fs/ceph/quota.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * quota.c - CephFS quota
   4  *
   5  * Copyright (C) 2017-2018 SUSE
   6  */
   7
   8 #include <linux/statfs.h>
   9
  10 #include "super.h"
  11 #include "mds_client.h"
  12
  13 void ceph_adjust_quota_realms_count(struct inode *inode, bool inc)
  14 {
  15         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
  16         if (inc)
  17                 atomic64_inc(&mdsc->quotarealms_count);
  18         else
  19                 atomic64_dec(&mdsc->quotarealms_count);
  20 }
  21
  22 static inline bool ceph_has_realms_with_quotas(struct inode *inode)
  23 {
  24         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
  25         struct super_block *sb = mdsc->fsc->sb;
  26
  27         if (atomic64_read(&mdsc->quotarealms_count) > 0)
  28                 return true;
  29         /* if root is the real CephFS root, we don't have quota realms */
  30         if (sb->s_root->d_inode &&
  31             (sb->s_root->d_inode->i_ino == CEPH_INO_ROOT))
  32                 return false;
  33         /* otherwise, we can't know for sure */
  34         return true;
  35 }
  36
  37 void ceph_handle_quota(struct ceph_mds_client *mdsc,
  38                        struct ceph_mds_session *session,
  39                        struct ceph_msg *msg)
  40 {
  41         struct super_block *sb = mdsc->fsc->sb;
  42         struct ceph_mds_quota *h = msg->front.iov_base;
  43         struct ceph_vino vino;
  44         struct inode *inode;
  45         struct ceph_inode_info *ci;
  46
  47         if (msg->front.iov_len < sizeof(*h)) {
  48                 pr_err("%s corrupt message mds%d len %d\n", __func__,
  49                        session->s_mds, (int)msg->front.iov_len);
  50                 ceph_msg_dump(msg);
  51                 return;
  52         }
  53
  54         /* increment msg sequence number */
  55         mutex_lock(&session->s_mutex);
  56         session->s_seq++;
  57         mutex_unlock(&session->s_mutex);
  58
  59         /* lookup inode */
  60         vino.ino = le64_to_cpu(h->ino);
  61         vino.snap = CEPH_NOSNAP;
  62         inode = ceph_find_inode(sb, vino);
  63         if (!inode) {
  64                 pr_warn("Failed to find inode %llu\n", vino.ino);
  65                 return;
  66         }
  67         ci = ceph_inode(inode);
  68
  69         spin_lock(&ci->i_ceph_lock);
  70         ci->i_rbytes = le64_to_cpu(h->rbytes);
  71         ci->i_rfiles = le64_to_cpu(h->rfiles);
  72         ci->i_rsubdirs = le64_to_cpu(h->rsubdirs);
  73         __ceph_update_quota(ci, le64_to_cpu(h->max_bytes),
  74                             le64_to_cpu(h->max_files));
  75         spin_unlock(&ci->i_ceph_lock);
  76
  77         iput(inode);
  78 }
  79
  80 static struct ceph_quotarealm_inode *
  81 find_quotarealm_inode(struct ceph_mds_client *mdsc, u64 ino)
  82 {
  83         struct ceph_quotarealm_inode *qri = NULL;
  84         struct rb_node **node, *parent = NULL;
  85
  86         mutex_lock(&mdsc->quotarealms_inodes_mutex);
  87         node = &(mdsc->quotarealms_inodes.rb_node);
  88         while (*node) {
  89                 parent = *node;
  90                 qri = container_of(*node, struct ceph_quotarealm_inode, node);
  91
  92                 if (ino < qri->ino)
  93                         node = &((*node)->rb_left);
  94                 else if (ino > qri->ino)
  95                         node = &((*node)->rb_right);
  96                 else
  97                         break;
  98         }
  99         if (!qri || (qri->ino != ino)) {
 100                 /* Not found, create a new one and insert it */
 101                 qri = kmalloc(sizeof(*qri), GFP_KERNEL);
 102                 if (qri) {
 103                         qri->ino = ino;
 104                         qri->inode = NULL;
 105                         qri->timeout = 0;
 106                         mutex_init(&qri->mutex);
 107                         rb_link_node(&qri->node, parent, node);
 108                         rb_insert_color(&qri->node, &mdsc->quotarealms_inodes);
 109                 } else
 110                         pr_warn("Failed to alloc quotarealms_inode\n");
 111         }
 112         mutex_unlock(&mdsc->quotarealms_inodes_mutex);
 113
 114         return qri;
 115 }
 116
 117 /*
 118  * This function will try to lookup a realm inode which isn't visible in the
 119  * filesystem mountpoint.  A list of these kind of inodes (not visible) is
 120  * maintained in the mdsc and freed only when the filesystem is umounted.
 121  *
 122  * Note that these inodes are kept in this list even if the lookup fails, which
 123  * allows to prevent useless lookup requests.
 124  */
 125 static struct inode *lookup_quotarealm_inode(struct ceph_mds_client *mdsc,
 126                                              struct super_block *sb,
 127                                              struct ceph_snap_realm *realm)
 128 {
 129         struct ceph_quotarealm_inode *qri;
 130         struct inode *in;
 131
 132         qri = find_quotarealm_inode(mdsc, realm->ino);
 133         if (!qri)
 134                 return NULL;
 135
 136         mutex_lock(&qri->mutex);
 137         if (qri->inode) {
 138                 /* A request has already returned the inode */
 139                 mutex_unlock(&qri->mutex);
 140                 return qri->inode;
 141         }
 142         /* Check if this inode lookup has failed recently */
 143         if (qri->timeout &&
 144             time_before_eq(jiffies, qri->timeout)) {
 145                 mutex_unlock(&qri->mutex);
 146                 return NULL;
 147         }
 148         in = ceph_lookup_inode(sb, realm->ino);
 149         if (IS_ERR(in)) {
 150                 pr_warn("Can't lookup inode %llx (err: %ld)\n",
 151                         realm->ino, PTR_ERR(in));
 152                 qri->timeout = jiffies + msecs_to_jiffies(60 * 1000); /* XXX */
 153         } else {
 154                 qri->timeout = 0;
 155                 qri->inode = in;
 156         }
 157         mutex_unlock(&qri->mutex);
 158
 159         return in;
 160 }
 161
 162 void ceph_cleanup_quotarealms_inodes(struct ceph_mds_client *mdsc)
 163 {
 164         struct ceph_quotarealm_inode *qri;
 165         struct rb_node *node;
 166
 167         /*
 168          * It should now be safe to clean quotarealms_inode tree without holding
 169          * mdsc->quotarealms_inodes_mutex...
 170          */
 171         mutex_lock(&mdsc->quotarealms_inodes_mutex);
 172         while (!RB_EMPTY_ROOT(&mdsc->quotarealms_inodes)) {
 173                 node = rb_first(&mdsc->quotarealms_inodes);
 174                 qri = rb_entry(node, struct ceph_quotarealm_inode, node);
 175                 rb_erase(node, &mdsc->quotarealms_inodes);
 176                 iput(qri->inode);
 177                 kfree(qri);
 178         }
 179         mutex_unlock(&mdsc->quotarealms_inodes_mutex);
 180 }
 181
 182 /*
 183  * This function walks through the snaprealm for an inode and returns the
 184  * ceph_snap_realm for the first snaprealm that has quotas set (either max_files
 185  * or max_bytes).  If the root is reached, return the root ceph_snap_realm
 186  * instead.
 187  *
 188  * Note that the caller is responsible for calling ceph_put_snap_realm() on the
 189  * returned realm.
 190  *
 191  * Callers of this function need to hold mdsc->snap_rwsem.  However, if there's
 192  * a need to do an inode lookup, this rwsem will be temporarily dropped.  Hence
 193  * the 'retry' argument: if rwsem needs to be dropped and 'retry' is 'false'
 194  * this function will return -EAGAIN; otherwise, the snaprealms walk-through
 195  * will be restarted.
 196  */
 197 static struct ceph_snap_realm *get_quota_realm(struct ceph_mds_client *mdsc,
 198                                                struct inode *inode, bool retry)
 199 {
 200         struct ceph_inode_info *ci = NULL;
 201         struct ceph_snap_realm *realm, *next;
 202         struct inode *in;
 203         bool has_quota;
 204
 205         if (ceph_snap(inode) != CEPH_NOSNAP)
 206                 return NULL;
 207
 208 restart:
 209         realm = ceph_inode(inode)->i_snap_realm;
 210         if (realm)
 211                 ceph_get_snap_realm(mdsc, realm);
 212         else
 213                 pr_err_ratelimited("get_quota_realm: ino (%llx.%llx) "
 214                                    "null i_snap_realm\n", ceph_vinop(inode));
 215         while (realm) {
 216                 bool has_inode;
 217
 218                 spin_lock(&realm->inodes_with_caps_lock);
 219                 has_inode = realm->inode;
 220                 in = has_inode ? igrab(realm->inode) : NULL;
 221                 spin_unlock(&realm->inodes_with_caps_lock);
 222                 if (has_inode && !in)
 223                         break;
 224                 if (!in) {
 225                         up_read(&mdsc->snap_rwsem);
 226                         in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
 227                         down_read(&mdsc->snap_rwsem);
 228                         if (IS_ERR_OR_NULL(in))
 229                                 break;
 230                         ceph_put_snap_realm(mdsc, realm);
 231                         if (!retry)
 232                                 return ERR_PTR(-EAGAIN);
 233                         goto restart;
 234                 }
 235
 236                 ci = ceph_inode(in);
 237                 has_quota = __ceph_has_any_quota(ci);
 238                 iput(in);
 239
 240                 next = realm->parent;
 241                 if (has_quota || !next)
 242                        return realm;
 243
 244                 ceph_get_snap_realm(mdsc, next);
 245                 ceph_put_snap_realm(mdsc, realm);
 246                 realm = next;
 247         }
 248         if (realm)
 249                 ceph_put_snap_realm(mdsc, realm);
 250
 251         return NULL;
 252 }
 253
 254 bool ceph_quota_is_same_realm(struct inode *old, struct inode *new)
 255 {
 256         struct ceph_mds_client *mdsc = ceph_inode_to_client(old)->mdsc;
 257         struct ceph_snap_realm *old_realm, *new_realm;
 258         bool is_same;
 259
 260 restart:
 261         /*
 262          * We need to lookup 2 quota realms atomically, i.e. with snap_rwsem.
 263          * However, get_quota_realm may drop it temporarily.  By setting the
 264          * 'retry' parameter to 'false', we'll get -EAGAIN if the rwsem was
 265          * dropped and we can then restart the whole operation.
 266          */
 267         down_read(&mdsc->snap_rwsem);
 268         old_realm = get_quota_realm(mdsc, old, true);
 269         new_realm = get_quota_realm(mdsc, new, false);
 270         if (PTR_ERR(new_realm) == -EAGAIN) {
 271                 up_read(&mdsc->snap_rwsem);
 272                 if (old_realm)
 273                         ceph_put_snap_realm(mdsc, old_realm);
 274                 goto restart;
 275         }
 276         is_same = (old_realm == new_realm);
 277         up_read(&mdsc->snap_rwsem);
 278
 279         if (old_realm)
 280                 ceph_put_snap_realm(mdsc, old_realm);
 281         if (new_realm)
 282                 ceph_put_snap_realm(mdsc, new_realm);
 283
 284         return is_same;
 285 }
 286
 287 enum quota_check_op {
 288         QUOTA_CHECK_MAX_FILES_OP,       /* check quota max_files limit */
 289         QUOTA_CHECK_MAX_BYTES_OP,       /* check quota max_files limit */
 290         QUOTA_CHECK_MAX_BYTES_APPROACHING_OP    /* check if quota max_files
 291                                                    limit is approaching */
 292 };
 293
 294 /*
 295  * check_quota_exceeded() will walk up the snaprealm hierarchy and, for each
 296  * realm, it will execute quota check operation defined by the 'op' parameter.
 297  * The snaprealm walk is interrupted if the quota check detects that the quota
 298  * is exceeded or if the root inode is reached.
 299  */
 300 static bool check_quota_exceeded(struct inode *inode, enum quota_check_op op,
 301                                  loff_t delta)
 302 {
 303         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 304         struct ceph_inode_info *ci;
 305         struct ceph_snap_realm *realm, *next;
 306         struct inode *in;
 307         u64 max, rvalue;
 308         bool exceeded = false;
 309
 310         if (ceph_snap(inode) != CEPH_NOSNAP)
 311                 return false;
 312
 313         down_read(&mdsc->snap_rwsem);
 314 restart:
 315         realm = ceph_inode(inode)->i_snap_realm;
 316         if (realm)
 317                 ceph_get_snap_realm(mdsc, realm);
 318         else
 319                 pr_err_ratelimited("check_quota_exceeded: ino (%llx.%llx) "
 320                                    "null i_snap_realm\n", ceph_vinop(inode));
 321         while (realm) {
 322                 bool has_inode;
 323
 324                 spin_lock(&realm->inodes_with_caps_lock);
 325                 has_inode = realm->inode;
 326                 in = has_inode ? igrab(realm->inode) : NULL;
 327                 spin_unlock(&realm->inodes_with_caps_lock);
 328                 if (has_inode && !in)
 329                         break;
 330                 if (!in) {
 331                         up_read(&mdsc->snap_rwsem);
 332                         in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
 333                         down_read(&mdsc->snap_rwsem);
 334                         if (IS_ERR_OR_NULL(in))
 335                                 break;
 336                         ceph_put_snap_realm(mdsc, realm);
 337                         goto restart;
 338                 }
 339                 ci = ceph_inode(in);
 340                 spin_lock(&ci->i_ceph_lock);
 341                 if (op == QUOTA_CHECK_MAX_FILES_OP) {
 342                         max = ci->i_max_files;
 343                         rvalue = ci->i_rfiles + ci->i_rsubdirs;
 344                 } else {
 345                         max = ci->i_max_bytes;
 346                         rvalue = ci->i_rbytes;
 347                 }
 348                 spin_unlock(&ci->i_ceph_lock);
 349                 switch (op) {
 350                 case QUOTA_CHECK_MAX_FILES_OP:
 351                         exceeded = (max && (rvalue >= max));
 352                         break;
 353                 case QUOTA_CHECK_MAX_BYTES_OP:
 354                         exceeded = (max && (rvalue + delta > max));
 355                         break;
 356                 case QUOTA_CHECK_MAX_BYTES_APPROACHING_OP:
 357                         if (max) {
 358                                 if (rvalue >= max)
 359                                         exceeded = true;
 360                                 else {
 361                                         /*
 362                                          * when we're writing more that 1/16th
 363                                          * of the available space
 364                                          */
 365                                         exceeded =
 366                                                 (((max - rvalue) >> 4) < delta);
 367                                 }
 368                         }
 369                         break;
 370                 default:
 371                         /* Shouldn't happen */
 372                         pr_warn("Invalid quota check op (%d)\n", op);
 373                         exceeded = true; /* Just break the loop */
 374                 }
 375                 iput(in);
 376
 377                 next = realm->parent;
 378                 if (exceeded || !next)
 379                         break;
 380                 ceph_get_snap_realm(mdsc, next);
 381                 ceph_put_snap_realm(mdsc, realm);
 382                 realm = next;
 383         }
 384         if (realm)
 385                 ceph_put_snap_realm(mdsc, realm);
 386         up_read(&mdsc->snap_rwsem);
 387
 388         return exceeded;
 389 }
 390
 391 /*
 392  * ceph_quota_is_max_files_exceeded - check if we can create a new file
 393  * @inode:      directory where a new file is being created
 394  *
 395  * This functions returns true is max_files quota allows a new file to be
 396  * created.  It is necessary to walk through the snaprealm hierarchy (until the
 397  * FS root) to check all realms with quotas set.
 398  */
 399 bool ceph_quota_is_max_files_exceeded(struct inode *inode)
 400 {
 401         if (!ceph_has_realms_with_quotas(inode))
 402                 return false;
 403
 404         WARN_ON(!S_ISDIR(inode->i_mode));
 405
 406         return check_quota_exceeded(inode, QUOTA_CHECK_MAX_FILES_OP, 0);
 407 }
 408
 409 /*
 410  * ceph_quota_is_max_bytes_exceeded - check if we can write to a file
 411  * @inode:      inode being written
 412  * @newsize:    new size if write succeeds
 413  *
 414  * This functions returns true is max_bytes quota allows a file size to reach
 415  * @newsize; it returns false otherwise.
 416  */
 417 bool ceph_quota_is_max_bytes_exceeded(struct inode *inode, loff_t newsize)
 418 {
 419         loff_t size = i_size_read(inode);
 420
 421         if (!ceph_has_realms_with_quotas(inode))
 422                 return false;
 423
 424         /* return immediately if we're decreasing file size */
 425         if (newsize <= size)
 426                 return false;
 427
 428         return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_OP, (newsize - size));
 429 }
 430
 431 /*
 432  * ceph_quota_is_max_bytes_approaching - check if we're reaching max_bytes
 433  * @inode:      inode being written
 434  * @newsize:    new size if write succeeds
 435  *
 436  * This function returns true if the new file size @newsize will be consuming
 437  * more than 1/16th of the available quota space; it returns false otherwise.
 438  */
 439 bool ceph_quota_is_max_bytes_approaching(struct inode *inode, loff_t newsize)
 440 {
 441         loff_t size = ceph_inode(inode)->i_reported_size;
 442
 443         if (!ceph_has_realms_with_quotas(inode))
 444                 return false;
 445
 446         /* return immediately if we're decreasing file size */
 447         if (newsize <= size)
 448                 return false;
 449
 450         return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_APPROACHING_OP,
 451                                     (newsize - size));
 452 }
 453
 454 /*
 455  * ceph_quota_update_statfs - if root has quota update statfs with quota status
 456  * @fsc:        filesystem client instance
 457  * @buf:        statfs to update
 458  *
 459  * If the mounted filesystem root has max_bytes quota set, update the filesystem
 460  * statistics with the quota status.
 461  *
 462  * This function returns true if the stats have been updated, false otherwise.
 463  */
 464 bool ceph_quota_update_statfs(struct ceph_fs_client *fsc, struct kstatfs *buf)
 465 {
 466         struct ceph_mds_client *mdsc = fsc->mdsc;
 467         struct ceph_inode_info *ci;
 468         struct ceph_snap_realm *realm;
 469         struct inode *in;
 470         u64 total = 0, used, free;
 471         bool is_updated = false;
 472
 473         down_read(&mdsc->snap_rwsem);
 474         realm = get_quota_realm(mdsc, d_inode(fsc->sb->s_root), true);
 475         up_read(&mdsc->snap_rwsem);
 476         if (!realm)
 477                 return false;
 478
 479         spin_lock(&realm->inodes_with_caps_lock);
 480         in = realm->inode ? igrab(realm->inode) : NULL;
 481         spin_unlock(&realm->inodes_with_caps_lock);
 482         if (in) {
 483                 ci = ceph_inode(in);
 484                 spin_lock(&ci->i_ceph_lock);
 485                 if (ci->i_max_bytes) {
 486                         total = ci->i_max_bytes >> CEPH_BLOCK_SHIFT;
 487                         used = ci->i_rbytes >> CEPH_BLOCK_SHIFT;
 488                         /* It is possible for a quota to be exceeded.
 489                          * Report 'zero' in that case
 490                          */
 491                         free = total > used ? total - used : 0;
 492                 }
 493                 spin_unlock(&ci->i_ceph_lock);
 494                 if (total) {
 495                         buf->f_blocks = total;
 496                         buf->f_bfree = free;
 497                         buf->f_bavail = free;
 498                         is_updated = true;
 499                 }
 500                 iput(in);
 501         }
 502         ceph_put_snap_realm(mdsc, realm);
 503
 504         return is_updated;
 505 }
 506