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         atomic_set(&ci->i_filelock_ref, 0);
 497         atomic_set(&ci->i_shared_gen, 0);
 498         ci->i_rdcache_gen = 0;
 499         ci->i_rdcache_revoking = 0;
 500
 501         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 502         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 503         spin_lock_init(&ci->i_unsafe_lock);
 504
 505         ci->i_snap_realm = NULL;
 506         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 507         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 508
 509         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 510         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 511
 512         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 513
 514         ceph_fscache_inode_init(ci);
 515
 516         return &ci->vfs_inode;
 517 }
 518
 519 static void ceph_i_callback(struct rcu_head *head)
 520 {
 521         struct inode *inode = container_of(head, struct inode, i_rcu);
 522         struct ceph_inode_info *ci = ceph_inode(inode);
 523
 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         kfree(ci->i_symlink);
 556         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 557                 frag = rb_entry(n, struct ceph_inode_frag, node);
 558                 rb_erase(n, &ci->i_fragtree);
 559                 kfree(frag);
 560         }
 561         ci->i_fragtree_nsplits = 0;
 562
 563         __ceph_destroy_xattrs(ci);
 564         if (ci->i_xattrs.blob)
 565                 ceph_buffer_put(ci->i_xattrs.blob);
 566         if (ci->i_xattrs.prealloc_blob)
 567                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 568
 569         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 570
 571         call_rcu(&inode->i_rcu, ceph_i_callback);
 572 }
 573
 574 int ceph_drop_inode(struct inode *inode)
 575 {
 576         /*
 577          * Positve dentry and corresponding inode are always accompanied
 578          * in MDS reply. So no need to keep inode in the cache after
 579          * dropping all its aliases.
 580          */
 581         return 1;
 582 }
 583
 584 static inline blkcnt_t calc_inode_blocks(u64 size)
 585 {
 586         return (size + (1<<9) - 1) >> 9;
 587 }
 588
 589 /*
 590  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 591  * careful because either the client or MDS may have more up to date
 592  * info, depending on which capabilities are held, and whether
 593  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 594  * and size are monotonically increasing, except when utimes() or
 595  * truncate() increments the corresponding _seq values.)
 596  */
 597 int ceph_fill_file_size(struct inode *inode, int issued,
 598                         u32 truncate_seq, u64 truncate_size, u64 size)
 599 {
 600         struct ceph_inode_info *ci = ceph_inode(inode);
 601         int queue_trunc = 0;
 602
 603         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 604             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 605                 dout("size %lld -> %llu\n", inode->i_size, size);
 606                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 607                         pr_err("fill_file_size non-zero size for directory\n");
 608                         size = 0;
 609                 }
 610                 i_size_write(inode, size);
 611                 inode->i_blocks = calc_inode_blocks(size);
 612                 ci->i_reported_size = size;
 613                 if (truncate_seq != ci->i_truncate_seq) {
 614                         dout("truncate_seq %u -> %u\n",
 615                              ci->i_truncate_seq, truncate_seq);
 616                         ci->i_truncate_seq = truncate_seq;
 617
 618                         /* the MDS should have revoked these caps */
 619                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 620                                                CEPH_CAP_FILE_RD |
 621                                                CEPH_CAP_FILE_WR |
 622                                                CEPH_CAP_FILE_LAZYIO));
 623                         /*
 624                          * If we hold relevant caps, or in the case where we're
 625                          * not the only client referencing this file and we
 626                          * don't hold those caps, then we need to check whether
 627                          * the file is either opened or mmaped
 628                          */
 629                         if ((issued & (CEPH_CAP_FILE_CACHE|
 630                                        CEPH_CAP_FILE_BUFFER)) ||
 631                             mapping_mapped(inode->i_mapping) ||
 632                             __ceph_caps_file_wanted(ci)) {
 633                                 ci->i_truncate_pending++;
 634                                 queue_trunc = 1;
 635                         }
 636                 }
 637         }
 638         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 639             ci->i_truncate_size != truncate_size) {
 640                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 641                      truncate_size);
 642                 ci->i_truncate_size = truncate_size;
 643         }
 644
 645         if (queue_trunc)
 646                 ceph_fscache_invalidate(inode);
 647
 648         return queue_trunc;
 649 }
 650
 651 void ceph_fill_file_time(struct inode *inode, int issued,
 652                          u64 time_warp_seq, struct timespec *ctime,
 653                          struct timespec *mtime, struct timespec *atime)
 654 {
 655         struct ceph_inode_info *ci = ceph_inode(inode);
 656         int warn = 0;
 657
 658         if (issued & (CEPH_CAP_FILE_EXCL|
 659                       CEPH_CAP_FILE_WR|
 660                       CEPH_CAP_FILE_BUFFER|
 661                       CEPH_CAP_AUTH_EXCL|
 662                       CEPH_CAP_XATTR_EXCL)) {
 663                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 664                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 665                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 666                              ctime->tv_sec, ctime->tv_nsec);
 667                         inode->i_ctime = *ctime;
 668                 }
 669                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 670                         /* the MDS did a utimes() */
 671                         dout("mtime %ld.%09ld -> %ld.%09ld "
 672                              "tw %d -> %d\n",
 673                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 674                              mtime->tv_sec, mtime->tv_nsec,
 675                              ci->i_time_warp_seq, (int)time_warp_seq);
 676
 677                         inode->i_mtime = *mtime;
 678                         inode->i_atime = *atime;
 679                         ci->i_time_warp_seq = time_warp_seq;
 680                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 681                         /* nobody did utimes(); take the max */
 682                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 683                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 684                                      inode->i_mtime.tv_sec,
 685                                      inode->i_mtime.tv_nsec,
 686                                      mtime->tv_sec, mtime->tv_nsec);
 687                                 inode->i_mtime = *mtime;
 688                         }
 689                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 690                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 691                                      inode->i_atime.tv_sec,
 692                                      inode->i_atime.tv_nsec,
 693                                      atime->tv_sec, atime->tv_nsec);
 694                                 inode->i_atime = *atime;
 695                         }
 696                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 697                         /* we did a utimes(); ignore mds values */
 698                 } else {
 699                         warn = 1;
 700                 }
 701         } else {
 702                 /* we have no write|excl caps; whatever the MDS says is true */
 703                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 704                         inode->i_ctime = *ctime;
 705                         inode->i_mtime = *mtime;
 706                         inode->i_atime = *atime;
 707                         ci->i_time_warp_seq = time_warp_seq;
 708                 } else {
 709                         warn = 1;
 710                 }
 711         }
 712         if (warn) /* time_warp_seq shouldn't go backwards */
 713                 dout("%p mds time_warp_seq %llu < %u\n",
 714                      inode, time_warp_seq, ci->i_time_warp_seq);
 715 }
 716
 717 /*
 718  * Populate an inode based on info from mds.  May be called on new or
 719  * existing inodes.
 720  */
 721 static int fill_inode(struct inode *inode, struct page *locked_page,
 722                       struct ceph_mds_reply_info_in *iinfo,
 723                       struct ceph_mds_reply_dirfrag *dirinfo,
 724                       struct ceph_mds_session *session,
 725                       unsigned long ttl_from, int cap_fmode,
 726                       struct ceph_cap_reservation *caps_reservation)
 727 {
 728         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 729         struct ceph_mds_reply_inode *info = iinfo->in;
 730         struct ceph_inode_info *ci = ceph_inode(inode);
 731         int issued = 0, implemented, new_issued;
 732         struct timespec mtime, atime, ctime;
 733         struct ceph_buffer *xattr_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_rdev = le32_to_cpu(info->rdev);
 791         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 792
 793         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 794             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 795                 inode->i_mode = le32_to_cpu(info->mode);
 796                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 797                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 798                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 799                      from_kuid(&init_user_ns, inode->i_uid),
 800                      from_kgid(&init_user_ns, inode->i_gid));
 801         }
 802
 803         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 804             (issued & CEPH_CAP_LINK_EXCL) == 0)
 805                 set_nlink(inode, le32_to_cpu(info->nlink));
 806
 807         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 808                 /* be careful with mtime, atime, size */
 809                 ceph_decode_timespec(&atime, &info->atime);
 810                 ceph_decode_timespec(&mtime, &info->mtime);
 811                 ceph_decode_timespec(&ctime, &info->ctime);
 812                 ceph_fill_file_time(inode, issued,
 813                                 le32_to_cpu(info->time_warp_seq),
 814                                 &ctime, &mtime, &atime);
 815         }
 816
 817         if (new_version ||
 818             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 819                 s64 old_pool = ci->i_layout.pool_id;
 820                 struct ceph_string *old_ns;
 821
 822                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 823                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 824                                         lockdep_is_held(&ci->i_ceph_lock));
 825                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 826
 827                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 828                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 829
 830                 pool_ns = old_ns;
 831
 832                 queue_trunc = ceph_fill_file_size(inode, issued,
 833                                         le32_to_cpu(info->truncate_seq),
 834                                         le64_to_cpu(info->truncate_size),
 835                                         le64_to_cpu(info->size));
 836                 /* only update max_size on auth cap */
 837                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 838                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 839                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 840                                         le64_to_cpu(info->max_size));
 841                         ci->i_max_size = le64_to_cpu(info->max_size);
 842                 }
 843         }
 844
 845         /* xattrs */
 846         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 847         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 848             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 849                 if (ci->i_xattrs.blob)
 850                         ceph_buffer_put(ci->i_xattrs.blob);
 851                 ci->i_xattrs.blob = xattr_blob;
 852                 if (xattr_blob)
 853                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 854                                iinfo->xattr_data, iinfo->xattr_len);
 855                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 856                 ceph_forget_all_cached_acls(inode);
 857                 xattr_blob = NULL;
 858         }
 859
 860         inode->i_mapping->a_ops = &ceph_aops;
 861
 862         switch (inode->i_mode & S_IFMT) {
 863         case S_IFIFO:
 864         case S_IFBLK:
 865         case S_IFCHR:
 866         case S_IFSOCK:
 867                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 868                 inode->i_op = &ceph_file_iops;
 869                 break;
 870         case S_IFREG:
 871                 inode->i_op = &ceph_file_iops;
 872                 inode->i_fop = &ceph_file_fops;
 873                 break;
 874         case S_IFLNK:
 875                 inode->i_op = &ceph_symlink_iops;
 876                 if (!ci->i_symlink) {
 877                         u32 symlen = iinfo->symlink_len;
 878                         char *sym;
 879
 880                         spin_unlock(&ci->i_ceph_lock);
 881
 882                         if (symlen != i_size_read(inode)) {
 883                                 pr_err("fill_inode %llx.%llx BAD symlink "
 884                                         "size %lld\n", ceph_vinop(inode),
 885                                         i_size_read(inode));
 886                                 i_size_write(inode, symlen);
 887                                 inode->i_blocks = calc_inode_blocks(symlen);
 888                         }
 889
 890                         err = -ENOMEM;
 891                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 892                         if (!sym)
 893                                 goto out;
 894
 895                         spin_lock(&ci->i_ceph_lock);
 896                         if (!ci->i_symlink)
 897                                 ci->i_symlink = sym;
 898                         else
 899                                 kfree(sym); /* lost a race */
 900                 }
 901                 inode->i_link = ci->i_symlink;
 902                 break;
 903         case S_IFDIR:
 904                 inode->i_op = &ceph_dir_iops;
 905                 inode->i_fop = &ceph_dir_fops;
 906
 907                 ci->i_dir_layout = iinfo->dir_layout;
 908
 909                 ci->i_files = le64_to_cpu(info->files);
 910                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 911                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 912                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 913                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 914                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 915                 break;
 916         default:
 917                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 918                        ceph_vinop(inode), inode->i_mode);
 919         }
 920
 921         /* were we issued a capability? */
 922         if (info->cap.caps) {
 923                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 924                         unsigned caps = le32_to_cpu(info->cap.caps);
 925                         ceph_add_cap(inode, session,
 926                                      le64_to_cpu(info->cap.cap_id),
 927                                      cap_fmode, caps,
 928                                      le32_to_cpu(info->cap.wanted),
 929                                      le32_to_cpu(info->cap.seq),
 930                                      le32_to_cpu(info->cap.mseq),
 931                                      le64_to_cpu(info->cap.realm),
 932                                      info->cap.flags, &new_cap);
 933
 934                         /* set dir completion flag? */
 935                         if (S_ISDIR(inode->i_mode) &&
 936                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 937                             (caps & CEPH_CAP_FILE_SHARED) &&
 938                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 939                             !__ceph_dir_is_complete(ci)) {
 940                                 dout(" marking %p complete (empty)\n", inode);
 941                                 i_size_write(inode, 0);
 942                                 __ceph_dir_set_complete(ci,
 943                                         atomic64_read(&ci->i_release_count),
 944                                         atomic64_read(&ci->i_ordered_count));
 945                         }
 946
 947                         wake = true;
 948                 } else {
 949                         dout(" %p got snap_caps %s\n", inode,
 950                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 951                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 952                         if (cap_fmode >= 0)
 953                                 __ceph_get_fmode(ci, cap_fmode);
 954                 }
 955         } else if (cap_fmode >= 0) {
 956                 pr_warn("mds issued no caps on %llx.%llx\n",
 957                            ceph_vinop(inode));
 958                 __ceph_get_fmode(ci, cap_fmode);
 959         }
 960
 961         if (iinfo->inline_version > 0 &&
 962             iinfo->inline_version >= ci->i_inline_version) {
 963                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 964                 ci->i_inline_version = iinfo->inline_version;
 965                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 966                     (locked_page ||
 967                      (le32_to_cpu(info->cap.caps) & cache_caps)))
 968                         fill_inline = true;
 969         }
 970
 971         spin_unlock(&ci->i_ceph_lock);
 972
 973         if (fill_inline)
 974                 ceph_fill_inline_data(inode, locked_page,
 975                                       iinfo->inline_data, iinfo->inline_len);
 976
 977         if (wake)
 978                 wake_up_all(&ci->i_cap_wq);
 979
 980         /* queue truncate if we saw i_size decrease */
 981         if (queue_trunc)
 982                 ceph_queue_vmtruncate(inode);
 983
 984         /* populate frag tree */
 985         if (S_ISDIR(inode->i_mode))
 986                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
 987
 988         /* update delegation info? */
 989         if (dirinfo)
 990                 ceph_fill_dirfrag(inode, dirinfo);
 991
 992         err = 0;
 993 out:
 994         if (new_cap)
 995                 ceph_put_cap(mdsc, new_cap);
 996         if (xattr_blob)
 997                 ceph_buffer_put(xattr_blob);
 998         ceph_put_string(pool_ns);
 999         return err;
1000 }
1001
1002 /*
1003  * caller should hold session s_mutex.
1004  */
1005 static void update_dentry_lease(struct dentry *dentry,
1006                                 struct ceph_mds_reply_lease *lease,
1007                                 struct ceph_mds_session *session,
1008                                 unsigned long from_time,
1009                                 struct ceph_vino *tgt_vino,
1010                                 struct ceph_vino *dir_vino)
1011 {
1012         struct ceph_dentry_info *di = ceph_dentry(dentry);
1013         long unsigned duration = le32_to_cpu(lease->duration_ms);
1014         long unsigned ttl = from_time + (duration * HZ) / 1000;
1015         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1016         struct inode *dir;
1017         struct ceph_mds_session *old_lease_session = NULL;
1018
1019         /*
1020          * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1021          * we expect a negative dentry.
1022          */
1023         if (!tgt_vino && d_really_is_positive(dentry))
1024                 return;
1025
1026         if (tgt_vino && (d_really_is_negative(dentry) ||
1027                         !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1028                 return;
1029
1030         spin_lock(&dentry->d_lock);
1031         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1032              dentry, duration, ttl);
1033
1034         dir = d_inode(dentry->d_parent);
1035
1036         /* make sure parent matches dir_vino */
1037         if (!ceph_ino_compare(dir, dir_vino))
1038                 goto out_unlock;
1039
1040         /* only track leases on regular dentries */
1041         if (ceph_snap(dir) != CEPH_NOSNAP)
1042                 goto out_unlock;
1043
1044         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1045
1046         if (duration == 0)
1047                 goto out_unlock;
1048
1049         if (di->lease_gen == session->s_cap_gen &&
1050             time_before(ttl, di->time))
1051                 goto out_unlock;  /* we already have a newer lease. */
1052
1053         if (di->lease_session && di->lease_session != session) {
1054                 old_lease_session = di->lease_session;
1055                 di->lease_session = NULL;
1056         }
1057
1058         ceph_dentry_lru_touch(dentry);
1059
1060         if (!di->lease_session)
1061                 di->lease_session = ceph_get_mds_session(session);
1062         di->lease_gen = session->s_cap_gen;
1063         di->lease_seq = le32_to_cpu(lease->seq);
1064         di->lease_renew_after = half_ttl;
1065         di->lease_renew_from = 0;
1066         di->time = ttl;
1067 out_unlock:
1068         spin_unlock(&dentry->d_lock);
1069         if (old_lease_session)
1070                 ceph_put_mds_session(old_lease_session);
1071 }
1072
1073 /*
1074  * splice a dentry to an inode.
1075  * caller must hold directory i_mutex for this to be safe.
1076  */
1077 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1078 {
1079         struct dentry *realdn;
1080
1081         BUG_ON(d_inode(dn));
1082
1083         if (S_ISDIR(in->i_mode)) {
1084                 /* If inode is directory, d_splice_alias() below will remove
1085                  * 'realdn' from its origin parent. We need to ensure that
1086                  * origin parent's readdir cache will not reference 'realdn'
1087                  */
1088                 realdn = d_find_any_alias(in);
1089                 if (realdn) {
1090                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1091                         spin_lock(&realdn->d_lock);
1092
1093                         realdn->d_op->d_prune(realdn);
1094
1095                         di->time = jiffies;
1096                         di->lease_shared_gen = 0;
1097                         di->offset = 0;
1098
1099                         spin_unlock(&realdn->d_lock);
1100                         dput(realdn);
1101                 }
1102         }
1103
1104         /* dn must be unhashed */
1105         if (!d_unhashed(dn))
1106                 d_drop(dn);
1107         realdn = d_splice_alias(in, dn);
1108         if (IS_ERR(realdn)) {
1109                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1110                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1111                 dn = realdn; /* note realdn contains the error */
1112                 goto out;
1113         } else if (realdn) {
1114                 dout("dn %p (%d) spliced with %p (%d) "
1115                      "inode %p ino %llx.%llx\n",
1116                      dn, d_count(dn),
1117                      realdn, d_count(realdn),
1118                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1119                 dput(dn);
1120                 dn = realdn;
1121         } else {
1122                 BUG_ON(!ceph_dentry(dn));
1123                 dout("dn %p attached to %p ino %llx.%llx\n",
1124                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1125         }
1126 out:
1127         return dn;
1128 }
1129
1130 /*
1131  * Incorporate results into the local cache.  This is either just
1132  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1133  * after a lookup).
1134  *
1135  * A reply may contain
1136  *         a directory inode along with a dentry.
1137  *  and/or a target inode
1138  *
1139  * Called with snap_rwsem (read).
1140  */
1141 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1142 {
1143         struct ceph_mds_session *session = req->r_session;
1144         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1145         struct inode *in = NULL;
1146         struct ceph_vino tvino, dvino;
1147         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1148         int err = 0;
1149
1150         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1151              rinfo->head->is_dentry, rinfo->head->is_target);
1152
1153         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1154                 dout("fill_trace reply is empty!\n");
1155                 if (rinfo->head->result == 0 && req->r_parent)
1156                         ceph_invalidate_dir_request(req);
1157                 return 0;
1158         }
1159
1160         if (rinfo->head->is_dentry) {
1161                 struct inode *dir = req->r_parent;
1162
1163                 if (dir) {
1164                         err = fill_inode(dir, NULL,
1165                                          &rinfo->diri, rinfo->dirfrag,
1166                                          session, req->r_request_started, -1,
1167                                          &req->r_caps_reservation);
1168                         if (err < 0)
1169                                 goto done;
1170                 } else {
1171                         WARN_ON_ONCE(1);
1172                 }
1173
1174                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1175                         struct qstr dname;
1176                         struct dentry *dn, *parent;
1177
1178                         BUG_ON(!rinfo->head->is_target);
1179                         BUG_ON(req->r_dentry);
1180
1181                         parent = d_find_any_alias(dir);
1182                         BUG_ON(!parent);
1183
1184                         dname.name = rinfo->dname;
1185                         dname.len = rinfo->dname_len;
1186                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1187                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1188                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1189 retry_lookup:
1190                         dn = d_lookup(parent, &dname);
1191                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1192                              parent, dname.len, dname.name, dn);
1193
1194                         if (!dn) {
1195                                 dn = d_alloc(parent, &dname);
1196                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1197                                      dname.len, dname.name, dn);
1198                                 if (!dn) {
1199                                         dput(parent);
1200                                         err = -ENOMEM;
1201                                         goto done;
1202                                 }
1203                                 err = 0;
1204                         } else if (d_really_is_positive(dn) &&
1205                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1206                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1207                                 dout(" dn %p points to wrong inode %p\n",
1208                                      dn, d_inode(dn));
1209                                 ceph_dir_clear_ordered(dir);
1210                                 d_delete(dn);
1211                                 dput(dn);
1212                                 goto retry_lookup;
1213                         }
1214
1215                         req->r_dentry = dn;
1216                         dput(parent);
1217                 }
1218         }
1219
1220         if (rinfo->head->is_target) {
1221                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1222                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1223
1224                 in = ceph_get_inode(sb, tvino);
1225                 if (IS_ERR(in)) {
1226                         err = PTR_ERR(in);
1227                         goto done;
1228                 }
1229                 req->r_target_inode = in;
1230
1231                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1232                                 session, req->r_request_started,
1233                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1234                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1235                                 &req->r_caps_reservation);
1236                 if (err < 0) {
1237                         pr_err("fill_inode badness %p %llx.%llx\n",
1238                                 in, ceph_vinop(in));
1239                         goto done;
1240                 }
1241         }
1242
1243         /*
1244          * ignore null lease/binding on snapdir ENOENT, or else we
1245          * will have trouble splicing in the virtual snapdir later
1246          */
1247         if (rinfo->head->is_dentry &&
1248             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1249             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1250             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1251                                                fsc->mount_options->snapdir_name,
1252                                                req->r_dentry->d_name.len))) {
1253                 /*
1254                  * lookup link rename   : null -> possibly existing inode
1255                  * mknod symlink mkdir  : null -> new inode
1256                  * unlink               : linked -> null
1257                  */
1258                 struct inode *dir = req->r_parent;
1259                 struct dentry *dn = req->r_dentry;
1260                 bool have_dir_cap, have_lease;
1261
1262                 BUG_ON(!dn);
1263                 BUG_ON(!dir);
1264                 BUG_ON(d_inode(dn->d_parent) != dir);
1265
1266                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1267                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1268
1269                 BUG_ON(ceph_ino(dir) != dvino.ino);
1270                 BUG_ON(ceph_snap(dir) != dvino.snap);
1271
1272                 /* do we have a lease on the whole dir? */
1273                 have_dir_cap =
1274                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1275                          CEPH_CAP_FILE_SHARED);
1276
1277                 /* do we have a dn lease? */
1278                 have_lease = have_dir_cap ||
1279                         le32_to_cpu(rinfo->dlease->duration_ms);
1280                 if (!have_lease)
1281                         dout("fill_trace  no dentry lease or dir cap\n");
1282
1283                 /* rename? */
1284                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1285                         struct inode *olddir = req->r_old_dentry_dir;
1286                         BUG_ON(!olddir);
1287
1288                         dout(" src %p '%pd' dst %p '%pd'\n",
1289                              req->r_old_dentry,
1290                              req->r_old_dentry,
1291                              dn, dn);
1292                         dout("fill_trace doing d_move %p -> %p\n",
1293                              req->r_old_dentry, dn);
1294
1295                         /* d_move screws up sibling dentries' offsets */
1296                         ceph_dir_clear_ordered(dir);
1297                         ceph_dir_clear_ordered(olddir);
1298
1299                         d_move(req->r_old_dentry, dn);
1300                         dout(" src %p '%pd' dst %p '%pd'\n",
1301                              req->r_old_dentry,
1302                              req->r_old_dentry,
1303                              dn, dn);
1304
1305                         /* ensure target dentry is invalidated, despite
1306                            rehashing bug in vfs_rename_dir */
1307                         ceph_invalidate_dentry_lease(dn);
1308
1309                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1310                              ceph_dentry(req->r_old_dentry)->offset);
1311
1312                         dn = req->r_old_dentry;  /* use old_dentry */
1313                 }
1314
1315                 /* null dentry? */
1316                 if (!rinfo->head->is_target) {
1317                         dout("fill_trace null dentry\n");
1318                         if (d_really_is_positive(dn)) {
1319                                 dout("d_delete %p\n", dn);
1320                                 ceph_dir_clear_ordered(dir);
1321                                 d_delete(dn);
1322                         } else if (have_lease) {
1323                                 if (d_unhashed(dn))
1324                                         d_add(dn, NULL);
1325                                 update_dentry_lease(dn, rinfo->dlease,
1326                                                     session,
1327                                                     req->r_request_started,
1328                                                     NULL, &dvino);
1329                         }
1330                         goto done;
1331                 }
1332
1333                 /* attach proper inode */
1334                 if (d_really_is_negative(dn)) {
1335                         ceph_dir_clear_ordered(dir);
1336                         ihold(in);
1337                         dn = splice_dentry(dn, in);
1338                         if (IS_ERR(dn)) {
1339                                 err = PTR_ERR(dn);
1340                                 goto done;
1341                         }
1342                         req->r_dentry = dn;  /* may have spliced */
1343                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1344                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1345                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1346                              ceph_vinop(in));
1347                         d_invalidate(dn);
1348                         have_lease = false;
1349                 }
1350
1351                 if (have_lease) {
1352                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1353                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1354                         update_dentry_lease(dn, rinfo->dlease, session,
1355                                             req->r_request_started,
1356                                             &tvino, &dvino);
1357                 }
1358                 dout(" final dn %p\n", dn);
1359         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1360                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1361                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1362                 struct dentry *dn = req->r_dentry;
1363                 struct inode *dir = req->r_parent;
1364
1365                 /* fill out a snapdir LOOKUPSNAP dentry */
1366                 BUG_ON(!dn);
1367                 BUG_ON(!dir);
1368                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1369                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1370                 ceph_dir_clear_ordered(dir);
1371                 ihold(in);
1372                 dn = splice_dentry(dn, in);
1373                 if (IS_ERR(dn)) {
1374                         err = PTR_ERR(dn);
1375                         goto done;
1376                 }
1377                 req->r_dentry = dn;  /* may have spliced */
1378         } else if (rinfo->head->is_dentry) {
1379                 struct ceph_vino *ptvino = NULL;
1380
1381                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1382                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1383                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1384                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1385
1386                         if (rinfo->head->is_target) {
1387                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1388                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1389                                 ptvino = &tvino;
1390                         }
1391
1392                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1393                                 session, req->r_request_started, ptvino,
1394                                 &dvino);
1395                 } else {
1396                         dout("%s: no dentry lease or dir cap\n", __func__);
1397                 }
1398         }
1399 done:
1400         dout("fill_trace done err=%d\n", err);
1401         return err;
1402 }
1403
1404 /*
1405  * Prepopulate our cache with readdir results, leases, etc.
1406  */
1407 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1408                                            struct ceph_mds_session *session)
1409 {
1410         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1411         int i, err = 0;
1412
1413         for (i = 0; i < rinfo->dir_nr; i++) {
1414                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1415                 struct ceph_vino vino;
1416                 struct inode *in;
1417                 int rc;
1418
1419                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1420                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1421
1422                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1423                 if (IS_ERR(in)) {
1424                         err = PTR_ERR(in);
1425                         dout("new_inode badness got %d\n", err);
1426                         continue;
1427                 }
1428                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1429                                 req->r_request_started, -1,
1430                                 &req->r_caps_reservation);
1431                 if (rc < 0) {
1432                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1433                         err = rc;
1434                 }
1435                 iput(in);
1436         }
1437
1438         return err;
1439 }
1440
1441 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1442 {
1443         if (ctl->page) {
1444                 kunmap(ctl->page);
1445                 put_page(ctl->page);
1446                 ctl->page = NULL;
1447         }
1448 }
1449
1450 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1451                               struct ceph_readdir_cache_control *ctl,
1452                               struct ceph_mds_request *req)
1453 {
1454         struct ceph_inode_info *ci = ceph_inode(dir);
1455         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1456         unsigned idx = ctl->index % nsize;
1457         pgoff_t pgoff = ctl->index / nsize;
1458
1459         if (!ctl->page || pgoff != page_index(ctl->page)) {
1460                 ceph_readdir_cache_release(ctl);
1461                 if (idx == 0)
1462                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1463                 else
1464                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1465                 if (!ctl->page) {
1466                         ctl->index = -1;
1467                         return idx == 0 ? -ENOMEM : 0;
1468                 }
1469                 /* reading/filling the cache are serialized by
1470                  * i_mutex, no need to use page lock */
1471                 unlock_page(ctl->page);
1472                 ctl->dentries = kmap(ctl->page);
1473                 if (idx == 0)
1474                         memset(ctl->dentries, 0, PAGE_SIZE);
1475         }
1476
1477         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1478             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1479                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1480                 ctl->dentries[idx] = dn;
1481                 ctl->index++;
1482         } else {
1483                 dout("disable readdir cache\n");
1484                 ctl->index = -1;
1485         }
1486         return 0;
1487 }
1488
1489 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1490                              struct ceph_mds_session *session)
1491 {
1492         struct dentry *parent = req->r_dentry;
1493         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1494         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1495         struct qstr dname;
1496         struct dentry *dn;
1497         struct inode *in;
1498         int err = 0, skipped = 0, ret, i;
1499         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1500         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1501         u32 last_hash = 0;
1502         u32 fpos_offset;
1503         struct ceph_readdir_cache_control cache_ctl = {};
1504
1505         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1506                 return readdir_prepopulate_inodes_only(req, session);
1507
1508         if (rinfo->hash_order) {
1509                 if (req->r_path2) {
1510                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1511                                                   req->r_path2,
1512                                                   strlen(req->r_path2));
1513                         last_hash = ceph_frag_value(last_hash);
1514                 } else if (rinfo->offset_hash) {
1515                         /* mds understands offset_hash */
1516                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1517                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1518                 }
1519         }
1520
1521         if (rinfo->dir_dir &&
1522             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1523                 dout("readdir_prepopulate got new frag %x -> %x\n",
1524                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1525                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1526                 if (!rinfo->hash_order)
1527                         req->r_readdir_offset = 2;
1528         }
1529
1530         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1531                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1532                      rinfo->dir_nr, parent);
1533         } else {
1534                 dout("readdir_prepopulate %d items under dn %p\n",
1535                      rinfo->dir_nr, parent);
1536                 if (rinfo->dir_dir)
1537                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1538
1539                 if (ceph_frag_is_leftmost(frag) &&
1540                     req->r_readdir_offset == 2 &&
1541                     !(rinfo->hash_order && last_hash)) {
1542                         /* note dir version at start of readdir so we can
1543                          * tell if any dentries get dropped */
1544                         req->r_dir_release_cnt =
1545                                 atomic64_read(&ci->i_release_count);
1546                         req->r_dir_ordered_cnt =
1547                                 atomic64_read(&ci->i_ordered_count);
1548                         req->r_readdir_cache_idx = 0;
1549                 }
1550         }
1551
1552         cache_ctl.index = req->r_readdir_cache_idx;
1553         fpos_offset = req->r_readdir_offset;
1554
1555         /* FIXME: release caps/leases if error occurs */
1556         for (i = 0; i < rinfo->dir_nr; i++) {
1557                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1558                 struct ceph_vino tvino, dvino;
1559
1560                 dname.name = rde->name;
1561                 dname.len = rde->name_len;
1562                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1563
1564                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1565                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1566
1567                 if (rinfo->hash_order) {
1568                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1569                                                  rde->name, rde->name_len);
1570                         hash = ceph_frag_value(hash);
1571                         if (hash != last_hash)
1572                                 fpos_offset = 2;
1573                         last_hash = hash;
1574                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1575                 } else {
1576                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1577                 }
1578
1579 retry_lookup:
1580                 dn = d_lookup(parent, &dname);
1581                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1582                      parent, dname.len, dname.name, dn);
1583
1584                 if (!dn) {
1585                         dn = d_alloc(parent, &dname);
1586                         dout("d_alloc %p '%.*s' = %p\n", parent,
1587                              dname.len, dname.name, dn);
1588                         if (!dn) {
1589                                 dout("d_alloc badness\n");
1590                                 err = -ENOMEM;
1591                                 goto out;
1592                         }
1593                 } else if (d_really_is_positive(dn) &&
1594                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1595                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1596                         struct ceph_dentry_info *di = ceph_dentry(dn);
1597                         dout(" dn %p points to wrong inode %p\n",
1598                              dn, d_inode(dn));
1599
1600                         spin_lock(&dn->d_lock);
1601                         if (di->offset > 0 &&
1602                             di->lease_shared_gen ==
1603                             atomic_read(&ci->i_shared_gen)) {
1604                                 __ceph_dir_clear_ordered(ci);
1605                                 di->offset = 0;
1606                         }
1607                         spin_unlock(&dn->d_lock);
1608
1609                         d_delete(dn);
1610                         dput(dn);
1611                         goto retry_lookup;
1612                 }
1613
1614                 /* inode */
1615                 if (d_really_is_positive(dn)) {
1616                         in = d_inode(dn);
1617                 } else {
1618                         in = ceph_get_inode(parent->d_sb, tvino);
1619                         if (IS_ERR(in)) {
1620                                 dout("new_inode badness\n");
1621                                 d_drop(dn);
1622                                 dput(dn);
1623                                 err = PTR_ERR(in);
1624                                 goto out;
1625                         }
1626                 }
1627
1628                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1629                                  req->r_request_started, -1,
1630                                  &req->r_caps_reservation);
1631                 if (ret < 0) {
1632                         pr_err("fill_inode badness on %p\n", in);
1633                         if (d_really_is_negative(dn))
1634                                 iput(in);
1635                         d_drop(dn);
1636                         err = ret;
1637                         goto next_item;
1638                 }
1639
1640                 if (d_really_is_negative(dn)) {
1641                         struct dentry *realdn;
1642
1643                         if (ceph_security_xattr_deadlock(in)) {
1644                                 dout(" skip splicing dn %p to inode %p"
1645                                      " (security xattr deadlock)\n", dn, in);
1646                                 iput(in);
1647                                 skipped++;
1648                                 goto next_item;
1649                         }
1650
1651                         realdn = splice_dentry(dn, in);
1652                         if (IS_ERR(realdn)) {
1653                                 err = PTR_ERR(realdn);
1654                                 d_drop(dn);
1655                                 dn = NULL;
1656                                 goto next_item;
1657                         }
1658                         dn = realdn;
1659                 }
1660
1661                 ceph_dentry(dn)->offset = rde->offset;
1662
1663                 dvino = ceph_vino(d_inode(parent));
1664                 update_dentry_lease(dn, rde->lease, req->r_session,
1665                                     req->r_request_started, &tvino, &dvino);
1666
1667                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1668                         ret = fill_readdir_cache(d_inode(parent), dn,
1669                                                  &cache_ctl, req);
1670                         if (ret < 0)
1671                                 err = ret;
1672                 }
1673 next_item:
1674                 if (dn)
1675                         dput(dn);
1676         }
1677 out:
1678         if (err == 0 && skipped == 0) {
1679                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1680                 req->r_readdir_cache_idx = cache_ctl.index;
1681         }
1682         ceph_readdir_cache_release(&cache_ctl);
1683         dout("readdir_prepopulate done\n");
1684         return err;
1685 }
1686
1687 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1688 {
1689         struct ceph_inode_info *ci = ceph_inode(inode);
1690         bool ret;
1691
1692         spin_lock(&ci->i_ceph_lock);
1693         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1694         i_size_write(inode, size);
1695         inode->i_blocks = calc_inode_blocks(size);
1696
1697         ret = __ceph_should_report_size(ci);
1698
1699         spin_unlock(&ci->i_ceph_lock);
1700         return ret;
1701 }
1702
1703 /*
1704  * Write back inode data in a worker thread.  (This can't be done
1705  * in the message handler context.)
1706  */
1707 void ceph_queue_writeback(struct inode *inode)
1708 {
1709         ihold(inode);
1710         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1711                        &ceph_inode(inode)->i_wb_work)) {
1712                 dout("ceph_queue_writeback %p\n", inode);
1713         } else {
1714                 dout("ceph_queue_writeback %p failed\n", inode);
1715                 iput(inode);
1716         }
1717 }
1718
1719 static void ceph_writeback_work(struct work_struct *work)
1720 {
1721         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1722                                                   i_wb_work);
1723         struct inode *inode = &ci->vfs_inode;
1724
1725         dout("writeback %p\n", inode);
1726         filemap_fdatawrite(&inode->i_data);
1727         iput(inode);
1728 }
1729
1730 /*
1731  * queue an async invalidation
1732  */
1733 void ceph_queue_invalidate(struct inode *inode)
1734 {
1735         ihold(inode);
1736         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1737                        &ceph_inode(inode)->i_pg_inv_work)) {
1738                 dout("ceph_queue_invalidate %p\n", inode);
1739         } else {
1740                 dout("ceph_queue_invalidate %p failed\n", inode);
1741                 iput(inode);
1742         }
1743 }
1744
1745 /*
1746  * Invalidate inode pages in a worker thread.  (This can't be done
1747  * in the message handler context.)
1748  */
1749 static void ceph_invalidate_work(struct work_struct *work)
1750 {
1751         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1752                                                   i_pg_inv_work);
1753         struct inode *inode = &ci->vfs_inode;
1754         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1755         u32 orig_gen;
1756         int check = 0;
1757
1758         mutex_lock(&ci->i_truncate_mutex);
1759
1760         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1761                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1762                                     inode, ceph_ino(inode));
1763                 mapping_set_error(inode->i_mapping, -EIO);
1764                 truncate_pagecache(inode, 0);
1765                 mutex_unlock(&ci->i_truncate_mutex);
1766                 goto out;
1767         }
1768
1769         spin_lock(&ci->i_ceph_lock);
1770         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1771              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1772         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1773                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1774                         check = 1;
1775                 spin_unlock(&ci->i_ceph_lock);
1776                 mutex_unlock(&ci->i_truncate_mutex);
1777                 goto out;
1778         }
1779         orig_gen = ci->i_rdcache_gen;
1780         spin_unlock(&ci->i_ceph_lock);
1781
1782         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1783                 pr_err("invalidate_pages %p fails\n", inode);
1784         }
1785
1786         spin_lock(&ci->i_ceph_lock);
1787         if (orig_gen == ci->i_rdcache_gen &&
1788             orig_gen == ci->i_rdcache_revoking) {
1789                 dout("invalidate_pages %p gen %d successful\n", inode,
1790                      ci->i_rdcache_gen);
1791                 ci->i_rdcache_revoking--;
1792                 check = 1;
1793         } else {
1794                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1795                      inode, orig_gen, ci->i_rdcache_gen,
1796                      ci->i_rdcache_revoking);
1797                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1798                         check = 1;
1799         }
1800         spin_unlock(&ci->i_ceph_lock);
1801         mutex_unlock(&ci->i_truncate_mutex);
1802 out:
1803         if (check)
1804                 ceph_check_caps(ci, 0, NULL);
1805         iput(inode);
1806 }
1807
1808
1809 /*
1810  * called by trunc_wq;
1811  *
1812  * We also truncate in a separate thread as well.
1813  */
1814 static void ceph_vmtruncate_work(struct work_struct *work)
1815 {
1816         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1817                                                   i_vmtruncate_work);
1818         struct inode *inode = &ci->vfs_inode;
1819
1820         dout("vmtruncate_work %p\n", inode);
1821         __ceph_do_pending_vmtruncate(inode);
1822         iput(inode);
1823 }
1824
1825 /*
1826  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1827  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1828  */
1829 void ceph_queue_vmtruncate(struct inode *inode)
1830 {
1831         struct ceph_inode_info *ci = ceph_inode(inode);
1832
1833         ihold(inode);
1834
1835         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1836                        &ci->i_vmtruncate_work)) {
1837                 dout("ceph_queue_vmtruncate %p\n", inode);
1838         } else {
1839                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1840                      inode, ci->i_truncate_pending);
1841                 iput(inode);
1842         }
1843 }
1844
1845 /*
1846  * Make sure any pending truncation is applied before doing anything
1847  * that may depend on it.
1848  */
1849 void __ceph_do_pending_vmtruncate(struct inode *inode)
1850 {
1851         struct ceph_inode_info *ci = ceph_inode(inode);
1852         u64 to;
1853         int wrbuffer_refs, finish = 0;
1854
1855         mutex_lock(&ci->i_truncate_mutex);
1856 retry:
1857         spin_lock(&ci->i_ceph_lock);
1858         if (ci->i_truncate_pending == 0) {
1859                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1860                 spin_unlock(&ci->i_ceph_lock);
1861                 mutex_unlock(&ci->i_truncate_mutex);
1862                 return;
1863         }
1864
1865         /*
1866          * make sure any dirty snapped pages are flushed before we
1867          * possibly truncate them.. so write AND block!
1868          */
1869         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1870                 struct ceph_cap_snap *capsnap;
1871                 to = ci->i_truncate_size;
1872                 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1873                         // MDS should have revoked Frw caps
1874                         WARN_ON_ONCE(capsnap->writing);
1875                         if (capsnap->dirty_pages && capsnap->size > to)
1876                                 to = capsnap->size;
1877                 }
1878                 spin_unlock(&ci->i_ceph_lock);
1879                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1880                      inode);
1881
1882                 truncate_pagecache(inode, to);
1883
1884                 filemap_write_and_wait_range(&inode->i_data, 0,
1885                                              inode->i_sb->s_maxbytes);
1886                 goto retry;
1887         }
1888
1889         /* there should be no reader or writer */
1890         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1891
1892         to = ci->i_truncate_size;
1893         wrbuffer_refs = ci->i_wrbuffer_ref;
1894         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1895              ci->i_truncate_pending, to);
1896         spin_unlock(&ci->i_ceph_lock);
1897
1898         truncate_pagecache(inode, to);
1899
1900         spin_lock(&ci->i_ceph_lock);
1901         if (to == ci->i_truncate_size) {
1902                 ci->i_truncate_pending = 0;
1903                 finish = 1;
1904         }
1905         spin_unlock(&ci->i_ceph_lock);
1906         if (!finish)
1907                 goto retry;
1908
1909         mutex_unlock(&ci->i_truncate_mutex);
1910
1911         if (wrbuffer_refs == 0)
1912                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1913
1914         wake_up_all(&ci->i_cap_wq);
1915 }
1916
1917 /*
1918  * symlinks
1919  */
1920 static const struct inode_operations ceph_symlink_iops = {
1921         .get_link = simple_get_link,
1922         .setattr = ceph_setattr,
1923         .getattr = ceph_getattr,
1924         .listxattr = ceph_listxattr,
1925 };
1926
1927 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1928 {
1929         struct ceph_inode_info *ci = ceph_inode(inode);
1930         const unsigned int ia_valid = attr->ia_valid;
1931         struct ceph_mds_request *req;
1932         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1933         struct ceph_cap_flush *prealloc_cf;
1934         int issued;
1935         int release = 0, dirtied = 0;
1936         int mask = 0;
1937         int err = 0;
1938         int inode_dirty_flags = 0;
1939         bool lock_snap_rwsem = false;
1940
1941         prealloc_cf = ceph_alloc_cap_flush();
1942         if (!prealloc_cf)
1943                 return -ENOMEM;
1944
1945         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1946                                        USE_AUTH_MDS);
1947         if (IS_ERR(req)) {
1948                 ceph_free_cap_flush(prealloc_cf);
1949                 return PTR_ERR(req);
1950         }
1951
1952         spin_lock(&ci->i_ceph_lock);
1953         issued = __ceph_caps_issued(ci, NULL);
1954
1955         if (!ci->i_head_snapc &&
1956             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1957                 lock_snap_rwsem = true;
1958                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1959                         spin_unlock(&ci->i_ceph_lock);
1960                         down_read(&mdsc->snap_rwsem);
1961                         spin_lock(&ci->i_ceph_lock);
1962                         issued = __ceph_caps_issued(ci, NULL);
1963                 }
1964         }
1965
1966         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1967
1968         if (ia_valid & ATTR_UID) {
1969                 dout("setattr %p uid %d -> %d\n", inode,
1970                      from_kuid(&init_user_ns, inode->i_uid),
1971                      from_kuid(&init_user_ns, attr->ia_uid));
1972                 if (issued & CEPH_CAP_AUTH_EXCL) {
1973                         inode->i_uid = attr->ia_uid;
1974                         dirtied |= CEPH_CAP_AUTH_EXCL;
1975                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1976                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1977                         req->r_args.setattr.uid = cpu_to_le32(
1978                                 from_kuid(&init_user_ns, attr->ia_uid));
1979                         mask |= CEPH_SETATTR_UID;
1980                         release |= CEPH_CAP_AUTH_SHARED;
1981                 }
1982         }
1983         if (ia_valid & ATTR_GID) {
1984                 dout("setattr %p gid %d -> %d\n", inode,
1985                      from_kgid(&init_user_ns, inode->i_gid),
1986                      from_kgid(&init_user_ns, attr->ia_gid));
1987                 if (issued & CEPH_CAP_AUTH_EXCL) {
1988                         inode->i_gid = attr->ia_gid;
1989                         dirtied |= CEPH_CAP_AUTH_EXCL;
1990                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1991                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1992                         req->r_args.setattr.gid = cpu_to_le32(
1993                                 from_kgid(&init_user_ns, attr->ia_gid));
1994                         mask |= CEPH_SETATTR_GID;
1995                         release |= CEPH_CAP_AUTH_SHARED;
1996                 }
1997         }
1998         if (ia_valid & ATTR_MODE) {
1999                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2000                      attr->ia_mode);
2001                 if (issued & CEPH_CAP_AUTH_EXCL) {
2002                         inode->i_mode = attr->ia_mode;
2003                         dirtied |= CEPH_CAP_AUTH_EXCL;
2004                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2005                            attr->ia_mode != inode->i_mode) {
2006                         inode->i_mode = attr->ia_mode;
2007                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2008                         mask |= CEPH_SETATTR_MODE;
2009                         release |= CEPH_CAP_AUTH_SHARED;
2010                 }
2011         }
2012
2013         if (ia_valid & ATTR_ATIME) {
2014                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
2015                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2016                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2017                 if (issued & CEPH_CAP_FILE_EXCL) {
2018                         ci->i_time_warp_seq++;
2019                         inode->i_atime = attr->ia_atime;
2020                         dirtied |= CEPH_CAP_FILE_EXCL;
2021                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2022                            timespec_compare(&inode->i_atime,
2023                                             &attr->ia_atime) < 0) {
2024                         inode->i_atime = attr->ia_atime;
2025                         dirtied |= CEPH_CAP_FILE_WR;
2026                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2027                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
2028                         ceph_encode_timespec(&req->r_args.setattr.atime,
2029                                              &attr->ia_atime);
2030                         mask |= CEPH_SETATTR_ATIME;
2031                         release |= CEPH_CAP_FILE_SHARED |
2032                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2033                 }
2034         }
2035         if (ia_valid & ATTR_MTIME) {
2036                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
2037                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2038                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2039                 if (issued & CEPH_CAP_FILE_EXCL) {
2040                         ci->i_time_warp_seq++;
2041                         inode->i_mtime = attr->ia_mtime;
2042                         dirtied |= CEPH_CAP_FILE_EXCL;
2043                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2044                            timespec_compare(&inode->i_mtime,
2045                                             &attr->ia_mtime) < 0) {
2046                         inode->i_mtime = attr->ia_mtime;
2047                         dirtied |= CEPH_CAP_FILE_WR;
2048                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2049                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2050                         ceph_encode_timespec(&req->r_args.setattr.mtime,
2051                                              &attr->ia_mtime);
2052                         mask |= CEPH_SETATTR_MTIME;
2053                         release |= CEPH_CAP_FILE_SHARED |
2054                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2055                 }
2056         }
2057         if (ia_valid & ATTR_SIZE) {
2058                 dout("setattr %p size %lld -> %lld\n", inode,
2059                      inode->i_size, attr->ia_size);
2060                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2061                     attr->ia_size > inode->i_size) {
2062                         i_size_write(inode, attr->ia_size);
2063                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2064                         ci->i_reported_size = attr->ia_size;
2065                         dirtied |= CEPH_CAP_FILE_EXCL;
2066                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2067                            attr->ia_size != inode->i_size) {
2068                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2069                         req->r_args.setattr.old_size =
2070                                 cpu_to_le64(inode->i_size);
2071                         mask |= CEPH_SETATTR_SIZE;
2072                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2073                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2074                 }
2075         }
2076
2077         /* these do nothing */
2078         if (ia_valid & ATTR_CTIME) {
2079                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2080                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2081                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2082                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2083                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2084                      only ? "ctime only" : "ignored");
2085                 if (only) {
2086                         /*
2087                          * if kernel wants to dirty ctime but nothing else,
2088                          * we need to choose a cap to dirty under, or do
2089                          * a almost-no-op setattr
2090                          */
2091                         if (issued & CEPH_CAP_AUTH_EXCL)
2092                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2093                         else if (issued & CEPH_CAP_FILE_EXCL)
2094                                 dirtied |= CEPH_CAP_FILE_EXCL;
2095                         else if (issued & CEPH_CAP_XATTR_EXCL)
2096                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2097                         else
2098                                 mask |= CEPH_SETATTR_CTIME;
2099                 }
2100         }
2101         if (ia_valid & ATTR_FILE)
2102                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2103
2104         if (dirtied) {
2105                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2106                                                            &prealloc_cf);
2107                 inode->i_ctime = attr->ia_ctime;
2108         }
2109
2110         release &= issued;
2111         spin_unlock(&ci->i_ceph_lock);
2112         if (lock_snap_rwsem)
2113                 up_read(&mdsc->snap_rwsem);
2114
2115         if (inode_dirty_flags)
2116                 __mark_inode_dirty(inode, inode_dirty_flags);
2117
2118
2119         if (mask) {
2120                 req->r_inode = inode;
2121                 ihold(inode);
2122                 req->r_inode_drop = release;
2123                 req->r_args.setattr.mask = cpu_to_le32(mask);
2124                 req->r_num_caps = 1;
2125                 req->r_stamp = attr->ia_ctime;
2126                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2127         }
2128         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2129              ceph_cap_string(dirtied), mask);
2130
2131         ceph_mdsc_put_request(req);
2132         ceph_free_cap_flush(prealloc_cf);
2133
2134         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2135                 __ceph_do_pending_vmtruncate(inode);
2136
2137         return err;
2138 }
2139
2140 /*
2141  * setattr
2142  */
2143 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2144 {
2145         struct inode *inode = d_inode(dentry);
2146         int err;
2147
2148         if (ceph_snap(inode) != CEPH_NOSNAP)
2149                 return -EROFS;
2150
2151         err = setattr_prepare(dentry, attr);
2152         if (err != 0)
2153                 return err;
2154
2155         err = __ceph_setattr(inode, attr);
2156
2157         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2158                 err = posix_acl_chmod(inode, attr->ia_mode);
2159
2160         return err;
2161 }
2162
2163 /*
2164  * Verify that we have a lease on the given mask.  If not,
2165  * do a getattr against an mds.
2166  */
2167 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2168                       int mask, bool force)
2169 {
2170         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2171         struct ceph_mds_client *mdsc = fsc->mdsc;
2172         struct ceph_mds_request *req;
2173         int err;
2174
2175         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2176                 dout("do_getattr inode %p SNAPDIR\n", inode);
2177                 return 0;
2178         }
2179
2180         dout("do_getattr inode %p mask %s mode 0%o\n",
2181              inode, ceph_cap_string(mask), inode->i_mode);
2182         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2183                 return 0;
2184
2185         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2186         if (IS_ERR(req))
2187                 return PTR_ERR(req);
2188         req->r_inode = inode;
2189         ihold(inode);
2190         req->r_num_caps = 1;
2191         req->r_args.getattr.mask = cpu_to_le32(mask);
2192         req->r_locked_page = locked_page;
2193         err = ceph_mdsc_do_request(mdsc, NULL, req);
2194         if (locked_page && err == 0) {
2195                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2196                 if (inline_version == 0) {
2197                         /* the reply is supposed to contain inline data */
2198                         err = -EINVAL;
2199                 } else if (inline_version == CEPH_INLINE_NONE) {
2200                         err = -ENODATA;
2201                 } else {
2202                         err = req->r_reply_info.targeti.inline_len;
2203                 }
2204         }
2205         ceph_mdsc_put_request(req);
2206         dout("do_getattr result=%d\n", err);
2207         return err;
2208 }
2209
2210
2211 /*
2212  * Check inode permissions.  We verify we have a valid value for
2213  * the AUTH cap, then call the generic handler.
2214  */
2215 int ceph_permission(struct inode *inode, int mask)
2216 {
2217         int err;
2218
2219         if (mask & MAY_NOT_BLOCK)
2220                 return -ECHILD;
2221
2222         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2223
2224         if (!err)
2225                 err = generic_permission(inode, mask);
2226         return err;
2227 }
2228
2229 /*
2230  * Get all attributes.  Hopefully somedata we'll have a statlite()
2231  * and can limit the fields we require to be accurate.
2232  */
2233 int ceph_getattr(const struct path *path, struct kstat *stat,
2234                  u32 request_mask, unsigned int flags)
2235 {
2236         struct inode *inode = d_inode(path->dentry);
2237         struct ceph_inode_info *ci = ceph_inode(inode);
2238         int err;
2239
2240         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2241         if (!err) {
2242                 generic_fillattr(inode, stat);
2243                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2244                 if (ceph_snap(inode) != CEPH_NOSNAP)
2245                         stat->dev = ceph_snap(inode);
2246                 else
2247                         stat->dev = 0;
2248                 if (S_ISDIR(inode->i_mode)) {
2249                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2250                                                 RBYTES))
2251                                 stat->size = ci->i_rbytes;
2252                         else
2253                                 stat->size = ci->i_files + ci->i_subdirs;
2254                         stat->blocks = 0;
2255                         stat->blksize = 65536;
2256                 }
2257         }
2258         return err;
2259 }