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         .get_acl = ceph_get_acl,
  99 };
 100
 101
 102 /*
 103  * We use a 'frag tree' to keep track of the MDS's directory fragments
 104  * for a given inode (usually there is just a single fragment).  We
 105  * need to know when a child frag is delegated to a new MDS, or when
 106  * it is flagged as replicated, so we can direct our requests
 107  * accordingly.
 108  */
 109
 110 /*
 111  * find/create a frag in the tree
 112  */
 113 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 114                                                     u32 f)
 115 {
 116         struct rb_node **p;
 117         struct rb_node *parent = NULL;
 118         struct ceph_inode_frag *frag;
 119         int c;
 120
 121         p = &ci->i_fragtree.rb_node;
 122         while (*p) {
 123                 parent = *p;
 124                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 125                 c = ceph_frag_compare(f, frag->frag);
 126                 if (c < 0)
 127                         p = &(*p)->rb_left;
 128                 else if (c > 0)
 129                         p = &(*p)->rb_right;
 130                 else
 131                         return frag;
 132         }
 133
 134         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 135         if (!frag) {
 136                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
 137                        "frag %x\n", &ci->vfs_inode,
 138                        ceph_vinop(&ci->vfs_inode), f);
 139                 return ERR_PTR(-ENOMEM);
 140         }
 141         frag->frag = f;
 142         frag->split_by = 0;
 143         frag->mds = -1;
 144         frag->ndist = 0;
 145
 146         rb_link_node(&frag->node, parent, p);
 147         rb_insert_color(&frag->node, &ci->i_fragtree);
 148
 149         dout("get_or_create_frag added %llx.%llx frag %x\n",
 150              ceph_vinop(&ci->vfs_inode), f);
 151         return frag;
 152 }
 153
 154 /*
 155  * find a specific frag @f
 156  */
 157 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 158 {
 159         struct rb_node *n = ci->i_fragtree.rb_node;
 160
 161         while (n) {
 162                 struct ceph_inode_frag *frag =
 163                         rb_entry(n, struct ceph_inode_frag, node);
 164                 int c = ceph_frag_compare(f, frag->frag);
 165                 if (c < 0)
 166                         n = n->rb_left;
 167                 else if (c > 0)
 168                         n = n->rb_right;
 169                 else
 170                         return frag;
 171         }
 172         return NULL;
 173 }
 174
 175 /*
 176  * Choose frag containing the given value @v.  If @pfrag is
 177  * specified, copy the frag delegation info to the caller if
 178  * it is present.
 179  */
 180 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 181                      struct ceph_inode_frag *pfrag,
 182                      int *found)
 183 {
 184         u32 t = ceph_frag_make(0, 0);
 185         struct ceph_inode_frag *frag;
 186         unsigned nway, i;
 187         u32 n;
 188
 189         if (found)
 190                 *found = 0;
 191
 192         mutex_lock(&ci->i_fragtree_mutex);
 193         while (1) {
 194                 WARN_ON(!ceph_frag_contains_value(t, v));
 195                 frag = __ceph_find_frag(ci, t);
 196                 if (!frag)
 197                         break; /* t is a leaf */
 198                 if (frag->split_by == 0) {
 199                         if (pfrag)
 200                                 memcpy(pfrag, frag, sizeof(*pfrag));
 201                         if (found)
 202                                 *found = 1;
 203                         break;
 204                 }
 205
 206                 /* choose child */
 207                 nway = 1 << frag->split_by;
 208                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 209                      frag->split_by, nway);
 210                 for (i = 0; i < nway; i++) {
 211                         n = ceph_frag_make_child(t, frag->split_by, i);
 212                         if (ceph_frag_contains_value(n, v)) {
 213                                 t = n;
 214                                 break;
 215                         }
 216                 }
 217                 BUG_ON(i == nway);
 218         }
 219         dout("choose_frag(%x) = %x\n", v, t);
 220
 221         mutex_unlock(&ci->i_fragtree_mutex);
 222         return t;
 223 }
 224
 225 /*
 226  * Process dirfrag (delegation) info from the mds.  Include leaf
 227  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 228  * branches/splits are included in i_fragtree)
 229  */
 230 static int ceph_fill_dirfrag(struct inode *inode,
 231                              struct ceph_mds_reply_dirfrag *dirinfo)
 232 {
 233         struct ceph_inode_info *ci = ceph_inode(inode);
 234         struct ceph_inode_frag *frag;
 235         u32 id = le32_to_cpu(dirinfo->frag);
 236         int mds = le32_to_cpu(dirinfo->auth);
 237         int ndist = le32_to_cpu(dirinfo->ndist);
 238         int i;
 239         int err = 0;
 240
 241         mutex_lock(&ci->i_fragtree_mutex);
 242         if (ndist == 0) {
 243                 /* no delegation info needed. */
 244                 frag = __ceph_find_frag(ci, id);
 245                 if (!frag)
 246                         goto out;
 247                 if (frag->split_by == 0) {
 248                         /* tree leaf, remove */
 249                         dout("fill_dirfrag removed %llx.%llx frag %x"
 250                              " (no ref)\n", ceph_vinop(inode), id);
 251                         rb_erase(&frag->node, &ci->i_fragtree);
 252                         kfree(frag);
 253                 } else {
 254                         /* tree branch, keep and clear */
 255                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 256                              " referral\n", ceph_vinop(inode), id);
 257                         frag->mds = -1;
 258                         frag->ndist = 0;
 259                 }
 260                 goto out;
 261         }
 262
 263
 264         /* find/add this frag to store mds delegation info */
 265         frag = __get_or_create_frag(ci, id);
 266         if (IS_ERR(frag)) {
 267                 /* this is not the end of the world; we can continue
 268                    with bad/inaccurate delegation info */
 269                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 270                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 271                 err = -ENOMEM;
 272                 goto out;
 273         }
 274
 275         frag->mds = mds;
 276         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 277         for (i = 0; i < frag->ndist; i++)
 278                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 279         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 280              ceph_vinop(inode), frag->frag, frag->ndist);
 281
 282 out:
 283         mutex_unlock(&ci->i_fragtree_mutex);
 284         return err;
 285 }
 286
 287
 288 /*
 289  * initialize a newly allocated inode.
 290  */
 291 struct inode *ceph_alloc_inode(struct super_block *sb)
 292 {
 293         struct ceph_inode_info *ci;
 294         int i;
 295
 296         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 297         if (!ci)
 298                 return NULL;
 299
 300         dout("alloc_inode %p\n", &ci->vfs_inode);
 301
 302         spin_lock_init(&ci->i_ceph_lock);
 303
 304         ci->i_version = 0;
 305         ci->i_time_warp_seq = 0;
 306         ci->i_ceph_flags = 0;
 307         atomic_set(&ci->i_release_count, 1);
 308         atomic_set(&ci->i_complete_count, 0);
 309         ci->i_symlink = NULL;
 310
 311         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 312
 313         ci->i_fragtree = RB_ROOT;
 314         mutex_init(&ci->i_fragtree_mutex);
 315
 316         ci->i_xattrs.blob = NULL;
 317         ci->i_xattrs.prealloc_blob = NULL;
 318         ci->i_xattrs.dirty = false;
 319         ci->i_xattrs.index = RB_ROOT;
 320         ci->i_xattrs.count = 0;
 321         ci->i_xattrs.names_size = 0;
 322         ci->i_xattrs.vals_size = 0;
 323         ci->i_xattrs.version = 0;
 324         ci->i_xattrs.index_version = 0;
 325
 326         ci->i_caps = RB_ROOT;
 327         ci->i_auth_cap = NULL;
 328         ci->i_dirty_caps = 0;
 329         ci->i_flushing_caps = 0;
 330         INIT_LIST_HEAD(&ci->i_dirty_item);
 331         INIT_LIST_HEAD(&ci->i_flushing_item);
 332         ci->i_cap_flush_seq = 0;
 333         ci->i_cap_flush_last_tid = 0;
 334         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
 335         init_waitqueue_head(&ci->i_cap_wq);
 336         ci->i_hold_caps_min = 0;
 337         ci->i_hold_caps_max = 0;
 338         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 339         INIT_LIST_HEAD(&ci->i_cap_snaps);
 340         ci->i_head_snapc = NULL;
 341         ci->i_snap_caps = 0;
 342         ci->i_cap_exporting_issued = 0;
 343
 344         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
 345                 ci->i_nr_by_mode[i] = 0;
 346
 347         mutex_init(&ci->i_truncate_mutex);
 348         ci->i_truncate_seq = 0;
 349         ci->i_truncate_size = 0;
 350         ci->i_truncate_pending = 0;
 351
 352         ci->i_max_size = 0;
 353         ci->i_reported_size = 0;
 354         ci->i_wanted_max_size = 0;
 355         ci->i_requested_max_size = 0;
 356
 357         ci->i_pin_ref = 0;
 358         ci->i_rd_ref = 0;
 359         ci->i_rdcache_ref = 0;
 360         ci->i_wr_ref = 0;
 361         ci->i_wb_ref = 0;
 362         ci->i_wrbuffer_ref = 0;
 363         ci->i_wrbuffer_ref_head = 0;
 364         ci->i_shared_gen = 0;
 365         ci->i_rdcache_gen = 0;
 366         ci->i_rdcache_revoking = 0;
 367
 368         INIT_LIST_HEAD(&ci->i_unsafe_writes);
 369         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 370         spin_lock_init(&ci->i_unsafe_lock);
 371
 372         ci->i_snap_realm = NULL;
 373         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 374         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 375
 376         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
 377         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
 378
 379         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
 380
 381         ceph_fscache_inode_init(ci);
 382
 383         return &ci->vfs_inode;
 384 }
 385
 386 static void ceph_i_callback(struct rcu_head *head)
 387 {
 388         struct inode *inode = container_of(head, struct inode, i_rcu);
 389         struct ceph_inode_info *ci = ceph_inode(inode);
 390
 391         kmem_cache_free(ceph_inode_cachep, ci);
 392 }
 393
 394 void ceph_destroy_inode(struct inode *inode)
 395 {
 396         struct ceph_inode_info *ci = ceph_inode(inode);
 397         struct ceph_inode_frag *frag;
 398         struct rb_node *n;
 399
 400         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 401
 402         ceph_fscache_unregister_inode_cookie(ci);
 403
 404         ceph_queue_caps_release(inode);
 405
 406         /*
 407          * we may still have a snap_realm reference if there are stray
 408          * caps in i_cap_exporting_issued or i_snap_caps.
 409          */
 410         if (ci->i_snap_realm) {
 411                 struct ceph_mds_client *mdsc =
 412                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
 413                 struct ceph_snap_realm *realm = ci->i_snap_realm;
 414
 415                 dout(" dropping residual ref to snap realm %p\n", realm);
 416                 spin_lock(&realm->inodes_with_caps_lock);
 417                 list_del_init(&ci->i_snap_realm_item);
 418                 spin_unlock(&realm->inodes_with_caps_lock);
 419                 ceph_put_snap_realm(mdsc, realm);
 420         }
 421
 422         kfree(ci->i_symlink);
 423         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 424                 frag = rb_entry(n, struct ceph_inode_frag, node);
 425                 rb_erase(n, &ci->i_fragtree);
 426                 kfree(frag);
 427         }
 428
 429         __ceph_destroy_xattrs(ci);
 430         if (ci->i_xattrs.blob)
 431                 ceph_buffer_put(ci->i_xattrs.blob);
 432         if (ci->i_xattrs.prealloc_blob)
 433                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 434
 435         call_rcu(&inode->i_rcu, ceph_i_callback);
 436 }
 437
 438 int ceph_drop_inode(struct inode *inode)
 439 {
 440         /*
 441          * Positve dentry and corresponding inode are always accompanied
 442          * in MDS reply. So no need to keep inode in the cache after
 443          * dropping all its aliases.
 444          */
 445         return 1;
 446 }
 447
 448 /*
 449  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 450  * careful because either the client or MDS may have more up to date
 451  * info, depending on which capabilities are held, and whether
 452  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 453  * and size are monotonically increasing, except when utimes() or
 454  * truncate() increments the corresponding _seq values.)
 455  */
 456 int ceph_fill_file_size(struct inode *inode, int issued,
 457                         u32 truncate_seq, u64 truncate_size, u64 size)
 458 {
 459         struct ceph_inode_info *ci = ceph_inode(inode);
 460         int queue_trunc = 0;
 461
 462         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 463             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
 464                 dout("size %lld -> %llu\n", inode->i_size, size);
 465                 inode->i_size = size;
 466                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
 467                 ci->i_reported_size = size;
 468                 if (truncate_seq != ci->i_truncate_seq) {
 469                         dout("truncate_seq %u -> %u\n",
 470                              ci->i_truncate_seq, truncate_seq);
 471                         ci->i_truncate_seq = truncate_seq;
 472
 473                         /* the MDS should have revoked these caps */
 474                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 475                                                CEPH_CAP_FILE_RD |
 476                                                CEPH_CAP_FILE_WR |
 477                                                CEPH_CAP_FILE_LAZYIO));
 478                         /*
 479                          * If we hold relevant caps, or in the case where we're
 480                          * not the only client referencing this file and we
 481                          * don't hold those caps, then we need to check whether
 482                          * the file is either opened or mmaped
 483                          */
 484                         if ((issued & (CEPH_CAP_FILE_CACHE|
 485                                        CEPH_CAP_FILE_BUFFER)) ||
 486                             mapping_mapped(inode->i_mapping) ||
 487                             __ceph_caps_file_wanted(ci)) {
 488                                 ci->i_truncate_pending++;
 489                                 queue_trunc = 1;
 490                         }
 491                 }
 492         }
 493         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 494             ci->i_truncate_size != truncate_size) {
 495                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 496                      truncate_size);
 497                 ci->i_truncate_size = truncate_size;
 498         }
 499
 500         if (queue_trunc)
 501                 ceph_fscache_invalidate(inode);
 502
 503         return queue_trunc;
 504 }
 505
 506 void ceph_fill_file_time(struct inode *inode, int issued,
 507                          u64 time_warp_seq, struct timespec *ctime,
 508                          struct timespec *mtime, struct timespec *atime)
 509 {
 510         struct ceph_inode_info *ci = ceph_inode(inode);
 511         int warn = 0;
 512
 513         if (issued & (CEPH_CAP_FILE_EXCL|
 514                       CEPH_CAP_FILE_WR|
 515                       CEPH_CAP_FILE_BUFFER|
 516                       CEPH_CAP_AUTH_EXCL|
 517                       CEPH_CAP_XATTR_EXCL)) {
 518                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
 519                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
 520                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 521                              ctime->tv_sec, ctime->tv_nsec);
 522                         inode->i_ctime = *ctime;
 523                 }
 524                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 525                         /* the MDS did a utimes() */
 526                         dout("mtime %ld.%09ld -> %ld.%09ld "
 527                              "tw %d -> %d\n",
 528                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 529                              mtime->tv_sec, mtime->tv_nsec,
 530                              ci->i_time_warp_seq, (int)time_warp_seq);
 531
 532                         inode->i_mtime = *mtime;
 533                         inode->i_atime = *atime;
 534                         ci->i_time_warp_seq = time_warp_seq;
 535                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 536                         /* nobody did utimes(); take the max */
 537                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
 538                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
 539                                      inode->i_mtime.tv_sec,
 540                                      inode->i_mtime.tv_nsec,
 541                                      mtime->tv_sec, mtime->tv_nsec);
 542                                 inode->i_mtime = *mtime;
 543                         }
 544                         if (timespec_compare(atime, &inode->i_atime) > 0) {
 545                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
 546                                      inode->i_atime.tv_sec,
 547                                      inode->i_atime.tv_nsec,
 548                                      atime->tv_sec, atime->tv_nsec);
 549                                 inode->i_atime = *atime;
 550                         }
 551                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 552                         /* we did a utimes(); ignore mds values */
 553                 } else {
 554                         warn = 1;
 555                 }
 556         } else {
 557                 /* we have no write|excl caps; whatever the MDS says is true */
 558                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 559                         inode->i_ctime = *ctime;
 560                         inode->i_mtime = *mtime;
 561                         inode->i_atime = *atime;
 562                         ci->i_time_warp_seq = time_warp_seq;
 563                 } else {
 564                         warn = 1;
 565                 }
 566         }
 567         if (warn) /* time_warp_seq shouldn't go backwards */
 568                 dout("%p mds time_warp_seq %llu < %u\n",
 569                      inode, time_warp_seq, ci->i_time_warp_seq);
 570 }
 571
 572 /*
 573  * Populate an inode based on info from mds.  May be called on new or
 574  * existing inodes.
 575  */
 576 static int fill_inode(struct inode *inode,
 577                       struct ceph_mds_reply_info_in *iinfo,
 578                       struct ceph_mds_reply_dirfrag *dirinfo,
 579                       struct ceph_mds_session *session,
 580                       unsigned long ttl_from, int cap_fmode,
 581                       struct ceph_cap_reservation *caps_reservation)
 582 {
 583         struct ceph_mds_reply_inode *info = iinfo->in;
 584         struct ceph_inode_info *ci = ceph_inode(inode);
 585         int i;
 586         int issued = 0, implemented;
 587         struct timespec mtime, atime, ctime;
 588         u32 nsplits;
 589         struct ceph_inode_frag *frag;
 590         struct rb_node *rb_node;
 591         struct ceph_buffer *xattr_blob = NULL;
 592         int err = 0;
 593         int queue_trunc = 0;
 594
 595         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
 596              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 597              ci->i_version);
 598
 599         /*
 600          * prealloc xattr data, if it looks like we'll need it.  only
 601          * if len > 4 (meaning there are actually xattrs; the first 4
 602          * bytes are the xattr count).
 603          */
 604         if (iinfo->xattr_len > 4) {
 605                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 606                 if (!xattr_blob)
 607                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
 608                                iinfo->xattr_len);
 609         }
 610
 611         spin_lock(&ci->i_ceph_lock);
 612
 613         /*
 614          * provided version will be odd if inode value is projected,
 615          * even if stable.  skip the update if we have newer stable
 616          * info (ours>=theirs, e.g. due to racing mds replies), unless
 617          * we are getting projected (unstable) info (in which case the
 618          * version is odd, and we want ours>theirs).
 619          *   us   them
 620          *   2    2     skip
 621          *   3    2     skip
 622          *   3    3     update
 623          */
 624         if (le64_to_cpu(info->version) > 0 &&
 625             (ci->i_version & ~1) >= le64_to_cpu(info->version))
 626                 goto no_change;
 627
 628         issued = __ceph_caps_issued(ci, &implemented);
 629         issued |= implemented | __ceph_caps_dirty(ci);
 630
 631         /* update inode */
 632         ci->i_version = le64_to_cpu(info->version);
 633         inode->i_version++;
 634         inode->i_rdev = le32_to_cpu(info->rdev);
 635
 636         if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
 637                 inode->i_mode = le32_to_cpu(info->mode);
 638                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 639                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 640                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 641                      from_kuid(&init_user_ns, inode->i_uid),
 642                      from_kgid(&init_user_ns, inode->i_gid));
 643         }
 644
 645         if ((issued & CEPH_CAP_LINK_EXCL) == 0)
 646                 set_nlink(inode, le32_to_cpu(info->nlink));
 647
 648         /* be careful with mtime, atime, size */
 649         ceph_decode_timespec(&atime, &info->atime);
 650         ceph_decode_timespec(&mtime, &info->mtime);
 651         ceph_decode_timespec(&ctime, &info->ctime);
 652         queue_trunc = ceph_fill_file_size(inode, issued,
 653                                           le32_to_cpu(info->truncate_seq),
 654                                           le64_to_cpu(info->truncate_size),
 655                                           le64_to_cpu(info->size));
 656         ceph_fill_file_time(inode, issued,
 657                             le32_to_cpu(info->time_warp_seq),
 658                             &ctime, &mtime, &atime);
 659
 660         /* only update max_size on auth cap */
 661         if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 662             ci->i_max_size != le64_to_cpu(info->max_size)) {
 663                 dout("max_size %lld -> %llu\n", ci->i_max_size,
 664                      le64_to_cpu(info->max_size));
 665                 ci->i_max_size = le64_to_cpu(info->max_size);
 666         }
 667
 668         ci->i_layout = info->layout;
 669         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 670
 671         /* xattrs */
 672         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 673         if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
 674             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 675                 if (ci->i_xattrs.blob)
 676                         ceph_buffer_put(ci->i_xattrs.blob);
 677                 ci->i_xattrs.blob = xattr_blob;
 678                 if (xattr_blob)
 679                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 680                                iinfo->xattr_data, iinfo->xattr_len);
 681                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 682                 ceph_forget_all_cached_acls(inode);
 683                 xattr_blob = NULL;
 684         }
 685
 686         inode->i_mapping->a_ops = &ceph_aops;
 687         inode->i_mapping->backing_dev_info =
 688                 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
 689
 690         switch (inode->i_mode & S_IFMT) {
 691         case S_IFIFO:
 692         case S_IFBLK:
 693         case S_IFCHR:
 694         case S_IFSOCK:
 695                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 696                 inode->i_op = &ceph_file_iops;
 697                 break;
 698         case S_IFREG:
 699                 inode->i_op = &ceph_file_iops;
 700                 inode->i_fop = &ceph_file_fops;
 701                 break;
 702         case S_IFLNK:
 703                 inode->i_op = &ceph_symlink_iops;
 704                 if (!ci->i_symlink) {
 705                         u32 symlen = iinfo->symlink_len;
 706                         char *sym;
 707
 708                         spin_unlock(&ci->i_ceph_lock);
 709
 710                         err = -EINVAL;
 711                         if (WARN_ON(symlen != inode->i_size))
 712                                 goto out;
 713
 714                         err = -ENOMEM;
 715                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 716                         if (!sym)
 717                                 goto out;
 718
 719                         spin_lock(&ci->i_ceph_lock);
 720                         if (!ci->i_symlink)
 721                                 ci->i_symlink = sym;
 722                         else
 723                                 kfree(sym); /* lost a race */
 724                 }
 725                 break;
 726         case S_IFDIR:
 727                 inode->i_op = &ceph_dir_iops;
 728                 inode->i_fop = &ceph_dir_fops;
 729
 730                 ci->i_dir_layout = iinfo->dir_layout;
 731
 732                 ci->i_files = le64_to_cpu(info->files);
 733                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 734                 ci->i_rbytes = le64_to_cpu(info->rbytes);
 735                 ci->i_rfiles = le64_to_cpu(info->rfiles);
 736                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 737                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
 738                 break;
 739         default:
 740                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
 741                        ceph_vinop(inode), inode->i_mode);
 742         }
 743
 744         /* set dir completion flag? */
 745         if (S_ISDIR(inode->i_mode) &&
 746             ci->i_files == 0 && ci->i_subdirs == 0 &&
 747             ceph_snap(inode) == CEPH_NOSNAP &&
 748             (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
 749             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
 750             !__ceph_dir_is_complete(ci)) {
 751                 dout(" marking %p complete (empty)\n", inode);
 752                 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count));
 753                 ci->i_max_offset = 2;
 754         }
 755 no_change:
 756         spin_unlock(&ci->i_ceph_lock);
 757
 758         /* queue truncate if we saw i_size decrease */
 759         if (queue_trunc)
 760                 ceph_queue_vmtruncate(inode);
 761
 762         /* populate frag tree */
 763         /* FIXME: move me up, if/when version reflects fragtree changes */
 764         nsplits = le32_to_cpu(info->fragtree.nsplits);
 765         mutex_lock(&ci->i_fragtree_mutex);
 766         rb_node = rb_first(&ci->i_fragtree);
 767         for (i = 0; i < nsplits; i++) {
 768                 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
 769                 frag = NULL;
 770                 while (rb_node) {
 771                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 772                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 773                                 if (frag->frag != id)
 774                                         frag = NULL;
 775                                 else
 776                                         rb_node = rb_next(rb_node);
 777                                 break;
 778                         }
 779                         rb_node = rb_next(rb_node);
 780                         rb_erase(&frag->node, &ci->i_fragtree);
 781                         kfree(frag);
 782                         frag = NULL;
 783                 }
 784                 if (!frag) {
 785                         frag = __get_or_create_frag(ci, id);
 786                         if (IS_ERR(frag))
 787                                 continue;
 788                 }
 789                 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
 790                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 791         }
 792         while (rb_node) {
 793                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 794                 rb_node = rb_next(rb_node);
 795                 rb_erase(&frag->node, &ci->i_fragtree);
 796                 kfree(frag);
 797         }
 798         mutex_unlock(&ci->i_fragtree_mutex);
 799
 800         /* were we issued a capability? */
 801         if (info->cap.caps) {
 802                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 803                         ceph_add_cap(inode, session,
 804                                      le64_to_cpu(info->cap.cap_id),
 805                                      cap_fmode,
 806                                      le32_to_cpu(info->cap.caps),
 807                                      le32_to_cpu(info->cap.wanted),
 808                                      le32_to_cpu(info->cap.seq),
 809                                      le32_to_cpu(info->cap.mseq),
 810                                      le64_to_cpu(info->cap.realm),
 811                                      info->cap.flags,
 812                                      caps_reservation);
 813                 } else {
 814                         spin_lock(&ci->i_ceph_lock);
 815                         dout(" %p got snap_caps %s\n", inode,
 816                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
 817                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
 818                         if (cap_fmode >= 0)
 819                                 __ceph_get_fmode(ci, cap_fmode);
 820                         spin_unlock(&ci->i_ceph_lock);
 821                 }
 822         } else if (cap_fmode >= 0) {
 823                 pr_warning("mds issued no caps on %llx.%llx\n",
 824                            ceph_vinop(inode));
 825                 __ceph_get_fmode(ci, cap_fmode);
 826         }
 827
 828         /* update delegation info? */
 829         if (dirinfo)
 830                 ceph_fill_dirfrag(inode, dirinfo);
 831
 832         err = 0;
 833
 834 out:
 835         if (xattr_blob)
 836                 ceph_buffer_put(xattr_blob);
 837         return err;
 838 }
 839
 840 /*
 841  * caller should hold session s_mutex.
 842  */
 843 static void update_dentry_lease(struct dentry *dentry,
 844                                 struct ceph_mds_reply_lease *lease,
 845                                 struct ceph_mds_session *session,
 846                                 unsigned long from_time)
 847 {
 848         struct ceph_dentry_info *di = ceph_dentry(dentry);
 849         long unsigned duration = le32_to_cpu(lease->duration_ms);
 850         long unsigned ttl = from_time + (duration * HZ) / 1000;
 851         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
 852         struct inode *dir;
 853
 854         /* only track leases on regular dentries */
 855         if (dentry->d_op != &ceph_dentry_ops)
 856                 return;
 857
 858         spin_lock(&dentry->d_lock);
 859         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
 860              dentry, duration, ttl);
 861
 862         /* make lease_rdcache_gen match directory */
 863         dir = dentry->d_parent->d_inode;
 864         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
 865
 866         if (duration == 0)
 867                 goto out_unlock;
 868
 869         if (di->lease_gen == session->s_cap_gen &&
 870             time_before(ttl, dentry->d_time))
 871                 goto out_unlock;  /* we already have a newer lease. */
 872
 873         if (di->lease_session && di->lease_session != session)
 874                 goto out_unlock;
 875
 876         ceph_dentry_lru_touch(dentry);
 877
 878         if (!di->lease_session)
 879                 di->lease_session = ceph_get_mds_session(session);
 880         di->lease_gen = session->s_cap_gen;
 881         di->lease_seq = le32_to_cpu(lease->seq);
 882         di->lease_renew_after = half_ttl;
 883         di->lease_renew_from = 0;
 884         dentry->d_time = ttl;
 885 out_unlock:
 886         spin_unlock(&dentry->d_lock);
 887         return;
 888 }
 889
 890 /*
 891  * Set dentry's directory position based on the current dir's max, and
 892  * order it in d_subdirs, so that dcache_readdir behaves.
 893  *
 894  * Always called under directory's i_mutex.
 895  */
 896 static void ceph_set_dentry_offset(struct dentry *dn)
 897 {
 898         struct dentry *dir = dn->d_parent;
 899         struct inode *inode = dir->d_inode;
 900         struct ceph_inode_info *ci;
 901         struct ceph_dentry_info *di;
 902
 903         BUG_ON(!inode);
 904
 905         ci = ceph_inode(inode);
 906         di = ceph_dentry(dn);
 907
 908         spin_lock(&ci->i_ceph_lock);
 909         if (!__ceph_dir_is_complete(ci)) {
 910                 spin_unlock(&ci->i_ceph_lock);
 911                 return;
 912         }
 913         di->offset = ceph_inode(inode)->i_max_offset++;
 914         spin_unlock(&ci->i_ceph_lock);
 915
 916         spin_lock(&dir->d_lock);
 917         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
 918         list_move(&dn->d_u.d_child, &dir->d_subdirs);
 919         dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
 920              dn->d_u.d_child.prev, dn->d_u.d_child.next);
 921         spin_unlock(&dn->d_lock);
 922         spin_unlock(&dir->d_lock);
 923 }
 924
 925 /*
 926  * splice a dentry to an inode.
 927  * caller must hold directory i_mutex for this to be safe.
 928  *
 929  * we will only rehash the resulting dentry if @prehash is
 930  * true; @prehash will be set to false (for the benefit of
 931  * the caller) if we fail.
 932  */
 933 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
 934                                     bool *prehash, bool set_offset)
 935 {
 936         struct dentry *realdn;
 937
 938         BUG_ON(dn->d_inode);
 939
 940         /* dn must be unhashed */
 941         if (!d_unhashed(dn))
 942                 d_drop(dn);
 943         realdn = d_materialise_unique(dn, in);
 944         if (IS_ERR(realdn)) {
 945                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
 946                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
 947                 if (prehash)
 948                         *prehash = false; /* don't rehash on error */
 949                 dn = realdn; /* note realdn contains the error */
 950                 goto out;
 951         } else if (realdn) {
 952                 dout("dn %p (%d) spliced with %p (%d) "
 953                      "inode %p ino %llx.%llx\n",
 954                      dn, d_count(dn),
 955                      realdn, d_count(realdn),
 956                      realdn->d_inode, ceph_vinop(realdn->d_inode));
 957                 dput(dn);
 958                 dn = realdn;
 959         } else {
 960                 BUG_ON(!ceph_dentry(dn));
 961                 dout("dn %p attached to %p ino %llx.%llx\n",
 962                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
 963         }
 964         if ((!prehash || *prehash) && d_unhashed(dn))
 965                 d_rehash(dn);
 966         if (set_offset)
 967                 ceph_set_dentry_offset(dn);
 968 out:
 969         return dn;
 970 }
 971
 972 /*
 973  * Incorporate results into the local cache.  This is either just
 974  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
 975  * after a lookup).
 976  *
 977  * A reply may contain
 978  *         a directory inode along with a dentry.
 979  *  and/or a target inode
 980  *
 981  * Called with snap_rwsem (read).
 982  */
 983 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
 984                     struct ceph_mds_session *session)
 985 {
 986         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
 987         struct inode *in = NULL;
 988         struct ceph_mds_reply_inode *ininfo;
 989         struct ceph_vino vino;
 990         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 991         int err = 0;
 992
 993         dout("fill_trace %p is_dentry %d is_target %d\n", req,
 994              rinfo->head->is_dentry, rinfo->head->is_target);
 995
 996 #if 0
 997         /*
 998          * Debugging hook:
 999          *
1000          * If we resend completed ops to a recovering mds, we get no
1001          * trace.  Since that is very rare, pretend this is the case
1002          * to ensure the 'no trace' handlers in the callers behave.
1003          *
1004          * Fill in inodes unconditionally to avoid breaking cap
1005          * invariants.
1006          */
1007         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1008                 pr_info("fill_trace faking empty trace on %lld %s\n",
1009                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1010                 if (rinfo->head->is_dentry) {
1011                         rinfo->head->is_dentry = 0;
1012                         err = fill_inode(req->r_locked_dir,
1013                                          &rinfo->diri, rinfo->dirfrag,
1014                                          session, req->r_request_started, -1);
1015                 }
1016                 if (rinfo->head->is_target) {
1017                         rinfo->head->is_target = 0;
1018                         ininfo = rinfo->targeti.in;
1019                         vino.ino = le64_to_cpu(ininfo->ino);
1020                         vino.snap = le64_to_cpu(ininfo->snapid);
1021                         in = ceph_get_inode(sb, vino);
1022                         err = fill_inode(in, &rinfo->targeti, NULL,
1023                                          session, req->r_request_started,
1024                                          req->r_fmode);
1025                         iput(in);
1026                 }
1027         }
1028 #endif
1029
1030         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1031                 dout("fill_trace reply is empty!\n");
1032                 if (rinfo->head->result == 0 && req->r_locked_dir)
1033                         ceph_invalidate_dir_request(req);
1034                 return 0;
1035         }
1036
1037         if (rinfo->head->is_dentry) {
1038                 struct inode *dir = req->r_locked_dir;
1039
1040                 if (dir) {
1041                         err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
1042                                          session, req->r_request_started, -1,
1043                                          &req->r_caps_reservation);
1044                         if (err < 0)
1045                                 return err;
1046                 } else {
1047                         WARN_ON_ONCE(1);
1048                 }
1049         }
1050
1051         if (rinfo->head->is_target) {
1052                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1053                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1054
1055                 in = ceph_get_inode(sb, vino);
1056                 if (IS_ERR(in)) {
1057                         err = PTR_ERR(in);
1058                         goto done;
1059                 }
1060                 req->r_target_inode = in;
1061
1062                 err = fill_inode(in, &rinfo->targeti, NULL,
1063                                 session, req->r_request_started,
1064                                 (le32_to_cpu(rinfo->head->result) == 0) ?
1065                                 req->r_fmode : -1,
1066                                 &req->r_caps_reservation);
1067                 if (err < 0) {
1068                         pr_err("fill_inode badness %p %llx.%llx\n",
1069                                 in, ceph_vinop(in));
1070                         goto done;
1071                 }
1072         }
1073
1074         /*
1075          * ignore null lease/binding on snapdir ENOENT, or else we
1076          * will have trouble splicing in the virtual snapdir later
1077          */
1078         if (rinfo->head->is_dentry && !req->r_aborted &&
1079             req->r_locked_dir &&
1080             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1081                                                fsc->mount_options->snapdir_name,
1082                                                req->r_dentry->d_name.len))) {
1083                 /*
1084                  * lookup link rename   : null -> possibly existing inode
1085                  * mknod symlink mkdir  : null -> new inode
1086                  * unlink               : linked -> null
1087                  */
1088                 struct inode *dir = req->r_locked_dir;
1089                 struct dentry *dn = req->r_dentry;
1090                 bool have_dir_cap, have_lease;
1091
1092                 BUG_ON(!dn);
1093                 BUG_ON(!dir);
1094                 BUG_ON(dn->d_parent->d_inode != dir);
1095                 BUG_ON(ceph_ino(dir) !=
1096                        le64_to_cpu(rinfo->diri.in->ino));
1097                 BUG_ON(ceph_snap(dir) !=
1098                        le64_to_cpu(rinfo->diri.in->snapid));
1099
1100                 /* do we have a lease on the whole dir? */
1101                 have_dir_cap =
1102                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1103                          CEPH_CAP_FILE_SHARED);
1104
1105                 /* do we have a dn lease? */
1106                 have_lease = have_dir_cap ||
1107                         le32_to_cpu(rinfo->dlease->duration_ms);
1108                 if (!have_lease)
1109                         dout("fill_trace  no dentry lease or dir cap\n");
1110
1111                 /* rename? */
1112                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1113                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1114                              req->r_old_dentry,
1115                              req->r_old_dentry->d_name.len,
1116                              req->r_old_dentry->d_name.name,
1117                              dn, dn->d_name.len, dn->d_name.name);
1118                         dout("fill_trace doing d_move %p -> %p\n",
1119                              req->r_old_dentry, dn);
1120
1121                         d_move(req->r_old_dentry, dn);
1122                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1123                              req->r_old_dentry,
1124                              req->r_old_dentry->d_name.len,
1125                              req->r_old_dentry->d_name.name,
1126                              dn, dn->d_name.len, dn->d_name.name);
1127
1128                         /* ensure target dentry is invalidated, despite
1129                            rehashing bug in vfs_rename_dir */
1130                         ceph_invalidate_dentry_lease(dn);
1131
1132                         /*
1133                          * d_move() puts the renamed dentry at the end of
1134                          * d_subdirs.  We need to assign it an appropriate
1135                          * directory offset so we can behave when dir is
1136                          * complete.
1137                          */
1138                         ceph_set_dentry_offset(req->r_old_dentry);
1139                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1140                              ceph_dentry(req->r_old_dentry)->offset);
1141
1142                         dn = req->r_old_dentry;  /* use old_dentry */
1143                 }
1144
1145                 /* null dentry? */
1146                 if (!rinfo->head->is_target) {
1147                         dout("fill_trace null dentry\n");
1148                         if (dn->d_inode) {
1149                                 dout("d_delete %p\n", dn);
1150                                 d_delete(dn);
1151                         } else {
1152                                 dout("d_instantiate %p NULL\n", dn);
1153                                 d_instantiate(dn, NULL);
1154                                 if (have_lease && d_unhashed(dn))
1155                                         d_rehash(dn);
1156                                 update_dentry_lease(dn, rinfo->dlease,
1157                                                     session,
1158                                                     req->r_request_started);
1159                         }
1160                         goto done;
1161                 }
1162
1163                 /* attach proper inode */
1164                 if (!dn->d_inode) {
1165                         ihold(in);
1166                         dn = splice_dentry(dn, in, &have_lease, true);
1167                         if (IS_ERR(dn)) {
1168                                 err = PTR_ERR(dn);
1169                                 goto done;
1170                         }
1171                         req->r_dentry = dn;  /* may have spliced */
1172                 } else if (dn->d_inode && dn->d_inode != in) {
1173                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1174                              dn, dn->d_inode, ceph_vinop(dn->d_inode),
1175                              ceph_vinop(in));
1176                         have_lease = false;
1177                 }
1178
1179                 if (have_lease)
1180                         update_dentry_lease(dn, rinfo->dlease, session,
1181                                             req->r_request_started);
1182                 dout(" final dn %p\n", dn);
1183         } else if (!req->r_aborted &&
1184                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1185                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1186                 struct dentry *dn = req->r_dentry;
1187
1188                 /* fill out a snapdir LOOKUPSNAP dentry */
1189                 BUG_ON(!dn);
1190                 BUG_ON(!req->r_locked_dir);
1191                 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1192                 ininfo = rinfo->targeti.in;
1193                 vino.ino = le64_to_cpu(ininfo->ino);
1194                 vino.snap = le64_to_cpu(ininfo->snapid);
1195                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1196                 ihold(in);
1197                 dn = splice_dentry(dn, in, NULL, true);
1198                 if (IS_ERR(dn)) {
1199                         err = PTR_ERR(dn);
1200                         goto done;
1201                 }
1202                 req->r_dentry = dn;  /* may have spliced */
1203         }
1204 done:
1205         dout("fill_trace done err=%d\n", err);
1206         return err;
1207 }
1208
1209 /*
1210  * Prepopulate our cache with readdir results, leases, etc.
1211  */
1212 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1213                                            struct ceph_mds_session *session)
1214 {
1215         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1216         int i, err = 0;
1217
1218         for (i = 0; i < rinfo->dir_nr; i++) {
1219                 struct ceph_vino vino;
1220                 struct inode *in;
1221                 int rc;
1222
1223                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1224                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1225
1226                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1227                 if (IS_ERR(in)) {
1228                         err = PTR_ERR(in);
1229                         dout("new_inode badness got %d\n", err);
1230                         continue;
1231                 }
1232                 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1233                                 req->r_request_started, -1,
1234                                 &req->r_caps_reservation);
1235                 if (rc < 0) {
1236                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1237                         err = rc;
1238                         continue;
1239                 }
1240         }
1241
1242         return err;
1243 }
1244
1245 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1246                              struct ceph_mds_session *session)
1247 {
1248         struct dentry *parent = req->r_dentry;
1249         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1250         struct qstr dname;
1251         struct dentry *dn;
1252         struct inode *in;
1253         int err = 0, ret, i;
1254         struct inode *snapdir = NULL;
1255         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1256         struct ceph_dentry_info *di;
1257         u64 r_readdir_offset = req->r_readdir_offset;
1258         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1259
1260         if (rinfo->dir_dir &&
1261             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1262                 dout("readdir_prepopulate got new frag %x -> %x\n",
1263                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1264                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1265                 if (ceph_frag_is_leftmost(frag))
1266                         r_readdir_offset = 2;
1267                 else
1268                         r_readdir_offset = 0;
1269         }
1270
1271         if (req->r_aborted)
1272                 return readdir_prepopulate_inodes_only(req, session);
1273
1274         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1275                 snapdir = ceph_get_snapdir(parent->d_inode);
1276                 parent = d_find_alias(snapdir);
1277                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1278                      rinfo->dir_nr, parent);
1279         } else {
1280                 dout("readdir_prepopulate %d items under dn %p\n",
1281                      rinfo->dir_nr, parent);
1282                 if (rinfo->dir_dir)
1283                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1284         }
1285
1286         /* FIXME: release caps/leases if error occurs */
1287         for (i = 0; i < rinfo->dir_nr; i++) {
1288                 struct ceph_vino vino;
1289
1290                 dname.name = rinfo->dir_dname[i];
1291                 dname.len = rinfo->dir_dname_len[i];
1292                 dname.hash = full_name_hash(dname.name, dname.len);
1293
1294                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1295                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1296
1297 retry_lookup:
1298                 dn = d_lookup(parent, &dname);
1299                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1300                      parent, dname.len, dname.name, dn);
1301
1302                 if (!dn) {
1303                         dn = d_alloc(parent, &dname);
1304                         dout("d_alloc %p '%.*s' = %p\n", parent,
1305                              dname.len, dname.name, dn);
1306                         if (dn == NULL) {
1307                                 dout("d_alloc badness\n");
1308                                 err = -ENOMEM;
1309                                 goto out;
1310                         }
1311                         ret = ceph_init_dentry(dn);
1312                         if (ret < 0) {
1313                                 dput(dn);
1314                                 err = ret;
1315                                 goto out;
1316                         }
1317                 } else if (dn->d_inode &&
1318                            (ceph_ino(dn->d_inode) != vino.ino ||
1319                             ceph_snap(dn->d_inode) != vino.snap)) {
1320                         dout(" dn %p points to wrong inode %p\n",
1321                              dn, dn->d_inode);
1322                         d_delete(dn);
1323                         dput(dn);
1324                         goto retry_lookup;
1325                 } else {
1326                         /* reorder parent's d_subdirs */
1327                         spin_lock(&parent->d_lock);
1328                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1329                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1330                         spin_unlock(&dn->d_lock);
1331                         spin_unlock(&parent->d_lock);
1332                 }
1333
1334                 /* inode */
1335                 if (dn->d_inode) {
1336                         in = dn->d_inode;
1337                 } else {
1338                         in = ceph_get_inode(parent->d_sb, vino);
1339                         if (IS_ERR(in)) {
1340                                 dout("new_inode badness\n");
1341                                 d_drop(dn);
1342                                 dput(dn);
1343                                 err = PTR_ERR(in);
1344                                 goto out;
1345                         }
1346                 }
1347
1348                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1349                                req->r_request_started, -1,
1350                                &req->r_caps_reservation) < 0) {
1351                         pr_err("fill_inode badness on %p\n", in);
1352                         if (!dn->d_inode)
1353                                 iput(in);
1354                         d_drop(dn);
1355                         goto next_item;
1356                 }
1357
1358                 if (!dn->d_inode) {
1359                         dn = splice_dentry(dn, in, NULL, false);
1360                         if (IS_ERR(dn)) {
1361                                 err = PTR_ERR(dn);
1362                                 dn = NULL;
1363                                 goto next_item;
1364                         }
1365                 }
1366
1367                 di = dn->d_fsdata;
1368                 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1369
1370                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1371                                     req->r_session,
1372                                     req->r_request_started);
1373 next_item:
1374                 if (dn)
1375                         dput(dn);
1376         }
1377         if (err == 0)
1378                 req->r_did_prepopulate = true;
1379
1380 out:
1381         if (snapdir) {
1382                 iput(snapdir);
1383                 dput(parent);
1384         }
1385         dout("readdir_prepopulate done\n");
1386         return err;
1387 }
1388
1389 int ceph_inode_set_size(struct inode *inode, loff_t size)
1390 {
1391         struct ceph_inode_info *ci = ceph_inode(inode);
1392         int ret = 0;
1393
1394         spin_lock(&ci->i_ceph_lock);
1395         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1396         inode->i_size = size;
1397         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1398
1399         /* tell the MDS if we are approaching max_size */
1400         if ((size << 1) >= ci->i_max_size &&
1401             (ci->i_reported_size << 1) < ci->i_max_size)
1402                 ret = 1;
1403
1404         spin_unlock(&ci->i_ceph_lock);
1405         return ret;
1406 }
1407
1408 /*
1409  * Write back inode data in a worker thread.  (This can't be done
1410  * in the message handler context.)
1411  */
1412 void ceph_queue_writeback(struct inode *inode)
1413 {
1414         ihold(inode);
1415         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1416                        &ceph_inode(inode)->i_wb_work)) {
1417                 dout("ceph_queue_writeback %p\n", inode);
1418         } else {
1419                 dout("ceph_queue_writeback %p failed\n", inode);
1420                 iput(inode);
1421         }
1422 }
1423
1424 static void ceph_writeback_work(struct work_struct *work)
1425 {
1426         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1427                                                   i_wb_work);
1428         struct inode *inode = &ci->vfs_inode;
1429
1430         dout("writeback %p\n", inode);
1431         filemap_fdatawrite(&inode->i_data);
1432         iput(inode);
1433 }
1434
1435 /*
1436  * queue an async invalidation
1437  */
1438 void ceph_queue_invalidate(struct inode *inode)
1439 {
1440         ihold(inode);
1441         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1442                        &ceph_inode(inode)->i_pg_inv_work)) {
1443                 dout("ceph_queue_invalidate %p\n", inode);
1444         } else {
1445                 dout("ceph_queue_invalidate %p failed\n", inode);
1446                 iput(inode);
1447         }
1448 }
1449
1450 /*
1451  * Invalidate inode pages in a worker thread.  (This can't be done
1452  * in the message handler context.)
1453  */
1454 static void ceph_invalidate_work(struct work_struct *work)
1455 {
1456         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1457                                                   i_pg_inv_work);
1458         struct inode *inode = &ci->vfs_inode;
1459         u32 orig_gen;
1460         int check = 0;
1461
1462         mutex_lock(&ci->i_truncate_mutex);
1463         spin_lock(&ci->i_ceph_lock);
1464         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1465              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1466         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1467                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1468                         check = 1;
1469                 spin_unlock(&ci->i_ceph_lock);
1470                 mutex_unlock(&ci->i_truncate_mutex);
1471                 goto out;
1472         }
1473         orig_gen = ci->i_rdcache_gen;
1474         spin_unlock(&ci->i_ceph_lock);
1475
1476         truncate_inode_pages(inode->i_mapping, 0);
1477
1478         spin_lock(&ci->i_ceph_lock);
1479         if (orig_gen == ci->i_rdcache_gen &&
1480             orig_gen == ci->i_rdcache_revoking) {
1481                 dout("invalidate_pages %p gen %d successful\n", inode,
1482                      ci->i_rdcache_gen);
1483                 ci->i_rdcache_revoking--;
1484                 check = 1;
1485         } else {
1486                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1487                      inode, orig_gen, ci->i_rdcache_gen,
1488                      ci->i_rdcache_revoking);
1489                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1490                         check = 1;
1491         }
1492         spin_unlock(&ci->i_ceph_lock);
1493         mutex_unlock(&ci->i_truncate_mutex);
1494 out:
1495         if (check)
1496                 ceph_check_caps(ci, 0, NULL);
1497         iput(inode);
1498 }
1499
1500
1501 /*
1502  * called by trunc_wq;
1503  *
1504  * We also truncate in a separate thread as well.
1505  */
1506 static void ceph_vmtruncate_work(struct work_struct *work)
1507 {
1508         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1509                                                   i_vmtruncate_work);
1510         struct inode *inode = &ci->vfs_inode;
1511
1512         dout("vmtruncate_work %p\n", inode);
1513         __ceph_do_pending_vmtruncate(inode);
1514         iput(inode);
1515 }
1516
1517 /*
1518  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1519  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1520  */
1521 void ceph_queue_vmtruncate(struct inode *inode)
1522 {
1523         struct ceph_inode_info *ci = ceph_inode(inode);
1524
1525         ihold(inode);
1526
1527         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1528                        &ci->i_vmtruncate_work)) {
1529                 dout("ceph_queue_vmtruncate %p\n", inode);
1530         } else {
1531                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1532                      inode, ci->i_truncate_pending);
1533                 iput(inode);
1534         }
1535 }
1536
1537 /*
1538  * Make sure any pending truncation is applied before doing anything
1539  * that may depend on it.
1540  */
1541 void __ceph_do_pending_vmtruncate(struct inode *inode)
1542 {
1543         struct ceph_inode_info *ci = ceph_inode(inode);
1544         u64 to;
1545         int wrbuffer_refs, finish = 0;
1546
1547         mutex_lock(&ci->i_truncate_mutex);
1548 retry:
1549         spin_lock(&ci->i_ceph_lock);
1550         if (ci->i_truncate_pending == 0) {
1551                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1552                 spin_unlock(&ci->i_ceph_lock);
1553                 mutex_unlock(&ci->i_truncate_mutex);
1554                 return;
1555         }
1556
1557         /*
1558          * make sure any dirty snapped pages are flushed before we
1559          * possibly truncate them.. so write AND block!
1560          */
1561         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1562                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1563                      inode);
1564                 spin_unlock(&ci->i_ceph_lock);
1565                 filemap_write_and_wait_range(&inode->i_data, 0,
1566                                              inode->i_sb->s_maxbytes);
1567                 goto retry;
1568         }
1569
1570         /* there should be no reader or writer */
1571         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1572
1573         to = ci->i_truncate_size;
1574         wrbuffer_refs = ci->i_wrbuffer_ref;
1575         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1576              ci->i_truncate_pending, to);
1577         spin_unlock(&ci->i_ceph_lock);
1578
1579         truncate_inode_pages(inode->i_mapping, to);
1580
1581         spin_lock(&ci->i_ceph_lock);
1582         if (to == ci->i_truncate_size) {
1583                 ci->i_truncate_pending = 0;
1584                 finish = 1;
1585         }
1586         spin_unlock(&ci->i_ceph_lock);
1587         if (!finish)
1588                 goto retry;
1589
1590         mutex_unlock(&ci->i_truncate_mutex);
1591
1592         if (wrbuffer_refs == 0)
1593                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1594
1595         wake_up_all(&ci->i_cap_wq);
1596 }
1597
1598 /*
1599  * symlinks
1600  */
1601 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1602 {
1603         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1604         nd_set_link(nd, ci->i_symlink);
1605         return NULL;
1606 }
1607
1608 static const struct inode_operations ceph_symlink_iops = {
1609         .readlink = generic_readlink,
1610         .follow_link = ceph_sym_follow_link,
1611         .setattr = ceph_setattr,
1612         .getattr = ceph_getattr,
1613         .setxattr = ceph_setxattr,
1614         .getxattr = ceph_getxattr,
1615         .listxattr = ceph_listxattr,
1616         .removexattr = ceph_removexattr,
1617         .get_acl = ceph_get_acl,
1618 };
1619
1620 /*
1621  * setattr
1622  */
1623 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1624 {
1625         struct inode *inode = dentry->d_inode;
1626         struct ceph_inode_info *ci = ceph_inode(inode);
1627         struct inode *parent_inode;
1628         const unsigned int ia_valid = attr->ia_valid;
1629         struct ceph_mds_request *req;
1630         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1631         int issued;
1632         int release = 0, dirtied = 0;
1633         int mask = 0;
1634         int err = 0;
1635         int inode_dirty_flags = 0;
1636
1637         if (ceph_snap(inode) != CEPH_NOSNAP)
1638                 return -EROFS;
1639
1640         err = inode_change_ok(inode, attr);
1641         if (err != 0)
1642                 return err;
1643
1644         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1645                                        USE_AUTH_MDS);
1646         if (IS_ERR(req))
1647                 return PTR_ERR(req);
1648
1649         spin_lock(&ci->i_ceph_lock);
1650         issued = __ceph_caps_issued(ci, NULL);
1651         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1652
1653         if (ia_valid & ATTR_UID) {
1654                 dout("setattr %p uid %d -> %d\n", inode,
1655                      from_kuid(&init_user_ns, inode->i_uid),
1656                      from_kuid(&init_user_ns, attr->ia_uid));
1657                 if (issued & CEPH_CAP_AUTH_EXCL) {
1658                         inode->i_uid = attr->ia_uid;
1659                         dirtied |= CEPH_CAP_AUTH_EXCL;
1660                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1661                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1662                         req->r_args.setattr.uid = cpu_to_le32(
1663                                 from_kuid(&init_user_ns, attr->ia_uid));
1664                         mask |= CEPH_SETATTR_UID;
1665                         release |= CEPH_CAP_AUTH_SHARED;
1666                 }
1667         }
1668         if (ia_valid & ATTR_GID) {
1669                 dout("setattr %p gid %d -> %d\n", inode,
1670                      from_kgid(&init_user_ns, inode->i_gid),
1671                      from_kgid(&init_user_ns, attr->ia_gid));
1672                 if (issued & CEPH_CAP_AUTH_EXCL) {
1673                         inode->i_gid = attr->ia_gid;
1674                         dirtied |= CEPH_CAP_AUTH_EXCL;
1675                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1676                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1677                         req->r_args.setattr.gid = cpu_to_le32(
1678                                 from_kgid(&init_user_ns, attr->ia_gid));
1679                         mask |= CEPH_SETATTR_GID;
1680                         release |= CEPH_CAP_AUTH_SHARED;
1681                 }
1682         }
1683         if (ia_valid & ATTR_MODE) {
1684                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1685                      attr->ia_mode);
1686                 if (issued & CEPH_CAP_AUTH_EXCL) {
1687                         inode->i_mode = attr->ia_mode;
1688                         dirtied |= CEPH_CAP_AUTH_EXCL;
1689                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1690                            attr->ia_mode != inode->i_mode) {
1691                         inode->i_mode = attr->ia_mode;
1692                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1693                         mask |= CEPH_SETATTR_MODE;
1694                         release |= CEPH_CAP_AUTH_SHARED;
1695                 }
1696         }
1697
1698         if (ia_valid & ATTR_ATIME) {
1699                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1700                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1701                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1702                 if (issued & CEPH_CAP_FILE_EXCL) {
1703                         ci->i_time_warp_seq++;
1704                         inode->i_atime = attr->ia_atime;
1705                         dirtied |= CEPH_CAP_FILE_EXCL;
1706                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1707                            timespec_compare(&inode->i_atime,
1708                                             &attr->ia_atime) < 0) {
1709                         inode->i_atime = attr->ia_atime;
1710                         dirtied |= CEPH_CAP_FILE_WR;
1711                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1712                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1713                         ceph_encode_timespec(&req->r_args.setattr.atime,
1714                                              &attr->ia_atime);
1715                         mask |= CEPH_SETATTR_ATIME;
1716                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1717                                 CEPH_CAP_FILE_WR;
1718                 }
1719         }
1720         if (ia_valid & ATTR_MTIME) {
1721                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1722                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1723                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1724                 if (issued & CEPH_CAP_FILE_EXCL) {
1725                         ci->i_time_warp_seq++;
1726                         inode->i_mtime = attr->ia_mtime;
1727                         dirtied |= CEPH_CAP_FILE_EXCL;
1728                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1729                            timespec_compare(&inode->i_mtime,
1730                                             &attr->ia_mtime) < 0) {
1731                         inode->i_mtime = attr->ia_mtime;
1732                         dirtied |= CEPH_CAP_FILE_WR;
1733                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1734                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1735                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1736                                              &attr->ia_mtime);
1737                         mask |= CEPH_SETATTR_MTIME;
1738                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1739                                 CEPH_CAP_FILE_WR;
1740                 }
1741         }
1742         if (ia_valid & ATTR_SIZE) {
1743                 dout("setattr %p size %lld -> %lld\n", inode,
1744                      inode->i_size, attr->ia_size);
1745                 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1746                         err = -EINVAL;
1747                         goto out;
1748                 }
1749                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1750                     attr->ia_size > inode->i_size) {
1751                         inode->i_size = attr->ia_size;
1752                         inode->i_blocks =
1753                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1754                         inode->i_ctime = attr->ia_ctime;
1755                         ci->i_reported_size = attr->ia_size;
1756                         dirtied |= CEPH_CAP_FILE_EXCL;
1757                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1758                            attr->ia_size != inode->i_size) {
1759                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1760                         req->r_args.setattr.old_size =
1761                                 cpu_to_le64(inode->i_size);
1762                         mask |= CEPH_SETATTR_SIZE;
1763                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1764                                 CEPH_CAP_FILE_WR;
1765                 }
1766         }
1767
1768         /* these do nothing */
1769         if (ia_valid & ATTR_CTIME) {
1770                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1771                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1772                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1773                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1774                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1775                      only ? "ctime only" : "ignored");
1776                 inode->i_ctime = attr->ia_ctime;
1777                 if (only) {
1778                         /*
1779                          * if kernel wants to dirty ctime but nothing else,
1780                          * we need to choose a cap to dirty under, or do
1781                          * a almost-no-op setattr
1782                          */
1783                         if (issued & CEPH_CAP_AUTH_EXCL)
1784                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1785                         else if (issued & CEPH_CAP_FILE_EXCL)
1786                                 dirtied |= CEPH_CAP_FILE_EXCL;
1787                         else if (issued & CEPH_CAP_XATTR_EXCL)
1788                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1789                         else
1790                                 mask |= CEPH_SETATTR_CTIME;
1791                 }
1792         }
1793         if (ia_valid & ATTR_FILE)
1794                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1795
1796         if (dirtied) {
1797                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1798                 inode->i_ctime = CURRENT_TIME;
1799         }
1800
1801         release &= issued;
1802         spin_unlock(&ci->i_ceph_lock);
1803
1804         if (inode_dirty_flags)
1805                 __mark_inode_dirty(inode, inode_dirty_flags);
1806
1807         if (ia_valid & ATTR_MODE) {
1808                 err = ceph_acl_chmod(dentry, inode);
1809                 if (err)
1810                         goto out_put;
1811         }
1812
1813         if (mask) {
1814                 req->r_inode = inode;
1815                 ihold(inode);
1816                 req->r_inode_drop = release;
1817                 req->r_args.setattr.mask = cpu_to_le32(mask);
1818                 req->r_num_caps = 1;
1819                 parent_inode = ceph_get_dentry_parent_inode(dentry);
1820                 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1821                 iput(parent_inode);
1822         }
1823         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1824              ceph_cap_string(dirtied), mask);
1825
1826         ceph_mdsc_put_request(req);
1827         if (mask & CEPH_SETATTR_SIZE)
1828                 __ceph_do_pending_vmtruncate(inode);
1829         return err;
1830 out:
1831         spin_unlock(&ci->i_ceph_lock);
1832 out_put:
1833         ceph_mdsc_put_request(req);
1834         return err;
1835 }
1836
1837 /*
1838  * Verify that we have a lease on the given mask.  If not,
1839  * do a getattr against an mds.
1840  */
1841 int ceph_do_getattr(struct inode *inode, int mask)
1842 {
1843         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1844         struct ceph_mds_client *mdsc = fsc->mdsc;
1845         struct ceph_mds_request *req;
1846         int err;
1847
1848         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1849                 dout("do_getattr inode %p SNAPDIR\n", inode);
1850                 return 0;
1851         }
1852
1853         dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1854         if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1855                 return 0;
1856
1857         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1858         if (IS_ERR(req))
1859                 return PTR_ERR(req);
1860         req->r_inode = inode;
1861         ihold(inode);
1862         req->r_num_caps = 1;
1863         req->r_args.getattr.mask = cpu_to_le32(mask);
1864         err = ceph_mdsc_do_request(mdsc, NULL, req);
1865         ceph_mdsc_put_request(req);
1866         dout("do_getattr result=%d\n", err);
1867         return err;
1868 }
1869
1870
1871 /*
1872  * Check inode permissions.  We verify we have a valid value for
1873  * the AUTH cap, then call the generic handler.
1874  */
1875 int ceph_permission(struct inode *inode, int mask)
1876 {
1877         int err;
1878
1879         if (mask & MAY_NOT_BLOCK)
1880                 return -ECHILD;
1881
1882         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1883
1884         if (!err)
1885                 err = generic_permission(inode, mask);
1886         return err;
1887 }
1888
1889 /*
1890  * Get all attributes.  Hopefully somedata we'll have a statlite()
1891  * and can limit the fields we require to be accurate.
1892  */
1893 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1894                  struct kstat *stat)
1895 {
1896         struct inode *inode = dentry->d_inode;
1897         struct ceph_inode_info *ci = ceph_inode(inode);
1898         int err;
1899
1900         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1901         if (!err) {
1902                 generic_fillattr(inode, stat);
1903                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1904                 if (ceph_snap(inode) != CEPH_NOSNAP)
1905                         stat->dev = ceph_snap(inode);
1906                 else
1907                         stat->dev = 0;
1908                 if (S_ISDIR(inode->i_mode)) {
1909                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1910                                                 RBYTES))
1911                                 stat->size = ci->i_rbytes;
1912                         else
1913                                 stat->size = ci->i_files + ci->i_subdirs;
1914                         stat->blocks = 0;
1915                         stat->blksize = 65536;
1916                 }
1917         }
1918         return err;
1919 }