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         ci->i_meta_err = 0;
 545
 546         return &ci->vfs_inode;
 547 }
 548
 549 void ceph_free_inode(struct inode *inode)
 550 {
 551         struct ceph_inode_info *ci = ceph_inode(inode);
 552
 553         kfree(ci->i_symlink);
 554         kmem_cache_free(ceph_inode_cachep, ci);
 555 }
 556
 557 void ceph_evict_inode(struct inode *inode)
 558 {
 559         struct ceph_inode_info *ci = ceph_inode(inode);
 560         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 561         struct ceph_inode_frag *frag;
 562         struct rb_node *n;
 563
 564         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 565
 566         percpu_counter_dec(&mdsc->metric.total_inodes);
 567
 568         truncate_inode_pages_final(&inode->i_data);
 569         clear_inode(inode);
 570
 571         ceph_fscache_unregister_inode_cookie(ci);
 572
 573         __ceph_remove_caps(ci);
 574
 575         if (__ceph_has_any_quota(ci))
 576                 ceph_adjust_quota_realms_count(inode, false);
 577
 578         /*
 579          * we may still have a snap_realm reference if there are stray
 580          * caps in i_snap_caps.
 581          */
 582         if (ci->i_snap_realm) {
 583                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 584                         dout(" dropping residual ref to snap realm %p\n",
 585                              ci->i_snap_realm);
 586                         ceph_change_snap_realm(inode, NULL);
 587                 } else {
 588                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 589                         ci->i_snap_realm = NULL;
 590                 }
 591         }
 592
 593         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 594                 frag = rb_entry(n, struct ceph_inode_frag, node);
 595                 rb_erase(n, &ci->i_fragtree);
 596                 kfree(frag);
 597         }
 598         ci->i_fragtree_nsplits = 0;
 599
 600         __ceph_destroy_xattrs(ci);
 601         if (ci->i_xattrs.blob)
 602                 ceph_buffer_put(ci->i_xattrs.blob);
 603         if (ci->i_xattrs.prealloc_blob)
 604                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 605
 606         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 607         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 608 }
 609
 610 static inline blkcnt_t calc_inode_blocks(u64 size)
 611 {
 612         return (size + (1<<9) - 1) >> 9;
 613 }
 614
 615 /*
 616  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 617  * careful because either the client or MDS may have more up to date
 618  * info, depending on which capabilities are held, and whether
 619  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 620  * and size are monotonically increasing, except when utimes() or
 621  * truncate() increments the corresponding _seq values.)
 622  */
 623 int ceph_fill_file_size(struct inode *inode, int issued,
 624                         u32 truncate_seq, u64 truncate_size, u64 size)
 625 {
 626         struct ceph_inode_info *ci = ceph_inode(inode);
 627         int queue_trunc = 0;
 628         loff_t isize = i_size_read(inode);
 629
 630         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 631             (truncate_seq == ci->i_truncate_seq && size > isize)) {
 632                 dout("size %lld -> %llu\n", isize, size);
 633                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 634                         pr_err("fill_file_size non-zero size for directory\n");
 635                         size = 0;
 636                 }
 637                 i_size_write(inode, size);
 638                 inode->i_blocks = calc_inode_blocks(size);
 639                 ci->i_reported_size = size;
 640                 if (truncate_seq != ci->i_truncate_seq) {
 641                         dout("truncate_seq %u -> %u\n",
 642                              ci->i_truncate_seq, truncate_seq);
 643                         ci->i_truncate_seq = truncate_seq;
 644
 645                         /* the MDS should have revoked these caps */
 646                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 647                                                CEPH_CAP_FILE_RD |
 648                                                CEPH_CAP_FILE_WR |
 649                                                CEPH_CAP_FILE_LAZYIO));
 650                         /*
 651                          * If we hold relevant caps, or in the case where we're
 652                          * not the only client referencing this file and we
 653                          * don't hold those caps, then we need to check whether
 654                          * the file is either opened or mmaped
 655                          */
 656                         if ((issued & (CEPH_CAP_FILE_CACHE|
 657                                        CEPH_CAP_FILE_BUFFER)) ||
 658                             mapping_mapped(inode->i_mapping) ||
 659                             __ceph_is_file_opened(ci)) {
 660                                 ci->i_truncate_pending++;
 661                                 queue_trunc = 1;
 662                         }
 663                 }
 664         }
 665         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 666             ci->i_truncate_size != truncate_size) {
 667                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 668                      truncate_size);
 669                 ci->i_truncate_size = truncate_size;
 670         }
 671
 672         if (queue_trunc)
 673                 ceph_fscache_invalidate(inode);
 674
 675         return queue_trunc;
 676 }
 677
 678 void ceph_fill_file_time(struct inode *inode, int issued,
 679                          u64 time_warp_seq, struct timespec64 *ctime,
 680                          struct timespec64 *mtime, struct timespec64 *atime)
 681 {
 682         struct ceph_inode_info *ci = ceph_inode(inode);
 683         int warn = 0;
 684
 685         if (issued & (CEPH_CAP_FILE_EXCL|
 686                       CEPH_CAP_FILE_WR|
 687                       CEPH_CAP_FILE_BUFFER|
 688                       CEPH_CAP_AUTH_EXCL|
 689                       CEPH_CAP_XATTR_EXCL)) {
 690                 if (ci->i_version == 0 ||
 691                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 692                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 693                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 694                              ctime->tv_sec, ctime->tv_nsec);
 695                         inode->i_ctime = *ctime;
 696                 }
 697                 if (ci->i_version == 0 ||
 698                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 699                         /* the MDS did a utimes() */
 700                         dout("mtime %lld.%09ld -> %lld.%09ld "
 701                              "tw %d -> %d\n",
 702                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 703                              mtime->tv_sec, mtime->tv_nsec,
 704                              ci->i_time_warp_seq, (int)time_warp_seq);
 705
 706                         inode->i_mtime = *mtime;
 707                         inode->i_atime = *atime;
 708                         ci->i_time_warp_seq = time_warp_seq;
 709                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 710                         /* nobody did utimes(); take the max */
 711                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 712                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 713                                      inode->i_mtime.tv_sec,
 714                                      inode->i_mtime.tv_nsec,
 715                                      mtime->tv_sec, mtime->tv_nsec);
 716                                 inode->i_mtime = *mtime;
 717                         }
 718                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 719                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 720                                      inode->i_atime.tv_sec,
 721                                      inode->i_atime.tv_nsec,
 722                                      atime->tv_sec, atime->tv_nsec);
 723                                 inode->i_atime = *atime;
 724                         }
 725                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 726                         /* we did a utimes(); ignore mds values */
 727                 } else {
 728                         warn = 1;
 729                 }
 730         } else {
 731                 /* we have no write|excl caps; whatever the MDS says is true */
 732                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 733                         inode->i_ctime = *ctime;
 734                         inode->i_mtime = *mtime;
 735                         inode->i_atime = *atime;
 736                         ci->i_time_warp_seq = time_warp_seq;
 737                 } else {
 738                         warn = 1;
 739                 }
 740         }
 741         if (warn) /* time_warp_seq shouldn't go backwards */
 742                 dout("%p mds time_warp_seq %llu < %u\n",
 743                      inode, time_warp_seq, ci->i_time_warp_seq);
 744 }
 745
 746 /*
 747  * Populate an inode based on info from mds.  May be called on new or
 748  * existing inodes.
 749  */
 750 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 751                     struct ceph_mds_reply_info_in *iinfo,
 752                     struct ceph_mds_reply_dirfrag *dirinfo,
 753                     struct ceph_mds_session *session, int cap_fmode,
 754                     struct ceph_cap_reservation *caps_reservation)
 755 {
 756         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 757         struct ceph_mds_reply_inode *info = iinfo->in;
 758         struct ceph_inode_info *ci = ceph_inode(inode);
 759         int issued, new_issued, info_caps;
 760         struct timespec64 mtime, atime, ctime;
 761         struct ceph_buffer *xattr_blob = NULL;
 762         struct ceph_buffer *old_blob = NULL;
 763         struct ceph_string *pool_ns = NULL;
 764         struct ceph_cap *new_cap = NULL;
 765         int err = 0;
 766         bool wake = false;
 767         bool queue_trunc = false;
 768         bool new_version = false;
 769         bool fill_inline = false;
 770         umode_t mode = le32_to_cpu(info->mode);
 771         dev_t rdev = le32_to_cpu(info->rdev);
 772
 773         lockdep_assert_held(&mdsc->snap_rwsem);
 774
 775         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 776              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 777              ci->i_version);
 778
 779         /* Once I_NEW is cleared, we can't change type or dev numbers */
 780         if (inode->i_state & I_NEW) {
 781                 inode->i_mode = mode;
 782         } else {
 783                 if (inode_wrong_type(inode, mode)) {
 784                         pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
 785                                      ceph_vinop(inode), inode->i_mode, mode);
 786                         return -ESTALE;
 787                 }
 788
 789                 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
 790                         pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
 791                                      ceph_vinop(inode), MAJOR(inode->i_rdev),
 792                                      MINOR(inode->i_rdev), MAJOR(rdev),
 793                                      MINOR(rdev));
 794                         return -ESTALE;
 795                 }
 796         }
 797
 798         info_caps = le32_to_cpu(info->cap.caps);
 799
 800         /* prealloc new cap struct */
 801         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 802                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 803                 if (!new_cap)
 804                         return -ENOMEM;
 805         }
 806
 807         /*
 808          * prealloc xattr data, if it looks like we'll need it.  only
 809          * if len > 4 (meaning there are actually xattrs; the first 4
 810          * bytes are the xattr count).
 811          */
 812         if (iinfo->xattr_len > 4) {
 813                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 814                 if (!xattr_blob)
 815                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 816                                iinfo->xattr_len);
 817         }
 818
 819         if (iinfo->pool_ns_len > 0)
 820                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 821                                                      iinfo->pool_ns_len);
 822
 823         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 824                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 825
 826         spin_lock(&ci->i_ceph_lock);
 827
 828         /*
 829          * provided version will be odd if inode value is projected,
 830          * even if stable.  skip the update if we have newer stable
 831          * info (ours>=theirs, e.g. due to racing mds replies), unless
 832          * we are getting projected (unstable) info (in which case the
 833          * version is odd, and we want ours>theirs).
 834          *   us   them
 835          *   2    2     skip
 836          *   3    2     skip
 837          *   3    3     update
 838          */
 839         if (ci->i_version == 0 ||
 840             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 841              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 842                 new_version = true;
 843
 844         /* Update change_attribute */
 845         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 846
 847         __ceph_caps_issued(ci, &issued);
 848         issued |= __ceph_caps_dirty(ci);
 849         new_issued = ~issued & info_caps;
 850
 851         /* directories have fl_stripe_unit set to zero */
 852         if (le32_to_cpu(info->layout.fl_stripe_unit))
 853                 inode->i_blkbits =
 854                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 855         else
 856                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 857
 858         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 859
 860         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 861             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 862                 inode->i_mode = mode;
 863                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 864                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 865                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 866                      from_kuid(&init_user_ns, inode->i_uid),
 867                      from_kgid(&init_user_ns, inode->i_gid));
 868                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 869                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 870         }
 871
 872         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 873             (issued & CEPH_CAP_LINK_EXCL) == 0)
 874                 set_nlink(inode, le32_to_cpu(info->nlink));
 875
 876         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 877                 /* be careful with mtime, atime, size */
 878                 ceph_decode_timespec64(&atime, &info->atime);
 879                 ceph_decode_timespec64(&mtime, &info->mtime);
 880                 ceph_decode_timespec64(&ctime, &info->ctime);
 881                 ceph_fill_file_time(inode, issued,
 882                                 le32_to_cpu(info->time_warp_seq),
 883                                 &ctime, &mtime, &atime);
 884         }
 885
 886         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 887                 ci->i_files = le64_to_cpu(info->files);
 888                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 889         }
 890
 891         if (new_version ||
 892             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 893                 s64 old_pool = ci->i_layout.pool_id;
 894                 struct ceph_string *old_ns;
 895
 896                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 897                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 898                                         lockdep_is_held(&ci->i_ceph_lock));
 899                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 900
 901                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 902                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 903
 904                 pool_ns = old_ns;
 905
 906                 queue_trunc = ceph_fill_file_size(inode, issued,
 907                                         le32_to_cpu(info->truncate_seq),
 908                                         le64_to_cpu(info->truncate_size),
 909                                         le64_to_cpu(info->size));
 910                 /* only update max_size on auth cap */
 911                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 912                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 913                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 914                                         le64_to_cpu(info->max_size));
 915                         ci->i_max_size = le64_to_cpu(info->max_size);
 916                 }
 917         }
 918
 919         /* layout and rstat are not tracked by capability, update them if
 920          * the inode info is from auth mds */
 921         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 922                 if (S_ISDIR(inode->i_mode)) {
 923                         ci->i_dir_layout = iinfo->dir_layout;
 924                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 925                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 926                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 927                         ci->i_dir_pin = iinfo->dir_pin;
 928                         ci->i_rsnaps = iinfo->rsnaps;
 929                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 930                 }
 931         }
 932
 933         /* xattrs */
 934         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 935         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 936             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 937                 if (ci->i_xattrs.blob)
 938                         old_blob = ci->i_xattrs.blob;
 939                 ci->i_xattrs.blob = xattr_blob;
 940                 if (xattr_blob)
 941                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 942                                iinfo->xattr_data, iinfo->xattr_len);
 943                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 944                 ceph_forget_all_cached_acls(inode);
 945                 ceph_security_invalidate_secctx(inode);
 946                 xattr_blob = NULL;
 947         }
 948
 949         /* finally update i_version */
 950         if (le64_to_cpu(info->version) > ci->i_version)
 951                 ci->i_version = le64_to_cpu(info->version);
 952
 953         inode->i_mapping->a_ops = &ceph_aops;
 954
 955         switch (inode->i_mode & S_IFMT) {
 956         case S_IFIFO:
 957         case S_IFBLK:
 958         case S_IFCHR:
 959         case S_IFSOCK:
 960                 inode->i_blkbits = PAGE_SHIFT;
 961                 init_special_inode(inode, inode->i_mode, rdev);
 962                 inode->i_op = &ceph_file_iops;
 963                 break;
 964         case S_IFREG:
 965                 inode->i_op = &ceph_file_iops;
 966                 inode->i_fop = &ceph_file_fops;
 967                 break;
 968         case S_IFLNK:
 969                 inode->i_op = &ceph_symlink_iops;
 970                 if (!ci->i_symlink) {
 971                         u32 symlen = iinfo->symlink_len;
 972                         char *sym;
 973
 974                         spin_unlock(&ci->i_ceph_lock);
 975
 976                         if (symlen != i_size_read(inode)) {
 977                                 pr_err("%s %llx.%llx BAD symlink "
 978                                         "size %lld\n", __func__,
 979                                         ceph_vinop(inode),
 980                                         i_size_read(inode));
 981                                 i_size_write(inode, symlen);
 982                                 inode->i_blocks = calc_inode_blocks(symlen);
 983                         }
 984
 985                         err = -ENOMEM;
 986                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 987                         if (!sym)
 988                                 goto out;
 989
 990                         spin_lock(&ci->i_ceph_lock);
 991                         if (!ci->i_symlink)
 992                                 ci->i_symlink = sym;
 993                         else
 994                                 kfree(sym); /* lost a race */
 995                 }
 996                 inode->i_link = ci->i_symlink;
 997                 break;
 998         case S_IFDIR:
 999                 inode->i_op = &ceph_dir_iops;
1000                 inode->i_fop = &ceph_dir_fops;
1001                 break;
1002         default:
1003                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1004                        ceph_vinop(inode), inode->i_mode);
1005         }
1006
1007         /* were we issued a capability? */
1008         if (info_caps) {
1009                 if (ceph_snap(inode) == CEPH_NOSNAP) {
1010                         ceph_add_cap(inode, session,
1011                                      le64_to_cpu(info->cap.cap_id),
1012                                      info_caps,
1013                                      le32_to_cpu(info->cap.wanted),
1014                                      le32_to_cpu(info->cap.seq),
1015                                      le32_to_cpu(info->cap.mseq),
1016                                      le64_to_cpu(info->cap.realm),
1017                                      info->cap.flags, &new_cap);
1018
1019                         /* set dir completion flag? */
1020                         if (S_ISDIR(inode->i_mode) &&
1021                             ci->i_files == 0 && ci->i_subdirs == 0 &&
1022                             (info_caps & CEPH_CAP_FILE_SHARED) &&
1023                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1024                             !__ceph_dir_is_complete(ci)) {
1025                                 dout(" marking %p complete (empty)\n", inode);
1026                                 i_size_write(inode, 0);
1027                                 __ceph_dir_set_complete(ci,
1028                                         atomic64_read(&ci->i_release_count),
1029                                         atomic64_read(&ci->i_ordered_count));
1030                         }
1031
1032                         wake = true;
1033                 } else {
1034                         dout(" %p got snap_caps %s\n", inode,
1035                              ceph_cap_string(info_caps));
1036                         ci->i_snap_caps |= info_caps;
1037                 }
1038         }
1039
1040         if (iinfo->inline_version > 0 &&
1041             iinfo->inline_version >= ci->i_inline_version) {
1042                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1043                 ci->i_inline_version = iinfo->inline_version;
1044                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1045                     (locked_page || (info_caps & cache_caps)))
1046                         fill_inline = true;
1047         }
1048
1049         if (cap_fmode >= 0) {
1050                 if (!info_caps)
1051                         pr_warn("mds issued no caps on %llx.%llx\n",
1052                                 ceph_vinop(inode));
1053                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1054         }
1055
1056         spin_unlock(&ci->i_ceph_lock);
1057
1058         if (fill_inline)
1059                 ceph_fill_inline_data(inode, locked_page,
1060                                       iinfo->inline_data, iinfo->inline_len);
1061
1062         if (wake)
1063                 wake_up_all(&ci->i_cap_wq);
1064
1065         /* queue truncate if we saw i_size decrease */
1066         if (queue_trunc)
1067                 ceph_queue_vmtruncate(inode);
1068
1069         /* populate frag tree */
1070         if (S_ISDIR(inode->i_mode))
1071                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1072
1073         /* update delegation info? */
1074         if (dirinfo)
1075                 ceph_fill_dirfrag(inode, dirinfo);
1076
1077         err = 0;
1078 out:
1079         if (new_cap)
1080                 ceph_put_cap(mdsc, new_cap);
1081         ceph_buffer_put(old_blob);
1082         ceph_buffer_put(xattr_blob);
1083         ceph_put_string(pool_ns);
1084         return err;
1085 }
1086
1087 /*
1088  * caller should hold session s_mutex and dentry->d_lock.
1089  */
1090 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1091                                   struct ceph_mds_reply_lease *lease,
1092                                   struct ceph_mds_session *session,
1093                                   unsigned long from_time,
1094                                   struct ceph_mds_session **old_lease_session)
1095 {
1096         struct ceph_dentry_info *di = ceph_dentry(dentry);
1097         unsigned mask = le16_to_cpu(lease->mask);
1098         long unsigned duration = le32_to_cpu(lease->duration_ms);
1099         long unsigned ttl = from_time + (duration * HZ) / 1000;
1100         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1101
1102         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1103              dentry, duration, ttl);
1104
1105         /* only track leases on regular dentries */
1106         if (ceph_snap(dir) != CEPH_NOSNAP)
1107                 return;
1108
1109         if (mask & CEPH_LEASE_PRIMARY_LINK)
1110                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1111         else
1112                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1113
1114         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1115         if (!(mask & CEPH_LEASE_VALID)) {
1116                 __ceph_dentry_dir_lease_touch(di);
1117                 return;
1118         }
1119
1120         if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1121             time_before(ttl, di->time))
1122                 return;  /* we already have a newer lease. */
1123
1124         if (di->lease_session && di->lease_session != session) {
1125                 *old_lease_session = di->lease_session;
1126                 di->lease_session = NULL;
1127         }
1128
1129         if (!di->lease_session)
1130                 di->lease_session = ceph_get_mds_session(session);
1131         di->lease_gen = atomic_read(&session->s_cap_gen);
1132         di->lease_seq = le32_to_cpu(lease->seq);
1133         di->lease_renew_after = half_ttl;
1134         di->lease_renew_from = 0;
1135         di->time = ttl;
1136
1137         __ceph_dentry_lease_touch(di);
1138 }
1139
1140 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1141                                         struct ceph_mds_reply_lease *lease,
1142                                         struct ceph_mds_session *session,
1143                                         unsigned long from_time)
1144 {
1145         struct ceph_mds_session *old_lease_session = NULL;
1146         spin_lock(&dentry->d_lock);
1147         __update_dentry_lease(dir, dentry, lease, session, from_time,
1148                               &old_lease_session);
1149         spin_unlock(&dentry->d_lock);
1150         ceph_put_mds_session(old_lease_session);
1151 }
1152
1153 /*
1154  * update dentry lease without having parent inode locked
1155  */
1156 static void update_dentry_lease_careful(struct dentry *dentry,
1157                                         struct ceph_mds_reply_lease *lease,
1158                                         struct ceph_mds_session *session,
1159                                         unsigned long from_time,
1160                                         char *dname, u32 dname_len,
1161                                         struct ceph_vino *pdvino,
1162                                         struct ceph_vino *ptvino)
1163
1164 {
1165         struct inode *dir;
1166         struct ceph_mds_session *old_lease_session = NULL;
1167
1168         spin_lock(&dentry->d_lock);
1169         /* make sure dentry's name matches target */
1170         if (dentry->d_name.len != dname_len ||
1171             memcmp(dentry->d_name.name, dname, dname_len))
1172                 goto out_unlock;
1173
1174         dir = d_inode(dentry->d_parent);
1175         /* make sure parent matches dvino */
1176         if (!ceph_ino_compare(dir, pdvino))
1177                 goto out_unlock;
1178
1179         /* make sure dentry's inode matches target. NULL ptvino means that
1180          * we expect a negative dentry */
1181         if (ptvino) {
1182                 if (d_really_is_negative(dentry))
1183                         goto out_unlock;
1184                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1185                         goto out_unlock;
1186         } else {
1187                 if (d_really_is_positive(dentry))
1188                         goto out_unlock;
1189         }
1190
1191         __update_dentry_lease(dir, dentry, lease, session,
1192                               from_time, &old_lease_session);
1193 out_unlock:
1194         spin_unlock(&dentry->d_lock);
1195         ceph_put_mds_session(old_lease_session);
1196 }
1197
1198 /*
1199  * splice a dentry to an inode.
1200  * caller must hold directory i_mutex for this to be safe.
1201  */
1202 static int splice_dentry(struct dentry **pdn, struct inode *in)
1203 {
1204         struct dentry *dn = *pdn;
1205         struct dentry *realdn;
1206
1207         BUG_ON(d_inode(dn));
1208
1209         if (S_ISDIR(in->i_mode)) {
1210                 /* If inode is directory, d_splice_alias() below will remove
1211                  * 'realdn' from its origin parent. We need to ensure that
1212                  * origin parent's readdir cache will not reference 'realdn'
1213                  */
1214                 realdn = d_find_any_alias(in);
1215                 if (realdn) {
1216                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1217                         spin_lock(&realdn->d_lock);
1218
1219                         realdn->d_op->d_prune(realdn);
1220
1221                         di->time = jiffies;
1222                         di->lease_shared_gen = 0;
1223                         di->offset = 0;
1224
1225                         spin_unlock(&realdn->d_lock);
1226                         dput(realdn);
1227                 }
1228         }
1229
1230         /* dn must be unhashed */
1231         if (!d_unhashed(dn))
1232                 d_drop(dn);
1233         realdn = d_splice_alias(in, dn);
1234         if (IS_ERR(realdn)) {
1235                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1236                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1237                 return PTR_ERR(realdn);
1238         }
1239
1240         if (realdn) {
1241                 dout("dn %p (%d) spliced with %p (%d) "
1242                      "inode %p ino %llx.%llx\n",
1243                      dn, d_count(dn),
1244                      realdn, d_count(realdn),
1245                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1246                 dput(dn);
1247                 *pdn = realdn;
1248         } else {
1249                 BUG_ON(!ceph_dentry(dn));
1250                 dout("dn %p attached to %p ino %llx.%llx\n",
1251                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1252         }
1253         return 0;
1254 }
1255
1256 /*
1257  * Incorporate results into the local cache.  This is either just
1258  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1259  * after a lookup).
1260  *
1261  * A reply may contain
1262  *         a directory inode along with a dentry.
1263  *  and/or a target inode
1264  *
1265  * Called with snap_rwsem (read).
1266  */
1267 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1268 {
1269         struct ceph_mds_session *session = req->r_session;
1270         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1271         struct inode *in = NULL;
1272         struct ceph_vino tvino, dvino;
1273         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1274         int err = 0;
1275
1276         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1277              rinfo->head->is_dentry, rinfo->head->is_target);
1278
1279         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1280                 dout("fill_trace reply is empty!\n");
1281                 if (rinfo->head->result == 0 && req->r_parent)
1282                         ceph_invalidate_dir_request(req);
1283                 return 0;
1284         }
1285
1286         if (rinfo->head->is_dentry) {
1287                 struct inode *dir = req->r_parent;
1288
1289                 if (dir) {
1290                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1291                                               rinfo->dirfrag, session, -1,
1292                                               &req->r_caps_reservation);
1293                         if (err < 0)
1294                                 goto done;
1295                 } else {
1296                         WARN_ON_ONCE(1);
1297                 }
1298
1299                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1300                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1301                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1302                         struct qstr dname;
1303                         struct dentry *dn, *parent;
1304
1305                         BUG_ON(!rinfo->head->is_target);
1306                         BUG_ON(req->r_dentry);
1307
1308                         parent = d_find_any_alias(dir);
1309                         BUG_ON(!parent);
1310
1311                         dname.name = rinfo->dname;
1312                         dname.len = rinfo->dname_len;
1313                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1314                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1315                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1316 retry_lookup:
1317                         dn = d_lookup(parent, &dname);
1318                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1319                              parent, dname.len, dname.name, dn);
1320
1321                         if (!dn) {
1322                                 dn = d_alloc(parent, &dname);
1323                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1324                                      dname.len, dname.name, dn);
1325                                 if (!dn) {
1326                                         dput(parent);
1327                                         err = -ENOMEM;
1328                                         goto done;
1329                                 }
1330                                 err = 0;
1331                         } else if (d_really_is_positive(dn) &&
1332                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1333                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1334                                 dout(" dn %p points to wrong inode %p\n",
1335                                      dn, d_inode(dn));
1336                                 ceph_dir_clear_ordered(dir);
1337                                 d_delete(dn);
1338                                 dput(dn);
1339                                 goto retry_lookup;
1340                         }
1341
1342                         req->r_dentry = dn;
1343                         dput(parent);
1344                 }
1345         }
1346
1347         if (rinfo->head->is_target) {
1348                 /* Should be filled in by handle_reply */
1349                 BUG_ON(!req->r_target_inode);
1350
1351                 in = req->r_target_inode;
1352                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1353                                 NULL, session,
1354                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1355                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1356                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1357                                 &req->r_caps_reservation);
1358                 if (err < 0) {
1359                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1360                                 in, ceph_vinop(in));
1361                         req->r_target_inode = NULL;
1362                         if (in->i_state & I_NEW)
1363                                 discard_new_inode(in);
1364                         else
1365                                 iput(in);
1366                         goto done;
1367                 }
1368                 if (in->i_state & I_NEW)
1369                         unlock_new_inode(in);
1370         }
1371
1372         /*
1373          * ignore null lease/binding on snapdir ENOENT, or else we
1374          * will have trouble splicing in the virtual snapdir later
1375          */
1376         if (rinfo->head->is_dentry &&
1377             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1378             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1379             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1380                                                fsc->mount_options->snapdir_name,
1381                                                req->r_dentry->d_name.len))) {
1382                 /*
1383                  * lookup link rename   : null -> possibly existing inode
1384                  * mknod symlink mkdir  : null -> new inode
1385                  * unlink               : linked -> null
1386                  */
1387                 struct inode *dir = req->r_parent;
1388                 struct dentry *dn = req->r_dentry;
1389                 bool have_dir_cap, have_lease;
1390
1391                 BUG_ON(!dn);
1392                 BUG_ON(!dir);
1393                 BUG_ON(d_inode(dn->d_parent) != dir);
1394
1395                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1396                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1397
1398                 BUG_ON(ceph_ino(dir) != dvino.ino);
1399                 BUG_ON(ceph_snap(dir) != dvino.snap);
1400
1401                 /* do we have a lease on the whole dir? */
1402                 have_dir_cap =
1403                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1404                          CEPH_CAP_FILE_SHARED);
1405
1406                 /* do we have a dn lease? */
1407                 have_lease = have_dir_cap ||
1408                         le32_to_cpu(rinfo->dlease->duration_ms);
1409                 if (!have_lease)
1410                         dout("fill_trace  no dentry lease or dir cap\n");
1411
1412                 /* rename? */
1413                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1414                         struct inode *olddir = req->r_old_dentry_dir;
1415                         BUG_ON(!olddir);
1416
1417                         dout(" src %p '%pd' dst %p '%pd'\n",
1418                              req->r_old_dentry,
1419                              req->r_old_dentry,
1420                              dn, dn);
1421                         dout("fill_trace doing d_move %p -> %p\n",
1422                              req->r_old_dentry, dn);
1423
1424                         /* d_move screws up sibling dentries' offsets */
1425                         ceph_dir_clear_ordered(dir);
1426                         ceph_dir_clear_ordered(olddir);
1427
1428                         d_move(req->r_old_dentry, dn);
1429                         dout(" src %p '%pd' dst %p '%pd'\n",
1430                              req->r_old_dentry,
1431                              req->r_old_dentry,
1432                              dn, dn);
1433
1434                         /* ensure target dentry is invalidated, despite
1435                            rehashing bug in vfs_rename_dir */
1436                         ceph_invalidate_dentry_lease(dn);
1437
1438                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1439                              ceph_dentry(req->r_old_dentry)->offset);
1440
1441                         /* swap r_dentry and r_old_dentry in case that
1442                          * splice_dentry() gets called later. This is safe
1443                          * because no other place will use them */
1444                         req->r_dentry = req->r_old_dentry;
1445                         req->r_old_dentry = dn;
1446                         dn = req->r_dentry;
1447                 }
1448
1449                 /* null dentry? */
1450                 if (!rinfo->head->is_target) {
1451                         dout("fill_trace null dentry\n");
1452                         if (d_really_is_positive(dn)) {
1453                                 dout("d_delete %p\n", dn);
1454                                 ceph_dir_clear_ordered(dir);
1455                                 d_delete(dn);
1456                         } else if (have_lease) {
1457                                 if (d_unhashed(dn))
1458                                         d_add(dn, NULL);
1459                                 update_dentry_lease(dir, dn,
1460                                                     rinfo->dlease, session,
1461                                                     req->r_request_started);
1462                         }
1463                         goto done;
1464                 }
1465
1466                 /* attach proper inode */
1467                 if (d_really_is_negative(dn)) {
1468                         ceph_dir_clear_ordered(dir);
1469                         ihold(in);
1470                         err = splice_dentry(&req->r_dentry, in);
1471                         if (err < 0)
1472                                 goto done;
1473                         dn = req->r_dentry;  /* may have spliced */
1474                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1475                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1476                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1477                              ceph_vinop(in));
1478                         d_invalidate(dn);
1479                         have_lease = false;
1480                 }
1481
1482                 if (have_lease) {
1483                         update_dentry_lease(dir, dn,
1484                                             rinfo->dlease, session,
1485                                             req->r_request_started);
1486                 }
1487                 dout(" final dn %p\n", dn);
1488         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1489                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1490                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1491                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1492                 struct inode *dir = req->r_parent;
1493
1494                 /* fill out a snapdir LOOKUPSNAP dentry */
1495                 BUG_ON(!dir);
1496                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1497                 BUG_ON(!req->r_dentry);
1498                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1499                 ceph_dir_clear_ordered(dir);
1500                 ihold(in);
1501                 err = splice_dentry(&req->r_dentry, in);
1502                 if (err < 0)
1503                         goto done;
1504         } else if (rinfo->head->is_dentry && req->r_dentry) {
1505                 /* parent inode is not locked, be carefull */
1506                 struct ceph_vino *ptvino = NULL;
1507                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1508                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1509                 if (rinfo->head->is_target) {
1510                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1511                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1512                         ptvino = &tvino;
1513                 }
1514                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1515                                             session, req->r_request_started,
1516                                             rinfo->dname, rinfo->dname_len,
1517                                             &dvino, ptvino);
1518         }
1519 done:
1520         dout("fill_trace done err=%d\n", err);
1521         return err;
1522 }
1523
1524 /*
1525  * Prepopulate our cache with readdir results, leases, etc.
1526  */
1527 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1528                                            struct ceph_mds_session *session)
1529 {
1530         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1531         int i, err = 0;
1532
1533         for (i = 0; i < rinfo->dir_nr; i++) {
1534                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1535                 struct ceph_vino vino;
1536                 struct inode *in;
1537                 int rc;
1538
1539                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1540                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1541
1542                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1543                 if (IS_ERR(in)) {
1544                         err = PTR_ERR(in);
1545                         dout("new_inode badness got %d\n", err);
1546                         continue;
1547                 }
1548                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1549                                      -1, &req->r_caps_reservation);
1550                 if (rc < 0) {
1551                         pr_err("ceph_fill_inode badness on %p got %d\n",
1552                                in, rc);
1553                         err = rc;
1554                         if (in->i_state & I_NEW) {
1555                                 ihold(in);
1556                                 discard_new_inode(in);
1557                         }
1558                 } else if (in->i_state & I_NEW) {
1559                         unlock_new_inode(in);
1560                 }
1561
1562                 iput(in);
1563         }
1564
1565         return err;
1566 }
1567
1568 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1569 {
1570         if (ctl->page) {
1571                 kunmap(ctl->page);
1572                 put_page(ctl->page);
1573                 ctl->page = NULL;
1574         }
1575 }
1576
1577 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1578                               struct ceph_readdir_cache_control *ctl,
1579                               struct ceph_mds_request *req)
1580 {
1581         struct ceph_inode_info *ci = ceph_inode(dir);
1582         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1583         unsigned idx = ctl->index % nsize;
1584         pgoff_t pgoff = ctl->index / nsize;
1585
1586         if (!ctl->page || pgoff != page_index(ctl->page)) {
1587                 ceph_readdir_cache_release(ctl);
1588                 if (idx == 0)
1589                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1590                 else
1591                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1592                 if (!ctl->page) {
1593                         ctl->index = -1;
1594                         return idx == 0 ? -ENOMEM : 0;
1595                 }
1596                 /* reading/filling the cache are serialized by
1597                  * i_mutex, no need to use page lock */
1598                 unlock_page(ctl->page);
1599                 ctl->dentries = kmap(ctl->page);
1600                 if (idx == 0)
1601                         memset(ctl->dentries, 0, PAGE_SIZE);
1602         }
1603
1604         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1605             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1606                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1607                 ctl->dentries[idx] = dn;
1608                 ctl->index++;
1609         } else {
1610                 dout("disable readdir cache\n");
1611                 ctl->index = -1;
1612         }
1613         return 0;
1614 }
1615
1616 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1617                              struct ceph_mds_session *session)
1618 {
1619         struct dentry *parent = req->r_dentry;
1620         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1621         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1622         struct qstr dname;
1623         struct dentry *dn;
1624         struct inode *in;
1625         int err = 0, skipped = 0, ret, i;
1626         u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1627         u32 last_hash = 0;
1628         u32 fpos_offset;
1629         struct ceph_readdir_cache_control cache_ctl = {};
1630
1631         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1632                 return readdir_prepopulate_inodes_only(req, session);
1633
1634         if (rinfo->hash_order) {
1635                 if (req->r_path2) {
1636                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1637                                                   req->r_path2,
1638                                                   strlen(req->r_path2));
1639                         last_hash = ceph_frag_value(last_hash);
1640                 } else if (rinfo->offset_hash) {
1641                         /* mds understands offset_hash */
1642                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1643                         last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1644                 }
1645         }
1646
1647         if (rinfo->dir_dir &&
1648             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1649                 dout("readdir_prepopulate got new frag %x -> %x\n",
1650                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1651                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1652                 if (!rinfo->hash_order)
1653                         req->r_readdir_offset = 2;
1654         }
1655
1656         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1657                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1658                      rinfo->dir_nr, parent);
1659         } else {
1660                 dout("readdir_prepopulate %d items under dn %p\n",
1661                      rinfo->dir_nr, parent);
1662                 if (rinfo->dir_dir)
1663                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1664
1665                 if (ceph_frag_is_leftmost(frag) &&
1666                     req->r_readdir_offset == 2 &&
1667                     !(rinfo->hash_order && last_hash)) {
1668                         /* note dir version at start of readdir so we can
1669                          * tell if any dentries get dropped */
1670                         req->r_dir_release_cnt =
1671                                 atomic64_read(&ci->i_release_count);
1672                         req->r_dir_ordered_cnt =
1673                                 atomic64_read(&ci->i_ordered_count);
1674                         req->r_readdir_cache_idx = 0;
1675                 }
1676         }
1677
1678         cache_ctl.index = req->r_readdir_cache_idx;
1679         fpos_offset = req->r_readdir_offset;
1680
1681         /* FIXME: release caps/leases if error occurs */
1682         for (i = 0; i < rinfo->dir_nr; i++) {
1683                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1684                 struct ceph_vino tvino;
1685
1686                 dname.name = rde->name;
1687                 dname.len = rde->name_len;
1688                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1689
1690                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1691                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1692
1693                 if (rinfo->hash_order) {
1694                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1695                                                  rde->name, rde->name_len);
1696                         hash = ceph_frag_value(hash);
1697                         if (hash != last_hash)
1698                                 fpos_offset = 2;
1699                         last_hash = hash;
1700                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1701                 } else {
1702                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1703                 }
1704
1705 retry_lookup:
1706                 dn = d_lookup(parent, &dname);
1707                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1708                      parent, dname.len, dname.name, dn);
1709
1710                 if (!dn) {
1711                         dn = d_alloc(parent, &dname);
1712                         dout("d_alloc %p '%.*s' = %p\n", parent,
1713                              dname.len, dname.name, dn);
1714                         if (!dn) {
1715                                 dout("d_alloc badness\n");
1716                                 err = -ENOMEM;
1717                                 goto out;
1718                         }
1719                 } else if (d_really_is_positive(dn) &&
1720                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1721                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1722                         struct ceph_dentry_info *di = ceph_dentry(dn);
1723                         dout(" dn %p points to wrong inode %p\n",
1724                              dn, d_inode(dn));
1725
1726                         spin_lock(&dn->d_lock);
1727                         if (di->offset > 0 &&
1728                             di->lease_shared_gen ==
1729                             atomic_read(&ci->i_shared_gen)) {
1730                                 __ceph_dir_clear_ordered(ci);
1731                                 di->offset = 0;
1732                         }
1733                         spin_unlock(&dn->d_lock);
1734
1735                         d_delete(dn);
1736                         dput(dn);
1737                         goto retry_lookup;
1738                 }
1739
1740                 /* inode */
1741                 if (d_really_is_positive(dn)) {
1742                         in = d_inode(dn);
1743                 } else {
1744                         in = ceph_get_inode(parent->d_sb, tvino);
1745                         if (IS_ERR(in)) {
1746                                 dout("new_inode badness\n");
1747                                 d_drop(dn);
1748                                 dput(dn);
1749                                 err = PTR_ERR(in);
1750                                 goto out;
1751                         }
1752                 }
1753
1754                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1755                                       -1, &req->r_caps_reservation);
1756                 if (ret < 0) {
1757                         pr_err("ceph_fill_inode badness on %p\n", in);
1758                         if (d_really_is_negative(dn)) {
1759                                 if (in->i_state & I_NEW) {
1760                                         ihold(in);
1761                                         discard_new_inode(in);
1762                                 }
1763                                 iput(in);
1764                         }
1765                         d_drop(dn);
1766                         err = ret;
1767                         goto next_item;
1768                 }
1769                 if (in->i_state & I_NEW)
1770                         unlock_new_inode(in);
1771
1772                 if (d_really_is_negative(dn)) {
1773                         if (ceph_security_xattr_deadlock(in)) {
1774                                 dout(" skip splicing dn %p to inode %p"
1775                                      " (security xattr deadlock)\n", dn, in);
1776                                 iput(in);
1777                                 skipped++;
1778                                 goto next_item;
1779                         }
1780
1781                         err = splice_dentry(&dn, in);
1782                         if (err < 0)
1783                                 goto next_item;
1784                 }
1785
1786                 ceph_dentry(dn)->offset = rde->offset;
1787
1788                 update_dentry_lease(d_inode(parent), dn,
1789                                     rde->lease, req->r_session,
1790                                     req->r_request_started);
1791
1792                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1793                         ret = fill_readdir_cache(d_inode(parent), dn,
1794                                                  &cache_ctl, req);
1795                         if (ret < 0)
1796                                 err = ret;
1797                 }
1798 next_item:
1799                 dput(dn);
1800         }
1801 out:
1802         if (err == 0 && skipped == 0) {
1803                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1804                 req->r_readdir_cache_idx = cache_ctl.index;
1805         }
1806         ceph_readdir_cache_release(&cache_ctl);
1807         dout("readdir_prepopulate done\n");
1808         return err;
1809 }
1810
1811 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1812 {
1813         struct ceph_inode_info *ci = ceph_inode(inode);
1814         bool ret;
1815
1816         spin_lock(&ci->i_ceph_lock);
1817         dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1818         i_size_write(inode, size);
1819         inode->i_blocks = calc_inode_blocks(size);
1820
1821         ret = __ceph_should_report_size(ci);
1822
1823         spin_unlock(&ci->i_ceph_lock);
1824         return ret;
1825 }
1826
1827 void ceph_queue_inode_work(struct inode *inode, int work_bit)
1828 {
1829         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1830         struct ceph_inode_info *ci = ceph_inode(inode);
1831         set_bit(work_bit, &ci->i_work_mask);
1832
1833         ihold(inode);
1834         if (queue_work(fsc->inode_wq, &ci->i_work)) {
1835                 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1836         } else {
1837                 dout("queue_inode_work %p already queued, mask=%lx\n",
1838                      inode, ci->i_work_mask);
1839                 iput(inode);
1840         }
1841 }
1842
1843 static void ceph_do_invalidate_pages(struct inode *inode)
1844 {
1845         struct ceph_inode_info *ci = ceph_inode(inode);
1846         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1847         u32 orig_gen;
1848         int check = 0;
1849
1850         mutex_lock(&ci->i_truncate_mutex);
1851
1852         if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1853                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1854                                     inode, ceph_ino(inode));
1855                 mapping_set_error(inode->i_mapping, -EIO);
1856                 truncate_pagecache(inode, 0);
1857                 mutex_unlock(&ci->i_truncate_mutex);
1858                 goto out;
1859         }
1860
1861         spin_lock(&ci->i_ceph_lock);
1862         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1863              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1864         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1865                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1866                         check = 1;
1867                 spin_unlock(&ci->i_ceph_lock);
1868                 mutex_unlock(&ci->i_truncate_mutex);
1869                 goto out;
1870         }
1871         orig_gen = ci->i_rdcache_gen;
1872         spin_unlock(&ci->i_ceph_lock);
1873
1874         ceph_fscache_invalidate(inode);
1875         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1876                 pr_err("invalidate_pages %p fails\n", inode);
1877         }
1878
1879         spin_lock(&ci->i_ceph_lock);
1880         if (orig_gen == ci->i_rdcache_gen &&
1881             orig_gen == ci->i_rdcache_revoking) {
1882                 dout("invalidate_pages %p gen %d successful\n", inode,
1883                      ci->i_rdcache_gen);
1884                 ci->i_rdcache_revoking--;
1885                 check = 1;
1886         } else {
1887                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1888                      inode, orig_gen, ci->i_rdcache_gen,
1889                      ci->i_rdcache_revoking);
1890                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1891                         check = 1;
1892         }
1893         spin_unlock(&ci->i_ceph_lock);
1894         mutex_unlock(&ci->i_truncate_mutex);
1895 out:
1896         if (check)
1897                 ceph_check_caps(ci, 0, NULL);
1898 }
1899
1900 /*
1901  * Make sure any pending truncation is applied before doing anything
1902  * that may depend on it.
1903  */
1904 void __ceph_do_pending_vmtruncate(struct inode *inode)
1905 {
1906         struct ceph_inode_info *ci = ceph_inode(inode);
1907         u64 to;
1908         int wrbuffer_refs, finish = 0;
1909
1910         mutex_lock(&ci->i_truncate_mutex);
1911 retry:
1912         spin_lock(&ci->i_ceph_lock);
1913         if (ci->i_truncate_pending == 0) {
1914                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1915                 spin_unlock(&ci->i_ceph_lock);
1916                 mutex_unlock(&ci->i_truncate_mutex);
1917                 return;
1918         }
1919
1920         /*
1921          * make sure any dirty snapped pages are flushed before we
1922          * possibly truncate them.. so write AND block!
1923          */
1924         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1925                 spin_unlock(&ci->i_ceph_lock);
1926                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1927                      inode);
1928                 filemap_write_and_wait_range(&inode->i_data, 0,
1929                                              inode->i_sb->s_maxbytes);
1930                 goto retry;
1931         }
1932
1933         /* there should be no reader or writer */
1934         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1935
1936         to = ci->i_truncate_size;
1937         wrbuffer_refs = ci->i_wrbuffer_ref;
1938         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1939              ci->i_truncate_pending, to);
1940         spin_unlock(&ci->i_ceph_lock);
1941
1942         truncate_pagecache(inode, to);
1943
1944         spin_lock(&ci->i_ceph_lock);
1945         if (to == ci->i_truncate_size) {
1946                 ci->i_truncate_pending = 0;
1947                 finish = 1;
1948         }
1949         spin_unlock(&ci->i_ceph_lock);
1950         if (!finish)
1951                 goto retry;
1952
1953         mutex_unlock(&ci->i_truncate_mutex);
1954
1955         if (wrbuffer_refs == 0)
1956                 ceph_check_caps(ci, 0, NULL);
1957
1958         wake_up_all(&ci->i_cap_wq);
1959 }
1960
1961 static void ceph_inode_work(struct work_struct *work)
1962 {
1963         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1964                                                  i_work);
1965         struct inode *inode = &ci->vfs_inode;
1966
1967         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1968                 dout("writeback %p\n", inode);
1969                 filemap_fdatawrite(&inode->i_data);
1970         }
1971         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1972                 ceph_do_invalidate_pages(inode);
1973
1974         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1975                 __ceph_do_pending_vmtruncate(inode);
1976
1977         if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
1978                 ceph_check_caps(ci, 0, NULL);
1979
1980         if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
1981                 ceph_flush_snaps(ci, NULL);
1982
1983         iput(inode);
1984 }
1985
1986 /*
1987  * symlinks
1988  */
1989 static const struct inode_operations ceph_symlink_iops = {
1990         .get_link = simple_get_link,
1991         .setattr = ceph_setattr,
1992         .getattr = ceph_getattr,
1993         .listxattr = ceph_listxattr,
1994 };
1995
1996 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1997 {
1998         struct ceph_inode_info *ci = ceph_inode(inode);
1999         unsigned int ia_valid = attr->ia_valid;
2000         struct ceph_mds_request *req;
2001         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2002         struct ceph_cap_flush *prealloc_cf;
2003         int issued;
2004         int release = 0, dirtied = 0;
2005         int mask = 0;
2006         int err = 0;
2007         int inode_dirty_flags = 0;
2008         bool lock_snap_rwsem = false;
2009
2010         prealloc_cf = ceph_alloc_cap_flush();
2011         if (!prealloc_cf)
2012                 return -ENOMEM;
2013
2014         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2015                                        USE_AUTH_MDS);
2016         if (IS_ERR(req)) {
2017                 ceph_free_cap_flush(prealloc_cf);
2018                 return PTR_ERR(req);
2019         }
2020
2021         spin_lock(&ci->i_ceph_lock);
2022         issued = __ceph_caps_issued(ci, NULL);
2023
2024         if (!ci->i_head_snapc &&
2025             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2026                 lock_snap_rwsem = true;
2027                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2028                         spin_unlock(&ci->i_ceph_lock);
2029                         down_read(&mdsc->snap_rwsem);
2030                         spin_lock(&ci->i_ceph_lock);
2031                         issued = __ceph_caps_issued(ci, NULL);
2032                 }
2033         }
2034
2035         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2036
2037         if (ia_valid & ATTR_UID) {
2038                 dout("setattr %p uid %d -> %d\n", inode,
2039                      from_kuid(&init_user_ns, inode->i_uid),
2040                      from_kuid(&init_user_ns, attr->ia_uid));
2041                 if (issued & CEPH_CAP_AUTH_EXCL) {
2042                         inode->i_uid = attr->ia_uid;
2043                         dirtied |= CEPH_CAP_AUTH_EXCL;
2044                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2045                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2046                         req->r_args.setattr.uid = cpu_to_le32(
2047                                 from_kuid(&init_user_ns, attr->ia_uid));
2048                         mask |= CEPH_SETATTR_UID;
2049                         release |= CEPH_CAP_AUTH_SHARED;
2050                 }
2051         }
2052         if (ia_valid & ATTR_GID) {
2053                 dout("setattr %p gid %d -> %d\n", inode,
2054                      from_kgid(&init_user_ns, inode->i_gid),
2055                      from_kgid(&init_user_ns, attr->ia_gid));
2056                 if (issued & CEPH_CAP_AUTH_EXCL) {
2057                         inode->i_gid = attr->ia_gid;
2058                         dirtied |= CEPH_CAP_AUTH_EXCL;
2059                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2060                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2061                         req->r_args.setattr.gid = cpu_to_le32(
2062                                 from_kgid(&init_user_ns, attr->ia_gid));
2063                         mask |= CEPH_SETATTR_GID;
2064                         release |= CEPH_CAP_AUTH_SHARED;
2065                 }
2066         }
2067         if (ia_valid & ATTR_MODE) {
2068                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2069                      attr->ia_mode);
2070                 if (issued & CEPH_CAP_AUTH_EXCL) {
2071                         inode->i_mode = attr->ia_mode;
2072                         dirtied |= CEPH_CAP_AUTH_EXCL;
2073                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2074                            attr->ia_mode != inode->i_mode) {
2075                         inode->i_mode = attr->ia_mode;
2076                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2077                         mask |= CEPH_SETATTR_MODE;
2078                         release |= CEPH_CAP_AUTH_SHARED;
2079                 }
2080         }
2081
2082         if (ia_valid & ATTR_ATIME) {
2083                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2084                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2085                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2086                 if (issued & CEPH_CAP_FILE_EXCL) {
2087                         ci->i_time_warp_seq++;
2088                         inode->i_atime = attr->ia_atime;
2089                         dirtied |= CEPH_CAP_FILE_EXCL;
2090                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2091                            timespec64_compare(&inode->i_atime,
2092                                             &attr->ia_atime) < 0) {
2093                         inode->i_atime = attr->ia_atime;
2094                         dirtied |= CEPH_CAP_FILE_WR;
2095                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2096                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2097                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2098                                                &attr->ia_atime);
2099                         mask |= CEPH_SETATTR_ATIME;
2100                         release |= CEPH_CAP_FILE_SHARED |
2101                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2102                 }
2103         }
2104         if (ia_valid & ATTR_SIZE) {
2105                 loff_t isize = i_size_read(inode);
2106
2107                 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2108                 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size > isize) {
2109                         i_size_write(inode, attr->ia_size);
2110                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2111                         ci->i_reported_size = attr->ia_size;
2112                         dirtied |= CEPH_CAP_FILE_EXCL;
2113                         ia_valid |= ATTR_MTIME;
2114                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2115                            attr->ia_size != isize) {
2116                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2117                         req->r_args.setattr.old_size = cpu_to_le64(isize);
2118                         mask |= CEPH_SETATTR_SIZE;
2119                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2120                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2121                 }
2122         }
2123         if (ia_valid & ATTR_MTIME) {
2124                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2125                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2126                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2127                 if (issued & CEPH_CAP_FILE_EXCL) {
2128                         ci->i_time_warp_seq++;
2129                         inode->i_mtime = attr->ia_mtime;
2130                         dirtied |= CEPH_CAP_FILE_EXCL;
2131                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2132                            timespec64_compare(&inode->i_mtime,
2133                                             &attr->ia_mtime) < 0) {
2134                         inode->i_mtime = attr->ia_mtime;
2135                         dirtied |= CEPH_CAP_FILE_WR;
2136                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2137                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2138                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2139                                                &attr->ia_mtime);
2140                         mask |= CEPH_SETATTR_MTIME;
2141                         release |= CEPH_CAP_FILE_SHARED |
2142                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2143                 }
2144         }
2145
2146         /* these do nothing */
2147         if (ia_valid & ATTR_CTIME) {
2148                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2149                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2150                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2151                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2152                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2153                      only ? "ctime only" : "ignored");
2154                 if (only) {
2155                         /*
2156                          * if kernel wants to dirty ctime but nothing else,
2157                          * we need to choose a cap to dirty under, or do
2158                          * a almost-no-op setattr
2159                          */
2160                         if (issued & CEPH_CAP_AUTH_EXCL)
2161                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2162                         else if (issued & CEPH_CAP_FILE_EXCL)
2163                                 dirtied |= CEPH_CAP_FILE_EXCL;
2164                         else if (issued & CEPH_CAP_XATTR_EXCL)
2165                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2166                         else
2167                                 mask |= CEPH_SETATTR_CTIME;
2168                 }
2169         }
2170         if (ia_valid & ATTR_FILE)
2171                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2172
2173         if (dirtied) {
2174                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2175                                                            &prealloc_cf);
2176                 inode->i_ctime = attr->ia_ctime;
2177         }
2178
2179         release &= issued;
2180         spin_unlock(&ci->i_ceph_lock);
2181         if (lock_snap_rwsem)
2182                 up_read(&mdsc->snap_rwsem);
2183
2184         if (inode_dirty_flags)
2185                 __mark_inode_dirty(inode, inode_dirty_flags);
2186
2187
2188         if (mask) {
2189                 req->r_inode = inode;
2190                 ihold(inode);
2191                 req->r_inode_drop = release;
2192                 req->r_args.setattr.mask = cpu_to_le32(mask);
2193                 req->r_num_caps = 1;
2194                 req->r_stamp = attr->ia_ctime;
2195                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2196         }
2197         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2198              ceph_cap_string(dirtied), mask);
2199
2200         ceph_mdsc_put_request(req);
2201         ceph_free_cap_flush(prealloc_cf);
2202
2203         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2204                 __ceph_do_pending_vmtruncate(inode);
2205
2206         return err;
2207 }
2208
2209 /*
2210  * setattr
2211  */
2212 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2213                  struct iattr *attr)
2214 {
2215         struct inode *inode = d_inode(dentry);
2216         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2217         int err;
2218
2219         if (ceph_snap(inode) != CEPH_NOSNAP)
2220                 return -EROFS;
2221
2222         err = setattr_prepare(&init_user_ns, dentry, attr);
2223         if (err != 0)
2224                 return err;
2225
2226         if ((attr->ia_valid & ATTR_SIZE) &&
2227             attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2228                 return -EFBIG;
2229
2230         if ((attr->ia_valid & ATTR_SIZE) &&
2231             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2232                 return -EDQUOT;
2233
2234         err = __ceph_setattr(inode, attr);
2235
2236         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2237                 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode);
2238
2239         return err;
2240 }
2241
2242 /*
2243  * Verify that we have a lease on the given mask.  If not,
2244  * do a getattr against an mds.
2245  */
2246 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2247                       int mask, bool force)
2248 {
2249         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2250         struct ceph_mds_client *mdsc = fsc->mdsc;
2251         struct ceph_mds_request *req;
2252         int mode;
2253         int err;
2254
2255         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2256                 dout("do_getattr inode %p SNAPDIR\n", inode);
2257                 return 0;
2258         }
2259
2260         dout("do_getattr inode %p mask %s mode 0%o\n",
2261              inode, ceph_cap_string(mask), inode->i_mode);
2262         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2263                         return 0;
2264
2265         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2266         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2267         if (IS_ERR(req))
2268                 return PTR_ERR(req);
2269         req->r_inode = inode;
2270         ihold(inode);
2271         req->r_num_caps = 1;
2272         req->r_args.getattr.mask = cpu_to_le32(mask);
2273         req->r_locked_page = locked_page;
2274         err = ceph_mdsc_do_request(mdsc, NULL, req);
2275         if (locked_page && err == 0) {
2276                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2277                 if (inline_version == 0) {
2278                         /* the reply is supposed to contain inline data */
2279                         err = -EINVAL;
2280                 } else if (inline_version == CEPH_INLINE_NONE) {
2281                         err = -ENODATA;
2282                 } else {
2283                         err = req->r_reply_info.targeti.inline_len;
2284                 }
2285         }
2286         ceph_mdsc_put_request(req);
2287         dout("do_getattr result=%d\n", err);
2288         return err;
2289 }
2290
2291
2292 /*
2293  * Check inode permissions.  We verify we have a valid value for
2294  * the AUTH cap, then call the generic handler.
2295  */
2296 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2297                     int mask)
2298 {
2299         int err;
2300
2301         if (mask & MAY_NOT_BLOCK)
2302                 return -ECHILD;
2303
2304         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2305
2306         if (!err)
2307                 err = generic_permission(&init_user_ns, inode, mask);
2308         return err;
2309 }
2310
2311 /* Craft a mask of needed caps given a set of requested statx attrs. */
2312 static int statx_to_caps(u32 want, umode_t mode)
2313 {
2314         int mask = 0;
2315
2316         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2317                 mask |= CEPH_CAP_AUTH_SHARED;
2318
2319         if (want & (STATX_NLINK|STATX_CTIME)) {
2320                 /*
2321                  * The link count for directories depends on inode->i_subdirs,
2322                  * and that is only updated when Fs caps are held.
2323                  */
2324                 if (S_ISDIR(mode))
2325                         mask |= CEPH_CAP_FILE_SHARED;
2326                 else
2327                         mask |= CEPH_CAP_LINK_SHARED;
2328         }
2329
2330         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2331                     STATX_BLOCKS))
2332                 mask |= CEPH_CAP_FILE_SHARED;
2333
2334         if (want & (STATX_CTIME))
2335                 mask |= CEPH_CAP_XATTR_SHARED;
2336
2337         return mask;
2338 }
2339
2340 /*
2341  * Get all the attributes. If we have sufficient caps for the requested attrs,
2342  * then we can avoid talking to the MDS at all.
2343  */
2344 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2345                  struct kstat *stat, u32 request_mask, unsigned int flags)
2346 {
2347         struct inode *inode = d_inode(path->dentry);
2348         struct ceph_inode_info *ci = ceph_inode(inode);
2349         u32 valid_mask = STATX_BASIC_STATS;
2350         int err = 0;
2351
2352         /* Skip the getattr altogether if we're asked not to sync */
2353         if (!(flags & AT_STATX_DONT_SYNC)) {
2354                 err = ceph_do_getattr(inode,
2355                                 statx_to_caps(request_mask, inode->i_mode),
2356                                 flags & AT_STATX_FORCE_SYNC);
2357                 if (err)
2358                         return err;
2359         }
2360
2361         generic_fillattr(&init_user_ns, inode, stat);
2362         stat->ino = ceph_present_inode(inode);
2363
2364         /*
2365          * btime on newly-allocated inodes is 0, so if this is still set to
2366          * that, then assume that it's not valid.
2367          */
2368         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2369                 stat->btime = ci->i_btime;
2370                 valid_mask |= STATX_BTIME;
2371         }
2372
2373         if (ceph_snap(inode) == CEPH_NOSNAP)
2374                 stat->dev = inode->i_sb->s_dev;
2375         else
2376                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2377
2378         if (S_ISDIR(inode->i_mode)) {
2379                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2380                                         RBYTES))
2381                         stat->size = ci->i_rbytes;
2382                 else
2383                         stat->size = ci->i_files + ci->i_subdirs;
2384                 stat->blocks = 0;
2385                 stat->blksize = 65536;
2386                 /*
2387                  * Some applications rely on the number of st_nlink
2388                  * value on directories to be either 0 (if unlinked)
2389                  * or 2 + number of subdirectories.
2390                  */
2391                 if (stat->nlink == 1)
2392                         /* '.' + '..' + subdirs */
2393                         stat->nlink = 1 + 1 + ci->i_subdirs;
2394         }
2395
2396         stat->result_mask = request_mask & valid_mask;
2397         return err;
2398 }