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         /* dn must be unhashed */
1084         if (!d_unhashed(dn))
1085                 d_drop(dn);
1086         realdn = d_splice_alias(in, dn);
1087         if (IS_ERR(realdn)) {
1088                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1089                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1090                 dn = realdn; /* note realdn contains the error */
1091                 goto out;
1092         } else if (realdn) {
1093                 dout("dn %p (%d) spliced with %p (%d) "
1094                      "inode %p ino %llx.%llx\n",
1095                      dn, d_count(dn),
1096                      realdn, d_count(realdn),
1097                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1098                 dput(dn);
1099                 dn = realdn;
1100         } else {
1101                 BUG_ON(!ceph_dentry(dn));
1102                 dout("dn %p attached to %p ino %llx.%llx\n",
1103                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1104         }
1105 out:
1106         return dn;
1107 }
1108
1109 /*
1110  * Incorporate results into the local cache.  This is either just
1111  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1112  * after a lookup).
1113  *
1114  * A reply may contain
1115  *         a directory inode along with a dentry.
1116  *  and/or a target inode
1117  *
1118  * Called with snap_rwsem (read).
1119  */
1120 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1121 {
1122         struct ceph_mds_session *session = req->r_session;
1123         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1124         struct inode *in = NULL;
1125         struct ceph_vino tvino, dvino;
1126         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1127         int err = 0;
1128
1129         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1130              rinfo->head->is_dentry, rinfo->head->is_target);
1131
1132         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1133                 dout("fill_trace reply is empty!\n");
1134                 if (rinfo->head->result == 0 && req->r_parent)
1135                         ceph_invalidate_dir_request(req);
1136                 return 0;
1137         }
1138
1139         if (rinfo->head->is_dentry) {
1140                 struct inode *dir = req->r_parent;
1141
1142                 if (dir) {
1143                         err = fill_inode(dir, NULL,
1144                                          &rinfo->diri, rinfo->dirfrag,
1145                                          session, req->r_request_started, -1,
1146                                          &req->r_caps_reservation);
1147                         if (err < 0)
1148                                 goto done;
1149                 } else {
1150                         WARN_ON_ONCE(1);
1151                 }
1152
1153                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1154                         struct qstr dname;
1155                         struct dentry *dn, *parent;
1156
1157                         BUG_ON(!rinfo->head->is_target);
1158                         BUG_ON(req->r_dentry);
1159
1160                         parent = d_find_any_alias(dir);
1161                         BUG_ON(!parent);
1162
1163                         dname.name = rinfo->dname;
1164                         dname.len = rinfo->dname_len;
1165                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1166                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1167                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1168 retry_lookup:
1169                         dn = d_lookup(parent, &dname);
1170                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1171                              parent, dname.len, dname.name, dn);
1172
1173                         if (!dn) {
1174                                 dn = d_alloc(parent, &dname);
1175                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1176                                      dname.len, dname.name, dn);
1177                                 if (!dn) {
1178                                         dput(parent);
1179                                         err = -ENOMEM;
1180                                         goto done;
1181                                 }
1182                                 err = 0;
1183                         } else if (d_really_is_positive(dn) &&
1184                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1185                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1186                                 dout(" dn %p points to wrong inode %p\n",
1187                                      dn, d_inode(dn));
1188                                 ceph_dir_clear_ordered(dir);
1189                                 d_delete(dn);
1190                                 dput(dn);
1191                                 goto retry_lookup;
1192                         }
1193
1194                         req->r_dentry = dn;
1195                         dput(parent);
1196                 }
1197         }
1198
1199         if (rinfo->head->is_target) {
1200                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1201                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1202
1203                 in = ceph_get_inode(sb, tvino);
1204                 if (IS_ERR(in)) {
1205                         err = PTR_ERR(in);
1206                         goto done;
1207                 }
1208                 req->r_target_inode = in;
1209
1210                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1211                                 session, req->r_request_started,
1212                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1213                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1214                                 &req->r_caps_reservation);
1215                 if (err < 0) {
1216                         pr_err("fill_inode badness %p %llx.%llx\n",
1217                                 in, ceph_vinop(in));
1218                         goto done;
1219                 }
1220         }
1221
1222         /*
1223          * ignore null lease/binding on snapdir ENOENT, or else we
1224          * will have trouble splicing in the virtual snapdir later
1225          */
1226         if (rinfo->head->is_dentry &&
1227             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1228             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1229             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1230                                                fsc->mount_options->snapdir_name,
1231                                                req->r_dentry->d_name.len))) {
1232                 /*
1233                  * lookup link rename   : null -> possibly existing inode
1234                  * mknod symlink mkdir  : null -> new inode
1235                  * unlink               : linked -> null
1236                  */
1237                 struct inode *dir = req->r_parent;
1238                 struct dentry *dn = req->r_dentry;
1239                 bool have_dir_cap, have_lease;
1240
1241                 BUG_ON(!dn);
1242                 BUG_ON(!dir);
1243                 BUG_ON(d_inode(dn->d_parent) != dir);
1244
1245                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1246                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1247
1248                 BUG_ON(ceph_ino(dir) != dvino.ino);
1249                 BUG_ON(ceph_snap(dir) != dvino.snap);
1250
1251                 /* do we have a lease on the whole dir? */
1252                 have_dir_cap =
1253                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1254                          CEPH_CAP_FILE_SHARED);
1255
1256                 /* do we have a dn lease? */
1257                 have_lease = have_dir_cap ||
1258                         le32_to_cpu(rinfo->dlease->duration_ms);
1259                 if (!have_lease)
1260                         dout("fill_trace  no dentry lease or dir cap\n");
1261
1262                 /* rename? */
1263                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1264                         struct inode *olddir = req->r_old_dentry_dir;
1265                         BUG_ON(!olddir);
1266
1267                         dout(" src %p '%pd' dst %p '%pd'\n",
1268                              req->r_old_dentry,
1269                              req->r_old_dentry,
1270                              dn, dn);
1271                         dout("fill_trace doing d_move %p -> %p\n",
1272                              req->r_old_dentry, dn);
1273
1274                         /* d_move screws up sibling dentries' offsets */
1275                         ceph_dir_clear_ordered(dir);
1276                         ceph_dir_clear_ordered(olddir);
1277
1278                         d_move(req->r_old_dentry, dn);
1279                         dout(" src %p '%pd' dst %p '%pd'\n",
1280                              req->r_old_dentry,
1281                              req->r_old_dentry,
1282                              dn, dn);
1283
1284                         /* ensure target dentry is invalidated, despite
1285                            rehashing bug in vfs_rename_dir */
1286                         ceph_invalidate_dentry_lease(dn);
1287
1288                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1289                              ceph_dentry(req->r_old_dentry)->offset);
1290
1291                         dn = req->r_old_dentry;  /* use old_dentry */
1292                 }
1293
1294                 /* null dentry? */
1295                 if (!rinfo->head->is_target) {
1296                         dout("fill_trace null dentry\n");
1297                         if (d_really_is_positive(dn)) {
1298                                 ceph_dir_clear_ordered(dir);
1299                                 dout("d_delete %p\n", dn);
1300                                 d_delete(dn);
1301                         } else if (have_lease) {
1302                                 if (d_unhashed(dn))
1303                                         d_add(dn, NULL);
1304                                 update_dentry_lease(dn, rinfo->dlease,
1305                                                     session,
1306                                                     req->r_request_started,
1307                                                     NULL, &dvino);
1308                         }
1309                         goto done;
1310                 }
1311
1312                 /* attach proper inode */
1313                 if (d_really_is_negative(dn)) {
1314                         ceph_dir_clear_ordered(dir);
1315                         ihold(in);
1316                         dn = splice_dentry(dn, in);
1317                         if (IS_ERR(dn)) {
1318                                 err = PTR_ERR(dn);
1319                                 goto done;
1320                         }
1321                         req->r_dentry = dn;  /* may have spliced */
1322                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1323                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1324                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1325                              ceph_vinop(in));
1326                         ceph_dir_clear_ordered(dir);
1327                         d_invalidate(dn);
1328                         have_lease = false;
1329                 }
1330
1331                 if (have_lease) {
1332                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1333                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1334                         update_dentry_lease(dn, rinfo->dlease, session,
1335                                             req->r_request_started,
1336                                             &tvino, &dvino);
1337                 }
1338                 dout(" final dn %p\n", dn);
1339         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1340                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1341                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1342                 struct dentry *dn = req->r_dentry;
1343                 struct inode *dir = req->r_parent;
1344
1345                 /* fill out a snapdir LOOKUPSNAP dentry */
1346                 BUG_ON(!dn);
1347                 BUG_ON(!dir);
1348                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1349                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1350                 ceph_dir_clear_ordered(dir);
1351                 ihold(in);
1352                 dn = splice_dentry(dn, in);
1353                 if (IS_ERR(dn)) {
1354                         err = PTR_ERR(dn);
1355                         goto done;
1356                 }
1357                 req->r_dentry = dn;  /* may have spliced */
1358         } else if (rinfo->head->is_dentry) {
1359                 struct ceph_vino *ptvino = NULL;
1360
1361                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1362                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1363                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1364                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1365
1366                         if (rinfo->head->is_target) {
1367                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1368                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1369                                 ptvino = &tvino;
1370                         }
1371
1372                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1373                                 session, req->r_request_started, ptvino,
1374                                 &dvino);
1375                 } else {
1376                         dout("%s: no dentry lease or dir cap\n", __func__);
1377                 }
1378         }
1379 done:
1380         dout("fill_trace done err=%d\n", err);
1381         return err;
1382 }
1383
1384 /*
1385  * Prepopulate our cache with readdir results, leases, etc.
1386  */
1387 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1388                                            struct ceph_mds_session *session)
1389 {
1390         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1391         int i, err = 0;
1392
1393         for (i = 0; i < rinfo->dir_nr; i++) {
1394                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1395                 struct ceph_vino vino;
1396                 struct inode *in;
1397                 int rc;
1398
1399                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1400                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1401
1402                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1403                 if (IS_ERR(in)) {
1404                         err = PTR_ERR(in);
1405                         dout("new_inode badness got %d\n", err);
1406                         continue;
1407                 }
1408                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1409                                 req->r_request_started, -1,
1410                                 &req->r_caps_reservation);
1411                 if (rc < 0) {
1412                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1413                         err = rc;
1414                 }
1415                 iput(in);
1416         }
1417
1418         return err;
1419 }
1420
1421 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1422 {
1423         if (ctl->page) {
1424                 kunmap(ctl->page);
1425                 put_page(ctl->page);
1426                 ctl->page = NULL;
1427         }
1428 }
1429
1430 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1431                               struct ceph_readdir_cache_control *ctl,
1432                               struct ceph_mds_request *req)
1433 {
1434         struct ceph_inode_info *ci = ceph_inode(dir);
1435         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1436         unsigned idx = ctl->index % nsize;
1437         pgoff_t pgoff = ctl->index / nsize;
1438
1439         if (!ctl->page || pgoff != page_index(ctl->page)) {
1440                 ceph_readdir_cache_release(ctl);
1441                 if (idx == 0)
1442                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1443                 else
1444                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1445                 if (!ctl->page) {
1446                         ctl->index = -1;
1447                         return idx == 0 ? -ENOMEM : 0;
1448                 }
1449                 /* reading/filling the cache are serialized by
1450                  * i_mutex, no need to use page lock */
1451                 unlock_page(ctl->page);
1452                 ctl->dentries = kmap(ctl->page);
1453                 if (idx == 0)
1454                         memset(ctl->dentries, 0, PAGE_SIZE);
1455         }
1456
1457         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1458             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1459                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1460                 ctl->dentries[idx] = dn;
1461                 ctl->index++;
1462         } else {
1463                 dout("disable readdir cache\n");
1464                 ctl->index = -1;
1465         }
1466         return 0;
1467 }
1468
1469 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1470                              struct ceph_mds_session *session)
1471 {
1472         struct dentry *parent = req->r_dentry;
1473         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1474         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1475         struct qstr dname;
1476         struct dentry *dn;
1477         struct inode *in;
1478         int err = 0, skipped = 0, ret, i;
1479         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1480         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1481         u32 last_hash = 0;
1482         u32 fpos_offset;
1483         struct ceph_readdir_cache_control cache_ctl = {};
1484
1485         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1486                 return readdir_prepopulate_inodes_only(req, session);
1487
1488         if (rinfo->hash_order) {
1489                 if (req->r_path2) {
1490                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1491                                                   req->r_path2,
1492                                                   strlen(req->r_path2));
1493                         last_hash = ceph_frag_value(last_hash);
1494                 } else if (rinfo->offset_hash) {
1495                         /* mds understands offset_hash */
1496                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1497                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1498                 }
1499         }
1500
1501         if (rinfo->dir_dir &&
1502             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1503                 dout("readdir_prepopulate got new frag %x -> %x\n",
1504                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1505                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1506                 if (!rinfo->hash_order)
1507                         req->r_readdir_offset = 2;
1508         }
1509
1510         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1511                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1512                      rinfo->dir_nr, parent);
1513         } else {
1514                 dout("readdir_prepopulate %d items under dn %p\n",
1515                      rinfo->dir_nr, parent);
1516                 if (rinfo->dir_dir)
1517                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1518
1519                 if (ceph_frag_is_leftmost(frag) &&
1520                     req->r_readdir_offset == 2 &&
1521                     !(rinfo->hash_order && last_hash)) {
1522                         /* note dir version at start of readdir so we can
1523                          * tell if any dentries get dropped */
1524                         req->r_dir_release_cnt =
1525                                 atomic64_read(&ci->i_release_count);
1526                         req->r_dir_ordered_cnt =
1527                                 atomic64_read(&ci->i_ordered_count);
1528                         req->r_readdir_cache_idx = 0;
1529                 }
1530         }
1531
1532         cache_ctl.index = req->r_readdir_cache_idx;
1533         fpos_offset = req->r_readdir_offset;
1534
1535         /* FIXME: release caps/leases if error occurs */
1536         for (i = 0; i < rinfo->dir_nr; i++) {
1537                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1538                 struct ceph_vino tvino, dvino;
1539
1540                 dname.name = rde->name;
1541                 dname.len = rde->name_len;
1542                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1543
1544                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1545                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1546
1547                 if (rinfo->hash_order) {
1548                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1549                                                  rde->name, rde->name_len);
1550                         hash = ceph_frag_value(hash);
1551                         if (hash != last_hash)
1552                                 fpos_offset = 2;
1553                         last_hash = hash;
1554                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1555                 } else {
1556                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1557                 }
1558
1559 retry_lookup:
1560                 dn = d_lookup(parent, &dname);
1561                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1562                      parent, dname.len, dname.name, dn);
1563
1564                 if (!dn) {
1565                         dn = d_alloc(parent, &dname);
1566                         dout("d_alloc %p '%.*s' = %p\n", parent,
1567                              dname.len, dname.name, dn);
1568                         if (!dn) {
1569                                 dout("d_alloc badness\n");
1570                                 err = -ENOMEM;
1571                                 goto out;
1572                         }
1573                 } else if (d_really_is_positive(dn) &&
1574                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1575                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1576                         dout(" dn %p points to wrong inode %p\n",
1577                              dn, d_inode(dn));
1578                         __ceph_dir_clear_ordered(ci);
1579                         d_delete(dn);
1580                         dput(dn);
1581                         goto retry_lookup;
1582                 }
1583
1584                 /* inode */
1585                 if (d_really_is_positive(dn)) {
1586                         in = d_inode(dn);
1587                 } else {
1588                         in = ceph_get_inode(parent->d_sb, tvino);
1589                         if (IS_ERR(in)) {
1590                                 dout("new_inode badness\n");
1591                                 d_drop(dn);
1592                                 dput(dn);
1593                                 err = PTR_ERR(in);
1594                                 goto out;
1595                         }
1596                 }
1597
1598                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1599                                  req->r_request_started, -1,
1600                                  &req->r_caps_reservation);
1601                 if (ret < 0) {
1602                         pr_err("fill_inode badness on %p\n", in);
1603                         if (d_really_is_positive(dn))
1604                                 __ceph_dir_clear_ordered(ci);
1605                         else
1606                                 iput(in);
1607                         d_drop(dn);
1608                         err = ret;
1609                         goto next_item;
1610                 }
1611
1612                 if (d_really_is_negative(dn)) {
1613                         struct dentry *realdn;
1614
1615                         if (ceph_security_xattr_deadlock(in)) {
1616                                 dout(" skip splicing dn %p to inode %p"
1617                                      " (security xattr deadlock)\n", dn, in);
1618                                 iput(in);
1619                                 skipped++;
1620                                 goto next_item;
1621                         }
1622
1623                         realdn = splice_dentry(dn, in);
1624                         if (IS_ERR(realdn)) {
1625                                 err = PTR_ERR(realdn);
1626                                 d_drop(dn);
1627                                 dn = NULL;
1628                                 goto next_item;
1629                         }
1630                         dn = realdn;
1631                 }
1632
1633                 ceph_dentry(dn)->offset = rde->offset;
1634
1635                 dvino = ceph_vino(d_inode(parent));
1636                 update_dentry_lease(dn, rde->lease, req->r_session,
1637                                     req->r_request_started, &tvino, &dvino);
1638
1639                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1640                         ret = fill_readdir_cache(d_inode(parent), dn,
1641                                                  &cache_ctl, req);
1642                         if (ret < 0)
1643                                 err = ret;
1644                 }
1645 next_item:
1646                 if (dn)
1647                         dput(dn);
1648         }
1649 out:
1650         if (err == 0 && skipped == 0) {
1651                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1652                 req->r_readdir_cache_idx = cache_ctl.index;
1653         }
1654         ceph_readdir_cache_release(&cache_ctl);
1655         dout("readdir_prepopulate done\n");
1656         return err;
1657 }
1658
1659 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1660 {
1661         struct ceph_inode_info *ci = ceph_inode(inode);
1662         bool ret;
1663
1664         spin_lock(&ci->i_ceph_lock);
1665         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1666         i_size_write(inode, size);
1667         inode->i_blocks = calc_inode_blocks(size);
1668
1669         ret = __ceph_should_report_size(ci);
1670
1671         spin_unlock(&ci->i_ceph_lock);
1672         return ret;
1673 }
1674
1675 /*
1676  * Write back inode data in a worker thread.  (This can't be done
1677  * in the message handler context.)
1678  */
1679 void ceph_queue_writeback(struct inode *inode)
1680 {
1681         ihold(inode);
1682         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1683                        &ceph_inode(inode)->i_wb_work)) {
1684                 dout("ceph_queue_writeback %p\n", inode);
1685         } else {
1686                 dout("ceph_queue_writeback %p failed\n", inode);
1687                 iput(inode);
1688         }
1689 }
1690
1691 static void ceph_writeback_work(struct work_struct *work)
1692 {
1693         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1694                                                   i_wb_work);
1695         struct inode *inode = &ci->vfs_inode;
1696
1697         dout("writeback %p\n", inode);
1698         filemap_fdatawrite(&inode->i_data);
1699         iput(inode);
1700 }
1701
1702 /*
1703  * queue an async invalidation
1704  */
1705 void ceph_queue_invalidate(struct inode *inode)
1706 {
1707         ihold(inode);
1708         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1709                        &ceph_inode(inode)->i_pg_inv_work)) {
1710                 dout("ceph_queue_invalidate %p\n", inode);
1711         } else {
1712                 dout("ceph_queue_invalidate %p failed\n", inode);
1713                 iput(inode);
1714         }
1715 }
1716
1717 /*
1718  * Invalidate inode pages in a worker thread.  (This can't be done
1719  * in the message handler context.)
1720  */
1721 static void ceph_invalidate_work(struct work_struct *work)
1722 {
1723         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1724                                                   i_pg_inv_work);
1725         struct inode *inode = &ci->vfs_inode;
1726         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1727         u32 orig_gen;
1728         int check = 0;
1729
1730         mutex_lock(&ci->i_truncate_mutex);
1731
1732         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1733                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1734                                     inode, ceph_ino(inode));
1735                 mapping_set_error(inode->i_mapping, -EIO);
1736                 truncate_pagecache(inode, 0);
1737                 mutex_unlock(&ci->i_truncate_mutex);
1738                 goto out;
1739         }
1740
1741         spin_lock(&ci->i_ceph_lock);
1742         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1743              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1744         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1745                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1746                         check = 1;
1747                 spin_unlock(&ci->i_ceph_lock);
1748                 mutex_unlock(&ci->i_truncate_mutex);
1749                 goto out;
1750         }
1751         orig_gen = ci->i_rdcache_gen;
1752         spin_unlock(&ci->i_ceph_lock);
1753
1754         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1755                 pr_err("invalidate_pages %p fails\n", inode);
1756         }
1757
1758         spin_lock(&ci->i_ceph_lock);
1759         if (orig_gen == ci->i_rdcache_gen &&
1760             orig_gen == ci->i_rdcache_revoking) {
1761                 dout("invalidate_pages %p gen %d successful\n", inode,
1762                      ci->i_rdcache_gen);
1763                 ci->i_rdcache_revoking--;
1764                 check = 1;
1765         } else {
1766                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1767                      inode, orig_gen, ci->i_rdcache_gen,
1768                      ci->i_rdcache_revoking);
1769                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1770                         check = 1;
1771         }
1772         spin_unlock(&ci->i_ceph_lock);
1773         mutex_unlock(&ci->i_truncate_mutex);
1774 out:
1775         if (check)
1776                 ceph_check_caps(ci, 0, NULL);
1777         iput(inode);
1778 }
1779
1780
1781 /*
1782  * called by trunc_wq;
1783  *
1784  * We also truncate in a separate thread as well.
1785  */
1786 static void ceph_vmtruncate_work(struct work_struct *work)
1787 {
1788         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1789                                                   i_vmtruncate_work);
1790         struct inode *inode = &ci->vfs_inode;
1791
1792         dout("vmtruncate_work %p\n", inode);
1793         __ceph_do_pending_vmtruncate(inode);
1794         iput(inode);
1795 }
1796
1797 /*
1798  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1799  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1800  */
1801 void ceph_queue_vmtruncate(struct inode *inode)
1802 {
1803         struct ceph_inode_info *ci = ceph_inode(inode);
1804
1805         ihold(inode);
1806
1807         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1808                        &ci->i_vmtruncate_work)) {
1809                 dout("ceph_queue_vmtruncate %p\n", inode);
1810         } else {
1811                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1812                      inode, ci->i_truncate_pending);
1813                 iput(inode);
1814         }
1815 }
1816
1817 /*
1818  * Make sure any pending truncation is applied before doing anything
1819  * that may depend on it.
1820  */
1821 void __ceph_do_pending_vmtruncate(struct inode *inode)
1822 {
1823         struct ceph_inode_info *ci = ceph_inode(inode);
1824         u64 to;
1825         int wrbuffer_refs, finish = 0;
1826
1827         mutex_lock(&ci->i_truncate_mutex);
1828 retry:
1829         spin_lock(&ci->i_ceph_lock);
1830         if (ci->i_truncate_pending == 0) {
1831                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1832                 spin_unlock(&ci->i_ceph_lock);
1833                 mutex_unlock(&ci->i_truncate_mutex);
1834                 return;
1835         }
1836
1837         /*
1838          * make sure any dirty snapped pages are flushed before we
1839          * possibly truncate them.. so write AND block!
1840          */
1841         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1842                 struct ceph_cap_snap *capsnap;
1843                 to = ci->i_truncate_size;
1844                 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1845                         // MDS should have revoked Frw caps
1846                         WARN_ON_ONCE(capsnap->writing);
1847                         if (capsnap->dirty_pages && capsnap->size > to)
1848                                 to = capsnap->size;
1849                 }
1850                 spin_unlock(&ci->i_ceph_lock);
1851                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1852                      inode);
1853
1854                 truncate_pagecache(inode, to);
1855
1856                 filemap_write_and_wait_range(&inode->i_data, 0,
1857                                              inode->i_sb->s_maxbytes);
1858                 goto retry;
1859         }
1860
1861         /* there should be no reader or writer */
1862         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1863
1864         to = ci->i_truncate_size;
1865         wrbuffer_refs = ci->i_wrbuffer_ref;
1866         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1867              ci->i_truncate_pending, to);
1868         spin_unlock(&ci->i_ceph_lock);
1869
1870         truncate_pagecache(inode, to);
1871
1872         spin_lock(&ci->i_ceph_lock);
1873         if (to == ci->i_truncate_size) {
1874                 ci->i_truncate_pending = 0;
1875                 finish = 1;
1876         }
1877         spin_unlock(&ci->i_ceph_lock);
1878         if (!finish)
1879                 goto retry;
1880
1881         mutex_unlock(&ci->i_truncate_mutex);
1882
1883         if (wrbuffer_refs == 0)
1884                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1885
1886         wake_up_all(&ci->i_cap_wq);
1887 }
1888
1889 /*
1890  * symlinks
1891  */
1892 static const struct inode_operations ceph_symlink_iops = {
1893         .get_link = simple_get_link,
1894         .setattr = ceph_setattr,
1895         .getattr = ceph_getattr,
1896         .listxattr = ceph_listxattr,
1897 };
1898
1899 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1900 {
1901         struct ceph_inode_info *ci = ceph_inode(inode);
1902         const unsigned int ia_valid = attr->ia_valid;
1903         struct ceph_mds_request *req;
1904         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1905         struct ceph_cap_flush *prealloc_cf;
1906         int issued;
1907         int release = 0, dirtied = 0;
1908         int mask = 0;
1909         int err = 0;
1910         int inode_dirty_flags = 0;
1911         bool lock_snap_rwsem = false;
1912
1913         prealloc_cf = ceph_alloc_cap_flush();
1914         if (!prealloc_cf)
1915                 return -ENOMEM;
1916
1917         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1918                                        USE_AUTH_MDS);
1919         if (IS_ERR(req)) {
1920                 ceph_free_cap_flush(prealloc_cf);
1921                 return PTR_ERR(req);
1922         }
1923
1924         spin_lock(&ci->i_ceph_lock);
1925         issued = __ceph_caps_issued(ci, NULL);
1926
1927         if (!ci->i_head_snapc &&
1928             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1929                 lock_snap_rwsem = true;
1930                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1931                         spin_unlock(&ci->i_ceph_lock);
1932                         down_read(&mdsc->snap_rwsem);
1933                         spin_lock(&ci->i_ceph_lock);
1934                         issued = __ceph_caps_issued(ci, NULL);
1935                 }
1936         }
1937
1938         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1939
1940         if (ia_valid & ATTR_UID) {
1941                 dout("setattr %p uid %d -> %d\n", inode,
1942                      from_kuid(&init_user_ns, inode->i_uid),
1943                      from_kuid(&init_user_ns, attr->ia_uid));
1944                 if (issued & CEPH_CAP_AUTH_EXCL) {
1945                         inode->i_uid = attr->ia_uid;
1946                         dirtied |= CEPH_CAP_AUTH_EXCL;
1947                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1948                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1949                         req->r_args.setattr.uid = cpu_to_le32(
1950                                 from_kuid(&init_user_ns, attr->ia_uid));
1951                         mask |= CEPH_SETATTR_UID;
1952                         release |= CEPH_CAP_AUTH_SHARED;
1953                 }
1954         }
1955         if (ia_valid & ATTR_GID) {
1956                 dout("setattr %p gid %d -> %d\n", inode,
1957                      from_kgid(&init_user_ns, inode->i_gid),
1958                      from_kgid(&init_user_ns, attr->ia_gid));
1959                 if (issued & CEPH_CAP_AUTH_EXCL) {
1960                         inode->i_gid = attr->ia_gid;
1961                         dirtied |= CEPH_CAP_AUTH_EXCL;
1962                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1963                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1964                         req->r_args.setattr.gid = cpu_to_le32(
1965                                 from_kgid(&init_user_ns, attr->ia_gid));
1966                         mask |= CEPH_SETATTR_GID;
1967                         release |= CEPH_CAP_AUTH_SHARED;
1968                 }
1969         }
1970         if (ia_valid & ATTR_MODE) {
1971                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1972                      attr->ia_mode);
1973                 if (issued & CEPH_CAP_AUTH_EXCL) {
1974                         inode->i_mode = attr->ia_mode;
1975                         dirtied |= CEPH_CAP_AUTH_EXCL;
1976                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1977                            attr->ia_mode != inode->i_mode) {
1978                         inode->i_mode = attr->ia_mode;
1979                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1980                         mask |= CEPH_SETATTR_MODE;
1981                         release |= CEPH_CAP_AUTH_SHARED;
1982                 }
1983         }
1984
1985         if (ia_valid & ATTR_ATIME) {
1986                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1987                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1988                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1989                 if (issued & CEPH_CAP_FILE_EXCL) {
1990                         ci->i_time_warp_seq++;
1991                         inode->i_atime = attr->ia_atime;
1992                         dirtied |= CEPH_CAP_FILE_EXCL;
1993                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1994                            timespec_compare(&inode->i_atime,
1995                                             &attr->ia_atime) < 0) {
1996                         inode->i_atime = attr->ia_atime;
1997                         dirtied |= CEPH_CAP_FILE_WR;
1998                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1999                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
2000                         ceph_encode_timespec(&req->r_args.setattr.atime,
2001                                              &attr->ia_atime);
2002                         mask |= CEPH_SETATTR_ATIME;
2003                         release |= CEPH_CAP_FILE_SHARED |
2004                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2005                 }
2006         }
2007         if (ia_valid & ATTR_MTIME) {
2008                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
2009                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2010                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2011                 if (issued & CEPH_CAP_FILE_EXCL) {
2012                         ci->i_time_warp_seq++;
2013                         inode->i_mtime = attr->ia_mtime;
2014                         dirtied |= CEPH_CAP_FILE_EXCL;
2015                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2016                            timespec_compare(&inode->i_mtime,
2017                                             &attr->ia_mtime) < 0) {
2018                         inode->i_mtime = attr->ia_mtime;
2019                         dirtied |= CEPH_CAP_FILE_WR;
2020                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2021                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2022                         ceph_encode_timespec(&req->r_args.setattr.mtime,
2023                                              &attr->ia_mtime);
2024                         mask |= CEPH_SETATTR_MTIME;
2025                         release |= CEPH_CAP_FILE_SHARED |
2026                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2027                 }
2028         }
2029         if (ia_valid & ATTR_SIZE) {
2030                 dout("setattr %p size %lld -> %lld\n", inode,
2031                      inode->i_size, attr->ia_size);
2032                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2033                     attr->ia_size > inode->i_size) {
2034                         i_size_write(inode, attr->ia_size);
2035                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2036                         ci->i_reported_size = attr->ia_size;
2037                         dirtied |= CEPH_CAP_FILE_EXCL;
2038                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2039                            attr->ia_size != inode->i_size) {
2040                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2041                         req->r_args.setattr.old_size =
2042                                 cpu_to_le64(inode->i_size);
2043                         mask |= CEPH_SETATTR_SIZE;
2044                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2045                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2046                 }
2047         }
2048
2049         /* these do nothing */
2050         if (ia_valid & ATTR_CTIME) {
2051                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2052                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2053                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2054                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2055                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2056                      only ? "ctime only" : "ignored");
2057                 if (only) {
2058                         /*
2059                          * if kernel wants to dirty ctime but nothing else,
2060                          * we need to choose a cap to dirty under, or do
2061                          * a almost-no-op setattr
2062                          */
2063                         if (issued & CEPH_CAP_AUTH_EXCL)
2064                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2065                         else if (issued & CEPH_CAP_FILE_EXCL)
2066                                 dirtied |= CEPH_CAP_FILE_EXCL;
2067                         else if (issued & CEPH_CAP_XATTR_EXCL)
2068                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2069                         else
2070                                 mask |= CEPH_SETATTR_CTIME;
2071                 }
2072         }
2073         if (ia_valid & ATTR_FILE)
2074                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2075
2076         if (dirtied) {
2077                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2078                                                            &prealloc_cf);
2079                 inode->i_ctime = attr->ia_ctime;
2080         }
2081
2082         release &= issued;
2083         spin_unlock(&ci->i_ceph_lock);
2084         if (lock_snap_rwsem)
2085                 up_read(&mdsc->snap_rwsem);
2086
2087         if (inode_dirty_flags)
2088                 __mark_inode_dirty(inode, inode_dirty_flags);
2089
2090
2091         if (mask) {
2092                 req->r_inode = inode;
2093                 ihold(inode);
2094                 req->r_inode_drop = release;
2095                 req->r_args.setattr.mask = cpu_to_le32(mask);
2096                 req->r_num_caps = 1;
2097                 req->r_stamp = attr->ia_ctime;
2098                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2099         }
2100         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2101              ceph_cap_string(dirtied), mask);
2102
2103         ceph_mdsc_put_request(req);
2104         ceph_free_cap_flush(prealloc_cf);
2105
2106         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2107                 __ceph_do_pending_vmtruncate(inode);
2108
2109         return err;
2110 }
2111
2112 /*
2113  * setattr
2114  */
2115 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2116 {
2117         struct inode *inode = d_inode(dentry);
2118         int err;
2119
2120         if (ceph_snap(inode) != CEPH_NOSNAP)
2121                 return -EROFS;
2122
2123         err = setattr_prepare(dentry, attr);
2124         if (err != 0)
2125                 return err;
2126
2127         err = __ceph_setattr(inode, attr);
2128
2129         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2130                 err = posix_acl_chmod(inode, attr->ia_mode);
2131
2132         return err;
2133 }
2134
2135 /*
2136  * Verify that we have a lease on the given mask.  If not,
2137  * do a getattr against an mds.
2138  */
2139 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2140                       int mask, bool force)
2141 {
2142         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2143         struct ceph_mds_client *mdsc = fsc->mdsc;
2144         struct ceph_mds_request *req;
2145         int err;
2146
2147         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2148                 dout("do_getattr inode %p SNAPDIR\n", inode);
2149                 return 0;
2150         }
2151
2152         dout("do_getattr inode %p mask %s mode 0%o\n",
2153              inode, ceph_cap_string(mask), inode->i_mode);
2154         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2155                 return 0;
2156
2157         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2158         if (IS_ERR(req))
2159                 return PTR_ERR(req);
2160         req->r_inode = inode;
2161         ihold(inode);
2162         req->r_num_caps = 1;
2163         req->r_args.getattr.mask = cpu_to_le32(mask);
2164         req->r_locked_page = locked_page;
2165         err = ceph_mdsc_do_request(mdsc, NULL, req);
2166         if (locked_page && err == 0) {
2167                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2168                 if (inline_version == 0) {
2169                         /* the reply is supposed to contain inline data */
2170                         err = -EINVAL;
2171                 } else if (inline_version == CEPH_INLINE_NONE) {
2172                         err = -ENODATA;
2173                 } else {
2174                         err = req->r_reply_info.targeti.inline_len;
2175                 }
2176         }
2177         ceph_mdsc_put_request(req);
2178         dout("do_getattr result=%d\n", err);
2179         return err;
2180 }
2181
2182
2183 /*
2184  * Check inode permissions.  We verify we have a valid value for
2185  * the AUTH cap, then call the generic handler.
2186  */
2187 int ceph_permission(struct inode *inode, int mask)
2188 {
2189         int err;
2190
2191         if (mask & MAY_NOT_BLOCK)
2192                 return -ECHILD;
2193
2194         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2195
2196         if (!err)
2197                 err = generic_permission(inode, mask);
2198         return err;
2199 }
2200
2201 /*
2202  * Get all attributes.  Hopefully somedata we'll have a statlite()
2203  * and can limit the fields we require to be accurate.
2204  */
2205 int ceph_getattr(const struct path *path, struct kstat *stat,
2206                  u32 request_mask, unsigned int flags)
2207 {
2208         struct inode *inode = d_inode(path->dentry);
2209         struct ceph_inode_info *ci = ceph_inode(inode);
2210         int err;
2211
2212         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2213         if (!err) {
2214                 generic_fillattr(inode, stat);
2215                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2216                 if (ceph_snap(inode) != CEPH_NOSNAP)
2217                         stat->dev = ceph_snap(inode);
2218                 else
2219                         stat->dev = 0;
2220                 if (S_ISDIR(inode->i_mode)) {
2221                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2222                                                 RBYTES))
2223                                 stat->size = ci->i_rbytes;
2224                         else
2225                                 stat->size = ci->i_files + ci->i_subdirs;
2226                         stat->blocks = 0;
2227                         stat->blksize = 65536;
2228                 }
2229         }
2230         return err;
2231 }