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