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