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         /* avoid calling iput_final() in dispatch thread */
  78         ceph_async_iput(inode);
  79 }
  80
  81 static struct ceph_quotarealm_inode *
  82 find_quotarealm_inode(struct ceph_mds_client *mdsc, u64 ino)
  83 {
  84         struct ceph_quotarealm_inode *qri = NULL;
  85         struct rb_node **node, *parent = NULL;
  86
  87         mutex_lock(&mdsc->quotarealms_inodes_mutex);
  88         node = &(mdsc->quotarealms_inodes.rb_node);
  89         while (*node) {
  90                 parent = *node;
  91                 qri = container_of(*node, struct ceph_quotarealm_inode, node);
  92
  93                 if (ino < qri->ino)
  94                         node = &((*node)->rb_left);
  95                 else if (ino > qri->ino)
  96                         node = &((*node)->rb_right);
  97                 else
  98                         break;
  99         }
 100         if (!qri || (qri->ino != ino)) {
 101                 /* Not found, create a new one and insert it */
 102                 qri = kmalloc(sizeof(*qri), GFP_KERNEL);
 103                 if (qri) {
 104                         qri->ino = ino;
 105                         qri->inode = NULL;
 106                         qri->timeout = 0;
 107                         mutex_init(&qri->mutex);
 108                         rb_link_node(&qri->node, parent, node);
 109                         rb_insert_color(&qri->node, &mdsc->quotarealms_inodes);
 110                 } else
 111                         pr_warn("Failed to alloc quotarealms_inode\n");
 112         }
 113         mutex_unlock(&mdsc->quotarealms_inodes_mutex);
 114
 115         return qri;
 116 }
 117
 118 /*
 119  * This function will try to lookup a realm inode which isn't visible in the
 120  * filesystem mountpoint.  A list of these kind of inodes (not visible) is
 121  * maintained in the mdsc and freed only when the filesystem is umounted.
 122  *
 123  * Note that these inodes are kept in this list even if the lookup fails, which
 124  * allows to prevent useless lookup requests.
 125  */
 126 static struct inode *lookup_quotarealm_inode(struct ceph_mds_client *mdsc,
 127                                              struct super_block *sb,
 128                                              struct ceph_snap_realm *realm)
 129 {
 130         struct ceph_quotarealm_inode *qri;
 131         struct inode *in;
 132
 133         qri = find_quotarealm_inode(mdsc, realm->ino);
 134         if (!qri)
 135                 return NULL;
 136
 137         mutex_lock(&qri->mutex);
 138         if (qri->inode) {
 139                 /* A request has already returned the inode */
 140                 mutex_unlock(&qri->mutex);
 141                 return qri->inode;
 142         }
 143         /* Check if this inode lookup has failed recently */
 144         if (qri->timeout &&
 145             time_before_eq(jiffies, qri->timeout)) {
 146                 mutex_unlock(&qri->mutex);
 147                 return NULL;
 148         }
 149         in = ceph_lookup_inode(sb, realm->ino);
 150         if (IS_ERR(in)) {
 151                 pr_warn("Can't lookup inode %llx (err: %ld)\n",
 152                         realm->ino, PTR_ERR(in));
 153                 qri->timeout = jiffies + msecs_to_jiffies(60 * 1000); /* XXX */
 154         } else {
 155                 qri->timeout = 0;
 156                 qri->inode = in;
 157         }
 158         mutex_unlock(&qri->mutex);
 159
 160         return in;
 161 }
 162
 163 void ceph_cleanup_quotarealms_inodes(struct ceph_mds_client *mdsc)
 164 {
 165         struct ceph_quotarealm_inode *qri;
 166         struct rb_node *node;
 167
 168         /*
 169          * It should now be safe to clean quotarealms_inode tree without holding
 170          * mdsc->quotarealms_inodes_mutex...
 171          */
 172         mutex_lock(&mdsc->quotarealms_inodes_mutex);
 173         while (!RB_EMPTY_ROOT(&mdsc->quotarealms_inodes)) {
 174                 node = rb_first(&mdsc->quotarealms_inodes);
 175                 qri = rb_entry(node, struct ceph_quotarealm_inode, node);
 176                 rb_erase(node, &mdsc->quotarealms_inodes);
 177                 iput(qri->inode);
 178                 kfree(qri);
 179         }
 180         mutex_unlock(&mdsc->quotarealms_inodes_mutex);
 181 }
 182
 183 /*
 184  * This function walks through the snaprealm for an inode and returns the
 185  * ceph_snap_realm for the first snaprealm that has quotas set (either max_files
 186  * or max_bytes).  If the root is reached, return the root ceph_snap_realm
 187  * instead.
 188  *
 189  * Note that the caller is responsible for calling ceph_put_snap_realm() on the
 190  * returned realm.
 191  *
 192  * Callers of this function need to hold mdsc->snap_rwsem.  However, if there's
 193  * a need to do an inode lookup, this rwsem will be temporarily dropped.  Hence
 194  * the 'retry' argument: if rwsem needs to be dropped and 'retry' is 'false'
 195  * this function will return -EAGAIN; otherwise, the snaprealms walk-through
 196  * will be restarted.
 197  */
 198 static struct ceph_snap_realm *get_quota_realm(struct ceph_mds_client *mdsc,
 199                                                struct inode *inode, bool retry)
 200 {
 201         struct ceph_inode_info *ci = NULL;
 202         struct ceph_snap_realm *realm, *next;
 203         struct inode *in;
 204         bool has_quota;
 205
 206         if (ceph_snap(inode) != CEPH_NOSNAP)
 207                 return NULL;
 208
 209 restart:
 210         realm = ceph_inode(inode)->i_snap_realm;
 211         if (realm)
 212                 ceph_get_snap_realm(mdsc, realm);
 213         else
 214                 pr_err_ratelimited("get_quota_realm: ino (%llx.%llx) "
 215                                    "null i_snap_realm\n", ceph_vinop(inode));
 216         while (realm) {
 217                 bool has_inode;
 218
 219                 spin_lock(&realm->inodes_with_caps_lock);
 220                 has_inode = realm->inode;
 221                 in = has_inode ? igrab(realm->inode) : NULL;
 222                 spin_unlock(&realm->inodes_with_caps_lock);
 223                 if (has_inode && !in)
 224                         break;
 225                 if (!in) {
 226                         up_read(&mdsc->snap_rwsem);
 227                         in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
 228                         down_read(&mdsc->snap_rwsem);
 229                         if (IS_ERR_OR_NULL(in))
 230                                 break;
 231                         ceph_put_snap_realm(mdsc, realm);
 232                         if (!retry)
 233                                 return ERR_PTR(-EAGAIN);
 234                         goto restart;
 235                 }
 236
 237                 ci = ceph_inode(in);
 238                 has_quota = __ceph_has_any_quota(ci);
 239                 /* avoid calling iput_final() while holding mdsc->snap_rwsem */
 240                 ceph_async_iput(in);
 241
 242                 next = realm->parent;
 243                 if (has_quota || !next)
 244                        return realm;
 245
 246                 ceph_get_snap_realm(mdsc, next);
 247                 ceph_put_snap_realm(mdsc, realm);
 248                 realm = next;
 249         }
 250         if (realm)
 251                 ceph_put_snap_realm(mdsc, realm);
 252
 253         return NULL;
 254 }
 255
 256 bool ceph_quota_is_same_realm(struct inode *old, struct inode *new)
 257 {
 258         struct ceph_mds_client *mdsc = ceph_inode_to_client(old)->mdsc;
 259         struct ceph_snap_realm *old_realm, *new_realm;
 260         bool is_same;
 261
 262 restart:
 263         /*
 264          * We need to lookup 2 quota realms atomically, i.e. with snap_rwsem.
 265          * However, get_quota_realm may drop it temporarily.  By setting the
 266          * 'retry' parameter to 'false', we'll get -EAGAIN if the rwsem was
 267          * dropped and we can then restart the whole operation.
 268          */
 269         down_read(&mdsc->snap_rwsem);
 270         old_realm = get_quota_realm(mdsc, old, true);
 271         new_realm = get_quota_realm(mdsc, new, false);
 272         if (PTR_ERR(new_realm) == -EAGAIN) {
 273                 up_read(&mdsc->snap_rwsem);
 274                 if (old_realm)
 275                         ceph_put_snap_realm(mdsc, old_realm);
 276                 goto restart;
 277         }
 278         is_same = (old_realm == new_realm);
 279         up_read(&mdsc->snap_rwsem);
 280
 281         if (old_realm)
 282                 ceph_put_snap_realm(mdsc, old_realm);
 283         if (new_realm)
 284                 ceph_put_snap_realm(mdsc, new_realm);
 285
 286         return is_same;
 287 }
 288
 289 enum quota_check_op {
 290         QUOTA_CHECK_MAX_FILES_OP,       /* check quota max_files limit */
 291         QUOTA_CHECK_MAX_BYTES_OP,       /* check quota max_files limit */
 292         QUOTA_CHECK_MAX_BYTES_APPROACHING_OP    /* check if quota max_files
 293                                                    limit is approaching */
 294 };
 295
 296 /*
 297  * check_quota_exceeded() will walk up the snaprealm hierarchy and, for each
 298  * realm, it will execute quota check operation defined by the 'op' parameter.
 299  * The snaprealm walk is interrupted if the quota check detects that the quota
 300  * is exceeded or if the root inode is reached.
 301  */
 302 static bool check_quota_exceeded(struct inode *inode, enum quota_check_op op,
 303                                  loff_t delta)
 304 {
 305         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 306         struct ceph_inode_info *ci;
 307         struct ceph_snap_realm *realm, *next;
 308         struct inode *in;
 309         u64 max, rvalue;
 310         bool exceeded = false;
 311
 312         if (ceph_snap(inode) != CEPH_NOSNAP)
 313                 return false;
 314
 315         down_read(&mdsc->snap_rwsem);
 316 restart:
 317         realm = ceph_inode(inode)->i_snap_realm;
 318         if (realm)
 319                 ceph_get_snap_realm(mdsc, realm);
 320         else
 321                 pr_err_ratelimited("check_quota_exceeded: ino (%llx.%llx) "
 322                                    "null i_snap_realm\n", ceph_vinop(inode));
 323         while (realm) {
 324                 bool has_inode;
 325
 326                 spin_lock(&realm->inodes_with_caps_lock);
 327                 has_inode = realm->inode;
 328                 in = has_inode ? igrab(realm->inode) : NULL;
 329                 spin_unlock(&realm->inodes_with_caps_lock);
 330                 if (has_inode && !in)
 331                         break;
 332                 if (!in) {
 333                         up_read(&mdsc->snap_rwsem);
 334                         in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
 335                         down_read(&mdsc->snap_rwsem);
 336                         if (IS_ERR_OR_NULL(in))
 337                                 break;
 338                         ceph_put_snap_realm(mdsc, realm);
 339                         goto restart;
 340                 }
 341                 ci = ceph_inode(in);
 342                 spin_lock(&ci->i_ceph_lock);
 343                 if (op == QUOTA_CHECK_MAX_FILES_OP) {
 344                         max = ci->i_max_files;
 345                         rvalue = ci->i_rfiles + ci->i_rsubdirs;
 346                 } else {
 347                         max = ci->i_max_bytes;
 348                         rvalue = ci->i_rbytes;
 349                 }
 350                 spin_unlock(&ci->i_ceph_lock);
 351                 switch (op) {
 352                 case QUOTA_CHECK_MAX_FILES_OP:
 353                         exceeded = (max && (rvalue >= max));
 354                         break;
 355                 case QUOTA_CHECK_MAX_BYTES_OP:
 356                         exceeded = (max && (rvalue + delta > max));
 357                         break;
 358                 case QUOTA_CHECK_MAX_BYTES_APPROACHING_OP:
 359                         if (max) {
 360                                 if (rvalue >= max)
 361                                         exceeded = true;
 362                                 else {
 363                                         /*
 364                                          * when we're writing more that 1/16th
 365                                          * of the available space
 366                                          */
 367                                         exceeded =
 368                                                 (((max - rvalue) >> 4) < delta);
 369                                 }
 370                         }
 371                         break;
 372                 default:
 373                         /* Shouldn't happen */
 374                         pr_warn("Invalid quota check op (%d)\n", op);
 375                         exceeded = true; /* Just break the loop */
 376                 }
 377                 /* avoid calling iput_final() while holding mdsc->snap_rwsem */
 378                 ceph_async_iput(in);
 379
 380                 next = realm->parent;
 381                 if (exceeded || !next)
 382                         break;
 383                 ceph_get_snap_realm(mdsc, next);
 384                 ceph_put_snap_realm(mdsc, realm);
 385                 realm = next;
 386         }
 387         if (realm)
 388                 ceph_put_snap_realm(mdsc, realm);
 389         up_read(&mdsc->snap_rwsem);
 390
 391         return exceeded;
 392 }
 393
 394 /*
 395  * ceph_quota_is_max_files_exceeded - check if we can create a new file
 396  * @inode:      directory where a new file is being created
 397  *
 398  * This functions returns true is max_files quota allows a new file to be
 399  * created.  It is necessary to walk through the snaprealm hierarchy (until the
 400  * FS root) to check all realms with quotas set.
 401  */
 402 bool ceph_quota_is_max_files_exceeded(struct inode *inode)
 403 {
 404         if (!ceph_has_realms_with_quotas(inode))
 405                 return false;
 406
 407         WARN_ON(!S_ISDIR(inode->i_mode));
 408
 409         return check_quota_exceeded(inode, QUOTA_CHECK_MAX_FILES_OP, 0);
 410 }
 411
 412 /*
 413  * ceph_quota_is_max_bytes_exceeded - check if we can write to a file
 414  * @inode:      inode being written
 415  * @newsize:    new size if write succeeds
 416  *
 417  * This functions returns true is max_bytes quota allows a file size to reach
 418  * @newsize; it returns false otherwise.
 419  */
 420 bool ceph_quota_is_max_bytes_exceeded(struct inode *inode, loff_t newsize)
 421 {
 422         loff_t size = i_size_read(inode);
 423
 424         if (!ceph_has_realms_with_quotas(inode))
 425                 return false;
 426
 427         /* return immediately if we're decreasing file size */
 428         if (newsize <= size)
 429                 return false;
 430
 431         return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_OP, (newsize - size));
 432 }
 433
 434 /*
 435  * ceph_quota_is_max_bytes_approaching - check if we're reaching max_bytes
 436  * @inode:      inode being written
 437  * @newsize:    new size if write succeeds
 438  *
 439  * This function returns true if the new file size @newsize will be consuming
 440  * more than 1/16th of the available quota space; it returns false otherwise.
 441  */
 442 bool ceph_quota_is_max_bytes_approaching(struct inode *inode, loff_t newsize)
 443 {
 444         loff_t size = ceph_inode(inode)->i_reported_size;
 445
 446         if (!ceph_has_realms_with_quotas(inode))
 447                 return false;
 448
 449         /* return immediately if we're decreasing file size */
 450         if (newsize <= size)
 451                 return false;
 452
 453         return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_APPROACHING_OP,
 454                                     (newsize - size));
 455 }
 456
 457 /*
 458  * ceph_quota_update_statfs - if root has quota update statfs with quota status
 459  * @fsc:        filesystem client instance
 460  * @buf:        statfs to update
 461  *
 462  * If the mounted filesystem root has max_bytes quota set, update the filesystem
 463  * statistics with the quota status.
 464  *
 465  * This function returns true if the stats have been updated, false otherwise.
 466  */
 467 bool ceph_quota_update_statfs(struct ceph_fs_client *fsc, struct kstatfs *buf)
 468 {
 469         struct ceph_mds_client *mdsc = fsc->mdsc;
 470         struct ceph_inode_info *ci;
 471         struct ceph_snap_realm *realm;
 472         struct inode *in;
 473         u64 total = 0, used, free;
 474         bool is_updated = false;
 475
 476         down_read(&mdsc->snap_rwsem);
 477         realm = get_quota_realm(mdsc, d_inode(fsc->sb->s_root), true);
 478         up_read(&mdsc->snap_rwsem);
 479         if (!realm)
 480                 return false;
 481
 482         spin_lock(&realm->inodes_with_caps_lock);
 483         in = realm->inode ? igrab(realm->inode) : NULL;
 484         spin_unlock(&realm->inodes_with_caps_lock);
 485         if (in) {
 486                 ci = ceph_inode(in);
 487                 spin_lock(&ci->i_ceph_lock);
 488                 if (ci->i_max_bytes) {
 489                         total = ci->i_max_bytes >> CEPH_BLOCK_SHIFT;
 490                         used = ci->i_rbytes >> CEPH_BLOCK_SHIFT;
 491                         /* It is possible for a quota to be exceeded.
 492                          * Report 'zero' in that case
 493                          */
 494                         free = total > used ? total - used : 0;
 495                 }
 496                 spin_unlock(&ci->i_ceph_lock);
 497                 if (total) {
 498                         buf->f_blocks = total;
 499                         buf->f_bfree = free;
 500                         buf->f_bavail = free;
 501                         is_updated = true;
 502                 }
 503                 iput(in);
 504         }
 505         ceph_put_snap_realm(mdsc, realm);
 506
 507         return is_updated;
 508 }
 509