Linux Audio

Check our new training course

Loading...
Note: File does not exist in v4.6.
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Simple file system for zoned block devices exposing zones as files.
   4 *
   5 * Copyright (C) 2019 Western Digital Corporation or its affiliates.
   6 */
   7#include <linux/module.h>
   8#include <linux/pagemap.h>
   9#include <linux/magic.h>
  10#include <linux/iomap.h>
  11#include <linux/init.h>
  12#include <linux/slab.h>
  13#include <linux/blkdev.h>
  14#include <linux/statfs.h>
  15#include <linux/writeback.h>
  16#include <linux/quotaops.h>
  17#include <linux/seq_file.h>
  18#include <linux/parser.h>
  19#include <linux/uio.h>
  20#include <linux/mman.h>
  21#include <linux/sched/mm.h>
  22#include <linux/crc32.h>
  23#include <linux/task_io_accounting_ops.h>
  24
  25#include "zonefs.h"
  26
  27#define CREATE_TRACE_POINTS
  28#include "trace.h"
  29
  30/*
  31 * Get the name of a zone group directory.
  32 */
  33static const char *zonefs_zgroup_name(enum zonefs_ztype ztype)
  34{
  35	switch (ztype) {
  36	case ZONEFS_ZTYPE_CNV:
  37		return "cnv";
  38	case ZONEFS_ZTYPE_SEQ:
  39		return "seq";
  40	default:
  41		WARN_ON_ONCE(1);
  42		return "???";
  43	}
  44}
  45
  46/*
  47 * Manage the active zone count.
  48 */
  49static void zonefs_account_active(struct super_block *sb,
  50				  struct zonefs_zone *z)
  51{
  52	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
  53
  54	if (zonefs_zone_is_cnv(z))
  55		return;
  56
  57	/*
  58	 * For zones that transitioned to the offline or readonly condition,
  59	 * we only need to clear the active state.
  60	 */
  61	if (z->z_flags & (ZONEFS_ZONE_OFFLINE | ZONEFS_ZONE_READONLY))
  62		goto out;
  63
  64	/*
  65	 * If the zone is active, that is, if it is explicitly open or
  66	 * partially written, check if it was already accounted as active.
  67	 */
  68	if ((z->z_flags & ZONEFS_ZONE_OPEN) ||
  69	    (z->z_wpoffset > 0 && z->z_wpoffset < z->z_capacity)) {
  70		if (!(z->z_flags & ZONEFS_ZONE_ACTIVE)) {
  71			z->z_flags |= ZONEFS_ZONE_ACTIVE;
  72			atomic_inc(&sbi->s_active_seq_files);
  73		}
  74		return;
  75	}
  76
  77out:
  78	/* The zone is not active. If it was, update the active count */
  79	if (z->z_flags & ZONEFS_ZONE_ACTIVE) {
  80		z->z_flags &= ~ZONEFS_ZONE_ACTIVE;
  81		atomic_dec(&sbi->s_active_seq_files);
  82	}
  83}
  84
  85/*
  86 * Manage the active zone count. Called with zi->i_truncate_mutex held.
  87 */
  88void zonefs_inode_account_active(struct inode *inode)
  89{
  90	lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
  91
  92	return zonefs_account_active(inode->i_sb, zonefs_inode_zone(inode));
  93}
  94
  95/*
  96 * Execute a zone management operation.
  97 */
  98static int zonefs_zone_mgmt(struct super_block *sb,
  99			    struct zonefs_zone *z, enum req_op op)
 100{
 101	int ret;
 102
 103	/*
 104	 * With ZNS drives, closing an explicitly open zone that has not been
 105	 * written will change the zone state to "closed", that is, the zone
 106	 * will remain active. Since this can then cause failure of explicit
 107	 * open operation on other zones if the drive active zone resources
 108	 * are exceeded, make sure that the zone does not remain active by
 109	 * resetting it.
 110	 */
 111	if (op == REQ_OP_ZONE_CLOSE && !z->z_wpoffset)
 112		op = REQ_OP_ZONE_RESET;
 113
 114	trace_zonefs_zone_mgmt(sb, z, op);
 115	ret = blkdev_zone_mgmt(sb->s_bdev, op, z->z_sector,
 116			       z->z_size >> SECTOR_SHIFT, GFP_NOFS);
 117	if (ret) {
 118		zonefs_err(sb,
 119			   "Zone management operation %s at %llu failed %d\n",
 120			   blk_op_str(op), z->z_sector, ret);
 121		return ret;
 122	}
 123
 124	return 0;
 125}
 126
 127int zonefs_inode_zone_mgmt(struct inode *inode, enum req_op op)
 128{
 129	lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
 130
 131	return zonefs_zone_mgmt(inode->i_sb, zonefs_inode_zone(inode), op);
 132}
 133
 134void zonefs_i_size_write(struct inode *inode, loff_t isize)
 135{
 136	struct zonefs_zone *z = zonefs_inode_zone(inode);
 137
 138	i_size_write(inode, isize);
 139
 140	/*
 141	 * A full zone is no longer open/active and does not need
 142	 * explicit closing.
 143	 */
 144	if (isize >= z->z_capacity) {
 145		struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
 146
 147		if (z->z_flags & ZONEFS_ZONE_ACTIVE)
 148			atomic_dec(&sbi->s_active_seq_files);
 149		z->z_flags &= ~(ZONEFS_ZONE_OPEN | ZONEFS_ZONE_ACTIVE);
 150	}
 151}
 152
 153void zonefs_update_stats(struct inode *inode, loff_t new_isize)
 154{
 155	struct super_block *sb = inode->i_sb;
 156	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 157	loff_t old_isize = i_size_read(inode);
 158	loff_t nr_blocks;
 159
 160	if (new_isize == old_isize)
 161		return;
 162
 163	spin_lock(&sbi->s_lock);
 164
 165	/*
 166	 * This may be called for an update after an IO error.
 167	 * So beware of the values seen.
 168	 */
 169	if (new_isize < old_isize) {
 170		nr_blocks = (old_isize - new_isize) >> sb->s_blocksize_bits;
 171		if (sbi->s_used_blocks > nr_blocks)
 172			sbi->s_used_blocks -= nr_blocks;
 173		else
 174			sbi->s_used_blocks = 0;
 175	} else {
 176		sbi->s_used_blocks +=
 177			(new_isize - old_isize) >> sb->s_blocksize_bits;
 178		if (sbi->s_used_blocks > sbi->s_blocks)
 179			sbi->s_used_blocks = sbi->s_blocks;
 180	}
 181
 182	spin_unlock(&sbi->s_lock);
 183}
 184
 185/*
 186 * Check a zone condition. Return the amount of written (and still readable)
 187 * data in the zone.
 188 */
 189static loff_t zonefs_check_zone_condition(struct super_block *sb,
 190					  struct zonefs_zone *z,
 191					  struct blk_zone *zone)
 192{
 193	switch (zone->cond) {
 194	case BLK_ZONE_COND_OFFLINE:
 195		zonefs_warn(sb, "Zone %llu: offline zone\n",
 196			    z->z_sector);
 197		z->z_flags |= ZONEFS_ZONE_OFFLINE;
 198		return 0;
 199	case BLK_ZONE_COND_READONLY:
 200		/*
 201		 * The write pointer of read-only zones is invalid, so we cannot
 202		 * determine the zone wpoffset (inode size). We thus keep the
 203		 * zone wpoffset as is, which leads to an empty file
 204		 * (wpoffset == 0) on mount. For a runtime error, this keeps
 205		 * the inode size as it was when last updated so that the user
 206		 * can recover data.
 207		 */
 208		zonefs_warn(sb, "Zone %llu: read-only zone\n",
 209			    z->z_sector);
 210		z->z_flags |= ZONEFS_ZONE_READONLY;
 211		if (zonefs_zone_is_cnv(z))
 212			return z->z_capacity;
 213		return z->z_wpoffset;
 214	case BLK_ZONE_COND_FULL:
 215		/* The write pointer of full zones is invalid. */
 216		return z->z_capacity;
 217	default:
 218		if (zonefs_zone_is_cnv(z))
 219			return z->z_capacity;
 220		return (zone->wp - zone->start) << SECTOR_SHIFT;
 221	}
 222}
 223
 224/*
 225 * Check a zone condition and adjust its inode access permissions for
 226 * offline and readonly zones.
 227 */
 228static void zonefs_inode_update_mode(struct inode *inode)
 229{
 230	struct zonefs_zone *z = zonefs_inode_zone(inode);
 231
 232	if (z->z_flags & ZONEFS_ZONE_OFFLINE) {
 233		/* Offline zones cannot be read nor written */
 234		inode->i_flags |= S_IMMUTABLE;
 235		inode->i_mode &= ~0777;
 236	} else if (z->z_flags & ZONEFS_ZONE_READONLY) {
 237		/* Readonly zones cannot be written */
 238		inode->i_flags |= S_IMMUTABLE;
 239		if (z->z_flags & ZONEFS_ZONE_INIT_MODE)
 240			inode->i_mode &= ~0777;
 241		else
 242			inode->i_mode &= ~0222;
 243	}
 244
 245	z->z_flags &= ~ZONEFS_ZONE_INIT_MODE;
 246	z->z_mode = inode->i_mode;
 247}
 248
 249static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
 250			      void *data)
 251{
 252	struct blk_zone *z = data;
 253
 254	*z = *zone;
 255	return 0;
 256}
 257
 258static void zonefs_handle_io_error(struct inode *inode, struct blk_zone *zone,
 259				   bool write)
 260{
 261	struct zonefs_zone *z = zonefs_inode_zone(inode);
 262	struct super_block *sb = inode->i_sb;
 263	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 264	loff_t isize, data_size;
 265
 266	/*
 267	 * Check the zone condition: if the zone is not "bad" (offline or
 268	 * read-only), read errors are simply signaled to the IO issuer as long
 269	 * as there is no inconsistency between the inode size and the amount of
 270	 * data writen in the zone (data_size).
 271	 */
 272	data_size = zonefs_check_zone_condition(sb, z, zone);
 273	isize = i_size_read(inode);
 274	if (!(z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)) &&
 275	    !write && isize == data_size)
 276		return;
 277
 278	/*
 279	 * At this point, we detected either a bad zone or an inconsistency
 280	 * between the inode size and the amount of data written in the zone.
 281	 * For the latter case, the cause may be a write IO error or an external
 282	 * action on the device. Two error patterns exist:
 283	 * 1) The inode size is lower than the amount of data in the zone:
 284	 *    a write operation partially failed and data was writen at the end
 285	 *    of the file. This can happen in the case of a large direct IO
 286	 *    needing several BIOs and/or write requests to be processed.
 287	 * 2) The inode size is larger than the amount of data in the zone:
 288	 *    this can happen with a deferred write error with the use of the
 289	 *    device side write cache after getting successful write IO
 290	 *    completions. Other possibilities are (a) an external corruption,
 291	 *    e.g. an application reset the zone directly, or (b) the device
 292	 *    has a serious problem (e.g. firmware bug).
 293	 *
 294	 * In all cases, warn about inode size inconsistency and handle the
 295	 * IO error according to the zone condition and to the mount options.
 296	 */
 297	if (isize != data_size)
 298		zonefs_warn(sb,
 299			    "inode %lu: invalid size %lld (should be %lld)\n",
 300			    inode->i_ino, isize, data_size);
 301
 302	/*
 303	 * First handle bad zones signaled by hardware. The mount options
 304	 * errors=zone-ro and errors=zone-offline result in changing the
 305	 * zone condition to read-only and offline respectively, as if the
 306	 * condition was signaled by the hardware.
 307	 */
 308	if ((z->z_flags & ZONEFS_ZONE_OFFLINE) ||
 309	    (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)) {
 310		zonefs_warn(sb, "inode %lu: read/write access disabled\n",
 311			    inode->i_ino);
 312		if (!(z->z_flags & ZONEFS_ZONE_OFFLINE))
 313			z->z_flags |= ZONEFS_ZONE_OFFLINE;
 314		zonefs_inode_update_mode(inode);
 315		data_size = 0;
 316	} else if ((z->z_flags & ZONEFS_ZONE_READONLY) ||
 317		   (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)) {
 318		zonefs_warn(sb, "inode %lu: write access disabled\n",
 319			    inode->i_ino);
 320		if (!(z->z_flags & ZONEFS_ZONE_READONLY))
 321			z->z_flags |= ZONEFS_ZONE_READONLY;
 322		zonefs_inode_update_mode(inode);
 323		data_size = isize;
 324	} else if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO &&
 325		   data_size > isize) {
 326		/* Do not expose garbage data */
 327		data_size = isize;
 328	}
 329
 330	/*
 331	 * If the filesystem is mounted with the explicit-open mount option, we
 332	 * need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to
 333	 * the read-only or offline condition, to avoid attempting an explicit
 334	 * close of the zone when the inode file is closed.
 335	 */
 336	if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) &&
 337	    (z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)))
 338		z->z_flags &= ~ZONEFS_ZONE_OPEN;
 339
 340	/*
 341	 * If error=remount-ro was specified, any error result in remounting
 342	 * the volume as read-only.
 343	 */
 344	if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) {
 345		zonefs_warn(sb, "remounting filesystem read-only\n");
 346		sb->s_flags |= SB_RDONLY;
 347	}
 348
 349	/*
 350	 * Update block usage stats and the inode size  to prevent access to
 351	 * invalid data.
 352	 */
 353	zonefs_update_stats(inode, data_size);
 354	zonefs_i_size_write(inode, data_size);
 355	z->z_wpoffset = data_size;
 356	zonefs_inode_account_active(inode);
 357}
 358
 359/*
 360 * When an file IO error occurs, check the file zone to see if there is a change
 361 * in the zone condition (e.g. offline or read-only). For a failed write to a
 362 * sequential zone, the zone write pointer position must also be checked to
 363 * eventually correct the file size and zonefs inode write pointer offset
 364 * (which can be out of sync with the drive due to partial write failures).
 365 */
 366void __zonefs_io_error(struct inode *inode, bool write)
 367{
 368	struct zonefs_zone *z = zonefs_inode_zone(inode);
 369	struct super_block *sb = inode->i_sb;
 370	unsigned int noio_flag;
 371	struct blk_zone zone;
 372	int ret;
 373
 374	/*
 375	 * Conventional zone have no write pointer and cannot become read-only
 376	 * or offline. So simply fake a report for a single or aggregated zone
 377	 * and let zonefs_handle_io_error() correct the zone inode information
 378	 * according to the mount options.
 379	 */
 380	if (!zonefs_zone_is_seq(z)) {
 381		zone.start = z->z_sector;
 382		zone.len = z->z_size >> SECTOR_SHIFT;
 383		zone.wp = zone.start + zone.len;
 384		zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
 385		zone.cond = BLK_ZONE_COND_NOT_WP;
 386		zone.capacity = zone.len;
 387		goto handle_io_error;
 388	}
 389
 390	/*
 391	 * Memory allocations in blkdev_report_zones() can trigger a memory
 392	 * reclaim which may in turn cause a recursion into zonefs as well as
 393	 * struct request allocations for the same device. The former case may
 394	 * end up in a deadlock on the inode truncate mutex, while the latter
 395	 * may prevent IO forward progress. Executing the report zones under
 396	 * the GFP_NOIO context avoids both problems.
 397	 */
 398	noio_flag = memalloc_noio_save();
 399	ret = blkdev_report_zones(sb->s_bdev, z->z_sector, 1,
 400				  zonefs_io_error_cb, &zone);
 401	memalloc_noio_restore(noio_flag);
 402
 403	if (ret != 1) {
 404		zonefs_err(sb, "Get inode %lu zone information failed %d\n",
 405			   inode->i_ino, ret);
 406		zonefs_warn(sb, "remounting filesystem read-only\n");
 407		sb->s_flags |= SB_RDONLY;
 408		return;
 409	}
 410
 411handle_io_error:
 412	zonefs_handle_io_error(inode, &zone, write);
 413}
 414
 415static struct kmem_cache *zonefs_inode_cachep;
 416
 417static struct inode *zonefs_alloc_inode(struct super_block *sb)
 418{
 419	struct zonefs_inode_info *zi;
 420
 421	zi = alloc_inode_sb(sb, zonefs_inode_cachep, GFP_KERNEL);
 422	if (!zi)
 423		return NULL;
 424
 425	inode_init_once(&zi->i_vnode);
 426	mutex_init(&zi->i_truncate_mutex);
 427	zi->i_wr_refcnt = 0;
 428
 429	return &zi->i_vnode;
 430}
 431
 432static void zonefs_free_inode(struct inode *inode)
 433{
 434	kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode));
 435}
 436
 437/*
 438 * File system stat.
 439 */
 440static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
 441{
 442	struct super_block *sb = dentry->d_sb;
 443	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 444	enum zonefs_ztype t;
 445
 446	buf->f_type = ZONEFS_MAGIC;
 447	buf->f_bsize = sb->s_blocksize;
 448	buf->f_namelen = ZONEFS_NAME_MAX;
 449
 450	spin_lock(&sbi->s_lock);
 451
 452	buf->f_blocks = sbi->s_blocks;
 453	if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks))
 454		buf->f_bfree = 0;
 455	else
 456		buf->f_bfree = buf->f_blocks - sbi->s_used_blocks;
 457	buf->f_bavail = buf->f_bfree;
 458
 459	for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
 460		if (sbi->s_zgroup[t].g_nr_zones)
 461			buf->f_files += sbi->s_zgroup[t].g_nr_zones + 1;
 462	}
 463	buf->f_ffree = 0;
 464
 465	spin_unlock(&sbi->s_lock);
 466
 467	buf->f_fsid = uuid_to_fsid(sbi->s_uuid.b);
 468
 469	return 0;
 470}
 471
 472enum {
 473	Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
 474	Opt_explicit_open, Opt_err,
 475};
 476
 477static const match_table_t tokens = {
 478	{ Opt_errors_ro,	"errors=remount-ro"},
 479	{ Opt_errors_zro,	"errors=zone-ro"},
 480	{ Opt_errors_zol,	"errors=zone-offline"},
 481	{ Opt_errors_repair,	"errors=repair"},
 482	{ Opt_explicit_open,	"explicit-open" },
 483	{ Opt_err,		NULL}
 484};
 485
 486static int zonefs_parse_options(struct super_block *sb, char *options)
 487{
 488	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 489	substring_t args[MAX_OPT_ARGS];
 490	char *p;
 491
 492	if (!options)
 493		return 0;
 494
 495	while ((p = strsep(&options, ",")) != NULL) {
 496		int token;
 497
 498		if (!*p)
 499			continue;
 500
 501		token = match_token(p, tokens, args);
 502		switch (token) {
 503		case Opt_errors_ro:
 504			sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
 505			sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_RO;
 506			break;
 507		case Opt_errors_zro:
 508			sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
 509			sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZRO;
 510			break;
 511		case Opt_errors_zol:
 512			sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
 513			sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZOL;
 514			break;
 515		case Opt_errors_repair:
 516			sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
 517			sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
 518			break;
 519		case Opt_explicit_open:
 520			sbi->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN;
 521			break;
 522		default:
 523			return -EINVAL;
 524		}
 525	}
 526
 527	return 0;
 528}
 529
 530static int zonefs_show_options(struct seq_file *seq, struct dentry *root)
 531{
 532	struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb);
 533
 534	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO)
 535		seq_puts(seq, ",errors=remount-ro");
 536	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)
 537		seq_puts(seq, ",errors=zone-ro");
 538	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)
 539		seq_puts(seq, ",errors=zone-offline");
 540	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR)
 541		seq_puts(seq, ",errors=repair");
 542
 543	return 0;
 544}
 545
 546static int zonefs_remount(struct super_block *sb, int *flags, char *data)
 547{
 548	sync_filesystem(sb);
 549
 550	return zonefs_parse_options(sb, data);
 551}
 552
 553static int zonefs_inode_setattr(struct mnt_idmap *idmap,
 554				struct dentry *dentry, struct iattr *iattr)
 555{
 556	struct inode *inode = d_inode(dentry);
 557	int ret;
 558
 559	if (unlikely(IS_IMMUTABLE(inode)))
 560		return -EPERM;
 561
 562	ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
 563	if (ret)
 564		return ret;
 565
 566	/*
 567	 * Since files and directories cannot be created nor deleted, do not
 568	 * allow setting any write attributes on the sub-directories grouping
 569	 * files by zone type.
 570	 */
 571	if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) &&
 572	    (iattr->ia_mode & 0222))
 573		return -EPERM;
 574
 575	if (((iattr->ia_valid & ATTR_UID) &&
 576	     !uid_eq(iattr->ia_uid, inode->i_uid)) ||
 577	    ((iattr->ia_valid & ATTR_GID) &&
 578	     !gid_eq(iattr->ia_gid, inode->i_gid))) {
 579		ret = dquot_transfer(&nop_mnt_idmap, inode, iattr);
 580		if (ret)
 581			return ret;
 582	}
 583
 584	if (iattr->ia_valid & ATTR_SIZE) {
 585		ret = zonefs_file_truncate(inode, iattr->ia_size);
 586		if (ret)
 587			return ret;
 588	}
 589
 590	setattr_copy(&nop_mnt_idmap, inode, iattr);
 591
 592	if (S_ISREG(inode->i_mode)) {
 593		struct zonefs_zone *z = zonefs_inode_zone(inode);
 594
 595		z->z_mode = inode->i_mode;
 596		z->z_uid = inode->i_uid;
 597		z->z_gid = inode->i_gid;
 598	}
 599
 600	return 0;
 601}
 602
 603static const struct inode_operations zonefs_file_inode_operations = {
 604	.setattr	= zonefs_inode_setattr,
 605};
 606
 607static long zonefs_fname_to_fno(const struct qstr *fname)
 608{
 609	const char *name = fname->name;
 610	unsigned int len = fname->len;
 611	long fno = 0, shift = 1;
 612	const char *rname;
 613	char c = *name;
 614	unsigned int i;
 615
 616	/*
 617	 * File names are always a base-10 number string without any
 618	 * leading 0s.
 619	 */
 620	if (!isdigit(c))
 621		return -ENOENT;
 622
 623	if (len > 1 && c == '0')
 624		return -ENOENT;
 625
 626	if (len == 1)
 627		return c - '0';
 628
 629	for (i = 0, rname = name + len - 1; i < len; i++, rname--) {
 630		c = *rname;
 631		if (!isdigit(c))
 632			return -ENOENT;
 633		fno += (c - '0') * shift;
 634		shift *= 10;
 635	}
 636
 637	return fno;
 638}
 639
 640static struct inode *zonefs_get_file_inode(struct inode *dir,
 641					   struct dentry *dentry)
 642{
 643	struct zonefs_zone_group *zgroup = dir->i_private;
 644	struct super_block *sb = dir->i_sb;
 645	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 646	struct zonefs_zone *z;
 647	struct inode *inode;
 648	ino_t ino;
 649	long fno;
 650
 651	/* Get the file number from the file name */
 652	fno = zonefs_fname_to_fno(&dentry->d_name);
 653	if (fno < 0)
 654		return ERR_PTR(fno);
 655
 656	if (!zgroup->g_nr_zones || fno >= zgroup->g_nr_zones)
 657		return ERR_PTR(-ENOENT);
 658
 659	z = &zgroup->g_zones[fno];
 660	ino = z->z_sector >> sbi->s_zone_sectors_shift;
 661	inode = iget_locked(sb, ino);
 662	if (!inode)
 663		return ERR_PTR(-ENOMEM);
 664	if (!(inode->i_state & I_NEW)) {
 665		WARN_ON_ONCE(inode->i_private != z);
 666		return inode;
 667	}
 668
 669	inode->i_ino = ino;
 670	inode->i_mode = z->z_mode;
 671	inode_set_mtime_to_ts(inode,
 672			      inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(dir))));
 673	inode->i_uid = z->z_uid;
 674	inode->i_gid = z->z_gid;
 675	inode->i_size = z->z_wpoffset;
 676	inode->i_blocks = z->z_capacity >> SECTOR_SHIFT;
 677	inode->i_private = z;
 678
 679	inode->i_op = &zonefs_file_inode_operations;
 680	inode->i_fop = &zonefs_file_operations;
 681	inode->i_mapping->a_ops = &zonefs_file_aops;
 682
 683	/* Update the inode access rights depending on the zone condition */
 684	zonefs_inode_update_mode(inode);
 685
 686	unlock_new_inode(inode);
 687
 688	return inode;
 689}
 690
 691static struct inode *zonefs_get_zgroup_inode(struct super_block *sb,
 692					     enum zonefs_ztype ztype)
 693{
 694	struct inode *root = d_inode(sb->s_root);
 695	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 696	struct inode *inode;
 697	ino_t ino = bdev_nr_zones(sb->s_bdev) + ztype + 1;
 698
 699	inode = iget_locked(sb, ino);
 700	if (!inode)
 701		return ERR_PTR(-ENOMEM);
 702	if (!(inode->i_state & I_NEW))
 703		return inode;
 704
 705	inode->i_ino = ino;
 706	inode_init_owner(&nop_mnt_idmap, inode, root, S_IFDIR | 0555);
 707	inode->i_size = sbi->s_zgroup[ztype].g_nr_zones;
 708	inode_set_mtime_to_ts(inode,
 709			      inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(root))));
 710	inode->i_private = &sbi->s_zgroup[ztype];
 711	set_nlink(inode, 2);
 712
 713	inode->i_op = &zonefs_dir_inode_operations;
 714	inode->i_fop = &zonefs_dir_operations;
 715
 716	unlock_new_inode(inode);
 717
 718	return inode;
 719}
 720
 721
 722static struct inode *zonefs_get_dir_inode(struct inode *dir,
 723					  struct dentry *dentry)
 724{
 725	struct super_block *sb = dir->i_sb;
 726	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 727	const char *name = dentry->d_name.name;
 728	enum zonefs_ztype ztype;
 729
 730	/*
 731	 * We only need to check for the "seq" directory and
 732	 * the "cnv" directory if we have conventional zones.
 733	 */
 734	if (dentry->d_name.len != 3)
 735		return ERR_PTR(-ENOENT);
 736
 737	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
 738		if (sbi->s_zgroup[ztype].g_nr_zones &&
 739		    memcmp(name, zonefs_zgroup_name(ztype), 3) == 0)
 740			break;
 741	}
 742	if (ztype == ZONEFS_ZTYPE_MAX)
 743		return ERR_PTR(-ENOENT);
 744
 745	return zonefs_get_zgroup_inode(sb, ztype);
 746}
 747
 748static struct dentry *zonefs_lookup(struct inode *dir, struct dentry *dentry,
 749				    unsigned int flags)
 750{
 751	struct inode *inode;
 752
 753	if (dentry->d_name.len > ZONEFS_NAME_MAX)
 754		return ERR_PTR(-ENAMETOOLONG);
 755
 756	if (dir == d_inode(dir->i_sb->s_root))
 757		inode = zonefs_get_dir_inode(dir, dentry);
 758	else
 759		inode = zonefs_get_file_inode(dir, dentry);
 760
 761	return d_splice_alias(inode, dentry);
 762}
 763
 764static int zonefs_readdir_root(struct file *file, struct dir_context *ctx)
 765{
 766	struct inode *inode = file_inode(file);
 767	struct super_block *sb = inode->i_sb;
 768	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 769	enum zonefs_ztype ztype = ZONEFS_ZTYPE_CNV;
 770	ino_t base_ino = bdev_nr_zones(sb->s_bdev) + 1;
 771
 772	if (ctx->pos >= inode->i_size)
 773		return 0;
 774
 775	if (!dir_emit_dots(file, ctx))
 776		return 0;
 777
 778	if (ctx->pos == 2) {
 779		if (!sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones)
 780			ztype = ZONEFS_ZTYPE_SEQ;
 781
 782		if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3,
 783			      base_ino + ztype, DT_DIR))
 784			return 0;
 785		ctx->pos++;
 786	}
 787
 788	if (ctx->pos == 3 && ztype != ZONEFS_ZTYPE_SEQ) {
 789		ztype = ZONEFS_ZTYPE_SEQ;
 790		if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3,
 791			      base_ino + ztype, DT_DIR))
 792			return 0;
 793		ctx->pos++;
 794	}
 795
 796	return 0;
 797}
 798
 799static int zonefs_readdir_zgroup(struct file *file,
 800				 struct dir_context *ctx)
 801{
 802	struct inode *inode = file_inode(file);
 803	struct zonefs_zone_group *zgroup = inode->i_private;
 804	struct super_block *sb = inode->i_sb;
 805	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 806	struct zonefs_zone *z;
 807	int fname_len;
 808	char *fname;
 809	ino_t ino;
 810	int f;
 811
 812	/*
 813	 * The size of zone group directories is equal to the number
 814	 * of zone files in the group and does note include the "." and
 815	 * ".." entries. Hence the "+ 2" here.
 816	 */
 817	if (ctx->pos >= inode->i_size + 2)
 818		return 0;
 819
 820	if (!dir_emit_dots(file, ctx))
 821		return 0;
 822
 823	fname = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL);
 824	if (!fname)
 825		return -ENOMEM;
 826
 827	for (f = ctx->pos - 2; f < zgroup->g_nr_zones; f++) {
 828		z = &zgroup->g_zones[f];
 829		ino = z->z_sector >> sbi->s_zone_sectors_shift;
 830		fname_len = snprintf(fname, ZONEFS_NAME_MAX - 1, "%u", f);
 831		if (!dir_emit(ctx, fname, fname_len, ino, DT_REG))
 832			break;
 833		ctx->pos++;
 834	}
 835
 836	kfree(fname);
 837
 838	return 0;
 839}
 840
 841static int zonefs_readdir(struct file *file, struct dir_context *ctx)
 842{
 843	struct inode *inode = file_inode(file);
 844
 845	if (inode == d_inode(inode->i_sb->s_root))
 846		return zonefs_readdir_root(file, ctx);
 847
 848	return zonefs_readdir_zgroup(file, ctx);
 849}
 850
 851const struct inode_operations zonefs_dir_inode_operations = {
 852	.lookup		= zonefs_lookup,
 853	.setattr	= zonefs_inode_setattr,
 854};
 855
 856const struct file_operations zonefs_dir_operations = {
 857	.llseek		= generic_file_llseek,
 858	.read		= generic_read_dir,
 859	.iterate_shared	= zonefs_readdir,
 860};
 861
 862struct zonefs_zone_data {
 863	struct super_block	*sb;
 864	unsigned int		nr_zones[ZONEFS_ZTYPE_MAX];
 865	sector_t		cnv_zone_start;
 866	struct blk_zone		*zones;
 867};
 868
 869static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx,
 870				   void *data)
 871{
 872	struct zonefs_zone_data *zd = data;
 873	struct super_block *sb = zd->sb;
 874	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 875
 876	/*
 877	 * We do not care about the first zone: it contains the super block
 878	 * and not exposed as a file.
 879	 */
 880	if (!idx)
 881		return 0;
 882
 883	/*
 884	 * Count the number of zones that will be exposed as files.
 885	 * For sequential zones, we always have as many files as zones.
 886	 * FOr conventional zones, the number of files depends on if we have
 887	 * conventional zones aggregation enabled.
 888	 */
 889	switch (zone->type) {
 890	case BLK_ZONE_TYPE_CONVENTIONAL:
 891		if (sbi->s_features & ZONEFS_F_AGGRCNV) {
 892			/* One file per set of contiguous conventional zones */
 893			if (!(sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones) ||
 894			    zone->start != zd->cnv_zone_start)
 895				sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
 896			zd->cnv_zone_start = zone->start + zone->len;
 897		} else {
 898			/* One file per zone */
 899			sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
 900		}
 901		break;
 902	case BLK_ZONE_TYPE_SEQWRITE_REQ:
 903	case BLK_ZONE_TYPE_SEQWRITE_PREF:
 904		sbi->s_zgroup[ZONEFS_ZTYPE_SEQ].g_nr_zones++;
 905		break;
 906	default:
 907		zonefs_err(zd->sb, "Unsupported zone type 0x%x\n",
 908			   zone->type);
 909		return -EIO;
 910	}
 911
 912	memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone));
 913
 914	return 0;
 915}
 916
 917static int zonefs_get_zone_info(struct zonefs_zone_data *zd)
 918{
 919	struct block_device *bdev = zd->sb->s_bdev;
 920	int ret;
 921
 922	zd->zones = kvcalloc(bdev_nr_zones(bdev), sizeof(struct blk_zone),
 923			     GFP_KERNEL);
 924	if (!zd->zones)
 925		return -ENOMEM;
 926
 927	/* Get zones information from the device */
 928	ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES,
 929				  zonefs_get_zone_info_cb, zd);
 930	if (ret < 0) {
 931		zonefs_err(zd->sb, "Zone report failed %d\n", ret);
 932		return ret;
 933	}
 934
 935	if (ret != bdev_nr_zones(bdev)) {
 936		zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n",
 937			   ret, bdev_nr_zones(bdev));
 938		return -EIO;
 939	}
 940
 941	return 0;
 942}
 943
 944static inline void zonefs_free_zone_info(struct zonefs_zone_data *zd)
 945{
 946	kvfree(zd->zones);
 947}
 948
 949/*
 950 * Create a zone group and populate it with zone files.
 951 */
 952static int zonefs_init_zgroup(struct super_block *sb,
 953			      struct zonefs_zone_data *zd,
 954			      enum zonefs_ztype ztype)
 955{
 956	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
 957	struct zonefs_zone_group *zgroup = &sbi->s_zgroup[ztype];
 958	struct blk_zone *zone, *next, *end;
 959	struct zonefs_zone *z;
 960	unsigned int n = 0;
 961	int ret;
 962
 963	/* Allocate the zone group. If it is empty, we have nothing to do. */
 964	if (!zgroup->g_nr_zones)
 965		return 0;
 966
 967	zgroup->g_zones = kvcalloc(zgroup->g_nr_zones,
 968				   sizeof(struct zonefs_zone), GFP_KERNEL);
 969	if (!zgroup->g_zones)
 970		return -ENOMEM;
 971
 972	/*
 973	 * Initialize the zone groups using the device zone information.
 974	 * We always skip the first zone as it contains the super block
 975	 * and is not use to back a file.
 976	 */
 977	end = zd->zones + bdev_nr_zones(sb->s_bdev);
 978	for (zone = &zd->zones[1]; zone < end; zone = next) {
 979
 980		next = zone + 1;
 981		if (zonefs_zone_type(zone) != ztype)
 982			continue;
 983
 984		if (WARN_ON_ONCE(n >= zgroup->g_nr_zones))
 985			return -EINVAL;
 986
 987		/*
 988		 * For conventional zones, contiguous zones can be aggregated
 989		 * together to form larger files. Note that this overwrites the
 990		 * length of the first zone of the set of contiguous zones
 991		 * aggregated together. If one offline or read-only zone is
 992		 * found, assume that all zones aggregated have the same
 993		 * condition.
 994		 */
 995		if (ztype == ZONEFS_ZTYPE_CNV &&
 996		    (sbi->s_features & ZONEFS_F_AGGRCNV)) {
 997			for (; next < end; next++) {
 998				if (zonefs_zone_type(next) != ztype)
 999					break;
1000				zone->len += next->len;
1001				zone->capacity += next->capacity;
1002				if (next->cond == BLK_ZONE_COND_READONLY &&
1003				    zone->cond != BLK_ZONE_COND_OFFLINE)
1004					zone->cond = BLK_ZONE_COND_READONLY;
1005				else if (next->cond == BLK_ZONE_COND_OFFLINE)
1006					zone->cond = BLK_ZONE_COND_OFFLINE;
1007			}
1008		}
1009
1010		z = &zgroup->g_zones[n];
1011		if (ztype == ZONEFS_ZTYPE_CNV)
1012			z->z_flags |= ZONEFS_ZONE_CNV;
1013		z->z_sector = zone->start;
1014		z->z_size = zone->len << SECTOR_SHIFT;
1015		if (z->z_size > bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT &&
1016		    !(sbi->s_features & ZONEFS_F_AGGRCNV)) {
1017			zonefs_err(sb,
1018				"Invalid zone size %llu (device zone sectors %llu)\n",
1019				z->z_size,
1020				bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT);
1021			return -EINVAL;
1022		}
1023
1024		z->z_capacity = min_t(loff_t, MAX_LFS_FILESIZE,
1025				      zone->capacity << SECTOR_SHIFT);
1026		z->z_wpoffset = zonefs_check_zone_condition(sb, z, zone);
1027
1028		z->z_mode = S_IFREG | sbi->s_perm;
1029		z->z_uid = sbi->s_uid;
1030		z->z_gid = sbi->s_gid;
1031
1032		/*
1033		 * Let zonefs_inode_update_mode() know that we will need
1034		 * special initialization of the inode mode the first time
1035		 * it is accessed.
1036		 */
1037		z->z_flags |= ZONEFS_ZONE_INIT_MODE;
1038
1039		sb->s_maxbytes = max(z->z_capacity, sb->s_maxbytes);
1040		sbi->s_blocks += z->z_capacity >> sb->s_blocksize_bits;
1041		sbi->s_used_blocks += z->z_wpoffset >> sb->s_blocksize_bits;
1042
1043		/*
1044		 * For sequential zones, make sure that any open zone is closed
1045		 * first to ensure that the initial number of open zones is 0,
1046		 * in sync with the open zone accounting done when the mount
1047		 * option ZONEFS_MNTOPT_EXPLICIT_OPEN is used.
1048		 */
1049		if (ztype == ZONEFS_ZTYPE_SEQ &&
1050		    (zone->cond == BLK_ZONE_COND_IMP_OPEN ||
1051		     zone->cond == BLK_ZONE_COND_EXP_OPEN)) {
1052			ret = zonefs_zone_mgmt(sb, z, REQ_OP_ZONE_CLOSE);
1053			if (ret)
1054				return ret;
1055		}
1056
1057		zonefs_account_active(sb, z);
1058
1059		n++;
1060	}
1061
1062	if (WARN_ON_ONCE(n != zgroup->g_nr_zones))
1063		return -EINVAL;
1064
1065	zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
1066		    zonefs_zgroup_name(ztype),
1067		    zgroup->g_nr_zones,
1068		    zgroup->g_nr_zones > 1 ? "s" : "");
1069
1070	return 0;
1071}
1072
1073static void zonefs_free_zgroups(struct super_block *sb)
1074{
1075	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1076	enum zonefs_ztype ztype;
1077
1078	if (!sbi)
1079		return;
1080
1081	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1082		kvfree(sbi->s_zgroup[ztype].g_zones);
1083		sbi->s_zgroup[ztype].g_zones = NULL;
1084	}
1085}
1086
1087/*
1088 * Create a zone group and populate it with zone files.
1089 */
1090static int zonefs_init_zgroups(struct super_block *sb)
1091{
1092	struct zonefs_zone_data zd;
1093	enum zonefs_ztype ztype;
1094	int ret;
1095
1096	/* First get the device zone information */
1097	memset(&zd, 0, sizeof(struct zonefs_zone_data));
1098	zd.sb = sb;
1099	ret = zonefs_get_zone_info(&zd);
1100	if (ret)
1101		goto cleanup;
1102
1103	/* Allocate and initialize the zone groups */
1104	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1105		ret = zonefs_init_zgroup(sb, &zd, ztype);
1106		if (ret) {
1107			zonefs_info(sb,
1108				    "Zone group \"%s\" initialization failed\n",
1109				    zonefs_zgroup_name(ztype));
1110			break;
1111		}
1112	}
1113
1114cleanup:
1115	zonefs_free_zone_info(&zd);
1116	if (ret)
1117		zonefs_free_zgroups(sb);
1118
1119	return ret;
1120}
1121
1122/*
1123 * Read super block information from the device.
1124 */
1125static int zonefs_read_super(struct super_block *sb)
1126{
1127	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1128	struct zonefs_super *super;
1129	u32 crc, stored_crc;
1130	struct page *page;
1131	struct bio_vec bio_vec;
1132	struct bio bio;
1133	int ret;
1134
1135	page = alloc_page(GFP_KERNEL);
1136	if (!page)
1137		return -ENOMEM;
1138
1139	bio_init(&bio, sb->s_bdev, &bio_vec, 1, REQ_OP_READ);
1140	bio.bi_iter.bi_sector = 0;
1141	__bio_add_page(&bio, page, PAGE_SIZE, 0);
1142
1143	ret = submit_bio_wait(&bio);
1144	if (ret)
1145		goto free_page;
1146
1147	super = page_address(page);
1148
1149	ret = -EINVAL;
1150	if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC)
1151		goto free_page;
1152
1153	stored_crc = le32_to_cpu(super->s_crc);
1154	super->s_crc = 0;
1155	crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super));
1156	if (crc != stored_crc) {
1157		zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)",
1158			   crc, stored_crc);
1159		goto free_page;
1160	}
1161
1162	sbi->s_features = le64_to_cpu(super->s_features);
1163	if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) {
1164		zonefs_err(sb, "Unknown features set 0x%llx\n",
1165			   sbi->s_features);
1166		goto free_page;
1167	}
1168
1169	if (sbi->s_features & ZONEFS_F_UID) {
1170		sbi->s_uid = make_kuid(current_user_ns(),
1171				       le32_to_cpu(super->s_uid));
1172		if (!uid_valid(sbi->s_uid)) {
1173			zonefs_err(sb, "Invalid UID feature\n");
1174			goto free_page;
1175		}
1176	}
1177
1178	if (sbi->s_features & ZONEFS_F_GID) {
1179		sbi->s_gid = make_kgid(current_user_ns(),
1180				       le32_to_cpu(super->s_gid));
1181		if (!gid_valid(sbi->s_gid)) {
1182			zonefs_err(sb, "Invalid GID feature\n");
1183			goto free_page;
1184		}
1185	}
1186
1187	if (sbi->s_features & ZONEFS_F_PERM)
1188		sbi->s_perm = le32_to_cpu(super->s_perm);
1189
1190	if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) {
1191		zonefs_err(sb, "Reserved area is being used\n");
1192		goto free_page;
1193	}
1194
1195	import_uuid(&sbi->s_uuid, super->s_uuid);
1196	ret = 0;
1197
1198free_page:
1199	__free_page(page);
1200
1201	return ret;
1202}
1203
1204static const struct super_operations zonefs_sops = {
1205	.alloc_inode	= zonefs_alloc_inode,
1206	.free_inode	= zonefs_free_inode,
1207	.statfs		= zonefs_statfs,
1208	.remount_fs	= zonefs_remount,
1209	.show_options	= zonefs_show_options,
1210};
1211
1212static int zonefs_get_zgroup_inodes(struct super_block *sb)
1213{
1214	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1215	struct inode *dir_inode;
1216	enum zonefs_ztype ztype;
1217
1218	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1219		if (!sbi->s_zgroup[ztype].g_nr_zones)
1220			continue;
1221
1222		dir_inode = zonefs_get_zgroup_inode(sb, ztype);
1223		if (IS_ERR(dir_inode))
1224			return PTR_ERR(dir_inode);
1225
1226		sbi->s_zgroup[ztype].g_inode = dir_inode;
1227	}
1228
1229	return 0;
1230}
1231
1232static void zonefs_release_zgroup_inodes(struct super_block *sb)
1233{
1234	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1235	enum zonefs_ztype ztype;
1236
1237	if (!sbi)
1238		return;
1239
1240	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1241		if (sbi->s_zgroup[ztype].g_inode) {
1242			iput(sbi->s_zgroup[ztype].g_inode);
1243			sbi->s_zgroup[ztype].g_inode = NULL;
1244		}
1245	}
1246}
1247
1248/*
1249 * Check that the device is zoned. If it is, get the list of zones and create
1250 * sub-directories and files according to the device zone configuration and
1251 * format options.
1252 */
1253static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
1254{
1255	struct zonefs_sb_info *sbi;
1256	struct inode *inode;
1257	enum zonefs_ztype ztype;
1258	int ret;
1259
1260	if (!bdev_is_zoned(sb->s_bdev)) {
1261		zonefs_err(sb, "Not a zoned block device\n");
1262		return -EINVAL;
1263	}
1264
1265	/*
1266	 * Initialize super block information: the maximum file size is updated
1267	 * when the zone files are created so that the format option
1268	 * ZONEFS_F_AGGRCNV which increases the maximum file size of a file
1269	 * beyond the zone size is taken into account.
1270	 */
1271	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1272	if (!sbi)
1273		return -ENOMEM;
1274
1275	spin_lock_init(&sbi->s_lock);
1276	sb->s_fs_info = sbi;
1277	sb->s_magic = ZONEFS_MAGIC;
1278	sb->s_maxbytes = 0;
1279	sb->s_op = &zonefs_sops;
1280	sb->s_time_gran	= 1;
1281
1282	/*
1283	 * The block size is set to the device zone write granularity to ensure
1284	 * that write operations are always aligned according to the device
1285	 * interface constraints.
1286	 */
1287	sb_set_blocksize(sb, bdev_zone_write_granularity(sb->s_bdev));
1288	sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev));
1289	sbi->s_uid = GLOBAL_ROOT_UID;
1290	sbi->s_gid = GLOBAL_ROOT_GID;
1291	sbi->s_perm = 0640;
1292	sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
1293
1294	atomic_set(&sbi->s_wro_seq_files, 0);
1295	sbi->s_max_wro_seq_files = bdev_max_open_zones(sb->s_bdev);
1296	atomic_set(&sbi->s_active_seq_files, 0);
1297	sbi->s_max_active_seq_files = bdev_max_active_zones(sb->s_bdev);
1298
1299	ret = zonefs_read_super(sb);
1300	if (ret)
1301		return ret;
1302
1303	ret = zonefs_parse_options(sb, data);
1304	if (ret)
1305		return ret;
1306
1307	zonefs_info(sb, "Mounting %u zones", bdev_nr_zones(sb->s_bdev));
1308
1309	if (!sbi->s_max_wro_seq_files &&
1310	    !sbi->s_max_active_seq_files &&
1311	    sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
1312		zonefs_info(sb,
1313			"No open and active zone limits. Ignoring explicit_open mount option\n");
1314		sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN;
1315	}
1316
1317	/* Initialize the zone groups */
1318	ret = zonefs_init_zgroups(sb);
1319	if (ret)
1320		goto cleanup;
1321
1322	/* Create the root directory inode */
1323	ret = -ENOMEM;
1324	inode = new_inode(sb);
1325	if (!inode)
1326		goto cleanup;
1327
1328	inode->i_ino = bdev_nr_zones(sb->s_bdev);
1329	inode->i_mode = S_IFDIR | 0555;
1330	simple_inode_init_ts(inode);
1331	inode->i_op = &zonefs_dir_inode_operations;
1332	inode->i_fop = &zonefs_dir_operations;
1333	inode->i_size = 2;
1334	set_nlink(inode, 2);
1335	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1336		if (sbi->s_zgroup[ztype].g_nr_zones) {
1337			inc_nlink(inode);
1338			inode->i_size++;
1339		}
1340	}
1341
1342	sb->s_root = d_make_root(inode);
1343	if (!sb->s_root)
1344		goto cleanup;
1345
1346	/*
1347	 * Take a reference on the zone groups directory inodes
1348	 * to keep them in the inode cache.
1349	 */
1350	ret = zonefs_get_zgroup_inodes(sb);
1351	if (ret)
1352		goto cleanup;
1353
1354	ret = zonefs_sysfs_register(sb);
1355	if (ret)
1356		goto cleanup;
1357
1358	return 0;
1359
1360cleanup:
1361	zonefs_release_zgroup_inodes(sb);
1362	zonefs_free_zgroups(sb);
1363
1364	return ret;
1365}
1366
1367static struct dentry *zonefs_mount(struct file_system_type *fs_type,
1368				   int flags, const char *dev_name, void *data)
1369{
1370	return mount_bdev(fs_type, flags, dev_name, data, zonefs_fill_super);
1371}
1372
1373static void zonefs_kill_super(struct super_block *sb)
1374{
1375	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1376
1377	/* Release the reference on the zone group directory inodes */
1378	zonefs_release_zgroup_inodes(sb);
1379
1380	kill_block_super(sb);
1381
1382	zonefs_sysfs_unregister(sb);
1383	zonefs_free_zgroups(sb);
1384	kfree(sbi);
1385}
1386
1387/*
1388 * File system definition and registration.
1389 */
1390static struct file_system_type zonefs_type = {
1391	.owner		= THIS_MODULE,
1392	.name		= "zonefs",
1393	.mount		= zonefs_mount,
1394	.kill_sb	= zonefs_kill_super,
1395	.fs_flags	= FS_REQUIRES_DEV,
1396};
1397
1398static int __init zonefs_init_inodecache(void)
1399{
1400	zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache",
1401			sizeof(struct zonefs_inode_info), 0,
1402			(SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1403			NULL);
1404	if (zonefs_inode_cachep == NULL)
1405		return -ENOMEM;
1406	return 0;
1407}
1408
1409static void zonefs_destroy_inodecache(void)
1410{
1411	/*
1412	 * Make sure all delayed rcu free inodes are flushed before we
1413	 * destroy the inode cache.
1414	 */
1415	rcu_barrier();
1416	kmem_cache_destroy(zonefs_inode_cachep);
1417}
1418
1419static int __init zonefs_init(void)
1420{
1421	int ret;
1422
1423	BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE);
1424
1425	ret = zonefs_init_inodecache();
1426	if (ret)
1427		return ret;
1428
1429	ret = zonefs_sysfs_init();
1430	if (ret)
1431		goto destroy_inodecache;
1432
1433	ret = register_filesystem(&zonefs_type);
1434	if (ret)
1435		goto sysfs_exit;
1436
1437	return 0;
1438
1439sysfs_exit:
1440	zonefs_sysfs_exit();
1441destroy_inodecache:
1442	zonefs_destroy_inodecache();
1443
1444	return ret;
1445}
1446
1447static void __exit zonefs_exit(void)
1448{
1449	unregister_filesystem(&zonefs_type);
1450	zonefs_sysfs_exit();
1451	zonefs_destroy_inodecache();
1452}
1453
1454MODULE_AUTHOR("Damien Le Moal");
1455MODULE_DESCRIPTION("Zone file system for zoned block devices");
1456MODULE_LICENSE("GPL");
1457MODULE_ALIAS_FS("zonefs");
1458module_init(zonefs_init);
1459module_exit(zonefs_exit);