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         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  45         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  46         inode_set_iversion_raw(inode, 0);
  47         return 0;
  48 }
  49
  50 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  51 {
  52         struct inode *inode;
  53         ino_t t = ceph_vino_to_ino(vino);
  54
  55         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  56         if (!inode)
  57                 return ERR_PTR(-ENOMEM);
  58         if (inode->i_state & I_NEW)
  59                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  60                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  61
  62         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  63              vino.snap, inode);
  64         return inode;
  65 }
  66
  67 /*
  68  * get/constuct snapdir inode for a given directory
  69  */
  70 struct inode *ceph_get_snapdir(struct inode *parent)
  71 {
  72         struct ceph_vino vino = {
  73                 .ino = ceph_ino(parent),
  74                 .snap = CEPH_SNAPDIR,
  75         };
  76         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  77         struct ceph_inode_info *ci = ceph_inode(inode);
  78
  79         BUG_ON(!S_ISDIR(parent->i_mode));
  80         if (IS_ERR(inode))
  81                 return inode;
  82         inode->i_mode = parent->i_mode;
  83         inode->i_uid = parent->i_uid;
  84         inode->i_gid = parent->i_gid;
  85         inode->i_op = &ceph_snapdir_iops;
  86         inode->i_fop = &ceph_snapdir_fops;
  87         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  88         ci->i_rbytes = 0;
  89
  90         if (inode->i_state & I_NEW)
  91                 unlock_new_inode(inode);
  92
  93         return inode;
  94 }
  95
  96 const struct inode_operations ceph_file_iops = {
  97         .permission = ceph_permission,
  98         .setattr = ceph_setattr,
  99         .getattr = ceph_getattr,
 100         .listxattr = ceph_listxattr,
 101         .get_acl = ceph_get_acl,
 102         .set_acl = ceph_set_acl,
 103 };
 104
 105
 106 /*
 107  * We use a 'frag tree' to keep track of the MDS's directory fragments
 108  * for a given inode (usually there is just a single fragment).  We
 109  * need to know when a child frag is delegated to a new MDS, or when
 110  * it is flagged as replicated, so we can direct our requests
 111  * accordingly.
 112  */
 113
 114 /*
 115  * find/create a frag in the tree
 116  */
 117 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 118                                                     u32 f)
 119 {
 120         struct rb_node **p;
 121         struct rb_node *parent = NULL;
 122         struct ceph_inode_frag *frag;
 123         int c;
 124
 125         p = &ci->i_fragtree.rb_node;
 126         while (*p) {
 127                 parent = *p;
 128                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 129                 c = ceph_frag_compare(f, frag->frag);
 130                 if (c < 0)
 131                         p = &(*p)->rb_left;
 132                 else if (c > 0)
 133                         p = &(*p)->rb_right;
 134                 else
 135                         return frag;
 136         }
 137
 138         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 139         if (!frag)
 140                 return ERR_PTR(-ENOMEM);
 141
 142         frag->frag = f;
 143         frag->split_by = 0;
 144         frag->mds = -1;
 145         frag->ndist = 0;
 146
 147         rb_link_node(&frag->node, parent, p);
 148         rb_insert_color(&frag->node, &ci->i_fragtree);
 149
 150         dout("get_or_create_frag added %llx.%llx frag %x\n",
 151              ceph_vinop(&ci->vfs_inode), f);
 152         return frag;
 153 }
 154
 155 /*
 156  * find a specific frag @f
 157  */
 158 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 159 {
 160         struct rb_node *n = ci->i_fragtree.rb_node;
 161
 162         while (n) {
 163                 struct ceph_inode_frag *frag =
 164                         rb_entry(n, struct ceph_inode_frag, node);
 165                 int c = ceph_frag_compare(f, frag->frag);
 166                 if (c < 0)
 167                         n = n->rb_left;
 168                 else if (c > 0)
 169                         n = n->rb_right;
 170                 else
 171                         return frag;
 172         }
 173         return NULL;
 174 }
 175
 176 /*
 177  * Choose frag containing the given value @v.  If @pfrag is
 178  * specified, copy the frag delegation info to the caller if
 179  * it is present.
 180  */
 181 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 182                               struct ceph_inode_frag *pfrag, int *found)
 183 {
 184         u32 t = ceph_frag_make(0, 0);
 185         struct ceph_inode_frag *frag;
 186         unsigned nway, i;
 187         u32 n;
 188
 189         if (found)
 190                 *found = 0;
 191
 192         while (1) {
 193                 WARN_ON(!ceph_frag_contains_value(t, v));
 194                 frag = __ceph_find_frag(ci, t);
 195                 if (!frag)
 196                         break; /* t is a leaf */
 197                 if (frag->split_by == 0) {
 198                         if (pfrag)
 199                                 memcpy(pfrag, frag, sizeof(*pfrag));
 200                         if (found)
 201                                 *found = 1;
 202                         break;
 203                 }
 204
 205                 /* choose child */
 206                 nway = 1 << frag->split_by;
 207                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 208                      frag->split_by, nway);
 209                 for (i = 0; i < nway; i++) {
 210                         n = ceph_frag_make_child(t, frag->split_by, i);
 211                         if (ceph_frag_contains_value(n, v)) {
 212                                 t = n;
 213                                 break;
 214                         }
 215                 }
 216                 BUG_ON(i == nway);
 217         }
 218         dout("choose_frag(%x) = %x\n", v, t);
 219
 220         return t;
 221 }
 222
 223 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 224                      struct ceph_inode_frag *pfrag, int *found)
 225 {
 226         u32 ret;
 227         mutex_lock(&ci->i_fragtree_mutex);
 228         ret = __ceph_choose_frag(ci, v, pfrag, found);
 229         mutex_unlock(&ci->i_fragtree_mutex);
 230         return ret;
 231 }
 232
 233 /*
 234  * Process dirfrag (delegation) info from the mds.  Include leaf
 235  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 236  * branches/splits are included in i_fragtree)
 237  */
 238 static int ceph_fill_dirfrag(struct inode *inode,
 239                              struct ceph_mds_reply_dirfrag *dirinfo)
 240 {
 241         struct ceph_inode_info *ci = ceph_inode(inode);
 242         struct ceph_inode_frag *frag;
 243         u32 id = le32_to_cpu(dirinfo->frag);
 244         int mds = le32_to_cpu(dirinfo->auth);
 245         int ndist = le32_to_cpu(dirinfo->ndist);
 246         int diri_auth = -1;
 247         int i;
 248         int err = 0;
 249
 250         spin_lock(&ci->i_ceph_lock);
 251         if (ci->i_auth_cap)
 252                 diri_auth = ci->i_auth_cap->mds;
 253         spin_unlock(&ci->i_ceph_lock);
 254
 255         if (mds == -1) /* CDIR_AUTH_PARENT */
 256                 mds = diri_auth;
 257
 258         mutex_lock(&ci->i_fragtree_mutex);
 259         if (ndist == 0 && mds == diri_auth) {
 260                 /* no delegation info needed. */
 261                 frag = __ceph_find_frag(ci, id);
 262                 if (!frag)
 263                         goto out;
 264                 if (frag->split_by == 0) {
 265                         /* tree leaf, remove */
 266                         dout("fill_dirfrag removed %llx.%llx frag %x"
 267                              " (no ref)\n", ceph_vinop(inode), id);
 268                         rb_erase(&frag->node, &ci->i_fragtree);
 269                         kfree(frag);
 270                 } else {
 271                         /* tree branch, keep and clear */
 272                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 273                              " referral\n", ceph_vinop(inode), id);
 274                         frag->mds = -1;
 275                         frag->ndist = 0;
 276                 }
 277                 goto out;
 278         }
 279
 280
 281         /* find/add this frag to store mds delegation info */
 282         frag = __get_or_create_frag(ci, id);
 283         if (IS_ERR(frag)) {
 284                 /* this is not the end of the world; we can continue
 285                    with bad/inaccurate delegation info */
 286                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 287                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 288                 err = -ENOMEM;
 289                 goto out;
 290         }
 291
 292         frag->mds = mds;
 293         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 294         for (i = 0; i < frag->ndist; i++)
 295                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 296         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 297              ceph_vinop(inode), frag->frag, frag->ndist);
 298
 299 out:
 300         mutex_unlock(&ci->i_fragtree_mutex);
 301         return err;
 302 }
 303
 304 static int frag_tree_split_cmp(const void *l, const void *r)
 305 {
 306         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 307         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 308         return ceph_frag_compare(le32_to_cpu(ls->frag),
 309                                  le32_to_cpu(rs->frag));
 310 }
 311
 312 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 313 {
 314         if (!frag)
 315                 return f == ceph_frag_make(0, 0);
 316         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 317                 return false;
 318         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 319 }
 320
 321 static int ceph_fill_fragtree(struct inode *inode,
 322                               struct ceph_frag_tree_head *fragtree,
 323                               struct ceph_mds_reply_dirfrag *dirinfo)
 324 {
 325         struct ceph_inode_info *ci = ceph_inode(inode);
 326         struct ceph_inode_frag *frag, *prev_frag = NULL;
 327         struct rb_node *rb_node;
 328         unsigned i, split_by, nsplits;
 329         u32 id;
 330         bool update = false;
 331
 332         mutex_lock(&ci->i_fragtree_mutex);
 333         nsplits = le32_to_cpu(fragtree->nsplits);
 334         if (nsplits != ci->i_fragtree_nsplits) {
 335                 update = true;
 336         } else if (nsplits) {
 337                 i = prandom_u32() % nsplits;
 338                 id = le32_to_cpu(fragtree->splits[i].frag);
 339                 if (!__ceph_find_frag(ci, id))
 340                         update = true;
 341         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 342                 rb_node = rb_first(&ci->i_fragtree);
 343                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 344                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 345                         update = true;
 346         }
 347         if (!update && dirinfo) {
 348                 id = le32_to_cpu(dirinfo->frag);
 349                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 350                         update = true;
 351         }
 352         if (!update)
 353                 goto out_unlock;
 354
 355         if (nsplits > 1) {
 356                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 357                      frag_tree_split_cmp, NULL);
 358         }
 359
 360         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 361         rb_node = rb_first(&ci->i_fragtree);
 362         for (i = 0; i < nsplits; i++) {
 363                 id = le32_to_cpu(fragtree->splits[i].frag);
 364                 split_by = le32_to_cpu(fragtree->splits[i].by);
 365                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 366                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 367                                "frag %x split by %d\n", ceph_vinop(inode),
 368                                i, nsplits, id, split_by);
 369                         continue;
 370                 }
 371                 frag = NULL;
 372                 while (rb_node) {
 373                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 374                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 375                                 if (frag->frag != id)
 376                                         frag = NULL;
 377                                 else
 378                                         rb_node = rb_next(rb_node);
 379                                 break;
 380                         }
 381                         rb_node = rb_next(rb_node);
 382                         /* delete stale split/leaf node */
 383                         if (frag->split_by > 0 ||
 384                             !is_frag_child(frag->frag, prev_frag)) {
 385                                 rb_erase(&frag->node, &ci->i_fragtree);
 386                                 if (frag->split_by > 0)
 387                                         ci->i_fragtree_nsplits--;
 388                                 kfree(frag);
 389                         }
 390                         frag = NULL;
 391                 }
 392                 if (!frag) {
 393                         frag = __get_or_create_frag(ci, id);
 394                         if (IS_ERR(frag))
 395                                 continue;
 396                 }
 397                 if (frag->split_by == 0)
 398                         ci->i_fragtree_nsplits++;
 399                 frag->split_by = split_by;
 400                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 401                 prev_frag = frag;
 402         }
 403         while (rb_node) {
 404                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 405                 rb_node = rb_next(rb_node);
 406                 /* delete stale split/leaf node */
 407                 if (frag->split_by > 0 ||
 408                     !is_frag_child(frag->frag, prev_frag)) {
 409                         rb_erase(&frag->node, &ci->i_fragtree);
 410                         if (frag->split_by > 0)
 411                                 ci->i_fragtree_nsplits--;
 412                         kfree(frag);
 413                 }
 414         }
 415 out_unlock:
 416         mutex_unlock(&ci->i_fragtree_mutex);
 417         return 0;
 418 }
 419
 420 /*
 421  * initialize a newly allocated inode.
 422  */
 423 struct inode *ceph_alloc_inode(struct super_block *sb)
 424 {
 425         struct ceph_inode_info *ci;
 426         int i;
 427
 428         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 429         if (!ci)
 430                 return NULL;
 431
 432         dout("alloc_inode %p\n", &ci->vfs_inode);
 433
 434         spin_lock_init(&ci->i_ceph_lock);
 435
 436         ci->i_version = 0;
 437         ci->i_inline_version = 0;
 438         ci->i_time_warp_seq = 0;
 439         ci->i_ceph_flags = 0;
 440         atomic64_set(&ci->i_ordered_count, 1);
 441         atomic64_set(&ci->i_release_count, 1);
 442         atomic64_set(&ci->i_complete_seq[0], 0);
 443         atomic64_set(&ci->i_complete_seq[1], 0);
 444         ci->i_symlink = NULL;
 445
 446         ci->i_max_bytes = 0;
 447         ci->i_max_files = 0;
 448
 449         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 450         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 451
 452         ci->i_fragtree = RB_ROOT;
 453         mutex_init(&ci->i_fragtree_mutex);
 454
 455         ci->i_xattrs.blob = NULL;
 456         ci->i_xattrs.prealloc_blob = NULL;
 457         ci->i_xattrs.dirty = false;
 458         ci->i_xattrs.index = RB_ROOT;
 459         ci->i_xattrs.count = 0;
 460         ci->i_xattrs.names_size = 0;
 461         ci->i_xattrs.vals_size = 0;
 462         ci->i_xattrs.version = 0;
 463         ci->i_xattrs.index_version = 0;
 464
 465         ci->i_caps = RB_ROOT;
 466         ci->i_auth_cap = NULL;
 467         ci->i_dirty_caps = 0;
 468         ci->i_flushing_caps = 0;
 469         INIT_LIST_HEAD(&ci->i_dirty_item);
 470         INIT_LIST_HEAD(&ci->i_flushing_item);
 471         ci->i_prealloc_cap_flush = NULL;
 472         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 473         init_waitqueue_head(&ci->i_cap_wq);
 474         ci->i_hold_caps_min = 0;
 475         ci->i_hold_caps_max = 0;
 476         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 477         INIT_LIST_HEAD(&ci->i_cap_snaps);
 478         ci->i_head_snapc = NULL;
 479         ci->i_snap_caps = 0;
 480
 481         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 482                 ci->i_nr_by_mode[i] = 0;
 483
 484         mutex_init(&ci->i_truncate_mutex);
 485         ci->i_truncate_seq = 0;
 486         ci->i_truncate_size = 0;
 487         ci->i_truncate_pending = 0;
 488
 489         ci->i_max_size = 0;
 490         ci->i_reported_size = 0;
 491         ci->i_wanted_max_size = 0;
 492         ci->i_requested_max_size = 0;
 493
 494         ci->i_pin_ref = 0;
 495         ci->i_rd_ref = 0;
 496         ci->i_rdcache_ref = 0;
 497         ci->i_wr_ref = 0;
 498         ci->i_wb_ref = 0;
 499         ci->i_wrbuffer_ref = 0;
 500         ci->i_wrbuffer_ref_head = 0;
 501         atomic_set(&ci->i_filelock_ref, 0);
 502         atomic_set(&ci->i_shared_gen, 1);
 503         ci->i_rdcache_gen = 0;
 504         ci->i_rdcache_revoking = 0;
 505
 506         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 507         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 508         spin_lock_init(&ci->i_unsafe_lock);
 509
 510         ci->i_snap_realm = NULL;
 511         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 512         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 513
 514         INIT_WORK(&ci->i_work, ceph_inode_work);
 515         ci->i_work_mask = 0;
 516         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 517
 518         ceph_fscache_inode_init(ci);
 519
 520         ci->i_meta_err = 0;
 521
 522         return &ci->vfs_inode;
 523 }
 524
 525 void ceph_free_inode(struct inode *inode)
 526 {
 527         struct ceph_inode_info *ci = ceph_inode(inode);
 528
 529         kfree(ci->i_symlink);
 530         kmem_cache_free(ceph_inode_cachep, ci);
 531 }
 532
 533 void ceph_evict_inode(struct inode *inode)
 534 {
 535         struct ceph_inode_info *ci = ceph_inode(inode);
 536         struct ceph_inode_frag *frag;
 537         struct rb_node *n;
 538
 539         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 540
 541         truncate_inode_pages_final(&inode->i_data);
 542         clear_inode(inode);
 543
 544         ceph_fscache_unregister_inode_cookie(ci);
 545
 546         __ceph_remove_caps(ci);
 547
 548         if (__ceph_has_any_quota(ci))
 549                 ceph_adjust_quota_realms_count(inode, false);
 550
 551         /*
 552          * we may still have a snap_realm reference if there are stray
 553          * caps in i_snap_caps.
 554          */
 555         if (ci->i_snap_realm) {
 556                 struct ceph_mds_client *mdsc =
 557                                         ceph_inode_to_client(inode)->mdsc;
 558                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 559                         struct ceph_snap_realm *realm = ci->i_snap_realm;
 560                         dout(" dropping residual ref to snap realm %p\n",
 561                              realm);
 562                         spin_lock(&realm->inodes_with_caps_lock);
 563                         list_del_init(&ci->i_snap_realm_item);
 564                         ci->i_snap_realm = NULL;
 565                         if (realm->ino == ci->i_vino.ino)
 566                                 realm->inode = NULL;
 567                         spin_unlock(&realm->inodes_with_caps_lock);
 568                         ceph_put_snap_realm(mdsc, realm);
 569                 } else {
 570                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 571                         ci->i_snap_realm = NULL;
 572                 }
 573         }
 574
 575         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 576                 frag = rb_entry(n, struct ceph_inode_frag, node);
 577                 rb_erase(n, &ci->i_fragtree);
 578                 kfree(frag);
 579         }
 580         ci->i_fragtree_nsplits = 0;
 581
 582         __ceph_destroy_xattrs(ci);
 583         if (ci->i_xattrs.blob)
 584                 ceph_buffer_put(ci->i_xattrs.blob);
 585         if (ci->i_xattrs.prealloc_blob)
 586                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 587
 588         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 589 }
 590
 591 static inline blkcnt_t calc_inode_blocks(u64 size)
 592 {
 593         return (size + (1<<9) - 1) >> 9;
 594 }
 595
 596 /*
 597  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 598  * careful because either the client or MDS may have more up to date
 599  * info, depending on which capabilities are held, and whether
 600  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 601  * and size are monotonically increasing, except when utimes() or
 602  * truncate() increments the corresponding _seq values.)
 603  */
 604 int ceph_fill_file_size(struct inode *inode, int issued,
 605                         u32 truncate_seq, u64 truncate_size, u64 size)
 606 {
 607         struct ceph_inode_info *ci = ceph_inode(inode);
 608         int queue_trunc = 0;
 609
 610         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 611             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 612                 dout("size %lld -> %llu\n", inode->i_size, size);
 613                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 614                         pr_err("fill_file_size non-zero size for directory\n");
 615                         size = 0;
 616                 }
 617                 i_size_write(inode, size);
 618                 inode->i_blocks = calc_inode_blocks(size);
 619                 ci->i_reported_size = size;
 620                 if (truncate_seq != ci->i_truncate_seq) {
 621                         dout("truncate_seq %u -> %u\n",
 622                              ci->i_truncate_seq, truncate_seq);
 623                         ci->i_truncate_seq = truncate_seq;
 624
 625                         /* the MDS should have revoked these caps */
 626                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 627                                                CEPH_CAP_FILE_RD |
 628                                                CEPH_CAP_FILE_WR |
 629                                                CEPH_CAP_FILE_LAZYIO));
 630                         /*
 631                          * If we hold relevant caps, or in the case where we're
 632                          * not the only client referencing this file and we
 633                          * don't hold those caps, then we need to check whether
 634                          * the file is either opened or mmaped
 635                          */
 636                         if ((issued & (CEPH_CAP_FILE_CACHE|
 637                                        CEPH_CAP_FILE_BUFFER)) ||
 638                             mapping_mapped(inode->i_mapping) ||
 639                             __ceph_caps_file_wanted(ci)) {
 640                                 ci->i_truncate_pending++;
 641                                 queue_trunc = 1;
 642                         }
 643                 }
 644         }
 645         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 646             ci->i_truncate_size != truncate_size) {
 647                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 648                      truncate_size);
 649                 ci->i_truncate_size = truncate_size;
 650         }
 651
 652         if (queue_trunc)
 653                 ceph_fscache_invalidate(inode);
 654
 655         return queue_trunc;
 656 }
 657
 658 void ceph_fill_file_time(struct inode *inode, int issued,
 659                          u64 time_warp_seq, struct timespec64 *ctime,
 660                          struct timespec64 *mtime, struct timespec64 *atime)
 661 {
 662         struct ceph_inode_info *ci = ceph_inode(inode);
 663         int warn = 0;
 664
 665         if (issued & (CEPH_CAP_FILE_EXCL|
 666                       CEPH_CAP_FILE_WR|
 667                       CEPH_CAP_FILE_BUFFER|
 668                       CEPH_CAP_AUTH_EXCL|
 669                       CEPH_CAP_XATTR_EXCL)) {
 670                 if (ci->i_version == 0 ||
 671                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 672                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 673                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 674                              ctime->tv_sec, ctime->tv_nsec);
 675                         inode->i_ctime = *ctime;
 676                 }
 677                 if (ci->i_version == 0 ||
 678                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 679                         /* the MDS did a utimes() */
 680                         dout("mtime %lld.%09ld -> %lld.%09ld "
 681                              "tw %d -> %d\n",
 682                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 683                              mtime->tv_sec, mtime->tv_nsec,
 684                              ci->i_time_warp_seq, (int)time_warp_seq);
 685
 686                         inode->i_mtime = *mtime;
 687                         inode->i_atime = *atime;
 688                         ci->i_time_warp_seq = time_warp_seq;
 689                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 690                         /* nobody did utimes(); take the max */
 691                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 692                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 693                                      inode->i_mtime.tv_sec,
 694                                      inode->i_mtime.tv_nsec,
 695                                      mtime->tv_sec, mtime->tv_nsec);
 696                                 inode->i_mtime = *mtime;
 697                         }
 698                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 699                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 700                                      inode->i_atime.tv_sec,
 701                                      inode->i_atime.tv_nsec,
 702                                      atime->tv_sec, atime->tv_nsec);
 703                                 inode->i_atime = *atime;
 704                         }
 705                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 706                         /* we did a utimes(); ignore mds values */
 707                 } else {
 708                         warn = 1;
 709                 }
 710         } else {
 711                 /* we have no write|excl caps; whatever the MDS says is true */
 712                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 713                         inode->i_ctime = *ctime;
 714                         inode->i_mtime = *mtime;
 715                         inode->i_atime = *atime;
 716                         ci->i_time_warp_seq = time_warp_seq;
 717                 } else {
 718                         warn = 1;
 719                 }
 720         }
 721         if (warn) /* time_warp_seq shouldn't go backwards */
 722                 dout("%p mds time_warp_seq %llu < %u\n",
 723                      inode, time_warp_seq, ci->i_time_warp_seq);
 724 }
 725
 726 /*
 727  * Populate an inode based on info from mds.  May be called on new or
 728  * existing inodes.
 729  */
 730 static int fill_inode(struct inode *inode, struct page *locked_page,
 731                       struct ceph_mds_reply_info_in *iinfo,
 732                       struct ceph_mds_reply_dirfrag *dirinfo,
 733                       struct ceph_mds_session *session, int cap_fmode,
 734                       struct ceph_cap_reservation *caps_reservation)
 735 {
 736         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 737         struct ceph_mds_reply_inode *info = iinfo->in;
 738         struct ceph_inode_info *ci = ceph_inode(inode);
 739         int issued, new_issued, info_caps;
 740         struct timespec64 mtime, atime, ctime;
 741         struct ceph_buffer *xattr_blob = NULL;
 742         struct ceph_buffer *old_blob = NULL;
 743         struct ceph_string *pool_ns = NULL;
 744         struct ceph_cap *new_cap = NULL;
 745         int err = 0;
 746         bool wake = false;
 747         bool queue_trunc = false;
 748         bool new_version = false;
 749         bool fill_inline = false;
 750
 751         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 752              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 753              ci->i_version);
 754
 755         info_caps = le32_to_cpu(info->cap.caps);
 756
 757         /* prealloc new cap struct */
 758         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 759                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 760                 if (!new_cap)
 761                         return -ENOMEM;
 762         }
 763
 764         /*
 765          * prealloc xattr data, if it looks like we'll need it.  only
 766          * if len > 4 (meaning there are actually xattrs; the first 4
 767          * bytes are the xattr count).
 768          */
 769         if (iinfo->xattr_len > 4) {
 770                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 771                 if (!xattr_blob)
 772                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 773                                iinfo->xattr_len);
 774         }
 775
 776         if (iinfo->pool_ns_len > 0)
 777                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 778                                                      iinfo->pool_ns_len);
 779
 780         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 781                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 782
 783         spin_lock(&ci->i_ceph_lock);
 784
 785         /*
 786          * provided version will be odd if inode value is projected,
 787          * even if stable.  skip the update if we have newer stable
 788          * info (ours>=theirs, e.g. due to racing mds replies), unless
 789          * we are getting projected (unstable) info (in which case the
 790          * version is odd, and we want ours>theirs).
 791          *   us   them
 792          *   2    2     skip
 793          *   3    2     skip
 794          *   3    3     update
 795          */
 796         if (ci->i_version == 0 ||
 797             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 798              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 799                 new_version = true;
 800
 801         /* Update change_attribute */
 802         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 803
 804         __ceph_caps_issued(ci, &issued);
 805         issued |= __ceph_caps_dirty(ci);
 806         new_issued = ~issued & info_caps;
 807
 808         /* update inode */
 809         inode->i_rdev = le32_to_cpu(info->rdev);
 810         /* directories have fl_stripe_unit set to zero */
 811         if (le32_to_cpu(info->layout.fl_stripe_unit))
 812                 inode->i_blkbits =
 813                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 814         else
 815                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 816
 817         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 818
 819         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 820             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 821                 inode->i_mode = le32_to_cpu(info->mode);
 822                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 823                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 824                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 825                      from_kuid(&init_user_ns, inode->i_uid),
 826                      from_kgid(&init_user_ns, inode->i_gid));
 827                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 828                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 829         }
 830
 831         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 832             (issued & CEPH_CAP_LINK_EXCL) == 0)
 833                 set_nlink(inode, le32_to_cpu(info->nlink));
 834
 835         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 836                 /* be careful with mtime, atime, size */
 837                 ceph_decode_timespec64(&atime, &info->atime);
 838                 ceph_decode_timespec64(&mtime, &info->mtime);
 839                 ceph_decode_timespec64(&ctime, &info->ctime);
 840                 ceph_fill_file_time(inode, issued,
 841                                 le32_to_cpu(info->time_warp_seq),
 842                                 &ctime, &mtime, &atime);
 843         }
 844
 845         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 846                 ci->i_files = le64_to_cpu(info->files);
 847                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 848         }
 849
 850         if (new_version ||
 851             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 852                 s64 old_pool = ci->i_layout.pool_id;
 853                 struct ceph_string *old_ns;
 854
 855                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 856                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 857                                         lockdep_is_held(&ci->i_ceph_lock));
 858                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 859
 860                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 861                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 862
 863                 pool_ns = old_ns;
 864
 865                 queue_trunc = ceph_fill_file_size(inode, issued,
 866                                         le32_to_cpu(info->truncate_seq),
 867                                         le64_to_cpu(info->truncate_size),
 868                                         le64_to_cpu(info->size));
 869                 /* only update max_size on auth cap */
 870                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 871                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 872                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 873                                         le64_to_cpu(info->max_size));
 874                         ci->i_max_size = le64_to_cpu(info->max_size);
 875                 }
 876         }
 877
 878         /* layout and rstat are not tracked by capability, update them if
 879          * the inode info is from auth mds */
 880         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 881                 if (S_ISDIR(inode->i_mode)) {
 882                         ci->i_dir_layout = iinfo->dir_layout;
 883                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 884                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 885                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 886                         ci->i_dir_pin = iinfo->dir_pin;
 887                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 888                 }
 889         }
 890
 891         /* xattrs */
 892         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 893         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 894             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 895                 if (ci->i_xattrs.blob)
 896                         old_blob = ci->i_xattrs.blob;
 897                 ci->i_xattrs.blob = xattr_blob;
 898                 if (xattr_blob)
 899                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 900                                iinfo->xattr_data, iinfo->xattr_len);
 901                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 902                 ceph_forget_all_cached_acls(inode);
 903                 ceph_security_invalidate_secctx(inode);
 904                 xattr_blob = NULL;
 905         }
 906
 907         /* finally update i_version */
 908         if (le64_to_cpu(info->version) > ci->i_version)
 909                 ci->i_version = le64_to_cpu(info->version);
 910
 911         inode->i_mapping->a_ops = &ceph_aops;
 912
 913         switch (inode->i_mode & S_IFMT) {
 914         case S_IFIFO:
 915         case S_IFBLK:
 916         case S_IFCHR:
 917         case S_IFSOCK:
 918                 inode->i_blkbits = PAGE_SHIFT;
 919                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 920                 inode->i_op = &ceph_file_iops;
 921                 break;
 922         case S_IFREG:
 923                 inode->i_op = &ceph_file_iops;
 924                 inode->i_fop = &ceph_file_fops;
 925                 break;
 926         case S_IFLNK:
 927                 inode->i_op = &ceph_symlink_iops;
 928                 if (!ci->i_symlink) {
 929                         u32 symlen = iinfo->symlink_len;
 930                         char *sym;
 931
 932                         spin_unlock(&ci->i_ceph_lock);
 933
 934                         if (symlen != i_size_read(inode)) {
 935                                 pr_err("fill_inode %llx.%llx BAD symlink "
 936                                         "size %lld\n", ceph_vinop(inode),
 937                                         i_size_read(inode));
 938                                 i_size_write(inode, symlen);
 939                                 inode->i_blocks = calc_inode_blocks(symlen);
 940                         }
 941
 942                         err = -ENOMEM;
 943                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 944                         if (!sym)
 945                                 goto out;
 946
 947                         spin_lock(&ci->i_ceph_lock);
 948                         if (!ci->i_symlink)
 949                                 ci->i_symlink = sym;
 950                         else
 951                                 kfree(sym); /* lost a race */
 952                 }
 953                 inode->i_link = ci->i_symlink;
 954                 break;
 955         case S_IFDIR:
 956                 inode->i_op = &ceph_dir_iops;
 957                 inode->i_fop = &ceph_dir_fops;
 958                 break;
 959         default:
 960                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 961                        ceph_vinop(inode), inode->i_mode);
 962         }
 963
 964         /* were we issued a capability? */
 965         if (info_caps) {
 966                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 967                         ceph_add_cap(inode, session,
 968                                      le64_to_cpu(info->cap.cap_id),
 969                                      cap_fmode, info_caps,
 970                                      le32_to_cpu(info->cap.wanted),
 971                                      le32_to_cpu(info->cap.seq),
 972                                      le32_to_cpu(info->cap.mseq),
 973                                      le64_to_cpu(info->cap.realm),
 974                                      info->cap.flags, &new_cap);
 975
 976                         /* set dir completion flag? */
 977                         if (S_ISDIR(inode->i_mode) &&
 978                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 979                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 980                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 981                             !__ceph_dir_is_complete(ci)) {
 982                                 dout(" marking %p complete (empty)\n", inode);
 983                                 i_size_write(inode, 0);
 984                                 __ceph_dir_set_complete(ci,
 985                                         atomic64_read(&ci->i_release_count),
 986                                         atomic64_read(&ci->i_ordered_count));
 987                         }
 988
 989                         wake = true;
 990                 } else {
 991                         dout(" %p got snap_caps %s\n", inode,
 992                              ceph_cap_string(info_caps));
 993                         ci->i_snap_caps |= info_caps;
 994                         if (cap_fmode >= 0)
 995                                 __ceph_get_fmode(ci, cap_fmode);
 996                 }
 997         } else if (cap_fmode >= 0) {
 998                 pr_warn("mds issued no caps on %llx.%llx\n",
 999                            ceph_vinop(inode));
1000                 __ceph_get_fmode(ci, cap_fmode);
1001         }
1002
1003         if (iinfo->inline_version > 0 &&
1004             iinfo->inline_version >= ci->i_inline_version) {
1005                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1006                 ci->i_inline_version = iinfo->inline_version;
1007                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1008                     (locked_page || (info_caps & cache_caps)))
1009                         fill_inline = true;
1010         }
1011
1012         spin_unlock(&ci->i_ceph_lock);
1013
1014         if (fill_inline)
1015                 ceph_fill_inline_data(inode, locked_page,
1016                                       iinfo->inline_data, iinfo->inline_len);
1017
1018         if (wake)
1019                 wake_up_all(&ci->i_cap_wq);
1020
1021         /* queue truncate if we saw i_size decrease */
1022         if (queue_trunc)
1023                 ceph_queue_vmtruncate(inode);
1024
1025         /* populate frag tree */
1026         if (S_ISDIR(inode->i_mode))
1027                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1028
1029         /* update delegation info? */
1030         if (dirinfo)
1031                 ceph_fill_dirfrag(inode, dirinfo);
1032
1033         err = 0;
1034 out:
1035         if (new_cap)
1036                 ceph_put_cap(mdsc, new_cap);
1037         ceph_buffer_put(old_blob);
1038         ceph_buffer_put(xattr_blob);
1039         ceph_put_string(pool_ns);
1040         return err;
1041 }
1042
1043 /*
1044  * caller should hold session s_mutex and dentry->d_lock.
1045  */
1046 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1047                                   struct ceph_mds_reply_lease *lease,
1048                                   struct ceph_mds_session *session,
1049                                   unsigned long from_time,
1050                                   struct ceph_mds_session **old_lease_session)
1051 {
1052         struct ceph_dentry_info *di = ceph_dentry(dentry);
1053         long unsigned duration = le32_to_cpu(lease->duration_ms);
1054         long unsigned ttl = from_time + (duration * HZ) / 1000;
1055         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1056
1057         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1058              dentry, duration, ttl);
1059
1060         /* only track leases on regular dentries */
1061         if (ceph_snap(dir) != CEPH_NOSNAP)
1062                 return;
1063
1064         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1065         if (duration == 0) {
1066                 __ceph_dentry_dir_lease_touch(di);
1067                 return;
1068         }
1069
1070         if (di->lease_gen == session->s_cap_gen &&
1071             time_before(ttl, di->time))
1072                 return;  /* we already have a newer lease. */
1073
1074         if (di->lease_session && di->lease_session != session) {
1075                 *old_lease_session = di->lease_session;
1076                 di->lease_session = NULL;
1077         }
1078
1079         if (!di->lease_session)
1080                 di->lease_session = ceph_get_mds_session(session);
1081         di->lease_gen = session->s_cap_gen;
1082         di->lease_seq = le32_to_cpu(lease->seq);
1083         di->lease_renew_after = half_ttl;
1084         di->lease_renew_from = 0;
1085         di->time = ttl;
1086
1087         __ceph_dentry_lease_touch(di);
1088 }
1089
1090 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1091                                         struct ceph_mds_reply_lease *lease,
1092                                         struct ceph_mds_session *session,
1093                                         unsigned long from_time)
1094 {
1095         struct ceph_mds_session *old_lease_session = NULL;
1096         spin_lock(&dentry->d_lock);
1097         __update_dentry_lease(dir, dentry, lease, session, from_time,
1098                               &old_lease_session);
1099         spin_unlock(&dentry->d_lock);
1100         if (old_lease_session)
1101                 ceph_put_mds_session(old_lease_session);
1102 }
1103
1104 /*
1105  * update dentry lease without having parent inode locked
1106  */
1107 static void update_dentry_lease_careful(struct dentry *dentry,
1108                                         struct ceph_mds_reply_lease *lease,
1109                                         struct ceph_mds_session *session,
1110                                         unsigned long from_time,
1111                                         char *dname, u32 dname_len,
1112                                         struct ceph_vino *pdvino,
1113                                         struct ceph_vino *ptvino)
1114
1115 {
1116         struct inode *dir;
1117         struct ceph_mds_session *old_lease_session = NULL;
1118
1119         spin_lock(&dentry->d_lock);
1120         /* make sure dentry's name matches target */
1121         if (dentry->d_name.len != dname_len ||
1122             memcmp(dentry->d_name.name, dname, dname_len))
1123                 goto out_unlock;
1124
1125         dir = d_inode(dentry->d_parent);
1126         /* make sure parent matches dvino */
1127         if (!ceph_ino_compare(dir, pdvino))
1128                 goto out_unlock;
1129
1130         /* make sure dentry's inode matches target. NULL ptvino means that
1131          * we expect a negative dentry */
1132         if (ptvino) {
1133                 if (d_really_is_negative(dentry))
1134                         goto out_unlock;
1135                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1136                         goto out_unlock;
1137         } else {
1138                 if (d_really_is_positive(dentry))
1139                         goto out_unlock;
1140         }
1141
1142         __update_dentry_lease(dir, dentry, lease, session,
1143                               from_time, &old_lease_session);
1144 out_unlock:
1145         spin_unlock(&dentry->d_lock);
1146         if (old_lease_session)
1147                 ceph_put_mds_session(old_lease_session);
1148 }
1149
1150 /*
1151  * splice a dentry to an inode.
1152  * caller must hold directory i_mutex for this to be safe.
1153  */
1154 static int splice_dentry(struct dentry **pdn, struct inode *in)
1155 {
1156         struct dentry *dn = *pdn;
1157         struct dentry *realdn;
1158
1159         BUG_ON(d_inode(dn));
1160
1161         if (S_ISDIR(in->i_mode)) {
1162                 /* If inode is directory, d_splice_alias() below will remove
1163                  * 'realdn' from its origin parent. We need to ensure that
1164                  * origin parent's readdir cache will not reference 'realdn'
1165                  */
1166                 realdn = d_find_any_alias(in);
1167                 if (realdn) {
1168                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1169                         spin_lock(&realdn->d_lock);
1170
1171                         realdn->d_op->d_prune(realdn);
1172
1173                         di->time = jiffies;
1174                         di->lease_shared_gen = 0;
1175                         di->offset = 0;
1176
1177                         spin_unlock(&realdn->d_lock);
1178                         dput(realdn);
1179                 }
1180         }
1181
1182         /* dn must be unhashed */
1183         if (!d_unhashed(dn))
1184                 d_drop(dn);
1185         realdn = d_splice_alias(in, dn);
1186         if (IS_ERR(realdn)) {
1187                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1188                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1189                 return PTR_ERR(realdn);
1190         }
1191
1192         if (realdn) {
1193                 dout("dn %p (%d) spliced with %p (%d) "
1194                      "inode %p ino %llx.%llx\n",
1195                      dn, d_count(dn),
1196                      realdn, d_count(realdn),
1197                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1198                 dput(dn);
1199                 *pdn = realdn;
1200         } else {
1201                 BUG_ON(!ceph_dentry(dn));
1202                 dout("dn %p attached to %p ino %llx.%llx\n",
1203                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1204         }
1205         return 0;
1206 }
1207
1208 /*
1209  * Incorporate results into the local cache.  This is either just
1210  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1211  * after a lookup).
1212  *
1213  * A reply may contain
1214  *         a directory inode along with a dentry.
1215  *  and/or a target inode
1216  *
1217  * Called with snap_rwsem (read).
1218  */
1219 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1220 {
1221         struct ceph_mds_session *session = req->r_session;
1222         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1223         struct inode *in = NULL;
1224         struct ceph_vino tvino, dvino;
1225         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1226         int err = 0;
1227
1228         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1229              rinfo->head->is_dentry, rinfo->head->is_target);
1230
1231         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1232                 dout("fill_trace reply is empty!\n");
1233                 if (rinfo->head->result == 0 && req->r_parent)
1234                         ceph_invalidate_dir_request(req);
1235                 return 0;
1236         }
1237
1238         if (rinfo->head->is_dentry) {
1239                 struct inode *dir = req->r_parent;
1240
1241                 if (dir) {
1242                         err = fill_inode(dir, NULL,
1243                                          &rinfo->diri, rinfo->dirfrag,
1244                                          session, -1,
1245                                          &req->r_caps_reservation);
1246                         if (err < 0)
1247                                 goto done;
1248                 } else {
1249                         WARN_ON_ONCE(1);
1250                 }
1251
1252                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1253                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1254                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1255                         struct qstr dname;
1256                         struct dentry *dn, *parent;
1257
1258                         BUG_ON(!rinfo->head->is_target);
1259                         BUG_ON(req->r_dentry);
1260
1261                         parent = d_find_any_alias(dir);
1262                         BUG_ON(!parent);
1263
1264                         dname.name = rinfo->dname;
1265                         dname.len = rinfo->dname_len;
1266                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1267                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1268                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1269 retry_lookup:
1270                         dn = d_lookup(parent, &dname);
1271                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1272                              parent, dname.len, dname.name, dn);
1273
1274                         if (!dn) {
1275                                 dn = d_alloc(parent, &dname);
1276                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1277                                      dname.len, dname.name, dn);
1278                                 if (!dn) {
1279                                         dput(parent);
1280                                         err = -ENOMEM;
1281                                         goto done;
1282                                 }
1283                                 err = 0;
1284                         } else if (d_really_is_positive(dn) &&
1285                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1286                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1287                                 dout(" dn %p points to wrong inode %p\n",
1288                                      dn, d_inode(dn));
1289                                 ceph_dir_clear_ordered(dir);
1290                                 d_delete(dn);
1291                                 dput(dn);
1292                                 goto retry_lookup;
1293                         }
1294
1295                         req->r_dentry = dn;
1296                         dput(parent);
1297                 }
1298         }
1299
1300         if (rinfo->head->is_target) {
1301                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1302                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1303
1304                 in = ceph_get_inode(sb, tvino);
1305                 if (IS_ERR(in)) {
1306                         err = PTR_ERR(in);
1307                         goto done;
1308                 }
1309
1310                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1311                                 session,
1312                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1313                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1314                                 &req->r_caps_reservation);
1315                 if (err < 0) {
1316                         pr_err("fill_inode badness %p %llx.%llx\n",
1317                                 in, ceph_vinop(in));
1318                         if (in->i_state & I_NEW)
1319                                 discard_new_inode(in);
1320                         goto done;
1321                 }
1322                 req->r_target_inode = in;
1323                 if (in->i_state & I_NEW)
1324                         unlock_new_inode(in);
1325         }
1326
1327         /*
1328          * ignore null lease/binding on snapdir ENOENT, or else we
1329          * will have trouble splicing in the virtual snapdir later
1330          */
1331         if (rinfo->head->is_dentry &&
1332             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1333             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1334             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1335                                                fsc->mount_options->snapdir_name,
1336                                                req->r_dentry->d_name.len))) {
1337                 /*
1338                  * lookup link rename   : null -> possibly existing inode
1339                  * mknod symlink mkdir  : null -> new inode
1340                  * unlink               : linked -> null
1341                  */
1342                 struct inode *dir = req->r_parent;
1343                 struct dentry *dn = req->r_dentry;
1344                 bool have_dir_cap, have_lease;
1345
1346                 BUG_ON(!dn);
1347                 BUG_ON(!dir);
1348                 BUG_ON(d_inode(dn->d_parent) != dir);
1349
1350                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1351                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1352
1353                 BUG_ON(ceph_ino(dir) != dvino.ino);
1354                 BUG_ON(ceph_snap(dir) != dvino.snap);
1355
1356                 /* do we have a lease on the whole dir? */
1357                 have_dir_cap =
1358                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1359                          CEPH_CAP_FILE_SHARED);
1360
1361                 /* do we have a dn lease? */
1362                 have_lease = have_dir_cap ||
1363                         le32_to_cpu(rinfo->dlease->duration_ms);
1364                 if (!have_lease)
1365                         dout("fill_trace  no dentry lease or dir cap\n");
1366
1367                 /* rename? */
1368                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1369                         struct inode *olddir = req->r_old_dentry_dir;
1370                         BUG_ON(!olddir);
1371
1372                         dout(" src %p '%pd' dst %p '%pd'\n",
1373                              req->r_old_dentry,
1374                              req->r_old_dentry,
1375                              dn, dn);
1376                         dout("fill_trace doing d_move %p -> %p\n",
1377                              req->r_old_dentry, dn);
1378
1379                         /* d_move screws up sibling dentries' offsets */
1380                         ceph_dir_clear_ordered(dir);
1381                         ceph_dir_clear_ordered(olddir);
1382
1383                         d_move(req->r_old_dentry, dn);
1384                         dout(" src %p '%pd' dst %p '%pd'\n",
1385                              req->r_old_dentry,
1386                              req->r_old_dentry,
1387                              dn, dn);
1388
1389                         /* ensure target dentry is invalidated, despite
1390                            rehashing bug in vfs_rename_dir */
1391                         ceph_invalidate_dentry_lease(dn);
1392
1393                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1394                              ceph_dentry(req->r_old_dentry)->offset);
1395
1396                         /* swap r_dentry and r_old_dentry in case that
1397                          * splice_dentry() gets called later. This is safe
1398                          * because no other place will use them */
1399                         req->r_dentry = req->r_old_dentry;
1400                         req->r_old_dentry = dn;
1401                         dn = req->r_dentry;
1402                 }
1403
1404                 /* null dentry? */
1405                 if (!rinfo->head->is_target) {
1406                         dout("fill_trace null dentry\n");
1407                         if (d_really_is_positive(dn)) {
1408                                 dout("d_delete %p\n", dn);
1409                                 ceph_dir_clear_ordered(dir);
1410                                 d_delete(dn);
1411                         } else if (have_lease) {
1412                                 if (d_unhashed(dn))
1413                                         d_add(dn, NULL);
1414                                 update_dentry_lease(dir, dn,
1415                                                     rinfo->dlease, session,
1416                                                     req->r_request_started);
1417                         }
1418                         goto done;
1419                 }
1420
1421                 /* attach proper inode */
1422                 if (d_really_is_negative(dn)) {
1423                         ceph_dir_clear_ordered(dir);
1424                         ihold(in);
1425                         err = splice_dentry(&req->r_dentry, in);
1426                         if (err < 0)
1427                                 goto done;
1428                         dn = req->r_dentry;  /* may have spliced */
1429                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1430                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1431                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1432                              ceph_vinop(in));
1433                         d_invalidate(dn);
1434                         have_lease = false;
1435                 }
1436
1437                 if (have_lease) {
1438                         update_dentry_lease(dir, dn,
1439                                             rinfo->dlease, session,
1440                                             req->r_request_started);
1441                 }
1442                 dout(" final dn %p\n", dn);
1443         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1444                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1445                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1446                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1447                 struct inode *dir = req->r_parent;
1448
1449                 /* fill out a snapdir LOOKUPSNAP dentry */
1450                 BUG_ON(!dir);
1451                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1452                 BUG_ON(!req->r_dentry);
1453                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1454                 ceph_dir_clear_ordered(dir);
1455                 ihold(in);
1456                 err = splice_dentry(&req->r_dentry, in);
1457                 if (err < 0)
1458                         goto done;
1459         } else if (rinfo->head->is_dentry && req->r_dentry) {
1460                 /* parent inode is not locked, be carefull */
1461                 struct ceph_vino *ptvino = NULL;
1462                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1463                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1464                 if (rinfo->head->is_target) {
1465                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1466                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1467                         ptvino = &tvino;
1468                 }
1469                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1470                                             session, req->r_request_started,
1471                                             rinfo->dname, rinfo->dname_len,
1472                                             &dvino, ptvino);
1473         }
1474 done:
1475         dout("fill_trace done err=%d\n", err);
1476         return err;
1477 }
1478
1479 /*
1480  * Prepopulate our cache with readdir results, leases, etc.
1481  */
1482 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1483                                            struct ceph_mds_session *session)
1484 {
1485         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1486         int i, err = 0;
1487
1488         for (i = 0; i < rinfo->dir_nr; i++) {
1489                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1490                 struct ceph_vino vino;
1491                 struct inode *in;
1492                 int rc;
1493
1494                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1495                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1496
1497                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1498                 if (IS_ERR(in)) {
1499                         err = PTR_ERR(in);
1500                         dout("new_inode badness got %d\n", err);
1501                         continue;
1502                 }
1503                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1504                                 -1, &req->r_caps_reservation);
1505                 if (rc < 0) {
1506                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1507                         err = rc;
1508                         if (in->i_state & I_NEW) {
1509                                 ihold(in);
1510                                 discard_new_inode(in);
1511                         }
1512                 } else if (in->i_state & I_NEW) {
1513                         unlock_new_inode(in);
1514                 }
1515
1516                 /* avoid calling iput_final() in mds dispatch threads */
1517                 ceph_async_iput(in);
1518         }
1519
1520         return err;
1521 }
1522
1523 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1524 {
1525         if (ctl->page) {
1526                 kunmap(ctl->page);
1527                 put_page(ctl->page);
1528                 ctl->page = NULL;
1529         }
1530 }
1531
1532 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1533                               struct ceph_readdir_cache_control *ctl,
1534                               struct ceph_mds_request *req)
1535 {
1536         struct ceph_inode_info *ci = ceph_inode(dir);
1537         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1538         unsigned idx = ctl->index % nsize;
1539         pgoff_t pgoff = ctl->index / nsize;
1540
1541         if (!ctl->page || pgoff != page_index(ctl->page)) {
1542                 ceph_readdir_cache_release(ctl);
1543                 if (idx == 0)
1544                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1545                 else
1546                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1547                 if (!ctl->page) {
1548                         ctl->index = -1;
1549                         return idx == 0 ? -ENOMEM : 0;
1550                 }
1551                 /* reading/filling the cache are serialized by
1552                  * i_mutex, no need to use page lock */
1553                 unlock_page(ctl->page);
1554                 ctl->dentries = kmap(ctl->page);
1555                 if (idx == 0)
1556                         memset(ctl->dentries, 0, PAGE_SIZE);
1557         }
1558
1559         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1560             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1561                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1562                 ctl->dentries[idx] = dn;
1563                 ctl->index++;
1564         } else {
1565                 dout("disable readdir cache\n");
1566                 ctl->index = -1;
1567         }
1568         return 0;
1569 }
1570
1571 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1572                              struct ceph_mds_session *session)
1573 {
1574         struct dentry *parent = req->r_dentry;
1575         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1576         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1577         struct qstr dname;
1578         struct dentry *dn;
1579         struct inode *in;
1580         int err = 0, skipped = 0, ret, i;
1581         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1582         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1583         u32 last_hash = 0;
1584         u32 fpos_offset;
1585         struct ceph_readdir_cache_control cache_ctl = {};
1586
1587         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1588                 return readdir_prepopulate_inodes_only(req, session);
1589
1590         if (rinfo->hash_order) {
1591                 if (req->r_path2) {
1592                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1593                                                   req->r_path2,
1594                                                   strlen(req->r_path2));
1595                         last_hash = ceph_frag_value(last_hash);
1596                 } else if (rinfo->offset_hash) {
1597                         /* mds understands offset_hash */
1598                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1599                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1600                 }
1601         }
1602
1603         if (rinfo->dir_dir &&
1604             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1605                 dout("readdir_prepopulate got new frag %x -> %x\n",
1606                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1607                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1608                 if (!rinfo->hash_order)
1609                         req->r_readdir_offset = 2;
1610         }
1611
1612         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1613                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1614                      rinfo->dir_nr, parent);
1615         } else {
1616                 dout("readdir_prepopulate %d items under dn %p\n",
1617                      rinfo->dir_nr, parent);
1618                 if (rinfo->dir_dir)
1619                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1620
1621                 if (ceph_frag_is_leftmost(frag) &&
1622                     req->r_readdir_offset == 2 &&
1623                     !(rinfo->hash_order && last_hash)) {
1624                         /* note dir version at start of readdir so we can
1625                          * tell if any dentries get dropped */
1626                         req->r_dir_release_cnt =
1627                                 atomic64_read(&ci->i_release_count);
1628                         req->r_dir_ordered_cnt =
1629                                 atomic64_read(&ci->i_ordered_count);
1630                         req->r_readdir_cache_idx = 0;
1631                 }
1632         }
1633
1634         cache_ctl.index = req->r_readdir_cache_idx;
1635         fpos_offset = req->r_readdir_offset;
1636
1637         /* FIXME: release caps/leases if error occurs */
1638         for (i = 0; i < rinfo->dir_nr; i++) {
1639                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1640                 struct ceph_vino tvino;
1641
1642                 dname.name = rde->name;
1643                 dname.len = rde->name_len;
1644                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1645
1646                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1647                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1648
1649                 if (rinfo->hash_order) {
1650                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1651                                                  rde->name, rde->name_len);
1652                         hash = ceph_frag_value(hash);
1653                         if (hash != last_hash)
1654                                 fpos_offset = 2;
1655                         last_hash = hash;
1656                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1657                 } else {
1658                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1659                 }
1660
1661 retry_lookup:
1662                 dn = d_lookup(parent, &dname);
1663                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1664                      parent, dname.len, dname.name, dn);
1665
1666                 if (!dn) {
1667                         dn = d_alloc(parent, &dname);
1668                         dout("d_alloc %p '%.*s' = %p\n", parent,
1669                              dname.len, dname.name, dn);
1670                         if (!dn) {
1671                                 dout("d_alloc badness\n");
1672                                 err = -ENOMEM;
1673                                 goto out;
1674                         }
1675                 } else if (d_really_is_positive(dn) &&
1676                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1677                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1678                         struct ceph_dentry_info *di = ceph_dentry(dn);
1679                         dout(" dn %p points to wrong inode %p\n",
1680                              dn, d_inode(dn));
1681
1682                         spin_lock(&dn->d_lock);
1683                         if (di->offset > 0 &&
1684                             di->lease_shared_gen ==
1685                             atomic_read(&ci->i_shared_gen)) {
1686                                 __ceph_dir_clear_ordered(ci);
1687                                 di->offset = 0;
1688                         }
1689                         spin_unlock(&dn->d_lock);
1690
1691                         d_delete(dn);
1692                         dput(dn);
1693                         goto retry_lookup;
1694                 }
1695
1696                 /* inode */
1697                 if (d_really_is_positive(dn)) {
1698                         in = d_inode(dn);
1699                 } else {
1700                         in = ceph_get_inode(parent->d_sb, tvino);
1701                         if (IS_ERR(in)) {
1702                                 dout("new_inode badness\n");
1703                                 d_drop(dn);
1704                                 dput(dn);
1705                                 err = PTR_ERR(in);
1706                                 goto out;
1707                         }
1708                 }
1709
1710                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1711                                  -1, &req->r_caps_reservation);
1712                 if (ret < 0) {
1713                         pr_err("fill_inode badness on %p\n", in);
1714                         if (d_really_is_negative(dn)) {
1715                                 /* avoid calling iput_final() in mds
1716                                  * dispatch threads */
1717                                 if (in->i_state & I_NEW) {
1718                                         ihold(in);
1719                                         discard_new_inode(in);
1720                                 }
1721                                 ceph_async_iput(in);
1722                         }
1723                         d_drop(dn);
1724                         err = ret;
1725                         goto next_item;
1726                 }
1727                 if (in->i_state & I_NEW)
1728                         unlock_new_inode(in);
1729
1730                 if (d_really_is_negative(dn)) {
1731                         if (ceph_security_xattr_deadlock(in)) {
1732                                 dout(" skip splicing dn %p to inode %p"
1733                                      " (security xattr deadlock)\n", dn, in);
1734                                 ceph_async_iput(in);
1735                                 skipped++;
1736                                 goto next_item;
1737                         }
1738
1739                         err = splice_dentry(&dn, in);
1740                         if (err < 0)
1741                                 goto next_item;
1742                 }
1743
1744                 ceph_dentry(dn)->offset = rde->offset;
1745
1746                 update_dentry_lease(d_inode(parent), dn,
1747                                     rde->lease, req->r_session,
1748                                     req->r_request_started);
1749
1750                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1751                         ret = fill_readdir_cache(d_inode(parent), dn,
1752                                                  &cache_ctl, req);
1753                         if (ret < 0)
1754                                 err = ret;
1755                 }
1756 next_item:
1757                 dput(dn);
1758         }
1759 out:
1760         if (err == 0 && skipped == 0) {
1761                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1762                 req->r_readdir_cache_idx = cache_ctl.index;
1763         }
1764         ceph_readdir_cache_release(&cache_ctl);
1765         dout("readdir_prepopulate done\n");
1766         return err;
1767 }
1768
1769 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1770 {
1771         struct ceph_inode_info *ci = ceph_inode(inode);
1772         bool ret;
1773
1774         spin_lock(&ci->i_ceph_lock);
1775         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1776         i_size_write(inode, size);
1777         inode->i_blocks = calc_inode_blocks(size);
1778
1779         ret = __ceph_should_report_size(ci);
1780
1781         spin_unlock(&ci->i_ceph_lock);
1782         return ret;
1783 }
1784
1785 /*
1786  * Put reference to inode, but avoid calling iput_final() in current thread.
1787  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1788  * some contexts.
1789  */
1790 void ceph_async_iput(struct inode *inode)
1791 {
1792         if (!inode)
1793                 return;
1794         for (;;) {
1795                 if (atomic_add_unless(&inode->i_count, -1, 1))
1796                         break;
1797                 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1798                                &ceph_inode(inode)->i_work))
1799                         break;
1800                 /* queue work failed, i_count must be at least 2 */
1801         }
1802 }
1803
1804 /*
1805  * Write back inode data in a worker thread.  (This can't be done
1806  * in the message handler context.)
1807  */
1808 void ceph_queue_writeback(struct inode *inode)
1809 {
1810         struct ceph_inode_info *ci = ceph_inode(inode);
1811         set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1812
1813         ihold(inode);
1814         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1815                        &ci->i_work)) {
1816                 dout("ceph_queue_writeback %p\n", inode);
1817         } else {
1818                 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1819                      inode, ci->i_work_mask);
1820                 iput(inode);
1821         }
1822 }
1823
1824 /*
1825  * queue an async invalidation
1826  */
1827 void ceph_queue_invalidate(struct inode *inode)
1828 {
1829         struct ceph_inode_info *ci = ceph_inode(inode);
1830         set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1831
1832         ihold(inode);
1833         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1834                        &ceph_inode(inode)->i_work)) {
1835                 dout("ceph_queue_invalidate %p\n", inode);
1836         } else {
1837                 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1838                      inode, ci->i_work_mask);
1839                 iput(inode);
1840         }
1841 }
1842
1843 /*
1844  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1845  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1846  */
1847 void ceph_queue_vmtruncate(struct inode *inode)
1848 {
1849         struct ceph_inode_info *ci = ceph_inode(inode);
1850         set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1851
1852         ihold(inode);
1853         if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1854                        &ci->i_work)) {
1855                 dout("ceph_queue_vmtruncate %p\n", inode);
1856         } else {
1857                 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1858                      inode, ci->i_work_mask);
1859                 iput(inode);
1860         }
1861 }
1862
1863 static void ceph_do_invalidate_pages(struct inode *inode)
1864 {
1865         struct ceph_inode_info *ci = ceph_inode(inode);
1866         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1867         u32 orig_gen;
1868         int check = 0;
1869
1870         mutex_lock(&ci->i_truncate_mutex);
1871
1872         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1873                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1874                                     inode, ceph_ino(inode));
1875                 mapping_set_error(inode->i_mapping, -EIO);
1876                 truncate_pagecache(inode, 0);
1877                 mutex_unlock(&ci->i_truncate_mutex);
1878                 goto out;
1879         }
1880
1881         spin_lock(&ci->i_ceph_lock);
1882         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1883              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1884         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1885                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1886                         check = 1;
1887                 spin_unlock(&ci->i_ceph_lock);
1888                 mutex_unlock(&ci->i_truncate_mutex);
1889                 goto out;
1890         }
1891         orig_gen = ci->i_rdcache_gen;
1892         spin_unlock(&ci->i_ceph_lock);
1893
1894         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1895                 pr_err("invalidate_pages %p fails\n", inode);
1896         }
1897
1898         spin_lock(&ci->i_ceph_lock);
1899         if (orig_gen == ci->i_rdcache_gen &&
1900             orig_gen == ci->i_rdcache_revoking) {
1901                 dout("invalidate_pages %p gen %d successful\n", inode,
1902                      ci->i_rdcache_gen);
1903                 ci->i_rdcache_revoking--;
1904                 check = 1;
1905         } else {
1906                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1907                      inode, orig_gen, ci->i_rdcache_gen,
1908                      ci->i_rdcache_revoking);
1909                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1910                         check = 1;
1911         }
1912         spin_unlock(&ci->i_ceph_lock);
1913         mutex_unlock(&ci->i_truncate_mutex);
1914 out:
1915         if (check)
1916                 ceph_check_caps(ci, 0, NULL);
1917 }
1918
1919 /*
1920  * Make sure any pending truncation is applied before doing anything
1921  * that may depend on it.
1922  */
1923 void __ceph_do_pending_vmtruncate(struct inode *inode)
1924 {
1925         struct ceph_inode_info *ci = ceph_inode(inode);
1926         u64 to;
1927         int wrbuffer_refs, finish = 0;
1928
1929         mutex_lock(&ci->i_truncate_mutex);
1930 retry:
1931         spin_lock(&ci->i_ceph_lock);
1932         if (ci->i_truncate_pending == 0) {
1933                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1934                 spin_unlock(&ci->i_ceph_lock);
1935                 mutex_unlock(&ci->i_truncate_mutex);
1936                 return;
1937         }
1938
1939         /*
1940          * make sure any dirty snapped pages are flushed before we
1941          * possibly truncate them.. so write AND block!
1942          */
1943         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1944                 spin_unlock(&ci->i_ceph_lock);
1945                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1946                      inode);
1947                 filemap_write_and_wait_range(&inode->i_data, 0,
1948                                              inode->i_sb->s_maxbytes);
1949                 goto retry;
1950         }
1951
1952         /* there should be no reader or writer */
1953         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1954
1955         to = ci->i_truncate_size;
1956         wrbuffer_refs = ci->i_wrbuffer_ref;
1957         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1958              ci->i_truncate_pending, to);
1959         spin_unlock(&ci->i_ceph_lock);
1960
1961         truncate_pagecache(inode, to);
1962
1963         spin_lock(&ci->i_ceph_lock);
1964         if (to == ci->i_truncate_size) {
1965                 ci->i_truncate_pending = 0;
1966                 finish = 1;
1967         }
1968         spin_unlock(&ci->i_ceph_lock);
1969         if (!finish)
1970                 goto retry;
1971
1972         mutex_unlock(&ci->i_truncate_mutex);
1973
1974         if (wrbuffer_refs == 0)
1975                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1976
1977         wake_up_all(&ci->i_cap_wq);
1978 }
1979
1980 static void ceph_inode_work(struct work_struct *work)
1981 {
1982         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1983                                                  i_work);
1984         struct inode *inode = &ci->vfs_inode;
1985
1986         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1987                 dout("writeback %p\n", inode);
1988                 filemap_fdatawrite(&inode->i_data);
1989         }
1990         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1991                 ceph_do_invalidate_pages(inode);
1992
1993         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1994                 __ceph_do_pending_vmtruncate(inode);
1995
1996         iput(inode);
1997 }
1998
1999 /*
2000  * symlinks
2001  */
2002 static const struct inode_operations ceph_symlink_iops = {
2003         .get_link = simple_get_link,
2004         .setattr = ceph_setattr,
2005         .getattr = ceph_getattr,
2006         .listxattr = ceph_listxattr,
2007 };
2008
2009 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2010 {
2011         struct ceph_inode_info *ci = ceph_inode(inode);
2012         unsigned int ia_valid = attr->ia_valid;
2013         struct ceph_mds_request *req;
2014         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2015         struct ceph_cap_flush *prealloc_cf;
2016         int issued;
2017         int release = 0, dirtied = 0;
2018         int mask = 0;
2019         int err = 0;
2020         int inode_dirty_flags = 0;
2021         bool lock_snap_rwsem = false;
2022
2023         prealloc_cf = ceph_alloc_cap_flush();
2024         if (!prealloc_cf)
2025                 return -ENOMEM;
2026
2027         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2028                                        USE_AUTH_MDS);
2029         if (IS_ERR(req)) {
2030                 ceph_free_cap_flush(prealloc_cf);
2031                 return PTR_ERR(req);
2032         }
2033
2034         spin_lock(&ci->i_ceph_lock);
2035         issued = __ceph_caps_issued(ci, NULL);
2036
2037         if (!ci->i_head_snapc &&
2038             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2039                 lock_snap_rwsem = true;
2040                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2041                         spin_unlock(&ci->i_ceph_lock);
2042                         down_read(&mdsc->snap_rwsem);
2043                         spin_lock(&ci->i_ceph_lock);
2044                         issued = __ceph_caps_issued(ci, NULL);
2045                 }
2046         }
2047
2048         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2049
2050         if (ia_valid & ATTR_UID) {
2051                 dout("setattr %p uid %d -> %d\n", inode,
2052                      from_kuid(&init_user_ns, inode->i_uid),
2053                      from_kuid(&init_user_ns, attr->ia_uid));
2054                 if (issued & CEPH_CAP_AUTH_EXCL) {
2055                         inode->i_uid = attr->ia_uid;
2056                         dirtied |= CEPH_CAP_AUTH_EXCL;
2057                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2058                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2059                         req->r_args.setattr.uid = cpu_to_le32(
2060                                 from_kuid(&init_user_ns, attr->ia_uid));
2061                         mask |= CEPH_SETATTR_UID;
2062                         release |= CEPH_CAP_AUTH_SHARED;
2063                 }
2064         }
2065         if (ia_valid & ATTR_GID) {
2066                 dout("setattr %p gid %d -> %d\n", inode,
2067                      from_kgid(&init_user_ns, inode->i_gid),
2068                      from_kgid(&init_user_ns, attr->ia_gid));
2069                 if (issued & CEPH_CAP_AUTH_EXCL) {
2070                         inode->i_gid = attr->ia_gid;
2071                         dirtied |= CEPH_CAP_AUTH_EXCL;
2072                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2073                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2074                         req->r_args.setattr.gid = cpu_to_le32(
2075                                 from_kgid(&init_user_ns, attr->ia_gid));
2076                         mask |= CEPH_SETATTR_GID;
2077                         release |= CEPH_CAP_AUTH_SHARED;
2078                 }
2079         }
2080         if (ia_valid & ATTR_MODE) {
2081                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2082                      attr->ia_mode);
2083                 if (issued & CEPH_CAP_AUTH_EXCL) {
2084                         inode->i_mode = attr->ia_mode;
2085                         dirtied |= CEPH_CAP_AUTH_EXCL;
2086                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2087                            attr->ia_mode != inode->i_mode) {
2088                         inode->i_mode = attr->ia_mode;
2089                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2090                         mask |= CEPH_SETATTR_MODE;
2091                         release |= CEPH_CAP_AUTH_SHARED;
2092                 }
2093         }
2094
2095         if (ia_valid & ATTR_ATIME) {
2096                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2097                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2098                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2099                 if (issued & CEPH_CAP_FILE_EXCL) {
2100                         ci->i_time_warp_seq++;
2101                         inode->i_atime = attr->ia_atime;
2102                         dirtied |= CEPH_CAP_FILE_EXCL;
2103                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2104                            timespec64_compare(&inode->i_atime,
2105                                             &attr->ia_atime) < 0) {
2106                         inode->i_atime = attr->ia_atime;
2107                         dirtied |= CEPH_CAP_FILE_WR;
2108                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2109                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2110                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2111                                                &attr->ia_atime);
2112                         mask |= CEPH_SETATTR_ATIME;
2113                         release |= CEPH_CAP_FILE_SHARED |
2114                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2115                 }
2116         }
2117         if (ia_valid & ATTR_SIZE) {
2118                 dout("setattr %p size %lld -> %lld\n", inode,
2119                      inode->i_size, attr->ia_size);
2120                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2121                     attr->ia_size > inode->i_size) {
2122                         i_size_write(inode, attr->ia_size);
2123                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2124                         ci->i_reported_size = attr->ia_size;
2125                         dirtied |= CEPH_CAP_FILE_EXCL;
2126                         ia_valid |= ATTR_MTIME;
2127                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2128                            attr->ia_size != inode->i_size) {
2129                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2130                         req->r_args.setattr.old_size =
2131                                 cpu_to_le64(inode->i_size);
2132                         mask |= CEPH_SETATTR_SIZE;
2133                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2134                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2135                 }
2136         }
2137         if (ia_valid & ATTR_MTIME) {
2138                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2139                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2140                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2141                 if (issued & CEPH_CAP_FILE_EXCL) {
2142                         ci->i_time_warp_seq++;
2143                         inode->i_mtime = attr->ia_mtime;
2144                         dirtied |= CEPH_CAP_FILE_EXCL;
2145                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2146                            timespec64_compare(&inode->i_mtime,
2147                                             &attr->ia_mtime) < 0) {
2148                         inode->i_mtime = attr->ia_mtime;
2149                         dirtied |= CEPH_CAP_FILE_WR;
2150                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2151                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2152                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2153                                                &attr->ia_mtime);
2154                         mask |= CEPH_SETATTR_MTIME;
2155                         release |= CEPH_CAP_FILE_SHARED |
2156                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2157                 }
2158         }
2159
2160         /* these do nothing */
2161         if (ia_valid & ATTR_CTIME) {
2162                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2163                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2164                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2165                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2166                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2167                      only ? "ctime only" : "ignored");
2168                 if (only) {
2169                         /*
2170                          * if kernel wants to dirty ctime but nothing else,
2171                          * we need to choose a cap to dirty under, or do
2172                          * a almost-no-op setattr
2173                          */
2174                         if (issued & CEPH_CAP_AUTH_EXCL)
2175                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2176                         else if (issued & CEPH_CAP_FILE_EXCL)
2177                                 dirtied |= CEPH_CAP_FILE_EXCL;
2178                         else if (issued & CEPH_CAP_XATTR_EXCL)
2179                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2180                         else
2181                                 mask |= CEPH_SETATTR_CTIME;
2182                 }
2183         }
2184         if (ia_valid & ATTR_FILE)
2185                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2186
2187         if (dirtied) {
2188                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2189                                                            &prealloc_cf);
2190                 inode->i_ctime = attr->ia_ctime;
2191         }
2192
2193         release &= issued;
2194         spin_unlock(&ci->i_ceph_lock);
2195         if (lock_snap_rwsem)
2196                 up_read(&mdsc->snap_rwsem);
2197
2198         if (inode_dirty_flags)
2199                 __mark_inode_dirty(inode, inode_dirty_flags);
2200
2201
2202         if (mask) {
2203                 req->r_inode = inode;
2204                 ihold(inode);
2205                 req->r_inode_drop = release;
2206                 req->r_args.setattr.mask = cpu_to_le32(mask);
2207                 req->r_num_caps = 1;
2208                 req->r_stamp = attr->ia_ctime;
2209                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2210         }
2211         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2212              ceph_cap_string(dirtied), mask);
2213
2214         ceph_mdsc_put_request(req);
2215         ceph_free_cap_flush(prealloc_cf);
2216
2217         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2218                 __ceph_do_pending_vmtruncate(inode);
2219
2220         return err;
2221 }
2222
2223 /*
2224  * setattr
2225  */
2226 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2227 {
2228         struct inode *inode = d_inode(dentry);
2229         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2230         int err;
2231
2232         if (ceph_snap(inode) != CEPH_NOSNAP)
2233                 return -EROFS;
2234
2235         err = setattr_prepare(dentry, attr);
2236         if (err != 0)
2237                 return err;
2238
2239         if ((attr->ia_valid & ATTR_SIZE) &&
2240             attr->ia_size > max(inode->i_size, 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(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(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
2305 /*
2306  * Check inode permissions.  We verify we have a valid value for
2307  * the AUTH cap, then call the generic handler.
2308  */
2309 int ceph_permission(struct inode *inode, int mask)
2310 {
2311         int err;
2312
2313         if (mask & MAY_NOT_BLOCK)
2314                 return -ECHILD;
2315
2316         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2317
2318         if (!err)
2319                 err = generic_permission(inode, mask);
2320         return err;
2321 }
2322
2323 /* Craft a mask of needed caps given a set of requested statx attrs. */
2324 static int statx_to_caps(u32 want)
2325 {
2326         int mask = 0;
2327
2328         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2329                 mask |= CEPH_CAP_AUTH_SHARED;
2330
2331         if (want & (STATX_NLINK|STATX_CTIME))
2332                 mask |= CEPH_CAP_LINK_SHARED;
2333
2334         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2335                     STATX_BLOCKS))
2336                 mask |= CEPH_CAP_FILE_SHARED;
2337
2338         if (want & (STATX_CTIME))
2339                 mask |= CEPH_CAP_XATTR_SHARED;
2340
2341         return mask;
2342 }
2343
2344 /*
2345  * Get all the attributes. If we have sufficient caps for the requested attrs,
2346  * then we can avoid talking to the MDS at all.
2347  */
2348 int ceph_getattr(const struct path *path, struct kstat *stat,
2349                  u32 request_mask, unsigned int flags)
2350 {
2351         struct inode *inode = d_inode(path->dentry);
2352         struct ceph_inode_info *ci = ceph_inode(inode);
2353         u32 valid_mask = STATX_BASIC_STATS;
2354         int err = 0;
2355
2356         /* Skip the getattr altogether if we're asked not to sync */
2357         if (!(flags & AT_STATX_DONT_SYNC)) {
2358                 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2359                                       flags & AT_STATX_FORCE_SYNC);
2360                 if (err)
2361                         return err;
2362         }
2363
2364         generic_fillattr(inode, stat);
2365         stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2366
2367         /*
2368          * btime on newly-allocated inodes is 0, so if this is still set to
2369          * that, then assume that it's not valid.
2370          */
2371         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2372                 stat->btime = ci->i_btime;
2373                 valid_mask |= STATX_BTIME;
2374         }
2375
2376         if (ceph_snap(inode) == CEPH_NOSNAP)
2377                 stat->dev = inode->i_sb->s_dev;
2378         else
2379                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2380
2381         if (S_ISDIR(inode->i_mode)) {
2382                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2383                                         RBYTES))
2384                         stat->size = ci->i_rbytes;
2385                 else
2386                         stat->size = ci->i_files + ci->i_subdirs;
2387                 stat->blocks = 0;
2388                 stat->blksize = 65536;
2389                 /*
2390                  * Some applications rely on the number of st_nlink
2391                  * value on directories to be either 0 (if unlinked)
2392                  * or 2 + number of subdirectories.
2393                  */
2394                 if (stat->nlink == 1)
2395                         /* '.' + '..' + subdirs */
2396                         stat->nlink = 1 + 1 + ci->i_subdirs;
2397         }
2398
2399         stat->result_mask = request_mask & valid_mask;
2400         return err;
2401 }