dir.c

   1// SPDX-License-Identifier: GPL-2.0
   2#include <linux/ceph/ceph_debug.h>
   3
   4#include <linux/spinlock.h>
   5#include <linux/namei.h>
   6#include <linux/slab.h>
   7#include <linux/sched.h>
   8#include <linux/xattr.h>
   9
  10#include "super.h"
  11#include "mds_client.h"
  12#include "crypto.h"
  13
  14/*
  15 * Directory operations: readdir, lookup, create, link, unlink,
  16 * rename, etc.
  17 */
  18
  19/*
  20 * Ceph MDS operations are specified in terms of a base ino and
  21 * relative path.  Thus, the client can specify an operation on a
  22 * specific inode (e.g., a getattr due to fstat(2)), or as a path
  23 * relative to, say, the root directory.
  24 *
  25 * Normally, we limit ourselves to strict inode ops (no path component)
  26 * or dentry operations (a single path component relative to an ino).  The
  27 * exception to this is open_root_dentry(), which will open the mount
  28 * point by name.
  29 */
  30
  31const struct dentry_operations ceph_dentry_ops;
  32
  33static bool __dentry_lease_is_valid(struct ceph_dentry_info *di);
  34static int __dir_lease_try_check(const struct dentry *dentry);
  35
  36/*
  37 * Initialize ceph dentry state.
  38 */
  39static int ceph_d_init(struct dentry *dentry)
  40{
  41	struct ceph_dentry_info *di;
  42	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb);
  43
  44	di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
  45	if (!di)
  46		return -ENOMEM;          /* oh well */
  47
  48	di->dentry = dentry;
  49	di->lease_session = NULL;
  50	di->time = jiffies;
  51	dentry->d_fsdata = di;
  52	INIT_LIST_HEAD(&di->lease_list);
  53
  54	atomic64_inc(&mdsc->metric.total_dentries);
  55
  56	return 0;
  57}
  58
  59/*
  60 * for f_pos for readdir:
  61 * - hash order:
  62 *	(0xff << 52) | ((24 bits hash) << 28) |
  63 *	(the nth entry has hash collision);
  64 * - frag+name order;
  65 *	((frag value) << 28) | (the nth entry in frag);
  66 */
  67#define OFFSET_BITS	28
  68#define OFFSET_MASK	((1 << OFFSET_BITS) - 1)
  69#define HASH_ORDER	(0xffull << (OFFSET_BITS + 24))
  70loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
  71{
  72	loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
  73	if (hash_order)
  74		fpos |= HASH_ORDER;
  75	return fpos;
  76}
  77
  78static bool is_hash_order(loff_t p)
  79{
  80	return (p & HASH_ORDER) == HASH_ORDER;
  81}
  82
  83static unsigned fpos_frag(loff_t p)
  84{
  85	return p >> OFFSET_BITS;
  86}
  87
  88static unsigned fpos_hash(loff_t p)
  89{
  90	return ceph_frag_value(fpos_frag(p));
  91}
  92
  93static unsigned fpos_off(loff_t p)
  94{
  95	return p & OFFSET_MASK;
  96}
  97
  98static int fpos_cmp(loff_t l, loff_t r)
  99{
 100	int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
 101	if (v)
 102		return v;
 103	return (int)(fpos_off(l) - fpos_off(r));
 104}
 105
 106/*
 107 * make note of the last dentry we read, so we can
 108 * continue at the same lexicographical point,
 109 * regardless of what dir changes take place on the
 110 * server.
 111 */
 112static int note_last_dentry(struct ceph_fs_client *fsc,
 113			    struct ceph_dir_file_info *dfi,
 114			    const char *name,
 115		            int len, unsigned next_offset)
 116{
 117	char *buf = kmalloc(len+1, GFP_KERNEL);
 118	if (!buf)
 119		return -ENOMEM;
 120	kfree(dfi->last_name);
 121	dfi->last_name = buf;
 122	memcpy(dfi->last_name, name, len);
 123	dfi->last_name[len] = 0;
 124	dfi->next_offset = next_offset;
 125	doutc(fsc->client, "'%s'\n", dfi->last_name);
 126	return 0;
 127}
 128
 129
 130static struct dentry *
 131__dcache_find_get_entry(struct dentry *parent, u64 idx,
 132			struct ceph_readdir_cache_control *cache_ctl)
 133{
 134	struct inode *dir = d_inode(parent);
 135	struct ceph_client *cl = ceph_inode_to_client(dir);
 136	struct dentry *dentry;
 137	unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
 138	loff_t ptr_pos = idx * sizeof(struct dentry *);
 139	pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
 140
 141	if (ptr_pos >= i_size_read(dir))
 142		return NULL;
 143
 144	if (!cache_ctl->folio || ptr_pgoff != cache_ctl->folio->index) {
 145		ceph_readdir_cache_release(cache_ctl);
 146		cache_ctl->folio = filemap_lock_folio(&dir->i_data, ptr_pgoff);
 147		if (IS_ERR(cache_ctl->folio)) {
 148			cache_ctl->folio = NULL;
 149			doutc(cl, " folio %lu not found\n", ptr_pgoff);
 150			return ERR_PTR(-EAGAIN);
 151		}
 152		/* reading/filling the cache are serialized by
 153		   i_rwsem, no need to use folio lock */
 154		folio_unlock(cache_ctl->folio);
 155		cache_ctl->dentries = kmap_local_folio(cache_ctl->folio, 0);
 156	}
 157
 158	cache_ctl->index = idx & idx_mask;
 159
 160	rcu_read_lock();
 161	spin_lock(&parent->d_lock);
 162	/* check i_size again here, because empty directory can be
 163	 * marked as complete while not holding the i_rwsem. */
 164	if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
 165		dentry = cache_ctl->dentries[cache_ctl->index];
 166	else
 167		dentry = NULL;
 168	spin_unlock(&parent->d_lock);
 169	if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
 170		dentry = NULL;
 171	rcu_read_unlock();
 172	return dentry ? : ERR_PTR(-EAGAIN);
 173}
 174
 175/*
 176 * When possible, we try to satisfy a readdir by peeking at the
 177 * dcache.  We make this work by carefully ordering dentries on
 178 * d_children when we initially get results back from the MDS, and
 179 * falling back to a "normal" sync readdir if any dentries in the dir
 180 * are dropped.
 181 *
 182 * Complete dir indicates that we have all dentries in the dir.  It is
 183 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
 184 * the MDS if/when the directory is modified).
 185 */
 186static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
 187			    int shared_gen)
 188{
 189	struct ceph_dir_file_info *dfi = file->private_data;
 190	struct dentry *parent = file->f_path.dentry;
 191	struct inode *dir = d_inode(parent);
 192	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(dir);
 193	struct ceph_client *cl = ceph_inode_to_client(dir);
 194	struct dentry *dentry, *last = NULL;
 195	struct ceph_dentry_info *di;
 196	struct ceph_readdir_cache_control cache_ctl = {};
 197	u64 idx = 0;
 198	int err = 0;
 199
 200	doutc(cl, "%p %llx.%llx v%u at %llx\n", dir, ceph_vinop(dir),
 201	      (unsigned)shared_gen, ctx->pos);
 202
 203	/* search start position */
 204	if (ctx->pos > 2) {
 205		u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
 206		while (count > 0) {
 207			u64 step = count >> 1;
 208			dentry = __dcache_find_get_entry(parent, idx + step,
 209							 &cache_ctl);
 210			if (!dentry) {
 211				/* use linear search */
 212				idx = 0;
 213				break;
 214			}
 215			if (IS_ERR(dentry)) {
 216				err = PTR_ERR(dentry);
 217				goto out;
 218			}
 219			di = ceph_dentry(dentry);
 220			spin_lock(&dentry->d_lock);
 221			if (fpos_cmp(di->offset, ctx->pos) < 0) {
 222				idx += step + 1;
 223				count -= step + 1;
 224			} else {
 225				count = step;
 226			}
 227			spin_unlock(&dentry->d_lock);
 228			dput(dentry);
 229		}
 230
 231		doutc(cl, "%p %llx.%llx cache idx %llu\n", dir,
 232		      ceph_vinop(dir), idx);
 233	}
 234
 235
 236	for (;;) {
 237		bool emit_dentry = false;
 238		dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
 239		if (!dentry) {
 240			dfi->file_info.flags |= CEPH_F_ATEND;
 241			err = 0;
 242			break;
 243		}
 244		if (IS_ERR(dentry)) {
 245			err = PTR_ERR(dentry);
 246			goto out;
 247		}
 248
 249		spin_lock(&dentry->d_lock);
 250		di = ceph_dentry(dentry);
 251		if (d_unhashed(dentry) ||
 252		    d_really_is_negative(dentry) ||
 253		    di->lease_shared_gen != shared_gen ||
 254		    ((dentry->d_flags & DCACHE_NOKEY_NAME) &&
 255		     fscrypt_has_encryption_key(dir))) {
 256			spin_unlock(&dentry->d_lock);
 257			dput(dentry);
 258			err = -EAGAIN;
 259			goto out;
 260		}
 261		if (fpos_cmp(ctx->pos, di->offset) <= 0) {
 262			__ceph_dentry_dir_lease_touch(di);
 263			emit_dentry = true;
 264		}
 265		spin_unlock(&dentry->d_lock);
 266
 267		if (emit_dentry) {
 268			doutc(cl, " %llx dentry %p %pd %p\n", di->offset,
 269			      dentry, dentry, d_inode(dentry));
 270			ctx->pos = di->offset;
 271			if (!dir_emit(ctx, dentry->d_name.name,
 272				      dentry->d_name.len, ceph_present_inode(d_inode(dentry)),
 273				      d_inode(dentry)->i_mode >> 12)) {
 274				dput(dentry);
 275				err = 0;
 276				break;
 277			}
 278			ctx->pos++;
 279
 280			if (last)
 281				dput(last);
 282			last = dentry;
 283		} else {
 284			dput(dentry);
 285		}
 286	}
 287out:
 288	ceph_readdir_cache_release(&cache_ctl);
 289	if (last) {
 290		int ret;
 291		di = ceph_dentry(last);
 292		ret = note_last_dentry(fsc, dfi, last->d_name.name,
 293				       last->d_name.len,
 294				       fpos_off(di->offset) + 1);
 295		if (ret < 0)
 296			err = ret;
 297		dput(last);
 298		/* last_name no longer match cache index */
 299		if (dfi->readdir_cache_idx >= 0) {
 300			dfi->readdir_cache_idx = -1;
 301			dfi->dir_release_count = 0;
 302		}
 303	}
 304	return err;
 305}
 306
 307static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
 308{
 309	if (!dfi->last_readdir)
 310		return true;
 311	if (is_hash_order(pos))
 312		return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
 313	else
 314		return dfi->frag != fpos_frag(pos);
 315}
 316
 317static int ceph_readdir(struct file *file, struct dir_context *ctx)
 318{
 319	struct ceph_dir_file_info *dfi = file->private_data;
 320	struct inode *inode = file_inode(file);
 321	struct ceph_inode_info *ci = ceph_inode(inode);
 322	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
 323	struct ceph_mds_client *mdsc = fsc->mdsc;
 324	struct ceph_client *cl = fsc->client;
 325	int i;
 326	int err;
 327	unsigned frag = -1;
 328	struct ceph_mds_reply_info_parsed *rinfo;
 329
 330	doutc(cl, "%p %llx.%llx file %p pos %llx\n", inode,
 331	      ceph_vinop(inode), file, ctx->pos);
 332	if (dfi->file_info.flags & CEPH_F_ATEND)
 333		return 0;
 334
 335	/* always start with . and .. */
 336	if (ctx->pos == 0) {
 337		doutc(cl, "%p %llx.%llx off 0 -> '.'\n", inode,
 338		      ceph_vinop(inode));
 339		if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode),
 340			    inode->i_mode >> 12))
 341			return 0;
 342		ctx->pos = 1;
 343	}
 344	if (ctx->pos == 1) {
 345		u64 ino;
 346		struct dentry *dentry = file->f_path.dentry;
 347
 348		spin_lock(&dentry->d_lock);
 349		ino = ceph_present_inode(dentry->d_parent->d_inode);
 350		spin_unlock(&dentry->d_lock);
 351
 352		doutc(cl, "%p %llx.%llx off 1 -> '..'\n", inode,
 353		      ceph_vinop(inode));
 354		if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12))
 355			return 0;
 356		ctx->pos = 2;
 357	}
 358
 359	err = ceph_fscrypt_prepare_readdir(inode);
 360	if (err < 0)
 361		return err;
 362
 363	spin_lock(&ci->i_ceph_lock);
 364	/* request Fx cap. if have Fx, we don't need to release Fs cap
 365	 * for later create/unlink. */
 366	__ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR);
 367	/* can we use the dcache? */
 368	if (ceph_test_mount_opt(fsc, DCACHE) &&
 369	    !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
 370	    ceph_snap(inode) != CEPH_SNAPDIR &&
 371	    __ceph_dir_is_complete_ordered(ci) &&
 372	    __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
 373		int shared_gen = atomic_read(&ci->i_shared_gen);
 374
 375		spin_unlock(&ci->i_ceph_lock);
 376		err = __dcache_readdir(file, ctx, shared_gen);
 377		if (err != -EAGAIN)
 378			return err;
 379	} else {
 380		spin_unlock(&ci->i_ceph_lock);
 381	}
 382
 383	/* proceed with a normal readdir */
 384more:
 385	/* do we have the correct frag content buffered? */
 386	if (need_send_readdir(dfi, ctx->pos)) {
 387		struct ceph_mds_request *req;
 388		int op = ceph_snap(inode) == CEPH_SNAPDIR ?
 389			CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
 390
 391		/* discard old result, if any */
 392		if (dfi->last_readdir) {
 393			ceph_mdsc_put_request(dfi->last_readdir);
 394			dfi->last_readdir = NULL;
 395		}
 396
 397		if (is_hash_order(ctx->pos)) {
 398			/* fragtree isn't always accurate. choose frag
 399			 * based on previous reply when possible. */
 400			if (frag == (unsigned)-1)
 401				frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
 402							NULL, NULL);
 403		} else {
 404			frag = fpos_frag(ctx->pos);
 405		}
 406
 407		doutc(cl, "fetching %p %llx.%llx frag %x offset '%s'\n",
 408		      inode, ceph_vinop(inode), frag, dfi->last_name);
 409		req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 410		if (IS_ERR(req))
 411			return PTR_ERR(req);
 412
 413		err = ceph_alloc_readdir_reply_buffer(req, inode);
 414		if (err) {
 415			ceph_mdsc_put_request(req);
 416			return err;
 417		}
 418		/* hints to request -> mds selection code */
 419		req->r_direct_mode = USE_AUTH_MDS;
 420		if (op == CEPH_MDS_OP_READDIR) {
 421			req->r_direct_hash = ceph_frag_value(frag);
 422			__set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
 423			req->r_inode_drop = CEPH_CAP_FILE_EXCL;
 424		}
 425		if (dfi->last_name) {
 426			int len = strlen(dfi->last_name);
 427
 428			req->r_path2 = kzalloc(NAME_MAX + 1, GFP_KERNEL);
 429			if (!req->r_path2) {
 430				ceph_mdsc_put_request(req);
 431				return -ENOMEM;
 432			}
 433			memcpy(req->r_path2, dfi->last_name, len);
 434
 435			err = ceph_encode_encrypted_dname(inode, req->r_path2, len);
 436			if (err < 0) {
 437				ceph_mdsc_put_request(req);
 438				return err;
 439			}
 440		} else if (is_hash_order(ctx->pos)) {
 441			req->r_args.readdir.offset_hash =
 442				cpu_to_le32(fpos_hash(ctx->pos));
 443		}
 444
 445		req->r_dir_release_cnt = dfi->dir_release_count;
 446		req->r_dir_ordered_cnt = dfi->dir_ordered_count;
 447		req->r_readdir_cache_idx = dfi->readdir_cache_idx;
 448		req->r_readdir_offset = dfi->next_offset;
 449		req->r_args.readdir.frag = cpu_to_le32(frag);
 450		req->r_args.readdir.flags =
 451				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
 452
 453		req->r_inode = inode;
 454		ihold(inode);
 455		req->r_dentry = dget(file->f_path.dentry);
 456		err = ceph_mdsc_do_request(mdsc, NULL, req);
 457		if (err < 0) {
 458			ceph_mdsc_put_request(req);
 459			return err;
 460		}
 461		doutc(cl, "%p %llx.%llx got and parsed readdir result=%d"
 462		      "on frag %x, end=%d, complete=%d, hash_order=%d\n",
 463		      inode, ceph_vinop(inode), err, frag,
 464		      (int)req->r_reply_info.dir_end,
 465		      (int)req->r_reply_info.dir_complete,
 466		      (int)req->r_reply_info.hash_order);
 467
 468		rinfo = &req->r_reply_info;
 469		if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
 470			frag = le32_to_cpu(rinfo->dir_dir->frag);
 471			if (!rinfo->hash_order) {
 472				dfi->next_offset = req->r_readdir_offset;
 473				/* adjust ctx->pos to beginning of frag */
 474				ctx->pos = ceph_make_fpos(frag,
 475							  dfi->next_offset,
 476							  false);
 477			}
 478		}
 479
 480		dfi->frag = frag;
 481		dfi->last_readdir = req;
 482
 483		if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
 484			dfi->readdir_cache_idx = req->r_readdir_cache_idx;
 485			if (dfi->readdir_cache_idx < 0) {
 486				/* preclude from marking dir ordered */
 487				dfi->dir_ordered_count = 0;
 488			} else if (ceph_frag_is_leftmost(frag) &&
 489				   dfi->next_offset == 2) {
 490				/* note dir version at start of readdir so
 491				 * we can tell if any dentries get dropped */
 492				dfi->dir_release_count = req->r_dir_release_cnt;
 493				dfi->dir_ordered_count = req->r_dir_ordered_cnt;
 494			}
 495		} else {
 496			doutc(cl, "%p %llx.%llx !did_prepopulate\n", inode,
 497			      ceph_vinop(inode));
 498			/* disable readdir cache */
 499			dfi->readdir_cache_idx = -1;
 500			/* preclude from marking dir complete */
 501			dfi->dir_release_count = 0;
 502		}
 503
 504		/* note next offset and last dentry name */
 505		if (rinfo->dir_nr > 0) {
 506			struct ceph_mds_reply_dir_entry *rde =
 507					rinfo->dir_entries + (rinfo->dir_nr-1);
 508			unsigned next_offset = req->r_reply_info.dir_end ?
 509					2 : (fpos_off(rde->offset) + 1);
 510			err = note_last_dentry(fsc, dfi, rde->name,
 511					       rde->name_len, next_offset);
 512			if (err) {
 513				ceph_mdsc_put_request(dfi->last_readdir);
 514				dfi->last_readdir = NULL;
 515				return err;
 516			}
 517		} else if (req->r_reply_info.dir_end) {
 518			dfi->next_offset = 2;
 519			/* keep last name */
 520		}
 521	}
 522
 523	rinfo = &dfi->last_readdir->r_reply_info;
 524	doutc(cl, "%p %llx.%llx frag %x num %d pos %llx chunk first %llx\n",
 525	      inode, ceph_vinop(inode), dfi->frag, rinfo->dir_nr, ctx->pos,
 526	      rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
 527
 528	i = 0;
 529	/* search start position */
 530	if (rinfo->dir_nr > 0) {
 531		int step, nr = rinfo->dir_nr;
 532		while (nr > 0) {
 533			step = nr >> 1;
 534			if (rinfo->dir_entries[i + step].offset < ctx->pos) {
 535				i +=  step + 1;
 536				nr -= step + 1;
 537			} else {
 538				nr = step;
 539			}
 540		}
 541	}
 542	for (; i < rinfo->dir_nr; i++) {
 543		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
 544
 545		if (rde->offset < ctx->pos) {
 546			pr_warn_client(cl,
 547				"%p %llx.%llx rde->offset 0x%llx ctx->pos 0x%llx\n",
 548				inode, ceph_vinop(inode), rde->offset, ctx->pos);
 549			return -EIO;
 550		}
 551
 552		if (WARN_ON_ONCE(!rde->inode.in))
 553			return -EIO;
 554
 555		ctx->pos = rde->offset;
 556		doutc(cl, "%p %llx.%llx (%d/%d) -> %llx '%.*s' %p\n", inode,
 557		      ceph_vinop(inode), i, rinfo->dir_nr, ctx->pos,
 558		      rde->name_len, rde->name, &rde->inode.in);
 559
 560		if (!dir_emit(ctx, rde->name, rde->name_len,
 561			      ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)),
 562			      le32_to_cpu(rde->inode.in->mode) >> 12)) {
 563			/*
 564			 * NOTE: Here no need to put the 'dfi->last_readdir',
 565			 * because when dir_emit stops us it's most likely
 566			 * doesn't have enough memory, etc. So for next readdir
 567			 * it will continue.
 568			 */
 569			doutc(cl, "filldir stopping us...\n");
 570			return 0;
 571		}
 572
 573		/* Reset the lengths to their original allocated vals */
 574		ctx->pos++;
 575	}
 576
 577	ceph_mdsc_put_request(dfi->last_readdir);
 578	dfi->last_readdir = NULL;
 579
 580	if (dfi->next_offset > 2) {
 581		frag = dfi->frag;
 582		goto more;
 583	}
 584
 585	/* more frags? */
 586	if (!ceph_frag_is_rightmost(dfi->frag)) {
 587		frag = ceph_frag_next(dfi->frag);
 588		if (is_hash_order(ctx->pos)) {
 589			loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
 590							dfi->next_offset, true);
 591			if (new_pos > ctx->pos)
 592				ctx->pos = new_pos;
 593			/* keep last_name */
 594		} else {
 595			ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
 596							false);
 597			kfree(dfi->last_name);
 598			dfi->last_name = NULL;
 599		}
 600		doutc(cl, "%p %llx.%llx next frag is %x\n", inode,
 601		      ceph_vinop(inode), frag);
 602		goto more;
 603	}
 604	dfi->file_info.flags |= CEPH_F_ATEND;
 605
 606	/*
 607	 * if dir_release_count still matches the dir, no dentries
 608	 * were released during the whole readdir, and we should have
 609	 * the complete dir contents in our cache.
 610	 */
 611	if (atomic64_read(&ci->i_release_count) ==
 612					dfi->dir_release_count) {
 613		spin_lock(&ci->i_ceph_lock);
 614		if (dfi->dir_ordered_count ==
 615				atomic64_read(&ci->i_ordered_count)) {
 616			doutc(cl, " marking %p %llx.%llx complete and ordered\n",
 617			      inode, ceph_vinop(inode));
 618			/* use i_size to track number of entries in
 619			 * readdir cache */
 620			BUG_ON(dfi->readdir_cache_idx < 0);
 621			i_size_write(inode, dfi->readdir_cache_idx *
 622				     sizeof(struct dentry*));
 623		} else {
 624			doutc(cl, " marking %llx.%llx complete\n",
 625			      ceph_vinop(inode));
 626		}
 627		__ceph_dir_set_complete(ci, dfi->dir_release_count,
 628					dfi->dir_ordered_count);
 629		spin_unlock(&ci->i_ceph_lock);
 630	}
 631	doutc(cl, "%p %llx.%llx file %p done.\n", inode, ceph_vinop(inode),
 632	      file);
 633	return 0;
 634}
 635
 636static void reset_readdir(struct ceph_dir_file_info *dfi)
 637{
 638	if (dfi->last_readdir) {
 639		ceph_mdsc_put_request(dfi->last_readdir);
 640		dfi->last_readdir = NULL;
 641	}
 642	kfree(dfi->last_name);
 643	dfi->last_name = NULL;
 644	dfi->dir_release_count = 0;
 645	dfi->readdir_cache_idx = -1;
 646	dfi->next_offset = 2;  /* compensate for . and .. */
 647	dfi->file_info.flags &= ~CEPH_F_ATEND;
 648}
 649
 650/*
 651 * discard buffered readdir content on seekdir(0), or seek to new frag,
 652 * or seek prior to current chunk
 653 */
 654static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
 655{
 656	struct ceph_mds_reply_info_parsed *rinfo;
 657	loff_t chunk_offset;
 658	if (new_pos == 0)
 659		return true;
 660	if (is_hash_order(new_pos)) {
 661		/* no need to reset last_name for a forward seek when
 662		 * dentries are sorted in hash order */
 663	} else if (dfi->frag != fpos_frag(new_pos)) {
 664		return true;
 665	}
 666	rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
 667	if (!rinfo || !rinfo->dir_nr)
 668		return true;
 669	chunk_offset = rinfo->dir_entries[0].offset;
 670	return new_pos < chunk_offset ||
 671	       is_hash_order(new_pos) != is_hash_order(chunk_offset);
 672}
 673
 674static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
 675{
 676	struct ceph_dir_file_info *dfi = file->private_data;
 677	struct inode *inode = file->f_mapping->host;
 678	struct ceph_client *cl = ceph_inode_to_client(inode);
 679	loff_t retval;
 680
 681	inode_lock(inode);
 682	retval = -EINVAL;
 683	switch (whence) {
 684	case SEEK_CUR:
 685		offset += file->f_pos;
 686		break;
 687	case SEEK_SET:
 688		break;
 689	case SEEK_END:
 690		retval = -EOPNOTSUPP;
 691		goto out;
 692	default:
 693		goto out;
 694	}
 695
 696	if (offset >= 0) {
 697		if (need_reset_readdir(dfi, offset)) {
 698			doutc(cl, "%p %llx.%llx dropping %p content\n",
 699			      inode, ceph_vinop(inode), file);
 700			reset_readdir(dfi);
 701		} else if (is_hash_order(offset) && offset > file->f_pos) {
 702			/* for hash offset, we don't know if a forward seek
 703			 * is within same frag */
 704			dfi->dir_release_count = 0;
 705			dfi->readdir_cache_idx = -1;
 706		}
 707
 708		if (offset != file->f_pos) {
 709			file->f_pos = offset;
 710			dfi->file_info.flags &= ~CEPH_F_ATEND;
 711		}
 712		retval = offset;
 713	}
 714out:
 715	inode_unlock(inode);
 716	return retval;
 717}
 718
 719/*
 720 * Handle lookups for the hidden .snap directory.
 721 */
 722struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req,
 723				   struct dentry *dentry)
 724{
 725	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
 726	struct inode *parent = d_inode(dentry->d_parent); /* we hold i_rwsem */
 727	struct ceph_client *cl = ceph_inode_to_client(parent);
 728
 729	/* .snap dir? */
 730	if (ceph_snap(parent) == CEPH_NOSNAP &&
 731	    strcmp(dentry->d_name.name, fsc->mount_options->snapdir_name) == 0) {
 732		struct dentry *res;
 733		struct inode *inode = ceph_get_snapdir(parent);
 734
 735		res = d_splice_alias(inode, dentry);
 736		doutc(cl, "ENOENT on snapdir %p '%pd', linking to "
 737		      "snapdir %p %llx.%llx. Spliced dentry %p\n",
 738		      dentry, dentry, inode, ceph_vinop(inode), res);
 739		if (res)
 740			dentry = res;
 741	}
 742	return dentry;
 743}
 744
 745/*
 746 * Figure out final result of a lookup/open request.
 747 *
 748 * Mainly, make sure we return the final req->r_dentry (if it already
 749 * existed) in place of the original VFS-provided dentry when they
 750 * differ.
 751 *
 752 * Gracefully handle the case where the MDS replies with -ENOENT and
 753 * no trace (which it may do, at its discretion, e.g., if it doesn't
 754 * care to issue a lease on the negative dentry).
 755 */
 756struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
 757				  struct dentry *dentry, int err)
 758{
 759	struct ceph_client *cl = req->r_mdsc->fsc->client;
 760
 761	if (err == -ENOENT) {
 762		/* no trace? */
 763		err = 0;
 764		if (!req->r_reply_info.head->is_dentry) {
 765			doutc(cl,
 766			      "ENOENT and no trace, dentry %p inode %llx.%llx\n",
 767			      dentry, ceph_vinop(d_inode(dentry)));
 768			if (d_really_is_positive(dentry)) {
 769				d_drop(dentry);
 770				err = -ENOENT;
 771			} else {
 772				d_add(dentry, NULL);
 773			}
 774		}
 775	}
 776	if (err)
 777		dentry = ERR_PTR(err);
 778	else if (dentry != req->r_dentry)
 779		dentry = dget(req->r_dentry);   /* we got spliced */
 780	else
 781		dentry = NULL;
 782	return dentry;
 783}
 784
 785static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
 786{
 787	return ceph_ino(inode) == CEPH_INO_ROOT &&
 788		strncmp(dentry->d_name.name, ".ceph", 5) == 0;
 789}
 790
 791/*
 792 * Look up a single dir entry.  If there is a lookup intent, inform
 793 * the MDS so that it gets our 'caps wanted' value in a single op.
 794 */
 795static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 796				  unsigned int flags)
 797{
 798	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dir->i_sb);
 799	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
 800	struct ceph_client *cl = fsc->client;
 801	struct ceph_mds_request *req;
 802	int op;
 803	int mask;
 804	int err;
 805
 806	doutc(cl, "%p %llx.%llx/'%pd' dentry %p\n", dir, ceph_vinop(dir),
 807	      dentry, dentry);
 808
 809	if (dentry->d_name.len > NAME_MAX)
 810		return ERR_PTR(-ENAMETOOLONG);
 811
 812	if (IS_ENCRYPTED(dir)) {
 813		bool had_key = fscrypt_has_encryption_key(dir);
 814
 815		err = fscrypt_prepare_lookup_partial(dir, dentry);
 816		if (err < 0)
 817			return ERR_PTR(err);
 818
 819		/* mark directory as incomplete if it has been unlocked */
 820		if (!had_key && fscrypt_has_encryption_key(dir))
 821			ceph_dir_clear_complete(dir);
 822	}
 823
 824	/* can we conclude ENOENT locally? */
 825	if (d_really_is_negative(dentry)) {
 826		struct ceph_inode_info *ci = ceph_inode(dir);
 827		struct ceph_dentry_info *di = ceph_dentry(dentry);
 828
 829		spin_lock(&ci->i_ceph_lock);
 830		doutc(cl, " dir %llx.%llx flags are 0x%lx\n",
 831		      ceph_vinop(dir), ci->i_ceph_flags);
 832		if (strncmp(dentry->d_name.name,
 833			    fsc->mount_options->snapdir_name,
 834			    dentry->d_name.len) &&
 835		    !is_root_ceph_dentry(dir, dentry) &&
 836		    ceph_test_mount_opt(fsc, DCACHE) &&
 837		    __ceph_dir_is_complete(ci) &&
 838		    __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
 839			__ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
 840			spin_unlock(&ci->i_ceph_lock);
 841			doutc(cl, " dir %llx.%llx complete, -ENOENT\n",
 842			      ceph_vinop(dir));
 843			d_add(dentry, NULL);
 844			di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
 845			return NULL;
 846		}
 847		spin_unlock(&ci->i_ceph_lock);
 848	}
 849
 850	op = ceph_snap(dir) == CEPH_SNAPDIR ?
 851		CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
 852	req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
 853	if (IS_ERR(req))
 854		return ERR_CAST(req);
 855	req->r_dentry = dget(dentry);
 856	req->r_num_caps = 2;
 857
 858	mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 859	if (ceph_security_xattr_wanted(dir))
 860		mask |= CEPH_CAP_XATTR_SHARED;
 861	req->r_args.getattr.mask = cpu_to_le32(mask);
 862
 863	ihold(dir);
 864	req->r_parent = dir;
 865	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 866	err = ceph_mdsc_do_request(mdsc, NULL, req);
 867	if (err == -ENOENT) {
 868		struct dentry *res;
 869
 870		res = ceph_handle_snapdir(req, dentry);
 871		if (IS_ERR(res)) {
 872			err = PTR_ERR(res);
 873		} else {
 874			dentry = res;
 875			err = 0;
 876		}
 877	}
 878	dentry = ceph_finish_lookup(req, dentry, err);
 879	ceph_mdsc_put_request(req);  /* will dput(dentry) */
 880	doutc(cl, "result=%p\n", dentry);
 881	return dentry;
 882}
 883
 884/*
 885 * If we do a create but get no trace back from the MDS, follow up with
 886 * a lookup (the VFS expects us to link up the provided dentry).
 887 */
 888int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
 889{
 890	struct dentry *result = ceph_lookup(dir, dentry, 0);
 891
 892	if (result && !IS_ERR(result)) {
 893		/*
 894		 * We created the item, then did a lookup, and found
 895		 * it was already linked to another inode we already
 896		 * had in our cache (and thus got spliced). To not
 897		 * confuse VFS (especially when inode is a directory),
 898		 * we don't link our dentry to that inode, return an
 899		 * error instead.
 900		 *
 901		 * This event should be rare and it happens only when
 902		 * we talk to old MDS. Recent MDS does not send traceless
 903		 * reply for request that creates new inode.
 904		 */
 905		d_drop(result);
 906		return -ESTALE;
 907	}
 908	return PTR_ERR(result);
 909}
 910
 911static int ceph_mknod(struct mnt_idmap *idmap, struct inode *dir,
 912		      struct dentry *dentry, umode_t mode, dev_t rdev)
 913{
 914	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
 915	struct ceph_client *cl = mdsc->fsc->client;
 916	struct ceph_mds_request *req;
 917	struct ceph_acl_sec_ctx as_ctx = {};
 918	int err;
 919
 920	if (ceph_snap(dir) != CEPH_NOSNAP)
 921		return -EROFS;
 922
 923	err = ceph_wait_on_conflict_unlink(dentry);
 924	if (err)
 925		return err;
 926
 927	if (ceph_quota_is_max_files_exceeded(dir)) {
 928		err = -EDQUOT;
 929		goto out;
 930	}
 931
 932	doutc(cl, "%p %llx.%llx/'%pd' dentry %p mode 0%ho rdev %d\n",
 933	      dir, ceph_vinop(dir), dentry, dentry, mode, rdev);
 934	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
 935	if (IS_ERR(req)) {
 936		err = PTR_ERR(req);
 937		goto out;
 938	}
 939
 940	req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
 941	if (IS_ERR(req->r_new_inode)) {
 942		err = PTR_ERR(req->r_new_inode);
 943		req->r_new_inode = NULL;
 944		goto out_req;
 945	}
 946
 947	if (S_ISREG(mode) && IS_ENCRYPTED(dir))
 948		set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
 949
 950	req->r_dentry = dget(dentry);
 951	req->r_num_caps = 2;
 952	req->r_parent = dir;
 953	ihold(dir);
 954	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 955	req->r_mnt_idmap = mnt_idmap_get(idmap);
 956	req->r_args.mknod.mode = cpu_to_le32(mode);
 957	req->r_args.mknod.rdev = cpu_to_le32(rdev);
 958	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL |
 959			     CEPH_CAP_XATTR_EXCL;
 960	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 961
 962	ceph_as_ctx_to_req(req, &as_ctx);
 963
 964	err = ceph_mdsc_do_request(mdsc, dir, req);
 965	if (!err && !req->r_reply_info.head->is_dentry)
 966		err = ceph_handle_notrace_create(dir, dentry);
 967out_req:
 968	ceph_mdsc_put_request(req);
 969out:
 970	if (!err)
 971		ceph_init_inode_acls(d_inode(dentry), &as_ctx);
 972	else
 973		d_drop(dentry);
 974	ceph_release_acl_sec_ctx(&as_ctx);
 975	return err;
 976}
 977
 978static int ceph_create(struct mnt_idmap *idmap, struct inode *dir,
 979		       struct dentry *dentry, umode_t mode, bool excl)
 980{
 981	return ceph_mknod(idmap, dir, dentry, mode, 0);
 982}
 983
 984#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
 985static int prep_encrypted_symlink_target(struct ceph_mds_request *req,
 986					 const char *dest)
 987{
 988	int err;
 989	int len = strlen(dest);
 990	struct fscrypt_str osd_link = FSTR_INIT(NULL, 0);
 991
 992	err = fscrypt_prepare_symlink(req->r_parent, dest, len, PATH_MAX,
 993				      &osd_link);
 994	if (err)
 995		goto out;
 996
 997	err = fscrypt_encrypt_symlink(req->r_new_inode, dest, len, &osd_link);
 998	if (err)
 999		goto out;
1000
1001	req->r_path2 = kmalloc(BASE64_CHARS(osd_link.len) + 1, GFP_KERNEL);
1002	if (!req->r_path2) {
1003		err = -ENOMEM;
1004		goto out;
1005	}
1006
1007	len = base64_encode(osd_link.name, osd_link.len,
1008			    req->r_path2, false, BASE64_IMAP);
1009	req->r_path2[len] = '\0';
1010out:
1011	fscrypt_fname_free_buffer(&osd_link);
1012	return err;
1013}
1014#else
1015static int prep_encrypted_symlink_target(struct ceph_mds_request *req,
1016					 const char *dest)
1017{
1018	return -EOPNOTSUPP;
1019}
1020#endif
1021
1022static int ceph_symlink(struct mnt_idmap *idmap, struct inode *dir,
1023			struct dentry *dentry, const char *dest)
1024{
1025	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1026	struct ceph_client *cl = mdsc->fsc->client;
1027	struct ceph_mds_request *req;
1028	struct ceph_acl_sec_ctx as_ctx = {};
1029	umode_t mode = S_IFLNK | 0777;
1030	int err;
1031
1032	if (ceph_snap(dir) != CEPH_NOSNAP)
1033		return -EROFS;
1034
1035	err = ceph_wait_on_conflict_unlink(dentry);
1036	if (err)
1037		return err;
1038
1039	if (ceph_quota_is_max_files_exceeded(dir)) {
1040		err = -EDQUOT;
1041		goto out;
1042	}
1043
1044	doutc(cl, "%p %llx.%llx/'%pd' to '%s'\n", dir, ceph_vinop(dir), dentry,
1045	      dest);
1046	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
1047	if (IS_ERR(req)) {
1048		err = PTR_ERR(req);
1049		goto out;
1050	}
1051
1052	req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
1053	if (IS_ERR(req->r_new_inode)) {
1054		err = PTR_ERR(req->r_new_inode);
1055		req->r_new_inode = NULL;
1056		goto out_req;
1057	}
1058
1059	req->r_parent = dir;
1060	ihold(dir);
1061
1062	if (IS_ENCRYPTED(req->r_new_inode)) {
1063		err = prep_encrypted_symlink_target(req, dest);
1064		if (err)
1065			goto out_req;
1066	} else {
1067		req->r_path2 = kstrdup(dest, GFP_KERNEL);
1068		if (!req->r_path2) {
1069			err = -ENOMEM;
1070			goto out_req;
1071		}
1072	}
1073
1074	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1075	req->r_mnt_idmap = mnt_idmap_get(idmap);
1076	req->r_dentry = dget(dentry);
1077	req->r_num_caps = 2;
1078	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL |
1079			     CEPH_CAP_XATTR_EXCL;
1080	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1081
1082	ceph_as_ctx_to_req(req, &as_ctx);
1083
1084	err = ceph_mdsc_do_request(mdsc, dir, req);
1085	if (!err && !req->r_reply_info.head->is_dentry)
1086		err = ceph_handle_notrace_create(dir, dentry);
1087out_req:
1088	ceph_mdsc_put_request(req);
1089out:
1090	if (err)
1091		d_drop(dentry);
1092	ceph_release_acl_sec_ctx(&as_ctx);
1093	return err;
1094}
1095
1096static struct dentry *ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
1097				 struct dentry *dentry, umode_t mode)
1098{
1099	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1100	struct ceph_client *cl = mdsc->fsc->client;
1101	struct ceph_mds_request *req;
1102	struct ceph_acl_sec_ctx as_ctx = {};
1103	struct dentry *ret;
1104	int err;
1105	int op;
1106
1107	err = ceph_wait_on_conflict_unlink(dentry);
1108	if (err)
1109		return ERR_PTR(err);
1110
1111	if (ceph_snap(dir) == CEPH_SNAPDIR) {
1112		/* mkdir .snap/foo is a MKSNAP */
1113		op = CEPH_MDS_OP_MKSNAP;
1114		doutc(cl, "mksnap %llx.%llx/'%pd' dentry %p\n",
1115		      ceph_vinop(dir), dentry, dentry);
1116	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
1117		doutc(cl, "mkdir %llx.%llx/'%pd' dentry %p mode 0%ho\n",
1118		      ceph_vinop(dir), dentry, dentry, mode);
1119		op = CEPH_MDS_OP_MKDIR;
1120	} else {
1121		ret = ERR_PTR(-EROFS);
1122		goto out;
1123	}
1124
1125	if (op == CEPH_MDS_OP_MKDIR &&
1126	    ceph_quota_is_max_files_exceeded(dir)) {
1127		ret = ERR_PTR(-EDQUOT);
1128		goto out;
1129	}
1130	if ((op == CEPH_MDS_OP_MKSNAP) && IS_ENCRYPTED(dir) &&
1131	    !fscrypt_has_encryption_key(dir)) {
1132		ret = ERR_PTR(-ENOKEY);
1133		goto out;
1134	}
1135
1136
1137	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1138	if (IS_ERR(req)) {
1139		ret = ERR_CAST(req);
1140		goto out;
1141	}
1142
1143	mode |= S_IFDIR;
1144	req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
1145	if (IS_ERR(req->r_new_inode)) {
1146		ret = ERR_CAST(req->r_new_inode);
1147		req->r_new_inode = NULL;
1148		goto out_req;
1149	}
1150
1151	req->r_dentry = dget(dentry);
1152	req->r_num_caps = 2;
1153	req->r_parent = dir;
1154	ihold(dir);
1155	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1156	if (op == CEPH_MDS_OP_MKDIR)
1157		req->r_mnt_idmap = mnt_idmap_get(idmap);
1158	req->r_args.mkdir.mode = cpu_to_le32(mode);
1159	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL |
1160			     CEPH_CAP_XATTR_EXCL;
1161	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1162
1163	ceph_as_ctx_to_req(req, &as_ctx);
1164
1165	err = ceph_mdsc_do_request(mdsc, dir, req);
1166	if (!err &&
1167	    !req->r_reply_info.head->is_target &&
1168	    !req->r_reply_info.head->is_dentry)
1169		err = ceph_handle_notrace_create(dir, dentry);
1170	ret = ERR_PTR(err);
1171out_req:
1172	if (!IS_ERR(ret) && req->r_dentry != dentry)
1173		/* Some other dentry was spliced in */
1174		ret = dget(req->r_dentry);
1175	ceph_mdsc_put_request(req);
1176out:
1177	if (!IS_ERR(ret)) {
1178		if (ret)
1179			dentry = ret;
1180		ceph_init_inode_acls(d_inode(dentry), &as_ctx);
1181	} else {
1182		d_drop(dentry);
1183	}
1184	ceph_release_acl_sec_ctx(&as_ctx);
1185	return ret;
1186}
1187
1188static int ceph_link(struct dentry *old_dentry, struct inode *dir,
1189		     struct dentry *dentry)
1190{
1191	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1192	struct ceph_client *cl = mdsc->fsc->client;
1193	struct ceph_mds_request *req;
1194	int err;
1195
1196	if (dentry->d_flags & DCACHE_DISCONNECTED)
1197		return -EINVAL;
1198
1199	err = ceph_wait_on_conflict_unlink(dentry);
1200	if (err)
1201		return err;
1202
1203	if (ceph_snap(dir) != CEPH_NOSNAP)
1204		return -EROFS;
1205
1206	err = fscrypt_prepare_link(old_dentry, dir, dentry);
1207	if (err)
1208		return err;
1209
1210	doutc(cl, "%p %llx.%llx/'%pd' to '%pd'\n", dir, ceph_vinop(dir),
1211	      old_dentry, dentry);
1212	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
1213	if (IS_ERR(req)) {
1214		d_drop(dentry);
1215		return PTR_ERR(req);
1216	}
1217	req->r_dentry = dget(dentry);
1218	req->r_num_caps = 2;
1219	req->r_old_dentry = dget(old_dentry);
1220	/*
1221	 * The old_dentry maybe a DCACHE_DISCONNECTED dentry, then we
1222	 * will just pass the ino# to MDSs.
1223	 */
1224	if (old_dentry->d_flags & DCACHE_DISCONNECTED)
1225		req->r_ino2 = ceph_vino(d_inode(old_dentry));
1226	req->r_parent = dir;
1227	ihold(dir);
1228	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1229	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1230	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1231	/* release LINK_SHARED on source inode (mds will lock it) */
1232	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1233	err = ceph_mdsc_do_request(mdsc, dir, req);
1234	if (err) {
1235		d_drop(dentry);
1236	} else if (!req->r_reply_info.head->is_dentry) {
1237		ihold(d_inode(old_dentry));
1238		d_instantiate(dentry, d_inode(old_dentry));
1239	}
1240	ceph_mdsc_put_request(req);
1241	return err;
1242}
1243
1244static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc,
1245				 struct ceph_mds_request *req)
1246{
1247	struct dentry *dentry = req->r_dentry;
1248	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
1249	struct ceph_client *cl = fsc->client;
1250	struct ceph_dentry_info *di = ceph_dentry(dentry);
1251	int result = req->r_err ? req->r_err :
1252			le32_to_cpu(req->r_reply_info.head->result);
1253
1254	if (!test_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags))
1255		pr_warn_client(cl,
1256			"dentry %p:%pd async unlink bit is not set\n",
1257			dentry, dentry);
1258
1259	spin_lock(&fsc->async_unlink_conflict_lock);
1260	hash_del_rcu(&di->hnode);
1261	spin_unlock(&fsc->async_unlink_conflict_lock);
1262
1263	spin_lock(&dentry->d_lock);
1264	clear_and_wake_up_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags);
1265	spin_unlock(&dentry->d_lock);
1266
1267	synchronize_rcu();
1268
1269	if (result == -EJUKEBOX)
1270		goto out;
1271
1272	/* If op failed, mark everyone involved for errors */
1273	if (result) {
1274		struct ceph_path_info path_info = {0};
1275		char *path = ceph_mdsc_build_path(mdsc, dentry, &path_info, 0);
1276
1277		/* mark error on parent + clear complete */
1278		mapping_set_error(req->r_parent->i_mapping, result);
1279		ceph_dir_clear_complete(req->r_parent);
1280
1281		/* drop the dentry -- we don't know its status */
1282		if (!d_unhashed(dentry))
1283			d_drop(dentry);
1284
1285		/* mark inode itself for an error (since metadata is bogus) */
1286		mapping_set_error(req->r_old_inode->i_mapping, result);
1287
1288		pr_warn_client(cl, "failure path=(%llx)%s result=%d!\n",
1289			       path_info.vino.ino, IS_ERR(path) ? "<<bad>>" : path, result);
1290		ceph_mdsc_free_path_info(&path_info);
1291	}
1292out:
1293	iput(req->r_old_inode);
1294	ceph_mdsc_release_dir_caps(req);
1295}
1296
1297static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry)
1298{
1299	struct ceph_inode_info *ci = ceph_inode(dir);
1300	struct ceph_dentry_info *di;
1301	int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK;
1302
1303	spin_lock(&ci->i_ceph_lock);
1304	if ((__ceph_caps_issued(ci, NULL) & want) == want) {
1305		ceph_take_cap_refs(ci, want, false);
1306		got = want;
1307	}
1308	spin_unlock(&ci->i_ceph_lock);
1309
1310	/* If we didn't get anything, return 0 */
1311	if (!got)
1312		return 0;
1313
1314        spin_lock(&dentry->d_lock);
1315        di = ceph_dentry(dentry);
1316	/*
1317	 * - We are holding Fx, which implies Fs caps.
1318	 * - Only support async unlink for primary linkage
1319	 */
1320	if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen ||
1321	    !(di->flags & CEPH_DENTRY_PRIMARY_LINK))
1322		want = 0;
1323        spin_unlock(&dentry->d_lock);
1324
1325	/* Do we still want what we've got? */
1326	if (want == got)
1327		return got;
1328
1329	ceph_put_cap_refs(ci, got);
1330	return 0;
1331}
1332
1333/*
1334 * rmdir and unlink are differ only by the metadata op code
1335 */
1336static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1337{
1338	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dir->i_sb);
1339	struct ceph_client *cl = fsc->client;
1340	struct ceph_mds_client *mdsc = fsc->mdsc;
1341	struct inode *inode = d_inode(dentry);
1342	struct ceph_inode_info *ci = ceph_inode(inode);
1343	struct ceph_mds_request *req;
1344	bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
1345	struct dentry *dn;
1346	int err = -EROFS;
1347	int op;
1348	char *path;
1349
1350	if (ceph_snap(dir) == CEPH_SNAPDIR) {
1351		/* rmdir .snap/foo is RMSNAP */
1352		doutc(cl, "rmsnap %llx.%llx/'%pd' dn\n", ceph_vinop(dir),
1353		      dentry);
1354		op = CEPH_MDS_OP_RMSNAP;
1355	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
1356		doutc(cl, "unlink/rmdir %llx.%llx/'%pd' inode %llx.%llx\n",
1357		      ceph_vinop(dir), dentry, ceph_vinop(inode));
1358		op = d_is_dir(dentry) ?
1359			CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1360	} else
1361		goto out;
1362
1363	dn = d_find_alias(dir);
1364	if (!dn) {
1365		try_async = false;
1366	} else {
1367		struct ceph_path_info path_info = {0};
1368		path = ceph_mdsc_build_path(mdsc, dn, &path_info, 0);
1369		if (IS_ERR(path)) {
1370			try_async = false;
1371			err = 0;
1372		} else {
1373			err = ceph_mds_check_access(mdsc, path, MAY_WRITE);
1374		}
1375		ceph_mdsc_free_path_info(&path_info);
1376		dput(dn);
1377
1378		/* For none EACCES cases will let the MDS do the mds auth check */
1379		if (err == -EACCES) {
1380			return err;
1381		} else if (err < 0) {
1382			try_async = false;
1383			err = 0;
1384		}
1385	}
1386
1387retry:
1388	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1389	if (IS_ERR(req)) {
1390		err = PTR_ERR(req);
1391		goto out;
1392	}
1393	req->r_dentry = dget(dentry);
1394	req->r_num_caps = 2;
1395	req->r_parent = dir;
1396	ihold(dir);
1397	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1398	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1399	req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1400
1401	if (try_async && op == CEPH_MDS_OP_UNLINK &&
1402	    (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) {
1403		struct ceph_dentry_info *di = ceph_dentry(dentry);
1404
1405		doutc(cl, "async unlink on %llx.%llx/'%pd' caps=%s",
1406		      ceph_vinop(dir), dentry,
1407		      ceph_cap_string(req->r_dir_caps));
1408		set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
1409		req->r_callback = ceph_async_unlink_cb;
1410		req->r_old_inode = d_inode(dentry);
1411		ihold(req->r_old_inode);
1412
1413		spin_lock(&dentry->d_lock);
1414		di->flags |= CEPH_DENTRY_ASYNC_UNLINK;
1415		spin_unlock(&dentry->d_lock);
1416
1417		spin_lock(&fsc->async_unlink_conflict_lock);
1418		hash_add_rcu(fsc->async_unlink_conflict, &di->hnode,
1419			     dentry->d_name.hash);
1420		spin_unlock(&fsc->async_unlink_conflict_lock);
1421
1422		err = ceph_mdsc_submit_request(mdsc, dir, req);
1423		if (!err) {
1424			/*
1425			 * We have enough caps, so we assume that the unlink
1426			 * will succeed. Fix up the target inode and dcache.
1427			 */
1428
1429			/*
1430			 * Protect the i_nlink update with i_ceph_lock
1431			 * to precent racing against ceph_fill_inode()
1432			 * handling our completion on a worker thread
1433			 * and don't decrement if i_nlink has already
1434			 * been updated to zero by this completion.
1435			 */
1436			spin_lock(&ci->i_ceph_lock);
1437			if (inode->i_nlink > 0)
1438				drop_nlink(inode);
1439			spin_unlock(&ci->i_ceph_lock);
1440
1441			d_delete(dentry);
1442		} else {
1443			spin_lock(&fsc->async_unlink_conflict_lock);
1444			hash_del_rcu(&di->hnode);
1445			spin_unlock(&fsc->async_unlink_conflict_lock);
1446
1447			spin_lock(&dentry->d_lock);
1448			di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK;
1449			spin_unlock(&dentry->d_lock);
1450
1451			if (err == -EJUKEBOX) {
1452				try_async = false;
1453				ceph_mdsc_put_request(req);
1454				goto retry;
1455			}
1456		}
1457	} else {
1458		set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1459		err = ceph_mdsc_do_request(mdsc, dir, req);
1460		if (!err && !req->r_reply_info.head->is_dentry)
1461			d_delete(dentry);
1462	}
1463
1464	ceph_mdsc_put_request(req);
1465out:
1466	return err;
1467}
1468
1469static int ceph_rename(struct mnt_idmap *idmap, struct inode *old_dir,
1470		       struct dentry *old_dentry, struct inode *new_dir,
1471		       struct dentry *new_dentry, unsigned int flags)
1472{
1473	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old_dir->i_sb);
1474	struct ceph_client *cl = mdsc->fsc->client;
1475	struct ceph_mds_request *req;
1476	int op = CEPH_MDS_OP_RENAME;
1477	int err;
1478
1479	if (flags)
1480		return -EINVAL;
1481
1482	if (ceph_snap(old_dir) != ceph_snap(new_dir))
1483		return -EXDEV;
1484	if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1485		if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1486			op = CEPH_MDS_OP_RENAMESNAP;
1487		else
1488			return -EROFS;
1489	}
1490	/* don't allow cross-quota renames */
1491	if ((old_dir != new_dir) &&
1492	    (!ceph_quota_is_same_realm(old_dir, new_dir)))
1493		return -EXDEV;
1494
1495	err = ceph_wait_on_conflict_unlink(new_dentry);
1496	if (err)
1497		return err;
1498
1499	err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
1500				     flags);
1501	if (err)
1502		return err;
1503
1504	doutc(cl, "%llx.%llx/'%pd' to %llx.%llx/'%pd'\n",
1505	      ceph_vinop(old_dir), old_dentry, ceph_vinop(new_dir),
1506	      new_dentry);
1507	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1508	if (IS_ERR(req))
1509		return PTR_ERR(req);
1510	ihold(old_dir);
1511	req->r_dentry = dget(new_dentry);
1512	req->r_num_caps = 2;
1513	req->r_old_dentry = dget(old_dentry);
1514	req->r_old_dentry_dir = old_dir;
1515	req->r_parent = new_dir;
1516	ihold(new_dir);
1517	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1518	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1519	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1520	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1521	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1522	/* release LINK_RDCACHE on source inode (mds will lock it) */
1523	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1524	if (d_really_is_positive(new_dentry)) {
1525		req->r_inode_drop =
1526			ceph_drop_caps_for_unlink(d_inode(new_dentry));
1527	}
1528	err = ceph_mdsc_do_request(mdsc, old_dir, req);
1529	if (!err && !req->r_reply_info.head->is_dentry) {
1530		/*
1531		 * Normally d_move() is done by fill_trace (called by
1532		 * do_request, above).  If there is no trace, we need
1533		 * to do it here.
1534		 */
1535		d_move(old_dentry, new_dentry);
1536	}
1537	ceph_mdsc_put_request(req);
1538	return err;
1539}
1540
1541/*
1542 * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
1543 * Leases at front of the list will expire first. (Assume all leases have
1544 * similar duration)
1545 *
1546 * Called under dentry->d_lock.
1547 */
1548void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
1549{
1550	struct dentry *dn = di->dentry;
1551	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dn->d_sb)->mdsc;
1552	struct ceph_client *cl = mdsc->fsc->client;
1553
1554	doutc(cl, "%p %p '%pd'\n", di, dn, dn);
1555
1556	di->flags |= CEPH_DENTRY_LEASE_LIST;
1557	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1558		di->flags |= CEPH_DENTRY_REFERENCED;
1559		return;
1560	}
1561
1562	spin_lock(&mdsc->dentry_list_lock);
1563	list_move_tail(&di->lease_list, &mdsc->dentry_leases);
1564	spin_unlock(&mdsc->dentry_list_lock);
1565}
1566
1567static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
1568				     struct ceph_dentry_info *di)
1569{
1570	di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
1571	di->lease_gen = 0;
1572	di->time = jiffies;
1573	list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
1574}
1575
1576/*
1577 * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
1578 * list if it's not in the list, otherwise set 'referenced' flag.
1579 *
1580 * Called under dentry->d_lock.
1581 */
1582void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
1583{
1584	struct dentry *dn = di->dentry;
1585	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dn->d_sb)->mdsc;
1586	struct ceph_client *cl = mdsc->fsc->client;
1587
1588	doutc(cl, "%p %p '%pd' (offset 0x%llx)\n", di, dn, dn, di->offset);
1589
1590	if (!list_empty(&di->lease_list)) {
1591		if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1592			/* don't remove dentry from dentry lease list
1593			 * if its lease is valid */
1594			if (__dentry_lease_is_valid(di))
1595				return;
1596		} else {
1597			di->flags |= CEPH_DENTRY_REFERENCED;
1598			return;
1599		}
1600	}
1601
1602	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1603		di->flags |= CEPH_DENTRY_REFERENCED;
1604		di->flags &= ~CEPH_DENTRY_LEASE_LIST;
1605		return;
1606	}
1607
1608	spin_lock(&mdsc->dentry_list_lock);
1609	__dentry_dir_lease_touch(mdsc, di);
1610	spin_unlock(&mdsc->dentry_list_lock);
1611}
1612
1613static void __dentry_lease_unlist(struct ceph_dentry_info *di)
1614{
1615	struct ceph_mds_client *mdsc;
1616	if (di->flags & CEPH_DENTRY_SHRINK_LIST)
1617		return;
1618	if (list_empty(&di->lease_list))
1619		return;
1620
1621	mdsc = ceph_sb_to_fs_client(di->dentry->d_sb)->mdsc;
1622	spin_lock(&mdsc->dentry_list_lock);
1623	list_del_init(&di->lease_list);
1624	spin_unlock(&mdsc->dentry_list_lock);
1625}
1626
1627enum {
1628	KEEP	= 0,
1629	DELETE	= 1,
1630	TOUCH	= 2,
1631	STOP	= 4,
1632};
1633
1634struct ceph_lease_walk_control {
1635	bool dir_lease;
1636	bool expire_dir_lease;
1637	unsigned long nr_to_scan;
1638	unsigned long dir_lease_ttl;
1639};
1640
1641static int __dir_lease_check(const struct dentry *, struct ceph_lease_walk_control *);
1642static int __dentry_lease_check(const struct dentry *);
1643
1644static unsigned long
1645__dentry_leases_walk(struct ceph_mds_client *mdsc,
1646		     struct ceph_lease_walk_control *lwc)
1647{
1648	struct ceph_dentry_info *di, *tmp;
1649	struct dentry *dentry, *last = NULL;
1650	struct list_head* list;
1651        LIST_HEAD(dispose);
1652	unsigned long freed = 0;
1653	int ret = 0;
1654
1655	list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
1656	spin_lock(&mdsc->dentry_list_lock);
1657	list_for_each_entry_safe(di, tmp, list, lease_list) {
1658		if (!lwc->nr_to_scan)
1659			break;
1660		--lwc->nr_to_scan;
1661
1662		dentry = di->dentry;
1663		if (last == dentry)
1664			break;
1665
1666		if (!spin_trylock(&dentry->d_lock))
1667			continue;
1668
1669		if (__lockref_is_dead(&dentry->d_lockref)) {
1670			list_del_init(&di->lease_list);
1671			goto next;
1672		}
1673
1674		if (lwc->dir_lease)
1675			ret = __dir_lease_check(dentry, lwc);
1676		else
1677			ret = __dentry_lease_check(dentry);
1678		if (ret & TOUCH) {
1679			/* move it into tail of dir lease list */
1680			__dentry_dir_lease_touch(mdsc, di);
1681			if (!last)
1682				last = dentry;
1683		}
1684		if (ret & DELETE) {
1685			/* stale lease */
1686			di->flags &= ~CEPH_DENTRY_REFERENCED;
1687			if (dentry->d_lockref.count > 0) {
1688				/* update_dentry_lease() will re-add
1689				 * it to lease list, or
1690				 * ceph_d_delete() will return 1 when
1691				 * last reference is dropped */
1692				list_del_init(&di->lease_list);
1693			} else {
1694				di->flags |= CEPH_DENTRY_SHRINK_LIST;
1695				list_move_tail(&di->lease_list, &dispose);
1696				dget_dlock(dentry);
1697			}
1698		}
1699next:
1700		spin_unlock(&dentry->d_lock);
1701		if (ret & STOP)
1702			break;
1703	}
1704	spin_unlock(&mdsc->dentry_list_lock);
1705
1706	while (!list_empty(&dispose)) {
1707		di = list_first_entry(&dispose, struct ceph_dentry_info,
1708				      lease_list);
1709		dentry = di->dentry;
1710		spin_lock(&dentry->d_lock);
1711
1712		list_del_init(&di->lease_list);
1713		di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
1714		if (di->flags & CEPH_DENTRY_REFERENCED) {
1715			spin_lock(&mdsc->dentry_list_lock);
1716			if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1717				list_add_tail(&di->lease_list,
1718					      &mdsc->dentry_leases);
1719			} else {
1720				__dentry_dir_lease_touch(mdsc, di);
1721			}
1722			spin_unlock(&mdsc->dentry_list_lock);
1723		} else {
1724			freed++;
1725		}
1726
1727		spin_unlock(&dentry->d_lock);
1728		/* ceph_d_delete() does the trick */
1729		dput(dentry);
1730	}
1731	return freed;
1732}
1733
1734static int __dentry_lease_check(const struct dentry *dentry)
1735{
1736	struct ceph_dentry_info *di = ceph_dentry(dentry);
1737	int ret;
1738
1739	if (__dentry_lease_is_valid(di))
1740		return STOP;
1741	ret = __dir_lease_try_check(dentry);
1742	if (ret == -EBUSY)
1743		return KEEP;
1744	if (ret > 0)
1745		return TOUCH;
1746	return DELETE;
1747}
1748
1749static int __dir_lease_check(const struct dentry *dentry,
1750			     struct ceph_lease_walk_control *lwc)
1751{
1752	struct ceph_dentry_info *di = ceph_dentry(dentry);
1753
1754	int ret = __dir_lease_try_check(dentry);
1755	if (ret == -EBUSY)
1756		return KEEP;
1757	if (ret > 0) {
1758		if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
1759			return STOP;
1760		/* Move dentry to tail of dir lease list if we don't want
1761		 * to delete it. So dentries in the list are checked in a
1762		 * round robin manner */
1763		if (!lwc->expire_dir_lease)
1764			return TOUCH;
1765		if (dentry->d_lockref.count > 0 ||
1766		    (di->flags & CEPH_DENTRY_REFERENCED))
1767			return TOUCH;
1768		/* invalidate dir lease */
1769		di->lease_shared_gen = 0;
1770	}
1771	return DELETE;
1772}
1773
1774int ceph_trim_dentries(struct ceph_mds_client *mdsc)
1775{
1776	struct ceph_lease_walk_control lwc;
1777	unsigned long count;
1778	unsigned long freed;
1779
1780	spin_lock(&mdsc->caps_list_lock);
1781        if (mdsc->caps_use_max > 0 &&
1782            mdsc->caps_use_count > mdsc->caps_use_max)
1783		count = mdsc->caps_use_count - mdsc->caps_use_max;
1784	else
1785		count = 0;
1786        spin_unlock(&mdsc->caps_list_lock);
1787
1788	lwc.dir_lease = false;
1789	lwc.nr_to_scan  = CEPH_CAPS_PER_RELEASE * 2;
1790	freed = __dentry_leases_walk(mdsc, &lwc);
1791	if (!lwc.nr_to_scan) /* more invalid leases */
1792		return -EAGAIN;
1793
1794	if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
1795		lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
1796
1797	lwc.dir_lease = true;
1798	lwc.expire_dir_lease = freed < count;
1799	lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ;
1800	freed +=__dentry_leases_walk(mdsc, &lwc);
1801	if (!lwc.nr_to_scan) /* more to check */
1802		return -EAGAIN;
1803
1804	return freed > 0 ? 1 : 0;
1805}
1806
1807/*
1808 * Ensure a dentry lease will no longer revalidate.
1809 */
1810void ceph_invalidate_dentry_lease(struct dentry *dentry)
1811{
1812	struct ceph_dentry_info *di = ceph_dentry(dentry);
1813	spin_lock(&dentry->d_lock);
1814	di->time = jiffies;
1815	di->lease_shared_gen = 0;
1816	di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1817	__dentry_lease_unlist(di);
1818	spin_unlock(&dentry->d_lock);
1819}
1820
1821/*
1822 * Check if dentry lease is valid.  If not, delete the lease.  Try to
1823 * renew if the least is more than half up.
1824 */
1825static bool __dentry_lease_is_valid(struct ceph_dentry_info *di)
1826{
1827	struct ceph_mds_session *session;
1828
1829	if (!di->lease_gen)
1830		return false;
1831
1832	session = di->lease_session;
1833	if (session) {
1834		u32 gen;
1835		unsigned long ttl;
1836
1837		gen = atomic_read(&session->s_cap_gen);
1838		ttl = session->s_cap_ttl;
1839
1840		if (di->lease_gen == gen &&
1841		    time_before(jiffies, ttl) &&
1842		    time_before(jiffies, di->time))
1843			return true;
1844	}
1845	di->lease_gen = 0;
1846	return false;
1847}
1848
1849static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags)
1850{
1851	struct ceph_dentry_info *di;
1852	struct ceph_mds_session *session = NULL;
1853	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dentry->d_sb)->mdsc;
1854	struct ceph_client *cl = mdsc->fsc->client;
1855	u32 seq = 0;
1856	int valid = 0;
1857
1858	spin_lock(&dentry->d_lock);
1859	di = ceph_dentry(dentry);
1860	if (di && __dentry_lease_is_valid(di)) {
1861		valid = 1;
1862
1863		if (di->lease_renew_after &&
1864		    time_after(jiffies, di->lease_renew_after)) {
1865			/*
1866			 * We should renew. If we're in RCU walk mode
1867			 * though, we can't do that so just return
1868			 * -ECHILD.
1869			 */
1870			if (flags & LOOKUP_RCU) {
1871				valid = -ECHILD;
1872			} else {
1873				session = ceph_get_mds_session(di->lease_session);
1874				seq = di->lease_seq;
1875				di->lease_renew_after = 0;
1876				di->lease_renew_from = jiffies;
1877			}
1878		}
1879	}
1880	spin_unlock(&dentry->d_lock);
1881
1882	if (session) {
1883		ceph_mdsc_lease_send_msg(session, dentry,
1884					 CEPH_MDS_LEASE_RENEW, seq);
1885		ceph_put_mds_session(session);
1886	}
1887	doutc(cl, "dentry %p = %d\n", dentry, valid);
1888	return valid;
1889}
1890
1891/*
1892 * Called under dentry->d_lock.
1893 */
1894static int __dir_lease_try_check(const struct dentry *dentry)
1895{
1896	struct ceph_dentry_info *di = ceph_dentry(dentry);
1897	struct inode *dir;
1898	struct ceph_inode_info *ci;
1899	int valid = 0;
1900
1901	if (!di->lease_shared_gen)
1902		return 0;
1903	if (IS_ROOT(dentry))
1904		return 0;
1905
1906	dir = d_inode(dentry->d_parent);
1907	ci = ceph_inode(dir);
1908
1909	if (spin_trylock(&ci->i_ceph_lock)) {
1910		if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen &&
1911		    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0))
1912			valid = 1;
1913		spin_unlock(&ci->i_ceph_lock);
1914	} else {
1915		valid = -EBUSY;
1916	}
1917
1918	if (!valid)
1919		di->lease_shared_gen = 0;
1920	return valid;
1921}
1922
1923/*
1924 * Check if directory-wide content lease/cap is valid.
1925 */
1926static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry,
1927			      struct ceph_mds_client *mdsc)
1928{
1929	struct ceph_inode_info *ci = ceph_inode(dir);
1930	struct ceph_client *cl = mdsc->fsc->client;
1931	int valid;
1932	int shared_gen;
1933
1934	spin_lock(&ci->i_ceph_lock);
1935	valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1936	if (valid) {
1937		__ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
1938		shared_gen = atomic_read(&ci->i_shared_gen);
1939	}
1940	spin_unlock(&ci->i_ceph_lock);
1941	if (valid) {
1942		struct ceph_dentry_info *di;
1943		spin_lock(&dentry->d_lock);
1944		di = ceph_dentry(dentry);
1945		if (dir == d_inode(dentry->d_parent) &&
1946		    di && di->lease_shared_gen == shared_gen)
1947			__ceph_dentry_dir_lease_touch(di);
1948		else
1949			valid = 0;
1950		spin_unlock(&dentry->d_lock);
1951	}
1952	doutc(cl, "dir %p %llx.%llx v%u dentry %p '%pd' = %d\n", dir,
1953	      ceph_vinop(dir), (unsigned)atomic_read(&ci->i_shared_gen),
1954	      dentry, dentry, valid);
1955	return valid;
1956}
1957
1958/*
1959 * Check if cached dentry can be trusted.
1960 */
1961static int ceph_d_revalidate(struct inode *dir, const struct qstr *name,
1962			     struct dentry *dentry, unsigned int flags)
1963{
1964	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dentry->d_sb)->mdsc;
1965	struct ceph_client *cl = mdsc->fsc->client;
1966	int valid = 0;
1967	struct inode *inode;
1968
1969	valid = fscrypt_d_revalidate(dir, name, dentry, flags);
1970	if (valid <= 0)
1971		return valid;
1972
1973	inode = d_inode_rcu(dentry);
1974
1975	doutc(cl, "%p '%pd' inode %p offset 0x%llx nokey %d\n",
1976	      dentry, dentry, inode, ceph_dentry(dentry)->offset,
1977	      !!(dentry->d_flags & DCACHE_NOKEY_NAME));
1978
1979	mdsc = ceph_sb_to_fs_client(dir->i_sb)->mdsc;
1980
1981	/* always trust cached snapped dentries, snapdir dentry */
1982	if (ceph_snap(dir) != CEPH_NOSNAP) {
1983		doutc(cl, "%p '%pd' inode %p is SNAPPED\n", dentry,
1984		      dentry, inode);
1985		valid = 1;
1986	} else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1987		valid = 1;
1988	} else {
1989		valid = dentry_lease_is_valid(dentry, flags);
1990		if (valid == -ECHILD)
1991			return valid;
1992		if (valid || dir_lease_is_valid(dir, dentry, mdsc)) {
1993			if (inode)
1994				valid = ceph_is_any_caps(inode);
1995			else
1996				valid = 1;
1997		}
1998	}
1999
2000	if (!valid) {
2001		struct ceph_mds_request *req;
2002		int op, err;
2003		u32 mask;
2004
2005		if (flags & LOOKUP_RCU)
2006			return -ECHILD;
2007
2008		percpu_counter_inc(&mdsc->metric.d_lease_mis);
2009
2010		op = ceph_snap(dir) == CEPH_SNAPDIR ?
2011			CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
2012		req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
2013		if (!IS_ERR(req)) {
2014			req->r_dentry = dget(dentry);
2015			req->r_num_caps = 2;
2016			req->r_parent = dir;
2017			ihold(dir);
2018
2019			req->r_dname = name;
2020
2021			mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
2022			if (ceph_security_xattr_wanted(dir))
2023				mask |= CEPH_CAP_XATTR_SHARED;
2024			req->r_args.getattr.mask = cpu_to_le32(mask);
2025
2026			err = ceph_mdsc_do_request(mdsc, NULL, req);
2027			switch (err) {
2028			case 0:
2029				if (d_really_is_positive(dentry) &&
2030				    d_inode(dentry) == req->r_target_inode)
2031					valid = 1;
2032				break;
2033			case -ENOENT:
2034				if (d_really_is_negative(dentry))
2035					valid = 1;
2036				fallthrough;
2037			default:
2038				break;
2039			}
2040			ceph_mdsc_put_request(req);
2041			doutc(cl, "%p '%pd', lookup result=%d\n", dentry,
2042			      dentry, err);
2043		}
2044	} else {
2045		percpu_counter_inc(&mdsc->metric.d_lease_hit);
2046	}
2047
2048	doutc(cl, "%p '%pd' %s\n", dentry, dentry, valid ? "valid" : "invalid");
2049	if (!valid)
2050		ceph_dir_clear_complete(dir);
2051	return valid;
2052}
2053
2054/*
2055 * Delete unused dentry that doesn't have valid lease
2056 *
2057 * Called under dentry->d_lock.
2058 */
2059static int ceph_d_delete(const struct dentry *dentry)
2060{
2061	struct ceph_dentry_info *di;
2062
2063	/* won't release caps */
2064	if (d_really_is_negative(dentry))
2065		return 0;
2066	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
2067		return 0;
2068	/* valid lease? */
2069	di = ceph_dentry(dentry);
2070	if (di) {
2071		if (__dentry_lease_is_valid(di))
2072			return 0;
2073		if (__dir_lease_try_check(dentry))
2074			return 0;
2075	}
2076	return 1;
2077}
2078
2079/*
2080 * Release our ceph_dentry_info.
2081 */
2082static void ceph_d_release(struct dentry *dentry)
2083{
2084	struct ceph_dentry_info *di = ceph_dentry(dentry);
2085	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
2086
2087	doutc(fsc->client, "dentry %p '%pd'\n", dentry, dentry);
2088
2089	atomic64_dec(&fsc->mdsc->metric.total_dentries);
2090
2091	spin_lock(&dentry->d_lock);
2092	__dentry_lease_unlist(di);
2093	dentry->d_fsdata = NULL;
2094	spin_unlock(&dentry->d_lock);
2095
2096	ceph_put_mds_session(di->lease_session);
2097	kmem_cache_free(ceph_dentry_cachep, di);
2098}
2099
2100/*
2101 * When the VFS prunes a dentry from the cache, we need to clear the
2102 * complete flag on the parent directory.
2103 *
2104 * Called under dentry->d_lock.
2105 */
2106static void ceph_d_prune(struct dentry *dentry)
2107{
2108	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb);
2109	struct ceph_client *cl = mdsc->fsc->client;
2110	struct ceph_inode_info *dir_ci;
2111	struct ceph_dentry_info *di;
2112
2113	doutc(cl, "dentry %p '%pd'\n", dentry, dentry);
2114
2115	/* do we have a valid parent? */
2116	if (IS_ROOT(dentry))
2117		return;
2118
2119	/* we hold d_lock, so d_parent is stable */
2120	dir_ci = ceph_inode(d_inode(dentry->d_parent));
2121	if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
2122		return;
2123
2124	/* who calls d_delete() should also disable dcache readdir */
2125	if (d_really_is_negative(dentry))
2126		return;
2127
2128	/* d_fsdata does not get cleared until d_release */
2129	if (!d_unhashed(dentry)) {
2130		__ceph_dir_clear_complete(dir_ci);
2131		return;
2132	}
2133
2134	/* Disable dcache readdir just in case that someone called d_drop()
2135	 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
2136	 * properly (dcache readdir is still enabled) */
2137	di = ceph_dentry(dentry);
2138	if (di->offset > 0 &&
2139	    di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
2140		__ceph_dir_clear_ordered(dir_ci);
2141}
2142
2143/*
2144 * read() on a dir.  This weird interface hack only works if mounted
2145 * with '-o dirstat'.
2146 */
2147static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
2148			     loff_t *ppos)
2149{
2150	struct ceph_dir_file_info *dfi = file->private_data;
2151	struct inode *inode = file_inode(file);
2152	struct ceph_inode_info *ci = ceph_inode(inode);
2153	int left;
2154	const int bufsize = 1024;
2155
2156	if (!ceph_test_mount_opt(ceph_sb_to_fs_client(inode->i_sb), DIRSTAT))
2157		return -EISDIR;
2158
2159	if (!dfi->dir_info) {
2160		dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
2161		if (!dfi->dir_info)
2162			return -ENOMEM;
2163		dfi->dir_info_len =
2164			snprintf(dfi->dir_info, bufsize,
2165				"entries:   %20lld\n"
2166				" files:    %20lld\n"
2167				" subdirs:  %20lld\n"
2168				"rentries:  %20lld\n"
2169				" rfiles:   %20lld\n"
2170				" rsubdirs: %20lld\n"
2171				"rbytes:    %20lld\n"
2172				"rctime:    %ptSp\n",
2173				ci->i_files + ci->i_subdirs,
2174				ci->i_files,
2175				ci->i_subdirs,
2176				ci->i_rfiles + ci->i_rsubdirs,
2177				ci->i_rfiles,
2178				ci->i_rsubdirs,
2179				ci->i_rbytes,
2180				&ci->i_rctime);
2181	}
2182
2183	if (*ppos >= dfi->dir_info_len)
2184		return 0;
2185	size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
2186	left = copy_to_user(buf, dfi->dir_info + *ppos, size);
2187	if (left == size)
2188		return -EFAULT;
2189	*ppos += (size - left);
2190	return size - left;
2191}
2192
2193
2194
2195/*
2196 * Return name hash for a given dentry.  This is dependent on
2197 * the parent directory's hash function.
2198 */
2199unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
2200{
2201	struct ceph_inode_info *dci = ceph_inode(dir);
2202	unsigned hash;
2203
2204	switch (dci->i_dir_layout.dl_dir_hash) {
2205	case 0:	/* for backward compat */
2206	case CEPH_STR_HASH_LINUX:
2207		return dn->d_name.hash;
2208
2209	default:
2210		spin_lock(&dn->d_lock);
2211		hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
2212				     dn->d_name.name, dn->d_name.len);
2213		spin_unlock(&dn->d_lock);
2214		return hash;
2215	}
2216}
2217
2218WRAP_DIR_ITER(ceph_readdir) // FIXME!
2219const struct file_operations ceph_dir_fops = {
2220	.read = ceph_read_dir,
2221	.iterate_shared = shared_ceph_readdir,
2222	.llseek = ceph_dir_llseek,
2223	.open = ceph_open,
2224	.release = ceph_release,
2225	.unlocked_ioctl = ceph_ioctl,
2226	.compat_ioctl = compat_ptr_ioctl,
2227	.fsync = ceph_fsync,
2228	.lock = ceph_lock,
2229	.flock = ceph_flock,
2230	.setlease = simple_nosetlease,
2231};
2232
2233const struct file_operations ceph_snapdir_fops = {
2234	.iterate_shared = shared_ceph_readdir,
2235	.llseek = ceph_dir_llseek,
2236	.open = ceph_open,
2237	.release = ceph_release,
2238	.setlease = simple_nosetlease,
2239};
2240
2241const struct inode_operations ceph_dir_iops = {
2242	.lookup = ceph_lookup,
2243	.permission = ceph_permission,
2244	.getattr = ceph_getattr,
2245	.setattr = ceph_setattr,
2246	.listxattr = ceph_listxattr,
2247	.get_inode_acl = ceph_get_acl,
2248	.set_acl = ceph_set_acl,
2249	.mknod = ceph_mknod,
2250	.symlink = ceph_symlink,
2251	.mkdir = ceph_mkdir,
2252	.link = ceph_link,
2253	.unlink = ceph_unlink,
2254	.rmdir = ceph_unlink,
2255	.rename = ceph_rename,
2256	.create = ceph_create,
2257	.atomic_open = ceph_atomic_open,
2258};
2259
2260const struct inode_operations ceph_snapdir_iops = {
2261	.lookup = ceph_lookup,
2262	.permission = ceph_permission,
2263	.getattr = ceph_getattr,
2264	.mkdir = ceph_mkdir,
2265	.rmdir = ceph_unlink,
2266	.rename = ceph_rename,
2267};
2268
2269const struct dentry_operations ceph_dentry_ops = {
2270	.d_revalidate = ceph_d_revalidate,
2271	.d_delete = ceph_d_delete,
2272	.d_release = ceph_d_release,
2273	.d_prune = ceph_d_prune,
2274	.d_init = ceph_d_init,
2275};