update two blocks, 1782 adding an extra block to the transaction to handle such quotas */ 1783 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0); 1784 if (error) 1785 goto out_release; 1786 1787 /* Apply changes */ 1788 error = gfs2_adjust_quota(sdp, offset, 0, qd, fdq); 1789 if (!error) 1790 clear_bit(QDF_QMSG_QUIET, &qd->qd_flags); 1791 1792 gfs2_trans_end(sdp); 1793 out_release: 1794 if (alloc_required) 1795 gfs2_inplace_release(ip); 1796 out_i: 1797 gfs2_glock_dq_uninit(&i_gh); 1798 out_q: 1799 gfs2_glock_dq_uninit(&q_gh); 1800 out_unlockput: 1801 gfs2_qa_put(ip); 1802 inode_unlock(&ip->i_inode); 1803 out_put: 1804 qd_put(qd); 1805 return error; 1806 } 1807 1808 const struct quotactl_ops gfs2_quotactl_ops = { 1809 .quota_sync = gfs2_quota_sync, 1810 .get_state = gfs2_quota_get_state, 1811 .get_dqblk = gfs2_get_dqblk, 1812 .set_dqblk = gfs2_set_dqblk, 1813 }; 1814 1815 void __init gfs2_quota_hash_init(void) 1816 { 1817 unsigned i; 1818 1819 for(i = 0; i < GFS2_QD_HASH_SIZE; i++) 1820 INIT_HLIST_BL_HEAD(&qd_hash_table[i]); 1821 } 1822