fs/ceph/inode.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/module.h>
   4 #include <linux/fs.h>
   5 #include <linux/slab.h>
   6 #include <linux/string.h>
   7 #include <linux/uaccess.h>
   8 #include <linux/kernel.h>
   9 #include <linux/namei.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12
  13 #include "super.h"
  14 #include "mds_client.h"
  15 #include "cache.h"
  16 #include <linux/ceph/decode.h>
  17
  18 /*
  19  * Ceph inode operations
  20  *
  21  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  22  * setattr, etc.), xattr helpers, and helpers for assimilating
  23  * metadata returned by the MDS into our cache.
  24  *
  25  * Also define helpers for doing asynchronous writeback, invalidation,
  26  * and truncation for the benefit of those who can't afford to block
  27  * (typically because they are in the message handler path).
  28  */
  29
  30 static const struct inode_operations ceph_symlink_iops;
  31
  32 static void ceph_invalidate_work(struct work_struct *work);
  33 static void ceph_writeback_work(struct work_struct *work);
  34 static void ceph_vmtruncate_work(struct work_struct *work);
  35
  36 /*
  37  * find or create an inode, given the ceph ino number
  38  */
  39 static int ceph_set_ino_cb(struct inode *inode, void *data)
  40 {
  41         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
  42         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
  43         return 0;
  44 }
  45
  46 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  47 {
  48         struct inode *inode;
  49         ino_t t = ceph_vino_to_ino(vino);
  50
  51         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
  52         if (inode == NULL)
  53                 return ERR_PTR(-ENOMEM);
  54         if (inode->i_state & I_NEW) {
  55                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
  56                      inode, ceph_vinop(inode), (u64)inode->i_ino);
  57                 unlock_new_inode(inode);
  58         }
  59
  60         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
  61              vino.snap, inode);
  62         return inode;
  63 }
  64
  65 /*
  66  * get/constuct snapdir inode for a given directory
  67  */
  68 struct inode *ceph_get_snapdir(struct inode *parent)
  69 {
  70         struct ceph_vino vino = {
  71                 .ino = ceph_ino(parent),
  72                 .snap = CEPH_SNAPDIR,
  73         };
  74         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  75         struct ceph_inode_info *ci = ceph_inode(inode);
  76
  77         BUG_ON(!S_ISDIR(parent->i_mode));
  78         if (IS_ERR(inode))
  79                 return inode;
  80         inode->i_mode = parent->i_mode;
  81         inode->i_uid = parent->i_uid;
  82         inode->i_gid = parent->i_gid;
  83         inode->i_op = &ceph_dir_iops;
  84         inode->i_fop = &ceph_dir_fops;
  85         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
  86         ci->i_rbytes = 0;
  87         return inode;
  88 }
  89
  90 const struct inode_operations ceph_file_iops = {
  91         .permission = ceph_permission,
  92         .setattr = ceph_setattr,
  93         .getattr = ceph_getattr,
  94         .setxattr = ceph_setxattr,
  95         .getxattr = ceph_getxattr,
  96         .listxattr = ceph_listxattr,
  97         .removexattr = ceph_removexattr,
  98 };
  99
 100
 101 /*
 102  * We use a 'frag tree' to keep track of the MDS's directory fragments
 103  * for a given inode (usually there is just a single fragment).  We
 104  * need to know when a child frag is delegated to a new MDS, or when
 105  * it is flagged as replicated, so we can direct our requests
 106  * accordingly.
 107  */
 108
 109 /*
 110  * find/create a frag in the tree
 111  */
 112 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 113                                                     u32 f)
 114 {
 115         struct rb_node **p;
 116         struct rb_node *parent = NULL;
 117         struct ceph_inode_frag *frag;
 118         int c;
 119
 120         p = &ci->i_fragtree.rb_node;
 121         while (*p) {
 122                 parent = *p;
 123                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 124                 c = ceph_frag_compare(f, frag->frag);
 125                 if (c < 0)
 126                         p = &(*p)->rb_left;
 127                 else if (c > 0)
 128                         p = &(*p)->rb_right;
 129                 else
 130                         return frag;
 131         }
 132
 133         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 134         if (!frag) {
 135                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 136                        "frag %x\n", &ci->vfs_inode,
 137                        ceph_vinop(&ci->vfs_inode), f);
 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 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 180                      struct ceph_inode_frag *pfrag,
 181                      int *found)
 182 {
 183         u32 t = ceph_frag_make(0, 0);
 184         struct ceph_inode_frag *frag;
 185         unsigned nway, i;
 186         u32 n;
 187
 188         if (found)
 189                 *found = 0;
 190
 191         mutex_lock(&ci->i_fragtree_mutex);
 192         while (1) {
 193                 WARN_ON(!ceph_frag_contains_value(t, v));
 194                 frag = __ceph_find_frag(ci, t);
 195                 if (!frag)
 196                         break; /* t is a leaf */
 197                 if (frag->split_by == 0) {
 198                         if (pfrag)
 199                                 memcpy(pfrag, frag, sizeof(*pfrag));
 200                         if (found)
 201                                 *found = 1;
 202                         break;
 203                 }
 204
 205                 /* choose child */
 206                 nway = 1 << frag->split_by;
 207                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 208                      frag->split_by, nway);
 209                 for (i = 0; i < nway; i++) {
 210                         n = ceph_frag_make_child(t, frag->split_by, i);
 211                         if (ceph_frag_contains_value(n, v)) {
 212                                 t = n;
 213                                 break;
 214                         }
 215                 }
 216                 BUG_ON(i == nway);
 217         }
 218         dout("choose_frag(%x) = %x\n", v, t);
 219
 220         mutex_unlock(&ci->i_fragtree_mutex);
 221         return t;
 222 }
 223
 224 /*
 225  * Process dirfrag (delegation) info from the mds.  Include leaf
 226  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 227  * branches/splits are included in i_fragtree)
 228  */
 229 static int ceph_fill_dirfrag(struct inode *inode,
 230                              struct ceph_mds_reply_dirfrag *dirinfo)
 231 {
 232         struct ceph_inode_info *ci = ceph_inode(inode);
 233         struct ceph_inode_frag *frag;
 234         u32 id = le32_to_cpu(dirinfo->frag);
 235         int mds = le32_to_cpu(dirinfo->auth);
 236         int ndist = le32_to_cpu(dirinfo->ndist);
 237         int i;
 238         int err = 0;
 239
 240         mutex_lock(&ci->i_fragtree_mutex);
 241         if (ndist == 0) {
 242                 /* no delegation info needed. */
 243                 frag = __ceph_find_frag(ci, id);
 244                 if (!frag)
 245                         goto out;
 246                 if (frag->split_by == 0) {
 247                         /* tree leaf, remove */
 248                         dout("fill_dirfrag removed %llx.%llx frag %x"
 249                              " (no ref)\n", ceph_vinop(inode), id);
 250                         rb_erase(&frag->node, &ci->i_fragtree);
 251                         kfree(frag);
 252                 } else {
 253                         /* tree branch, keep and clear */
 254                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 255                              " referral\n", ceph_vinop(inode), id);
 256                         frag->mds = -1;
 257                         frag->ndist = 0;
 258                 }
 259                 goto out;
 260         }
 261
 262
 263         /* find/add this frag to store mds delegation info */
 264         frag = __get_or_create_frag(ci, id);
 265         if (IS_ERR(frag)) {
 266                 /* this is not the end of the world; we can continue
 267                    with bad/inaccurate delegation info */
 268                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 269                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 270                 err = -ENOMEM;
 271                 goto out;
 272         }
 273
 274         frag->mds = mds;
 275         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 276         for (i = 0; i < frag->ndist; i++)
 277                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 278         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 279              ceph_vinop(inode), frag->frag, frag->ndist);
 280
 281 out:
 282         mutex_unlock(&ci->i_fragtree_mutex);
 283         return err;
 284 }
 285
 286
 287 /*
 288  * initialize a newly allocated inode.
 289  */
 290 struct inode *ceph_alloc_inode(struct super_block *sb)
 291 {
 292         struct ceph_inode_info *ci;
 293         int i;
 294
 295         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 296         if (!ci)
 297                 return NULL;
 298
 299         dout("alloc_inode %p\n", &ci->vfs_inode);
 300
 301         spin_lock_init(&ci->i_ceph_lock);
 302
 303         ci->i_version = 0;
 304         ci->i_time_warp_seq = 0;
 305         ci->i_ceph_flags = 0;
 306         atomic_set(&ci->i_release_count, 1);
 307         atomic_set(&ci->i_complete_count, 0);
 308         ci->i_symlink = NULL;
 309
 310         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 311
 312         ci->i_fragtree = RB_ROOT;
 313         mutex_init(&ci->i_fragtree_mutex);
 314
 315         ci->i_xattrs.blob = NULL;
 316         ci->i_xattrs.prealloc_blob = NULL;
 317         ci->i_xattrs.dirty = false;
 318         ci->i_xattrs.index = RB_ROOT;
 319         ci->i_xattrs.count = 0;
 320         ci->i_xattrs.names_size = 0;
 321         ci->i_xattrs.vals_size = 0;
 322         ci->i_xattrs.version = 0;
 323         ci->i_xattrs.index_version = 0;
 324
 325         ci->i_caps = RB_ROOT;
 326         ci->i_auth_cap = NULL;
 327         ci->i_dirty_caps = 0;
 328         ci->i_flushing_caps = 0;
 329         INIT_LIST_HEAD(&ci->i_dirty_item);
 330         INIT_LIST_HEAD(&ci->i_flushing_item);
 331         ci->i_cap_flush_seq = 0;
 332         ci->i_cap_flush_last_tid = 0;
 333         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 334         init_waitqueue_head(&ci->i_cap_wq);
 335         ci->i_hold_caps_min = 0;
 336         ci->i_hold_caps_max = 0;
 337         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 338         ci->i_cap_exporting_mds = 0;
 339         ci->i_cap_exporting_mseq = 0;
 340         ci->i_cap_exporting_issued = 0;
 341         INIT_LIST_HEAD(&ci->i_cap_snaps);
 342         ci->i_head_snapc = NULL;
 343         ci->i_snap_caps = 0;
 344
 345         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 346                 ci->i_nr_by_mode[i] = 0;
 347
 348         mutex_init(&ci->i_truncate_mutex);
 349         ci->i_truncate_seq = 0;
 350         ci->i_truncate_size = 0;
 351         ci->i_truncate_pending = 0;
 352
 353         ci->i_max_size = 0;
 354         ci->i_reported_size = 0;
 355         ci->i_wanted_max_size = 0;
 356         ci->i_requested_max_size = 0;
 357
 358         ci->i_pin_ref = 0;
 359         ci->i_rd_ref = 0;
 360         ci->i_rdcache_ref = 0;
 361         ci->i_wr_ref = 0;
 362         ci->i_wb_ref = 0;
 363         ci->i_wrbuffer_ref = 0;
 364         ci->i_wrbuffer_ref_head = 0;
 365         ci->i_shared_gen = 0;
 366         ci->i_rdcache_gen = 0;
 367         ci->i_rdcache_revoking = 0;
 368
 369         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 370         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 371         spin_lock_init(&ci->i_unsafe_lock);
 372
 373         ci->i_snap_realm = NULL;
 374         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 375         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 376
 377         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 378         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 379
 380         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 381
 382         ceph_fscache_inode_init(ci);
 383
 384         return &ci->vfs_inode;
 385 }
 386
 387 static void ceph_i_callback(struct rcu_head *head)
 388 {
 389         struct inode *inode = container_of(head, struct inode, i_rcu);
 390         struct ceph_inode_info *ci = ceph_inode(inode);
 391
 392         kmem_cache_free(ceph_inode_cachep, ci);
 393 }
 394
 395 void ceph_destroy_inode(struct inode *inode)
 396 {
 397         struct ceph_inode_info *ci = ceph_inode(inode);
 398         struct ceph_inode_frag *frag;
 399         struct rb_node *n;
 400
 401         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 402
 403         ceph_fscache_unregister_inode_cookie(ci);
 404
 405         ceph_queue_caps_release(inode);
 406
 407         /*
 408          * we may still have a snap_realm reference if there are stray
 409          * caps in i_cap_exporting_issued or i_snap_caps.
 410          */
 411         if (ci->i_snap_realm) {
 412                 struct ceph_mds_client *mdsc =
 413                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 414                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 415
 416                 dout(" dropping residual ref to snap realm %p\n", realm);
 417                 spin_lock(&realm->inodes_with_caps_lock);
 418                 list_del_init(&ci->i_snap_realm_item);
 419                 spin_unlock(&realm->inodes_with_caps_lock);
 420                 ceph_put_snap_realm(mdsc, realm);
 421         }
 422
 423         kfree(ci->i_symlink);
 424         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 425                 frag = rb_entry(n, struct ceph_inode_frag, node);
 426                 rb_erase(n, &ci->i_fragtree);
 427                 kfree(frag);
 428         }
 429
 430         __ceph_destroy_xattrs(ci);
 431         if (ci->i_xattrs.blob)
 432                 ceph_buffer_put(ci->i_xattrs.blob);
 433         if (ci->i_xattrs.prealloc_blob)
 434                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 435
 436         call_rcu(&inode->i_rcu, ceph_i_callback);
 437 }
 438
 439 /*
 440  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 441  * careful because either the client or MDS may have more up to date
 442  * info, depending on which capabilities are held, and whether
 443  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 444  * and size are monotonically increasing, except when utimes() or
 445  * truncate() increments the corresponding _seq values.)
 446  */
 447 int ceph_fill_file_size(struct inode *inode, int issued,
 448                         u32 truncate_seq, u64 truncate_size, u64 size)
 449 {
 450         struct ceph_inode_info *ci = ceph_inode(inode);
 451         int queue_trunc = 0;
 452
 453         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 454             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 455                 dout("size %lld -> %llu\n", inode->i_size, size);
 456                 inode->i_size = size;
 457                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 458                 ci->i_reported_size = size;
 459                 if (truncate_seq != ci->i_truncate_seq) {
 460                         dout("truncate_seq %u -> %u\n",
 461                              ci->i_truncate_seq, truncate_seq);
 462                         ci->i_truncate_seq = truncate_seq;
 463
 464                         /* the MDS should have revoked these caps */
 465                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 466                                                CEPH_CAP_FILE_RD |
 467                                                CEPH_CAP_FILE_WR |
 468                                                CEPH_CAP_FILE_LAZYIO));
 469                         /*
 470                          * If we hold relevant caps, or in the case where we're
 471                          * not the only client referencing this file and we
 472                          * don't hold those caps, then we need to check whether
 473                          * the file is either opened or mmaped
 474                          */
 475                         if ((issued & (CEPH_CAP_FILE_CACHE|
 476                                        CEPH_CAP_FILE_BUFFER)) ||
 477                             mapping_mapped(inode->i_mapping) ||
 478                             __ceph_caps_file_wanted(ci)) {
 479                                 ci->i_truncate_pending++;
 480                                 queue_trunc = 1;
 481                         }
 482                 }
 483         }
 484         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 485             ci->i_truncate_size != truncate_size) {
 486                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 487                      truncate_size);
 488                 ci->i_truncate_size = truncate_size;
 489         }
 490
 491         if (queue_trunc)
 492                 ceph_fscache_invalidate(inode);
 493
 494         return queue_trunc;
 495 }
 496
 497 void ceph_fill_file_time(struct inode *inode, int issued,
 498                          u64 time_warp_seq, struct timespec *ctime,
 499                          struct timespec *mtime, struct timespec *atime)
 500 {
 501         struct ceph_inode_info *ci = ceph_inode(inode);
 502         int warn = 0;
 503
 504         if (issued & (CEPH_CAP_FILE_EXCL|
 505                       CEPH_CAP_FILE_WR|
 506                       CEPH_CAP_FILE_BUFFER|
 507                       CEPH_CAP_AUTH_EXCL|
 508                       CEPH_CAP_XATTR_EXCL)) {
 509                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 510                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 511                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 512                              ctime->tv_sec, ctime->tv_nsec);
 513                         inode->i_ctime = *ctime;
 514                 }
 515                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 516                         /* the MDS did a utimes() */
 517                         dout("mtime %ld.%09ld -> %ld.%09ld "
 518                              "tw %d -> %d\n",
 519                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 520                              mtime->tv_sec, mtime->tv_nsec,
 521                              ci->i_time_warp_seq, (int)time_warp_seq);
 522
 523                         inode->i_mtime = *mtime;
 524                         inode->i_atime = *atime;
 525                         ci->i_time_warp_seq = time_warp_seq;
 526                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 527                         /* nobody did utimes(); take the max */
 528                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 529                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 530                                      inode->i_mtime.tv_sec,
 531                                      inode->i_mtime.tv_nsec,
 532                                      mtime->tv_sec, mtime->tv_nsec);
 533                                 inode->i_mtime = *mtime;
 534                         }
 535                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 536                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 537                                      inode->i_atime.tv_sec,
 538                                      inode->i_atime.tv_nsec,
 539                                      atime->tv_sec, atime->tv_nsec);
 540                                 inode->i_atime = *atime;
 541                         }
 542                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 543                         /* we did a utimes(); ignore mds values */
 544                 } else {
 545                         warn = 1;
 546                 }
 547         } else {
 548                 /* we have no write|excl caps; whatever the MDS says is true */
 549                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 550                         inode->i_ctime = *ctime;
 551                         inode->i_mtime = *mtime;
 552                         inode->i_atime = *atime;
 553                         ci->i_time_warp_seq = time_warp_seq;
 554                 } else {
 555                         warn = 1;
 556                 }
 557         }
 558         if (warn) /* time_warp_seq shouldn't go backwards */
 559                 dout("%p mds time_warp_seq %llu < %u\n",
 560                      inode, time_warp_seq, ci->i_time_warp_seq);
 561 }
 562
 563 /*
 564  * Populate an inode based on info from mds.  May be called on new or
 565  * existing inodes.
 566  */
 567 static int fill_inode(struct inode *inode,
 568                       struct ceph_mds_reply_info_in *iinfo,
 569                       struct ceph_mds_reply_dirfrag *dirinfo,
 570                       struct ceph_mds_session *session,
 571                       unsigned long ttl_from, int cap_fmode,
 572                       struct ceph_cap_reservation *caps_reservation)
 573 {
 574         struct ceph_mds_reply_inode *info = iinfo->in;
 575         struct ceph_inode_info *ci = ceph_inode(inode);
 576         int i;
 577         int issued = 0, implemented;
 578         struct timespec mtime, atime, ctime;
 579         u32 nsplits;
 580         struct ceph_buffer *xattr_blob = NULL;
 581         int err = 0;
 582         int queue_trunc = 0;
 583
 584         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 585              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 586              ci->i_version);
 587
 588         /*
 589          * prealloc xattr data, if it looks like we'll need it.  only
 590          * if len > 4 (meaning there are actually xattrs; the first 4
 591          * bytes are the xattr count).
 592          */
 593         if (iinfo->xattr_len > 4) {
 594                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 595                 if (!xattr_blob)
 596                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 597                                iinfo->xattr_len);
 598         }
 599
 600         spin_lock(&ci->i_ceph_lock);
 601
 602         /*
 603          * provided version will be odd if inode value is projected,
 604          * even if stable.  skip the update if we have newer stable
 605          * info (ours>=theirs, e.g. due to racing mds replies), unless
 606          * we are getting projected (unstable) info (in which case the
 607          * version is odd, and we want ours>theirs).
 608          *   us   them
 609          *   2    2     skip
 610          *   3    2     skip
 611          *   3    3     update
 612          */
 613         if (le64_to_cpu(info->version) > 0 &&
 614             (ci->i_version & ~1) >= le64_to_cpu(info->version))
 615                 goto no_change;
 616
 617         issued = __ceph_caps_issued(ci, &implemented);
 618         issued |= implemented | __ceph_caps_dirty(ci);
 619
 620         /* update inode */
 621         ci->i_version = le64_to_cpu(info->version);
 622         inode->i_version++;
 623         inode->i_rdev = le32_to_cpu(info->rdev);
 624
 625         if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
 626                 inode->i_mode = le32_to_cpu(info->mode);
 627                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 628                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 629                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 630                      from_kuid(&init_user_ns, inode->i_uid),
 631                      from_kgid(&init_user_ns, inode->i_gid));
 632         }
 633
 634         if ((issued & CEPH_CAP_LINK_EXCL) == 0)
 635                 set_nlink(inode, le32_to_cpu(info->nlink));
 636
 637         /* be careful with mtime, atime, size */
 638         ceph_decode_timespec(&atime, &info->atime);
 639         ceph_decode_timespec(&mtime, &info->mtime);
 640         ceph_decode_timespec(&ctime, &info->ctime);
 641         queue_trunc = ceph_fill_file_size(inode, issued,
 642                                           le32_to_cpu(info->truncate_seq),
 643                                           le64_to_cpu(info->truncate_size),
 644                                           le64_to_cpu(info->size));
 645         ceph_fill_file_time(inode, issued,
 646                             le32_to_cpu(info->time_warp_seq),
 647                             &ctime, &mtime, &atime);
 648
 649         /* only update max_size on auth cap */
 650         if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 651             ci->i_max_size != le64_to_cpu(info->max_size)) {
 652                 dout("max_size %lld -> %llu\n", ci->i_max_size,
 653                      le64_to_cpu(info->max_size));
 654                 ci->i_max_size = le64_to_cpu(info->max_size);
 655         }
 656
 657         ci->i_layout = info->layout;
 658         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 659
 660         /* xattrs */
 661         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 662         if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
 663             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 664                 if (ci->i_xattrs.blob)
 665                         ceph_buffer_put(ci->i_xattrs.blob);
 666                 ci->i_xattrs.blob = xattr_blob;
 667                 if (xattr_blob)
 668                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 669                                iinfo->xattr_data, iinfo->xattr_len);
 670                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 671                 xattr_blob = NULL;
 672         }
 673
 674         inode->i_mapping->a_ops = &ceph_aops;
 675         inode->i_mapping->backing_dev_info =
 676                 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
 677
 678         switch (inode->i_mode & S_IFMT) {
 679         case S_IFIFO:
 680         case S_IFBLK:
 681         case S_IFCHR:
 682         case S_IFSOCK:
 683                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 684                 inode->i_op = &ceph_file_iops;
 685                 break;
 686         case S_IFREG:
 687                 inode->i_op = &ceph_file_iops;
 688                 inode->i_fop = &ceph_file_fops;
 689                 break;
 690         case S_IFLNK:
 691                 inode->i_op = &ceph_symlink_iops;
 692                 if (!ci->i_symlink) {
 693                         u32 symlen = iinfo->symlink_len;
 694                         char *sym;
 695
 696                         spin_unlock(&ci->i_ceph_lock);
 697
 698                         err = -EINVAL;
 699                         if (WARN_ON(symlen != inode->i_size))
 700                                 goto out;
 701
 702                         err = -ENOMEM;
 703                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 704                         if (!sym)
 705                                 goto out;
 706
 707                         spin_lock(&ci->i_ceph_lock);
 708                         if (!ci->i_symlink)
 709                                 ci->i_symlink = sym;
 710                         else
 711                                 kfree(sym); /* lost a race */
 712                 }
 713                 break;
 714         case S_IFDIR:
 715                 inode->i_op = &ceph_dir_iops;
 716                 inode->i_fop = &ceph_dir_fops;
 717
 718                 ci->i_dir_layout = iinfo->dir_layout;
 719
 720                 ci->i_files = le64_to_cpu(info->files);
 721                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 722                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 723                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 724                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 725                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 726                 break;
 727         default:
 728                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 729                        ceph_vinop(inode), inode->i_mode);
 730         }
 731
 732         /* set dir completion flag? */
 733         if (S_ISDIR(inode->i_mode) &&
 734             ci->i_files == 0 && ci->i_subdirs == 0 &&
 735             ceph_snap(inode) == CEPH_NOSNAP &&
 736             (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
 737             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 738             !__ceph_dir_is_complete(ci)) {
 739                 dout(" marking %p complete (empty)\n", inode);
 740                 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count));
 741                 ci->i_max_offset = 2;
 742         }
 743 no_change:
 744         spin_unlock(&ci->i_ceph_lock);
 745
 746         /* queue truncate if we saw i_size decrease */
 747         if (queue_trunc)
 748                 ceph_queue_vmtruncate(inode);
 749
 750         /* populate frag tree */
 751         /* FIXME: move me up, if/when version reflects fragtree changes */
 752         nsplits = le32_to_cpu(info->fragtree.nsplits);
 753         mutex_lock(&ci->i_fragtree_mutex);
 754         for (i = 0; i < nsplits; i++) {
 755                 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
 756                 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
 757
 758                 if (IS_ERR(frag))
 759                         continue;
 760                 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
 761                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 762         }
 763         mutex_unlock(&ci->i_fragtree_mutex);
 764
 765         /* were we issued a capability? */
 766         if (info->cap.caps) {
 767                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 768                         ceph_add_cap(inode, session,
 769                                      le64_to_cpu(info->cap.cap_id),
 770                                      cap_fmode,
 771                                      le32_to_cpu(info->cap.caps),
 772                                      le32_to_cpu(info->cap.wanted),
 773                                      le32_to_cpu(info->cap.seq),
 774                                      le32_to_cpu(info->cap.mseq),
 775                                      le64_to_cpu(info->cap.realm),
 776                                      info->cap.flags,
 777                                      caps_reservation);
 778                 } else {
 779                         spin_lock(&ci->i_ceph_lock);
 780                         dout(" %p got snap_caps %s\n", inode,
 781                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 782                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 783                         if (cap_fmode >= 0)
 784                                 __ceph_get_fmode(ci, cap_fmode);
 785                         spin_unlock(&ci->i_ceph_lock);
 786                 }
 787         } else if (cap_fmode >= 0) {
 788                 pr_warning("mds issued no caps on %llx.%llx\n",
 789                            ceph_vinop(inode));
 790                 __ceph_get_fmode(ci, cap_fmode);
 791         }
 792
 793         /* update delegation info? */
 794         if (dirinfo)
 795                 ceph_fill_dirfrag(inode, dirinfo);
 796
 797         err = 0;
 798
 799 out:
 800         if (xattr_blob)
 801                 ceph_buffer_put(xattr_blob);
 802         return err;
 803 }
 804
 805 /*
 806  * caller should hold session s_mutex.
 807  */
 808 static void update_dentry_lease(struct dentry *dentry,
 809                                 struct ceph_mds_reply_lease *lease,
 810                                 struct ceph_mds_session *session,
 811                                 unsigned long from_time)
 812 {
 813         struct ceph_dentry_info *di = ceph_dentry(dentry);
 814         long unsigned duration = le32_to_cpu(lease->duration_ms);
 815         long unsigned ttl = from_time + (duration * HZ) / 1000;
 816         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 817         struct inode *dir;
 818
 819         /* only track leases on regular dentries */
 820         if (dentry->d_op != &ceph_dentry_ops)
 821                 return;
 822
 823         spin_lock(&dentry->d_lock);
 824         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 825              dentry, duration, ttl);
 826
 827         /* make lease_rdcache_gen match directory */
 828         dir = dentry->d_parent->d_inode;
 829         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 830
 831         if (duration == 0)
 832                 goto out_unlock;
 833
 834         if (di->lease_gen == session->s_cap_gen &&
 835             time_before(ttl, dentry->d_time))
 836                 goto out_unlock;  /* we already have a newer lease. */
 837
 838         if (di->lease_session && di->lease_session != session)
 839                 goto out_unlock;
 840
 841         ceph_dentry_lru_touch(dentry);
 842
 843         if (!di->lease_session)
 844                 di->lease_session = ceph_get_mds_session(session);
 845         di->lease_gen = session->s_cap_gen;
 846         di->lease_seq = le32_to_cpu(lease->seq);
 847         di->lease_renew_after = half_ttl;
 848         di->lease_renew_from = 0;
 849         dentry->d_time = ttl;
 850 out_unlock:
 851         spin_unlock(&dentry->d_lock);
 852         return;
 853 }
 854
 855 /*
 856  * Set dentry's directory position based on the current dir's max, and
 857  * order it in d_subdirs, so that dcache_readdir behaves.
 858  *
 859  * Always called under directory's i_mutex.
 860  */
 861 static void ceph_set_dentry_offset(struct dentry *dn)
 862 {
 863         struct dentry *dir = dn->d_parent;
 864         struct inode *inode = dir->d_inode;
 865         struct ceph_inode_info *ci;
 866         struct ceph_dentry_info *di;
 867
 868         BUG_ON(!inode);
 869
 870         ci = ceph_inode(inode);
 871         di = ceph_dentry(dn);
 872
 873         spin_lock(&ci->i_ceph_lock);
 874         if (!__ceph_dir_is_complete(ci)) {
 875                 spin_unlock(&ci->i_ceph_lock);
 876                 return;
 877         }
 878         di->offset = ceph_inode(inode)->i_max_offset++;
 879         spin_unlock(&ci->i_ceph_lock);
 880
 881         spin_lock(&dir->d_lock);
 882         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
 883         list_move(&dn->d_u.d_child, &dir->d_subdirs);
 884         dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
 885              dn->d_u.d_child.prev, dn->d_u.d_child.next);
 886         spin_unlock(&dn->d_lock);
 887         spin_unlock(&dir->d_lock);
 888 }
 889
 890 /*
 891  * splice a dentry to an inode.
 892  * caller must hold directory i_mutex for this to be safe.
 893  *
 894  * we will only rehash the resulting dentry if @prehash is
 895  * true; @prehash will be set to false (for the benefit of
 896  * the caller) if we fail.
 897  */
 898 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 899                                     bool *prehash, bool set_offset)
 900 {
 901         struct dentry *realdn;
 902
 903         BUG_ON(dn->d_inode);
 904
 905         /* dn must be unhashed */
 906         if (!d_unhashed(dn))
 907                 d_drop(dn);
 908         realdn = d_materialise_unique(dn, in);
 909         if (IS_ERR(realdn)) {
 910                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 911                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 912                 if (prehash)
 913                         *prehash = false; /* don't rehash on error */
 914                 dn = realdn; /* note realdn contains the error */
 915                 goto out;
 916         } else if (realdn) {
 917                 dout("dn %p (%d) spliced with %p (%d) "
 918                      "inode %p ino %llx.%llx\n",
 919                      dn, d_count(dn),
 920                      realdn, d_count(realdn),
 921                      realdn->d_inode, ceph_vinop(realdn->d_inode));
 922                 dput(dn);
 923                 dn = realdn;
 924         } else {
 925                 BUG_ON(!ceph_dentry(dn));
 926                 dout("dn %p attached to %p ino %llx.%llx\n",
 927                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
 928         }
 929         if ((!prehash || *prehash) && d_unhashed(dn))
 930                 d_rehash(dn);
 931         if (set_offset)
 932                 ceph_set_dentry_offset(dn);
 933 out:
 934         return dn;
 935 }
 936
 937 /*
 938  * Incorporate results into the local cache.  This is either just
 939  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
 940  * after a lookup).
 941  *
 942  * A reply may contain
 943  *         a directory inode along with a dentry.
 944  *  and/or a target inode
 945  *
 946  * Called with snap_rwsem (read).
 947  */
 948 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
 949                     struct ceph_mds_session *session)
 950 {
 951         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
 952         struct inode *in = NULL;
 953         struct ceph_mds_reply_inode *ininfo;
 954         struct ceph_vino vino;
 955         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 956         int i = 0;
 957         int err = 0;
 958
 959         dout("fill_trace %p is_dentry %d is_target %d\n", req,
 960              rinfo->head->is_dentry, rinfo->head->is_target);
 961
 962 #if 0
 963         /*
 964          * Debugging hook:
 965          *
 966          * If we resend completed ops to a recovering mds, we get no
 967          * trace.  Since that is very rare, pretend this is the case
 968          * to ensure the 'no trace' handlers in the callers behave.
 969          *
 970          * Fill in inodes unconditionally to avoid breaking cap
 971          * invariants.
 972          */
 973         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
 974                 pr_info("fill_trace faking empty trace on %lld %s\n",
 975                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
 976                 if (rinfo->head->is_dentry) {
 977                         rinfo->head->is_dentry = 0;
 978                         err = fill_inode(req->r_locked_dir,
 979                                          &rinfo->diri, rinfo->dirfrag,
 980                                          session, req->r_request_started, -1);
 981                 }
 982                 if (rinfo->head->is_target) {
 983                         rinfo->head->is_target = 0;
 984                         ininfo = rinfo->targeti.in;
 985                         vino.ino = le64_to_cpu(ininfo->ino);
 986                         vino.snap = le64_to_cpu(ininfo->snapid);
 987                         in = ceph_get_inode(sb, vino);
 988                         err = fill_inode(in, &rinfo->targeti, NULL,
 989                                          session, req->r_request_started,
 990                                          req->r_fmode);
 991                         iput(in);
 992                 }
 993         }
 994 #endif
 995
 996         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
 997                 dout("fill_trace reply is empty!\n");
 998                 if (rinfo->head->result == 0 && req->r_locked_dir)
 999                         ceph_invalidate_dir_request(req);
1000                 return 0;
1001         }
1002
1003         if (rinfo->head->is_dentry) {
1004                 struct inode *dir = req->r_locked_dir;
1005
1006                 if (dir) {
1007                         err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
1008                                          session, req->r_request_started, -1,
1009                                          &req->r_caps_reservation);
1010                         if (err < 0)
1011                                 return err;
1012                 } else {
1013                         WARN_ON_ONCE(1);
1014                 }
1015         }
1016
1017         /*
1018          * ignore null lease/binding on snapdir ENOENT, or else we
1019          * will have trouble splicing in the virtual snapdir later
1020          */
1021         if (rinfo->head->is_dentry && !req->r_aborted &&
1022             req->r_locked_dir &&
1023             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1024                                                fsc->mount_options->snapdir_name,
1025                                                req->r_dentry->d_name.len))) {
1026                 /*
1027                  * lookup link rename   : null -> possibly existing inode
1028                  * mknod symlink mkdir  : null -> new inode
1029                  * unlink               : linked -> null
1030                  */
1031                 struct inode *dir = req->r_locked_dir;
1032                 struct dentry *dn = req->r_dentry;
1033                 bool have_dir_cap, have_lease;
1034
1035                 BUG_ON(!dn);
1036                 BUG_ON(!dir);
1037                 BUG_ON(dn->d_parent->d_inode != dir);
1038                 BUG_ON(ceph_ino(dir) !=
1039                        le64_to_cpu(rinfo->diri.in->ino));
1040                 BUG_ON(ceph_snap(dir) !=
1041                        le64_to_cpu(rinfo->diri.in->snapid));
1042
1043                 /* do we have a lease on the whole dir? */
1044                 have_dir_cap =
1045                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1046                          CEPH_CAP_FILE_SHARED);
1047
1048                 /* do we have a dn lease? */
1049                 have_lease = have_dir_cap ||
1050                         le32_to_cpu(rinfo->dlease->duration_ms);
1051                 if (!have_lease)
1052                         dout("fill_trace  no dentry lease or dir cap\n");
1053
1054                 /* rename? */
1055                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1056                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1057                              req->r_old_dentry,
1058                              req->r_old_dentry->d_name.len,
1059                              req->r_old_dentry->d_name.name,
1060                              dn, dn->d_name.len, dn->d_name.name);
1061                         dout("fill_trace doing d_move %p -> %p\n",
1062                              req->r_old_dentry, dn);
1063
1064                         d_move(req->r_old_dentry, dn);
1065                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1066                              req->r_old_dentry,
1067                              req->r_old_dentry->d_name.len,
1068                              req->r_old_dentry->d_name.name,
1069                              dn, dn->d_name.len, dn->d_name.name);
1070
1071                         /* ensure target dentry is invalidated, despite
1072                            rehashing bug in vfs_rename_dir */
1073                         ceph_invalidate_dentry_lease(dn);
1074
1075                         /*
1076                          * d_move() puts the renamed dentry at the end of
1077                          * d_subdirs.  We need to assign it an appropriate
1078                          * directory offset so we can behave when dir is
1079                          * complete.
1080                          */
1081                         ceph_set_dentry_offset(req->r_old_dentry);
1082                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1083                              ceph_dentry(req->r_old_dentry)->offset);
1084
1085                         dn = req->r_old_dentry;  /* use old_dentry */
1086                         in = dn->d_inode;
1087                 }
1088
1089                 /* null dentry? */
1090                 if (!rinfo->head->is_target) {
1091                         dout("fill_trace null dentry\n");
1092                         if (dn->d_inode) {
1093                                 dout("d_delete %p\n", dn);
1094                                 d_delete(dn);
1095                         } else {
1096                                 dout("d_instantiate %p NULL\n", dn);
1097                                 d_instantiate(dn, NULL);
1098                                 if (have_lease && d_unhashed(dn))
1099                                         d_rehash(dn);
1100                                 update_dentry_lease(dn, rinfo->dlease,
1101                                                     session,
1102                                                     req->r_request_started);
1103                         }
1104                         goto done;
1105                 }
1106
1107                 /* attach proper inode */
1108                 ininfo = rinfo->targeti.in;
1109                 vino.ino = le64_to_cpu(ininfo->ino);
1110                 vino.snap = le64_to_cpu(ininfo->snapid);
1111                 in = dn->d_inode;
1112                 if (!in) {
1113                         in = ceph_get_inode(sb, vino);
1114                         if (IS_ERR(in)) {
1115                                 pr_err("fill_trace bad get_inode "
1116                                        "%llx.%llx\n", vino.ino, vino.snap);
1117                                 err = PTR_ERR(in);
1118                                 d_drop(dn);
1119                                 goto done;
1120                         }
1121                         dn = splice_dentry(dn, in, &have_lease, true);
1122                         if (IS_ERR(dn)) {
1123                                 err = PTR_ERR(dn);
1124                                 goto done;
1125                         }
1126                         req->r_dentry = dn;  /* may have spliced */
1127                         ihold(in);
1128                 } else if (ceph_ino(in) == vino.ino &&
1129                            ceph_snap(in) == vino.snap) {
1130                         ihold(in);
1131                 } else {
1132                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1133                              dn, in, ceph_ino(in), ceph_snap(in),
1134                              vino.ino, vino.snap);
1135                         have_lease = false;
1136                         in = NULL;
1137                 }
1138
1139                 if (have_lease)
1140                         update_dentry_lease(dn, rinfo->dlease, session,
1141                                             req->r_request_started);
1142                 dout(" final dn %p\n", dn);
1143                 i++;
1144         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1145                    req->r_op == CEPH_MDS_OP_MKSNAP) && !req->r_aborted) {
1146                 struct dentry *dn = req->r_dentry;
1147
1148                 /* fill out a snapdir LOOKUPSNAP dentry */
1149                 BUG_ON(!dn);
1150                 BUG_ON(!req->r_locked_dir);
1151                 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1152                 ininfo = rinfo->targeti.in;
1153                 vino.ino = le64_to_cpu(ininfo->ino);
1154                 vino.snap = le64_to_cpu(ininfo->snapid);
1155                 in = ceph_get_inode(sb, vino);
1156                 if (IS_ERR(in)) {
1157                         pr_err("fill_inode get_inode badness %llx.%llx\n",
1158                                vino.ino, vino.snap);
1159                         err = PTR_ERR(in);
1160                         d_delete(dn);
1161                         goto done;
1162                 }
1163                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1164                 dn = splice_dentry(dn, in, NULL, true);
1165                 if (IS_ERR(dn)) {
1166                         err = PTR_ERR(dn);
1167                         goto done;
1168                 }
1169                 req->r_dentry = dn;  /* may have spliced */
1170                 ihold(in);
1171                 rinfo->head->is_dentry = 1;  /* fool notrace handlers */
1172         }
1173
1174         if (rinfo->head->is_target) {
1175                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1176                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1177
1178                 if (in == NULL || ceph_ino(in) != vino.ino ||
1179                     ceph_snap(in) != vino.snap) {
1180                         in = ceph_get_inode(sb, vino);
1181                         if (IS_ERR(in)) {
1182                                 err = PTR_ERR(in);
1183                                 goto done;
1184                         }
1185                 }
1186                 req->r_target_inode = in;
1187
1188                 err = fill_inode(in,
1189                                  &rinfo->targeti, NULL,
1190                                  session, req->r_request_started,
1191                                  (le32_to_cpu(rinfo->head->result) == 0) ?
1192                                  req->r_fmode : -1,
1193                                  &req->r_caps_reservation);
1194                 if (err < 0) {
1195                         pr_err("fill_inode badness %p %llx.%llx\n",
1196                                in, ceph_vinop(in));
1197                         goto done;
1198                 }
1199         }
1200
1201 done:
1202         dout("fill_trace done err=%d\n", err);
1203         return err;
1204 }
1205
1206 /*
1207  * Prepopulate our cache with readdir results, leases, etc.
1208  */
1209 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1210                                            struct ceph_mds_session *session)
1211 {
1212         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1213         int i, err = 0;
1214
1215         for (i = 0; i < rinfo->dir_nr; i++) {
1216                 struct ceph_vino vino;
1217                 struct inode *in;
1218                 int rc;
1219
1220                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1221                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1222
1223                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1224                 if (IS_ERR(in)) {
1225                         err = PTR_ERR(in);
1226                         dout("new_inode badness got %d\n", err);
1227                         continue;
1228                 }
1229                 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1230                                 req->r_request_started, -1,
1231                                 &req->r_caps_reservation);
1232                 if (rc < 0) {
1233                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1234                         err = rc;
1235                         continue;
1236                 }
1237         }
1238
1239         return err;
1240 }
1241
1242 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1243                              struct ceph_mds_session *session)
1244 {
1245         struct dentry *parent = req->r_dentry;
1246         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1247         struct qstr dname;
1248         struct dentry *dn;
1249         struct inode *in;
1250         int err = 0, i;
1251         struct inode *snapdir = NULL;
1252         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1253         u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1254         struct ceph_dentry_info *di;
1255
1256         if (req->r_aborted)
1257                 return readdir_prepopulate_inodes_only(req, session);
1258
1259         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1260                 snapdir = ceph_get_snapdir(parent->d_inode);
1261                 parent = d_find_alias(snapdir);
1262                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1263                      rinfo->dir_nr, parent);
1264         } else {
1265                 dout("readdir_prepopulate %d items under dn %p\n",
1266                      rinfo->dir_nr, parent);
1267                 if (rinfo->dir_dir)
1268                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1269         }
1270
1271         for (i = 0; i < rinfo->dir_nr; i++) {
1272                 struct ceph_vino vino;
1273
1274                 dname.name = rinfo->dir_dname[i];
1275                 dname.len = rinfo->dir_dname_len[i];
1276                 dname.hash = full_name_hash(dname.name, dname.len);
1277
1278                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1279                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1280
1281 retry_lookup:
1282                 dn = d_lookup(parent, &dname);
1283                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1284                      parent, dname.len, dname.name, dn);
1285
1286                 if (!dn) {
1287                         dn = d_alloc(parent, &dname);
1288                         dout("d_alloc %p '%.*s' = %p\n", parent,
1289                              dname.len, dname.name, dn);
1290                         if (dn == NULL) {
1291                                 dout("d_alloc badness\n");
1292                                 err = -ENOMEM;
1293                                 goto out;
1294                         }
1295                         err = ceph_init_dentry(dn);
1296                         if (err < 0) {
1297                                 dput(dn);
1298                                 goto out;
1299                         }
1300                 } else if (dn->d_inode &&
1301                            (ceph_ino(dn->d_inode) != vino.ino ||
1302                             ceph_snap(dn->d_inode) != vino.snap)) {
1303                         dout(" dn %p points to wrong inode %p\n",
1304                              dn, dn->d_inode);
1305                         d_delete(dn);
1306                         dput(dn);
1307                         goto retry_lookup;
1308                 } else {
1309                         /* reorder parent's d_subdirs */
1310                         spin_lock(&parent->d_lock);
1311                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1312                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1313                         spin_unlock(&dn->d_lock);
1314                         spin_unlock(&parent->d_lock);
1315                 }
1316
1317                 di = dn->d_fsdata;
1318                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1319
1320                 /* inode */
1321                 if (dn->d_inode) {
1322                         in = dn->d_inode;
1323                 } else {
1324                         in = ceph_get_inode(parent->d_sb, vino);
1325                         if (IS_ERR(in)) {
1326                                 dout("new_inode badness\n");
1327                                 d_drop(dn);
1328                                 dput(dn);
1329                                 err = PTR_ERR(in);
1330                                 goto out;
1331                         }
1332                         dn = splice_dentry(dn, in, NULL, false);
1333                         if (IS_ERR(dn))
1334                                 dn = NULL;
1335                 }
1336
1337                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1338                                req->r_request_started, -1,
1339                                &req->r_caps_reservation) < 0) {
1340                         pr_err("fill_inode badness on %p\n", in);
1341                         goto next_item;
1342                 }
1343                 if (dn)
1344                         update_dentry_lease(dn, rinfo->dir_dlease[i],
1345                                             req->r_session,
1346                                             req->r_request_started);
1347 next_item:
1348                 if (dn)
1349                         dput(dn);
1350         }
1351         req->r_did_prepopulate = true;
1352
1353 out:
1354         if (snapdir) {
1355                 iput(snapdir);
1356                 dput(parent);
1357         }
1358         dout("readdir_prepopulate done\n");
1359         return err;
1360 }
1361
1362 int ceph_inode_set_size(struct inode *inode, loff_t size)
1363 {
1364         struct ceph_inode_info *ci = ceph_inode(inode);
1365         int ret = 0;
1366
1367         spin_lock(&ci->i_ceph_lock);
1368         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1369         inode->i_size = size;
1370         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1371
1372         /* tell the MDS if we are approaching max_size */
1373         if ((size << 1) >= ci->i_max_size &&
1374             (ci->i_reported_size << 1) < ci->i_max_size)
1375                 ret = 1;
1376
1377         spin_unlock(&ci->i_ceph_lock);
1378         return ret;
1379 }
1380
1381 /*
1382  * Write back inode data in a worker thread.  (This can't be done
1383  * in the message handler context.)
1384  */
1385 void ceph_queue_writeback(struct inode *inode)
1386 {
1387         ihold(inode);
1388         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1389                        &ceph_inode(inode)->i_wb_work)) {
1390                 dout("ceph_queue_writeback %p\n", inode);
1391         } else {
1392                 dout("ceph_queue_writeback %p failed\n", inode);
1393                 iput(inode);
1394         }
1395 }
1396
1397 static void ceph_writeback_work(struct work_struct *work)
1398 {
1399         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1400                                                   i_wb_work);
1401         struct inode *inode = &ci->vfs_inode;
1402
1403         dout("writeback %p\n", inode);
1404         filemap_fdatawrite(&inode->i_data);
1405         iput(inode);
1406 }
1407
1408 /*
1409  * queue an async invalidation
1410  */
1411 void ceph_queue_invalidate(struct inode *inode)
1412 {
1413         ihold(inode);
1414         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1415                        &ceph_inode(inode)->i_pg_inv_work)) {
1416                 dout("ceph_queue_invalidate %p\n", inode);
1417         } else {
1418                 dout("ceph_queue_invalidate %p failed\n", inode);
1419                 iput(inode);
1420         }
1421 }
1422
1423 /*
1424  * Invalidate inode pages in a worker thread.  (This can't be done
1425  * in the message handler context.)
1426  */
1427 static void ceph_invalidate_work(struct work_struct *work)
1428 {
1429         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1430                                                   i_pg_inv_work);
1431         struct inode *inode = &ci->vfs_inode;
1432         u32 orig_gen;
1433         int check = 0;
1434
1435         mutex_lock(&ci->i_truncate_mutex);
1436         spin_lock(&ci->i_ceph_lock);
1437         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1438              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1439         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1440                 /* nevermind! */
1441                 spin_unlock(&ci->i_ceph_lock);
1442                 mutex_unlock(&ci->i_truncate_mutex);
1443                 goto out;
1444         }
1445         orig_gen = ci->i_rdcache_gen;
1446         spin_unlock(&ci->i_ceph_lock);
1447
1448         truncate_inode_pages(inode->i_mapping, 0);
1449
1450         spin_lock(&ci->i_ceph_lock);
1451         if (orig_gen == ci->i_rdcache_gen &&
1452             orig_gen == ci->i_rdcache_revoking) {
1453                 dout("invalidate_pages %p gen %d successful\n", inode,
1454                      ci->i_rdcache_gen);
1455                 ci->i_rdcache_revoking--;
1456                 check = 1;
1457         } else {
1458                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1459                      inode, orig_gen, ci->i_rdcache_gen,
1460                      ci->i_rdcache_revoking);
1461         }
1462         spin_unlock(&ci->i_ceph_lock);
1463         mutex_unlock(&ci->i_truncate_mutex);
1464
1465         if (check)
1466                 ceph_check_caps(ci, 0, NULL);
1467 out:
1468         iput(inode);
1469 }
1470
1471
1472 /*
1473  * called by trunc_wq;
1474  *
1475  * We also truncate in a separate thread as well.
1476  */
1477 static void ceph_vmtruncate_work(struct work_struct *work)
1478 {
1479         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1480                                                   i_vmtruncate_work);
1481         struct inode *inode = &ci->vfs_inode;
1482
1483         dout("vmtruncate_work %p\n", inode);
1484         __ceph_do_pending_vmtruncate(inode);
1485         iput(inode);
1486 }
1487
1488 /*
1489  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1490  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1491  */
1492 void ceph_queue_vmtruncate(struct inode *inode)
1493 {
1494         struct ceph_inode_info *ci = ceph_inode(inode);
1495
1496         ihold(inode);
1497
1498         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1499                        &ci->i_vmtruncate_work)) {
1500                 dout("ceph_queue_vmtruncate %p\n", inode);
1501         } else {
1502                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1503                      inode, ci->i_truncate_pending);
1504                 iput(inode);
1505         }
1506 }
1507
1508 /*
1509  * Make sure any pending truncation is applied before doing anything
1510  * that may depend on it.
1511  */
1512 void __ceph_do_pending_vmtruncate(struct inode *inode)
1513 {
1514         struct ceph_inode_info *ci = ceph_inode(inode);
1515         u64 to;
1516         int wrbuffer_refs, finish = 0;
1517
1518         mutex_lock(&ci->i_truncate_mutex);
1519 retry:
1520         spin_lock(&ci->i_ceph_lock);
1521         if (ci->i_truncate_pending == 0) {
1522                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1523                 spin_unlock(&ci->i_ceph_lock);
1524                 mutex_unlock(&ci->i_truncate_mutex);
1525                 return;
1526         }
1527
1528         /*
1529          * make sure any dirty snapped pages are flushed before we
1530          * possibly truncate them.. so write AND block!
1531          */
1532         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1533                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1534                      inode);
1535                 spin_unlock(&ci->i_ceph_lock);
1536                 filemap_write_and_wait_range(&inode->i_data, 0,
1537                                              inode->i_sb->s_maxbytes);
1538                 goto retry;
1539         }
1540
1541         /* there should be no reader or writer */
1542         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1543
1544         to = ci->i_truncate_size;
1545         wrbuffer_refs = ci->i_wrbuffer_ref;
1546         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1547              ci->i_truncate_pending, to);
1548         spin_unlock(&ci->i_ceph_lock);
1549
1550         truncate_inode_pages(inode->i_mapping, to);
1551
1552         spin_lock(&ci->i_ceph_lock);
1553         if (to == ci->i_truncate_size) {
1554                 ci->i_truncate_pending = 0;
1555                 finish = 1;
1556         }
1557         spin_unlock(&ci->i_ceph_lock);
1558         if (!finish)
1559                 goto retry;
1560
1561         mutex_unlock(&ci->i_truncate_mutex);
1562
1563         if (wrbuffer_refs == 0)
1564                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1565
1566         wake_up_all(&ci->i_cap_wq);
1567 }
1568
1569 /*
1570  * symlinks
1571  */
1572 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1573 {
1574         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1575         nd_set_link(nd, ci->i_symlink);
1576         return NULL;
1577 }
1578
1579 static const struct inode_operations ceph_symlink_iops = {
1580         .readlink = generic_readlink,
1581         .follow_link = ceph_sym_follow_link,
1582         .setattr = ceph_setattr,
1583         .getattr = ceph_getattr,
1584         .setxattr = ceph_setxattr,
1585         .getxattr = ceph_getxattr,
1586         .listxattr = ceph_listxattr,
1587         .removexattr = ceph_removexattr,
1588 };
1589
1590 /*
1591  * setattr
1592  */
1593 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1594 {
1595         struct inode *inode = dentry->d_inode;
1596         struct ceph_inode_info *ci = ceph_inode(inode);
1597         struct inode *parent_inode;
1598         const unsigned int ia_valid = attr->ia_valid;
1599         struct ceph_mds_request *req;
1600         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1601         int issued;
1602         int release = 0, dirtied = 0;
1603         int mask = 0;
1604         int err = 0;
1605         int inode_dirty_flags = 0;
1606
1607         if (ceph_snap(inode) != CEPH_NOSNAP)
1608                 return -EROFS;
1609
1610         err = inode_change_ok(inode, attr);
1611         if (err != 0)
1612                 return err;
1613
1614         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1615                                        USE_AUTH_MDS);
1616         if (IS_ERR(req))
1617                 return PTR_ERR(req);
1618
1619         spin_lock(&ci->i_ceph_lock);
1620         issued = __ceph_caps_issued(ci, NULL);
1621         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1622
1623         if (ia_valid & ATTR_UID) {
1624                 dout("setattr %p uid %d -> %d\n", inode,
1625                      from_kuid(&init_user_ns, inode->i_uid),
1626                      from_kuid(&init_user_ns, attr->ia_uid));
1627                 if (issued & CEPH_CAP_AUTH_EXCL) {
1628                         inode->i_uid = attr->ia_uid;
1629                         dirtied |= CEPH_CAP_AUTH_EXCL;
1630                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1631                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1632                         req->r_args.setattr.uid = cpu_to_le32(
1633                                 from_kuid(&init_user_ns, attr->ia_uid));
1634                         mask |= CEPH_SETATTR_UID;
1635                         release |= CEPH_CAP_AUTH_SHARED;
1636                 }
1637         }
1638         if (ia_valid & ATTR_GID) {
1639                 dout("setattr %p gid %d -> %d\n", inode,
1640                      from_kgid(&init_user_ns, inode->i_gid),
1641                      from_kgid(&init_user_ns, attr->ia_gid));
1642                 if (issued & CEPH_CAP_AUTH_EXCL) {
1643                         inode->i_gid = attr->ia_gid;
1644                         dirtied |= CEPH_CAP_AUTH_EXCL;
1645                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1646                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1647                         req->r_args.setattr.gid = cpu_to_le32(
1648                                 from_kgid(&init_user_ns, attr->ia_gid));
1649                         mask |= CEPH_SETATTR_GID;
1650                         release |= CEPH_CAP_AUTH_SHARED;
1651                 }
1652         }
1653         if (ia_valid & ATTR_MODE) {
1654                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1655                      attr->ia_mode);
1656                 if (issued & CEPH_CAP_AUTH_EXCL) {
1657                         inode->i_mode = attr->ia_mode;
1658                         dirtied |= CEPH_CAP_AUTH_EXCL;
1659                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1660                            attr->ia_mode != inode->i_mode) {
1661                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1662                         mask |= CEPH_SETATTR_MODE;
1663                         release |= CEPH_CAP_AUTH_SHARED;
1664                 }
1665         }
1666
1667         if (ia_valid & ATTR_ATIME) {
1668                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1669                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1670                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1671                 if (issued & CEPH_CAP_FILE_EXCL) {
1672                         ci->i_time_warp_seq++;
1673                         inode->i_atime = attr->ia_atime;
1674                         dirtied |= CEPH_CAP_FILE_EXCL;
1675                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1676                            timespec_compare(&inode->i_atime,
1677                                             &attr->ia_atime) < 0) {
1678                         inode->i_atime = attr->ia_atime;
1679                         dirtied |= CEPH_CAP_FILE_WR;
1680                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1681                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1682                         ceph_encode_timespec(&req->r_args.setattr.atime,
1683                                              &attr->ia_atime);
1684                         mask |= CEPH_SETATTR_ATIME;
1685                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1686                                 CEPH_CAP_FILE_WR;
1687                 }
1688         }
1689         if (ia_valid & ATTR_MTIME) {
1690                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1691                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1692                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1693                 if (issued & CEPH_CAP_FILE_EXCL) {
1694                         ci->i_time_warp_seq++;
1695                         inode->i_mtime = attr->ia_mtime;
1696                         dirtied |= CEPH_CAP_FILE_EXCL;
1697                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1698                            timespec_compare(&inode->i_mtime,
1699                                             &attr->ia_mtime) < 0) {
1700                         inode->i_mtime = attr->ia_mtime;
1701                         dirtied |= CEPH_CAP_FILE_WR;
1702                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1703                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1704                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1705                                              &attr->ia_mtime);
1706                         mask |= CEPH_SETATTR_MTIME;
1707                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1708                                 CEPH_CAP_FILE_WR;
1709                 }
1710         }
1711         if (ia_valid & ATTR_SIZE) {
1712                 dout("setattr %p size %lld -> %lld\n", inode,
1713                      inode->i_size, attr->ia_size);
1714                 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1715                         err = -EINVAL;
1716                         goto out;
1717                 }
1718                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1719                     attr->ia_size > inode->i_size) {
1720                         inode->i_size = attr->ia_size;
1721                         inode->i_blocks =
1722                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1723                         inode->i_ctime = attr->ia_ctime;
1724                         ci->i_reported_size = attr->ia_size;
1725                         dirtied |= CEPH_CAP_FILE_EXCL;
1726                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1727                            attr->ia_size != inode->i_size) {
1728                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1729                         req->r_args.setattr.old_size =
1730                                 cpu_to_le64(inode->i_size);
1731                         mask |= CEPH_SETATTR_SIZE;
1732                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1733                                 CEPH_CAP_FILE_WR;
1734                 }
1735         }
1736
1737         /* these do nothing */
1738         if (ia_valid & ATTR_CTIME) {
1739                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1740                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1741                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1742                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1743                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1744                      only ? "ctime only" : "ignored");
1745                 inode->i_ctime = attr->ia_ctime;
1746                 if (only) {
1747                         /*
1748                          * if kernel wants to dirty ctime but nothing else,
1749                          * we need to choose a cap to dirty under, or do
1750                          * a almost-no-op setattr
1751                          */
1752                         if (issued & CEPH_CAP_AUTH_EXCL)
1753                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1754                         else if (issued & CEPH_CAP_FILE_EXCL)
1755                                 dirtied |= CEPH_CAP_FILE_EXCL;
1756                         else if (issued & CEPH_CAP_XATTR_EXCL)
1757                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1758                         else
1759                                 mask |= CEPH_SETATTR_CTIME;
1760                 }
1761         }
1762         if (ia_valid & ATTR_FILE)
1763                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1764
1765         if (dirtied) {
1766                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1767                 inode->i_ctime = CURRENT_TIME;
1768         }
1769
1770         release &= issued;
1771         spin_unlock(&ci->i_ceph_lock);
1772
1773         if (inode_dirty_flags)
1774                 __mark_inode_dirty(inode, inode_dirty_flags);
1775
1776         if (mask) {
1777                 req->r_inode = inode;
1778                 ihold(inode);
1779                 req->r_inode_drop = release;
1780                 req->r_args.setattr.mask = cpu_to_le32(mask);
1781                 req->r_num_caps = 1;
1782                 parent_inode = ceph_get_dentry_parent_inode(dentry);
1783                 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1784                 iput(parent_inode);
1785         }
1786         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1787              ceph_cap_string(dirtied), mask);
1788
1789         ceph_mdsc_put_request(req);
1790         if (mask & CEPH_SETATTR_SIZE)
1791                 __ceph_do_pending_vmtruncate(inode);
1792         return err;
1793 out:
1794         spin_unlock(&ci->i_ceph_lock);
1795         ceph_mdsc_put_request(req);
1796         return err;
1797 }
1798
1799 /*
1800  * Verify that we have a lease on the given mask.  If not,
1801  * do a getattr against an mds.
1802  */
1803 int ceph_do_getattr(struct inode *inode, int mask)
1804 {
1805         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1806         struct ceph_mds_client *mdsc = fsc->mdsc;
1807         struct ceph_mds_request *req;
1808         int err;
1809
1810         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1811                 dout("do_getattr inode %p SNAPDIR\n", inode);
1812                 return 0;
1813         }
1814
1815         dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1816         if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1817                 return 0;
1818
1819         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1820         if (IS_ERR(req))
1821                 return PTR_ERR(req);
1822         req->r_inode = inode;
1823         ihold(inode);
1824         req->r_num_caps = 1;
1825         req->r_args.getattr.mask = cpu_to_le32(mask);
1826         err = ceph_mdsc_do_request(mdsc, NULL, req);
1827         ceph_mdsc_put_request(req);
1828         dout("do_getattr result=%d\n", err);
1829         return err;
1830 }
1831
1832
1833 /*
1834  * Check inode permissions.  We verify we have a valid value for
1835  * the AUTH cap, then call the generic handler.
1836  */
1837 int ceph_permission(struct inode *inode, int mask)
1838 {
1839         int err;
1840
1841         if (mask & MAY_NOT_BLOCK)
1842                 return -ECHILD;
1843
1844         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1845
1846         if (!err)
1847                 err = generic_permission(inode, mask);
1848         return err;
1849 }
1850
1851 /*
1852  * Get all attributes.  Hopefully somedata we'll have a statlite()
1853  * and can limit the fields we require to be accurate.
1854  */
1855 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1856                  struct kstat *stat)
1857 {
1858         struct inode *inode = dentry->d_inode;
1859         struct ceph_inode_info *ci = ceph_inode(inode);
1860         int err;
1861
1862         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1863         if (!err) {
1864                 generic_fillattr(inode, stat);
1865                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1866                 if (ceph_snap(inode) != CEPH_NOSNAP)
1867                         stat->dev = ceph_snap(inode);
1868                 else
1869                         stat->dev = 0;
1870                 if (S_ISDIR(inode->i_mode)) {
1871                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1872                                                 RBYTES))
1873                                 stat->size = ci->i_rbytes;
1874                         else
1875                                 stat->size = ci->i_files + ci->i_subdirs;
1876                         stat->blocks = 0;
1877                         stat->blksize = 65536;
1878                 }
1879         }
1880         return err;
1881 }