fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16
  17 #include "super.h"
  18 #include "mds_client.h"
  19 #include "cache.h"
  20 #include <linux/ceph/decode.h>
  21
  22 /*
  23  * Ceph inode operations
  24  *
  25  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  26  * setattr, etc.), xattr helpers, and helpers for assimilating
  27  * metadata returned by the MDS into our cache.
  28  *
  29  * Also define helpers for doing asynchronous writeback, invalidation,
  30  * and truncation for the benefit of those who can't afford to block
  31  * (typically because they are in the message handler path).
  32  */
  33
  34 static const struct inode_operations ceph_symlink_iops;
  35
  36 static void ceph_invalidate_work(struct work_struct *work);
  37 static void ceph_writeback_work(struct work_struct *work);
  38 static void ceph_vmtruncate_work(struct work_struct *work);
  39
  40 /*
  41  * find or create an inode, given the ceph ino number
  42  */
  43 static int ceph_set_ino_cb(struct inode *inode, void *data)
  44 {
  45         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  46         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  47         return 0;
  48 }
  49
  50 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  51 {
  52         struct inode *inode;
  53         ino_t t = ceph_vino_to_ino(vino);
  54
  55         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  56         if (!inode)
  57                 return ERR_PTR(-ENOMEM);
  58         if (inode->i_state & I_NEW) {
  59                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  60                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  61                 unlock_new_inode(inode);
  62         }
  63
  64         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  65              vino.snap, inode);
  66         return inode;
  67 }
  68
  69 /*
  70  * get/constuct snapdir inode for a given directory
  71  */
  72 struct inode *ceph_get_snapdir(struct inode *parent)
  73 {
  74         struct ceph_vino vino = {
  75                 .ino = ceph_ino(parent),
  76                 .snap = CEPH_SNAPDIR,
  77         };
  78         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  79         struct ceph_inode_info *ci = ceph_inode(inode);
  80
  81         BUG_ON(!S_ISDIR(parent->i_mode));
  82         if (IS_ERR(inode))
  83                 return inode;
  84         inode->i_mode = parent->i_mode;
  85         inode->i_uid = parent->i_uid;
  86         inode->i_gid = parent->i_gid;
  87         inode->i_op = &ceph_snapdir_iops;
  88         inode->i_fop = &ceph_snapdir_fops;
  89         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  90         ci->i_rbytes = 0;
  91         return inode;
  92 }
  93
  94 const struct inode_operations ceph_file_iops = {
  95         .permission = ceph_permission,
  96         .setattr = ceph_setattr,
  97         .getattr = ceph_getattr,
  98         .listxattr = ceph_listxattr,
  99         .get_acl = ceph_get_acl,
 100         .set_acl = ceph_set_acl,
 101 };
 102
 103
 104 /*
 105  * We use a 'frag tree' to keep track of the MDS's directory fragments
 106  * for a given inode (usually there is just a single fragment).  We
 107  * need to know when a child frag is delegated to a new MDS, or when
 108  * it is flagged as replicated, so we can direct our requests
 109  * accordingly.
 110  */
 111
 112 /*
 113  * find/create a frag in the tree
 114  */
 115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 116                                                     u32 f)
 117 {
 118         struct rb_node **p;
 119         struct rb_node *parent = NULL;
 120         struct ceph_inode_frag *frag;
 121         int c;
 122
 123         p = &ci->i_fragtree.rb_node;
 124         while (*p) {
 125                 parent = *p;
 126                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 127                 c = ceph_frag_compare(f, frag->frag);
 128                 if (c < 0)
 129                         p = &(*p)->rb_left;
 130                 else if (c > 0)
 131                         p = &(*p)->rb_right;
 132                 else
 133                         return frag;
 134         }
 135
 136         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 137         if (!frag)
 138                 return ERR_PTR(-ENOMEM);
 139
 140         frag->frag = f;
 141         frag->split_by = 0;
 142         frag->mds = -1;
 143         frag->ndist = 0;
 144
 145         rb_link_node(&frag->node, parent, p);
 146         rb_insert_color(&frag->node, &ci->i_fragtree);
 147
 148         dout("get_or_create_frag added %llx.%llx frag %x\n",
 149              ceph_vinop(&ci->vfs_inode), f);
 150         return frag;
 151 }
 152
 153 /*
 154  * find a specific frag @f
 155  */
 156 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 157 {
 158         struct rb_node *n = ci->i_fragtree.rb_node;
 159
 160         while (n) {
 161                 struct ceph_inode_frag *frag =
 162                         rb_entry(n, struct ceph_inode_frag, node);
 163                 int c = ceph_frag_compare(f, frag->frag);
 164                 if (c < 0)
 165                         n = n->rb_left;
 166                 else if (c > 0)
 167                         n = n->rb_right;
 168                 else
 169                         return frag;
 170         }
 171         return NULL;
 172 }
 173
 174 /*
 175  * Choose frag containing the given value @v.  If @pfrag is
 176  * specified, copy the frag delegation info to the caller if
 177  * it is present.
 178  */
 179 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 180                               struct ceph_inode_frag *pfrag, int *found)
 181 {
 182         u32 t = ceph_frag_make(0, 0);
 183         struct ceph_inode_frag *frag;
 184         unsigned nway, i;
 185         u32 n;
 186
 187         if (found)
 188                 *found = 0;
 189
 190         while (1) {
 191                 WARN_ON(!ceph_frag_contains_value(t, v));
 192                 frag = __ceph_find_frag(ci, t);
 193                 if (!frag)
 194                         break; /* t is a leaf */
 195                 if (frag->split_by == 0) {
 196                         if (pfrag)
 197                                 memcpy(pfrag, frag, sizeof(*pfrag));
 198                         if (found)
 199                                 *found = 1;
 200                         break;
 201                 }
 202
 203                 /* choose child */
 204                 nway = 1 << frag->split_by;
 205                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 206                      frag->split_by, nway);
 207                 for (i = 0; i < nway; i++) {
 208                         n = ceph_frag_make_child(t, frag->split_by, i);
 209                         if (ceph_frag_contains_value(n, v)) {
 210                                 t = n;
 211                                 break;
 212                         }
 213                 }
 214                 BUG_ON(i == nway);
 215         }
 216         dout("choose_frag(%x) = %x\n", v, t);
 217
 218         return t;
 219 }
 220
 221 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 222                      struct ceph_inode_frag *pfrag, int *found)
 223 {
 224         u32 ret;
 225         mutex_lock(&ci->i_fragtree_mutex);
 226         ret = __ceph_choose_frag(ci, v, pfrag, found);
 227         mutex_unlock(&ci->i_fragtree_mutex);
 228         return ret;
 229 }
 230
 231 /*
 232  * Process dirfrag (delegation) info from the mds.  Include leaf
 233  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 234  * branches/splits are included in i_fragtree)
 235  */
 236 static int ceph_fill_dirfrag(struct inode *inode,
 237                              struct ceph_mds_reply_dirfrag *dirinfo)
 238 {
 239         struct ceph_inode_info *ci = ceph_inode(inode);
 240         struct ceph_inode_frag *frag;
 241         u32 id = le32_to_cpu(dirinfo->frag);
 242         int mds = le32_to_cpu(dirinfo->auth);
 243         int ndist = le32_to_cpu(dirinfo->ndist);
 244         int diri_auth = -1;
 245         int i;
 246         int err = 0;
 247
 248         spin_lock(&ci->i_ceph_lock);
 249         if (ci->i_auth_cap)
 250                 diri_auth = ci->i_auth_cap->mds;
 251         spin_unlock(&ci->i_ceph_lock);
 252
 253         if (mds == -1) /* CDIR_AUTH_PARENT */
 254                 mds = diri_auth;
 255
 256         mutex_lock(&ci->i_fragtree_mutex);
 257         if (ndist == 0 && mds == diri_auth) {
 258                 /* no delegation info needed. */
 259                 frag = __ceph_find_frag(ci, id);
 260                 if (!frag)
 261                         goto out;
 262                 if (frag->split_by == 0) {
 263                         /* tree leaf, remove */
 264                         dout("fill_dirfrag removed %llx.%llx frag %x"
 265                              " (no ref)\n", ceph_vinop(inode), id);
 266                         rb_erase(&frag->node, &ci->i_fragtree);
 267                         kfree(frag);
 268                 } else {
 269                         /* tree branch, keep and clear */
 270                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 271                              " referral\n", ceph_vinop(inode), id);
 272                         frag->mds = -1;
 273                         frag->ndist = 0;
 274                 }
 275                 goto out;
 276         }
 277
 278
 279         /* find/add this frag to store mds delegation info */
 280         frag = __get_or_create_frag(ci, id);
 281         if (IS_ERR(frag)) {
 282                 /* this is not the end of the world; we can continue
 283                    with bad/inaccurate delegation info */
 284                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 285                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 286                 err = -ENOMEM;
 287                 goto out;
 288         }
 289
 290         frag->mds = mds;
 291         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 292         for (i = 0; i < frag->ndist; i++)
 293                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 294         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 295              ceph_vinop(inode), frag->frag, frag->ndist);
 296
 297 out:
 298         mutex_unlock(&ci->i_fragtree_mutex);
 299         return err;
 300 }
 301
 302 static int frag_tree_split_cmp(const void *l, const void *r)
 303 {
 304         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 305         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 306         return ceph_frag_compare(le32_to_cpu(ls->frag),
 307                                  le32_to_cpu(rs->frag));
 308 }
 309
 310 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 311 {
 312         if (!frag)
 313                 return f == ceph_frag_make(0, 0);
 314         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 315                 return false;
 316         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 317 }
 318
 319 static int ceph_fill_fragtree(struct inode *inode,
 320                               struct ceph_frag_tree_head *fragtree,
 321                               struct ceph_mds_reply_dirfrag *dirinfo)
 322 {
 323         struct ceph_inode_info *ci = ceph_inode(inode);
 324         struct ceph_inode_frag *frag, *prev_frag = NULL;
 325         struct rb_node *rb_node;
 326         unsigned i, split_by, nsplits;
 327         u32 id;
 328         bool update = false;
 329
 330         mutex_lock(&ci->i_fragtree_mutex);
 331         nsplits = le32_to_cpu(fragtree->nsplits);
 332         if (nsplits != ci->i_fragtree_nsplits) {
 333                 update = true;
 334         } else if (nsplits) {
 335                 i = prandom_u32() % nsplits;
 336                 id = le32_to_cpu(fragtree->splits[i].frag);
 337                 if (!__ceph_find_frag(ci, id))
 338                         update = true;
 339         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 340                 rb_node = rb_first(&ci->i_fragtree);
 341                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 342                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 343                         update = true;
 344         }
 345         if (!update && dirinfo) {
 346                 id = le32_to_cpu(dirinfo->frag);
 347                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 348                         update = true;
 349         }
 350         if (!update)
 351                 goto out_unlock;
 352
 353         if (nsplits > 1) {
 354                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 355                      frag_tree_split_cmp, NULL);
 356         }
 357
 358         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 359         rb_node = rb_first(&ci->i_fragtree);
 360         for (i = 0; i < nsplits; i++) {
 361                 id = le32_to_cpu(fragtree->splits[i].frag);
 362                 split_by = le32_to_cpu(fragtree->splits[i].by);
 363                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 364                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 365                                "frag %x split by %d\n", ceph_vinop(inode),
 366                                i, nsplits, id, split_by);
 367                         continue;
 368                 }
 369                 frag = NULL;
 370                 while (rb_node) {
 371                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 372                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 373                                 if (frag->frag != id)
 374                                         frag = NULL;
 375                                 else
 376                                         rb_node = rb_next(rb_node);
 377                                 break;
 378                         }
 379                         rb_node = rb_next(rb_node);
 380                         /* delete stale split/leaf node */
 381                         if (frag->split_by > 0 ||
 382                             !is_frag_child(frag->frag, prev_frag)) {
 383                                 rb_erase(&frag->node, &ci->i_fragtree);
 384                                 if (frag->split_by > 0)
 385                                         ci->i_fragtree_nsplits--;
 386                                 kfree(frag);
 387                         }
 388                         frag = NULL;
 389                 }
 390                 if (!frag) {
 391                         frag = __get_or_create_frag(ci, id);
 392                         if (IS_ERR(frag))
 393                                 continue;
 394                 }
 395                 if (frag->split_by == 0)
 396                         ci->i_fragtree_nsplits++;
 397                 frag->split_by = split_by;
 398                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 399                 prev_frag = frag;
 400         }
 401         while (rb_node) {
 402                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 403                 rb_node = rb_next(rb_node);
 404                 /* delete stale split/leaf node */
 405                 if (frag->split_by > 0 ||
 406                     !is_frag_child(frag->frag, prev_frag)) {
 407                         rb_erase(&frag->node, &ci->i_fragtree);
 408                         if (frag->split_by > 0)
 409                                 ci->i_fragtree_nsplits--;
 410                         kfree(frag);
 411                 }
 412         }
 413 out_unlock:
 414         mutex_unlock(&ci->i_fragtree_mutex);
 415         return 0;
 416 }
 417
 418 /*
 419  * initialize a newly allocated inode.
 420  */
 421 struct inode *ceph_alloc_inode(struct super_block *sb)
 422 {
 423         struct ceph_inode_info *ci;
 424         int i;
 425
 426         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 427         if (!ci)
 428                 return NULL;
 429
 430         dout("alloc_inode %p\n", &ci->vfs_inode);
 431
 432         spin_lock_init(&ci->i_ceph_lock);
 433
 434         ci->i_version = 0;
 435         ci->i_inline_version = 0;
 436         ci->i_time_warp_seq = 0;
 437         ci->i_ceph_flags = 0;
 438         atomic64_set(&ci->i_ordered_count, 1);
 439         atomic64_set(&ci->i_release_count, 1);
 440         atomic64_set(&ci->i_complete_seq[0], 0);
 441         atomic64_set(&ci->i_complete_seq[1], 0);
 442         ci->i_symlink = NULL;
 443
 444         ci->i_max_bytes = 0;
 445         ci->i_max_files = 0;
 446
 447         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 448         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 449
 450         ci->i_fragtree = RB_ROOT;
 451         mutex_init(&ci->i_fragtree_mutex);
 452
 453         ci->i_xattrs.blob = NULL;
 454         ci->i_xattrs.prealloc_blob = NULL;
 455         ci->i_xattrs.dirty = false;
 456         ci->i_xattrs.index = RB_ROOT;
 457         ci->i_xattrs.count = 0;
 458         ci->i_xattrs.names_size = 0;
 459         ci->i_xattrs.vals_size = 0;
 460         ci->i_xattrs.version = 0;
 461         ci->i_xattrs.index_version = 0;
 462
 463         ci->i_caps = RB_ROOT;
 464         ci->i_auth_cap = NULL;
 465         ci->i_dirty_caps = 0;
 466         ci->i_flushing_caps = 0;
 467         INIT_LIST_HEAD(&ci->i_dirty_item);
 468         INIT_LIST_HEAD(&ci->i_flushing_item);
 469         ci->i_prealloc_cap_flush = NULL;
 470         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 471         init_waitqueue_head(&ci->i_cap_wq);
 472         ci->i_hold_caps_min = 0;
 473         ci->i_hold_caps_max = 0;
 474         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 475         INIT_LIST_HEAD(&ci->i_cap_snaps);
 476         ci->i_head_snapc = NULL;
 477         ci->i_snap_caps = 0;
 478
 479         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 480                 ci->i_nr_by_mode[i] = 0;
 481
 482         mutex_init(&ci->i_truncate_mutex);
 483         ci->i_truncate_seq = 0;
 484         ci->i_truncate_size = 0;
 485         ci->i_truncate_pending = 0;
 486
 487         ci->i_max_size = 0;
 488         ci->i_reported_size = 0;
 489         ci->i_wanted_max_size = 0;
 490         ci->i_requested_max_size = 0;
 491
 492         ci->i_pin_ref = 0;
 493         ci->i_rd_ref = 0;
 494         ci->i_rdcache_ref = 0;
 495         ci->i_wr_ref = 0;
 496         ci->i_wb_ref = 0;
 497         ci->i_wrbuffer_ref = 0;
 498         ci->i_wrbuffer_ref_head = 0;
 499         atomic_set(&ci->i_filelock_ref, 0);
 500         atomic_set(&ci->i_shared_gen, 0);
 501         ci->i_rdcache_gen = 0;
 502         ci->i_rdcache_revoking = 0;
 503
 504         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 505         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 506         spin_lock_init(&ci->i_unsafe_lock);
 507
 508         ci->i_snap_realm = NULL;
 509         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 510         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 511
 512         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 513         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 514
 515         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 516
 517         ceph_fscache_inode_init(ci);
 518
 519         return &ci->vfs_inode;
 520 }
 521
 522 static void ceph_i_callback(struct rcu_head *head)
 523 {
 524         struct inode *inode = container_of(head, struct inode, i_rcu);
 525         struct ceph_inode_info *ci = ceph_inode(inode);
 526
 527         kmem_cache_free(ceph_inode_cachep, ci);
 528 }
 529
 530 void ceph_destroy_inode(struct inode *inode)
 531 {
 532         struct ceph_inode_info *ci = ceph_inode(inode);
 533         struct ceph_inode_frag *frag;
 534         struct rb_node *n;
 535
 536         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 537
 538         ceph_fscache_unregister_inode_cookie(ci);
 539
 540         ceph_queue_caps_release(inode);
 541
 542         if (__ceph_has_any_quota(ci))
 543                 ceph_adjust_quota_realms_count(inode, false);
 544
 545         /*
 546          * we may still have a snap_realm reference if there are stray
 547          * caps in i_snap_caps.
 548          */
 549         if (ci->i_snap_realm) {
 550                 struct ceph_mds_client *mdsc =
 551                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 552                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 553
 554                 dout(" dropping residual ref to snap realm %p\n", realm);
 555                 spin_lock(&realm->inodes_with_caps_lock);
 556                 list_del_init(&ci->i_snap_realm_item);
 557                 ci->i_snap_realm = NULL;
 558                 if (realm->ino == ci->i_vino.ino)
 559                         realm->inode = NULL;
 560                 spin_unlock(&realm->inodes_with_caps_lock);
 561                 ceph_put_snap_realm(mdsc, realm);
 562         }
 563
 564         kfree(ci->i_symlink);
 565         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 566                 frag = rb_entry(n, struct ceph_inode_frag, node);
 567                 rb_erase(n, &ci->i_fragtree);
 568                 kfree(frag);
 569         }
 570         ci->i_fragtree_nsplits = 0;
 571
 572         __ceph_destroy_xattrs(ci);
 573         if (ci->i_xattrs.blob)
 574                 ceph_buffer_put(ci->i_xattrs.blob);
 575         if (ci->i_xattrs.prealloc_blob)
 576                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 577
 578         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 579
 580         call_rcu(&inode->i_rcu, ceph_i_callback);
 581 }
 582
 583 int ceph_drop_inode(struct inode *inode)
 584 {
 585         /*
 586          * Positve dentry and corresponding inode are always accompanied
 587          * in MDS reply. So no need to keep inode in the cache after
 588          * dropping all its aliases.
 589          */
 590         return 1;
 591 }
 592
 593 static inline blkcnt_t calc_inode_blocks(u64 size)
 594 {
 595         return (size + (1<<9) - 1) >> 9;
 596 }
 597
 598 /*
 599  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 600  * careful because either the client or MDS may have more up to date
 601  * info, depending on which capabilities are held, and whether
 602  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 603  * and size are monotonically increasing, except when utimes() or
 604  * truncate() increments the corresponding _seq values.)
 605  */
 606 int ceph_fill_file_size(struct inode *inode, int issued,
 607                         u32 truncate_seq, u64 truncate_size, u64 size)
 608 {
 609         struct ceph_inode_info *ci = ceph_inode(inode);
 610         int queue_trunc = 0;
 611
 612         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 613             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 614                 dout("size %lld -> %llu\n", inode->i_size, size);
 615                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 616                         pr_err("fill_file_size non-zero size for directory\n");
 617                         size = 0;
 618                 }
 619                 i_size_write(inode, size);
 620                 inode->i_blocks = calc_inode_blocks(size);
 621                 ci->i_reported_size = size;
 622                 if (truncate_seq != ci->i_truncate_seq) {
 623                         dout("truncate_seq %u -> %u\n",
 624                              ci->i_truncate_seq, truncate_seq);
 625                         ci->i_truncate_seq = truncate_seq;
 626
 627                         /* the MDS should have revoked these caps */
 628                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 629                                                CEPH_CAP_FILE_RD |
 630                                                CEPH_CAP_FILE_WR |
 631                                                CEPH_CAP_FILE_LAZYIO));
 632                         /*
 633                          * If we hold relevant caps, or in the case where we're
 634                          * not the only client referencing this file and we
 635                          * don't hold those caps, then we need to check whether
 636                          * the file is either opened or mmaped
 637                          */
 638                         if ((issued & (CEPH_CAP_FILE_CACHE|
 639                                        CEPH_CAP_FILE_BUFFER)) ||
 640                             mapping_mapped(inode->i_mapping) ||
 641                             __ceph_caps_file_wanted(ci)) {
 642                                 ci->i_truncate_pending++;
 643                                 queue_trunc = 1;
 644                         }
 645                 }
 646         }
 647         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 648             ci->i_truncate_size != truncate_size) {
 649                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 650                      truncate_size);
 651                 ci->i_truncate_size = truncate_size;
 652         }
 653
 654         if (queue_trunc)
 655                 ceph_fscache_invalidate(inode);
 656
 657         return queue_trunc;
 658 }
 659
 660 void ceph_fill_file_time(struct inode *inode, int issued,
 661                          u64 time_warp_seq, struct timespec64 *ctime,
 662                          struct timespec64 *mtime, struct timespec64 *atime)
 663 {
 664         struct ceph_inode_info *ci = ceph_inode(inode);
 665         int warn = 0;
 666
 667         if (issued & (CEPH_CAP_FILE_EXCL|
 668                       CEPH_CAP_FILE_WR|
 669                       CEPH_CAP_FILE_BUFFER|
 670                       CEPH_CAP_AUTH_EXCL|
 671                       CEPH_CAP_XATTR_EXCL)) {
 672                 if (ci->i_version == 0 ||
 673                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 674                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 675                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 676                              ctime->tv_sec, ctime->tv_nsec);
 677                         inode->i_ctime = *ctime;
 678                 }
 679                 if (ci->i_version == 0 ||
 680                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 681                         /* the MDS did a utimes() */
 682                         dout("mtime %lld.%09ld -> %lld.%09ld "
 683                              "tw %d -> %d\n",
 684                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 685                              mtime->tv_sec, mtime->tv_nsec,
 686                              ci->i_time_warp_seq, (int)time_warp_seq);
 687
 688                         inode->i_mtime = *mtime;
 689                         inode->i_atime = *atime;
 690                         ci->i_time_warp_seq = time_warp_seq;
 691                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 692                         /* nobody did utimes(); take the max */
 693                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 694                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 695                                      inode->i_mtime.tv_sec,
 696                                      inode->i_mtime.tv_nsec,
 697                                      mtime->tv_sec, mtime->tv_nsec);
 698                                 inode->i_mtime = *mtime;
 699                         }
 700                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 701                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 702                                      inode->i_atime.tv_sec,
 703                                      inode->i_atime.tv_nsec,
 704                                      atime->tv_sec, atime->tv_nsec);
 705                                 inode->i_atime = *atime;
 706                         }
 707                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 708                         /* we did a utimes(); ignore mds values */
 709                 } else {
 710                         warn = 1;
 711                 }
 712         } else {
 713                 /* we have no write|excl caps; whatever the MDS says is true */
 714                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 715                         inode->i_ctime = *ctime;
 716                         inode->i_mtime = *mtime;
 717                         inode->i_atime = *atime;
 718                         ci->i_time_warp_seq = time_warp_seq;
 719                 } else {
 720                         warn = 1;
 721                 }
 722         }
 723         if (warn) /* time_warp_seq shouldn't go backwards */
 724                 dout("%p mds time_warp_seq %llu < %u\n",
 725                      inode, time_warp_seq, ci->i_time_warp_seq);
 726 }
 727
 728 /*
 729  * Populate an inode based on info from mds.  May be called on new or
 730  * existing inodes.
 731  */
 732 static int fill_inode(struct inode *inode, struct page *locked_page,
 733                       struct ceph_mds_reply_info_in *iinfo,
 734                       struct ceph_mds_reply_dirfrag *dirinfo,
 735                       struct ceph_mds_session *session,
 736                       unsigned long ttl_from, int cap_fmode,
 737                       struct ceph_cap_reservation *caps_reservation)
 738 {
 739         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
 740         struct ceph_mds_reply_inode *info = iinfo->in;
 741         struct ceph_inode_info *ci = ceph_inode(inode);
 742         int issued, new_issued, info_caps;
 743         struct timespec64 mtime, atime, ctime;
 744         struct ceph_buffer *xattr_blob = NULL;
 745         struct ceph_string *pool_ns = NULL;
 746         struct ceph_cap *new_cap = NULL;
 747         int err = 0;
 748         bool wake = false;
 749         bool queue_trunc = false;
 750         bool new_version = false;
 751         bool fill_inline = false;
 752
 753         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 754              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 755              ci->i_version);
 756
 757         info_caps = le32_to_cpu(info->cap.caps);
 758
 759         /* prealloc new cap struct */
 760         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
 761                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 762
 763         /*
 764          * prealloc xattr data, if it looks like we'll need it.  only
 765          * if len > 4 (meaning there are actually xattrs; the first 4
 766          * bytes are the xattr count).
 767          */
 768         if (iinfo->xattr_len > 4) {
 769                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 770                 if (!xattr_blob)
 771                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 772                                iinfo->xattr_len);
 773         }
 774
 775         if (iinfo->pool_ns_len > 0)
 776                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 777                                                      iinfo->pool_ns_len);
 778
 779         spin_lock(&ci->i_ceph_lock);
 780
 781         /*
 782          * provided version will be odd if inode value is projected,
 783          * even if stable.  skip the update if we have newer stable
 784          * info (ours>=theirs, e.g. due to racing mds replies), unless
 785          * we are getting projected (unstable) info (in which case the
 786          * version is odd, and we want ours>theirs).
 787          *   us   them
 788          *   2    2     skip
 789          *   3    2     skip
 790          *   3    3     update
 791          */
 792         if (ci->i_version == 0 ||
 793             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 794              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 795                 new_version = true;
 796
 797         __ceph_caps_issued(ci, &issued);
 798         issued |= __ceph_caps_dirty(ci);
 799         new_issued = ~issued & info_caps;
 800
 801         /* update inode */
 802         inode->i_rdev = le32_to_cpu(info->rdev);
 803         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 804
 805         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 806
 807         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 808             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 809                 inode->i_mode = le32_to_cpu(info->mode);
 810                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 811                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 812                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 813                      from_kuid(&init_user_ns, inode->i_uid),
 814                      from_kgid(&init_user_ns, inode->i_gid));
 815         }
 816
 817         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 818             (issued & CEPH_CAP_LINK_EXCL) == 0)
 819                 set_nlink(inode, le32_to_cpu(info->nlink));
 820
 821         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 822                 /* be careful with mtime, atime, size */
 823                 ceph_decode_timespec64(&atime, &info->atime);
 824                 ceph_decode_timespec64(&mtime, &info->mtime);
 825                 ceph_decode_timespec64(&ctime, &info->ctime);
 826                 ceph_fill_file_time(inode, issued,
 827                                 le32_to_cpu(info->time_warp_seq),
 828                                 &ctime, &mtime, &atime);
 829         }
 830
 831         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 832                 ci->i_files = le64_to_cpu(info->files);
 833                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 834         }
 835
 836         if (new_version ||
 837             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 838                 s64 old_pool = ci->i_layout.pool_id;
 839                 struct ceph_string *old_ns;
 840
 841                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 842                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 843                                         lockdep_is_held(&ci->i_ceph_lock));
 844                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 845
 846                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 847                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 848
 849                 pool_ns = old_ns;
 850
 851                 queue_trunc = ceph_fill_file_size(inode, issued,
 852                                         le32_to_cpu(info->truncate_seq),
 853                                         le64_to_cpu(info->truncate_size),
 854                                         le64_to_cpu(info->size));
 855                 /* only update max_size on auth cap */
 856                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 857                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 858                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 859                                         le64_to_cpu(info->max_size));
 860                         ci->i_max_size = le64_to_cpu(info->max_size);
 861                 }
 862         }
 863
 864         /* layout and rstat are not tracked by capability, update them if
 865          * the inode info is from auth mds */
 866         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 867                 if (S_ISDIR(inode->i_mode)) {
 868                         ci->i_dir_layout = iinfo->dir_layout;
 869                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 870                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 871                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 872                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 873                 }
 874         }
 875
 876         /* xattrs */
 877         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 878         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 879             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 880                 if (ci->i_xattrs.blob)
 881                         ceph_buffer_put(ci->i_xattrs.blob);
 882                 ci->i_xattrs.blob = xattr_blob;
 883                 if (xattr_blob)
 884                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 885                                iinfo->xattr_data, iinfo->xattr_len);
 886                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 887                 ceph_forget_all_cached_acls(inode);
 888                 xattr_blob = NULL;
 889         }
 890
 891         /* finally update i_version */
 892         if (le64_to_cpu(info->version) > ci->i_version)
 893                 ci->i_version = le64_to_cpu(info->version);
 894
 895         inode->i_mapping->a_ops = &ceph_aops;
 896
 897         switch (inode->i_mode & S_IFMT) {
 898         case S_IFIFO:
 899         case S_IFBLK:
 900         case S_IFCHR:
 901         case S_IFSOCK:
 902                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 903                 inode->i_op = &ceph_file_iops;
 904                 break;
 905         case S_IFREG:
 906                 inode->i_op = &ceph_file_iops;
 907                 inode->i_fop = &ceph_file_fops;
 908                 break;
 909         case S_IFLNK:
 910                 inode->i_op = &ceph_symlink_iops;
 911                 if (!ci->i_symlink) {
 912                         u32 symlen = iinfo->symlink_len;
 913                         char *sym;
 914
 915                         spin_unlock(&ci->i_ceph_lock);
 916
 917                         if (symlen != i_size_read(inode)) {
 918                                 pr_err("fill_inode %llx.%llx BAD symlink "
 919                                         "size %lld\n", ceph_vinop(inode),
 920                                         i_size_read(inode));
 921                                 i_size_write(inode, symlen);
 922                                 inode->i_blocks = calc_inode_blocks(symlen);
 923                         }
 924
 925                         err = -ENOMEM;
 926                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 927                         if (!sym)
 928                                 goto out;
 929
 930                         spin_lock(&ci->i_ceph_lock);
 931                         if (!ci->i_symlink)
 932                                 ci->i_symlink = sym;
 933                         else
 934                                 kfree(sym); /* lost a race */
 935                 }
 936                 inode->i_link = ci->i_symlink;
 937                 break;
 938         case S_IFDIR:
 939                 inode->i_op = &ceph_dir_iops;
 940                 inode->i_fop = &ceph_dir_fops;
 941                 break;
 942         default:
 943                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 944                        ceph_vinop(inode), inode->i_mode);
 945         }
 946
 947         /* were we issued a capability? */
 948         if (info_caps) {
 949                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 950                         ceph_add_cap(inode, session,
 951                                      le64_to_cpu(info->cap.cap_id),
 952                                      cap_fmode, info_caps,
 953                                      le32_to_cpu(info->cap.wanted),
 954                                      le32_to_cpu(info->cap.seq),
 955                                      le32_to_cpu(info->cap.mseq),
 956                                      le64_to_cpu(info->cap.realm),
 957                                      info->cap.flags, &new_cap);
 958
 959                         /* set dir completion flag? */
 960                         if (S_ISDIR(inode->i_mode) &&
 961                             ci->i_files == 0 && ci->i_subdirs == 0 &&
 962                             (info_caps & CEPH_CAP_FILE_SHARED) &&
 963                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 964                             !__ceph_dir_is_complete(ci)) {
 965                                 dout(" marking %p complete (empty)\n", inode);
 966                                 i_size_write(inode, 0);
 967                                 __ceph_dir_set_complete(ci,
 968                                         atomic64_read(&ci->i_release_count),
 969                                         atomic64_read(&ci->i_ordered_count));
 970                         }
 971
 972                         wake = true;
 973                 } else {
 974                         dout(" %p got snap_caps %s\n", inode,
 975                              ceph_cap_string(info_caps));
 976                         ci->i_snap_caps |= info_caps;
 977                         if (cap_fmode >= 0)
 978                                 __ceph_get_fmode(ci, cap_fmode);
 979                 }
 980         } else if (cap_fmode >= 0) {
 981                 pr_warn("mds issued no caps on %llx.%llx\n",
 982                            ceph_vinop(inode));
 983                 __ceph_get_fmode(ci, cap_fmode);
 984         }
 985
 986         if (iinfo->inline_version > 0 &&
 987             iinfo->inline_version >= ci->i_inline_version) {
 988                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
 989                 ci->i_inline_version = iinfo->inline_version;
 990                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
 991                     (locked_page || (info_caps & cache_caps)))
 992                         fill_inline = true;
 993         }
 994
 995         spin_unlock(&ci->i_ceph_lock);
 996
 997         if (fill_inline)
 998                 ceph_fill_inline_data(inode, locked_page,
 999                                       iinfo->inline_data, iinfo->inline_len);
1000
1001         if (wake)
1002                 wake_up_all(&ci->i_cap_wq);
1003
1004         /* queue truncate if we saw i_size decrease */
1005         if (queue_trunc)
1006                 ceph_queue_vmtruncate(inode);
1007
1008         /* populate frag tree */
1009         if (S_ISDIR(inode->i_mode))
1010                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1011
1012         /* update delegation info? */
1013         if (dirinfo)
1014                 ceph_fill_dirfrag(inode, dirinfo);
1015
1016         err = 0;
1017 out:
1018         if (new_cap)
1019                 ceph_put_cap(mdsc, new_cap);
1020         if (xattr_blob)
1021                 ceph_buffer_put(xattr_blob);
1022         ceph_put_string(pool_ns);
1023         return err;
1024 }
1025
1026 /*
1027  * caller should hold session s_mutex.
1028  */
1029 static void update_dentry_lease(struct dentry *dentry,
1030                                 struct ceph_mds_reply_lease *lease,
1031                                 struct ceph_mds_session *session,
1032                                 unsigned long from_time,
1033                                 struct ceph_vino *tgt_vino,
1034                                 struct ceph_vino *dir_vino)
1035 {
1036         struct ceph_dentry_info *di = ceph_dentry(dentry);
1037         long unsigned duration = le32_to_cpu(lease->duration_ms);
1038         long unsigned ttl = from_time + (duration * HZ) / 1000;
1039         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1040         struct inode *dir;
1041         struct ceph_mds_session *old_lease_session = NULL;
1042
1043         /*
1044          * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1045          * we expect a negative dentry.
1046          */
1047         if (!tgt_vino && d_really_is_positive(dentry))
1048                 return;
1049
1050         if (tgt_vino && (d_really_is_negative(dentry) ||
1051                         !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1052                 return;
1053
1054         spin_lock(&dentry->d_lock);
1055         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1056              dentry, duration, ttl);
1057
1058         dir = d_inode(dentry->d_parent);
1059
1060         /* make sure parent matches dir_vino */
1061         if (!ceph_ino_compare(dir, dir_vino))
1062                 goto out_unlock;
1063
1064         /* only track leases on regular dentries */
1065         if (ceph_snap(dir) != CEPH_NOSNAP)
1066                 goto out_unlock;
1067
1068         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1069
1070         if (duration == 0)
1071                 goto out_unlock;
1072
1073         if (di->lease_gen == session->s_cap_gen &&
1074             time_before(ttl, di->time))
1075                 goto out_unlock;  /* we already have a newer lease. */
1076
1077         if (di->lease_session && di->lease_session != session) {
1078                 old_lease_session = di->lease_session;
1079                 di->lease_session = NULL;
1080         }
1081
1082         ceph_dentry_lru_touch(dentry);
1083
1084         if (!di->lease_session)
1085                 di->lease_session = ceph_get_mds_session(session);
1086         di->lease_gen = session->s_cap_gen;
1087         di->lease_seq = le32_to_cpu(lease->seq);
1088         di->lease_renew_after = half_ttl;
1089         di->lease_renew_from = 0;
1090         di->time = ttl;
1091 out_unlock:
1092         spin_unlock(&dentry->d_lock);
1093         if (old_lease_session)
1094                 ceph_put_mds_session(old_lease_session);
1095 }
1096
1097 /*
1098  * splice a dentry to an inode.
1099  * caller must hold directory i_mutex for this to be safe.
1100  */
1101 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1102 {
1103         struct dentry *realdn;
1104
1105         BUG_ON(d_inode(dn));
1106
1107         if (S_ISDIR(in->i_mode)) {
1108                 /* If inode is directory, d_splice_alias() below will remove
1109                  * 'realdn' from its origin parent. We need to ensure that
1110                  * origin parent's readdir cache will not reference 'realdn'
1111                  */
1112                 realdn = d_find_any_alias(in);
1113                 if (realdn) {
1114                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1115                         spin_lock(&realdn->d_lock);
1116
1117                         realdn->d_op->d_prune(realdn);
1118
1119                         di->time = jiffies;
1120                         di->lease_shared_gen = 0;
1121                         di->offset = 0;
1122
1123                         spin_unlock(&realdn->d_lock);
1124                         dput(realdn);
1125                 }
1126         }
1127
1128         /* dn must be unhashed */
1129         if (!d_unhashed(dn))
1130                 d_drop(dn);
1131         realdn = d_splice_alias(in, dn);
1132         if (IS_ERR(realdn)) {
1133                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1134                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1135                 dput(dn);
1136                 dn = realdn; /* note realdn contains the error */
1137                 goto out;
1138         } else if (realdn) {
1139                 dout("dn %p (%d) spliced with %p (%d) "
1140                      "inode %p ino %llx.%llx\n",
1141                      dn, d_count(dn),
1142                      realdn, d_count(realdn),
1143                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1144                 dput(dn);
1145                 dn = realdn;
1146         } else {
1147                 BUG_ON(!ceph_dentry(dn));
1148                 dout("dn %p attached to %p ino %llx.%llx\n",
1149                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1150         }
1151 out:
1152         return dn;
1153 }
1154
1155 /*
1156  * Incorporate results into the local cache.  This is either just
1157  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1158  * after a lookup).
1159  *
1160  * A reply may contain
1161  *         a directory inode along with a dentry.
1162  *  and/or a target inode
1163  *
1164  * Called with snap_rwsem (read).
1165  */
1166 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1167 {
1168         struct ceph_mds_session *session = req->r_session;
1169         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1170         struct inode *in = NULL;
1171         struct ceph_vino tvino, dvino;
1172         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1173         int err = 0;
1174
1175         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1176              rinfo->head->is_dentry, rinfo->head->is_target);
1177
1178         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1179                 dout("fill_trace reply is empty!\n");
1180                 if (rinfo->head->result == 0 && req->r_parent)
1181                         ceph_invalidate_dir_request(req);
1182                 return 0;
1183         }
1184
1185         if (rinfo->head->is_dentry) {
1186                 struct inode *dir = req->r_parent;
1187
1188                 if (dir) {
1189                         err = fill_inode(dir, NULL,
1190                                          &rinfo->diri, rinfo->dirfrag,
1191                                          session, req->r_request_started, -1,
1192                                          &req->r_caps_reservation);
1193                         if (err < 0)
1194                                 goto done;
1195                 } else {
1196                         WARN_ON_ONCE(1);
1197                 }
1198
1199                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1200                         struct qstr dname;
1201                         struct dentry *dn, *parent;
1202
1203                         BUG_ON(!rinfo->head->is_target);
1204                         BUG_ON(req->r_dentry);
1205
1206                         parent = d_find_any_alias(dir);
1207                         BUG_ON(!parent);
1208
1209                         dname.name = rinfo->dname;
1210                         dname.len = rinfo->dname_len;
1211                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1212                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1213                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1214 retry_lookup:
1215                         dn = d_lookup(parent, &dname);
1216                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1217                              parent, dname.len, dname.name, dn);
1218
1219                         if (!dn) {
1220                                 dn = d_alloc(parent, &dname);
1221                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1222                                      dname.len, dname.name, dn);
1223                                 if (!dn) {
1224                                         dput(parent);
1225                                         err = -ENOMEM;
1226                                         goto done;
1227                                 }
1228                                 err = 0;
1229                         } else if (d_really_is_positive(dn) &&
1230                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1231                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1232                                 dout(" dn %p points to wrong inode %p\n",
1233                                      dn, d_inode(dn));
1234                                 ceph_dir_clear_ordered(dir);
1235                                 d_delete(dn);
1236                                 dput(dn);
1237                                 goto retry_lookup;
1238                         }
1239
1240                         req->r_dentry = dn;
1241                         dput(parent);
1242                 }
1243         }
1244
1245         if (rinfo->head->is_target) {
1246                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1247                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1248
1249                 in = ceph_get_inode(sb, tvino);
1250                 if (IS_ERR(in)) {
1251                         err = PTR_ERR(in);
1252                         goto done;
1253                 }
1254                 req->r_target_inode = in;
1255
1256                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1257                                 session, req->r_request_started,
1258                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1259                                 rinfo->head->result == 0) ?  req->r_fmode : -1,
1260                                 &req->r_caps_reservation);
1261                 if (err < 0) {
1262                         pr_err("fill_inode badness %p %llx.%llx\n",
1263                                 in, ceph_vinop(in));
1264                         goto done;
1265                 }
1266         }
1267
1268         /*
1269          * ignore null lease/binding on snapdir ENOENT, or else we
1270          * will have trouble splicing in the virtual snapdir later
1271          */
1272         if (rinfo->head->is_dentry &&
1273             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1274             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1275             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1276                                                fsc->mount_options->snapdir_name,
1277                                                req->r_dentry->d_name.len))) {
1278                 /*
1279                  * lookup link rename   : null -> possibly existing inode
1280                  * mknod symlink mkdir  : null -> new inode
1281                  * unlink               : linked -> null
1282                  */
1283                 struct inode *dir = req->r_parent;
1284                 struct dentry *dn = req->r_dentry;
1285                 bool have_dir_cap, have_lease;
1286
1287                 BUG_ON(!dn);
1288                 BUG_ON(!dir);
1289                 BUG_ON(d_inode(dn->d_parent) != dir);
1290
1291                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1292                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1293
1294                 BUG_ON(ceph_ino(dir) != dvino.ino);
1295                 BUG_ON(ceph_snap(dir) != dvino.snap);
1296
1297                 /* do we have a lease on the whole dir? */
1298                 have_dir_cap =
1299                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1300                          CEPH_CAP_FILE_SHARED);
1301
1302                 /* do we have a dn lease? */
1303                 have_lease = have_dir_cap ||
1304                         le32_to_cpu(rinfo->dlease->duration_ms);
1305                 if (!have_lease)
1306                         dout("fill_trace  no dentry lease or dir cap\n");
1307
1308                 /* rename? */
1309                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1310                         struct inode *olddir = req->r_old_dentry_dir;
1311                         BUG_ON(!olddir);
1312
1313                         dout(" src %p '%pd' dst %p '%pd'\n",
1314                              req->r_old_dentry,
1315                              req->r_old_dentry,
1316                              dn, dn);
1317                         dout("fill_trace doing d_move %p -> %p\n",
1318                              req->r_old_dentry, dn);
1319
1320                         /* d_move screws up sibling dentries' offsets */
1321                         ceph_dir_clear_ordered(dir);
1322                         ceph_dir_clear_ordered(olddir);
1323
1324                         d_move(req->r_old_dentry, dn);
1325                         dout(" src %p '%pd' dst %p '%pd'\n",
1326                              req->r_old_dentry,
1327                              req->r_old_dentry,
1328                              dn, dn);
1329
1330                         /* ensure target dentry is invalidated, despite
1331                            rehashing bug in vfs_rename_dir */
1332                         ceph_invalidate_dentry_lease(dn);
1333
1334                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1335                              ceph_dentry(req->r_old_dentry)->offset);
1336
1337                         dn = req->r_old_dentry;  /* use old_dentry */
1338                 }
1339
1340                 /* null dentry? */
1341                 if (!rinfo->head->is_target) {
1342                         dout("fill_trace null dentry\n");
1343                         if (d_really_is_positive(dn)) {
1344                                 dout("d_delete %p\n", dn);
1345                                 ceph_dir_clear_ordered(dir);
1346                                 d_delete(dn);
1347                         } else if (have_lease) {
1348                                 if (d_unhashed(dn))
1349                                         d_add(dn, NULL);
1350                                 update_dentry_lease(dn, rinfo->dlease,
1351                                                     session,
1352                                                     req->r_request_started,
1353                                                     NULL, &dvino);
1354                         }
1355                         goto done;
1356                 }
1357
1358                 /* attach proper inode */
1359                 if (d_really_is_negative(dn)) {
1360                         ceph_dir_clear_ordered(dir);
1361                         ihold(in);
1362                         dn = splice_dentry(dn, in);
1363                         if (IS_ERR(dn)) {
1364                                 err = PTR_ERR(dn);
1365                                 goto done;
1366                         }
1367                         req->r_dentry = dn;  /* may have spliced */
1368                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1369                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1370                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1371                              ceph_vinop(in));
1372                         d_invalidate(dn);
1373                         have_lease = false;
1374                 }
1375
1376                 if (have_lease) {
1377                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1378                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1379                         update_dentry_lease(dn, rinfo->dlease, session,
1380                                             req->r_request_started,
1381                                             &tvino, &dvino);
1382                 }
1383                 dout(" final dn %p\n", dn);
1384         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1385                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1386                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1387                 struct dentry *dn = req->r_dentry;
1388                 struct inode *dir = req->r_parent;
1389
1390                 /* fill out a snapdir LOOKUPSNAP dentry */
1391                 BUG_ON(!dn);
1392                 BUG_ON(!dir);
1393                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1394                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1395                 ceph_dir_clear_ordered(dir);
1396                 ihold(in);
1397                 dn = splice_dentry(dn, in);
1398                 if (IS_ERR(dn)) {
1399                         err = PTR_ERR(dn);
1400                         goto done;
1401                 }
1402                 req->r_dentry = dn;  /* may have spliced */
1403         } else if (rinfo->head->is_dentry) {
1404                 struct ceph_vino *ptvino = NULL;
1405
1406                 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1407                     le32_to_cpu(rinfo->dlease->duration_ms)) {
1408                         dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1409                         dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1410
1411                         if (rinfo->head->is_target) {
1412                                 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1413                                 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1414                                 ptvino = &tvino;
1415                         }
1416
1417                         update_dentry_lease(req->r_dentry, rinfo->dlease,
1418                                 session, req->r_request_started, ptvino,
1419                                 &dvino);
1420                 } else {
1421                         dout("%s: no dentry lease or dir cap\n", __func__);
1422                 }
1423         }
1424 done:
1425         dout("fill_trace done err=%d\n", err);
1426         return err;
1427 }
1428
1429 /*
1430  * Prepopulate our cache with readdir results, leases, etc.
1431  */
1432 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1433                                            struct ceph_mds_session *session)
1434 {
1435         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1436         int i, err = 0;
1437
1438         for (i = 0; i < rinfo->dir_nr; i++) {
1439                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1440                 struct ceph_vino vino;
1441                 struct inode *in;
1442                 int rc;
1443
1444                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1445                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1446
1447                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1448                 if (IS_ERR(in)) {
1449                         err = PTR_ERR(in);
1450                         dout("new_inode badness got %d\n", err);
1451                         continue;
1452                 }
1453                 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1454                                 req->r_request_started, -1,
1455                                 &req->r_caps_reservation);
1456                 if (rc < 0) {
1457                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1458                         err = rc;
1459                 }
1460                 iput(in);
1461         }
1462
1463         return err;
1464 }
1465
1466 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1467 {
1468         if (ctl->page) {
1469                 kunmap(ctl->page);
1470                 put_page(ctl->page);
1471                 ctl->page = NULL;
1472         }
1473 }
1474
1475 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1476                               struct ceph_readdir_cache_control *ctl,
1477                               struct ceph_mds_request *req)
1478 {
1479         struct ceph_inode_info *ci = ceph_inode(dir);
1480         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1481         unsigned idx = ctl->index % nsize;
1482         pgoff_t pgoff = ctl->index / nsize;
1483
1484         if (!ctl->page || pgoff != page_index(ctl->page)) {
1485                 ceph_readdir_cache_release(ctl);
1486                 if (idx == 0)
1487                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1488                 else
1489                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1490                 if (!ctl->page) {
1491                         ctl->index = -1;
1492                         return idx == 0 ? -ENOMEM : 0;
1493                 }
1494                 /* reading/filling the cache are serialized by
1495                  * i_mutex, no need to use page lock */
1496                 unlock_page(ctl->page);
1497                 ctl->dentries = kmap(ctl->page);
1498                 if (idx == 0)
1499                         memset(ctl->dentries, 0, PAGE_SIZE);
1500         }
1501
1502         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1503             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1504                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1505                 ctl->dentries[idx] = dn;
1506                 ctl->index++;
1507         } else {
1508                 dout("disable readdir cache\n");
1509                 ctl->index = -1;
1510         }
1511         return 0;
1512 }
1513
1514 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1515                              struct ceph_mds_session *session)
1516 {
1517         struct dentry *parent = req->r_dentry;
1518         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1519         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1520         struct qstr dname;
1521         struct dentry *dn;
1522         struct inode *in;
1523         int err = 0, skipped = 0, ret, i;
1524         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1525         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1526         u32 last_hash = 0;
1527         u32 fpos_offset;
1528         struct ceph_readdir_cache_control cache_ctl = {};
1529
1530         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1531                 return readdir_prepopulate_inodes_only(req, session);
1532
1533         if (rinfo->hash_order) {
1534                 if (req->r_path2) {
1535                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1536                                                   req->r_path2,
1537                                                   strlen(req->r_path2));
1538                         last_hash = ceph_frag_value(last_hash);
1539                 } else if (rinfo->offset_hash) {
1540                         /* mds understands offset_hash */
1541                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1542                         last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1543                 }
1544         }
1545
1546         if (rinfo->dir_dir &&
1547             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1548                 dout("readdir_prepopulate got new frag %x -> %x\n",
1549                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1550                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1551                 if (!rinfo->hash_order)
1552                         req->r_readdir_offset = 2;
1553         }
1554
1555         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1556                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1557                      rinfo->dir_nr, parent);
1558         } else {
1559                 dout("readdir_prepopulate %d items under dn %p\n",
1560                      rinfo->dir_nr, parent);
1561                 if (rinfo->dir_dir)
1562                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1563
1564                 if (ceph_frag_is_leftmost(frag) &&
1565                     req->r_readdir_offset == 2 &&
1566                     !(rinfo->hash_order && last_hash)) {
1567                         /* note dir version at start of readdir so we can
1568                          * tell if any dentries get dropped */
1569                         req->r_dir_release_cnt =
1570                                 atomic64_read(&ci->i_release_count);
1571                         req->r_dir_ordered_cnt =
1572                                 atomic64_read(&ci->i_ordered_count);
1573                         req->r_readdir_cache_idx = 0;
1574                 }
1575         }
1576
1577         cache_ctl.index = req->r_readdir_cache_idx;
1578         fpos_offset = req->r_readdir_offset;
1579
1580         /* FIXME: release caps/leases if error occurs */
1581         for (i = 0; i < rinfo->dir_nr; i++) {
1582                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1583                 struct ceph_vino tvino, dvino;
1584
1585                 dname.name = rde->name;
1586                 dname.len = rde->name_len;
1587                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1588
1589                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1590                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1591
1592                 if (rinfo->hash_order) {
1593                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1594                                                  rde->name, rde->name_len);
1595                         hash = ceph_frag_value(hash);
1596                         if (hash != last_hash)
1597                                 fpos_offset = 2;
1598                         last_hash = hash;
1599                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1600                 } else {
1601                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1602                 }
1603
1604 retry_lookup:
1605                 dn = d_lookup(parent, &dname);
1606                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1607                      parent, dname.len, dname.name, dn);
1608
1609                 if (!dn) {
1610                         dn = d_alloc(parent, &dname);
1611                         dout("d_alloc %p '%.*s' = %p\n", parent,
1612                              dname.len, dname.name, dn);
1613                         if (!dn) {
1614                                 dout("d_alloc badness\n");
1615                                 err = -ENOMEM;
1616                                 goto out;
1617                         }
1618                 } else if (d_really_is_positive(dn) &&
1619                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1620                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1621                         struct ceph_dentry_info *di = ceph_dentry(dn);
1622                         dout(" dn %p points to wrong inode %p\n",
1623                              dn, d_inode(dn));
1624
1625                         spin_lock(&dn->d_lock);
1626                         if (di->offset > 0 &&
1627                             di->lease_shared_gen ==
1628                             atomic_read(&ci->i_shared_gen)) {
1629                                 __ceph_dir_clear_ordered(ci);
1630                                 di->offset = 0;
1631                         }
1632                         spin_unlock(&dn->d_lock);
1633
1634                         d_delete(dn);
1635                         dput(dn);
1636                         goto retry_lookup;
1637                 }
1638
1639                 /* inode */
1640                 if (d_really_is_positive(dn)) {
1641                         in = d_inode(dn);
1642                 } else {
1643                         in = ceph_get_inode(parent->d_sb, tvino);
1644                         if (IS_ERR(in)) {
1645                                 dout("new_inode badness\n");
1646                                 d_drop(dn);
1647                                 dput(dn);
1648                                 err = PTR_ERR(in);
1649                                 goto out;
1650                         }
1651                 }
1652
1653                 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1654                                  req->r_request_started, -1,
1655                                  &req->r_caps_reservation);
1656                 if (ret < 0) {
1657                         pr_err("fill_inode badness on %p\n", in);
1658                         if (d_really_is_negative(dn))
1659                                 iput(in);
1660                         d_drop(dn);
1661                         err = ret;
1662                         goto next_item;
1663                 }
1664
1665                 if (d_really_is_negative(dn)) {
1666                         struct dentry *realdn;
1667
1668                         if (ceph_security_xattr_deadlock(in)) {
1669                                 dout(" skip splicing dn %p to inode %p"
1670                                      " (security xattr deadlock)\n", dn, in);
1671                                 iput(in);
1672                                 skipped++;
1673                                 goto next_item;
1674                         }
1675
1676                         realdn = splice_dentry(dn, in);
1677                         if (IS_ERR(realdn)) {
1678                                 err = PTR_ERR(realdn);
1679                                 d_drop(dn);
1680                                 dn = NULL;
1681                                 goto next_item;
1682                         }
1683                         dn = realdn;
1684                 }
1685
1686                 ceph_dentry(dn)->offset = rde->offset;
1687
1688                 dvino = ceph_vino(d_inode(parent));
1689                 update_dentry_lease(dn, rde->lease, req->r_session,
1690                                     req->r_request_started, &tvino, &dvino);
1691
1692                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1693                         ret = fill_readdir_cache(d_inode(parent), dn,
1694                                                  &cache_ctl, req);
1695                         if (ret < 0)
1696                                 err = ret;
1697                 }
1698 next_item:
1699                 if (dn)
1700                         dput(dn);
1701         }
1702 out:
1703         if (err == 0 && skipped == 0) {
1704                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1705                 req->r_readdir_cache_idx = cache_ctl.index;
1706         }
1707         ceph_readdir_cache_release(&cache_ctl);
1708         dout("readdir_prepopulate done\n");
1709         return err;
1710 }
1711
1712 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1713 {
1714         struct ceph_inode_info *ci = ceph_inode(inode);
1715         bool ret;
1716
1717         spin_lock(&ci->i_ceph_lock);
1718         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1719         i_size_write(inode, size);
1720         inode->i_blocks = calc_inode_blocks(size);
1721
1722         ret = __ceph_should_report_size(ci);
1723
1724         spin_unlock(&ci->i_ceph_lock);
1725         return ret;
1726 }
1727
1728 /*
1729  * Write back inode data in a worker thread.  (This can't be done
1730  * in the message handler context.)
1731  */
1732 void ceph_queue_writeback(struct inode *inode)
1733 {
1734         ihold(inode);
1735         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1736                        &ceph_inode(inode)->i_wb_work)) {
1737                 dout("ceph_queue_writeback %p\n", inode);
1738         } else {
1739                 dout("ceph_queue_writeback %p failed\n", inode);
1740                 iput(inode);
1741         }
1742 }
1743
1744 static void ceph_writeback_work(struct work_struct *work)
1745 {
1746         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1747                                                   i_wb_work);
1748         struct inode *inode = &ci->vfs_inode;
1749
1750         dout("writeback %p\n", inode);
1751         filemap_fdatawrite(&inode->i_data);
1752         iput(inode);
1753 }
1754
1755 /*
1756  * queue an async invalidation
1757  */
1758 void ceph_queue_invalidate(struct inode *inode)
1759 {
1760         ihold(inode);
1761         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1762                        &ceph_inode(inode)->i_pg_inv_work)) {
1763                 dout("ceph_queue_invalidate %p\n", inode);
1764         } else {
1765                 dout("ceph_queue_invalidate %p failed\n", inode);
1766                 iput(inode);
1767         }
1768 }
1769
1770 /*
1771  * Invalidate inode pages in a worker thread.  (This can't be done
1772  * in the message handler context.)
1773  */
1774 static void ceph_invalidate_work(struct work_struct *work)
1775 {
1776         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1777                                                   i_pg_inv_work);
1778         struct inode *inode = &ci->vfs_inode;
1779         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1780         u32 orig_gen;
1781         int check = 0;
1782
1783         mutex_lock(&ci->i_truncate_mutex);
1784
1785         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1786                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1787                                     inode, ceph_ino(inode));
1788                 mapping_set_error(inode->i_mapping, -EIO);
1789                 truncate_pagecache(inode, 0);
1790                 mutex_unlock(&ci->i_truncate_mutex);
1791                 goto out;
1792         }
1793
1794         spin_lock(&ci->i_ceph_lock);
1795         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1796              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1797         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1798                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1799                         check = 1;
1800                 spin_unlock(&ci->i_ceph_lock);
1801                 mutex_unlock(&ci->i_truncate_mutex);
1802                 goto out;
1803         }
1804         orig_gen = ci->i_rdcache_gen;
1805         spin_unlock(&ci->i_ceph_lock);
1806
1807         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1808                 pr_err("invalidate_pages %p fails\n", inode);
1809         }
1810
1811         spin_lock(&ci->i_ceph_lock);
1812         if (orig_gen == ci->i_rdcache_gen &&
1813             orig_gen == ci->i_rdcache_revoking) {
1814                 dout("invalidate_pages %p gen %d successful\n", inode,
1815                      ci->i_rdcache_gen);
1816                 ci->i_rdcache_revoking--;
1817                 check = 1;
1818         } else {
1819                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1820                      inode, orig_gen, ci->i_rdcache_gen,
1821                      ci->i_rdcache_revoking);
1822                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1823                         check = 1;
1824         }
1825         spin_unlock(&ci->i_ceph_lock);
1826         mutex_unlock(&ci->i_truncate_mutex);
1827 out:
1828         if (check)
1829                 ceph_check_caps(ci, 0, NULL);
1830         iput(inode);
1831 }
1832
1833
1834 /*
1835  * called by trunc_wq;
1836  *
1837  * We also truncate in a separate thread as well.
1838  */
1839 static void ceph_vmtruncate_work(struct work_struct *work)
1840 {
1841         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1842                                                   i_vmtruncate_work);
1843         struct inode *inode = &ci->vfs_inode;
1844
1845         dout("vmtruncate_work %p\n", inode);
1846         __ceph_do_pending_vmtruncate(inode);
1847         iput(inode);
1848 }
1849
1850 /*
1851  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1852  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1853  */
1854 void ceph_queue_vmtruncate(struct inode *inode)
1855 {
1856         struct ceph_inode_info *ci = ceph_inode(inode);
1857
1858         ihold(inode);
1859
1860         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1861                        &ci->i_vmtruncate_work)) {
1862                 dout("ceph_queue_vmtruncate %p\n", inode);
1863         } else {
1864                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1865                      inode, ci->i_truncate_pending);
1866                 iput(inode);
1867         }
1868 }
1869
1870 /*
1871  * Make sure any pending truncation is applied before doing anything
1872  * that may depend on it.
1873  */
1874 void __ceph_do_pending_vmtruncate(struct inode *inode)
1875 {
1876         struct ceph_inode_info *ci = ceph_inode(inode);
1877         u64 to;
1878         int wrbuffer_refs, finish = 0;
1879
1880         mutex_lock(&ci->i_truncate_mutex);
1881 retry:
1882         spin_lock(&ci->i_ceph_lock);
1883         if (ci->i_truncate_pending == 0) {
1884                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1885                 spin_unlock(&ci->i_ceph_lock);
1886                 mutex_unlock(&ci->i_truncate_mutex);
1887                 return;
1888         }
1889
1890         /*
1891          * make sure any dirty snapped pages are flushed before we
1892          * possibly truncate them.. so write AND block!
1893          */
1894         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1895                 spin_unlock(&ci->i_ceph_lock);
1896                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1897                      inode);
1898                 filemap_write_and_wait_range(&inode->i_data, 0,
1899                                              inode->i_sb->s_maxbytes);
1900                 goto retry;
1901         }
1902
1903         /* there should be no reader or writer */
1904         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1905
1906         to = ci->i_truncate_size;
1907         wrbuffer_refs = ci->i_wrbuffer_ref;
1908         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1909              ci->i_truncate_pending, to);
1910         spin_unlock(&ci->i_ceph_lock);
1911
1912         truncate_pagecache(inode, to);
1913
1914         spin_lock(&ci->i_ceph_lock);
1915         if (to == ci->i_truncate_size) {
1916                 ci->i_truncate_pending = 0;
1917                 finish = 1;
1918         }
1919         spin_unlock(&ci->i_ceph_lock);
1920         if (!finish)
1921                 goto retry;
1922
1923         mutex_unlock(&ci->i_truncate_mutex);
1924
1925         if (wrbuffer_refs == 0)
1926                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1927
1928         wake_up_all(&ci->i_cap_wq);
1929 }
1930
1931 /*
1932  * symlinks
1933  */
1934 static const struct inode_operations ceph_symlink_iops = {
1935         .get_link = simple_get_link,
1936         .setattr = ceph_setattr,
1937         .getattr = ceph_getattr,
1938         .listxattr = ceph_listxattr,
1939 };
1940
1941 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1942 {
1943         struct ceph_inode_info *ci = ceph_inode(inode);
1944         const unsigned int ia_valid = attr->ia_valid;
1945         struct ceph_mds_request *req;
1946         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1947         struct ceph_cap_flush *prealloc_cf;
1948         int issued;
1949         int release = 0, dirtied = 0;
1950         int mask = 0;
1951         int err = 0;
1952         int inode_dirty_flags = 0;
1953         bool lock_snap_rwsem = false;
1954
1955         prealloc_cf = ceph_alloc_cap_flush();
1956         if (!prealloc_cf)
1957                 return -ENOMEM;
1958
1959         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1960                                        USE_AUTH_MDS);
1961         if (IS_ERR(req)) {
1962                 ceph_free_cap_flush(prealloc_cf);
1963                 return PTR_ERR(req);
1964         }
1965
1966         spin_lock(&ci->i_ceph_lock);
1967         issued = __ceph_caps_issued(ci, NULL);
1968
1969         if (!ci->i_head_snapc &&
1970             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1971                 lock_snap_rwsem = true;
1972                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1973                         spin_unlock(&ci->i_ceph_lock);
1974                         down_read(&mdsc->snap_rwsem);
1975                         spin_lock(&ci->i_ceph_lock);
1976                         issued = __ceph_caps_issued(ci, NULL);
1977                 }
1978         }
1979
1980         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1981
1982         if (ia_valid & ATTR_UID) {
1983                 dout("setattr %p uid %d -> %d\n", inode,
1984                      from_kuid(&init_user_ns, inode->i_uid),
1985                      from_kuid(&init_user_ns, attr->ia_uid));
1986                 if (issued & CEPH_CAP_AUTH_EXCL) {
1987                         inode->i_uid = attr->ia_uid;
1988                         dirtied |= CEPH_CAP_AUTH_EXCL;
1989                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1990                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1991                         req->r_args.setattr.uid = cpu_to_le32(
1992                                 from_kuid(&init_user_ns, attr->ia_uid));
1993                         mask |= CEPH_SETATTR_UID;
1994                         release |= CEPH_CAP_AUTH_SHARED;
1995                 }
1996         }
1997         if (ia_valid & ATTR_GID) {
1998                 dout("setattr %p gid %d -> %d\n", inode,
1999                      from_kgid(&init_user_ns, inode->i_gid),
2000                      from_kgid(&init_user_ns, attr->ia_gid));
2001                 if (issued & CEPH_CAP_AUTH_EXCL) {
2002                         inode->i_gid = attr->ia_gid;
2003                         dirtied |= CEPH_CAP_AUTH_EXCL;
2004                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2005                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2006                         req->r_args.setattr.gid = cpu_to_le32(
2007                                 from_kgid(&init_user_ns, attr->ia_gid));
2008                         mask |= CEPH_SETATTR_GID;
2009                         release |= CEPH_CAP_AUTH_SHARED;
2010                 }
2011         }
2012         if (ia_valid & ATTR_MODE) {
2013                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2014                      attr->ia_mode);
2015                 if (issued & CEPH_CAP_AUTH_EXCL) {
2016                         inode->i_mode = attr->ia_mode;
2017                         dirtied |= CEPH_CAP_AUTH_EXCL;
2018                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2019                            attr->ia_mode != inode->i_mode) {
2020                         inode->i_mode = attr->ia_mode;
2021                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2022                         mask |= CEPH_SETATTR_MODE;
2023                         release |= CEPH_CAP_AUTH_SHARED;
2024                 }
2025         }
2026
2027         if (ia_valid & ATTR_ATIME) {
2028                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2029                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2030                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2031                 if (issued & CEPH_CAP_FILE_EXCL) {
2032                         ci->i_time_warp_seq++;
2033                         inode->i_atime = attr->ia_atime;
2034                         dirtied |= CEPH_CAP_FILE_EXCL;
2035                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2036                            timespec64_compare(&inode->i_atime,
2037                                             &attr->ia_atime) < 0) {
2038                         inode->i_atime = attr->ia_atime;
2039                         dirtied |= CEPH_CAP_FILE_WR;
2040                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2041                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2042                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2043                                                &attr->ia_atime);
2044                         mask |= CEPH_SETATTR_ATIME;
2045                         release |= CEPH_CAP_FILE_SHARED |
2046                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2047                 }
2048         }
2049         if (ia_valid & ATTR_MTIME) {
2050                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2051                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2052                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2053                 if (issued & CEPH_CAP_FILE_EXCL) {
2054                         ci->i_time_warp_seq++;
2055                         inode->i_mtime = attr->ia_mtime;
2056                         dirtied |= CEPH_CAP_FILE_EXCL;
2057                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2058                            timespec64_compare(&inode->i_mtime,
2059                                             &attr->ia_mtime) < 0) {
2060                         inode->i_mtime = attr->ia_mtime;
2061                         dirtied |= CEPH_CAP_FILE_WR;
2062                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2063                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2064                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2065                                                &attr->ia_mtime);
2066                         mask |= CEPH_SETATTR_MTIME;
2067                         release |= CEPH_CAP_FILE_SHARED |
2068                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2069                 }
2070         }
2071         if (ia_valid & ATTR_SIZE) {
2072                 dout("setattr %p size %lld -> %lld\n", inode,
2073                      inode->i_size, attr->ia_size);
2074                 if ((issued & CEPH_CAP_FILE_EXCL) &&
2075                     attr->ia_size > inode->i_size) {
2076                         i_size_write(inode, attr->ia_size);
2077                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2078                         ci->i_reported_size = attr->ia_size;
2079                         dirtied |= CEPH_CAP_FILE_EXCL;
2080                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2081                            attr->ia_size != inode->i_size) {
2082                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2083                         req->r_args.setattr.old_size =
2084                                 cpu_to_le64(inode->i_size);
2085                         mask |= CEPH_SETATTR_SIZE;
2086                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2087                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2088                 }
2089         }
2090
2091         /* these do nothing */
2092         if (ia_valid & ATTR_CTIME) {
2093                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2094                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2095                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2096                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2097                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2098                      only ? "ctime only" : "ignored");
2099                 if (only) {
2100                         /*
2101                          * if kernel wants to dirty ctime but nothing else,
2102                          * we need to choose a cap to dirty under, or do
2103                          * a almost-no-op setattr
2104                          */
2105                         if (issued & CEPH_CAP_AUTH_EXCL)
2106                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2107                         else if (issued & CEPH_CAP_FILE_EXCL)
2108                                 dirtied |= CEPH_CAP_FILE_EXCL;
2109                         else if (issued & CEPH_CAP_XATTR_EXCL)
2110                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2111                         else
2112                                 mask |= CEPH_SETATTR_CTIME;
2113                 }
2114         }
2115         if (ia_valid & ATTR_FILE)
2116                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2117
2118         if (dirtied) {
2119                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2120                                                            &prealloc_cf);
2121                 inode->i_ctime = attr->ia_ctime;
2122         }
2123
2124         release &= issued;
2125         spin_unlock(&ci->i_ceph_lock);
2126         if (lock_snap_rwsem)
2127                 up_read(&mdsc->snap_rwsem);
2128
2129         if (inode_dirty_flags)
2130                 __mark_inode_dirty(inode, inode_dirty_flags);
2131
2132
2133         if (mask) {
2134                 req->r_inode = inode;
2135                 ihold(inode);
2136                 req->r_inode_drop = release;
2137                 req->r_args.setattr.mask = cpu_to_le32(mask);
2138                 req->r_num_caps = 1;
2139                 req->r_stamp = attr->ia_ctime;
2140                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2141         }
2142         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2143              ceph_cap_string(dirtied), mask);
2144
2145         ceph_mdsc_put_request(req);
2146         ceph_free_cap_flush(prealloc_cf);
2147
2148         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2149                 __ceph_do_pending_vmtruncate(inode);
2150
2151         return err;
2152 }
2153
2154 /*
2155  * setattr
2156  */
2157 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2158 {
2159         struct inode *inode = d_inode(dentry);
2160         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2161         int err;
2162
2163         if (ceph_snap(inode) != CEPH_NOSNAP)
2164                 return -EROFS;
2165
2166         err = setattr_prepare(dentry, attr);
2167         if (err != 0)
2168                 return err;
2169
2170         if ((attr->ia_valid & ATTR_SIZE) &&
2171             attr->ia_size > max(inode->i_size, fsc->max_file_size))
2172                 return -EFBIG;
2173
2174         if ((attr->ia_valid & ATTR_SIZE) &&
2175             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2176                 return -EDQUOT;
2177
2178         err = __ceph_setattr(inode, attr);
2179
2180         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2181                 err = posix_acl_chmod(inode, attr->ia_mode);
2182
2183         return err;
2184 }
2185
2186 /*
2187  * Verify that we have a lease on the given mask.  If not,
2188  * do a getattr against an mds.
2189  */
2190 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2191                       int mask, bool force)
2192 {
2193         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2194         struct ceph_mds_client *mdsc = fsc->mdsc;
2195         struct ceph_mds_request *req;
2196         int mode;
2197         int err;
2198
2199         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2200                 dout("do_getattr inode %p SNAPDIR\n", inode);
2201                 return 0;
2202         }
2203
2204         dout("do_getattr inode %p mask %s mode 0%o\n",
2205              inode, ceph_cap_string(mask), inode->i_mode);
2206         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2207                 return 0;
2208
2209         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2210         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2211         if (IS_ERR(req))
2212                 return PTR_ERR(req);
2213         req->r_inode = inode;
2214         ihold(inode);
2215         req->r_num_caps = 1;
2216         req->r_args.getattr.mask = cpu_to_le32(mask);
2217         req->r_locked_page = locked_page;
2218         err = ceph_mdsc_do_request(mdsc, NULL, req);
2219         if (locked_page && err == 0) {
2220                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2221                 if (inline_version == 0) {
2222                         /* the reply is supposed to contain inline data */
2223                         err = -EINVAL;
2224                 } else if (inline_version == CEPH_INLINE_NONE) {
2225                         err = -ENODATA;
2226                 } else {
2227                         err = req->r_reply_info.targeti.inline_len;
2228                 }
2229         }
2230         ceph_mdsc_put_request(req);
2231         dout("do_getattr result=%d\n", err);
2232         return err;
2233 }
2234
2235
2236 /*
2237  * Check inode permissions.  We verify we have a valid value for
2238  * the AUTH cap, then call the generic handler.
2239  */
2240 int ceph_permission(struct inode *inode, int mask)
2241 {
2242         int err;
2243
2244         if (mask & MAY_NOT_BLOCK)
2245                 return -ECHILD;
2246
2247         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2248
2249         if (!err)
2250                 err = generic_permission(inode, mask);
2251         return err;
2252 }
2253
2254 /*
2255  * Get all attributes.  Hopefully somedata we'll have a statlite()
2256  * and can limit the fields we require to be accurate.
2257  */
2258 int ceph_getattr(const struct path *path, struct kstat *stat,
2259                  u32 request_mask, unsigned int flags)
2260 {
2261         struct inode *inode = d_inode(path->dentry);
2262         struct ceph_inode_info *ci = ceph_inode(inode);
2263         int err;
2264
2265         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2266         if (!err) {
2267                 generic_fillattr(inode, stat);
2268                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2269                 if (ceph_snap(inode) != CEPH_NOSNAP)
2270                         stat->dev = ceph_snap(inode);
2271                 else
2272                         stat->dev = 0;
2273                 if (S_ISDIR(inode->i_mode)) {
2274                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2275                                                 RBYTES))
2276                                 stat->size = ci->i_rbytes;
2277                         else
2278                                 stat->size = ci->i_files + ci->i_subdirs;
2279                         stat->blocks = 0;
2280                         stat->blksize = 65536;
2281                         /*
2282                          * Some applications rely on the number of st_nlink
2283                          * value on directories to be either 0 (if unlinked)
2284                          * or 2 + number of subdirectories.
2285                          */
2286                         if (stat->nlink == 1)
2287                                 /* '.' + '..' + subdirs */
2288                                 stat->nlink = 1 + 1 + ci->i_subdirs;
2289                 }
2290         }
2291         return err;
2292 }