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