inode.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * (C) 2001 Clemson University and The University of Chicago
   4 * Copyright 2018 Omnibond Systems, L.L.C.
   5 *
   6 * See COPYING in top-level directory.
   7 */
   8
   9/*
  10 *  Linux VFS inode operations.
  11 */
  12
  13#include <linux/blkdev.h>
  14#include <linux/fileattr.h>
  15#include "protocol.h"
  16#include "orangefs-kernel.h"
  17#include "orangefs-bufmap.h"
  18
  19static int orangefs_writepage_locked(struct folio *folio,
  20		struct writeback_control *wbc)
  21{
  22	struct inode *inode = folio->mapping->host;
  23	struct orangefs_write_range *wr = NULL;
  24	struct iov_iter iter;
  25	struct bio_vec bv;
  26	size_t wlen;
  27	ssize_t ret;
  28	loff_t len, off;
  29
  30	folio_start_writeback(folio);
  31
  32	len = i_size_read(inode);
  33	if (folio->private) {
  34		wr = folio->private;
  35		off = wr->pos;
  36		if ((off + wr->len > len) && (off <= len))
  37			wlen = len - off;
  38		else
  39			wlen = wr->len;
  40		if (wlen == 0)
  41			wlen = wr->len;
  42	} else {
  43		WARN_ON(1);
  44		off = folio_pos(folio);
  45		wlen = folio_size(folio);
  46
  47		if (wlen > len - off)
  48			wlen = len - off;
  49	}
  50
  51	WARN_ON(wlen == 0);
  52	bvec_set_folio(&bv, folio, wlen, offset_in_folio(folio, off));
  53	iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, wlen);
  54
  55	ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
  56	    len, wr, NULL, NULL);
  57	if (ret < 0) {
  58		mapping_set_error(folio->mapping, ret);
  59	} else {
  60		ret = 0;
  61	}
  62	kfree(folio_detach_private(folio));
  63	return ret;
  64}
  65
  66struct orangefs_writepages {
  67	loff_t off;
  68	size_t len;
  69	kuid_t uid;
  70	kgid_t gid;
  71	int maxpages;
  72	int nfolios;
  73	struct address_space *mapping;
  74	struct folio **folios;
  75	struct bio_vec *bv;
  76};
  77
  78static int orangefs_writepages_work(struct orangefs_writepages *ow,
  79		struct writeback_control *wbc)
  80{
  81	struct inode *inode = ow->mapping->host;
  82	struct orangefs_write_range *wrp, wr;
  83	struct iov_iter iter;
  84	ssize_t ret;
  85	size_t start;
  86	loff_t len, off;
  87	int i;
  88
  89	len = i_size_read(inode);
  90
  91	start = offset_in_folio(ow->folios[0], ow->off);
  92	for (i = 0; i < ow->nfolios; i++) {
  93		folio_start_writeback(ow->folios[i]);
  94		bvec_set_folio(&ow->bv[i], ow->folios[i],
  95				folio_size(ow->folios[i]) - start, start);
  96		start = 0;
  97	}
  98	iov_iter_bvec(&iter, ITER_SOURCE, ow->bv, ow->nfolios, ow->len);
  99
 100	WARN_ON(ow->off >= len);
 101	if (ow->off + ow->len > len)
 102		ow->len = len - ow->off;
 103
 104	off = ow->off;
 105	wr.uid = ow->uid;
 106	wr.gid = ow->gid;
 107	ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len,
 108	    0, &wr, NULL, NULL);
 109	if (ret < 0)
 110		mapping_set_error(ow->mapping, ret);
 111	else
 112		ret = 0;
 113
 114	for (i = 0; i < ow->nfolios; i++) {
 115		wrp = folio_detach_private(ow->folios[i]);
 116		kfree(wrp);
 117		folio_end_writeback(ow->folios[i]);
 118		folio_unlock(ow->folios[i]);
 119	}
 120
 121	return ret;
 122}
 123
 124static int orangefs_writepages_callback(struct folio *folio,
 125		struct writeback_control *wbc, struct orangefs_writepages *ow)
 126{
 127	struct orangefs_write_range *wr = folio->private;
 128	int ret;
 129
 130	if (!wr) {
 131		folio_unlock(folio);
 132		/* It's not private so there's nothing to write, right? */
 133		printk("writepages_callback not private!\n");
 134		BUG();
 135		return 0;
 136	}
 137
 138	ret = -1;
 139	if (ow->nfolios == 0) {
 140		ow->off = wr->pos;
 141		ow->len = wr->len;
 142		ow->uid = wr->uid;
 143		ow->gid = wr->gid;
 144		ow->folios[ow->nfolios++] = folio;
 145		ret = 0;
 146		goto done;
 147	}
 148	if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
 149		orangefs_writepages_work(ow, wbc);
 150		ow->nfolios = 0;
 151		ret = -1;
 152		goto done;
 153	}
 154	if (ow->off + ow->len == wr->pos) {
 155		ow->len += wr->len;
 156		ow->folios[ow->nfolios++] = folio;
 157		ret = 0;
 158		goto done;
 159	}
 160done:
 161	if (ret == -1) {
 162		if (ow->nfolios) {
 163			orangefs_writepages_work(ow, wbc);
 164			ow->nfolios = 0;
 165		}
 166		ret = orangefs_writepage_locked(folio, wbc);
 167		mapping_set_error(folio->mapping, ret);
 168		folio_unlock(folio);
 169		folio_end_writeback(folio);
 170	} else {
 171		if (ow->nfolios == ow->maxpages) {
 172			orangefs_writepages_work(ow, wbc);
 173			ow->nfolios = 0;
 174		}
 175	}
 176	return ret;
 177}
 178
 179static int orangefs_writepages(struct address_space *mapping,
 180    struct writeback_control *wbc)
 181{
 182	struct orangefs_writepages *ow;
 183	struct blk_plug plug;
 184	int error;
 185	struct folio *folio = NULL;
 186
 187	ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
 188	if (!ow)
 189		return -ENOMEM;
 190	ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
 191	ow->folios = kcalloc(ow->maxpages, sizeof(struct folio *), GFP_KERNEL);
 192	if (!ow->folios) {
 193		kfree(ow);
 194		return -ENOMEM;
 195	}
 196	ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
 197	if (!ow->bv) {
 198		kfree(ow->folios);
 199		kfree(ow);
 200		return -ENOMEM;
 201	}
 202	ow->mapping = mapping;
 203	blk_start_plug(&plug);
 204	while ((folio = writeback_iter(mapping, wbc, folio, &error)))
 205		error = orangefs_writepages_callback(folio, wbc, ow);
 206	if (ow->nfolios)
 207		error = orangefs_writepages_work(ow, wbc);
 208	blk_finish_plug(&plug);
 209	kfree(ow->folios);
 210	kfree(ow->bv);
 211	kfree(ow);
 212	return error;
 213}
 214
 215static int orangefs_launder_folio(struct folio *);
 216
 217static void orangefs_readahead(struct readahead_control *rac)
 218{
 219	loff_t offset;
 220	struct iov_iter iter;
 221	struct inode *inode = rac->mapping->host;
 222	struct xarray *i_pages;
 223	struct folio *folio;
 224	loff_t new_start = readahead_pos(rac);
 225	int ret;
 226	size_t new_len = 0;
 227
 228	loff_t bytes_remaining = inode->i_size - readahead_pos(rac);
 229	loff_t pages_remaining = bytes_remaining / PAGE_SIZE;
 230
 231	if (pages_remaining >= 1024)
 232		new_len = 4194304;
 233	else if (pages_remaining > readahead_count(rac))
 234		new_len = bytes_remaining;
 235
 236	if (new_len)
 237		readahead_expand(rac, new_start, new_len);
 238
 239	offset = readahead_pos(rac);
 240	i_pages = &rac->mapping->i_pages;
 241
 242	iov_iter_xarray(&iter, ITER_DEST, i_pages, offset, readahead_length(rac));
 243
 244	/* read in the pages. */
 245	if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode,
 246			&offset, &iter, readahead_length(rac),
 247			inode->i_size, NULL, NULL, rac->file)) < 0)
 248		gossip_debug(GOSSIP_FILE_DEBUG,
 249			"%s: wait_for_direct_io failed. \n", __func__);
 250	else
 251		ret = 0;
 252
 253	/* clean up. */
 254	while ((folio = readahead_folio(rac))) {
 255		if (!ret)
 256			folio_mark_uptodate(folio);
 257		folio_unlock(folio);
 258	}
 259}
 260
 261static int orangefs_read_folio(struct file *file, struct folio *folio)
 262{
 263	struct inode *inode = folio->mapping->host;
 264	struct iov_iter iter;
 265	struct bio_vec bv;
 266	ssize_t ret;
 267	loff_t off; /* offset of this folio in the file */
 268
 269	if (folio_test_dirty(folio))
 270		orangefs_launder_folio(folio);
 271
 272	off = folio_pos(folio);
 273	bvec_set_folio(&bv, folio, folio_size(folio), 0);
 274	iov_iter_bvec(&iter, ITER_DEST, &bv, 1, folio_size(folio));
 275
 276	ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
 277			folio_size(folio), inode->i_size, NULL, NULL, file);
 278	/* this will only zero remaining unread portions of the folio data */
 279	iov_iter_zero(~0U, &iter);
 280	/* takes care of potential aliasing */
 281	flush_dcache_folio(folio);
 282	if (ret > 0)
 283		ret = 0;
 284	folio_end_read(folio, ret == 0);
 285	return ret;
 286}
 287
 288static int orangefs_write_begin(const struct kiocb *iocb,
 289				struct address_space *mapping, loff_t pos,
 290				unsigned len, struct folio **foliop,
 291				void **fsdata)
 292{
 293	struct orangefs_write_range *wr;
 294	struct folio *folio;
 295	int ret;
 296
 297	folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
 298			mapping_gfp_mask(mapping));
 299	if (IS_ERR(folio))
 300		return PTR_ERR(folio);
 301
 302	*foliop = folio;
 303
 304	if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 305		/*
 306		 * Should be impossible.  If it happens, launder the page
 307		 * since we don't know what's dirty.  This will WARN in
 308		 * orangefs_writepage_locked.
 309		 */
 310		ret = orangefs_launder_folio(folio);
 311		if (ret)
 312			return ret;
 313	}
 314	if (folio_test_private(folio)) {
 315		struct orangefs_write_range *wr;
 316		wr = folio_get_private(folio);
 317		if (wr->pos + wr->len == pos &&
 318		    uid_eq(wr->uid, current_fsuid()) &&
 319		    gid_eq(wr->gid, current_fsgid())) {
 320			wr->len += len;
 321			goto okay;
 322		} else {
 323			wr->pos = pos;
 324			wr->len = len;
 325			ret = orangefs_launder_folio(folio);
 326			if (ret)
 327				return ret;
 328		}
 329	}
 330
 331	wr = kmalloc(sizeof *wr, GFP_KERNEL);
 332	if (!wr)
 333		return -ENOMEM;
 334
 335	wr->pos = pos;
 336	wr->len = len;
 337	wr->uid = current_fsuid();
 338	wr->gid = current_fsgid();
 339	folio_attach_private(folio, wr);
 340okay:
 341	return 0;
 342}
 343
 344static int orangefs_write_end(const struct kiocb *iocb,
 345			      struct address_space *mapping,
 346			      loff_t pos, unsigned len, unsigned copied,
 347			      struct folio *folio, void *fsdata)
 348{
 349	struct inode *inode = folio->mapping->host;
 350	loff_t last_pos = pos + copied;
 351
 352	/*
 353	 * No need to use i_size_read() here, the i_size
 354	 * cannot change under us because we hold the i_mutex.
 355	 */
 356	if (last_pos > inode->i_size)
 357		i_size_write(inode, last_pos);
 358
 359	/* zero the stale part of the folio if we did a short copy */
 360	if (!folio_test_uptodate(folio)) {
 361		unsigned from = pos & (PAGE_SIZE - 1);
 362		if (copied < len) {
 363			folio_zero_range(folio, from + copied, len - copied);
 364		}
 365		/* Set fully written pages uptodate. */
 366		if (pos == folio_pos(folio) &&
 367		    (len == PAGE_SIZE || pos + len == inode->i_size)) {
 368			folio_zero_segment(folio, from + copied, PAGE_SIZE);
 369			folio_mark_uptodate(folio);
 370		}
 371	}
 372
 373	folio_mark_dirty(folio);
 374	folio_unlock(folio);
 375	folio_put(folio);
 376
 377	mark_inode_dirty_sync(file_inode(iocb->ki_filp));
 378	return copied;
 379}
 380
 381static void orangefs_invalidate_folio(struct folio *folio,
 382				 size_t offset, size_t length)
 383{
 384	struct orangefs_write_range *wr = folio_get_private(folio);
 385
 386	if (offset == 0 && length == PAGE_SIZE) {
 387		kfree(folio_detach_private(folio));
 388		return;
 389	/* write range entirely within invalidate range (or equal) */
 390	} else if (folio_pos(folio) + offset <= wr->pos &&
 391	    wr->pos + wr->len <= folio_pos(folio) + offset + length) {
 392		kfree(folio_detach_private(folio));
 393		/* XXX is this right? only caller in fs */
 394		folio_cancel_dirty(folio);
 395		return;
 396	/* invalidate range chops off end of write range */
 397	} else if (wr->pos < folio_pos(folio) + offset &&
 398	    wr->pos + wr->len <= folio_pos(folio) + offset + length &&
 399	     folio_pos(folio) + offset < wr->pos + wr->len) {
 400		size_t x;
 401		x = wr->pos + wr->len - (folio_pos(folio) + offset);
 402		WARN_ON(x > wr->len);
 403		wr->len -= x;
 404		wr->uid = current_fsuid();
 405		wr->gid = current_fsgid();
 406	/* invalidate range chops off beginning of write range */
 407	} else if (folio_pos(folio) + offset <= wr->pos &&
 408	    folio_pos(folio) + offset + length < wr->pos + wr->len &&
 409	    wr->pos < folio_pos(folio) + offset + length) {
 410		size_t x;
 411		x = folio_pos(folio) + offset + length - wr->pos;
 412		WARN_ON(x > wr->len);
 413		wr->pos += x;
 414		wr->len -= x;
 415		wr->uid = current_fsuid();
 416		wr->gid = current_fsgid();
 417	/* invalidate range entirely within write range (punch hole) */
 418	} else if (wr->pos < folio_pos(folio) + offset &&
 419	    folio_pos(folio) + offset + length < wr->pos + wr->len) {
 420		/* XXX what do we do here... should not WARN_ON */
 421		WARN_ON(1);
 422		/* punch hole */
 423		/*
 424		 * should we just ignore this and write it out anyway?
 425		 * it hardly makes sense
 426		 */
 427		return;
 428	/* non-overlapping ranges */
 429	} else {
 430		/* WARN if they do overlap */
 431		if (!((folio_pos(folio) + offset + length <= wr->pos) ^
 432		    (wr->pos + wr->len <= folio_pos(folio) + offset))) {
 433			WARN_ON(1);
 434			printk("invalidate range offset %llu length %zu\n",
 435			    folio_pos(folio) + offset, length);
 436			printk("write range offset %llu length %zu\n",
 437			    wr->pos, wr->len);
 438		}
 439		return;
 440	}
 441
 442	/*
 443	 * Above there are returns where wr is freed or where we WARN.
 444	 * Thus the following runs if wr was modified above.
 445	 */
 446
 447	orangefs_launder_folio(folio);
 448}
 449
 450static bool orangefs_release_folio(struct folio *folio, gfp_t foo)
 451{
 452	return !folio_test_private(folio);
 453}
 454
 455static void orangefs_free_folio(struct folio *folio)
 456{
 457	kfree(folio_detach_private(folio));
 458}
 459
 460static int orangefs_launder_folio(struct folio *folio)
 461{
 462	int r = 0;
 463	struct writeback_control wbc = {
 464		.sync_mode = WB_SYNC_ALL,
 465		.nr_to_write = 0,
 466	};
 467	folio_wait_writeback(folio);
 468	if (folio_clear_dirty_for_io(folio)) {
 469		r = orangefs_writepage_locked(folio, &wbc);
 470		folio_end_writeback(folio);
 471	}
 472	return r;
 473}
 474
 475static ssize_t orangefs_direct_IO(struct kiocb *iocb,
 476				  struct iov_iter *iter)
 477{
 478	/*
 479	 * Comment from original do_readv_writev:
 480	 * Common entry point for read/write/readv/writev
 481	 * This function will dispatch it to either the direct I/O
 482	 * or buffered I/O path depending on the mount options and/or
 483	 * augmented/extended metadata attached to the file.
 484	 * Note: File extended attributes override any mount options.
 485	 */
 486	struct file *file = iocb->ki_filp;
 487	loff_t pos = iocb->ki_pos;
 488	enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
 489            ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
 490	loff_t *offset = &pos;
 491	struct inode *inode = file->f_mapping->host;
 492	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 493	struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
 494	size_t count = iov_iter_count(iter);
 495	ssize_t total_count = 0;
 496	ssize_t ret = -EINVAL;
 497
 498	gossip_debug(GOSSIP_FILE_DEBUG,
 499		"%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
 500		__func__,
 501		handle,
 502		(int)count);
 503
 504	if (type == ORANGEFS_IO_WRITE) {
 505		gossip_debug(GOSSIP_FILE_DEBUG,
 506			     "%s(%pU): proceeding with offset : %llu, "
 507			     "size %d\n",
 508			     __func__,
 509			     handle,
 510			     llu(*offset),
 511			     (int)count);
 512	}
 513
 514	if (count == 0) {
 515		ret = 0;
 516		goto out;
 517	}
 518
 519	while (iov_iter_count(iter)) {
 520		size_t each_count = iov_iter_count(iter);
 521		size_t amt_complete;
 522
 523		/* how much to transfer in this loop iteration */
 524		if (each_count > orangefs_bufmap_size_query())
 525			each_count = orangefs_bufmap_size_query();
 526
 527		gossip_debug(GOSSIP_FILE_DEBUG,
 528			     "%s(%pU): size of each_count(%d)\n",
 529			     __func__,
 530			     handle,
 531			     (int)each_count);
 532		gossip_debug(GOSSIP_FILE_DEBUG,
 533			     "%s(%pU): BEFORE wait_for_io: offset is %d\n",
 534			     __func__,
 535			     handle,
 536			     (int)*offset);
 537
 538		ret = wait_for_direct_io(type, inode, offset, iter,
 539				each_count, 0, NULL, NULL, file);
 540		gossip_debug(GOSSIP_FILE_DEBUG,
 541			     "%s(%pU): return from wait_for_io:%d\n",
 542			     __func__,
 543			     handle,
 544			     (int)ret);
 545
 546		if (ret < 0)
 547			goto out;
 548
 549		*offset += ret;
 550		total_count += ret;
 551		amt_complete = ret;
 552
 553		gossip_debug(GOSSIP_FILE_DEBUG,
 554			     "%s(%pU): AFTER wait_for_io: offset is %d\n",
 555			     __func__,
 556			     handle,
 557			     (int)*offset);
 558
 559		/*
 560		 * if we got a short I/O operations,
 561		 * fall out and return what we got so far
 562		 */
 563		if (amt_complete < each_count)
 564			break;
 565	} /*end while */
 566
 567out:
 568	if (total_count > 0)
 569		ret = total_count;
 570	if (ret > 0) {
 571		if (type == ORANGEFS_IO_READ) {
 572			file_accessed(file);
 573		} else {
 574			file_update_time(file);
 575			if (*offset > i_size_read(inode))
 576				i_size_write(inode, *offset);
 577		}
 578	}
 579
 580	gossip_debug(GOSSIP_FILE_DEBUG,
 581		     "%s(%pU): Value(%d) returned.\n",
 582		     __func__,
 583		     handle,
 584		     (int)ret);
 585
 586	return ret;
 587}
 588
 589/** ORANGEFS2 implementation of address space operations */
 590static const struct address_space_operations orangefs_address_operations = {
 591	.readahead = orangefs_readahead,
 592	.read_folio = orangefs_read_folio,
 593	.writepages = orangefs_writepages,
 594	.dirty_folio = filemap_dirty_folio,
 595	.write_begin = orangefs_write_begin,
 596	.write_end = orangefs_write_end,
 597	.invalidate_folio = orangefs_invalidate_folio,
 598	.release_folio = orangefs_release_folio,
 599	.free_folio = orangefs_free_folio,
 600	.migrate_folio = filemap_migrate_folio,
 601	.launder_folio = orangefs_launder_folio,
 602	.direct_IO = orangefs_direct_IO,
 603};
 604
 605vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
 606{
 607	struct folio *folio = page_folio(vmf->page);
 608	struct inode *inode = file_inode(vmf->vma->vm_file);
 609	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 610	unsigned long *bitlock = &orangefs_inode->bitlock;
 611	vm_fault_t ret;
 612	struct orangefs_write_range *wr;
 613
 614	sb_start_pagefault(inode->i_sb);
 615
 616	if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
 617		ret = VM_FAULT_RETRY;
 618		goto out;
 619	}
 620
 621	folio_lock(folio);
 622	if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 623		/*
 624		 * Should be impossible.  If it happens, launder the folio
 625		 * since we don't know what's dirty.  This will WARN in
 626		 * orangefs_writepage_locked.
 627		 */
 628		if (orangefs_launder_folio(folio)) {
 629			ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 630			goto out;
 631		}
 632	}
 633	if (folio_test_private(folio)) {
 634		wr = folio_get_private(folio);
 635		if (uid_eq(wr->uid, current_fsuid()) &&
 636		    gid_eq(wr->gid, current_fsgid())) {
 637			wr->pos = page_offset(vmf->page);
 638			wr->len = PAGE_SIZE;
 639			goto okay;
 640		} else {
 641			if (orangefs_launder_folio(folio)) {
 642				ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 643				goto out;
 644			}
 645		}
 646	}
 647	wr = kmalloc(sizeof *wr, GFP_KERNEL);
 648	if (!wr) {
 649		ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 650		goto out;
 651	}
 652	wr->pos = page_offset(vmf->page);
 653	wr->len = PAGE_SIZE;
 654	wr->uid = current_fsuid();
 655	wr->gid = current_fsgid();
 656	folio_attach_private(folio, wr);
 657okay:
 658
 659	file_update_time(vmf->vma->vm_file);
 660	if (folio->mapping != inode->i_mapping) {
 661		folio_unlock(folio);
 662		ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
 663		goto out;
 664	}
 665
 666	/*
 667	 * We mark the folio dirty already here so that when freeze is in
 668	 * progress, we are guaranteed that writeback during freezing will
 669	 * see the dirty folio and writeprotect it again.
 670	 */
 671	folio_mark_dirty(folio);
 672	folio_wait_stable(folio);
 673	ret = VM_FAULT_LOCKED;
 674out:
 675	sb_end_pagefault(inode->i_sb);
 676	return ret;
 677}
 678
 679static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 680{
 681	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 682	struct orangefs_kernel_op_s *new_op;
 683	loff_t orig_size;
 684	int ret = -EINVAL;
 685
 686	gossip_debug(GOSSIP_INODE_DEBUG,
 687		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 688		     __func__,
 689		     get_khandle_from_ino(inode),
 690		     &orangefs_inode->refn.khandle,
 691		     orangefs_inode->refn.fs_id,
 692		     iattr->ia_size);
 693
 694	/* Ensure that we have a up to date size, so we know if it changed. */
 695	ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
 696	if (ret == -ESTALE)
 697		ret = -EIO;
 698	if (ret) {
 699		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 700		    __func__, ret);
 701		return ret;
 702	}
 703	orig_size = i_size_read(inode);
 704
 705	/* This is truncate_setsize in a different order. */
 706	truncate_pagecache(inode, iattr->ia_size);
 707	i_size_write(inode, iattr->ia_size);
 708	if (iattr->ia_size > orig_size)
 709		pagecache_isize_extended(inode, orig_size, iattr->ia_size);
 710
 711	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 712	if (!new_op)
 713		return -ENOMEM;
 714
 715	new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 716	new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 717
 718	ret = service_operation(new_op,
 719		__func__,
 720		get_interruptible_flag(inode));
 721
 722	/*
 723	 * the truncate has no downcall members to retrieve, but
 724	 * the status value tells us if it went through ok or not
 725	 */
 726	gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
 727
 728	op_release(new_op);
 729
 730	if (ret != 0)
 731		return ret;
 732
 733	if (orig_size != i_size_read(inode))
 734		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 735
 736	return ret;
 737}
 738
 739int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
 740{
 741	int ret;
 742
 743	if (iattr->ia_valid & ATTR_MODE) {
 744		if (iattr->ia_mode & (S_ISVTX)) {
 745			if (is_root_handle(inode)) {
 746				/*
 747				 * allow sticky bit to be set on root (since
 748				 * it shows up that way by default anyhow),
 749				 * but don't show it to the server
 750				 */
 751				iattr->ia_mode -= S_ISVTX;
 752			} else {
 753				gossip_debug(GOSSIP_UTILS_DEBUG,
 754					     "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
 755				ret = -EINVAL;
 756				goto out;
 757			}
 758		}
 759		if (iattr->ia_mode & (S_ISUID)) {
 760			gossip_debug(GOSSIP_UTILS_DEBUG,
 761				     "Attempting to set setuid bit (not supported); returning EINVAL.\n");
 762			ret = -EINVAL;
 763			goto out;
 764		}
 765	}
 766
 767	if (iattr->ia_valid & ATTR_SIZE) {
 768		ret = orangefs_setattr_size(inode, iattr);
 769		if (ret)
 770			goto out;
 771	}
 772
 773again:
 774	spin_lock(&inode->i_lock);
 775	if (ORANGEFS_I(inode)->attr_valid) {
 776		if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
 777		    gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
 778			ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 779		} else {
 780			spin_unlock(&inode->i_lock);
 781			write_inode_now(inode, 1);
 782			goto again;
 783		}
 784	} else {
 785		ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 786		ORANGEFS_I(inode)->attr_uid = current_fsuid();
 787		ORANGEFS_I(inode)->attr_gid = current_fsgid();
 788	}
 789	setattr_copy(&nop_mnt_idmap, inode, iattr);
 790	spin_unlock(&inode->i_lock);
 791	mark_inode_dirty(inode);
 792
 793	ret = 0;
 794out:
 795	return ret;
 796}
 797
 798int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr)
 799{
 800	int ret;
 801	struct inode *inode = d_inode(dentry);
 802
 803	ret = __orangefs_setattr(inode, iattr);
 804	/* change mode on a file that has ACLs */
 805	if (!ret && (iattr->ia_valid & ATTR_MODE))
 806		ret = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode);
 807	return ret;
 808}
 809
 810/*
 811 * Change attributes of an object referenced by dentry.
 812 */
 813int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 814		     struct iattr *iattr)
 815{
 816	int ret;
 817	gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
 818	    dentry);
 819	ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
 820	if (ret)
 821	        goto out;
 822	ret = __orangefs_setattr_mode(dentry, iattr);
 823	sync_inode_metadata(d_inode(dentry), 1);
 824out:
 825	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
 826	    ret);
 827	return ret;
 828}
 829
 830/*
 831 * Obtain attributes of an object given a dentry
 832 */
 833int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path,
 834		     struct kstat *stat, u32 request_mask, unsigned int flags)
 835{
 836	int ret;
 837	struct inode *inode = path->dentry->d_inode;
 838
 839	gossip_debug(GOSSIP_INODE_DEBUG,
 840		     "orangefs_getattr: called on %pd mask %u\n",
 841		     path->dentry, request_mask);
 842
 843	ret = orangefs_inode_getattr(inode,
 844	    request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
 845	if (ret == 0) {
 846		generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
 847
 848		/* override block size reported to stat */
 849		if (!(request_mask & STATX_SIZE))
 850			stat->result_mask &= ~STATX_SIZE;
 851
 852		generic_fill_statx_attr(inode, stat);
 853	}
 854	return ret;
 855}
 856
 857int orangefs_permission(struct mnt_idmap *idmap,
 858			struct inode *inode, int mask)
 859{
 860	int ret;
 861
 862	if (mask & MAY_NOT_BLOCK)
 863		return -ECHILD;
 864
 865	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 866
 867	/* Make sure the permission (and other common attrs) are up to date. */
 868	ret = orangefs_inode_getattr(inode, 0);
 869	if (ret < 0)
 870		return ret;
 871
 872	return generic_permission(&nop_mnt_idmap, inode, mask);
 873}
 874
 875int orangefs_update_time(struct inode *inode, int flags)
 876{
 877	struct iattr iattr;
 878
 879	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
 880	    get_khandle_from_ino(inode));
 881
 882	flags = inode_update_timestamps(inode, flags);
 883
 884	memset(&iattr, 0, sizeof iattr);
 885        if (flags & S_ATIME)
 886		iattr.ia_valid |= ATTR_ATIME;
 887	if (flags & S_CTIME)
 888		iattr.ia_valid |= ATTR_CTIME;
 889	if (flags & S_MTIME)
 890		iattr.ia_valid |= ATTR_MTIME;
 891	return __orangefs_setattr(inode, &iattr);
 892}
 893
 894static int orangefs_fileattr_get(struct dentry *dentry, struct file_kattr *fa)
 895{
 896	u64 val = 0;
 897	int ret;
 898
 899	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 900		     dentry);
 901
 902	ret = orangefs_inode_getxattr(d_inode(dentry),
 903				      "user.pvfs2.meta_hint",
 904				      &val, sizeof(val));
 905	if (ret < 0 && ret != -ENODATA)
 906		return ret;
 907
 908	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 909
 910	fileattr_fill_flags(fa, val);
 911	return 0;
 912}
 913
 914static int orangefs_fileattr_set(struct mnt_idmap *idmap,
 915				 struct dentry *dentry, struct file_kattr *fa)
 916{
 917	u64 val = 0;
 918
 919	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 920		     dentry);
 921	/*
 922	 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
 923	 * turned on for a file. The user is not allowed to turn on this bit,
 924	 * but the bit is present if the user first gets the flags and then
 925	 * updates the flags with some new settings. So, we ignore it in the
 926	 * following edit. bligon.
 927	 */
 928	if (fileattr_has_fsx(fa) ||
 929	    (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
 930		gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
 931			   __func__);
 932		return -EOPNOTSUPP;
 933	}
 934	val = fa->flags;
 935	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 936	return orangefs_inode_setxattr(d_inode(dentry),
 937				       "user.pvfs2.meta_hint",
 938				       &val, sizeof(val), 0);
 939}
 940
 941/* ORANGEFS2 implementation of VFS inode operations for files */
 942static const struct inode_operations orangefs_file_inode_operations = {
 943	.get_inode_acl = orangefs_get_acl,
 944	.set_acl = orangefs_set_acl,
 945	.setattr = orangefs_setattr,
 946	.getattr = orangefs_getattr,
 947	.listxattr = orangefs_listxattr,
 948	.permission = orangefs_permission,
 949	.update_time = orangefs_update_time,
 950	.fileattr_get = orangefs_fileattr_get,
 951	.fileattr_set = orangefs_fileattr_set,
 952};
 953
 954static int orangefs_init_iops(struct inode *inode)
 955{
 956	inode->i_mapping->a_ops = &orangefs_address_operations;
 957
 958	switch (inode->i_mode & S_IFMT) {
 959	case S_IFREG:
 960		inode->i_op = &orangefs_file_inode_operations;
 961		inode->i_fop = &orangefs_file_operations;
 962		break;
 963	case S_IFLNK:
 964		inode->i_op = &orangefs_symlink_inode_operations;
 965		break;
 966	case S_IFDIR:
 967		inode->i_op = &orangefs_dir_inode_operations;
 968		inode->i_fop = &orangefs_dir_operations;
 969		break;
 970	default:
 971		gossip_debug(GOSSIP_INODE_DEBUG,
 972			     "%s: unsupported mode\n",
 973			     __func__);
 974		return -EINVAL;
 975	}
 976
 977	return 0;
 978}
 979
 980/*
 981 * Given an ORANGEFS object identifier (fsid, handle), convert it into
 982 * a ino_t type that will be used as a hash-index from where the handle will
 983 * be searched for in the VFS hash table of inodes.
 984 */
 985static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
 986{
 987	if (!ref)
 988		return 0;
 989	return orangefs_khandle_to_ino(&(ref->khandle));
 990}
 991
 992/*
 993 * Called to set up an inode from iget5_locked.
 994 */
 995static int orangefs_set_inode(struct inode *inode, void *data)
 996{
 997	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 998	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
 999	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1000	ORANGEFS_I(inode)->attr_valid = 0;
1001	hash_init(ORANGEFS_I(inode)->xattr_cache);
1002	ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1003	ORANGEFS_I(inode)->bitlock = 0;
1004	return 0;
1005}
1006
1007/*
1008 * Called to determine if handles match.
1009 */
1010static int orangefs_test_inode(struct inode *inode, void *data)
1011{
1012	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1013	struct orangefs_inode_s *orangefs_inode = NULL;
1014
1015	orangefs_inode = ORANGEFS_I(inode);
1016	/* test handles and fs_ids... */
1017	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1018				&(ref->khandle)) &&
1019			orangefs_inode->refn.fs_id == ref->fs_id);
1020}
1021
1022/*
1023 * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1024 * file handle.
1025 *
1026 * @sb: the file system super block instance.
1027 * @ref: The ORANGEFS object for which we are trying to locate an inode.
1028 */
1029struct inode *orangefs_iget(struct super_block *sb,
1030		struct orangefs_object_kref *ref)
1031{
1032	struct inode *inode = NULL;
1033	unsigned long hash;
1034	int error;
1035
1036	hash = orangefs_handle_hash(ref);
1037	inode = iget5_locked(sb,
1038			hash,
1039			orangefs_test_inode,
1040			orangefs_set_inode,
1041			ref);
1042
1043	if (!inode)
1044		return ERR_PTR(-ENOMEM);
1045
1046	if (!(inode_state_read_once(inode) & I_NEW))
1047		return inode;
1048
1049	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1050	if (error) {
1051		iget_failed(inode);
1052		return ERR_PTR(error);
1053	}
1054
1055	inode->i_ino = hash;	/* needed for stat etc */
1056	orangefs_init_iops(inode);
1057	unlock_new_inode(inode);
1058
1059	gossip_debug(GOSSIP_INODE_DEBUG,
1060		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1061		     &ref->khandle,
1062		     ref->fs_id,
1063		     hash,
1064		     inode->i_ino);
1065
1066	return inode;
1067}
1068
1069/*
1070 * Allocate an inode for a newly created file and insert it into the inode hash.
1071 */
1072struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1073		umode_t mode, dev_t dev, struct orangefs_object_kref *ref)
1074{
1075	struct posix_acl *acl = NULL, *default_acl = NULL;
1076	unsigned long hash = orangefs_handle_hash(ref);
1077	struct inode *inode;
1078	int error;
1079
1080	gossip_debug(GOSSIP_INODE_DEBUG,
1081		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1082		     __func__,
1083		     sb,
1084		     MAJOR(dev),
1085		     MINOR(dev),
1086		     mode);
1087
1088	inode = new_inode(sb);
1089	if (!inode)
1090		return ERR_PTR(-ENOMEM);
1091
1092	error = posix_acl_create(dir, &mode, &default_acl, &acl);
1093	if (error)
1094		goto out_iput;
1095
1096	orangefs_set_inode(inode, ref);
1097	inode->i_ino = hash;	/* needed for stat etc */
1098
1099	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1100	if (error)
1101		goto out_iput;
1102
1103	orangefs_init_iops(inode);
1104	inode->i_rdev = dev;
1105
1106	if (default_acl) {
1107		error = __orangefs_set_acl(inode, default_acl,
1108					   ACL_TYPE_DEFAULT);
1109		if (error)
1110			goto out_iput;
1111	}
1112
1113	if (acl) {
1114		error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS);
1115		if (error)
1116			goto out_iput;
1117	}
1118
1119	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1120	if (error < 0)
1121		goto out_iput;
1122
1123	gossip_debug(GOSSIP_INODE_DEBUG,
1124		     "Initializing ACL's for inode %pU\n",
1125		     get_khandle_from_ino(inode));
1126	if (mode != inode->i_mode) {
1127		struct iattr iattr = {
1128			.ia_mode = mode,
1129			.ia_valid = ATTR_MODE,
1130		};
1131		inode->i_mode = mode;
1132		__orangefs_setattr(inode, &iattr);
1133		__posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode);
1134	}
1135	posix_acl_release(acl);
1136	posix_acl_release(default_acl);
1137	return inode;
1138
1139out_iput:
1140	iput(inode);
1141	posix_acl_release(acl);
1142	posix_acl_release(default_acl);
1143	return ERR_PTR(error);
1144}