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
  17 #include "super.h"
  18 #include "mds_client.h"
  19 #include "cache.h"
  20 #include <linux/ceph/decode.h>
  21
  22 /*
  23  * Ceph inode operations
  24  *
  25  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  26  * setattr, etc.), xattr helpers, and helpers for assimilating
  27  * metadata returned by the MDS into our cache.
  28  *
  29  * Also define helpers for doing asynchronous writeback, invalidation,
  30  * and truncation for the benefit of those who can't afford to block
  31  * (typically because they are in the message handler path).
  32  */
  33
  34 static const struct inode_operations ceph_symlink_iops;
  35
  36 static void ceph_invalidate_work(struct work_struct *work);
  37 static void ceph_writeback_work(struct work_struct *work);
  38 static void ceph_vmtruncate_work(struct work_struct *work);
  39
  40 /*
  41  * find or create an inode, given the ceph ino number
  42  */
  43 static int ceph_set_ino_cb(struct inode *inode, void *data)
  44 {
  45         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  46         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  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                 unlock_new_inode(inode);
  62         }
  63
  64         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  65              vino.snap, inode);
  66         return inode;
  67 }
  68
  69 /*
  70  * get/constuct snapdir inode for a given directory
  71  */
  72 struct inode *ceph_get_snapdir(struct inode *parent)
  73 {
  74         struct ceph_vino vino = {
  75                 .ino = ceph_ino(parent),
  76                 .snap = CEPH_SNAPDIR,
  77         };
  78         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  79         struct ceph_inode_info *ci = ceph_inode(inode);
  80
  81         BUG_ON(!S_ISDIR(parent->i_mode));
  82         if (IS_ERR(inode))
  83                 return inode;
  84         inode->i_mode = parent->i_mode;
  85         inode->i_uid = parent->i_uid;
  86         inode->i_gid = parent->i_gid;
  87         inode->i_op = &ceph_snapdir_iops;
  88         inode->i_fop = &ceph_snapdir_fops;
  89         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  90         ci->i_rbytes = 0;
  91         return inode;
  92 }
  93
  94 const struct inode_operations ceph_file_iops = {
  95         .permission = ceph_permission,
  96         .setattr = ceph_setattr,
  97         .getattr = ceph_getattr,
  98         .listxattr = ceph_listxattr,
  99         .get_acl = ceph_get_acl,
 100         .set_acl = ceph_set_acl,
 101 };
 102
 103
 104 /*
 105  * We use a 'frag tree' to keep track of the MDS's directory fragments
 106  * for a given inode (usually there is just a single fragment).  We
 107  * need to know when a child frag is delegated to a new MDS, or when
 108  * it is flagged as replicated, so we can direct our requests
 109  * accordingly.
 110  */
 111
 112 /*
 113  * find/create a frag in the tree
 114  */
 115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 116                                                     u32 f)
 117 {
 118         struct rb_node **p;
 119         struct rb_node *parent = NULL;
 120         struct ceph_inode_frag *frag;
 121         int c;
 122
 123         p = &ci->i_fragtree.rb_node;
 124         while (*p) {
 125                 parent = *p;
 126                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 127                 c = ceph_frag_compare(f, frag->frag);
 128                 if (c < 0)
 129                         p = &(*p)->rb_left;
 130                 else if (c > 0)
 131                         p = &(*p)->rb_right;
 132                 else
 133                         return frag;
 134         }
 135
 136         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 137         if (!frag)
 138                 return ERR_PTR(-ENOMEM);
 139
 140         frag->frag = f;
 141         frag->split_by = 0;
 142         frag->mds = -1;
 143         frag->ndist = 0;
 144
 145         rb_link_node(&frag->node, parent, p);
 146         rb_insert_color(&frag->node, &ci->i_fragtree);
 147
 148         dout("get_or_create_frag added %llx.%llx frag %x\n",
 149              ceph_vinop(&ci->vfs_inode), f);
 150         return frag;
 151 }
 152
 153 /*
 154  * find a specific frag @f
 155  */
 156 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 157 {
 158         struct rb_node *n = ci->i_fragtree.rb_node;
 159
 160         while (n) {
 161                 struct ceph_inode_frag *frag =
 162                         rb_entry(n, struct ceph_inode_frag, node);
 163                 int c = ceph_frag_compare(f, frag->frag);
 164                 if (c < 0)
 165                         n = n->rb_left;
 166                 else if (c > 0)
 167                         n = n->rb_right;
 168                 else
 169                         return frag;
 170         }
 171         return NULL;
 172 }
 173
 174 /*
 175  * Choose frag containing the given value @v.  If @pfrag is
 176  * specified, copy the frag delegation info to the caller if
 177  * it is present.
 178  */
 179 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 180                               struct ceph_inode_frag *pfrag, int *found)
 181 {
 182         u32 t = ceph_frag_make(0, 0);
 183         struct ceph_inode_frag *frag;
 184         unsigned nway, i;
 185         u32 n;
 186
 187         if (found)
 188                 *found = 0;
 189
 190         while (1) {
 191                 WARN_ON(!ceph_frag_contains_value(t, v));
 192                 frag = __ceph_find_frag(ci, t);
 193                 if (!frag)
 194                         break; /* t is a leaf */
 195                 if (frag->split_by == 0) {
 196                         if (pfrag)
 197                                 memcpy(pfrag, frag, sizeof(*pfrag));
 198                         if (found)
 199                                 *found = 1;
 200                         break;
 201                 }
 202
 203                 /* choose child */
 204                 nway = 1 << frag->split_by;
 205                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 206                      frag->split_by, nway);
 207                 for (i = 0; i < nway; i++) {
 208                         n = ceph_frag_make_child(t, frag->split_by, i);
 209                         if (ceph_frag_contains_value(n, v)) {
 210                                 t = n;
 211                                 break;
 212                         }
 213                 }
 214                 BUG_ON(i == nway);
 215         }
 216         dout("choose_frag(%x) = %x\n", v, t);
 217
 218         return t;
 219 }
 220
 221 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 222                      struct ceph_inode_frag *pfrag, int *found)
 223 {
 224         u32 ret;
 225         mutex_lock(&ci->i_fragtree_mutex);
 226         ret = __ceph_choose_frag(ci, v, pfrag, found);
 227         mutex_unlock(&ci->i_fragtree_mutex);
 228         return ret;
 229 }
 230
 231 /*
 232  * Process dirfrag (delegation) info from the mds.  Include leaf
 233  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 234  * branches/splits are included in i_fragtree)
 235  */
 236 static int ceph_fill_dirfrag(struct inode *inode,
 237                              struct ceph_mds_reply_dirfrag *dirinfo)
 238 {
 239         struct ceph_inode_info *ci = ceph_inode(inode);
 240         struct ceph_inode_frag *frag;
 241         u32 id = le32_to_cpu(dirinfo->frag);
 242         int mds = le32_to_cpu(dirinfo->auth);
 243         int ndist = le32_to_cpu(dirinfo->ndist);
 244         int diri_auth = -1;
 245         int i;
 246         int err = 0;
 247
 248         spin_lock(&ci->i_ceph_lock);
 249         if (ci->i_auth_cap)
 250                 diri_auth = ci->i_auth_cap->mds;
 251         spin_unlock(&ci->i_ceph_lock);
 252
 253         if (mds == -1) /* CDIR_AUTH_PARENT */
 254                 mds = diri_auth;
 255
 256         mutex_lock(&ci->i_fragtree_mutex);
 257         if (ndist == 0 && mds == diri_auth) {
 258                 /* no delegation info needed. */
 259                 frag = __ceph_find_frag(ci, id);
 260                 if (!frag)
 261                         goto out;
 262                 if (frag->split_by == 0) {
 263                         /* tree leaf, remove */
 264                         dout("fill_dirfrag removed %llx.%llx frag %x"
 265                              " (no ref)\n", ceph_vinop(inode), id);
 266                         rb_erase(&frag->node, &ci->i_fragtree);
 267                         kfree(frag);
 268                 } else {
 269                         /* tree branch, keep and clear */
 270                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 271                              " referral\n", ceph_vinop(inode), id);
 272                         frag->mds = -1;
 273                         frag->ndist = 0;
 274                 }
 275                 goto out;
 276         }
 277
 278
 279         /* find/add this frag to store mds delegation info */
 280         frag = __get_or_create_frag(ci, id);
 281         if (IS_ERR(frag)) {
 282                 /* this is not the end of the world; we can continue
 283                    with bad/inaccurate delegation info */
 284                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 285                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 286                 err = -ENOMEM;
 287                 goto out;
 288         }
 289
 290         frag->mds = mds;
 291         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 292         for (i = 0; i < frag->ndist; i++)
 293                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 294         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 295              ceph_vinop(inode), frag->frag, frag->ndist);
 296
 297 out:
 298         mutex_unlock(&ci->i_fragtree_mutex);
 299         return err;
 300 }
 301
 302 static int frag_tree_split_cmp(const void *l, const void *r)
 303 {
 304         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 305         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 306         return ceph_frag_compare(le32_to_cpu(ls->frag),
 307                                  le32_to_cpu(rs->frag));
 308 }
 309
 310 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 311 {
 312         if (!frag)
 313                 return f == ceph_frag_make(0, 0);
 314         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 315                 return false;
 316         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 317 }
 318
 319 static int ceph_fill_fragtree(struct inode *inode,
 320                               struct ceph_frag_tree_head *fragtree,
 321                               struct ceph_mds_reply_dirfrag *dirinfo)
 322 {
 323         struct ceph_inode_info *ci = ceph_inode(inode);
 324         struct ceph_inode_frag *frag, *prev_frag = NULL;
 325         struct rb_node *rb_node;
 326         unsigned i, split_by, nsplits;
 327         u32 id;
 328         bool update = false;
 329
 330         mutex_lock(&ci->i_fragtree_mutex);
 331         nsplits = le32_to_cpu(fragtree->nsplits);
 332         if (nsplits != ci->i_fragtree_nsplits) {
 333                 update = true;
 334         } else if (nsplits) {
 335                 i = prandom_u32() % nsplits;
 336                 id = le32_to_cpu(fragtree->splits[i].frag);
 337                 if (!__ceph_find_frag(ci, id))
 338                         update = true;
 339         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 340                 rb_node = rb_first(&ci->i_fragtree);
 341                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 342                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 343                         update = true;
 344         }
 345         if (!update && dirinfo) {
 346                 id = le32_to_cpu(dirinfo->frag);
 347                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 348                         update = true;
 349         }
 350         if (!update)
 351                 goto out_unlock;
 352
 353         if (nsplits > 1) {
 354                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 355                      frag_tree_split_cmp, NULL);
 356         }
 357
 358         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 359         rb_node = rb_first(&ci->i_fragtree);
 360         for (i = 0; i < nsplits; i++) {
 361                 id = le32_to_cpu(fragtree->splits[i].frag);
 362                 split_by = le32_to_cpu(fragtree->splits[i].by);
 363                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 364                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 365                                "frag %x split by %d\n", ceph_vinop(inode),
 366                                i, nsplits, id, split_by);
 367                         continue;
 368                 }
 369                 frag = NULL;
 370                 while (rb_node) {
 371                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 372                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 373                                 if (frag->frag != id)
 374                                         frag = NULL;
 375                                 else
 376                                         rb_node = rb_next(rb_node);
 377                                 break;
 378                         }
 379                         rb_node = rb_next(rb_node);
 380                         /* delete stale split/leaf node */
 381                         if (frag->split_by > 0 ||
 382                             !is_frag_child(frag->frag, prev_frag)) {
 383                                 rb_erase(&frag->node, &ci->i_fragtree);
 384                                 if (frag->split_by > 0)
 385                                         ci->i_fragtree_nsplits--;
 386                                 kfree(frag);
 387                         }
 388                         frag = NULL;
 389                 }
 390                 if (!frag) {
 391                         frag = __get_or_create_frag(ci, id);
 392                         if (IS_ERR(frag))
 393                                 continue;
 394                 }
 395                 if (frag->split_by == 0)
 396                         ci->i_fragtree_nsplits++;
 397                 frag->split_by = split_by;
 398                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 399                 prev_frag = frag;
 400         }
 401         while (rb_node) {
 402                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 403                 rb_node = rb_next(rb_node);
 404                 /* delete stale split/leaf node */
 405                 if (frag->split_by > 0 ||
 406                     !is_frag_child(frag->frag, prev_frag)) {
 407                         rb_erase(&frag->node, &ci->i_fragtree);
 408                         if (frag->split_by > 0)
 409                                 ci->i_fragtree_nsplits--;
 410                         kfree(frag);
 411                 }
 412         }
 413 out_unlock:
 414         mutex_unlock(&ci->i_fragtree_mutex);
 415         return 0;
 416 }
 417
 418 /*
 419  * initialize a newly allocated inode.
 420  */
 421 struct inode *ceph_alloc_inode(struct super_block *sb)
 422 {
 423         struct ceph_inode_info *ci;
 424         int i;
 425
 426         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 427         if (!ci)
 428                 return NULL;
 429
 430         dout("alloc_inode %p\n", &ci->vfs_inode);
 431
 432         spin_lock_init(&ci->i_ceph_lock);
 433
 434         ci->i_version = 0;
 435         ci->i_inline_version = 0;
 436         ci->i_time_warp_seq = 0;
 437         ci->i_ceph_flags = 0;
 438         atomic64_set(&ci->i_ordered_count, 1);
 439         atomic64_set(&ci->i_release_count, 1);
 440         atomic64_set(&ci->i_complete_seq[0], 0);
 441         atomic64_set(&ci->i_complete_seq[1], 0);
 442         ci->i_symlink = NULL;
 443
 444         ci->i_max_bytes = 0;
 445         ci->i_max_files = 0;
 446
 447         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 448         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 449
 450         ci->i_fragtree = RB_ROOT;
 451         mutex_init(&ci->i_fragtree_mutex);
 452
 453         ci->i_xattrs.blob = NULL;
 454         ci->i_xattrs.prealloc_blob = NULL;
 455         ci->i_xattrs.dirty = false;
 456         ci->i_xattrs.index = RB_ROOT;
 457         ci->i_xattrs.count = 0;
 458         ci->i_xattrs.names_size = 0;
 459         ci->i_xattrs.vals_size = 0;
 460         ci->i_xattrs.version = 0;
 461         ci->i_xattrs.index_version = 0;
 462
 463         ci->i_caps = RB_ROOT;
 464         ci->i_auth_cap = NULL;
 465         ci->i_dirty_caps = 0;
 466         ci->i_flushing_caps = 0;
 467         INIT_LIST_HEAD(&ci->i_dirty_item);
 468         INIT_LIST_HEAD(&ci->i_flushing_item);
 469         ci->i_prealloc_cap_flush = NULL;
 470         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 471         init_waitqueue_head(&ci->i_cap_wq);
 472         ci->i_hold_caps_min = 0;
 473         ci->i_hold_caps_max = 0;
 474         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 475         INIT_LIST_HEAD(&ci->i_cap_snaps);
 476         ci->i_head_snapc = NULL;
 477         ci->i_snap_caps = 0;
 478
 479         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 480                 ci->i_nr_by_mode[i] = 0;
 481
 482         mutex_init(&ci->i_truncate_mutex);
 483         ci->i_truncate_seq = 0;
 484         ci->i_truncate_size = 0;
 485         ci->i_truncate_pending = 0;
 486
 487         ci->i_max_size = 0;
 488         ci->i_reported_size = 0;
 489         ci->i_wanted_max_size = 0;
 490         ci->i_requested_max_size = 0;
 491
 492         ci->i_pin_ref = 0;
 493         ci->i_rd_ref = 0;
 494         ci->i_rdcache_ref = 0;
 495         ci->i_wr_ref = 0;
 496         ci->i_wb_ref = 0;
 497         ci->i_wrbuffer_ref = 0;
 498         ci->i_wrbuffer_ref_head = 0;
 499         atomic_set(&ci->i_filelock_ref, 0);
 500         atomic_set(&ci->i_shared_gen, 0);
 501         ci->i_rdcache_gen = 0;
 502         ci->i_rdcache_revoking = 0;
 503
 504         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 505         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 506         spin_lock_init(&ci->i_unsafe_lock);
 507
 508         ci->i_snap_realm = NULL;
 509         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 510         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 511
 512         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 513         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 514
 515         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 516
 517         ceph_fscache_inode_init(ci);
 518
 519         return &ci->vfs_inode;
 520 }
 521
 522 static void ceph_i_callback(struct rcu_head *head)
 523 {
 524         struct inode *inode = container_of(head, struct inode, i_rcu);
 525         struct ceph_inode_info *ci = ceph_inode(inode);
 526
 527         kmem_cache_free(ceph_inode_cachep, ci);
 528 }
 529
 530 void ceph_destroy_inode(struct inode *inode)
 531 {
 532         struct ceph_inode_info *ci = ceph_inode(inode);
 533         struct ceph_inode_frag *frag;
 534         struct rb_node *n;
 535
 536         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 537
 538         ceph_fscache_unregister_inode_cookie(ci);
 539
 540         ceph_queue_caps_release(inode);
 541
 542         if (__ceph_has_any_quota(ci))
 543                 ceph_adjust_quota_realms_count(inode, false);
 544
 545         /*
 546          * we may still have a snap_realm reference if there are stray
 547          * caps in i_snap_caps.
 548          */
 549         if (ci->i_snap_realm) {
 550                 struct ceph_mds_client *mdsc =
 551                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 552                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 553
 554                 dout(" dropping residual ref to snap realm %p\n", realm);
 555                 spin_lock(&realm->inodes_with_caps_lock);
 556                 list_del_init(&ci->i_snap_realm_item);
 557                 ci->i_snap_realm = NULL;
 558                 if (realm->ino == ci->i_vino.ino)
 559                         realm->inode = NULL;
 560                 spin_unlock(&realm->inodes_with_caps_lock);
 561                 ceph_put_snap_realm(mdsc, realm);
 562         }
 563
 564         kfree(ci->i_symlink);
 565         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 566                 frag = rb_entry(n, struct ceph_inode_frag, node);
 567                 rb_erase(n, &ci->i_fragtree);
 568                 kfree(frag);
 569         }
 570         ci->i_fragtree_nsplits = 0;
 571
 572         __ceph_destroy_xattrs(ci);
 573         if (ci->i_xattrs.blob)
 574                 ceph_buffer_put(ci->i_xattrs.blob);
 575         if (ci->i_xattrs.prealloc_blob)
 576                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 577
 578         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 579
 580         call_rcu(&inode->i_rcu, ceph_i_callback);
 581 }
 582
 583 int ceph_drop_inode(struct inode *inode)
 584 {
 585         /*
 586          * Positve dentry and corresponding inode are always accompanied
 587          * in MDS reply. So no need to keep inode in the cache after
 588          * dropping all its aliases.
 589          */
 590         return 1;
 591 }
 592
 593 static inline blkcnt_t calc_inode_blocks(u64 size)
 594 {
 595         return (size + (1<<9) - 1) >> 9;
 596 }
 597
 598 /*
 599  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 600  * careful because either the client or MDS may have more up to date
 601  * info, depending on which capabilities are held, and whether
 602  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 603  * and size are monotonically increasing, except when utimes() or
 604  * truncate() increments the corresponding _seq values.)
 605  */
 606 int ceph_fill_file_size(struct inode *inode, int issued,
 607                         u32 truncate_seq, u64 truncate_size, u64 size)
 608 {
 609         struct ceph_inode_info *ci = ceph_inode(inode);
 610         int queue_trunc = 0;
 611
 612         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 613             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 614                 dout("size %lld -> %llu\n", inode->i_size, size);
 615                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 616                         pr_err("fill_file_size non-zero size for directory\n");
 617                         size = 0;
 618                 }
 619                 i_size_write(inode, size);
 620                 inode->i_blocks = calc_inode_blocks(size);
 621                 ci->i_reported_size = size;
 622                 if (truncate_seq != ci->i_truncate_seq) {
 623                         dout("truncate_seq %u -> %u\n",
 624                              ci->i_truncate_seq, truncate_seq);
 625                         ci->i_truncate_seq = truncate_seq;
 626
 627                         /* the MDS should have revoked these caps */
 628                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 629                                                CEPH_CAP_FILE_RD |
 630                                                CEPH_CAP_FILE_WR |
 631                                                CEPH_CAP_FILE_LAZYIO));
 632                         /*
 633                          * If we hold relevant caps, or in the case where we're
 634                          * not the only client referencing this file and we
 635                          * don't hold those caps, then we need to check whether
 636                          * the file is either opened or mmaped
 637                          */
 638                         if ((issued & (CEPH_CAP_FILE_CACHE|
 639                                        CEPH_CAP_FILE_BUFFER)) ||
 640                             mapping_mapped(inode->i_mapping) ||
 641                             __ceph_caps_file_wanted(ci)) {
 642                                 ci->i_truncate_pending++;
 643                                 queue_trunc = 1;
 644                         }
 645                 }
 646         }
 647         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 648             ci->i_truncate_size != truncate_size) {
 649                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 650                      truncate_size);
 651                 ci->i_truncate_size = truncate_size;
 652         }
 653
 654         if (queue_trunc)
 655                 ceph_fscache_invalidate(inode);
 656
 657         return queue_trunc;
 658 }
 659
 660 void ceph_fill_file_time(struct inode *inode, int issued,
 661                          u64 time_warp_seq, struct timespec *ctime,
 662                          struct timespec *mtime, struct timespec *atime)
 663 {
 664         struct ceph_inode_info *ci = ceph_inode(inode);
 665         struct timespec64 ctime64 = timespec_to_timespec64(*ctime);
 666         struct timespec64 mtime64 = timespec_to_timespec64(*mtime);
 667         struct timespec64 atime64 = timespec_to_timespec64(*atime);
 668         int warn = 0;
 669
 670         if (issued & (CEPH_CAP_FILE_EXCL|
 671                       CEPH_CAP_FILE_WR|
 672                       CEPH_CAP_FILE_BUFFER|
 673                       CEPH_CAP_AUTH_EXCL|
 674                       CEPH_CAP_XATTR_EXCL)) {
 675                 if (ci->i_version == 0 ||
 676                     timespec64_compare(&ctime64, &inode->i_ctime) > 0) {
 677                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 678                              (long long)inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 679                              (long long)ctime->tv_sec, ctime->tv_nsec);
 680                         inode->i_ctime = ctime64;
 681                 }
 682                 if (ci->i_version == 0 ||
 683                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 684                         /* the MDS did a utimes() */
 685                         dout("mtime %lld.%09ld -> %lld.%09ld "
 686                              "tw %d -> %d\n",
 687                              (long long)inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 688                              (long long)mtime->tv_sec, mtime->tv_nsec,
 689                              ci->i_time_warp_seq, (int)time_warp_seq);
 690
 691                         inode->i_mtime = mtime64;
 692                         inode->i_atime = atime64;
 693                         ci->i_time_warp_seq = time_warp_seq;
 694                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 695                         /* nobody did utimes(); take the max */
 696                         if (timespec64_compare(&mtime64, &inode->i_mtime) > 0) {
 697                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 698                                      (long long)inode->i_mtime.tv_sec,
 699                                      inode->i_mtime.tv_nsec,
 700                                      (long long)mtime->tv_sec, mtime->tv_nsec);
 701                                 inode->i_mtime = mtime64;
 702                         }
 703                         if (timespec64_compare(&atime64, &inode->i_atime) > 0) {
 704                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 705                                      (long long)inode->i_atime.tv_sec,
 706                                      inode->i_atime.tv_nsec,
 707                                      (long long)atime->tv_sec, atime->tv_nsec);
 708                                 inode->i_atime = atime64;
 709                         }
 710                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 711                         /* we did a utimes(); ignore mds values */
 712                 } else {
 713                         warn = 1;
 714                 }
 715         } else {
 716                 /* we have no write|excl caps; whatever the MDS says is true */
 717                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 718                         inode->i_ctime = ctime64;
 719                         inode->i_mtime = mtime64;
 720                         inode->i_atime = atime64;
 721                         ci->i_time_warp_seq = time_warp_seq;
 722                 } else {
 723                         warn = 1;
 724                 }
 725         }
 726         if (warn) /* time_warp_seq shouldn't go backwards */
 727                 dout("%p mds time_warp_seq %llu < %u\n",
 728                      inode, time_warp_seq, ci->i_time_warp_seq);
 729 }
 730
 731 /*
 732  * Populate an inode based on info from mds.  May be called on new or
 733  * existing inodes.
 734  */
 735 static int fill_inode(struct inode *inode, struct page *locked_page,
 736                       struct ceph_mds_reply_info_in *iinfo,
 737                       struct ceph_mds_reply_dirfrag *dirinfo,
 738                       struct ceph_mds_session *session,
 739                       unsigned long ttl_from, int cap_fmode,
 740                       struct ceph_cap_reservation *caps_reservation)
 741 {
 742         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 743         struct ceph_mds_reply_inode *info = iinfo->in;
 744         struct ceph_inode_info *ci = ceph_inode(inode);
 745         int issued, new_issued, info_caps;
 746         struct timespec mtime, atime, ctime;
 747         struct ceph_buffer *xattr_blob = NULL;
 748         struct ceph_string *pool_ns = NULL;
 749         struct ceph_cap *new_cap = NULL;
 750         int err = 0;
 751         bool wake = false;
 752         bool queue_trunc = false;
 753         bool new_version = false;
 754         bool fill_inline = false;
 755
 756         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 757              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 758              ci->i_version);
 759
 760         info_caps = le32_to_cpu(info->cap.caps);
 761
 762         /* prealloc new cap struct */
 763         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
 764                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 765
 766         /*
 767          * prealloc xattr data, if it looks like we'll need it.  only
 768          * if len > 4 (meaning there are actually xattrs; the first 4
 769          * bytes are the xattr count).
 770          */
 771         if (iinfo->xattr_len > 4) {
 772                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 773                 if (!xattr_blob)
 774                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 775                                iinfo->xattr_len);
 776         }
 777
 778         if (iinfo->pool_ns_len > 0)
 779                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 780                                                      iinfo->pool_ns_len);
 781
 782         spin_lock(&ci->i_ceph_lock);
 783
 784         /*
 785          * provided version will be odd if inode value is projected,
 786          * even if stable.  skip the update if we have newer stable
 787          * info (ours>=theirs, e.g. due to racing mds replies), unless
 788          * we are getting projected (unstable) info (in which case the
 789          * version is odd, and we want ours>theirs).
 790          *   us   them
 791          *   2    2     skip
 792          *   3    2     skip
 793          *   3    3     update
 794          */
 795         if (ci->i_version == 0 ||
 796             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 797              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 798                 new_version = true;
 799
 800         __ceph_caps_issued(ci, &issued);
 801         issued |= __ceph_caps_dirty(ci);
 802         new_issued = ~issued & info_caps;
 803
 804         /* update inode */
 805         inode->i_rdev = le32_to_cpu(info->rdev);
 806         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 807
 808         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 809
 810         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 811             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 812                 inode->i_mode = le32_to_cpu(info->mode);
 813                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 814                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 815                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 816                      from_kuid(&init_user_ns, inode->i_uid),
 817                      from_kgid(&init_user_ns, inode->i_gid));
 818         }
 819
 820         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 821             (issued & CEPH_CAP_LINK_EXCL) == 0)
 822                 set_nlink(inode, le32_to_cpu(info->nlink));
 823
 824         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 825                 /* be careful with mtime, atime, size */
 826                 ceph_decode_timespec(&atime, &info->atime);
 827                 ceph_decode_timespec(&mtime, &info->mtime);
 828                 ceph_decode_timespec(&ctime, &info->ctime);
 829                 ceph_fill_file_time(inode, issued,
 830                                 le32_to_cpu(info->time_warp_seq),
 831                                 &ctime, &mtime, &atime);
 832         }
 833
 834         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 835                 ci->i_files = le64_to_cpu(info->files);
 836                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 837         }
 838
 839         if (new_version ||
 840             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 841                 s64 old_pool = ci->i_layout.pool_id;
 842                 struct ceph_string *old_ns;
 843
 844                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 845                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 846                                         lockdep_is_held(&ci->i_ceph_lock));
 847                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 848
 849                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 850                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 851
 852                 pool_ns = old_ns;
 853
 854                 queue_trunc = ceph_fill_file_size(inode, issued,
 855                                         le32_to_cpu(info->truncate_seq),
 856                                         le64_to_cpu(info->truncate_size),
 857                                         le64_to_cpu(info->size));
 858                 /* only update max_size on auth cap */
 859                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 860                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 861                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 862                                         le64_to_cpu(info->max_size));
 863                         ci->i_max_size = le64_to_cpu(info->max_size);
 864                 }
 865         }
 866
 867         /* layout and rstat are not tracked by capability, update them if
 868          * the inode info is from auth mds */
 869         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 870                 if (S_ISDIR(inode->i_mode)) {
 871                         ci->i_dir_layout = iinfo->dir_layout;
 872                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 873                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 874                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 875                         ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 876                 }
 877         }
 878
 879         /* xattrs */
 880         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 881         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 882             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 883                 if (ci->i_xattrs.blob)
 884                         ceph_buffer_put(ci->i_xattrs.blob);
 885                 ci->i_xattrs.blob = xattr_blob;
 886                 if (xattr_blob)
 887                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 888                                iinfo->xattr_data, iinfo->xattr_len);
 889                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 890                 ceph_forget_all_cached_acls(inode);
 891                 xattr_blob = NULL;
 892         }
 893
 894         /* finally update i_version */
 895         if (le64_to_cpu(info->version) > ci->i_version)
 896                 ci->i_version = le64_to_cpu(info->version);
 897
 898         inode->i_mapping->a_ops = &ceph_aops;
 899
 900         switch (inode->i_mode & S_IFMT) {
 901         case S_IFIFO:
 902         case S_IFBLK:
 903         case S_IFCHR:
 904         case S_IFSOCK:
 905                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 906                 inode->i_op = &ceph_file_iops;
 907                 break;
 908         case S_IFREG:
 909                 inode->i_op = &ceph_file_iops;
 910                 inode->i_fop = &ceph_file_fops;
 911                 break;
 912         case S_IFLNK:
 913                 inode->i_op = &ceph_symlink_iops;
 914                 if (!ci->i_symlink) {
 915                         u32 symlen = iinfo->symlink_len;
 916                         char *sym;
 917
 918                         spin_unlock(&ci->i_ceph_lock);
 919
 920                         if (symlen != i_size_read(inode)) {
 921                                 pr_err("fill_inode %llx.%llx BAD symlink "
 922                                         "size %lld\n", ceph_vinop(inode),
 923                                         i_size_read(inode));
 924                                 i_size_write(inode, symlen);
 925                                 inode->i_blocks = calc_inode_blocks(symlen);
 926                         }
 927
 928                         err = -ENOMEM;
 929                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 930                         if (!sym)
 931                                 goto out;
 932
 933                         spin_lock(&ci->i_ceph_lock);
 934                         if (!ci->i_symlink)
 935                                 ci->i_symlink = sym;
 936                         else
 937                                 kfree(sym); /* lost a race */
 938                 }
 939                 inode->i_link = ci->i_symlink;
 940                 break;
 941         case S_IFDIR:
 942                 inode->i_op = &ceph_dir_iops;
 943                 inode->i_fop = &ceph_dir_fops;
 944                 break;
 945         default:
 946                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 947                        ceph_vinop(inode), inode->i_mode);
 948         }
 949
 950         /* were we issued a capability? */
 951         if (info_caps) {
 952                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 953                         ceph_add_cap(inode, session,
 954                                      le64_to_cpu(info->cap.cap_id),
 955                                      cap_fmode, info_caps,
 956                                      le32_to_cpu(info->cap.wanted),
 957                                      le32_to_cpu(info->cap.seq),
 958                                      le32_to_cpu(info->cap.mseq),
 959                                      le64_to_cpu(info->cap.realm),
 960                                      info->cap.flags, &new_cap);
 961
 962                         /* set dir completion flag? */
 963                         if (S_ISDIR(inode->i_mode) &&
 964                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 965                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 966                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 967                             !__ceph_dir_is_complete(ci)) {
 968                                 dout(" marking %p complete (empty)\n", inode);
 969                                 i_size_write(inode, 0);
 970                                 __ceph_dir_set_complete(ci,
 971                                         atomic64_read(&ci->i_release_count),
 972                                         atomic64_read(&ci->i_ordered_count));
 973                         }
 974
 975                         wake = true;
 976                 } else {
 977                         dout(" %p got snap_caps %s\n", inode,
 978                              ceph_cap_string(info_caps));
 979                         ci->i_snap_caps |= info_caps;
 980                         if (cap_fmode >= 0)
 981                                 __ceph_get_fmode(ci, cap_fmode);
 982                 }
 983         } else if (cap_fmode >= 0) {
 984                 pr_warn("mds issued no caps on %llx.%llx\n",
 985                            ceph_vinop(inode));
 986                 __ceph_get_fmode(ci, cap_fmode);
 987         }
 988
 989         if (iinfo->inline_version > 0 &&
 990             iinfo->inline_version >= ci->i_inline_version) {
 991                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 992                 ci->i_inline_version = iinfo->inline_version;
 993                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 994                     (locked_page || (info_caps & cache_caps)))
 995                         fill_inline = true;
 996         }
 997
 998         spin_unlock(&ci->i_ceph_lock);
 999
1000         if (fill_inline)
1001                 ceph_fill_inline_data(inode, locked_page,
1002                                       iinfo->inline_data, iinfo->inline_len);
1003
1004         if (wake)
1005                 wake_up_all(&ci->i_cap_wq);
1006
1007         /* queue truncate if we saw i_size decrease */
1008         if (queue_trunc)
1009                 ceph_queue_vmtruncate(inode);
1010
1011         /* populate frag tree */
1012         if (S_ISDIR(inode->i_mode))
1013                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1014
1015         /* update delegation info? */
1016         if (dirinfo)
1017                 ceph_fill_dirfrag(inode, dirinfo);
1018
1019         err = 0;
1020 out:
1021         if (new_cap)
1022                 ceph_put_cap(mdsc, new_cap);
1023         if (xattr_blob)
1024                 ceph_buffer_put(xattr_blob);
1025         ceph_put_string(pool_ns);
1026         return err;
1027 }
1028
1029 /*
1030  * caller should hold session s_mutex.
1031  */
1032 static void update_dentry_lease(struct dentry *dentry,
1033                                 struct ceph_mds_reply_lease *lease,
1034                                 struct ceph_mds_session *session,
1035                                 unsigned long from_time,
1036                                 struct ceph_vino *tgt_vino,
1037                                 struct ceph_vino *dir_vino)
1038 {
1039         struct ceph_dentry_info *di = ceph_dentry(dentry);
1040         long unsigned duration = le32_to_cpu(lease->duration_ms);
1041         long unsigned ttl = from_time + (duration * HZ) / 1000;
1042         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1043         struct inode *dir;
1044         struct ceph_mds_session *old_lease_session = NULL;
1045
1046         /*
1047          * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1048          * we expect a negative dentry.
1049          */
1050         if (!tgt_vino && d_really_is_positive(dentry))
1051                 return;
1052
1053         if (tgt_vino && (d_really_is_negative(dentry) ||
1054                         !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1055                 return;
1056
1057         spin_lock(&dentry->d_lock);
1058         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1059              dentry, duration, ttl);
1060
1061         dir = d_inode(dentry->d_parent);
1062
1063         /* make sure parent matches dir_vino */
1064         if (!ceph_ino_compare(dir, dir_vino))
1065                 goto out_unlock;
1066
1067         /* only track leases on regular dentries */
1068         if (ceph_snap(dir) != CEPH_NOSNAP)
1069                 goto out_unlock;
1070
1071         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1072
1073         if (duration == 0)
1074                 goto out_unlock;
1075
1076         if (di->lease_gen == session->s_cap_gen &&
1077             time_before(ttl, di->time))
1078                 goto out_unlock;  /* we already have a newer lease. */
1079
1080         if (di->lease_session && di->lease_session != session) {
1081                 old_lease_session = di->lease_session;
1082                 di->lease_session = NULL;
1083         }
1084
1085         ceph_dentry_lru_touch(dentry);
1086
1087         if (!di->lease_session)
1088                 di->lease_session = ceph_get_mds_session(session);
1089         di->lease_gen = session->s_cap_gen;
1090         di->lease_seq = le32_to_cpu(lease->seq);
1091         di->lease_renew_after = half_ttl;
1092         di->lease_renew_from = 0;
1093         di->time = ttl;
1094 out_unlock:
1095         spin_unlock(&dentry->d_lock);
1096         if (old_lease_session)
1097                 ceph_put_mds_session(old_lease_session);
1098 }
1099
1100 /*
1101  * splice a dentry to an inode.
1102  * caller must hold directory i_mutex for this to be safe.
1103  */
1104 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1105 {
1106         struct dentry *realdn;
1107
1108         BUG_ON(d_inode(dn));
1109
1110         if (S_ISDIR(in->i_mode)) {
1111                 /* If inode is directory, d_splice_alias() below will remove
1112                  * 'realdn' from its origin parent. We need to ensure that
1113                  * origin parent's readdir cache will not reference 'realdn'
1114                  */
1115                 realdn = d_find_any_alias(in);
1116                 if (realdn) {
1117                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1118                         spin_lock(&realdn->d_lock);
1119
1120                         realdn->d_op->d_prune(realdn);
1121
1122                         di->time = jiffies;
1123                         di->lease_shared_gen = 0;
1124                         di->offset = 0;
1125
1126                         spin_unlock(&realdn->d_lock);
1127                         dput(realdn);
1128                 }
1129         }
1130
1131         /* dn must be unhashed */
1132         if (!d_unhashed(dn))
1133                 d_drop(dn);
1134         realdn = d_splice_alias(in, dn);
1135         if (IS_ERR(realdn)) {
1136                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1137                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1138                 dput(dn);
1139                 dn = realdn; /* note realdn contains the error */
1140                 goto out;
1141         } else if (realdn) {
1142                 dout("dn %p (%d) spliced with %p (%d) "
1143                      "inode %p ino %llx.%llx\n",
1144                      dn, d_count(dn),
1145                      realdn, d_count(realdn),
1146                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1147                 dput(dn);
1148                 dn = realdn;
1149         } else {
1150                 BUG_ON(!ceph_dentry(dn));
1151                 dout("dn %p attached to %p ino %llx.%llx\n",
1152                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1153         }
1154 out:
1155         return dn;
1156 }
1157
1158 /*
1159  * Incorporate results into the local cache.  This is either just
1160  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1161  * after a lookup).
1162  *
1163  * A reply may contain
1164  *         a directory inode along with a dentry.
1165  *  and/or a target inode
1166  *
1167  * Called with snap_rwsem (read).
1168  */
1169 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1170 {
1171         struct ceph_mds_session *session = req->r_session;
1172         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1173         struct inode *in = NULL;
1174         struct ceph_vino tvino, dvino;
1175         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1176         int err = 0;
1177
1178         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1179              rinfo->head->is_dentry, rinfo->head->is_target);
1180
1181         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1182                 dout("fill_trace reply is empty!\n");
1183                 if (rinfo->head->result == 0 && req->r_parent)
1184                         ceph_invalidate_dir_request(req);
1185                 return 0;
1186         }
1187
1188         if (rinfo->head->is_dentry) {
1189                 struct inode *dir = req->r_parent;
1190
1191                 if (dir) {
1192                         err = fill_inode(dir, NULL,
1193                                          &rinfo->diri, rinfo->dirfrag,
1194                                          session, req->r_request_started, -1,
1195                                          &req->r_caps_reservation);
1196                         if (err < 0)
1197                                 goto done;
1198                 } else {
1199                         WARN_ON_ONCE(1);
1200                 }
1201
1202                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1203                         struct qstr dname;
1204                         struct dentry *dn, *parent;
1205
1206                         BUG_ON(!rinfo->head->is_target);
1207                         BUG_ON(req->r_dentry);
1208
1209                         parent = d_find_any_alias(dir);
1210                         BUG_ON(!parent);
1211
1212                         dname.name = rinfo->dname;
1213                         dname.len = rinfo->dname_len;
1214                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1215                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1216                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1217 retry_lookup:
1218                         dn = d_lookup(parent, &dname);
1219                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1220                              parent, dname.len, dname.name, dn);
1221
1222                         if (!dn) {
1223                                 dn = d_alloc(parent, &dname);
1224                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1225                                      dname.len, dname.name, dn);
1226                                 if (!dn) {
1227                                         dput(parent);
1228                                         err = -ENOMEM;
1229                                         goto done;
1230                                 }
1231                                 err = 0;
1232                         } else if (d_really_is_positive(dn) &&
1233                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1234                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1235                                 dout(" dn %p points to wrong inode %p\n",
1236                                      dn, d_inode(dn));
1237                                 ceph_dir_clear_ordered(dir);
1238                                 d_delete(dn);
1239                                 dput(dn);
1240                                 goto retry_lookup;
1241                         }
1242
1243                         req->r_dentry = dn;
1244                         dput(parent);
1245                 }
1246         }
1247
1248         if (rinfo->head->is_target) {
1249                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1250                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1251
1252                 in = ceph_get_inode(sb, tvino);
1253                 if (IS_ERR(in)) {
1254                         err = PTR_ERR(in);
1255                         goto done;
1256                 }
1257                 req->r_target_inode = in;
1258
1259                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1260                                 session, req->r_request_started,
1261                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1262                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1263                                 &req->r_caps_reservation);
1264                 if (err < 0) {
1265                         pr_err("fill_inode badness %p %llx.%llx\n",
1266                                 in, ceph_vinop(in));
1267                         goto done;
1268                 }
1269         }
1270
1271         /*
1272          * ignore null lease/binding on snapdir ENOENT, or else we
1273          * will have trouble splicing in the virtual snapdir later
1274          */
1275         if (rinfo->head->is_dentry &&
1276             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1277             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1278             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1279                                                fsc->mount_options->snapdir_name,
1280                                                req->r_dentry->d_name.len))) {
1281                 /*
1282                  * lookup link rename   : null -> possibly existing inode
1283                  * mknod symlink mkdir  : null -> new inode
1284                  * unlink               : linked -> null
1285                  */
1286                 struct inode *dir = req->r_parent;
1287                 struct dentry *dn = req->r_dentry;
1288                 bool have_dir_cap, have_lease;
1289
1290                 BUG_ON(!dn);
1291                 BUG_ON(!dir);
1292                 BUG_ON(d_inode(dn->d_parent) != dir);
1293
1294                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1295                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1296
1297                 BUG_ON(ceph_ino(dir) != dvino.ino);
1298                 BUG_ON(ceph_snap(dir) != dvino.snap);
1299
1300                 /* do we have a lease on the whole dir? */
1301                 have_dir_cap =
1302                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1303                          CEPH_CAP_FILE_SHARED);
1304
1305                 /* do we have a dn lease? */
1306                 have_lease = have_dir_cap ||
1307                         le32_to_cpu(rinfo->dlease->duration_ms);
1308                 if (!have_lease)
1309                         dout("fill_trace  no dentry lease or dir cap\n");
1310
1311                 /* rename? */
1312                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1313                         struct inode *olddir = req->r_old_dentry_dir;
1314                         BUG_ON(!olddir);
1315
1316                         dout(" src %p '%pd' dst %p '%pd'\n",
1317                              req->r_old_dentry,
1318                              req->r_old_dentry,
1319                              dn, dn);
1320                         dout("fill_trace doing d_move %p -> %p\n",
1321                              req->r_old_dentry, dn);
1322
1323                         /* d_move screws up sibling dentries' offsets */
1324                         ceph_dir_clear_ordered(dir);
1325                         ceph_dir_clear_ordered(olddir);
1326
1327                         d_move(req->r_old_dentry, dn);
1328                         dout(" src %p '%pd' dst %p '%pd'\n",
1329                              req->r_old_dentry,
1330                              req->r_old_dentry,
1331                              dn, dn);
1332
1333                         /* ensure target dentry is invalidated, despite
1334                            rehashing bug in vfs_rename_dir */
1335                         ceph_invalidate_dentry_lease(dn);
1336
1337                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1338                              ceph_dentry(req->r_old_dentry)->offset);
1339
1340                         dn = req->r_old_dentry;  /* use old_dentry */
1341                 }
1342
1343                 /* null dentry? */
1344                 if (!rinfo->head->is_target) {
1345                         dout("fill_trace null dentry\n");
1346                         if (d_really_is_positive(dn)) {
1347                                 dout("d_delete %p\n", dn);
1348                                 ceph_dir_clear_ordered(dir);
1349                                 d_delete(dn);
1350                         } else if (have_lease) {
1351                                 if (d_unhashed(dn))
1352                                         d_add(dn, NULL);
1353                                 update_dentry_lease(dn, rinfo->dlease,
1354                                                     session,
1355                                                     req->r_request_started,
1356                                                     NULL, &dvino);
1357                         }
1358                         goto done;
1359                 }
1360
1361                 /* attach proper inode */
1362                 if (d_really_is_negative(dn)) {
1363                         ceph_dir_clear_ordered(dir);
1364                         ihold(in);
1365                         dn = splice_dentry(dn, in);
1366                         if (IS_ERR(dn)) {
1367                                 err = PTR_ERR(dn);
1368                                 goto done;
1369                         }
1370                         req->r_dentry = dn;  /* may have spliced */
1371                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1372                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1373                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1374                              ceph_vinop(in));
1375                         d_invalidate(dn);
1376                         have_lease = false;
1377                 }
1378
1379                 if (have_lease) {
1380                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1381                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1382                         update_dentry_lease(dn, rinfo->dlease, session,
1383                                             req->r_request_started,
1384                                             &tvino, &dvino);
1385                 }
1386                 dout(" final dn %p\n", dn);
1387         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1388                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1389                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1390                 struct dentry *dn = req->r_dentry;
1391                 struct inode *dir = req->r_parent;
1392
1393                 /* fill out a snapdir LOOKUPSNAP dentry */
1394                 BUG_ON(!dn);
1395                 BUG_ON(!dir);
1396                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1397                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1398                 ceph_dir_clear_ordered(dir);
1399                 ihold(in);
1400                 dn = splice_dentry(dn, in);
1401                 if (IS_ERR(dn)) {
1402                         err = PTR_ERR(dn);
1403                         goto done;
1404                 }
1405                 req->r_dentry = dn;  /* may have spliced */
1406         } else if (rinfo->head->is_dentry) {
1407                 struct ceph_vino *ptvino = NULL;
1408
1409                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1410                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1411                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1412                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1413
1414                         if (rinfo->head->is_target) {
1415                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1416                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1417                                 ptvino = &tvino;
1418                         }
1419
1420                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1421                                 session, req->r_request_started, ptvino,
1422                                 &dvino);
1423                 } else {
1424                         dout("%s: no dentry lease or dir cap\n", __func__);
1425                 }
1426         }
1427 done:
1428         dout("fill_trace done err=%d\n", err);
1429         return err;
1430 }
1431
1432 /*
1433  * Prepopulate our cache with readdir results, leases, etc.
1434  */
1435 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1436                                            struct ceph_mds_session *session)
1437 {
1438         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1439         int i, err = 0;
1440
1441         for (i = 0; i < rinfo->dir_nr; i++) {
1442                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1443                 struct ceph_vino vino;
1444                 struct inode *in;
1445                 int rc;
1446
1447                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1448                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1449
1450                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1451                 if (IS_ERR(in)) {
1452                         err = PTR_ERR(in);
1453                         dout("new_inode badness got %d\n", err);
1454                         continue;
1455                 }
1456                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1457                                 req->r_request_started, -1,
1458                                 &req->r_caps_reservation);
1459                 if (rc < 0) {
1460                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1461                         err = rc;
1462                 }
1463                 iput(in);
1464         }
1465
1466         return err;
1467 }
1468
1469 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1470 {
1471         if (ctl->page) {
1472                 kunmap(ctl->page);
1473                 put_page(ctl->page);
1474                 ctl->page = NULL;
1475         }
1476 }
1477
1478 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1479                               struct ceph_readdir_cache_control *ctl,
1480                               struct ceph_mds_request *req)
1481 {
1482         struct ceph_inode_info *ci = ceph_inode(dir);
1483         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1484         unsigned idx = ctl->index % nsize;
1485         pgoff_t pgoff = ctl->index / nsize;
1486
1487         if (!ctl->page || pgoff != page_index(ctl->page)) {
1488                 ceph_readdir_cache_release(ctl);
1489                 if (idx == 0)
1490                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1491                 else
1492                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1493                 if (!ctl->page) {
1494                         ctl->index = -1;
1495                         return idx == 0 ? -ENOMEM : 0;
1496                 }
1497                 /* reading/filling the cache are serialized by
1498                  * i_mutex, no need to use page lock */
1499                 unlock_page(ctl->page);
1500                 ctl->dentries = kmap(ctl->page);
1501                 if (idx == 0)
1502                         memset(ctl->dentries, 0, PAGE_SIZE);
1503         }
1504
1505         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1506             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1507                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1508                 ctl->dentries[idx] = dn;
1509                 ctl->index++;
1510         } else {
1511                 dout("disable readdir cache\n");
1512                 ctl->index = -1;
1513         }
1514         return 0;
1515 }
1516
1517 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1518                              struct ceph_mds_session *session)
1519 {
1520         struct dentry *parent = req->r_dentry;
1521         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1522         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1523         struct qstr dname;
1524         struct dentry *dn;
1525         struct inode *in;
1526         int err = 0, skipped = 0, ret, i;
1527         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1528         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1529         u32 last_hash = 0;
1530         u32 fpos_offset;
1531         struct ceph_readdir_cache_control cache_ctl = {};
1532
1533         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1534                 return readdir_prepopulate_inodes_only(req, session);
1535
1536         if (rinfo->hash_order) {
1537                 if (req->r_path2) {
1538                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1539                                                   req->r_path2,
1540                                                   strlen(req->r_path2));
1541                         last_hash = ceph_frag_value(last_hash);
1542                 } else if (rinfo->offset_hash) {
1543                         /* mds understands offset_hash */
1544                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1545                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1546                 }
1547         }
1548
1549         if (rinfo->dir_dir &&
1550             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1551                 dout("readdir_prepopulate got new frag %x -> %x\n",
1552                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1553                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1554                 if (!rinfo->hash_order)
1555                         req->r_readdir_offset = 2;
1556         }
1557
1558         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1559                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1560                      rinfo->dir_nr, parent);
1561         } else {
1562                 dout("readdir_prepopulate %d items under dn %p\n",
1563                      rinfo->dir_nr, parent);
1564                 if (rinfo->dir_dir)
1565                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1566
1567                 if (ceph_frag_is_leftmost(frag) &&
1568                     req->r_readdir_offset == 2 &&
1569                     !(rinfo->hash_order && last_hash)) {
1570                         /* note dir version at start of readdir so we can
1571                          * tell if any dentries get dropped */
1572                         req->r_dir_release_cnt =
1573                                 atomic64_read(&ci->i_release_count);
1574                         req->r_dir_ordered_cnt =
1575                                 atomic64_read(&ci->i_ordered_count);
1576                         req->r_readdir_cache_idx = 0;
1577                 }
1578         }
1579
1580         cache_ctl.index = req->r_readdir_cache_idx;
1581         fpos_offset = req->r_readdir_offset;
1582
1583         /* FIXME: release caps/leases if error occurs */
1584         for (i = 0; i < rinfo->dir_nr; i++) {
1585                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1586                 struct ceph_vino tvino, dvino;
1587
1588                 dname.name = rde->name;
1589                 dname.len = rde->name_len;
1590                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1591
1592                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1593                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1594
1595                 if (rinfo->hash_order) {
1596                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1597                                                  rde->name, rde->name_len);
1598                         hash = ceph_frag_value(hash);
1599                         if (hash != last_hash)
1600                                 fpos_offset = 2;
1601                         last_hash = hash;
1602                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1603                 } else {
1604                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1605                 }
1606
1607 retry_lookup:
1608                 dn = d_lookup(parent, &dname);
1609                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1610                      parent, dname.len, dname.name, dn);
1611
1612                 if (!dn) {
1613                         dn = d_alloc(parent, &dname);
1614                         dout("d_alloc %p '%.*s' = %p\n", parent,
1615                              dname.len, dname.name, dn);
1616                         if (!dn) {
1617                                 dout("d_alloc badness\n");
1618                                 err = -ENOMEM;
1619                                 goto out;
1620                         }
1621                 } else if (d_really_is_positive(dn) &&
1622                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1623                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1624                         struct ceph_dentry_info *di = ceph_dentry(dn);
1625                         dout(" dn %p points to wrong inode %p\n",
1626                              dn, d_inode(dn));
1627
1628                         spin_lock(&dn->d_lock);
1629                         if (di->offset > 0 &&
1630                             di->lease_shared_gen ==
1631                             atomic_read(&ci->i_shared_gen)) {
1632                                 __ceph_dir_clear_ordered(ci);
1633                                 di->offset = 0;
1634                         }
1635                         spin_unlock(&dn->d_lock);
1636
1637                         d_delete(dn);
1638                         dput(dn);
1639                         goto retry_lookup;
1640                 }
1641
1642                 /* inode */
1643                 if (d_really_is_positive(dn)) {
1644                         in = d_inode(dn);
1645                 } else {
1646                         in = ceph_get_inode(parent->d_sb, tvino);
1647                         if (IS_ERR(in)) {
1648                                 dout("new_inode badness\n");
1649                                 d_drop(dn);
1650                                 dput(dn);
1651                                 err = PTR_ERR(in);
1652                                 goto out;
1653                         }
1654                 }
1655
1656                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1657                                  req->r_request_started, -1,
1658                                  &req->r_caps_reservation);
1659                 if (ret < 0) {
1660                         pr_err("fill_inode badness on %p\n", in);
1661                         if (d_really_is_negative(dn))
1662                                 iput(in);
1663                         d_drop(dn);
1664                         err = ret;
1665                         goto next_item;
1666                 }
1667
1668                 if (d_really_is_negative(dn)) {
1669                         struct dentry *realdn;
1670
1671                         if (ceph_security_xattr_deadlock(in)) {
1672                                 dout(" skip splicing dn %p to inode %p"
1673                                      " (security xattr deadlock)\n", dn, in);
1674                                 iput(in);
1675                                 skipped++;
1676                                 goto next_item;
1677                         }
1678
1679                         realdn = splice_dentry(dn, in);
1680                         if (IS_ERR(realdn)) {
1681                                 err = PTR_ERR(realdn);
1682                                 d_drop(dn);
1683                                 dn = NULL;
1684                                 goto next_item;
1685                         }
1686                         dn = realdn;
1687                 }
1688
1689                 ceph_dentry(dn)->offset = rde->offset;
1690
1691                 dvino = ceph_vino(d_inode(parent));
1692                 update_dentry_lease(dn, rde->lease, req->r_session,
1693                                     req->r_request_started, &tvino, &dvino);
1694
1695                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1696                         ret = fill_readdir_cache(d_inode(parent), dn,
1697                                                  &cache_ctl, req);
1698                         if (ret < 0)
1699                                 err = ret;
1700                 }
1701 next_item:
1702                 if (dn)
1703                         dput(dn);
1704         }
1705 out:
1706         if (err == 0 && skipped == 0) {
1707                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1708                 req->r_readdir_cache_idx = cache_ctl.index;
1709         }
1710         ceph_readdir_cache_release(&cache_ctl);
1711         dout("readdir_prepopulate done\n");
1712         return err;
1713 }
1714
1715 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1716 {
1717         struct ceph_inode_info *ci = ceph_inode(inode);
1718         bool ret;
1719
1720         spin_lock(&ci->i_ceph_lock);
1721         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1722         i_size_write(inode, size);
1723         inode->i_blocks = calc_inode_blocks(size);
1724
1725         ret = __ceph_should_report_size(ci);
1726
1727         spin_unlock(&ci->i_ceph_lock);
1728         return ret;
1729 }
1730
1731 /*
1732  * Write back inode data in a worker thread.  (This can't be done
1733  * in the message handler context.)
1734  */
1735 void ceph_queue_writeback(struct inode *inode)
1736 {
1737         ihold(inode);
1738         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1739                        &ceph_inode(inode)->i_wb_work)) {
1740                 dout("ceph_queue_writeback %p\n", inode);
1741         } else {
1742                 dout("ceph_queue_writeback %p failed\n", inode);
1743                 iput(inode);
1744         }
1745 }
1746
1747 static void ceph_writeback_work(struct work_struct *work)
1748 {
1749         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1750                                                   i_wb_work);
1751         struct inode *inode = &ci->vfs_inode;
1752
1753         dout("writeback %p\n", inode);
1754         filemap_fdatawrite(&inode->i_data);
1755         iput(inode);
1756 }
1757
1758 /*
1759  * queue an async invalidation
1760  */
1761 void ceph_queue_invalidate(struct inode *inode)
1762 {
1763         ihold(inode);
1764         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1765                        &ceph_inode(inode)->i_pg_inv_work)) {
1766                 dout("ceph_queue_invalidate %p\n", inode);
1767         } else {
1768                 dout("ceph_queue_invalidate %p failed\n", inode);
1769                 iput(inode);
1770         }
1771 }
1772
1773 /*
1774  * Invalidate inode pages in a worker thread.  (This can't be done
1775  * in the message handler context.)
1776  */
1777 static void ceph_invalidate_work(struct work_struct *work)
1778 {
1779         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1780                                                   i_pg_inv_work);
1781         struct inode *inode = &ci->vfs_inode;
1782         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1783         u32 orig_gen;
1784         int check = 0;
1785
1786         mutex_lock(&ci->i_truncate_mutex);
1787
1788         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1789                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1790                                     inode, ceph_ino(inode));
1791                 mapping_set_error(inode->i_mapping, -EIO);
1792                 truncate_pagecache(inode, 0);
1793                 mutex_unlock(&ci->i_truncate_mutex);
1794                 goto out;
1795         }
1796
1797         spin_lock(&ci->i_ceph_lock);
1798         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1799              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1800         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1801                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1802                         check = 1;
1803                 spin_unlock(&ci->i_ceph_lock);
1804                 mutex_unlock(&ci->i_truncate_mutex);
1805                 goto out;
1806         }
1807         orig_gen = ci->i_rdcache_gen;
1808         spin_unlock(&ci->i_ceph_lock);
1809
1810         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1811                 pr_err("invalidate_pages %p fails\n", inode);
1812         }
1813
1814         spin_lock(&ci->i_ceph_lock);
1815         if (orig_gen == ci->i_rdcache_gen &&
1816             orig_gen == ci->i_rdcache_revoking) {
1817                 dout("invalidate_pages %p gen %d successful\n", inode,
1818                      ci->i_rdcache_gen);
1819                 ci->i_rdcache_revoking--;
1820                 check = 1;
1821         } else {
1822                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1823                      inode, orig_gen, ci->i_rdcache_gen,
1824                      ci->i_rdcache_revoking);
1825                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1826                         check = 1;
1827         }
1828         spin_unlock(&ci->i_ceph_lock);
1829         mutex_unlock(&ci->i_truncate_mutex);
1830 out:
1831         if (check)
1832                 ceph_check_caps(ci, 0, NULL);
1833         iput(inode);
1834 }
1835
1836
1837 /*
1838  * called by trunc_wq;
1839  *
1840  * We also truncate in a separate thread as well.
1841  */
1842 static void ceph_vmtruncate_work(struct work_struct *work)
1843 {
1844         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1845                                                   i_vmtruncate_work);
1846         struct inode *inode = &ci->vfs_inode;
1847
1848         dout("vmtruncate_work %p\n", inode);
1849         __ceph_do_pending_vmtruncate(inode);
1850         iput(inode);
1851 }
1852
1853 /*
1854  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1855  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1856  */
1857 void ceph_queue_vmtruncate(struct inode *inode)
1858 {
1859         struct ceph_inode_info *ci = ceph_inode(inode);
1860
1861         ihold(inode);
1862
1863         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1864                        &ci->i_vmtruncate_work)) {
1865                 dout("ceph_queue_vmtruncate %p\n", inode);
1866         } else {
1867                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1868                      inode, ci->i_truncate_pending);
1869                 iput(inode);
1870         }
1871 }
1872
1873 /*
1874  * Make sure any pending truncation is applied before doing anything
1875  * that may depend on it.
1876  */
1877 void __ceph_do_pending_vmtruncate(struct inode *inode)
1878 {
1879         struct ceph_inode_info *ci = ceph_inode(inode);
1880         u64 to;
1881         int wrbuffer_refs, finish = 0;
1882
1883         mutex_lock(&ci->i_truncate_mutex);
1884 retry:
1885         spin_lock(&ci->i_ceph_lock);
1886         if (ci->i_truncate_pending == 0) {
1887                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1888                 spin_unlock(&ci->i_ceph_lock);
1889                 mutex_unlock(&ci->i_truncate_mutex);
1890                 return;
1891         }
1892
1893         /*
1894          * make sure any dirty snapped pages are flushed before we
1895          * possibly truncate them.. so write AND block!
1896          */
1897         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1898                 spin_unlock(&ci->i_ceph_lock);
1899                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1900                      inode);
1901                 filemap_write_and_wait_range(&inode->i_data, 0,
1902                                              inode->i_sb->s_maxbytes);
1903                 goto retry;
1904         }
1905
1906         /* there should be no reader or writer */
1907         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1908
1909         to = ci->i_truncate_size;
1910         wrbuffer_refs = ci->i_wrbuffer_ref;
1911         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1912              ci->i_truncate_pending, to);
1913         spin_unlock(&ci->i_ceph_lock);
1914
1915         truncate_pagecache(inode, to);
1916
1917         spin_lock(&ci->i_ceph_lock);
1918         if (to == ci->i_truncate_size) {
1919                 ci->i_truncate_pending = 0;
1920                 finish = 1;
1921         }
1922         spin_unlock(&ci->i_ceph_lock);
1923         if (!finish)
1924                 goto retry;
1925
1926         mutex_unlock(&ci->i_truncate_mutex);
1927
1928         if (wrbuffer_refs == 0)
1929                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1930
1931         wake_up_all(&ci->i_cap_wq);
1932 }
1933
1934 /*
1935  * symlinks
1936  */
1937 static const struct inode_operations ceph_symlink_iops = {
1938         .get_link = simple_get_link,
1939         .setattr = ceph_setattr,
1940         .getattr = ceph_getattr,
1941         .listxattr = ceph_listxattr,
1942 };
1943
1944 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1945 {
1946         struct ceph_inode_info *ci = ceph_inode(inode);
1947         const unsigned int ia_valid = attr->ia_valid;
1948         struct ceph_mds_request *req;
1949         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1950         struct ceph_cap_flush *prealloc_cf;
1951         int issued;
1952         int release = 0, dirtied = 0;
1953         int mask = 0;
1954         int err = 0;
1955         int inode_dirty_flags = 0;
1956         bool lock_snap_rwsem = false;
1957         struct timespec ts;
1958
1959         prealloc_cf = ceph_alloc_cap_flush();
1960         if (!prealloc_cf)
1961                 return -ENOMEM;
1962
1963         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1964                                        USE_AUTH_MDS);
1965         if (IS_ERR(req)) {
1966                 ceph_free_cap_flush(prealloc_cf);
1967                 return PTR_ERR(req);
1968         }
1969
1970         spin_lock(&ci->i_ceph_lock);
1971         issued = __ceph_caps_issued(ci, NULL);
1972
1973         if (!ci->i_head_snapc &&
1974             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1975                 lock_snap_rwsem = true;
1976                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1977                         spin_unlock(&ci->i_ceph_lock);
1978                         down_read(&mdsc->snap_rwsem);
1979                         spin_lock(&ci->i_ceph_lock);
1980                         issued = __ceph_caps_issued(ci, NULL);
1981                 }
1982         }
1983
1984         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1985
1986         if (ia_valid & ATTR_UID) {
1987                 dout("setattr %p uid %d -> %d\n", inode,
1988                      from_kuid(&init_user_ns, inode->i_uid),
1989                      from_kuid(&init_user_ns, attr->ia_uid));
1990                 if (issued & CEPH_CAP_AUTH_EXCL) {
1991                         inode->i_uid = attr->ia_uid;
1992                         dirtied |= CEPH_CAP_AUTH_EXCL;
1993                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1994                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1995                         req->r_args.setattr.uid = cpu_to_le32(
1996                                 from_kuid(&init_user_ns, attr->ia_uid));
1997                         mask |= CEPH_SETATTR_UID;
1998                         release |= CEPH_CAP_AUTH_SHARED;
1999                 }
2000         }
2001         if (ia_valid & ATTR_GID) {
2002                 dout("setattr %p gid %d -> %d\n", inode,
2003                      from_kgid(&init_user_ns, inode->i_gid),
2004                      from_kgid(&init_user_ns, attr->ia_gid));
2005                 if (issued & CEPH_CAP_AUTH_EXCL) {
2006                         inode->i_gid = attr->ia_gid;
2007                         dirtied |= CEPH_CAP_AUTH_EXCL;
2008                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2009                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2010                         req->r_args.setattr.gid = cpu_to_le32(
2011                                 from_kgid(&init_user_ns, attr->ia_gid));
2012                         mask |= CEPH_SETATTR_GID;
2013                         release |= CEPH_CAP_AUTH_SHARED;
2014                 }
2015         }
2016         if (ia_valid & ATTR_MODE) {
2017                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2018                      attr->ia_mode);
2019                 if (issued & CEPH_CAP_AUTH_EXCL) {
2020                         inode->i_mode = attr->ia_mode;
2021                         dirtied |= CEPH_CAP_AUTH_EXCL;
2022                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2023                            attr->ia_mode != inode->i_mode) {
2024                         inode->i_mode = attr->ia_mode;
2025                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2026                         mask |= CEPH_SETATTR_MODE;
2027                         release |= CEPH_CAP_AUTH_SHARED;
2028                 }
2029         }
2030
2031         if (ia_valid & ATTR_ATIME) {
2032                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2033                      (long long)inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2034                      (long long)attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2035                 if (issued & CEPH_CAP_FILE_EXCL) {
2036                         ci->i_time_warp_seq++;
2037                         inode->i_atime = attr->ia_atime;
2038                         dirtied |= CEPH_CAP_FILE_EXCL;
2039                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2040                            timespec64_compare(&inode->i_atime,
2041                                             &attr->ia_atime) < 0) {
2042                         inode->i_atime = attr->ia_atime;
2043                         dirtied |= CEPH_CAP_FILE_WR;
2044                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2045                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2046                         ts = timespec64_to_timespec(attr->ia_atime);
2047                         ceph_encode_timespec(&req->r_args.setattr.atime, &ts);
2048                         mask |= CEPH_SETATTR_ATIME;
2049                         release |= CEPH_CAP_FILE_SHARED |
2050                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2051                 }
2052         }
2053         if (ia_valid & ATTR_MTIME) {
2054                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2055                      (long long)inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2056                      (long long)attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2057                 if (issued & CEPH_CAP_FILE_EXCL) {
2058                         ci->i_time_warp_seq++;
2059                         inode->i_mtime = attr->ia_mtime;
2060                         dirtied |= CEPH_CAP_FILE_EXCL;
2061                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2062                            timespec64_compare(&inode->i_mtime,
2063                                             &attr->ia_mtime) < 0) {
2064                         inode->i_mtime = attr->ia_mtime;
2065                         dirtied |= CEPH_CAP_FILE_WR;
2066                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2067                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2068                         ts = timespec64_to_timespec(attr->ia_mtime);
2069                         ceph_encode_timespec(&req->r_args.setattr.mtime, &ts);
2070                         mask |= CEPH_SETATTR_MTIME;
2071                         release |= CEPH_CAP_FILE_SHARED |
2072                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2073                 }
2074         }
2075         if (ia_valid & ATTR_SIZE) {
2076                 dout("setattr %p size %lld -> %lld\n", inode,
2077                      inode->i_size, attr->ia_size);
2078                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2079                     attr->ia_size > inode->i_size) {
2080                         i_size_write(inode, attr->ia_size);
2081                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2082                         ci->i_reported_size = attr->ia_size;
2083                         dirtied |= CEPH_CAP_FILE_EXCL;
2084                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2085                            attr->ia_size != inode->i_size) {
2086                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2087                         req->r_args.setattr.old_size =
2088                                 cpu_to_le64(inode->i_size);
2089                         mask |= CEPH_SETATTR_SIZE;
2090                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2091                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2092                 }
2093         }
2094
2095         /* these do nothing */
2096         if (ia_valid & ATTR_CTIME) {
2097                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2098                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2099                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2100                      (long long)inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2101                      (long long)attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2102                      only ? "ctime only" : "ignored");
2103                 if (only) {
2104                         /*
2105                          * if kernel wants to dirty ctime but nothing else,
2106                          * we need to choose a cap to dirty under, or do
2107                          * a almost-no-op setattr
2108                          */
2109                         if (issued & CEPH_CAP_AUTH_EXCL)
2110                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2111                         else if (issued & CEPH_CAP_FILE_EXCL)
2112                                 dirtied |= CEPH_CAP_FILE_EXCL;
2113                         else if (issued & CEPH_CAP_XATTR_EXCL)
2114                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2115                         else
2116                                 mask |= CEPH_SETATTR_CTIME;
2117                 }
2118         }
2119         if (ia_valid & ATTR_FILE)
2120                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2121
2122         if (dirtied) {
2123                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2124                                                            &prealloc_cf);
2125                 inode->i_ctime = attr->ia_ctime;
2126         }
2127
2128         release &= issued;
2129         spin_unlock(&ci->i_ceph_lock);
2130         if (lock_snap_rwsem)
2131                 up_read(&mdsc->snap_rwsem);
2132
2133         if (inode_dirty_flags)
2134                 __mark_inode_dirty(inode, inode_dirty_flags);
2135
2136
2137         if (mask) {
2138                 req->r_inode = inode;
2139                 ihold(inode);
2140                 req->r_inode_drop = release;
2141                 req->r_args.setattr.mask = cpu_to_le32(mask);
2142                 req->r_num_caps = 1;
2143                 req->r_stamp = timespec64_to_timespec(attr->ia_ctime);
2144                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2145         }
2146         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2147              ceph_cap_string(dirtied), mask);
2148
2149         ceph_mdsc_put_request(req);
2150         ceph_free_cap_flush(prealloc_cf);
2151
2152         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2153                 __ceph_do_pending_vmtruncate(inode);
2154
2155         return err;
2156 }
2157
2158 /*
2159  * setattr
2160  */
2161 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2162 {
2163         struct inode *inode = d_inode(dentry);
2164         int err;
2165
2166         if (ceph_snap(inode) != CEPH_NOSNAP)
2167                 return -EROFS;
2168
2169         err = setattr_prepare(dentry, attr);
2170         if (err != 0)
2171                 return err;
2172
2173         if ((attr->ia_valid & ATTR_SIZE) &&
2174             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2175                 return -EDQUOT;
2176
2177         err = __ceph_setattr(inode, attr);
2178
2179         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2180                 err = posix_acl_chmod(inode, attr->ia_mode);
2181
2182         return err;
2183 }
2184
2185 /*
2186  * Verify that we have a lease on the given mask.  If not,
2187  * do a getattr against an mds.
2188  */
2189 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2190                       int mask, bool force)
2191 {
2192         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2193         struct ceph_mds_client *mdsc = fsc->mdsc;
2194         struct ceph_mds_request *req;
2195         int mode;
2196         int err;
2197
2198         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2199                 dout("do_getattr inode %p SNAPDIR\n", inode);
2200                 return 0;
2201         }
2202
2203         dout("do_getattr inode %p mask %s mode 0%o\n",
2204              inode, ceph_cap_string(mask), inode->i_mode);
2205         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2206                 return 0;
2207
2208         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2209         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2210         if (IS_ERR(req))
2211                 return PTR_ERR(req);
2212         req->r_inode = inode;
2213         ihold(inode);
2214         req->r_num_caps = 1;
2215         req->r_args.getattr.mask = cpu_to_le32(mask);
2216         req->r_locked_page = locked_page;
2217         err = ceph_mdsc_do_request(mdsc, NULL, req);
2218         if (locked_page && err == 0) {
2219                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2220                 if (inline_version == 0) {
2221                         /* the reply is supposed to contain inline data */
2222                         err = -EINVAL;
2223                 } else if (inline_version == CEPH_INLINE_NONE) {
2224                         err = -ENODATA;
2225                 } else {
2226                         err = req->r_reply_info.targeti.inline_len;
2227                 }
2228         }
2229         ceph_mdsc_put_request(req);
2230         dout("do_getattr result=%d\n", err);
2231         return err;
2232 }
2233
2234
2235 /*
2236  * Check inode permissions.  We verify we have a valid value for
2237  * the AUTH cap, then call the generic handler.
2238  */
2239 int ceph_permission(struct inode *inode, int mask)
2240 {
2241         int err;
2242
2243         if (mask & MAY_NOT_BLOCK)
2244                 return -ECHILD;
2245
2246         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2247
2248         if (!err)
2249                 err = generic_permission(inode, mask);
2250         return err;
2251 }
2252
2253 /*
2254  * Get all attributes.  Hopefully somedata we'll have a statlite()
2255  * and can limit the fields we require to be accurate.
2256  */
2257 int ceph_getattr(const struct path *path, struct kstat *stat,
2258                  u32 request_mask, unsigned int flags)
2259 {
2260         struct inode *inode = d_inode(path->dentry);
2261         struct ceph_inode_info *ci = ceph_inode(inode);
2262         int err;
2263
2264         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2265         if (!err) {
2266                 generic_fillattr(inode, stat);
2267                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2268                 if (ceph_snap(inode) != CEPH_NOSNAP)
2269                         stat->dev = ceph_snap(inode);
2270                 else
2271                         stat->dev = 0;
2272                 if (S_ISDIR(inode->i_mode)) {
2273                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2274                                                 RBYTES))
2275                                 stat->size = ci->i_rbytes;
2276                         else
2277                                 stat->size = ci->i_files + ci->i_subdirs;
2278                         stat->blocks = 0;
2279                         stat->blksize = 65536;
2280                         /*
2281                          * Some applications rely on the number of st_nlink
2282                          * value on directories to be either 0 (if unlinked)
2283                          * or 2 + number of subdirectories.
2284                          */
2285                         if (stat->nlink == 1)
2286                                 /* '.' + '..' + subdirs */
2287                                 stat->nlink = 1 + 1 + ci->i_subdirs;
2288                 }
2289         }
2290         return err;
2291 }