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         /* Set parameters for the netfs library */
 457         netfs_i_context_init(&ci->vfs_inode, &ceph_netfs_ops);
 458
 459         spin_lock_init(&ci->i_ceph_lock);
 460
 461         ci->i_version = 0;
 462         ci->i_inline_version = 0;
 463         ci->i_time_warp_seq = 0;
 464         ci->i_ceph_flags = 0;
 465         atomic64_set(&ci->i_ordered_count, 1);
 466         atomic64_set(&ci->i_release_count, 1);
 467         atomic64_set(&ci->i_complete_seq[0], 0);
 468         atomic64_set(&ci->i_complete_seq[1], 0);
 469         ci->i_symlink = NULL;
 470
 471         ci->i_max_bytes = 0;
 472         ci->i_max_files = 0;
 473
 474         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 475         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 476         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 477
 478         ci->i_fragtree = RB_ROOT;
 479         mutex_init(&ci->i_fragtree_mutex);
 480
 481         ci->i_xattrs.blob = NULL;
 482         ci->i_xattrs.prealloc_blob = NULL;
 483         ci->i_xattrs.dirty = false;
 484         ci->i_xattrs.index = RB_ROOT;
 485         ci->i_xattrs.count = 0;
 486         ci->i_xattrs.names_size = 0;
 487         ci->i_xattrs.vals_size = 0;
 488         ci->i_xattrs.version = 0;
 489         ci->i_xattrs.index_version = 0;
 490
 491         ci->i_caps = RB_ROOT;
 492         ci->i_auth_cap = NULL;
 493         ci->i_dirty_caps = 0;
 494         ci->i_flushing_caps = 0;
 495         INIT_LIST_HEAD(&ci->i_dirty_item);
 496         INIT_LIST_HEAD(&ci->i_flushing_item);
 497         ci->i_prealloc_cap_flush = NULL;
 498         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 499         init_waitqueue_head(&ci->i_cap_wq);
 500         ci->i_hold_caps_max = 0;
 501         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 502         INIT_LIST_HEAD(&ci->i_cap_snaps);
 503         ci->i_head_snapc = NULL;
 504         ci->i_snap_caps = 0;
 505
 506         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 507         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 508                 ci->i_nr_by_mode[i] = 0;
 509
 510         mutex_init(&ci->i_truncate_mutex);
 511         ci->i_truncate_seq = 0;
 512         ci->i_truncate_size = 0;
 513         ci->i_truncate_pending = 0;
 514
 515         ci->i_max_size = 0;
 516         ci->i_reported_size = 0;
 517         ci->i_wanted_max_size = 0;
 518         ci->i_requested_max_size = 0;
 519
 520         ci->i_pin_ref = 0;
 521         ci->i_rd_ref = 0;
 522         ci->i_rdcache_ref = 0;
 523         ci->i_wr_ref = 0;
 524         ci->i_wb_ref = 0;
 525         ci->i_fx_ref = 0;
 526         ci->i_wrbuffer_ref = 0;
 527         ci->i_wrbuffer_ref_head = 0;
 528         atomic_set(&ci->i_filelock_ref, 0);
 529         atomic_set(&ci->i_shared_gen, 1);
 530         ci->i_rdcache_gen = 0;
 531         ci->i_rdcache_revoking = 0;
 532
 533         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 534         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 535         spin_lock_init(&ci->i_unsafe_lock);
 536
 537         ci->i_snap_realm = NULL;
 538         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 539         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 540
 541         INIT_WORK(&ci->i_work, ceph_inode_work);
 542         ci->i_work_mask = 0;
 543         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 544         return &ci->vfs_inode;
 545 }
 546
 547 void ceph_free_inode(struct inode *inode)
 548 {
 549         struct ceph_inode_info *ci = ceph_inode(inode);
 550
 551         kfree(ci->i_symlink);
 552         kmem_cache_free(ceph_inode_cachep, ci);
 553 }
 554
 555 void ceph_evict_inode(struct inode *inode)
 556 {
 557         struct ceph_inode_info *ci = ceph_inode(inode);
 558         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 559         struct ceph_inode_frag *frag;
 560         struct rb_node *n;
 561
 562         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 563
 564         percpu_counter_dec(&mdsc->metric.total_inodes);
 565
 566         truncate_inode_pages_final(&inode->i_data);
 567         if (inode->i_state & I_PINNING_FSCACHE_WB)
 568                 ceph_fscache_unuse_cookie(inode, true);
 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                 /*
 640                  * If we're expanding, then we should be able to just update
 641                  * the existing cookie.
 642                  */
 643                 if (size > isize)
 644                         ceph_fscache_update(inode);
 645                 ci->i_reported_size = size;
 646                 if (truncate_seq != ci->i_truncate_seq) {
 647                         dout("truncate_seq %u -> %u\n",
 648                              ci->i_truncate_seq, truncate_seq);
 649                         ci->i_truncate_seq = truncate_seq;
 650
 651                         /* the MDS should have revoked these caps */
 652                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 653                                                CEPH_CAP_FILE_RD |
 654                                                CEPH_CAP_FILE_WR |
 655                                                CEPH_CAP_FILE_LAZYIO));
 656                         /*
 657                          * If we hold relevant caps, or in the case where we're
 658                          * not the only client referencing this file and we
 659                          * don't hold those caps, then we need to check whether
 660                          * the file is either opened or mmaped
 661                          */
 662                         if ((issued & (CEPH_CAP_FILE_CACHE|
 663                                        CEPH_CAP_FILE_BUFFER)) ||
 664                             mapping_mapped(inode->i_mapping) ||
 665                             __ceph_is_file_opened(ci)) {
 666                                 ci->i_truncate_pending++;
 667                                 queue_trunc = 1;
 668                         }
 669                 }
 670         }
 671         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 672             ci->i_truncate_size != truncate_size) {
 673                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 674                      truncate_size);
 675                 ci->i_truncate_size = truncate_size;
 676         }
 677         return queue_trunc;
 678 }
 679
 680 void ceph_fill_file_time(struct inode *inode, int issued,
 681                          u64 time_warp_seq, struct timespec64 *ctime,
 682                          struct timespec64 *mtime, struct timespec64 *atime)
 683 {
 684         struct ceph_inode_info *ci = ceph_inode(inode);
 685         int warn = 0;
 686
 687         if (issued & (CEPH_CAP_FILE_EXCL|
 688                       CEPH_CAP_FILE_WR|
 689                       CEPH_CAP_FILE_BUFFER|
 690                       CEPH_CAP_AUTH_EXCL|
 691                       CEPH_CAP_XATTR_EXCL)) {
 692                 if (ci->i_version == 0 ||
 693                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 694                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 695                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 696                              ctime->tv_sec, ctime->tv_nsec);
 697                         inode->i_ctime = *ctime;
 698                 }
 699                 if (ci->i_version == 0 ||
 700                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 701                         /* the MDS did a utimes() */
 702                         dout("mtime %lld.%09ld -> %lld.%09ld "
 703                              "tw %d -> %d\n",
 704                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 705                              mtime->tv_sec, mtime->tv_nsec,
 706                              ci->i_time_warp_seq, (int)time_warp_seq);
 707
 708                         inode->i_mtime = *mtime;
 709                         inode->i_atime = *atime;
 710                         ci->i_time_warp_seq = time_warp_seq;
 711                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 712                         /* nobody did utimes(); take the max */
 713                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 714                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 715                                      inode->i_mtime.tv_sec,
 716                                      inode->i_mtime.tv_nsec,
 717                                      mtime->tv_sec, mtime->tv_nsec);
 718                                 inode->i_mtime = *mtime;
 719                         }
 720                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 721                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 722                                      inode->i_atime.tv_sec,
 723                                      inode->i_atime.tv_nsec,
 724                                      atime->tv_sec, atime->tv_nsec);
 725                                 inode->i_atime = *atime;
 726                         }
 727                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 728                         /* we did a utimes(); ignore mds values */
 729                 } else {
 730                         warn = 1;
 731                 }
 732         } else {
 733                 /* we have no write|excl caps; whatever the MDS says is true */
 734                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 735                         inode->i_ctime = *ctime;
 736                         inode->i_mtime = *mtime;
 737                         inode->i_atime = *atime;
 738                         ci->i_time_warp_seq = time_warp_seq;
 739                 } else {
 740                         warn = 1;
 741                 }
 742         }
 743         if (warn) /* time_warp_seq shouldn't go backwards */
 744                 dout("%p mds time_warp_seq %llu < %u\n",
 745                      inode, time_warp_seq, ci->i_time_warp_seq);
 746 }
 747
 748 /*
 749  * Populate an inode based on info from mds.  May be called on new or
 750  * existing inodes.
 751  */
 752 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 753                     struct ceph_mds_reply_info_in *iinfo,
 754                     struct ceph_mds_reply_dirfrag *dirinfo,
 755                     struct ceph_mds_session *session, int cap_fmode,
 756                     struct ceph_cap_reservation *caps_reservation)
 757 {
 758         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 759         struct ceph_mds_reply_inode *info = iinfo->in;
 760         struct ceph_inode_info *ci = ceph_inode(inode);
 761         int issued, new_issued, info_caps;
 762         struct timespec64 mtime, atime, ctime;
 763         struct ceph_buffer *xattr_blob = NULL;
 764         struct ceph_buffer *old_blob = NULL;
 765         struct ceph_string *pool_ns = NULL;
 766         struct ceph_cap *new_cap = NULL;
 767         int err = 0;
 768         bool wake = false;
 769         bool queue_trunc = false;
 770         bool new_version = false;
 771         bool fill_inline = false;
 772         umode_t mode = le32_to_cpu(info->mode);
 773         dev_t rdev = le32_to_cpu(info->rdev);
 774
 775         lockdep_assert_held(&mdsc->snap_rwsem);
 776
 777         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 778              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 779              ci->i_version);
 780
 781         /* Once I_NEW is cleared, we can't change type or dev numbers */
 782         if (inode->i_state & I_NEW) {
 783                 inode->i_mode = mode;
 784         } else {
 785                 if (inode_wrong_type(inode, mode)) {
 786                         pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
 787                                      ceph_vinop(inode), inode->i_mode, mode);
 788                         return -ESTALE;
 789                 }
 790
 791                 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
 792                         pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
 793                                      ceph_vinop(inode), MAJOR(inode->i_rdev),
 794                                      MINOR(inode->i_rdev), MAJOR(rdev),
 795                                      MINOR(rdev));
 796                         return -ESTALE;
 797                 }
 798         }
 799
 800         info_caps = le32_to_cpu(info->cap.caps);
 801
 802         /* prealloc new cap struct */
 803         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 804                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 805                 if (!new_cap)
 806                         return -ENOMEM;
 807         }
 808
 809         /*
 810          * prealloc xattr data, if it looks like we'll need it.  only
 811          * if len > 4 (meaning there are actually xattrs; the first 4
 812          * bytes are the xattr count).
 813          */
 814         if (iinfo->xattr_len > 4) {
 815                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 816                 if (!xattr_blob)
 817                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 818                                iinfo->xattr_len);
 819         }
 820
 821         if (iinfo->pool_ns_len > 0)
 822                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 823                                                      iinfo->pool_ns_len);
 824
 825         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 826                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 827
 828         spin_lock(&ci->i_ceph_lock);
 829
 830         /*
 831          * provided version will be odd if inode value is projected,
 832          * even if stable.  skip the update if we have newer stable
 833          * info (ours>=theirs, e.g. due to racing mds replies), unless
 834          * we are getting projected (unstable) info (in which case the
 835          * version is odd, and we want ours>theirs).
 836          *   us   them
 837          *   2    2     skip
 838          *   3    2     skip
 839          *   3    3     update
 840          */
 841         if (ci->i_version == 0 ||
 842             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 843              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 844                 new_version = true;
 845
 846         /* Update change_attribute */
 847         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 848
 849         __ceph_caps_issued(ci, &issued);
 850         issued |= __ceph_caps_dirty(ci);
 851         new_issued = ~issued & info_caps;
 852
 853         /* directories have fl_stripe_unit set to zero */
 854         if (le32_to_cpu(info->layout.fl_stripe_unit))
 855                 inode->i_blkbits =
 856                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 857         else
 858                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 859
 860         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 861
 862         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 863             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 864                 inode->i_mode = mode;
 865                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 866                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 867                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 868                      from_kuid(&init_user_ns, inode->i_uid),
 869                      from_kgid(&init_user_ns, inode->i_gid));
 870                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 871                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 872         }
 873
 874         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 875             (issued & CEPH_CAP_LINK_EXCL) == 0)
 876                 set_nlink(inode, le32_to_cpu(info->nlink));
 877
 878         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 879                 /* be careful with mtime, atime, size */
 880                 ceph_decode_timespec64(&atime, &info->atime);
 881                 ceph_decode_timespec64(&mtime, &info->mtime);
 882                 ceph_decode_timespec64(&ctime, &info->ctime);
 883                 ceph_fill_file_time(inode, issued,
 884                                 le32_to_cpu(info->time_warp_seq),
 885                                 &ctime, &mtime, &atime);
 886         }
 887
 888         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 889                 ci->i_files = le64_to_cpu(info->files);
 890                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 891         }
 892
 893         if (new_version ||
 894             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 895                 s64 old_pool = ci->i_layout.pool_id;
 896                 struct ceph_string *old_ns;
 897
 898                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 899                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 900                                         lockdep_is_held(&ci->i_ceph_lock));
 901                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 902
 903                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 904                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 905
 906                 pool_ns = old_ns;
 907
 908                 queue_trunc = ceph_fill_file_size(inode, issued,
 909                                         le32_to_cpu(info->truncate_seq),
 910                                         le64_to_cpu(info->truncate_size),
 911                                         le64_to_cpu(info->size));
 912                 /* only update max_size on auth cap */
 913                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 914                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 915                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 916                                         le64_to_cpu(info->max_size));
 917                         ci->i_max_size = le64_to_cpu(info->max_size);
 918                 }
 919         }
 920
 921         /* layout and rstat are not tracked by capability, update them if
 922          * the inode info is from auth mds */
 923         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 924                 if (S_ISDIR(inode->i_mode)) {
 925                         ci->i_dir_layout = iinfo->dir_layout;
 926                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 927                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 928                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 929                         ci->i_dir_pin = iinfo->dir_pin;
 930                         ci->i_rsnaps = iinfo->rsnaps;
 931                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 932                 }
 933         }
 934
 935         /* xattrs */
 936         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 937         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 938             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 939                 if (ci->i_xattrs.blob)
 940                         old_blob = ci->i_xattrs.blob;
 941                 ci->i_xattrs.blob = xattr_blob;
 942                 if (xattr_blob)
 943                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 944                                iinfo->xattr_data, iinfo->xattr_len);
 945                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 946                 ceph_forget_all_cached_acls(inode);
 947                 ceph_security_invalidate_secctx(inode);
 948                 xattr_blob = NULL;
 949         }
 950
 951         /* finally update i_version */
 952         if (le64_to_cpu(info->version) > ci->i_version)
 953                 ci->i_version = le64_to_cpu(info->version);
 954
 955         inode->i_mapping->a_ops = &ceph_aops;
 956
 957         switch (inode->i_mode & S_IFMT) {
 958         case S_IFIFO:
 959         case S_IFBLK:
 960         case S_IFCHR:
 961         case S_IFSOCK:
 962                 inode->i_blkbits = PAGE_SHIFT;
 963                 init_special_inode(inode, inode->i_mode, rdev);
 964                 inode->i_op = &ceph_file_iops;
 965                 break;
 966         case S_IFREG:
 967                 inode->i_op = &ceph_file_iops;
 968                 inode->i_fop = &ceph_file_fops;
 969                 break;
 970         case S_IFLNK:
 971                 inode->i_op = &ceph_symlink_iops;
 972                 if (!ci->i_symlink) {
 973                         u32 symlen = iinfo->symlink_len;
 974                         char *sym;
 975
 976                         spin_unlock(&ci->i_ceph_lock);
 977
 978                         if (symlen != i_size_read(inode)) {
 979                                 pr_err("%s %llx.%llx BAD symlink "
 980                                         "size %lld\n", __func__,
 981                                         ceph_vinop(inode),
 982                                         i_size_read(inode));
 983                                 i_size_write(inode, symlen);
 984                                 inode->i_blocks = calc_inode_blocks(symlen);
 985                         }
 986
 987                         err = -ENOMEM;
 988                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 989                         if (!sym)
 990                                 goto out;
 991
 992                         spin_lock(&ci->i_ceph_lock);
 993                         if (!ci->i_symlink)
 994                                 ci->i_symlink = sym;
 995                         else
 996                                 kfree(sym); /* lost a race */
 997                 }
 998                 inode->i_link = ci->i_symlink;
 999                 break;
1000         case S_IFDIR:
1001                 inode->i_op = &ceph_dir_iops;
1002                 inode->i_fop = &ceph_dir_fops;
1003                 break;
1004         default:
1005                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1006                        ceph_vinop(inode), inode->i_mode);
1007         }
1008
1009         /* were we issued a capability? */
1010         if (info_caps) {
1011                 if (ceph_snap(inode) == CEPH_NOSNAP) {
1012                         ceph_add_cap(inode, session,
1013                                      le64_to_cpu(info->cap.cap_id),
1014                                      info_caps,
1015                                      le32_to_cpu(info->cap.wanted),
1016                                      le32_to_cpu(info->cap.seq),
1017                                      le32_to_cpu(info->cap.mseq),
1018                                      le64_to_cpu(info->cap.realm),
1019                                      info->cap.flags, &new_cap);
1020
1021                         /* set dir completion flag? */
1022                         if (S_ISDIR(inode->i_mode) &&
1023                             ci->i_files == 0 && ci->i_subdirs == 0 &&
1024                             (info_caps & CEPH_CAP_FILE_SHARED) &&
1025                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1026                             !__ceph_dir_is_complete(ci)) {
1027                                 dout(" marking %p complete (empty)\n", inode);
1028                                 i_size_write(inode, 0);
1029                                 __ceph_dir_set_complete(ci,
1030                                         atomic64_read(&ci->i_release_count),
1031                                         atomic64_read(&ci->i_ordered_count));
1032                         }
1033
1034                         wake = true;
1035                 } else {
1036                         dout(" %p got snap_caps %s\n", inode,
1037                              ceph_cap_string(info_caps));
1038                         ci->i_snap_caps |= info_caps;
1039                 }
1040         }
1041
1042         if (iinfo->inline_version > 0 &&
1043             iinfo->inline_version >= ci->i_inline_version) {
1044                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1045                 ci->i_inline_version = iinfo->inline_version;
1046                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1047                     (locked_page || (info_caps & cache_caps)))
1048                         fill_inline = true;
1049         }
1050
1051         if (cap_fmode >= 0) {
1052                 if (!info_caps)
1053                         pr_warn("mds issued no caps on %llx.%llx\n",
1054                                 ceph_vinop(inode));
1055                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1056         }
1057
1058         spin_unlock(&ci->i_ceph_lock);
1059
1060         ceph_fscache_register_inode_cookie(inode);
1061
1062         if (fill_inline)
1063                 ceph_fill_inline_data(inode, locked_page,
1064                                       iinfo->inline_data, iinfo->inline_len);
1065
1066         if (wake)
1067                 wake_up_all(&ci->i_cap_wq);
1068
1069         /* queue truncate if we saw i_size decrease */
1070         if (queue_trunc)
1071                 ceph_queue_vmtruncate(inode);
1072
1073         /* populate frag tree */
1074         if (S_ISDIR(inode->i_mode))
1075                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1076
1077         /* update delegation info? */
1078         if (dirinfo)
1079                 ceph_fill_dirfrag(inode, dirinfo);
1080
1081         err = 0;
1082 out:
1083         if (new_cap)
1084                 ceph_put_cap(mdsc, new_cap);
1085         ceph_buffer_put(old_blob);
1086         ceph_buffer_put(xattr_blob);
1087         ceph_put_string(pool_ns);
1088         return err;
1089 }
1090
1091 /*
1092  * caller should hold session s_mutex and dentry->d_lock.
1093  */
1094 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1095                                   struct ceph_mds_reply_lease *lease,
1096                                   struct ceph_mds_session *session,
1097                                   unsigned long from_time,
1098                                   struct ceph_mds_session **old_lease_session)
1099 {
1100         struct ceph_dentry_info *di = ceph_dentry(dentry);
1101         unsigned mask = le16_to_cpu(lease->mask);
1102         long unsigned duration = le32_to_cpu(lease->duration_ms);
1103         long unsigned ttl = from_time + (duration * HZ) / 1000;
1104         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1105
1106         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1107              dentry, duration, ttl);
1108
1109         /* only track leases on regular dentries */
1110         if (ceph_snap(dir) != CEPH_NOSNAP)
1111                 return;
1112
1113         if (mask & CEPH_LEASE_PRIMARY_LINK)
1114                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1115         else
1116                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1117
1118         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1119         if (!(mask & CEPH_LEASE_VALID)) {
1120                 __ceph_dentry_dir_lease_touch(di);
1121                 return;
1122         }
1123
1124         if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1125             time_before(ttl, di->time))
1126                 return;  /* we already have a newer lease. */
1127
1128         if (di->lease_session && di->lease_session != session) {
1129                 *old_lease_session = di->lease_session;
1130                 di->lease_session = NULL;
1131         }
1132
1133         if (!di->lease_session)
1134                 di->lease_session = ceph_get_mds_session(session);
1135         di->lease_gen = atomic_read(&session->s_cap_gen);
1136         di->lease_seq = le32_to_cpu(lease->seq);
1137         di->lease_renew_after = half_ttl;
1138         di->lease_renew_from = 0;
1139         di->time = ttl;
1140
1141         __ceph_dentry_lease_touch(di);
1142 }
1143
1144 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1145                                         struct ceph_mds_reply_lease *lease,
1146                                         struct ceph_mds_session *session,
1147                                         unsigned long from_time)
1148 {
1149         struct ceph_mds_session *old_lease_session = NULL;
1150         spin_lock(&dentry->d_lock);
1151         __update_dentry_lease(dir, dentry, lease, session, from_time,
1152                               &old_lease_session);
1153         spin_unlock(&dentry->d_lock);
1154         ceph_put_mds_session(old_lease_session);
1155 }
1156
1157 /*
1158  * update dentry lease without having parent inode locked
1159  */
1160 static void update_dentry_lease_careful(struct dentry *dentry,
1161                                         struct ceph_mds_reply_lease *lease,
1162                                         struct ceph_mds_session *session,
1163                                         unsigned long from_time,
1164                                         char *dname, u32 dname_len,
1165                                         struct ceph_vino *pdvino,
1166                                         struct ceph_vino *ptvino)
1167
1168 {
1169         struct inode *dir;
1170         struct ceph_mds_session *old_lease_session = NULL;
1171
1172         spin_lock(&dentry->d_lock);
1173         /* make sure dentry's name matches target */
1174         if (dentry->d_name.len != dname_len ||
1175             memcmp(dentry->d_name.name, dname, dname_len))
1176                 goto out_unlock;
1177
1178         dir = d_inode(dentry->d_parent);
1179         /* make sure parent matches dvino */
1180         if (!ceph_ino_compare(dir, pdvino))
1181                 goto out_unlock;
1182
1183         /* make sure dentry's inode matches target. NULL ptvino means that
1184          * we expect a negative dentry */
1185         if (ptvino) {
1186                 if (d_really_is_negative(dentry))
1187                         goto out_unlock;
1188                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1189                         goto out_unlock;
1190         } else {
1191                 if (d_really_is_positive(dentry))
1192                         goto out_unlock;
1193         }
1194
1195         __update_dentry_lease(dir, dentry, lease, session,
1196                               from_time, &old_lease_session);
1197 out_unlock:
1198         spin_unlock(&dentry->d_lock);
1199         ceph_put_mds_session(old_lease_session);
1200 }
1201
1202 /*
1203  * splice a dentry to an inode.
1204  * caller must hold directory i_rwsem for this to be safe.
1205  */
1206 static int splice_dentry(struct dentry **pdn, struct inode *in)
1207 {
1208         struct dentry *dn = *pdn;
1209         struct dentry *realdn;
1210
1211         BUG_ON(d_inode(dn));
1212
1213         if (S_ISDIR(in->i_mode)) {
1214                 /* If inode is directory, d_splice_alias() below will remove
1215                  * 'realdn' from its origin parent. We need to ensure that
1216                  * origin parent's readdir cache will not reference 'realdn'
1217                  */
1218                 realdn = d_find_any_alias(in);
1219                 if (realdn) {
1220                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1221                         spin_lock(&realdn->d_lock);
1222
1223                         realdn->d_op->d_prune(realdn);
1224
1225                         di->time = jiffies;
1226                         di->lease_shared_gen = 0;
1227                         di->offset = 0;
1228
1229                         spin_unlock(&realdn->d_lock);
1230                         dput(realdn);
1231                 }
1232         }
1233
1234         /* dn must be unhashed */
1235         if (!d_unhashed(dn))
1236                 d_drop(dn);
1237         realdn = d_splice_alias(in, dn);
1238         if (IS_ERR(realdn)) {
1239                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1240                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1241                 return PTR_ERR(realdn);
1242         }
1243
1244         if (realdn) {
1245                 dout("dn %p (%d) spliced with %p (%d) "
1246                      "inode %p ino %llx.%llx\n",
1247                      dn, d_count(dn),
1248                      realdn, d_count(realdn),
1249                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1250                 dput(dn);
1251                 *pdn = realdn;
1252         } else {
1253                 BUG_ON(!ceph_dentry(dn));
1254                 dout("dn %p attached to %p ino %llx.%llx\n",
1255                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1256         }
1257         return 0;
1258 }
1259
1260 /*
1261  * Incorporate results into the local cache.  This is either just
1262  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1263  * after a lookup).
1264  *
1265  * A reply may contain
1266  *         a directory inode along with a dentry.
1267  *  and/or a target inode
1268  *
1269  * Called with snap_rwsem (read).
1270  */
1271 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1272 {
1273         struct ceph_mds_session *session = req->r_session;
1274         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1275         struct inode *in = NULL;
1276         struct ceph_vino tvino, dvino;
1277         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1278         int err = 0;
1279
1280         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1281              rinfo->head->is_dentry, rinfo->head->is_target);
1282
1283         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1284                 dout("fill_trace reply is empty!\n");
1285                 if (rinfo->head->result == 0 && req->r_parent)
1286                         ceph_invalidate_dir_request(req);
1287                 return 0;
1288         }
1289
1290         if (rinfo->head->is_dentry) {
1291                 struct inode *dir = req->r_parent;
1292
1293                 if (dir) {
1294                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1295                                               rinfo->dirfrag, session, -1,
1296                                               &req->r_caps_reservation);
1297                         if (err < 0)
1298                                 goto done;
1299                 } else {
1300                         WARN_ON_ONCE(1);
1301                 }
1302
1303                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1304                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1305                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1306                         struct qstr dname;
1307                         struct dentry *dn, *parent;
1308
1309                         BUG_ON(!rinfo->head->is_target);
1310                         BUG_ON(req->r_dentry);
1311
1312                         parent = d_find_any_alias(dir);
1313                         BUG_ON(!parent);
1314
1315                         dname.name = rinfo->dname;
1316                         dname.len = rinfo->dname_len;
1317                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1318                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1319                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1320 retry_lookup:
1321                         dn = d_lookup(parent, &dname);
1322                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1323                              parent, dname.len, dname.name, dn);
1324
1325                         if (!dn) {
1326                                 dn = d_alloc(parent, &dname);
1327                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1328                                      dname.len, dname.name, dn);
1329                                 if (!dn) {
1330                                         dput(parent);
1331                                         err = -ENOMEM;
1332                                         goto done;
1333                                 }
1334                                 err = 0;
1335                         } else if (d_really_is_positive(dn) &&
1336                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1337                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1338                                 dout(" dn %p points to wrong inode %p\n",
1339                                      dn, d_inode(dn));
1340                                 ceph_dir_clear_ordered(dir);
1341                                 d_delete(dn);
1342                                 dput(dn);
1343                                 goto retry_lookup;
1344                         }
1345
1346                         req->r_dentry = dn;
1347                         dput(parent);
1348                 }
1349         }
1350
1351         if (rinfo->head->is_target) {
1352                 /* Should be filled in by handle_reply */
1353                 BUG_ON(!req->r_target_inode);
1354
1355                 in = req->r_target_inode;
1356                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1357                                 NULL, session,
1358                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1359                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1360                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1361                                 &req->r_caps_reservation);
1362                 if (err < 0) {
1363                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1364                                 in, ceph_vinop(in));
1365                         req->r_target_inode = NULL;
1366                         if (in->i_state & I_NEW)
1367                                 discard_new_inode(in);
1368                         else
1369                                 iput(in);
1370                         goto done;
1371                 }
1372                 if (in->i_state & I_NEW)
1373                         unlock_new_inode(in);
1374         }
1375
1376         /*
1377          * ignore null lease/binding on snapdir ENOENT, or else we
1378          * will have trouble splicing in the virtual snapdir later
1379          */
1380         if (rinfo->head->is_dentry &&
1381             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1382             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1383             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1384                                                fsc->mount_options->snapdir_name,
1385                                                req->r_dentry->d_name.len))) {
1386                 /*
1387                  * lookup link rename   : null -> possibly existing inode
1388                  * mknod symlink mkdir  : null -> new inode
1389                  * unlink               : linked -> null
1390                  */
1391                 struct inode *dir = req->r_parent;
1392                 struct dentry *dn = req->r_dentry;
1393                 bool have_dir_cap, have_lease;
1394
1395                 BUG_ON(!dn);
1396                 BUG_ON(!dir);
1397                 BUG_ON(d_inode(dn->d_parent) != dir);
1398
1399                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1400                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1401
1402                 BUG_ON(ceph_ino(dir) != dvino.ino);
1403                 BUG_ON(ceph_snap(dir) != dvino.snap);
1404
1405                 /* do we have a lease on the whole dir? */
1406                 have_dir_cap =
1407                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1408                          CEPH_CAP_FILE_SHARED);
1409
1410                 /* do we have a dn lease? */
1411                 have_lease = have_dir_cap ||
1412                         le32_to_cpu(rinfo->dlease->duration_ms);
1413                 if (!have_lease)
1414                         dout("fill_trace  no dentry lease or dir cap\n");
1415
1416                 /* rename? */
1417                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1418                         struct inode *olddir = req->r_old_dentry_dir;
1419                         BUG_ON(!olddir);
1420
1421                         dout(" src %p '%pd' dst %p '%pd'\n",
1422                              req->r_old_dentry,
1423                              req->r_old_dentry,
1424                              dn, dn);
1425                         dout("fill_trace doing d_move %p -> %p\n",
1426                              req->r_old_dentry, dn);
1427
1428                         /* d_move screws up sibling dentries' offsets */
1429                         ceph_dir_clear_ordered(dir);
1430                         ceph_dir_clear_ordered(olddir);
1431
1432                         d_move(req->r_old_dentry, dn);
1433                         dout(" src %p '%pd' dst %p '%pd'\n",
1434                              req->r_old_dentry,
1435                              req->r_old_dentry,
1436                              dn, dn);
1437
1438                         /* ensure target dentry is invalidated, despite
1439                            rehashing bug in vfs_rename_dir */
1440                         ceph_invalidate_dentry_lease(dn);
1441
1442                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1443                              ceph_dentry(req->r_old_dentry)->offset);
1444
1445                         /* swap r_dentry and r_old_dentry in case that
1446                          * splice_dentry() gets called later. This is safe
1447                          * because no other place will use them */
1448                         req->r_dentry = req->r_old_dentry;
1449                         req->r_old_dentry = dn;
1450                         dn = req->r_dentry;
1451                 }
1452
1453                 /* null dentry? */
1454                 if (!rinfo->head->is_target) {
1455                         dout("fill_trace null dentry\n");
1456                         if (d_really_is_positive(dn)) {
1457                                 dout("d_delete %p\n", dn);
1458                                 ceph_dir_clear_ordered(dir);
1459                                 d_delete(dn);
1460                         } else if (have_lease) {
1461                                 if (d_unhashed(dn))
1462                                         d_add(dn, NULL);
1463                                 update_dentry_lease(dir, dn,
1464                                                     rinfo->dlease, session,
1465                                                     req->r_request_started);
1466                         }
1467                         goto done;
1468                 }
1469
1470                 /* attach proper inode */
1471                 if (d_really_is_negative(dn)) {
1472                         ceph_dir_clear_ordered(dir);
1473                         ihold(in);
1474                         err = splice_dentry(&req->r_dentry, in);
1475                         if (err < 0)
1476                                 goto done;
1477                         dn = req->r_dentry;  /* may have spliced */
1478                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1479                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1480                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1481                              ceph_vinop(in));
1482                         d_invalidate(dn);
1483                         have_lease = false;
1484                 }
1485
1486                 if (have_lease) {
1487                         update_dentry_lease(dir, dn,
1488                                             rinfo->dlease, session,
1489                                             req->r_request_started);
1490                 }
1491                 dout(" final dn %p\n", dn);
1492         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1493                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1494                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1495                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1496                 struct inode *dir = req->r_parent;
1497
1498                 /* fill out a snapdir LOOKUPSNAP dentry */
1499                 BUG_ON(!dir);
1500                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1501                 BUG_ON(!req->r_dentry);
1502                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1503                 ceph_dir_clear_ordered(dir);
1504                 ihold(in);
1505                 err = splice_dentry(&req->r_dentry, in);
1506                 if (err < 0)
1507                         goto done;
1508         } else if (rinfo->head->is_dentry && req->r_dentry) {
1509                 /* parent inode is not locked, be carefull */
1510                 struct ceph_vino *ptvino = NULL;
1511                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1512                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1513                 if (rinfo->head->is_target) {
1514                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1515                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1516                         ptvino = &tvino;
1517                 }
1518                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1519                                             session, req->r_request_started,
1520                                             rinfo->dname, rinfo->dname_len,
1521                                             &dvino, ptvino);
1522         }
1523 done:
1524         dout("fill_trace done err=%d\n", err);
1525         return err;
1526 }
1527
1528 /*
1529  * Prepopulate our cache with readdir results, leases, etc.
1530  */
1531 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1532                                            struct ceph_mds_session *session)
1533 {
1534         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1535         int i, err = 0;
1536
1537         for (i = 0; i < rinfo->dir_nr; i++) {
1538                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1539                 struct ceph_vino vino;
1540                 struct inode *in;
1541                 int rc;
1542
1543                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1544                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1545
1546                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1547                 if (IS_ERR(in)) {
1548                         err = PTR_ERR(in);
1549                         dout("new_inode badness got %d\n", err);
1550                         continue;
1551                 }
1552                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1553                                      -1, &req->r_caps_reservation);
1554                 if (rc < 0) {
1555                         pr_err("ceph_fill_inode badness on %p got %d\n",
1556                                in, rc);
1557                         err = rc;
1558                         if (in->i_state & I_NEW) {
1559                                 ihold(in);
1560                                 discard_new_inode(in);
1561                         }
1562                 } else if (in->i_state & I_NEW) {
1563                         unlock_new_inode(in);
1564                 }
1565
1566                 iput(in);
1567         }
1568
1569         return err;
1570 }
1571
1572 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1573 {
1574         if (ctl->page) {
1575                 kunmap(ctl->page);
1576                 put_page(ctl->page);
1577                 ctl->page = NULL;
1578         }
1579 }
1580
1581 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1582                               struct ceph_readdir_cache_control *ctl,
1583                               struct ceph_mds_request *req)
1584 {
1585         struct ceph_inode_info *ci = ceph_inode(dir);
1586         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1587         unsigned idx = ctl->index % nsize;
1588         pgoff_t pgoff = ctl->index / nsize;
1589
1590         if (!ctl->page || pgoff != page_index(ctl->page)) {
1591                 ceph_readdir_cache_release(ctl);
1592                 if (idx == 0)
1593                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1594                 else
1595                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1596                 if (!ctl->page) {
1597                         ctl->index = -1;
1598                         return idx == 0 ? -ENOMEM : 0;
1599                 }
1600                 /* reading/filling the cache are serialized by
1601                  * i_rwsem, no need to use page lock */
1602                 unlock_page(ctl->page);
1603                 ctl->dentries = kmap(ctl->page);
1604                 if (idx == 0)
1605                         memset(ctl->dentries, 0, PAGE_SIZE);
1606         }
1607
1608         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1609             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1610                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1611                 ctl->dentries[idx] = dn;
1612                 ctl->index++;
1613         } else {
1614                 dout("disable readdir cache\n");
1615                 ctl->index = -1;
1616         }
1617         return 0;
1618 }
1619
1620 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1621                              struct ceph_mds_session *session)
1622 {
1623         struct dentry *parent = req->r_dentry;
1624         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1625         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1626         struct qstr dname;
1627         struct dentry *dn;
1628         struct inode *in;
1629         int err = 0, skipped = 0, ret, i;
1630         u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1631         u32 last_hash = 0;
1632         u32 fpos_offset;
1633         struct ceph_readdir_cache_control cache_ctl = {};
1634
1635         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1636                 return readdir_prepopulate_inodes_only(req, session);
1637
1638         if (rinfo->hash_order) {
1639                 if (req->r_path2) {
1640                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1641                                                   req->r_path2,
1642                                                   strlen(req->r_path2));
1643                         last_hash = ceph_frag_value(last_hash);
1644                 } else if (rinfo->offset_hash) {
1645                         /* mds understands offset_hash */
1646                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1647                         last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1648                 }
1649         }
1650
1651         if (rinfo->dir_dir &&
1652             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1653                 dout("readdir_prepopulate got new frag %x -> %x\n",
1654                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1655                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1656                 if (!rinfo->hash_order)
1657                         req->r_readdir_offset = 2;
1658         }
1659
1660         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1661                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1662                      rinfo->dir_nr, parent);
1663         } else {
1664                 dout("readdir_prepopulate %d items under dn %p\n",
1665                      rinfo->dir_nr, parent);
1666                 if (rinfo->dir_dir)
1667                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1668
1669                 if (ceph_frag_is_leftmost(frag) &&
1670                     req->r_readdir_offset == 2 &&
1671                     !(rinfo->hash_order && last_hash)) {
1672                         /* note dir version at start of readdir so we can
1673                          * tell if any dentries get dropped */
1674                         req->r_dir_release_cnt =
1675                                 atomic64_read(&ci->i_release_count);
1676                         req->r_dir_ordered_cnt =
1677                                 atomic64_read(&ci->i_ordered_count);
1678                         req->r_readdir_cache_idx = 0;
1679                 }
1680         }
1681
1682         cache_ctl.index = req->r_readdir_cache_idx;
1683         fpos_offset = req->r_readdir_offset;
1684
1685         /* FIXME: release caps/leases if error occurs */
1686         for (i = 0; i < rinfo->dir_nr; i++) {
1687                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1688                 struct ceph_vino tvino;
1689
1690                 dname.name = rde->name;
1691                 dname.len = rde->name_len;
1692                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1693
1694                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1695                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1696
1697                 if (rinfo->hash_order) {
1698                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1699                                                  rde->name, rde->name_len);
1700                         hash = ceph_frag_value(hash);
1701                         if (hash != last_hash)
1702                                 fpos_offset = 2;
1703                         last_hash = hash;
1704                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1705                 } else {
1706                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1707                 }
1708
1709 retry_lookup:
1710                 dn = d_lookup(parent, &dname);
1711                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1712                      parent, dname.len, dname.name, dn);
1713
1714                 if (!dn) {
1715                         dn = d_alloc(parent, &dname);
1716                         dout("d_alloc %p '%.*s' = %p\n", parent,
1717                              dname.len, dname.name, dn);
1718                         if (!dn) {
1719                                 dout("d_alloc badness\n");
1720                                 err = -ENOMEM;
1721                                 goto out;
1722                         }
1723                 } else if (d_really_is_positive(dn) &&
1724                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1725                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1726                         struct ceph_dentry_info *di = ceph_dentry(dn);
1727                         dout(" dn %p points to wrong inode %p\n",
1728                              dn, d_inode(dn));
1729
1730                         spin_lock(&dn->d_lock);
1731                         if (di->offset > 0 &&
1732                             di->lease_shared_gen ==
1733                             atomic_read(&ci->i_shared_gen)) {
1734                                 __ceph_dir_clear_ordered(ci);
1735                                 di->offset = 0;
1736                         }
1737                         spin_unlock(&dn->d_lock);
1738
1739                         d_delete(dn);
1740                         dput(dn);
1741                         goto retry_lookup;
1742                 }
1743
1744                 /* inode */
1745                 if (d_really_is_positive(dn)) {
1746                         in = d_inode(dn);
1747                 } else {
1748                         in = ceph_get_inode(parent->d_sb, tvino);
1749                         if (IS_ERR(in)) {
1750                                 dout("new_inode badness\n");
1751                                 d_drop(dn);
1752                                 dput(dn);
1753                                 err = PTR_ERR(in);
1754                                 goto out;
1755                         }
1756                 }
1757
1758                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1759                                       -1, &req->r_caps_reservation);
1760                 if (ret < 0) {
1761                         pr_err("ceph_fill_inode badness on %p\n", in);
1762                         if (d_really_is_negative(dn)) {
1763                                 if (in->i_state & I_NEW) {
1764                                         ihold(in);
1765                                         discard_new_inode(in);
1766                                 }
1767                                 iput(in);
1768                         }
1769                         d_drop(dn);
1770                         err = ret;
1771                         goto next_item;
1772                 }
1773                 if (in->i_state & I_NEW)
1774                         unlock_new_inode(in);
1775
1776                 if (d_really_is_negative(dn)) {
1777                         if (ceph_security_xattr_deadlock(in)) {
1778                                 dout(" skip splicing dn %p to inode %p"
1779                                      " (security xattr deadlock)\n", dn, in);
1780                                 iput(in);
1781                                 skipped++;
1782                                 goto next_item;
1783                         }
1784
1785                         err = splice_dentry(&dn, in);
1786                         if (err < 0)
1787                                 goto next_item;
1788                 }
1789
1790                 ceph_dentry(dn)->offset = rde->offset;
1791
1792                 update_dentry_lease(d_inode(parent), dn,
1793                                     rde->lease, req->r_session,
1794                                     req->r_request_started);
1795
1796                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1797                         ret = fill_readdir_cache(d_inode(parent), dn,
1798                                                  &cache_ctl, req);
1799                         if (ret < 0)
1800                                 err = ret;
1801                 }
1802 next_item:
1803                 dput(dn);
1804         }
1805 out:
1806         if (err == 0 && skipped == 0) {
1807                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1808                 req->r_readdir_cache_idx = cache_ctl.index;
1809         }
1810         ceph_readdir_cache_release(&cache_ctl);
1811         dout("readdir_prepopulate done\n");
1812         return err;
1813 }
1814
1815 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1816 {
1817         struct ceph_inode_info *ci = ceph_inode(inode);
1818         bool ret;
1819
1820         spin_lock(&ci->i_ceph_lock);
1821         dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1822         i_size_write(inode, size);
1823         ceph_fscache_update(inode);
1824         inode->i_blocks = calc_inode_blocks(size);
1825
1826         ret = __ceph_should_report_size(ci);
1827
1828         spin_unlock(&ci->i_ceph_lock);
1829
1830         return ret;
1831 }
1832
1833 void ceph_queue_inode_work(struct inode *inode, int work_bit)
1834 {
1835         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1836         struct ceph_inode_info *ci = ceph_inode(inode);
1837         set_bit(work_bit, &ci->i_work_mask);
1838
1839         ihold(inode);
1840         if (queue_work(fsc->inode_wq, &ci->i_work)) {
1841                 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1842         } else {
1843                 dout("queue_inode_work %p already queued, mask=%lx\n",
1844                      inode, ci->i_work_mask);
1845                 iput(inode);
1846         }
1847 }
1848
1849 static void ceph_do_invalidate_pages(struct inode *inode)
1850 {
1851         struct ceph_inode_info *ci = ceph_inode(inode);
1852         u32 orig_gen;
1853         int check = 0;
1854
1855         ceph_fscache_invalidate(inode, false);
1856
1857         mutex_lock(&ci->i_truncate_mutex);
1858
1859         if (ceph_inode_is_shutdown(inode)) {
1860                 pr_warn_ratelimited("%s: inode %llx.%llx is shut down\n",
1861                                     __func__, ceph_vinop(inode));
1862                 mapping_set_error(inode->i_mapping, -EIO);
1863                 truncate_pagecache(inode, 0);
1864                 mutex_unlock(&ci->i_truncate_mutex);
1865                 goto out;
1866         }
1867
1868         spin_lock(&ci->i_ceph_lock);
1869         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1870              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1871         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1872                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1873                         check = 1;
1874                 spin_unlock(&ci->i_ceph_lock);
1875                 mutex_unlock(&ci->i_truncate_mutex);
1876                 goto out;
1877         }
1878         orig_gen = ci->i_rdcache_gen;
1879         spin_unlock(&ci->i_ceph_lock);
1880
1881         ceph_fscache_invalidate(inode, false);
1882         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1883                 pr_err("invalidate_inode_pages2 %llx.%llx failed\n",
1884                        ceph_vinop(inode));
1885         }
1886
1887         spin_lock(&ci->i_ceph_lock);
1888         if (orig_gen == ci->i_rdcache_gen &&
1889             orig_gen == ci->i_rdcache_revoking) {
1890                 dout("invalidate_pages %p gen %d successful\n", inode,
1891                      ci->i_rdcache_gen);
1892                 ci->i_rdcache_revoking--;
1893                 check = 1;
1894         } else {
1895                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1896                      inode, orig_gen, ci->i_rdcache_gen,
1897                      ci->i_rdcache_revoking);
1898                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1899                         check = 1;
1900         }
1901         spin_unlock(&ci->i_ceph_lock);
1902         mutex_unlock(&ci->i_truncate_mutex);
1903 out:
1904         if (check)
1905                 ceph_check_caps(ci, 0, NULL);
1906 }
1907
1908 /*
1909  * Make sure any pending truncation is applied before doing anything
1910  * that may depend on it.
1911  */
1912 void __ceph_do_pending_vmtruncate(struct inode *inode)
1913 {
1914         struct ceph_inode_info *ci = ceph_inode(inode);
1915         u64 to;
1916         int wrbuffer_refs, finish = 0;
1917
1918         mutex_lock(&ci->i_truncate_mutex);
1919 retry:
1920         spin_lock(&ci->i_ceph_lock);
1921         if (ci->i_truncate_pending == 0) {
1922                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1923                 spin_unlock(&ci->i_ceph_lock);
1924                 mutex_unlock(&ci->i_truncate_mutex);
1925                 return;
1926         }
1927
1928         /*
1929          * make sure any dirty snapped pages are flushed before we
1930          * possibly truncate them.. so write AND block!
1931          */
1932         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1933                 spin_unlock(&ci->i_ceph_lock);
1934                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1935                      inode);
1936                 filemap_write_and_wait_range(&inode->i_data, 0,
1937                                              inode->i_sb->s_maxbytes);
1938                 goto retry;
1939         }
1940
1941         /* there should be no reader or writer */
1942         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1943
1944         to = ci->i_truncate_size;
1945         wrbuffer_refs = ci->i_wrbuffer_ref;
1946         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1947              ci->i_truncate_pending, to);
1948         spin_unlock(&ci->i_ceph_lock);
1949
1950         ceph_fscache_resize(inode, to);
1951         truncate_pagecache(inode, to);
1952
1953         spin_lock(&ci->i_ceph_lock);
1954         if (to == ci->i_truncate_size) {
1955                 ci->i_truncate_pending = 0;
1956                 finish = 1;
1957         }
1958         spin_unlock(&ci->i_ceph_lock);
1959         if (!finish)
1960                 goto retry;
1961
1962         mutex_unlock(&ci->i_truncate_mutex);
1963
1964         if (wrbuffer_refs == 0)
1965                 ceph_check_caps(ci, 0, NULL);
1966
1967         wake_up_all(&ci->i_cap_wq);
1968 }
1969
1970 static void ceph_inode_work(struct work_struct *work)
1971 {
1972         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1973                                                  i_work);
1974         struct inode *inode = &ci->vfs_inode;
1975
1976         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1977                 dout("writeback %p\n", inode);
1978                 filemap_fdatawrite(&inode->i_data);
1979         }
1980         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1981                 ceph_do_invalidate_pages(inode);
1982
1983         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1984                 __ceph_do_pending_vmtruncate(inode);
1985
1986         if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
1987                 ceph_check_caps(ci, 0, NULL);
1988
1989         if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
1990                 ceph_flush_snaps(ci, NULL);
1991
1992         iput(inode);
1993 }
1994
1995 /*
1996  * symlinks
1997  */
1998 static const struct inode_operations ceph_symlink_iops = {
1999         .get_link = simple_get_link,
2000         .setattr = ceph_setattr,
2001         .getattr = ceph_getattr,
2002         .listxattr = ceph_listxattr,
2003 };
2004
2005 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2006 {
2007         struct ceph_inode_info *ci = ceph_inode(inode);
2008         unsigned int ia_valid = attr->ia_valid;
2009         struct ceph_mds_request *req;
2010         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2011         struct ceph_cap_flush *prealloc_cf;
2012         int issued;
2013         int release = 0, dirtied = 0;
2014         int mask = 0;
2015         int err = 0;
2016         int inode_dirty_flags = 0;
2017         bool lock_snap_rwsem = false;
2018
2019         prealloc_cf = ceph_alloc_cap_flush();
2020         if (!prealloc_cf)
2021                 return -ENOMEM;
2022
2023         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2024                                        USE_AUTH_MDS);
2025         if (IS_ERR(req)) {
2026                 ceph_free_cap_flush(prealloc_cf);
2027                 return PTR_ERR(req);
2028         }
2029
2030         spin_lock(&ci->i_ceph_lock);
2031         issued = __ceph_caps_issued(ci, NULL);
2032
2033         if (!ci->i_head_snapc &&
2034             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2035                 lock_snap_rwsem = true;
2036                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2037                         spin_unlock(&ci->i_ceph_lock);
2038                         down_read(&mdsc->snap_rwsem);
2039                         spin_lock(&ci->i_ceph_lock);
2040                         issued = __ceph_caps_issued(ci, NULL);
2041                 }
2042         }
2043
2044         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2045
2046         if (ia_valid & ATTR_UID) {
2047                 dout("setattr %p uid %d -> %d\n", inode,
2048                      from_kuid(&init_user_ns, inode->i_uid),
2049                      from_kuid(&init_user_ns, attr->ia_uid));
2050                 if (issued & CEPH_CAP_AUTH_EXCL) {
2051                         inode->i_uid = attr->ia_uid;
2052                         dirtied |= CEPH_CAP_AUTH_EXCL;
2053                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2054                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2055                         req->r_args.setattr.uid = cpu_to_le32(
2056                                 from_kuid(&init_user_ns, attr->ia_uid));
2057                         mask |= CEPH_SETATTR_UID;
2058                         release |= CEPH_CAP_AUTH_SHARED;
2059                 }
2060         }
2061         if (ia_valid & ATTR_GID) {
2062                 dout("setattr %p gid %d -> %d\n", inode,
2063                      from_kgid(&init_user_ns, inode->i_gid),
2064                      from_kgid(&init_user_ns, attr->ia_gid));
2065                 if (issued & CEPH_CAP_AUTH_EXCL) {
2066                         inode->i_gid = attr->ia_gid;
2067                         dirtied |= CEPH_CAP_AUTH_EXCL;
2068                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2069                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2070                         req->r_args.setattr.gid = cpu_to_le32(
2071                                 from_kgid(&init_user_ns, attr->ia_gid));
2072                         mask |= CEPH_SETATTR_GID;
2073                         release |= CEPH_CAP_AUTH_SHARED;
2074                 }
2075         }
2076         if (ia_valid & ATTR_MODE) {
2077                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2078                      attr->ia_mode);
2079                 if (issued & CEPH_CAP_AUTH_EXCL) {
2080                         inode->i_mode = attr->ia_mode;
2081                         dirtied |= CEPH_CAP_AUTH_EXCL;
2082                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2083                            attr->ia_mode != inode->i_mode) {
2084                         inode->i_mode = attr->ia_mode;
2085                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2086                         mask |= CEPH_SETATTR_MODE;
2087                         release |= CEPH_CAP_AUTH_SHARED;
2088                 }
2089         }
2090
2091         if (ia_valid & ATTR_ATIME) {
2092                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2093                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2094                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2095                 if (issued & CEPH_CAP_FILE_EXCL) {
2096                         ci->i_time_warp_seq++;
2097                         inode->i_atime = attr->ia_atime;
2098                         dirtied |= CEPH_CAP_FILE_EXCL;
2099                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2100                            timespec64_compare(&inode->i_atime,
2101                                             &attr->ia_atime) < 0) {
2102                         inode->i_atime = attr->ia_atime;
2103                         dirtied |= CEPH_CAP_FILE_WR;
2104                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2105                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2106                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2107                                                &attr->ia_atime);
2108                         mask |= CEPH_SETATTR_ATIME;
2109                         release |= CEPH_CAP_FILE_SHARED |
2110                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2111                 }
2112         }
2113         if (ia_valid & ATTR_SIZE) {
2114                 loff_t isize = i_size_read(inode);
2115
2116                 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2117                 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2118                         if (attr->ia_size > isize) {
2119                                 i_size_write(inode, attr->ia_size);
2120                                 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2121                                 ci->i_reported_size = attr->ia_size;
2122                                 dirtied |= CEPH_CAP_FILE_EXCL;
2123                                 ia_valid |= ATTR_MTIME;
2124                         }
2125                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2126                            attr->ia_size != isize) {
2127                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2128                         req->r_args.setattr.old_size = cpu_to_le64(isize);
2129                         mask |= CEPH_SETATTR_SIZE;
2130                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2131                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2132                 }
2133         }
2134         if (ia_valid & ATTR_MTIME) {
2135                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2136                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2137                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2138                 if (issued & CEPH_CAP_FILE_EXCL) {
2139                         ci->i_time_warp_seq++;
2140                         inode->i_mtime = attr->ia_mtime;
2141                         dirtied |= CEPH_CAP_FILE_EXCL;
2142                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2143                            timespec64_compare(&inode->i_mtime,
2144                                             &attr->ia_mtime) < 0) {
2145                         inode->i_mtime = attr->ia_mtime;
2146                         dirtied |= CEPH_CAP_FILE_WR;
2147                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2148                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2149                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2150                                                &attr->ia_mtime);
2151                         mask |= CEPH_SETATTR_MTIME;
2152                         release |= CEPH_CAP_FILE_SHARED |
2153                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2154                 }
2155         }
2156
2157         /* these do nothing */
2158         if (ia_valid & ATTR_CTIME) {
2159                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2160                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2161                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2162                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2163                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2164                      only ? "ctime only" : "ignored");
2165                 if (only) {
2166                         /*
2167                          * if kernel wants to dirty ctime but nothing else,
2168                          * we need to choose a cap to dirty under, or do
2169                          * a almost-no-op setattr
2170                          */
2171                         if (issued & CEPH_CAP_AUTH_EXCL)
2172                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2173                         else if (issued & CEPH_CAP_FILE_EXCL)
2174                                 dirtied |= CEPH_CAP_FILE_EXCL;
2175                         else if (issued & CEPH_CAP_XATTR_EXCL)
2176                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2177                         else
2178                                 mask |= CEPH_SETATTR_CTIME;
2179                 }
2180         }
2181         if (ia_valid & ATTR_FILE)
2182                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2183
2184         if (dirtied) {
2185                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2186                                                            &prealloc_cf);
2187                 inode->i_ctime = attr->ia_ctime;
2188         }
2189
2190         release &= issued;
2191         spin_unlock(&ci->i_ceph_lock);
2192         if (lock_snap_rwsem)
2193                 up_read(&mdsc->snap_rwsem);
2194
2195         if (inode_dirty_flags)
2196                 __mark_inode_dirty(inode, inode_dirty_flags);
2197
2198         if (mask) {
2199                 req->r_inode = inode;
2200                 ihold(inode);
2201                 req->r_inode_drop = release;
2202                 req->r_args.setattr.mask = cpu_to_le32(mask);
2203                 req->r_num_caps = 1;
2204                 req->r_stamp = attr->ia_ctime;
2205                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2206         }
2207         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2208              ceph_cap_string(dirtied), mask);
2209
2210         ceph_mdsc_put_request(req);
2211         ceph_free_cap_flush(prealloc_cf);
2212
2213         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2214                 __ceph_do_pending_vmtruncate(inode);
2215
2216         return err;
2217 }
2218
2219 /*
2220  * setattr
2221  */
2222 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2223                  struct iattr *attr)
2224 {
2225         struct inode *inode = d_inode(dentry);
2226         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2227         int err;
2228
2229         if (ceph_snap(inode) != CEPH_NOSNAP)
2230                 return -EROFS;
2231
2232         if (ceph_inode_is_shutdown(inode))
2233                 return -ESTALE;
2234
2235         err = setattr_prepare(&init_user_ns, dentry, attr);
2236         if (err != 0)
2237                 return err;
2238
2239         if ((attr->ia_valid & ATTR_SIZE) &&
2240             attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2241                 return -EFBIG;
2242
2243         if ((attr->ia_valid & ATTR_SIZE) &&
2244             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2245                 return -EDQUOT;
2246
2247         err = __ceph_setattr(inode, attr);
2248
2249         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2250                 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode);
2251
2252         return err;
2253 }
2254
2255 /*
2256  * Verify that we have a lease on the given mask.  If not,
2257  * do a getattr against an mds.
2258  */
2259 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2260                       int mask, bool force)
2261 {
2262         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2263         struct ceph_mds_client *mdsc = fsc->mdsc;
2264         struct ceph_mds_request *req;
2265         int mode;
2266         int err;
2267
2268         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2269                 dout("do_getattr inode %p SNAPDIR\n", inode);
2270                 return 0;
2271         }
2272
2273         dout("do_getattr inode %p mask %s mode 0%o\n",
2274              inode, ceph_cap_string(mask), inode->i_mode);
2275         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2276                         return 0;
2277
2278         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2279         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2280         if (IS_ERR(req))
2281                 return PTR_ERR(req);
2282         req->r_inode = inode;
2283         ihold(inode);
2284         req->r_num_caps = 1;
2285         req->r_args.getattr.mask = cpu_to_le32(mask);
2286         req->r_locked_page = locked_page;
2287         err = ceph_mdsc_do_request(mdsc, NULL, req);
2288         if (locked_page && err == 0) {
2289                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2290                 if (inline_version == 0) {
2291                         /* the reply is supposed to contain inline data */
2292                         err = -EINVAL;
2293                 } else if (inline_version == CEPH_INLINE_NONE) {
2294                         err = -ENODATA;
2295                 } else {
2296                         err = req->r_reply_info.targeti.inline_len;
2297                 }
2298         }
2299         ceph_mdsc_put_request(req);
2300         dout("do_getattr result=%d\n", err);
2301         return err;
2302 }
2303
2304 int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2305                       size_t size)
2306 {
2307         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2308         struct ceph_mds_client *mdsc = fsc->mdsc;
2309         struct ceph_mds_request *req;
2310         int mode = USE_AUTH_MDS;
2311         int err;
2312         char *xattr_value;
2313         size_t xattr_value_len;
2314
2315         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2316         if (IS_ERR(req)) {
2317                 err = -ENOMEM;
2318                 goto out;
2319         }
2320
2321         req->r_path2 = kstrdup(name, GFP_NOFS);
2322         if (!req->r_path2) {
2323                 err = -ENOMEM;
2324                 goto put;
2325         }
2326
2327         ihold(inode);
2328         req->r_inode = inode;
2329         err = ceph_mdsc_do_request(mdsc, NULL, req);
2330         if (err < 0)
2331                 goto put;
2332
2333         xattr_value = req->r_reply_info.xattr_info.xattr_value;
2334         xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2335
2336         dout("do_getvxattr xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
2337
2338         err = (int)xattr_value_len;
2339         if (size == 0)
2340                 goto put;
2341
2342         if (xattr_value_len > size) {
2343                 err = -ERANGE;
2344                 goto put;
2345         }
2346
2347         memcpy(value, xattr_value, xattr_value_len);
2348 put:
2349         ceph_mdsc_put_request(req);
2350 out:
2351         dout("do_getvxattr result=%d\n", err);
2352         return err;
2353 }
2354
2355
2356 /*
2357  * Check inode permissions.  We verify we have a valid value for
2358  * the AUTH cap, then call the generic handler.
2359  */
2360 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2361                     int mask)
2362 {
2363         int err;
2364
2365         if (mask & MAY_NOT_BLOCK)
2366                 return -ECHILD;
2367
2368         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2369
2370         if (!err)
2371                 err = generic_permission(&init_user_ns, inode, mask);
2372         return err;
2373 }
2374
2375 /* Craft a mask of needed caps given a set of requested statx attrs. */
2376 static int statx_to_caps(u32 want, umode_t mode)
2377 {
2378         int mask = 0;
2379
2380         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2381                 mask |= CEPH_CAP_AUTH_SHARED;
2382
2383         if (want & (STATX_NLINK|STATX_CTIME)) {
2384                 /*
2385                  * The link count for directories depends on inode->i_subdirs,
2386                  * and that is only updated when Fs caps are held.
2387                  */
2388                 if (S_ISDIR(mode))
2389                         mask |= CEPH_CAP_FILE_SHARED;
2390                 else
2391                         mask |= CEPH_CAP_LINK_SHARED;
2392         }
2393
2394         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2395                     STATX_BLOCKS))
2396                 mask |= CEPH_CAP_FILE_SHARED;
2397
2398         if (want & (STATX_CTIME))
2399                 mask |= CEPH_CAP_XATTR_SHARED;
2400
2401         return mask;
2402 }
2403
2404 /*
2405  * Get all the attributes. If we have sufficient caps for the requested attrs,
2406  * then we can avoid talking to the MDS at all.
2407  */
2408 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2409                  struct kstat *stat, u32 request_mask, unsigned int flags)
2410 {
2411         struct inode *inode = d_inode(path->dentry);
2412         struct ceph_inode_info *ci = ceph_inode(inode);
2413         u32 valid_mask = STATX_BASIC_STATS;
2414         int err = 0;
2415
2416         if (ceph_inode_is_shutdown(inode))
2417                 return -ESTALE;
2418
2419         /* Skip the getattr altogether if we're asked not to sync */
2420         if (!(flags & AT_STATX_DONT_SYNC)) {
2421                 err = ceph_do_getattr(inode,
2422                                 statx_to_caps(request_mask, inode->i_mode),
2423                                 flags & AT_STATX_FORCE_SYNC);
2424                 if (err)
2425                         return err;
2426         }
2427
2428         generic_fillattr(&init_user_ns, inode, stat);
2429         stat->ino = ceph_present_inode(inode);
2430
2431         /*
2432          * btime on newly-allocated inodes is 0, so if this is still set to
2433          * that, then assume that it's not valid.
2434          */
2435         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2436                 stat->btime = ci->i_btime;
2437                 valid_mask |= STATX_BTIME;
2438         }
2439
2440         if (ceph_snap(inode) == CEPH_NOSNAP)
2441                 stat->dev = inode->i_sb->s_dev;
2442         else
2443                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2444
2445         if (S_ISDIR(inode->i_mode)) {
2446                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2447                                         RBYTES))
2448                         stat->size = ci->i_rbytes;
2449                 else
2450                         stat->size = ci->i_files + ci->i_subdirs;
2451                 stat->blocks = 0;
2452                 stat->blksize = 65536;
2453                 /*
2454                  * Some applications rely on the number of st_nlink
2455                  * value on directories to be either 0 (if unlinked)
2456                  * or 2 + number of subdirectories.
2457                  */
2458                 if (stat->nlink == 1)
2459                         /* '.' + '..' + subdirs */
2460                         stat->nlink = 1 + 1 + ci->i_subdirs;
2461         }
2462
2463         stat->result_mask = request_mask & valid_mask;
2464         return err;
2465 }
2466
2467 void ceph_inode_shutdown(struct inode *inode)
2468 {
2469         struct ceph_inode_info *ci = ceph_inode(inode);
2470         struct rb_node *p;
2471         int iputs = 0;
2472         bool invalidate = false;
2473
2474         spin_lock(&ci->i_ceph_lock);
2475         ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
2476         p = rb_first(&ci->i_caps);
2477         while (p) {
2478                 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
2479
2480                 p = rb_next(p);
2481                 iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
2482         }
2483         spin_unlock(&ci->i_ceph_lock);
2484
2485         if (invalidate)
2486                 ceph_queue_invalidate(inode);
2487         while (iputs--)
2488                 iput(inode);
2489 }