fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16 #include <linux/iversion.h>
  17
  18 #include "super.h"
  19 #include "mds_client.h"
  20 #include "cache.h"
  21 #include <linux/ceph/decode.h>
  22
  23 /*
  24  * Ceph inode operations
  25  *
  26  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  27  * setattr, etc.), xattr helpers, and helpers for assimilating
  28  * metadata returned by the MDS into our cache.
  29  *
  30  * Also define helpers for doing asynchronous writeback, invalidation,
  31  * and truncation for the benefit of those who can't afford to block
  32  * (typically because they are in the message handler path).
  33  */
  34
  35 static const struct inode_operations ceph_symlink_iops;
  36
  37 static void ceph_inode_work(struct work_struct *work);
  38
  39 /*
  40  * find or create an inode, given the ceph ino number
  41  */
  42 static int ceph_set_ino_cb(struct inode *inode, void *data)
  43 {
  44         struct ceph_inode_info *ci = ceph_inode(inode);
  45         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
  46
  47         ci->i_vino = *(struct ceph_vino *)data;
  48         inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
  49         inode_set_iversion_raw(inode, 0);
  50         percpu_counter_inc(&mdsc->metric.total_inodes);
  51
  52         return 0;
  53 }
  54
  55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  56 {
  57         struct inode *inode;
  58
  59         if (ceph_vino_is_reserved(vino))
  60                 return ERR_PTR(-EREMOTEIO);
  61
  62         inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
  63                              ceph_set_ino_cb, &vino);
  64         if (!inode)
  65                 return ERR_PTR(-ENOMEM);
  66
  67         dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
  68              ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
  69         return inode;
  70 }
  71
  72 /*
  73  * get/constuct snapdir inode for a given directory
  74  */
  75 struct inode *ceph_get_snapdir(struct inode *parent)
  76 {
  77         struct ceph_vino vino = {
  78                 .ino = ceph_ino(parent),
  79                 .snap = CEPH_SNAPDIR,
  80         };
  81         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  82         struct ceph_inode_info *ci = ceph_inode(inode);
  83
  84         if (IS_ERR(inode))
  85                 return inode;
  86
  87         if (!S_ISDIR(parent->i_mode)) {
  88                 pr_warn_once("bad snapdir parent type (mode=0%o)\n",
  89                              parent->i_mode);
  90                 return ERR_PTR(-ENOTDIR);
  91         }
  92
  93         if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
  94                 pr_warn_once("bad snapdir inode type (mode=0%o)\n",
  95                              inode->i_mode);
  96                 return ERR_PTR(-ENOTDIR);
  97         }
  98
  99         inode->i_mode = parent->i_mode;
 100         inode->i_uid = parent->i_uid;
 101         inode->i_gid = parent->i_gid;
 102         inode->i_mtime = parent->i_mtime;
 103         inode->i_ctime = parent->i_ctime;
 104         inode->i_atime = parent->i_atime;
 105         ci->i_rbytes = 0;
 106         ci->i_btime = ceph_inode(parent)->i_btime;
 107
 108         if (inode->i_state & I_NEW) {
 109                 inode->i_op = &ceph_snapdir_iops;
 110                 inode->i_fop = &ceph_snapdir_fops;
 111                 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
 112                 unlock_new_inode(inode);
 113         }
 114
 115         return inode;
 116 }
 117
 118 const struct inode_operations ceph_file_iops = {
 119         .permission = ceph_permission,
 120         .setattr = ceph_setattr,
 121         .getattr = ceph_getattr,
 122         .listxattr = ceph_listxattr,
 123         .get_acl = ceph_get_acl,
 124         .set_acl = ceph_set_acl,
 125 };
 126
 127
 128 /*
 129  * We use a 'frag tree' to keep track of the MDS's directory fragments
 130  * for a given inode (usually there is just a single fragment).  We
 131  * need to know when a child frag is delegated to a new MDS, or when
 132  * it is flagged as replicated, so we can direct our requests
 133  * accordingly.
 134  */
 135
 136 /*
 137  * find/create a frag in the tree
 138  */
 139 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 140                                                     u32 f)
 141 {
 142         struct rb_node **p;
 143         struct rb_node *parent = NULL;
 144         struct ceph_inode_frag *frag;
 145         int c;
 146
 147         p = &ci->i_fragtree.rb_node;
 148         while (*p) {
 149                 parent = *p;
 150                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 151                 c = ceph_frag_compare(f, frag->frag);
 152                 if (c < 0)
 153                         p = &(*p)->rb_left;
 154                 else if (c > 0)
 155                         p = &(*p)->rb_right;
 156                 else
 157                         return frag;
 158         }
 159
 160         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 161         if (!frag)
 162                 return ERR_PTR(-ENOMEM);
 163
 164         frag->frag = f;
 165         frag->split_by = 0;
 166         frag->mds = -1;
 167         frag->ndist = 0;
 168
 169         rb_link_node(&frag->node, parent, p);
 170         rb_insert_color(&frag->node, &ci->i_fragtree);
 171
 172         dout("get_or_create_frag added %llx.%llx frag %x\n",
 173              ceph_vinop(&ci->vfs_inode), f);
 174         return frag;
 175 }
 176
 177 /*
 178  * find a specific frag @f
 179  */
 180 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 181 {
 182         struct rb_node *n = ci->i_fragtree.rb_node;
 183
 184         while (n) {
 185                 struct ceph_inode_frag *frag =
 186                         rb_entry(n, struct ceph_inode_frag, node);
 187                 int c = ceph_frag_compare(f, frag->frag);
 188                 if (c < 0)
 189                         n = n->rb_left;
 190                 else if (c > 0)
 191                         n = n->rb_right;
 192                 else
 193                         return frag;
 194         }
 195         return NULL;
 196 }
 197
 198 /*
 199  * Choose frag containing the given value @v.  If @pfrag is
 200  * specified, copy the frag delegation info to the caller if
 201  * it is present.
 202  */
 203 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 204                               struct ceph_inode_frag *pfrag, int *found)
 205 {
 206         u32 t = ceph_frag_make(0, 0);
 207         struct ceph_inode_frag *frag;
 208         unsigned nway, i;
 209         u32 n;
 210
 211         if (found)
 212                 *found = 0;
 213
 214         while (1) {
 215                 WARN_ON(!ceph_frag_contains_value(t, v));
 216                 frag = __ceph_find_frag(ci, t);
 217                 if (!frag)
 218                         break; /* t is a leaf */
 219                 if (frag->split_by == 0) {
 220                         if (pfrag)
 221                                 memcpy(pfrag, frag, sizeof(*pfrag));
 222                         if (found)
 223                                 *found = 1;
 224                         break;
 225                 }
 226
 227                 /* choose child */
 228                 nway = 1 << frag->split_by;
 229                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 230                      frag->split_by, nway);
 231                 for (i = 0; i < nway; i++) {
 232                         n = ceph_frag_make_child(t, frag->split_by, i);
 233                         if (ceph_frag_contains_value(n, v)) {
 234                                 t = n;
 235                                 break;
 236                         }
 237                 }
 238                 BUG_ON(i == nway);
 239         }
 240         dout("choose_frag(%x) = %x\n", v, t);
 241
 242         return t;
 243 }
 244
 245 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 246                      struct ceph_inode_frag *pfrag, int *found)
 247 {
 248         u32 ret;
 249         mutex_lock(&ci->i_fragtree_mutex);
 250         ret = __ceph_choose_frag(ci, v, pfrag, found);
 251         mutex_unlock(&ci->i_fragtree_mutex);
 252         return ret;
 253 }
 254
 255 /*
 256  * Process dirfrag (delegation) info from the mds.  Include leaf
 257  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 258  * branches/splits are included in i_fragtree)
 259  */
 260 static int ceph_fill_dirfrag(struct inode *inode,
 261                              struct ceph_mds_reply_dirfrag *dirinfo)
 262 {
 263         struct ceph_inode_info *ci = ceph_inode(inode);
 264         struct ceph_inode_frag *frag;
 265         u32 id = le32_to_cpu(dirinfo->frag);
 266         int mds = le32_to_cpu(dirinfo->auth);
 267         int ndist = le32_to_cpu(dirinfo->ndist);
 268         int diri_auth = -1;
 269         int i;
 270         int err = 0;
 271
 272         spin_lock(&ci->i_ceph_lock);
 273         if (ci->i_auth_cap)
 274                 diri_auth = ci->i_auth_cap->mds;
 275         spin_unlock(&ci->i_ceph_lock);
 276
 277         if (mds == -1) /* CDIR_AUTH_PARENT */
 278                 mds = diri_auth;
 279
 280         mutex_lock(&ci->i_fragtree_mutex);
 281         if (ndist == 0 && mds == diri_auth) {
 282                 /* no delegation info needed. */
 283                 frag = __ceph_find_frag(ci, id);
 284                 if (!frag)
 285                         goto out;
 286                 if (frag->split_by == 0) {
 287                         /* tree leaf, remove */
 288                         dout("fill_dirfrag removed %llx.%llx frag %x"
 289                              " (no ref)\n", ceph_vinop(inode), id);
 290                         rb_erase(&frag->node, &ci->i_fragtree);
 291                         kfree(frag);
 292                 } else {
 293                         /* tree branch, keep and clear */
 294                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 295                              " referral\n", ceph_vinop(inode), id);
 296                         frag->mds = -1;
 297                         frag->ndist = 0;
 298                 }
 299                 goto out;
 300         }
 301
 302
 303         /* find/add this frag to store mds delegation info */
 304         frag = __get_or_create_frag(ci, id);
 305         if (IS_ERR(frag)) {
 306                 /* this is not the end of the world; we can continue
 307                    with bad/inaccurate delegation info */
 308                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 309                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 310                 err = -ENOMEM;
 311                 goto out;
 312         }
 313
 314         frag->mds = mds;
 315         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 316         for (i = 0; i < frag->ndist; i++)
 317                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 318         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 319              ceph_vinop(inode), frag->frag, frag->ndist);
 320
 321 out:
 322         mutex_unlock(&ci->i_fragtree_mutex);
 323         return err;
 324 }
 325
 326 static int frag_tree_split_cmp(const void *l, const void *r)
 327 {
 328         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 329         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 330         return ceph_frag_compare(le32_to_cpu(ls->frag),
 331                                  le32_to_cpu(rs->frag));
 332 }
 333
 334 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 335 {
 336         if (!frag)
 337                 return f == ceph_frag_make(0, 0);
 338         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 339                 return false;
 340         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 341 }
 342
 343 static int ceph_fill_fragtree(struct inode *inode,
 344                               struct ceph_frag_tree_head *fragtree,
 345                               struct ceph_mds_reply_dirfrag *dirinfo)
 346 {
 347         struct ceph_inode_info *ci = ceph_inode(inode);
 348         struct ceph_inode_frag *frag, *prev_frag = NULL;
 349         struct rb_node *rb_node;
 350         unsigned i, split_by, nsplits;
 351         u32 id;
 352         bool update = false;
 353
 354         mutex_lock(&ci->i_fragtree_mutex);
 355         nsplits = le32_to_cpu(fragtree->nsplits);
 356         if (nsplits != ci->i_fragtree_nsplits) {
 357                 update = true;
 358         } else if (nsplits) {
 359                 i = prandom_u32() % nsplits;
 360                 id = le32_to_cpu(fragtree->splits[i].frag);
 361                 if (!__ceph_find_frag(ci, id))
 362                         update = true;
 363         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 364                 rb_node = rb_first(&ci->i_fragtree);
 365                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 366                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 367                         update = true;
 368         }
 369         if (!update && dirinfo) {
 370                 id = le32_to_cpu(dirinfo->frag);
 371                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 372                         update = true;
 373         }
 374         if (!update)
 375                 goto out_unlock;
 376
 377         if (nsplits > 1) {
 378                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 379                      frag_tree_split_cmp, NULL);
 380         }
 381
 382         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 383         rb_node = rb_first(&ci->i_fragtree);
 384         for (i = 0; i < nsplits; i++) {
 385                 id = le32_to_cpu(fragtree->splits[i].frag);
 386                 split_by = le32_to_cpu(fragtree->splits[i].by);
 387                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 388                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 389                                "frag %x split by %d\n", ceph_vinop(inode),
 390                                i, nsplits, id, split_by);
 391                         continue;
 392                 }
 393                 frag = NULL;
 394                 while (rb_node) {
 395                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 396                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 397                                 if (frag->frag != id)
 398                                         frag = NULL;
 399                                 else
 400                                         rb_node = rb_next(rb_node);
 401                                 break;
 402                         }
 403                         rb_node = rb_next(rb_node);
 404                         /* delete stale split/leaf node */
 405                         if (frag->split_by > 0 ||
 406                             !is_frag_child(frag->frag, prev_frag)) {
 407                                 rb_erase(&frag->node, &ci->i_fragtree);
 408                                 if (frag->split_by > 0)
 409                                         ci->i_fragtree_nsplits--;
 410                                 kfree(frag);
 411                         }
 412                         frag = NULL;
 413                 }
 414                 if (!frag) {
 415                         frag = __get_or_create_frag(ci, id);
 416                         if (IS_ERR(frag))
 417                                 continue;
 418                 }
 419                 if (frag->split_by == 0)
 420                         ci->i_fragtree_nsplits++;
 421                 frag->split_by = split_by;
 422                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 423                 prev_frag = frag;
 424         }
 425         while (rb_node) {
 426                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 427                 rb_node = rb_next(rb_node);
 428                 /* delete stale split/leaf node */
 429                 if (frag->split_by > 0 ||
 430                     !is_frag_child(frag->frag, prev_frag)) {
 431                         rb_erase(&frag->node, &ci->i_fragtree);
 432                         if (frag->split_by > 0)
 433                                 ci->i_fragtree_nsplits--;
 434                         kfree(frag);
 435                 }
 436         }
 437 out_unlock:
 438         mutex_unlock(&ci->i_fragtree_mutex);
 439         return 0;
 440 }
 441
 442 /*
 443  * initialize a newly allocated inode.
 444  */
 445 struct inode *ceph_alloc_inode(struct super_block *sb)
 446 {
 447         struct ceph_inode_info *ci;
 448         int i;
 449
 450         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 451         if (!ci)
 452                 return NULL;
 453
 454         dout("alloc_inode %p\n", &ci->vfs_inode);
 455
 456         spin_lock_init(&ci->i_ceph_lock);
 457
 458         ci->i_version = 0;
 459         ci->i_inline_version = 0;
 460         ci->i_time_warp_seq = 0;
 461         ci->i_ceph_flags = 0;
 462         atomic64_set(&ci->i_ordered_count, 1);
 463         atomic64_set(&ci->i_release_count, 1);
 464         atomic64_set(&ci->i_complete_seq[0], 0);
 465         atomic64_set(&ci->i_complete_seq[1], 0);
 466         ci->i_symlink = NULL;
 467
 468         ci->i_max_bytes = 0;
 469         ci->i_max_files = 0;
 470
 471         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 472         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 473         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 474
 475         ci->i_fragtree = RB_ROOT;
 476         mutex_init(&ci->i_fragtree_mutex);
 477
 478         ci->i_xattrs.blob = NULL;
 479         ci->i_xattrs.prealloc_blob = NULL;
 480         ci->i_xattrs.dirty = false;
 481         ci->i_xattrs.index = RB_ROOT;
 482         ci->i_xattrs.count = 0;
 483         ci->i_xattrs.names_size = 0;
 484         ci->i_xattrs.vals_size = 0;
 485         ci->i_xattrs.version = 0;
 486         ci->i_xattrs.index_version = 0;
 487
 488         ci->i_caps = RB_ROOT;
 489         ci->i_auth_cap = NULL;
 490         ci->i_dirty_caps = 0;
 491         ci->i_flushing_caps = 0;
 492         INIT_LIST_HEAD(&ci->i_dirty_item);
 493         INIT_LIST_HEAD(&ci->i_flushing_item);
 494         ci->i_prealloc_cap_flush = NULL;
 495         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 496         init_waitqueue_head(&ci->i_cap_wq);
 497         ci->i_hold_caps_max = 0;
 498         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 499         INIT_LIST_HEAD(&ci->i_cap_snaps);
 500         ci->i_head_snapc = NULL;
 501         ci->i_snap_caps = 0;
 502
 503         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 504         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 505                 ci->i_nr_by_mode[i] = 0;
 506
 507         mutex_init(&ci->i_truncate_mutex);
 508         ci->i_truncate_seq = 0;
 509         ci->i_truncate_size = 0;
 510         ci->i_truncate_pending = 0;
 511
 512         ci->i_max_size = 0;
 513         ci->i_reported_size = 0;
 514         ci->i_wanted_max_size = 0;
 515         ci->i_requested_max_size = 0;
 516
 517         ci->i_pin_ref = 0;
 518         ci->i_rd_ref = 0;
 519         ci->i_rdcache_ref = 0;
 520         ci->i_wr_ref = 0;
 521         ci->i_wb_ref = 0;
 522         ci->i_fx_ref = 0;
 523         ci->i_wrbuffer_ref = 0;
 524         ci->i_wrbuffer_ref_head = 0;
 525         atomic_set(&ci->i_filelock_ref, 0);
 526         atomic_set(&ci->i_shared_gen, 1);
 527         ci->i_rdcache_gen = 0;
 528         ci->i_rdcache_revoking = 0;
 529
 530         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 531         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 532         spin_lock_init(&ci->i_unsafe_lock);
 533
 534         ci->i_snap_realm = NULL;
 535         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 536         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 537
 538         INIT_WORK(&ci->i_work, ceph_inode_work);
 539         ci->i_work_mask = 0;
 540         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 541
 542         ceph_fscache_inode_init(ci);
 543
 544         return &ci->vfs_inode;
 545 }
 546
 547 void ceph_free_inode(struct inode *inode)
 548 {
 549         struct ceph_inode_info *ci = ceph_inode(inode);
 550
 551         kfree(ci->i_symlink);
 552         kmem_cache_free(ceph_inode_cachep, ci);
 553 }
 554
 555 void ceph_evict_inode(struct inode *inode)
 556 {
 557         struct ceph_inode_info *ci = ceph_inode(inode);
 558         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 559         struct ceph_inode_frag *frag;
 560         struct rb_node *n;
 561
 562         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 563
 564         percpu_counter_dec(&mdsc->metric.total_inodes);
 565
 566         truncate_inode_pages_final(&inode->i_data);
 567         clear_inode(inode);
 568
 569         ceph_fscache_unregister_inode_cookie(ci);
 570
 571         __ceph_remove_caps(ci);
 572
 573         if (__ceph_has_any_quota(ci))
 574                 ceph_adjust_quota_realms_count(inode, false);
 575
 576         /*
 577          * we may still have a snap_realm reference if there are stray
 578          * caps in i_snap_caps.
 579          */
 580         if (ci->i_snap_realm) {
 581                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 582                         dout(" dropping residual ref to snap realm %p\n",
 583                              ci->i_snap_realm);
 584                         ceph_change_snap_realm(inode, NULL);
 585                 } else {
 586                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 587                         ci->i_snap_realm = NULL;
 588                 }
 589         }
 590
 591         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 592                 frag = rb_entry(n, struct ceph_inode_frag, node);
 593                 rb_erase(n, &ci->i_fragtree);
 594                 kfree(frag);
 595         }
 596         ci->i_fragtree_nsplits = 0;
 597
 598         __ceph_destroy_xattrs(ci);
 599         if (ci->i_xattrs.blob)
 600                 ceph_buffer_put(ci->i_xattrs.blob);
 601         if (ci->i_xattrs.prealloc_blob)
 602                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 603
 604         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 605         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 606 }
 607
 608 static inline blkcnt_t calc_inode_blocks(u64 size)
 609 {
 610         return (size + (1<<9) - 1) >> 9;
 611 }
 612
 613 /*
 614  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 615  * careful because either the client or MDS may have more up to date
 616  * info, depending on which capabilities are held, and whether
 617  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 618  * and size are monotonically increasing, except when utimes() or
 619  * truncate() increments the corresponding _seq values.)
 620  */
 621 int ceph_fill_file_size(struct inode *inode, int issued,
 622                         u32 truncate_seq, u64 truncate_size, u64 size)
 623 {
 624         struct ceph_inode_info *ci = ceph_inode(inode);
 625         int queue_trunc = 0;
 626         loff_t isize = i_size_read(inode);
 627
 628         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 629             (truncate_seq == ci->i_truncate_seq && size > isize)) {
 630                 dout("size %lld -> %llu\n", isize, size);
 631                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 632                         pr_err("fill_file_size non-zero size for directory\n");
 633                         size = 0;
 634                 }
 635                 i_size_write(inode, size);
 636                 inode->i_blocks = calc_inode_blocks(size);
 637                 ci->i_reported_size = size;
 638                 if (truncate_seq != ci->i_truncate_seq) {
 639                         dout("truncate_seq %u -> %u\n",
 640                              ci->i_truncate_seq, truncate_seq);
 641                         ci->i_truncate_seq = truncate_seq;
 642
 643                         /* the MDS should have revoked these caps */
 644                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 645                                                CEPH_CAP_FILE_RD |
 646                                                CEPH_CAP_FILE_WR |
 647                                                CEPH_CAP_FILE_LAZYIO));
 648                         /*
 649                          * If we hold relevant caps, or in the case where we're
 650                          * not the only client referencing this file and we
 651                          * don't hold those caps, then we need to check whether
 652                          * the file is either opened or mmaped
 653                          */
 654                         if ((issued & (CEPH_CAP_FILE_CACHE|
 655                                        CEPH_CAP_FILE_BUFFER)) ||
 656                             mapping_mapped(inode->i_mapping) ||
 657                             __ceph_is_file_opened(ci)) {
 658                                 ci->i_truncate_pending++;
 659                                 queue_trunc = 1;
 660                         }
 661                 }
 662         }
 663         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 664             ci->i_truncate_size != truncate_size) {
 665                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 666                      truncate_size);
 667                 ci->i_truncate_size = truncate_size;
 668         }
 669
 670         if (queue_trunc)
 671                 ceph_fscache_invalidate(inode);
 672
 673         return queue_trunc;
 674 }
 675
 676 void ceph_fill_file_time(struct inode *inode, int issued,
 677                          u64 time_warp_seq, struct timespec64 *ctime,
 678                          struct timespec64 *mtime, struct timespec64 *atime)
 679 {
 680         struct ceph_inode_info *ci = ceph_inode(inode);
 681         int warn = 0;
 682
 683         if (issued & (CEPH_CAP_FILE_EXCL|
 684                       CEPH_CAP_FILE_WR|
 685                       CEPH_CAP_FILE_BUFFER|
 686                       CEPH_CAP_AUTH_EXCL|
 687                       CEPH_CAP_XATTR_EXCL)) {
 688                 if (ci->i_version == 0 ||
 689                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 690                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 691                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 692                              ctime->tv_sec, ctime->tv_nsec);
 693                         inode->i_ctime = *ctime;
 694                 }
 695                 if (ci->i_version == 0 ||
 696                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 697                         /* the MDS did a utimes() */
 698                         dout("mtime %lld.%09ld -> %lld.%09ld "
 699                              "tw %d -> %d\n",
 700                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 701                              mtime->tv_sec, mtime->tv_nsec,
 702                              ci->i_time_warp_seq, (int)time_warp_seq);
 703
 704                         inode->i_mtime = *mtime;
 705                         inode->i_atime = *atime;
 706                         ci->i_time_warp_seq = time_warp_seq;
 707                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 708                         /* nobody did utimes(); take the max */
 709                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 710                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 711                                      inode->i_mtime.tv_sec,
 712                                      inode->i_mtime.tv_nsec,
 713                                      mtime->tv_sec, mtime->tv_nsec);
 714                                 inode->i_mtime = *mtime;
 715                         }
 716                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 717                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 718                                      inode->i_atime.tv_sec,
 719                                      inode->i_atime.tv_nsec,
 720                                      atime->tv_sec, atime->tv_nsec);
 721                                 inode->i_atime = *atime;
 722                         }
 723                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 724                         /* we did a utimes(); ignore mds values */
 725                 } else {
 726                         warn = 1;
 727                 }
 728         } else {
 729                 /* we have no write|excl caps; whatever the MDS says is true */
 730                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 731                         inode->i_ctime = *ctime;
 732                         inode->i_mtime = *mtime;
 733                         inode->i_atime = *atime;
 734                         ci->i_time_warp_seq = time_warp_seq;
 735                 } else {
 736                         warn = 1;
 737                 }
 738         }
 739         if (warn) /* time_warp_seq shouldn't go backwards */
 740                 dout("%p mds time_warp_seq %llu < %u\n",
 741                      inode, time_warp_seq, ci->i_time_warp_seq);
 742 }
 743
 744 /*
 745  * Populate an inode based on info from mds.  May be called on new or
 746  * existing inodes.
 747  */
 748 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 749                     struct ceph_mds_reply_info_in *iinfo,
 750                     struct ceph_mds_reply_dirfrag *dirinfo,
 751                     struct ceph_mds_session *session, int cap_fmode,
 752                     struct ceph_cap_reservation *caps_reservation)
 753 {
 754         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 755         struct ceph_mds_reply_inode *info = iinfo->in;
 756         struct ceph_inode_info *ci = ceph_inode(inode);
 757         int issued, new_issued, info_caps;
 758         struct timespec64 mtime, atime, ctime;
 759         struct ceph_buffer *xattr_blob = NULL;
 760         struct ceph_buffer *old_blob = NULL;
 761         struct ceph_string *pool_ns = NULL;
 762         struct ceph_cap *new_cap = NULL;
 763         int err = 0;
 764         bool wake = false;
 765         bool queue_trunc = false;
 766         bool new_version = false;
 767         bool fill_inline = false;
 768         umode_t mode = le32_to_cpu(info->mode);
 769         dev_t rdev = le32_to_cpu(info->rdev);
 770
 771         lockdep_assert_held(&mdsc->snap_rwsem);
 772
 773         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 774              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 775              ci->i_version);
 776
 777         /* Once I_NEW is cleared, we can't change type or dev numbers */
 778         if (inode->i_state & I_NEW) {
 779                 inode->i_mode = mode;
 780         } else {
 781                 if (inode_wrong_type(inode, mode)) {
 782                         pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
 783                                      ceph_vinop(inode), inode->i_mode, mode);
 784                         return -ESTALE;
 785                 }
 786
 787                 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
 788                         pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
 789                                      ceph_vinop(inode), MAJOR(inode->i_rdev),
 790                                      MINOR(inode->i_rdev), MAJOR(rdev),
 791                                      MINOR(rdev));
 792                         return -ESTALE;
 793                 }
 794         }
 795
 796         info_caps = le32_to_cpu(info->cap.caps);
 797
 798         /* prealloc new cap struct */
 799         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 800                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 801                 if (!new_cap)
 802                         return -ENOMEM;
 803         }
 804
 805         /*
 806          * prealloc xattr data, if it looks like we'll need it.  only
 807          * if len > 4 (meaning there are actually xattrs; the first 4
 808          * bytes are the xattr count).
 809          */
 810         if (iinfo->xattr_len > 4) {
 811                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 812                 if (!xattr_blob)
 813                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 814                                iinfo->xattr_len);
 815         }
 816
 817         if (iinfo->pool_ns_len > 0)
 818                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 819                                                      iinfo->pool_ns_len);
 820
 821         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 822                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 823
 824         spin_lock(&ci->i_ceph_lock);
 825
 826         /*
 827          * provided version will be odd if inode value is projected,
 828          * even if stable.  skip the update if we have newer stable
 829          * info (ours>=theirs, e.g. due to racing mds replies), unless
 830          * we are getting projected (unstable) info (in which case the
 831          * version is odd, and we want ours>theirs).
 832          *   us   them
 833          *   2    2     skip
 834          *   3    2     skip
 835          *   3    3     update
 836          */
 837         if (ci->i_version == 0 ||
 838             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 839              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 840                 new_version = true;
 841
 842         /* Update change_attribute */
 843         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 844
 845         __ceph_caps_issued(ci, &issued);
 846         issued |= __ceph_caps_dirty(ci);
 847         new_issued = ~issued & info_caps;
 848
 849         /* directories have fl_stripe_unit set to zero */
 850         if (le32_to_cpu(info->layout.fl_stripe_unit))
 851                 inode->i_blkbits =
 852                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 853         else
 854                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 855
 856         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 857
 858         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 859             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 860                 inode->i_mode = mode;
 861                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 862                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 863                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 864                      from_kuid(&init_user_ns, inode->i_uid),
 865                      from_kgid(&init_user_ns, inode->i_gid));
 866                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 867                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 868         }
 869
 870         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 871             (issued & CEPH_CAP_LINK_EXCL) == 0)
 872                 set_nlink(inode, le32_to_cpu(info->nlink));
 873
 874         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 875                 /* be careful with mtime, atime, size */
 876                 ceph_decode_timespec64(&atime, &info->atime);
 877                 ceph_decode_timespec64(&mtime, &info->mtime);
 878                 ceph_decode_timespec64(&ctime, &info->ctime);
 879                 ceph_fill_file_time(inode, issued,
 880                                 le32_to_cpu(info->time_warp_seq),
 881                                 &ctime, &mtime, &atime);
 882         }
 883
 884         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 885                 ci->i_files = le64_to_cpu(info->files);
 886                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 887         }
 888
 889         if (new_version ||
 890             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 891                 s64 old_pool = ci->i_layout.pool_id;
 892                 struct ceph_string *old_ns;
 893
 894                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 895                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 896                                         lockdep_is_held(&ci->i_ceph_lock));
 897                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 898
 899                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 900                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 901
 902                 pool_ns = old_ns;
 903
 904                 queue_trunc = ceph_fill_file_size(inode, issued,
 905                                         le32_to_cpu(info->truncate_seq),
 906                                         le64_to_cpu(info->truncate_size),
 907                                         le64_to_cpu(info->size));
 908                 /* only update max_size on auth cap */
 909                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 910                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 911                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 912                                         le64_to_cpu(info->max_size));
 913                         ci->i_max_size = le64_to_cpu(info->max_size);
 914                 }
 915         }
 916
 917         /* layout and rstat are not tracked by capability, update them if
 918          * the inode info is from auth mds */
 919         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 920                 if (S_ISDIR(inode->i_mode)) {
 921                         ci->i_dir_layout = iinfo->dir_layout;
 922                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 923                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 924                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 925                         ci->i_dir_pin = iinfo->dir_pin;
 926                         ci->i_rsnaps = iinfo->rsnaps;
 927                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 928                 }
 929         }
 930
 931         /* xattrs */
 932         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 933         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 934             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 935                 if (ci->i_xattrs.blob)
 936                         old_blob = ci->i_xattrs.blob;
 937                 ci->i_xattrs.blob = xattr_blob;
 938                 if (xattr_blob)
 939                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 940                                iinfo->xattr_data, iinfo->xattr_len);
 941                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 942                 ceph_forget_all_cached_acls(inode);
 943                 ceph_security_invalidate_secctx(inode);
 944                 xattr_blob = NULL;
 945         }
 946
 947         /* finally update i_version */
 948         if (le64_to_cpu(info->version) > ci->i_version)
 949                 ci->i_version = le64_to_cpu(info->version);
 950
 951         inode->i_mapping->a_ops = &ceph_aops;
 952
 953         switch (inode->i_mode & S_IFMT) {
 954         case S_IFIFO:
 955         case S_IFBLK:
 956         case S_IFCHR:
 957         case S_IFSOCK:
 958                 inode->i_blkbits = PAGE_SHIFT;
 959                 init_special_inode(inode, inode->i_mode, rdev);
 960                 inode->i_op = &ceph_file_iops;
 961                 break;
 962         case S_IFREG:
 963                 inode->i_op = &ceph_file_iops;
 964                 inode->i_fop = &ceph_file_fops;
 965                 break;
 966         case S_IFLNK:
 967                 inode->i_op = &ceph_symlink_iops;
 968                 if (!ci->i_symlink) {
 969                         u32 symlen = iinfo->symlink_len;
 970                         char *sym;
 971
 972                         spin_unlock(&ci->i_ceph_lock);
 973
 974                         if (symlen != i_size_read(inode)) {
 975                                 pr_err("%s %llx.%llx BAD symlink "
 976                                         "size %lld\n", __func__,
 977                                         ceph_vinop(inode),
 978                                         i_size_read(inode));
 979                                 i_size_write(inode, symlen);
 980                                 inode->i_blocks = calc_inode_blocks(symlen);
 981                         }
 982
 983                         err = -ENOMEM;
 984                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 985                         if (!sym)
 986                                 goto out;
 987
 988                         spin_lock(&ci->i_ceph_lock);
 989                         if (!ci->i_symlink)
 990                                 ci->i_symlink = sym;
 991                         else
 992                                 kfree(sym); /* lost a race */
 993                 }
 994                 inode->i_link = ci->i_symlink;
 995                 break;
 996         case S_IFDIR:
 997                 inode->i_op = &ceph_dir_iops;
 998                 inode->i_fop = &ceph_dir_fops;
 999                 break;
1000         default:
1001                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1002                        ceph_vinop(inode), inode->i_mode);
1003         }
1004
1005         /* were we issued a capability? */
1006         if (info_caps) {
1007                 if (ceph_snap(inode) == CEPH_NOSNAP) {
1008                         ceph_add_cap(inode, session,
1009                                      le64_to_cpu(info->cap.cap_id),
1010                                      info_caps,
1011                                      le32_to_cpu(info->cap.wanted),
1012                                      le32_to_cpu(info->cap.seq),
1013                                      le32_to_cpu(info->cap.mseq),
1014                                      le64_to_cpu(info->cap.realm),
1015                                      info->cap.flags, &new_cap);
1016
1017                         /* set dir completion flag? */
1018                         if (S_ISDIR(inode->i_mode) &&
1019                             ci->i_files == 0 && ci->i_subdirs == 0 &&
1020                             (info_caps & CEPH_CAP_FILE_SHARED) &&
1021                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1022                             !__ceph_dir_is_complete(ci)) {
1023                                 dout(" marking %p complete (empty)\n", inode);
1024                                 i_size_write(inode, 0);
1025                                 __ceph_dir_set_complete(ci,
1026                                         atomic64_read(&ci->i_release_count),
1027                                         atomic64_read(&ci->i_ordered_count));
1028                         }
1029
1030                         wake = true;
1031                 } else {
1032                         dout(" %p got snap_caps %s\n", inode,
1033                              ceph_cap_string(info_caps));
1034                         ci->i_snap_caps |= info_caps;
1035                 }
1036         }
1037
1038         if (iinfo->inline_version > 0 &&
1039             iinfo->inline_version >= ci->i_inline_version) {
1040                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1041                 ci->i_inline_version = iinfo->inline_version;
1042                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1043                     (locked_page || (info_caps & cache_caps)))
1044                         fill_inline = true;
1045         }
1046
1047         if (cap_fmode >= 0) {
1048                 if (!info_caps)
1049                         pr_warn("mds issued no caps on %llx.%llx\n",
1050                                 ceph_vinop(inode));
1051                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1052         }
1053
1054         spin_unlock(&ci->i_ceph_lock);
1055
1056         if (fill_inline)
1057                 ceph_fill_inline_data(inode, locked_page,
1058                                       iinfo->inline_data, iinfo->inline_len);
1059
1060         if (wake)
1061                 wake_up_all(&ci->i_cap_wq);
1062
1063         /* queue truncate if we saw i_size decrease */
1064         if (queue_trunc)
1065                 ceph_queue_vmtruncate(inode);
1066
1067         /* populate frag tree */
1068         if (S_ISDIR(inode->i_mode))
1069                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1070
1071         /* update delegation info? */
1072         if (dirinfo)
1073                 ceph_fill_dirfrag(inode, dirinfo);
1074
1075         err = 0;
1076 out:
1077         if (new_cap)
1078                 ceph_put_cap(mdsc, new_cap);
1079         ceph_buffer_put(old_blob);
1080         ceph_buffer_put(xattr_blob);
1081         ceph_put_string(pool_ns);
1082         return err;
1083 }
1084
1085 /*
1086  * caller should hold session s_mutex and dentry->d_lock.
1087  */
1088 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1089                                   struct ceph_mds_reply_lease *lease,
1090                                   struct ceph_mds_session *session,
1091                                   unsigned long from_time,
1092                                   struct ceph_mds_session **old_lease_session)
1093 {
1094         struct ceph_dentry_info *di = ceph_dentry(dentry);
1095         unsigned mask = le16_to_cpu(lease->mask);
1096         long unsigned duration = le32_to_cpu(lease->duration_ms);
1097         long unsigned ttl = from_time + (duration * HZ) / 1000;
1098         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1099
1100         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1101              dentry, duration, ttl);
1102
1103         /* only track leases on regular dentries */
1104         if (ceph_snap(dir) != CEPH_NOSNAP)
1105                 return;
1106
1107         if (mask & CEPH_LEASE_PRIMARY_LINK)
1108                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1109         else
1110                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1111
1112         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1113         if (!(mask & CEPH_LEASE_VALID)) {
1114                 __ceph_dentry_dir_lease_touch(di);
1115                 return;
1116         }
1117
1118         if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1119             time_before(ttl, di->time))
1120                 return;  /* we already have a newer lease. */
1121
1122         if (di->lease_session && di->lease_session != session) {
1123                 *old_lease_session = di->lease_session;
1124                 di->lease_session = NULL;
1125         }
1126
1127         if (!di->lease_session)
1128                 di->lease_session = ceph_get_mds_session(session);
1129         di->lease_gen = atomic_read(&session->s_cap_gen);
1130         di->lease_seq = le32_to_cpu(lease->seq);
1131         di->lease_renew_after = half_ttl;
1132         di->lease_renew_from = 0;
1133         di->time = ttl;
1134
1135         __ceph_dentry_lease_touch(di);
1136 }
1137
1138 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1139                                         struct ceph_mds_reply_lease *lease,
1140                                         struct ceph_mds_session *session,
1141                                         unsigned long from_time)
1142 {
1143         struct ceph_mds_session *old_lease_session = NULL;
1144         spin_lock(&dentry->d_lock);
1145         __update_dentry_lease(dir, dentry, lease, session, from_time,
1146                               &old_lease_session);
1147         spin_unlock(&dentry->d_lock);
1148         ceph_put_mds_session(old_lease_session);
1149 }
1150
1151 /*
1152  * update dentry lease without having parent inode locked
1153  */
1154 static void update_dentry_lease_careful(struct dentry *dentry,
1155                                         struct ceph_mds_reply_lease *lease,
1156                                         struct ceph_mds_session *session,
1157                                         unsigned long from_time,
1158                                         char *dname, u32 dname_len,
1159                                         struct ceph_vino *pdvino,
1160                                         struct ceph_vino *ptvino)
1161
1162 {
1163         struct inode *dir;
1164         struct ceph_mds_session *old_lease_session = NULL;
1165
1166         spin_lock(&dentry->d_lock);
1167         /* make sure dentry's name matches target */
1168         if (dentry->d_name.len != dname_len ||
1169             memcmp(dentry->d_name.name, dname, dname_len))
1170                 goto out_unlock;
1171
1172         dir = d_inode(dentry->d_parent);
1173         /* make sure parent matches dvino */
1174         if (!ceph_ino_compare(dir, pdvino))
1175                 goto out_unlock;
1176
1177         /* make sure dentry's inode matches target. NULL ptvino means that
1178          * we expect a negative dentry */
1179         if (ptvino) {
1180                 if (d_really_is_negative(dentry))
1181                         goto out_unlock;
1182                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1183                         goto out_unlock;
1184         } else {
1185                 if (d_really_is_positive(dentry))
1186                         goto out_unlock;
1187         }
1188
1189         __update_dentry_lease(dir, dentry, lease, session,
1190                               from_time, &old_lease_session);
1191 out_unlock:
1192         spin_unlock(&dentry->d_lock);
1193         ceph_put_mds_session(old_lease_session);
1194 }
1195
1196 /*
1197  * splice a dentry to an inode.
1198  * caller must hold directory i_mutex for this to be safe.
1199  */
1200 static int splice_dentry(struct dentry **pdn, struct inode *in)
1201 {
1202         struct dentry *dn = *pdn;
1203         struct dentry *realdn;
1204
1205         BUG_ON(d_inode(dn));
1206
1207         if (S_ISDIR(in->i_mode)) {
1208                 /* If inode is directory, d_splice_alias() below will remove
1209                  * 'realdn' from its origin parent. We need to ensure that
1210                  * origin parent's readdir cache will not reference 'realdn'
1211                  */
1212                 realdn = d_find_any_alias(in);
1213                 if (realdn) {
1214                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1215                         spin_lock(&realdn->d_lock);
1216
1217                         realdn->d_op->d_prune(realdn);
1218
1219                         di->time = jiffies;
1220                         di->lease_shared_gen = 0;
1221                         di->offset = 0;
1222
1223                         spin_unlock(&realdn->d_lock);
1224                         dput(realdn);
1225                 }
1226         }
1227
1228         /* dn must be unhashed */
1229         if (!d_unhashed(dn))
1230                 d_drop(dn);
1231         realdn = d_splice_alias(in, dn);
1232         if (IS_ERR(realdn)) {
1233                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1234                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1235                 return PTR_ERR(realdn);
1236         }
1237
1238         if (realdn) {
1239                 dout("dn %p (%d) spliced with %p (%d) "
1240                      "inode %p ino %llx.%llx\n",
1241                      dn, d_count(dn),
1242                      realdn, d_count(realdn),
1243                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1244                 dput(dn);
1245                 *pdn = realdn;
1246         } else {
1247                 BUG_ON(!ceph_dentry(dn));
1248                 dout("dn %p attached to %p ino %llx.%llx\n",
1249                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1250         }
1251         return 0;
1252 }
1253
1254 /*
1255  * Incorporate results into the local cache.  This is either just
1256  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1257  * after a lookup).
1258  *
1259  * A reply may contain
1260  *         a directory inode along with a dentry.
1261  *  and/or a target inode
1262  *
1263  * Called with snap_rwsem (read).
1264  */
1265 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1266 {
1267         struct ceph_mds_session *session = req->r_session;
1268         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1269         struct inode *in = NULL;
1270         struct ceph_vino tvino, dvino;
1271         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1272         int err = 0;
1273
1274         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1275              rinfo->head->is_dentry, rinfo->head->is_target);
1276
1277         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1278                 dout("fill_trace reply is empty!\n");
1279                 if (rinfo->head->result == 0 && req->r_parent)
1280                         ceph_invalidate_dir_request(req);
1281                 return 0;
1282         }
1283
1284         if (rinfo->head->is_dentry) {
1285                 struct inode *dir = req->r_parent;
1286
1287                 if (dir) {
1288                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1289                                               rinfo->dirfrag, session, -1,
1290                                               &req->r_caps_reservation);
1291                         if (err < 0)
1292                                 goto done;
1293                 } else {
1294                         WARN_ON_ONCE(1);
1295                 }
1296
1297                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1298                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1299                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1300                         struct qstr dname;
1301                         struct dentry *dn, *parent;
1302
1303                         BUG_ON(!rinfo->head->is_target);
1304                         BUG_ON(req->r_dentry);
1305
1306                         parent = d_find_any_alias(dir);
1307                         BUG_ON(!parent);
1308
1309                         dname.name = rinfo->dname;
1310                         dname.len = rinfo->dname_len;
1311                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1312                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1313                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1314 retry_lookup:
1315                         dn = d_lookup(parent, &dname);
1316                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1317                              parent, dname.len, dname.name, dn);
1318
1319                         if (!dn) {
1320                                 dn = d_alloc(parent, &dname);
1321                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1322                                      dname.len, dname.name, dn);
1323                                 if (!dn) {
1324                                         dput(parent);
1325                                         err = -ENOMEM;
1326                                         goto done;
1327                                 }
1328                                 err = 0;
1329                         } else if (d_really_is_positive(dn) &&
1330                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1331                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1332                                 dout(" dn %p points to wrong inode %p\n",
1333                                      dn, d_inode(dn));
1334                                 ceph_dir_clear_ordered(dir);
1335                                 d_delete(dn);
1336                                 dput(dn);
1337                                 goto retry_lookup;
1338                         }
1339
1340                         req->r_dentry = dn;
1341                         dput(parent);
1342                 }
1343         }
1344
1345         if (rinfo->head->is_target) {
1346                 /* Should be filled in by handle_reply */
1347                 BUG_ON(!req->r_target_inode);
1348
1349                 in = req->r_target_inode;
1350                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1351                                 NULL, session,
1352                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1353                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1354                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1355                                 &req->r_caps_reservation);
1356                 if (err < 0) {
1357                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1358                                 in, ceph_vinop(in));
1359                         req->r_target_inode = NULL;
1360                         if (in->i_state & I_NEW)
1361                                 discard_new_inode(in);
1362                         else
1363                                 iput(in);
1364                         goto done;
1365                 }
1366                 if (in->i_state & I_NEW)
1367                         unlock_new_inode(in);
1368         }
1369
1370         /*
1371          * ignore null lease/binding on snapdir ENOENT, or else we
1372          * will have trouble splicing in the virtual snapdir later
1373          */
1374         if (rinfo->head->is_dentry &&
1375             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1376             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1377             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1378                                                fsc->mount_options->snapdir_name,
1379                                                req->r_dentry->d_name.len))) {
1380                 /*
1381                  * lookup link rename   : null -> possibly existing inode
1382                  * mknod symlink mkdir  : null -> new inode
1383                  * unlink               : linked -> null
1384                  */
1385                 struct inode *dir = req->r_parent;
1386                 struct dentry *dn = req->r_dentry;
1387                 bool have_dir_cap, have_lease;
1388
1389                 BUG_ON(!dn);
1390                 BUG_ON(!dir);
1391                 BUG_ON(d_inode(dn->d_parent) != dir);
1392
1393                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1394                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1395
1396                 BUG_ON(ceph_ino(dir) != dvino.ino);
1397                 BUG_ON(ceph_snap(dir) != dvino.snap);
1398
1399                 /* do we have a lease on the whole dir? */
1400                 have_dir_cap =
1401                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1402                          CEPH_CAP_FILE_SHARED);
1403
1404                 /* do we have a dn lease? */
1405                 have_lease = have_dir_cap ||
1406                         le32_to_cpu(rinfo->dlease->duration_ms);
1407                 if (!have_lease)
1408                         dout("fill_trace  no dentry lease or dir cap\n");
1409
1410                 /* rename? */
1411                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1412                         struct inode *olddir = req->r_old_dentry_dir;
1413                         BUG_ON(!olddir);
1414
1415                         dout(" src %p '%pd' dst %p '%pd'\n",
1416                              req->r_old_dentry,
1417                              req->r_old_dentry,
1418                              dn, dn);
1419                         dout("fill_trace doing d_move %p -> %p\n",
1420                              req->r_old_dentry, dn);
1421
1422                         /* d_move screws up sibling dentries' offsets */
1423                         ceph_dir_clear_ordered(dir);
1424                         ceph_dir_clear_ordered(olddir);
1425
1426                         d_move(req->r_old_dentry, dn);
1427                         dout(" src %p '%pd' dst %p '%pd'\n",
1428                              req->r_old_dentry,
1429                              req->r_old_dentry,
1430                              dn, dn);
1431
1432                         /* ensure target dentry is invalidated, despite
1433                            rehashing bug in vfs_rename_dir */
1434                         ceph_invalidate_dentry_lease(dn);
1435
1436                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1437                              ceph_dentry(req->r_old_dentry)->offset);
1438
1439                         /* swap r_dentry and r_old_dentry in case that
1440                          * splice_dentry() gets called later. This is safe
1441                          * because no other place will use them */
1442                         req->r_dentry = req->r_old_dentry;
1443                         req->r_old_dentry = dn;
1444                         dn = req->r_dentry;
1445                 }
1446
1447                 /* null dentry? */
1448                 if (!rinfo->head->is_target) {
1449                         dout("fill_trace null dentry\n");
1450                         if (d_really_is_positive(dn)) {
1451                                 dout("d_delete %p\n", dn);
1452                                 ceph_dir_clear_ordered(dir);
1453                                 d_delete(dn);
1454                         } else if (have_lease) {
1455                                 if (d_unhashed(dn))
1456                                         d_add(dn, NULL);
1457                                 update_dentry_lease(dir, dn,
1458                                                     rinfo->dlease, session,
1459                                                     req->r_request_started);
1460                         }
1461                         goto done;
1462                 }
1463
1464                 /* attach proper inode */
1465                 if (d_really_is_negative(dn)) {
1466                         ceph_dir_clear_ordered(dir);
1467                         ihold(in);
1468                         err = splice_dentry(&req->r_dentry, in);
1469                         if (err < 0)
1470                                 goto done;
1471                         dn = req->r_dentry;  /* may have spliced */
1472                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1473                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1474                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1475                              ceph_vinop(in));
1476                         d_invalidate(dn);
1477                         have_lease = false;
1478                 }
1479
1480                 if (have_lease) {
1481                         update_dentry_lease(dir, dn,
1482                                             rinfo->dlease, session,
1483                                             req->r_request_started);
1484                 }
1485                 dout(" final dn %p\n", dn);
1486         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1487                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1488                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1489                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1490                 struct inode *dir = req->r_parent;
1491
1492                 /* fill out a snapdir LOOKUPSNAP dentry */
1493                 BUG_ON(!dir);
1494                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1495                 BUG_ON(!req->r_dentry);
1496                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1497                 ceph_dir_clear_ordered(dir);
1498                 ihold(in);
1499                 err = splice_dentry(&req->r_dentry, in);
1500                 if (err < 0)
1501                         goto done;
1502         } else if (rinfo->head->is_dentry && req->r_dentry) {
1503                 /* parent inode is not locked, be carefull */
1504                 struct ceph_vino *ptvino = NULL;
1505                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1506                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1507                 if (rinfo->head->is_target) {
1508                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1509                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1510                         ptvino = &tvino;
1511                 }
1512                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1513                                             session, req->r_request_started,
1514                                             rinfo->dname, rinfo->dname_len,
1515                                             &dvino, ptvino);
1516         }
1517 done:
1518         dout("fill_trace done err=%d\n", err);
1519         return err;
1520 }
1521
1522 /*
1523  * Prepopulate our cache with readdir results, leases, etc.
1524  */
1525 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1526                                            struct ceph_mds_session *session)
1527 {
1528         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1529         int i, err = 0;
1530
1531         for (i = 0; i < rinfo->dir_nr; i++) {
1532                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1533                 struct ceph_vino vino;
1534                 struct inode *in;
1535                 int rc;
1536
1537                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1538                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1539
1540                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1541                 if (IS_ERR(in)) {
1542                         err = PTR_ERR(in);
1543                         dout("new_inode badness got %d\n", err);
1544                         continue;
1545                 }
1546                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1547                                      -1, &req->r_caps_reservation);
1548                 if (rc < 0) {
1549                         pr_err("ceph_fill_inode badness on %p got %d\n",
1550                                in, rc);
1551                         err = rc;
1552                         if (in->i_state & I_NEW) {
1553                                 ihold(in);
1554                                 discard_new_inode(in);
1555                         }
1556                 } else if (in->i_state & I_NEW) {
1557                         unlock_new_inode(in);
1558                 }
1559
1560                 iput(in);
1561         }
1562
1563         return err;
1564 }
1565
1566 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1567 {
1568         if (ctl->page) {
1569                 kunmap(ctl->page);
1570                 put_page(ctl->page);
1571                 ctl->page = NULL;
1572         }
1573 }
1574
1575 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1576                               struct ceph_readdir_cache_control *ctl,
1577                               struct ceph_mds_request *req)
1578 {
1579         struct ceph_inode_info *ci = ceph_inode(dir);
1580         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1581         unsigned idx = ctl->index % nsize;
1582         pgoff_t pgoff = ctl->index / nsize;
1583
1584         if (!ctl->page || pgoff != page_index(ctl->page)) {
1585                 ceph_readdir_cache_release(ctl);
1586                 if (idx == 0)
1587                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1588                 else
1589                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1590                 if (!ctl->page) {
1591                         ctl->index = -1;
1592                         return idx == 0 ? -ENOMEM : 0;
1593                 }
1594                 /* reading/filling the cache are serialized by
1595                  * i_mutex, no need to use page lock */
1596                 unlock_page(ctl->page);
1597                 ctl->dentries = kmap(ctl->page);
1598                 if (idx == 0)
1599                         memset(ctl->dentries, 0, PAGE_SIZE);
1600         }
1601
1602         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1603             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1604                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1605                 ctl->dentries[idx] = dn;
1606                 ctl->index++;
1607         } else {
1608                 dout("disable readdir cache\n");
1609                 ctl->index = -1;
1610         }
1611         return 0;
1612 }
1613
1614 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1615                              struct ceph_mds_session *session)
1616 {
1617         struct dentry *parent = req->r_dentry;
1618         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1619         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1620         struct qstr dname;
1621         struct dentry *dn;
1622         struct inode *in;
1623         int err = 0, skipped = 0, ret, i;
1624         u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1625         u32 last_hash = 0;
1626         u32 fpos_offset;
1627         struct ceph_readdir_cache_control cache_ctl = {};
1628
1629         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1630                 return readdir_prepopulate_inodes_only(req, session);
1631
1632         if (rinfo->hash_order) {
1633                 if (req->r_path2) {
1634                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1635                                                   req->r_path2,
1636                                                   strlen(req->r_path2));
1637                         last_hash = ceph_frag_value(last_hash);
1638                 } else if (rinfo->offset_hash) {
1639                         /* mds understands offset_hash */
1640                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1641                         last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1642                 }
1643         }
1644
1645         if (rinfo->dir_dir &&
1646             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1647                 dout("readdir_prepopulate got new frag %x -> %x\n",
1648                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1649                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1650                 if (!rinfo->hash_order)
1651                         req->r_readdir_offset = 2;
1652         }
1653
1654         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1655                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1656                      rinfo->dir_nr, parent);
1657         } else {
1658                 dout("readdir_prepopulate %d items under dn %p\n",
1659                      rinfo->dir_nr, parent);
1660                 if (rinfo->dir_dir)
1661                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1662
1663                 if (ceph_frag_is_leftmost(frag) &&
1664                     req->r_readdir_offset == 2 &&
1665                     !(rinfo->hash_order && last_hash)) {
1666                         /* note dir version at start of readdir so we can
1667                          * tell if any dentries get dropped */
1668                         req->r_dir_release_cnt =
1669                                 atomic64_read(&ci->i_release_count);
1670                         req->r_dir_ordered_cnt =
1671                                 atomic64_read(&ci->i_ordered_count);
1672                         req->r_readdir_cache_idx = 0;
1673                 }
1674         }
1675
1676         cache_ctl.index = req->r_readdir_cache_idx;
1677         fpos_offset = req->r_readdir_offset;
1678
1679         /* FIXME: release caps/leases if error occurs */
1680         for (i = 0; i < rinfo->dir_nr; i++) {
1681                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1682                 struct ceph_vino tvino;
1683
1684                 dname.name = rde->name;
1685                 dname.len = rde->name_len;
1686                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1687
1688                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1689                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1690
1691                 if (rinfo->hash_order) {
1692                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1693                                                  rde->name, rde->name_len);
1694                         hash = ceph_frag_value(hash);
1695                         if (hash != last_hash)
1696                                 fpos_offset = 2;
1697                         last_hash = hash;
1698                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1699                 } else {
1700                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1701                 }
1702
1703 retry_lookup:
1704                 dn = d_lookup(parent, &dname);
1705                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1706                      parent, dname.len, dname.name, dn);
1707
1708                 if (!dn) {
1709                         dn = d_alloc(parent, &dname);
1710                         dout("d_alloc %p '%.*s' = %p\n", parent,
1711                              dname.len, dname.name, dn);
1712                         if (!dn) {
1713                                 dout("d_alloc badness\n");
1714                                 err = -ENOMEM;
1715                                 goto out;
1716                         }
1717                 } else if (d_really_is_positive(dn) &&
1718                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1719                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1720                         struct ceph_dentry_info *di = ceph_dentry(dn);
1721                         dout(" dn %p points to wrong inode %p\n",
1722                              dn, d_inode(dn));
1723
1724                         spin_lock(&dn->d_lock);
1725                         if (di->offset > 0 &&
1726                             di->lease_shared_gen ==
1727                             atomic_read(&ci->i_shared_gen)) {
1728                                 __ceph_dir_clear_ordered(ci);
1729                                 di->offset = 0;
1730                         }
1731                         spin_unlock(&dn->d_lock);
1732
1733                         d_delete(dn);
1734                         dput(dn);
1735                         goto retry_lookup;
1736                 }
1737
1738                 /* inode */
1739                 if (d_really_is_positive(dn)) {
1740                         in = d_inode(dn);
1741                 } else {
1742                         in = ceph_get_inode(parent->d_sb, tvino);
1743                         if (IS_ERR(in)) {
1744                                 dout("new_inode badness\n");
1745                                 d_drop(dn);
1746                                 dput(dn);
1747                                 err = PTR_ERR(in);
1748                                 goto out;
1749                         }
1750                 }
1751
1752                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1753                                       -1, &req->r_caps_reservation);
1754                 if (ret < 0) {
1755                         pr_err("ceph_fill_inode badness on %p\n", in);
1756                         if (d_really_is_negative(dn)) {
1757                                 if (in->i_state & I_NEW) {
1758                                         ihold(in);
1759                                         discard_new_inode(in);
1760                                 }
1761                                 iput(in);
1762                         }
1763                         d_drop(dn);
1764                         err = ret;
1765                         goto next_item;
1766                 }
1767                 if (in->i_state & I_NEW)
1768                         unlock_new_inode(in);
1769
1770                 if (d_really_is_negative(dn)) {
1771                         if (ceph_security_xattr_deadlock(in)) {
1772                                 dout(" skip splicing dn %p to inode %p"
1773                                      " (security xattr deadlock)\n", dn, in);
1774                                 iput(in);
1775                                 skipped++;
1776                                 goto next_item;
1777                         }
1778
1779                         err = splice_dentry(&dn, in);
1780                         if (err < 0)
1781                                 goto next_item;
1782                 }
1783
1784                 ceph_dentry(dn)->offset = rde->offset;
1785
1786                 update_dentry_lease(d_inode(parent), dn,
1787                                     rde->lease, req->r_session,
1788                                     req->r_request_started);
1789
1790                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1791                         ret = fill_readdir_cache(d_inode(parent), dn,
1792                                                  &cache_ctl, req);
1793                         if (ret < 0)
1794                                 err = ret;
1795                 }
1796 next_item:
1797                 dput(dn);
1798         }
1799 out:
1800         if (err == 0 && skipped == 0) {
1801                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1802                 req->r_readdir_cache_idx = cache_ctl.index;
1803         }
1804         ceph_readdir_cache_release(&cache_ctl);
1805         dout("readdir_prepopulate done\n");
1806         return err;
1807 }
1808
1809 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1810 {
1811         struct ceph_inode_info *ci = ceph_inode(inode);
1812         bool ret;
1813
1814         spin_lock(&ci->i_ceph_lock);
1815         dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1816         i_size_write(inode, size);
1817         inode->i_blocks = calc_inode_blocks(size);
1818
1819         ret = __ceph_should_report_size(ci);
1820
1821         spin_unlock(&ci->i_ceph_lock);
1822         return ret;
1823 }
1824
1825 void ceph_queue_inode_work(struct inode *inode, int work_bit)
1826 {
1827         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1828         struct ceph_inode_info *ci = ceph_inode(inode);
1829         set_bit(work_bit, &ci->i_work_mask);
1830
1831         ihold(inode);
1832         if (queue_work(fsc->inode_wq, &ci->i_work)) {
1833                 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1834         } else {
1835                 dout("queue_inode_work %p already queued, mask=%lx\n",
1836                      inode, ci->i_work_mask);
1837                 iput(inode);
1838         }
1839 }
1840
1841 static void ceph_do_invalidate_pages(struct inode *inode)
1842 {
1843         struct ceph_inode_info *ci = ceph_inode(inode);
1844         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1845         u32 orig_gen;
1846         int check = 0;
1847
1848         mutex_lock(&ci->i_truncate_mutex);
1849
1850         if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1851                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1852                                     inode, ceph_ino(inode));
1853                 mapping_set_error(inode->i_mapping, -EIO);
1854                 truncate_pagecache(inode, 0);
1855                 mutex_unlock(&ci->i_truncate_mutex);
1856                 goto out;
1857         }
1858
1859         spin_lock(&ci->i_ceph_lock);
1860         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1861              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1862         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1863                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1864                         check = 1;
1865                 spin_unlock(&ci->i_ceph_lock);
1866                 mutex_unlock(&ci->i_truncate_mutex);
1867                 goto out;
1868         }
1869         orig_gen = ci->i_rdcache_gen;
1870         spin_unlock(&ci->i_ceph_lock);
1871
1872         ceph_fscache_invalidate(inode);
1873         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1874                 pr_err("invalidate_pages %p fails\n", inode);
1875         }
1876
1877         spin_lock(&ci->i_ceph_lock);
1878         if (orig_gen == ci->i_rdcache_gen &&
1879             orig_gen == ci->i_rdcache_revoking) {
1880                 dout("invalidate_pages %p gen %d successful\n", inode,
1881                      ci->i_rdcache_gen);
1882                 ci->i_rdcache_revoking--;
1883                 check = 1;
1884         } else {
1885                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1886                      inode, orig_gen, ci->i_rdcache_gen,
1887                      ci->i_rdcache_revoking);
1888                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1889                         check = 1;
1890         }
1891         spin_unlock(&ci->i_ceph_lock);
1892         mutex_unlock(&ci->i_truncate_mutex);
1893 out:
1894         if (check)
1895                 ceph_check_caps(ci, 0, NULL);
1896 }
1897
1898 /*
1899  * Make sure any pending truncation is applied before doing anything
1900  * that may depend on it.
1901  */
1902 void __ceph_do_pending_vmtruncate(struct inode *inode)
1903 {
1904         struct ceph_inode_info *ci = ceph_inode(inode);
1905         u64 to;
1906         int wrbuffer_refs, finish = 0;
1907
1908         mutex_lock(&ci->i_truncate_mutex);
1909 retry:
1910         spin_lock(&ci->i_ceph_lock);
1911         if (ci->i_truncate_pending == 0) {
1912                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1913                 spin_unlock(&ci->i_ceph_lock);
1914                 mutex_unlock(&ci->i_truncate_mutex);
1915                 return;
1916         }
1917
1918         /*
1919          * make sure any dirty snapped pages are flushed before we
1920          * possibly truncate them.. so write AND block!
1921          */
1922         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1923                 spin_unlock(&ci->i_ceph_lock);
1924                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1925                      inode);
1926                 filemap_write_and_wait_range(&inode->i_data, 0,
1927                                              inode->i_sb->s_maxbytes);
1928                 goto retry;
1929         }
1930
1931         /* there should be no reader or writer */
1932         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1933
1934         to = ci->i_truncate_size;
1935         wrbuffer_refs = ci->i_wrbuffer_ref;
1936         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1937              ci->i_truncate_pending, to);
1938         spin_unlock(&ci->i_ceph_lock);
1939
1940         truncate_pagecache(inode, to);
1941
1942         spin_lock(&ci->i_ceph_lock);
1943         if (to == ci->i_truncate_size) {
1944                 ci->i_truncate_pending = 0;
1945                 finish = 1;
1946         }
1947         spin_unlock(&ci->i_ceph_lock);
1948         if (!finish)
1949                 goto retry;
1950
1951         mutex_unlock(&ci->i_truncate_mutex);
1952
1953         if (wrbuffer_refs == 0)
1954                 ceph_check_caps(ci, 0, NULL);
1955
1956         wake_up_all(&ci->i_cap_wq);
1957 }
1958
1959 static void ceph_inode_work(struct work_struct *work)
1960 {
1961         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1962                                                  i_work);
1963         struct inode *inode = &ci->vfs_inode;
1964
1965         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1966                 dout("writeback %p\n", inode);
1967                 filemap_fdatawrite(&inode->i_data);
1968         }
1969         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1970                 ceph_do_invalidate_pages(inode);
1971
1972         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1973                 __ceph_do_pending_vmtruncate(inode);
1974
1975         if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
1976                 ceph_check_caps(ci, 0, NULL);
1977
1978         if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
1979                 ceph_flush_snaps(ci, NULL);
1980
1981         iput(inode);
1982 }
1983
1984 /*
1985  * symlinks
1986  */
1987 static const struct inode_operations ceph_symlink_iops = {
1988         .get_link = simple_get_link,
1989         .setattr = ceph_setattr,
1990         .getattr = ceph_getattr,
1991         .listxattr = ceph_listxattr,
1992 };
1993
1994 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1995 {
1996         struct ceph_inode_info *ci = ceph_inode(inode);
1997         unsigned int ia_valid = attr->ia_valid;
1998         struct ceph_mds_request *req;
1999         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2000         struct ceph_cap_flush *prealloc_cf;
2001         int issued;
2002         int release = 0, dirtied = 0;
2003         int mask = 0;
2004         int err = 0;
2005         int inode_dirty_flags = 0;
2006         bool lock_snap_rwsem = false;
2007
2008         prealloc_cf = ceph_alloc_cap_flush();
2009         if (!prealloc_cf)
2010                 return -ENOMEM;
2011
2012         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2013                                        USE_AUTH_MDS);
2014         if (IS_ERR(req)) {
2015                 ceph_free_cap_flush(prealloc_cf);
2016                 return PTR_ERR(req);
2017         }
2018
2019         spin_lock(&ci->i_ceph_lock);
2020         issued = __ceph_caps_issued(ci, NULL);
2021
2022         if (!ci->i_head_snapc &&
2023             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2024                 lock_snap_rwsem = true;
2025                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2026                         spin_unlock(&ci->i_ceph_lock);
2027                         down_read(&mdsc->snap_rwsem);
2028                         spin_lock(&ci->i_ceph_lock);
2029                         issued = __ceph_caps_issued(ci, NULL);
2030                 }
2031         }
2032
2033         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2034
2035         if (ia_valid & ATTR_UID) {
2036                 dout("setattr %p uid %d -> %d\n", inode,
2037                      from_kuid(&init_user_ns, inode->i_uid),
2038                      from_kuid(&init_user_ns, attr->ia_uid));
2039                 if (issued & CEPH_CAP_AUTH_EXCL) {
2040                         inode->i_uid = attr->ia_uid;
2041                         dirtied |= CEPH_CAP_AUTH_EXCL;
2042                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2043                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2044                         req->r_args.setattr.uid = cpu_to_le32(
2045                                 from_kuid(&init_user_ns, attr->ia_uid));
2046                         mask |= CEPH_SETATTR_UID;
2047                         release |= CEPH_CAP_AUTH_SHARED;
2048                 }
2049         }
2050         if (ia_valid & ATTR_GID) {
2051                 dout("setattr %p gid %d -> %d\n", inode,
2052                      from_kgid(&init_user_ns, inode->i_gid),
2053                      from_kgid(&init_user_ns, attr->ia_gid));
2054                 if (issued & CEPH_CAP_AUTH_EXCL) {
2055                         inode->i_gid = attr->ia_gid;
2056                         dirtied |= CEPH_CAP_AUTH_EXCL;
2057                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2058                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2059                         req->r_args.setattr.gid = cpu_to_le32(
2060                                 from_kgid(&init_user_ns, attr->ia_gid));
2061                         mask |= CEPH_SETATTR_GID;
2062                         release |= CEPH_CAP_AUTH_SHARED;
2063                 }
2064         }
2065         if (ia_valid & ATTR_MODE) {
2066                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2067                      attr->ia_mode);
2068                 if (issued & CEPH_CAP_AUTH_EXCL) {
2069                         inode->i_mode = attr->ia_mode;
2070                         dirtied |= CEPH_CAP_AUTH_EXCL;
2071                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2072                            attr->ia_mode != inode->i_mode) {
2073                         inode->i_mode = attr->ia_mode;
2074                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2075                         mask |= CEPH_SETATTR_MODE;
2076                         release |= CEPH_CAP_AUTH_SHARED;
2077                 }
2078         }
2079
2080         if (ia_valid & ATTR_ATIME) {
2081                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2082                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2083                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2084                 if (issued & CEPH_CAP_FILE_EXCL) {
2085                         ci->i_time_warp_seq++;
2086                         inode->i_atime = attr->ia_atime;
2087                         dirtied |= CEPH_CAP_FILE_EXCL;
2088                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2089                            timespec64_compare(&inode->i_atime,
2090                                             &attr->ia_atime) < 0) {
2091                         inode->i_atime = attr->ia_atime;
2092                         dirtied |= CEPH_CAP_FILE_WR;
2093                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2094                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2095                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2096                                                &attr->ia_atime);
2097                         mask |= CEPH_SETATTR_ATIME;
2098                         release |= CEPH_CAP_FILE_SHARED |
2099                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2100                 }
2101         }
2102         if (ia_valid & ATTR_SIZE) {
2103                 loff_t isize = i_size_read(inode);
2104
2105                 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2106                 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size > isize) {
2107                         i_size_write(inode, attr->ia_size);
2108                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2109                         ci->i_reported_size = attr->ia_size;
2110                         dirtied |= CEPH_CAP_FILE_EXCL;
2111                         ia_valid |= ATTR_MTIME;
2112                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2113                            attr->ia_size != isize) {
2114                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2115                         req->r_args.setattr.old_size = cpu_to_le64(isize);
2116                         mask |= CEPH_SETATTR_SIZE;
2117                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2118                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2119                 }
2120         }
2121         if (ia_valid & ATTR_MTIME) {
2122                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2123                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2124                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2125                 if (issued & CEPH_CAP_FILE_EXCL) {
2126                         ci->i_time_warp_seq++;
2127                         inode->i_mtime = attr->ia_mtime;
2128                         dirtied |= CEPH_CAP_FILE_EXCL;
2129                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2130                            timespec64_compare(&inode->i_mtime,
2131                                             &attr->ia_mtime) < 0) {
2132                         inode->i_mtime = attr->ia_mtime;
2133                         dirtied |= CEPH_CAP_FILE_WR;
2134                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2135                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2136                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2137                                                &attr->ia_mtime);
2138                         mask |= CEPH_SETATTR_MTIME;
2139                         release |= CEPH_CAP_FILE_SHARED |
2140                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2141                 }
2142         }
2143
2144         /* these do nothing */
2145         if (ia_valid & ATTR_CTIME) {
2146                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2147                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2148                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2149                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2150                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2151                      only ? "ctime only" : "ignored");
2152                 if (only) {
2153                         /*
2154                          * if kernel wants to dirty ctime but nothing else,
2155                          * we need to choose a cap to dirty under, or do
2156                          * a almost-no-op setattr
2157                          */
2158                         if (issued & CEPH_CAP_AUTH_EXCL)
2159                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2160                         else if (issued & CEPH_CAP_FILE_EXCL)
2161                                 dirtied |= CEPH_CAP_FILE_EXCL;
2162                         else if (issued & CEPH_CAP_XATTR_EXCL)
2163                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2164                         else
2165                                 mask |= CEPH_SETATTR_CTIME;
2166                 }
2167         }
2168         if (ia_valid & ATTR_FILE)
2169                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2170
2171         if (dirtied) {
2172                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2173                                                            &prealloc_cf);
2174                 inode->i_ctime = attr->ia_ctime;
2175         }
2176
2177         release &= issued;
2178         spin_unlock(&ci->i_ceph_lock);
2179         if (lock_snap_rwsem)
2180                 up_read(&mdsc->snap_rwsem);
2181
2182         if (inode_dirty_flags)
2183                 __mark_inode_dirty(inode, inode_dirty_flags);
2184
2185
2186         if (mask) {
2187                 req->r_inode = inode;
2188                 ihold(inode);
2189                 req->r_inode_drop = release;
2190                 req->r_args.setattr.mask = cpu_to_le32(mask);
2191                 req->r_num_caps = 1;
2192                 req->r_stamp = attr->ia_ctime;
2193                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2194         }
2195         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2196              ceph_cap_string(dirtied), mask);
2197
2198         ceph_mdsc_put_request(req);
2199         ceph_free_cap_flush(prealloc_cf);
2200
2201         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2202                 __ceph_do_pending_vmtruncate(inode);
2203
2204         return err;
2205 }
2206
2207 /*
2208  * setattr
2209  */
2210 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2211                  struct iattr *attr)
2212 {
2213         struct inode *inode = d_inode(dentry);
2214         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2215         int err;
2216
2217         if (ceph_snap(inode) != CEPH_NOSNAP)
2218                 return -EROFS;
2219
2220         err = setattr_prepare(&init_user_ns, dentry, attr);
2221         if (err != 0)
2222                 return err;
2223
2224         if ((attr->ia_valid & ATTR_SIZE) &&
2225             attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2226                 return -EFBIG;
2227
2228         if ((attr->ia_valid & ATTR_SIZE) &&
2229             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2230                 return -EDQUOT;
2231
2232         err = __ceph_setattr(inode, attr);
2233
2234         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2235                 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode);
2236
2237         return err;
2238 }
2239
2240 /*
2241  * Verify that we have a lease on the given mask.  If not,
2242  * do a getattr against an mds.
2243  */
2244 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2245                       int mask, bool force)
2246 {
2247         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2248         struct ceph_mds_client *mdsc = fsc->mdsc;
2249         struct ceph_mds_request *req;
2250         int mode;
2251         int err;
2252
2253         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2254                 dout("do_getattr inode %p SNAPDIR\n", inode);
2255                 return 0;
2256         }
2257
2258         dout("do_getattr inode %p mask %s mode 0%o\n",
2259              inode, ceph_cap_string(mask), inode->i_mode);
2260         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2261                         return 0;
2262
2263         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2264         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2265         if (IS_ERR(req))
2266                 return PTR_ERR(req);
2267         req->r_inode = inode;
2268         ihold(inode);
2269         req->r_num_caps = 1;
2270         req->r_args.getattr.mask = cpu_to_le32(mask);
2271         req->r_locked_page = locked_page;
2272         err = ceph_mdsc_do_request(mdsc, NULL, req);
2273         if (locked_page && err == 0) {
2274                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2275                 if (inline_version == 0) {
2276                         /* the reply is supposed to contain inline data */
2277                         err = -EINVAL;
2278                 } else if (inline_version == CEPH_INLINE_NONE) {
2279                         err = -ENODATA;
2280                 } else {
2281                         err = req->r_reply_info.targeti.inline_len;
2282                 }
2283         }
2284         ceph_mdsc_put_request(req);
2285         dout("do_getattr result=%d\n", err);
2286         return err;
2287 }
2288
2289
2290 /*
2291  * Check inode permissions.  We verify we have a valid value for
2292  * the AUTH cap, then call the generic handler.
2293  */
2294 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2295                     int mask)
2296 {
2297         int err;
2298
2299         if (mask & MAY_NOT_BLOCK)
2300                 return -ECHILD;
2301
2302         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2303
2304         if (!err)
2305                 err = generic_permission(&init_user_ns, inode, mask);
2306         return err;
2307 }
2308
2309 /* Craft a mask of needed caps given a set of requested statx attrs. */
2310 static int statx_to_caps(u32 want, umode_t mode)
2311 {
2312         int mask = 0;
2313
2314         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2315                 mask |= CEPH_CAP_AUTH_SHARED;
2316
2317         if (want & (STATX_NLINK|STATX_CTIME)) {
2318                 /*
2319                  * The link count for directories depends on inode->i_subdirs,
2320                  * and that is only updated when Fs caps are held.
2321                  */
2322                 if (S_ISDIR(mode))
2323                         mask |= CEPH_CAP_FILE_SHARED;
2324                 else
2325                         mask |= CEPH_CAP_LINK_SHARED;
2326         }
2327
2328         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2329                     STATX_BLOCKS))
2330                 mask |= CEPH_CAP_FILE_SHARED;
2331
2332         if (want & (STATX_CTIME))
2333                 mask |= CEPH_CAP_XATTR_SHARED;
2334
2335         return mask;
2336 }
2337
2338 /*
2339  * Get all the attributes. If we have sufficient caps for the requested attrs,
2340  * then we can avoid talking to the MDS at all.
2341  */
2342 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2343                  struct kstat *stat, u32 request_mask, unsigned int flags)
2344 {
2345         struct inode *inode = d_inode(path->dentry);
2346         struct ceph_inode_info *ci = ceph_inode(inode);
2347         u32 valid_mask = STATX_BASIC_STATS;
2348         int err = 0;
2349
2350         /* Skip the getattr altogether if we're asked not to sync */
2351         if (!(flags & AT_STATX_DONT_SYNC)) {
2352                 err = ceph_do_getattr(inode,
2353                                 statx_to_caps(request_mask, inode->i_mode),
2354                                 flags & AT_STATX_FORCE_SYNC);
2355                 if (err)
2356                         return err;
2357         }
2358
2359         generic_fillattr(&init_user_ns, inode, stat);
2360         stat->ino = ceph_present_inode(inode);
2361
2362         /*
2363          * btime on newly-allocated inodes is 0, so if this is still set to
2364          * that, then assume that it's not valid.
2365          */
2366         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2367                 stat->btime = ci->i_btime;
2368                 valid_mask |= STATX_BTIME;
2369         }
2370
2371         if (ceph_snap(inode) == CEPH_NOSNAP)
2372                 stat->dev = inode->i_sb->s_dev;
2373         else
2374                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2375
2376         if (S_ISDIR(inode->i_mode)) {
2377                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2378                                         RBYTES))
2379                         stat->size = ci->i_rbytes;
2380                 else
2381                         stat->size = ci->i_files + ci->i_subdirs;
2382                 stat->blocks = 0;
2383                 stat->blksize = 65536;
2384                 /*
2385                  * Some applications rely on the number of st_nlink
2386                  * value on directories to be either 0 (if unlinked)
2387                  * or 2 + number of subdirectories.
2388                  */
2389                 if (stat->nlink == 1)
2390                         /* '.' + '..' + subdirs */
2391                         stat->nlink = 1 + 1 + ci->i_subdirs;
2392         }
2393
2394         stat->result_mask = request_mask & valid_mask;
2395         return err;
2396 }