super.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2
   3#include <linux/ceph/ceph_debug.h>
   4
   5#include <linux/backing-dev.h>
   6#include <linux/ctype.h>
   7#include <linux/fs.h>
   8#include <linux/inet.h>
   9#include <linux/in6.h>
  10#include <linux/module.h>
  11#include <linux/mount.h>
  12#include <linux/fs_context.h>
  13#include <linux/fs_parser.h>
  14#include <linux/sched.h>
  15#include <linux/seq_file.h>
  16#include <linux/slab.h>
  17#include <linux/statfs.h>
  18#include <linux/string.h>
  19
  20#include "super.h"
  21#include "mds_client.h"
  22#include "cache.h"
  23#include "crypto.h"
  24
  25#include <linux/ceph/ceph_features.h>
  26#include <linux/ceph/decode.h>
  27#include <linux/ceph/mon_client.h>
  28#include <linux/ceph/auth.h>
  29#include <linux/ceph/debugfs.h>
  30
  31#include <uapi/linux/magic.h>
  32
  33#define CREATE_TRACE_POINTS
  34#include <trace/events/ceph.h>
  35
  36static DEFINE_SPINLOCK(ceph_fsc_lock);
  37static LIST_HEAD(ceph_fsc_list);
  38
  39/*
  40 * Ceph superblock operations
  41 *
  42 * Handle the basics of mounting, unmounting.
  43 */
  44
  45/*
  46 * super ops
  47 */
  48static void ceph_put_super(struct super_block *s)
  49{
  50	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
  51
  52	doutc(fsc->client, "begin\n");
  53	ceph_fscrypt_free_dummy_policy(fsc);
  54	ceph_mdsc_close_sessions(fsc->mdsc);
  55	doutc(fsc->client, "done\n");
  56}
  57
  58static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  59{
  60	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
  61	struct ceph_mon_client *monc = &fsc->client->monc;
  62	struct ceph_statfs st;
  63	int i, err;
  64	u64 data_pool;
  65
  66	doutc(fsc->client, "begin\n");
  67	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  68		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  69	} else {
  70		data_pool = CEPH_NOPOOL;
  71	}
  72
  73	err = ceph_monc_do_statfs(monc, data_pool, &st);
  74	if (err < 0)
  75		return err;
  76
  77	/* fill in kstatfs */
  78	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  79
  80	/*
  81	 * Express utilization in terms of large blocks to avoid
  82	 * overflow on 32-bit machines.
  83	 */
  84	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  85
  86	/*
  87	 * By default use root quota for stats; fallback to overall filesystem
  88	 * usage if using 'noquotadf' mount option or if the root dir doesn't
  89	 * have max_bytes quota set.
  90	 */
  91	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  92	    !ceph_quota_update_statfs(fsc, buf)) {
  93		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  94		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  95		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  96	}
  97
  98	/*
  99	 * NOTE: for the time being, we make bsize == frsize to humor
 100	 * not-yet-ancient versions of glibc that are broken.
 101	 * Someday, we will probably want to report a real block
 102	 * size...  whatever that may mean for a network file system!
 103	 */
 104	buf->f_bsize = buf->f_frsize;
 105
 106	buf->f_files = le64_to_cpu(st.num_objects);
 107	buf->f_ffree = -1;
 108	buf->f_namelen = NAME_MAX;
 109
 110	/* Must convert the fsid, for consistent values across arches */
 111	buf->f_fsid.val[0] = 0;
 112	mutex_lock(&monc->mutex);
 113	for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
 114		buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
 115	mutex_unlock(&monc->mutex);
 116
 117	/* fold the fs_cluster_id into the upper bits */
 118	buf->f_fsid.val[1] = monc->fs_cluster_id;
 119
 120	doutc(fsc->client, "done\n");
 121	return 0;
 122}
 123
 124static int ceph_sync_fs(struct super_block *sb, int wait)
 125{
 126	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
 127	struct ceph_client *cl = fsc->client;
 128
 129	if (!wait) {
 130		doutc(cl, "(non-blocking)\n");
 131		ceph_flush_dirty_caps(fsc->mdsc);
 132		ceph_flush_cap_releases(fsc->mdsc);
 133		doutc(cl, "(non-blocking) done\n");
 134		return 0;
 135	}
 136
 137	doutc(cl, "(blocking)\n");
 138	ceph_osdc_sync(&fsc->client->osdc);
 139	ceph_mdsc_sync(fsc->mdsc);
 140	doutc(cl, "(blocking) done\n");
 141	return 0;
 142}
 143
 144/*
 145 * mount options
 146 */
 147enum {
 148	Opt_wsize,
 149	Opt_rsize,
 150	Opt_rasize,
 151	Opt_caps_wanted_delay_min,
 152	Opt_caps_wanted_delay_max,
 153	Opt_caps_max,
 154	Opt_readdir_max_entries,
 155	Opt_readdir_max_bytes,
 156	Opt_congestion_kb,
 157	/* int args above */
 158	Opt_snapdirname,
 159	Opt_mds_namespace,
 160	Opt_recover_session,
 161	Opt_source,
 162	Opt_mon_addr,
 163	Opt_test_dummy_encryption,
 164	/* string args above */
 165	Opt_dirstat,
 166	Opt_rbytes,
 167	Opt_asyncreaddir,
 168	Opt_dcache,
 169	Opt_ino32,
 170	Opt_fscache,
 171	Opt_poolperm,
 172	Opt_require_active_mds,
 173	Opt_acl,
 174	Opt_quotadf,
 175	Opt_copyfrom,
 176	Opt_wsync,
 177	Opt_pagecache,
 178	Opt_sparseread,
 179};
 180
 181enum ceph_recover_session_mode {
 182	ceph_recover_session_no,
 183	ceph_recover_session_clean
 184};
 185
 186static const struct constant_table ceph_param_recover[] = {
 187	{ "no",		ceph_recover_session_no },
 188	{ "clean",	ceph_recover_session_clean },
 189	{}
 190};
 191
 192static const struct fs_parameter_spec ceph_mount_parameters[] = {
 193	fsparam_flag_no ("acl",				Opt_acl),
 194	fsparam_flag_no ("asyncreaddir",		Opt_asyncreaddir),
 195	fsparam_s32	("caps_max",			Opt_caps_max),
 196	fsparam_u32	("caps_wanted_delay_max",	Opt_caps_wanted_delay_max),
 197	fsparam_u32	("caps_wanted_delay_min",	Opt_caps_wanted_delay_min),
 198	fsparam_u32	("write_congestion_kb",		Opt_congestion_kb),
 199	fsparam_flag_no ("copyfrom",			Opt_copyfrom),
 200	fsparam_flag_no ("dcache",			Opt_dcache),
 201	fsparam_flag_no ("dirstat",			Opt_dirstat),
 202	fsparam_flag_no	("fsc",				Opt_fscache), // fsc|nofsc
 203	fsparam_string	("fsc",				Opt_fscache), // fsc=...
 204	fsparam_flag_no ("ino32",			Opt_ino32),
 205	fsparam_string	("mds_namespace",		Opt_mds_namespace),
 206	fsparam_string	("mon_addr",			Opt_mon_addr),
 207	fsparam_flag_no ("poolperm",			Opt_poolperm),
 208	fsparam_flag_no ("quotadf",			Opt_quotadf),
 209	fsparam_u32	("rasize",			Opt_rasize),
 210	fsparam_flag_no ("rbytes",			Opt_rbytes),
 211	fsparam_u32	("readdir_max_bytes",		Opt_readdir_max_bytes),
 212	fsparam_u32	("readdir_max_entries",		Opt_readdir_max_entries),
 213	fsparam_enum	("recover_session",		Opt_recover_session, ceph_param_recover),
 214	fsparam_flag_no ("require_active_mds",		Opt_require_active_mds),
 215	fsparam_u32	("rsize",			Opt_rsize),
 216	fsparam_string	("snapdirname",			Opt_snapdirname),
 217	fsparam_string	("source",			Opt_source),
 218	fsparam_flag	("test_dummy_encryption",	Opt_test_dummy_encryption),
 219	fsparam_string	("test_dummy_encryption",	Opt_test_dummy_encryption),
 220	fsparam_u32	("wsize",			Opt_wsize),
 221	fsparam_flag_no	("wsync",			Opt_wsync),
 222	fsparam_flag_no	("pagecache",			Opt_pagecache),
 223	fsparam_flag_no	("sparseread",			Opt_sparseread),
 224	{}
 225};
 226
 227struct ceph_parse_opts_ctx {
 228	struct ceph_options		*copts;
 229	struct ceph_mount_options	*opts;
 230};
 231
 232/*
 233 * Remove adjacent slashes and then the trailing slash, unless it is
 234 * the only remaining character.
 235 *
 236 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 237 */
 238static void canonicalize_path(char *path)
 239{
 240	int i, j = 0;
 241
 242	for (i = 0; path[i] != '\0'; i++) {
 243		if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 244			path[j++] = path[i];
 245	}
 246
 247	if (j > 1 && path[j - 1] == '/')
 248		j--;
 249	path[j] = '\0';
 250}
 251
 252static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
 253				 struct fs_context *fc)
 254{
 255	int r;
 256	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 257	struct ceph_mount_options *fsopt = pctx->opts;
 258
 259	if (*dev_name_end != ':')
 260		return invalfc(fc, "separator ':' missing in source");
 261
 262	r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
 263			       pctx->copts, fc->log.log, ',');
 264	if (r)
 265		return r;
 266
 267	fsopt->new_dev_syntax = false;
 268	return 0;
 269}
 270
 271static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
 272				 struct fs_context *fc)
 273{
 274	size_t len;
 275	struct ceph_fsid fsid;
 276	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 277	struct ceph_options *opts = pctx->copts;
 278	struct ceph_mount_options *fsopt = pctx->opts;
 279	const char *name_start = dev_name;
 280	const char *fsid_start, *fs_name_start;
 281
 282	if (*dev_name_end != '=') {
 283		dout("separator '=' missing in source");
 284		return -EINVAL;
 285	}
 286
 287	fsid_start = strchr(dev_name, '@');
 288	if (!fsid_start)
 289		return invalfc(fc, "missing cluster fsid");
 290	len = fsid_start - name_start;
 291	kfree(opts->name);
 292	opts->name = kstrndup(name_start, len, GFP_KERNEL);
 293	if (!opts->name)
 294		return -ENOMEM;
 295	dout("using %s entity name", opts->name);
 296
 297	++fsid_start; /* start of cluster fsid */
 298	fs_name_start = strchr(fsid_start, '.');
 299	if (!fs_name_start)
 300		return invalfc(fc, "missing file system name");
 301
 302	if (ceph_parse_fsid(fsid_start, &fsid))
 303		return invalfc(fc, "Invalid FSID");
 304
 305	++fs_name_start; /* start of file system name */
 306	len = dev_name_end - fs_name_start;
 307
 308	if (!namespace_equals(fsopt, fs_name_start, len))
 309		return invalfc(fc, "Mismatching mds_namespace");
 310	kfree(fsopt->mds_namespace);
 311	fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
 312	if (!fsopt->mds_namespace)
 313		return -ENOMEM;
 314	dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
 315
 316	fsopt->new_dev_syntax = true;
 317	return 0;
 318}
 319
 320/*
 321 * Parse the source parameter for new device format. Distinguish the device
 322 * spec from the path. Try parsing new device format and fallback to old
 323 * format if needed.
 324 *
 325 * New device syntax will looks like:
 326 *     <device_spec>=/<path>
 327 * where
 328 *     <device_spec> is name@fsid.fsname
 329 *     <path> is optional, but if present must begin with '/'
 330 * (monitor addresses are passed via mount option)
 331 *
 332 * Old device syntax is:
 333 *     <server_spec>[,<server_spec>...]:[<path>]
 334 * where
 335 *     <server_spec> is <ip>[:<port>]
 336 *     <path> is optional, but if present must begin with '/'
 337 */
 338static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
 339{
 340	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 341	struct ceph_mount_options *fsopt = pctx->opts;
 342	char *dev_name = param->string, *dev_name_end;
 343	int ret;
 344
 345	dout("'%s'\n", dev_name);
 346	if (!dev_name || !*dev_name)
 347		return invalfc(fc, "Empty source");
 348
 349	dev_name_end = strchr(dev_name, '/');
 350	if (dev_name_end) {
 351		/*
 352		 * The server_path will include the whole chars from userland
 353		 * including the leading '/'.
 354		 */
 355		kfree(fsopt->server_path);
 356		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 357		if (!fsopt->server_path)
 358			return -ENOMEM;
 359
 360		canonicalize_path(fsopt->server_path);
 361	} else {
 362		dev_name_end = dev_name + strlen(dev_name);
 363	}
 364
 365	dev_name_end--;		/* back up to separator */
 366	if (dev_name_end < dev_name)
 367		return invalfc(fc, "Path missing in source");
 368
 369	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 370	if (fsopt->server_path)
 371		dout("server path '%s'\n", fsopt->server_path);
 372
 373	dout("trying new device syntax");
 374	ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
 375	if (ret) {
 376		if (ret != -EINVAL)
 377			return ret;
 378		dout("trying old device syntax");
 379		ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
 380		if (ret)
 381			return ret;
 382	}
 383
 384	fc->source = param->string;
 385	param->string = NULL;
 386	return 0;
 387}
 388
 389static int ceph_parse_mon_addr(struct fs_parameter *param,
 390			       struct fs_context *fc)
 391{
 392	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 393	struct ceph_mount_options *fsopt = pctx->opts;
 394
 395	kfree(fsopt->mon_addr);
 396	fsopt->mon_addr = param->string;
 397	param->string = NULL;
 398
 399	return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
 400				  pctx->copts, fc->log.log, '/');
 401}
 402
 403static int ceph_parse_mount_param(struct fs_context *fc,
 404				  struct fs_parameter *param)
 405{
 406	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 407	struct ceph_mount_options *fsopt = pctx->opts;
 408	struct fs_parse_result result;
 409	unsigned int mode;
 410	int token, ret;
 411
 412	ret = ceph_parse_param(param, pctx->copts, fc->log.log);
 413	if (ret != -ENOPARAM)
 414		return ret;
 415
 416	token = fs_parse(fc, ceph_mount_parameters, param, &result);
 417	dout("%s: fs_parse '%s' token %d\n",__func__, param->key, token);
 418	if (token < 0)
 419		return token;
 420
 421	switch (token) {
 422	case Opt_snapdirname:
 423		if (strlen(param->string) > NAME_MAX)
 424			return invalfc(fc, "snapdirname too long");
 425		kfree(fsopt->snapdir_name);
 426		fsopt->snapdir_name = param->string;
 427		param->string = NULL;
 428		break;
 429	case Opt_mds_namespace:
 430		if (!namespace_equals(fsopt, param->string, strlen(param->string)))
 431			return invalfc(fc, "Mismatching mds_namespace");
 432		kfree(fsopt->mds_namespace);
 433		fsopt->mds_namespace = param->string;
 434		param->string = NULL;
 435		break;
 436	case Opt_recover_session:
 437		mode = result.uint_32;
 438		if (mode == ceph_recover_session_no)
 439			fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
 440		else if (mode == ceph_recover_session_clean)
 441			fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
 442		else
 443			BUG();
 444		break;
 445	case Opt_source:
 446		if (fc->source)
 447			return invalfc(fc, "Multiple sources specified");
 448		return ceph_parse_source(param, fc);
 449	case Opt_mon_addr:
 450		return ceph_parse_mon_addr(param, fc);
 451	case Opt_wsize:
 452		if (result.uint_32 < PAGE_SIZE ||
 453		    result.uint_32 > CEPH_MAX_WRITE_SIZE)
 454			goto out_of_range;
 455		fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
 456		break;
 457	case Opt_rsize:
 458		if (result.uint_32 < PAGE_SIZE ||
 459		    result.uint_32 > CEPH_MAX_READ_SIZE)
 460			goto out_of_range;
 461		fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
 462		break;
 463	case Opt_rasize:
 464		fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
 465		break;
 466	case Opt_caps_wanted_delay_min:
 467		if (result.uint_32 < 1)
 468			goto out_of_range;
 469		fsopt->caps_wanted_delay_min = result.uint_32;
 470		break;
 471	case Opt_caps_wanted_delay_max:
 472		if (result.uint_32 < 1)
 473			goto out_of_range;
 474		fsopt->caps_wanted_delay_max = result.uint_32;
 475		break;
 476	case Opt_caps_max:
 477		if (result.int_32 < 0)
 478			goto out_of_range;
 479		fsopt->caps_max = result.int_32;
 480		break;
 481	case Opt_readdir_max_entries:
 482		if (result.uint_32 < 1)
 483			goto out_of_range;
 484		fsopt->max_readdir = result.uint_32;
 485		break;
 486	case Opt_readdir_max_bytes:
 487		if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
 488			goto out_of_range;
 489		fsopt->max_readdir_bytes = result.uint_32;
 490		break;
 491	case Opt_congestion_kb:
 492		if (result.uint_32 < 1024) /* at least 1M */
 493			goto out_of_range;
 494		fsopt->congestion_kb = result.uint_32;
 495		break;
 496	case Opt_dirstat:
 497		if (!result.negated)
 498			fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 499		else
 500			fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 501		break;
 502	case Opt_rbytes:
 503		if (!result.negated)
 504			fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 505		else
 506			fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 507		break;
 508	case Opt_asyncreaddir:
 509		if (!result.negated)
 510			fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 511		else
 512			fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 513		break;
 514	case Opt_dcache:
 515		if (!result.negated)
 516			fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 517		else
 518			fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 519		break;
 520	case Opt_ino32:
 521		if (!result.negated)
 522			fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 523		else
 524			fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 525		break;
 526
 527	case Opt_fscache:
 528#ifdef CONFIG_CEPH_FSCACHE
 529		kfree(fsopt->fscache_uniq);
 530		fsopt->fscache_uniq = NULL;
 531		if (result.negated) {
 532			fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 533		} else {
 534			fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 535			fsopt->fscache_uniq = param->string;
 536			param->string = NULL;
 537		}
 538		break;
 539#else
 540		return invalfc(fc, "fscache support is disabled");
 541#endif
 542	case Opt_poolperm:
 543		if (!result.negated)
 544			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 545		else
 546			fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 547		break;
 548	case Opt_require_active_mds:
 549		if (!result.negated)
 550			fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 551		else
 552			fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 553		break;
 554	case Opt_quotadf:
 555		if (!result.negated)
 556			fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
 557		else
 558			fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
 559		break;
 560	case Opt_copyfrom:
 561		if (!result.negated)
 562			fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
 563		else
 564			fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
 565		break;
 566	case Opt_acl:
 567		if (!result.negated) {
 568#ifdef CONFIG_CEPH_FS_POSIX_ACL
 569			fc->sb_flags |= SB_POSIXACL;
 570#else
 571			return invalfc(fc, "POSIX ACL support is disabled");
 572#endif
 573		} else {
 574			fc->sb_flags &= ~SB_POSIXACL;
 575		}
 576		break;
 577	case Opt_wsync:
 578		if (!result.negated)
 579			fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
 580		else
 581			fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
 582		break;
 583	case Opt_pagecache:
 584		if (result.negated)
 585			fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
 586		else
 587			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
 588		break;
 589	case Opt_sparseread:
 590		if (result.negated)
 591			fsopt->flags &= ~CEPH_MOUNT_OPT_SPARSEREAD;
 592		else
 593			fsopt->flags |= CEPH_MOUNT_OPT_SPARSEREAD;
 594		break;
 595	case Opt_test_dummy_encryption:
 596#ifdef CONFIG_FS_ENCRYPTION
 597		fscrypt_free_dummy_policy(&fsopt->dummy_enc_policy);
 598		ret = fscrypt_parse_test_dummy_encryption(param,
 599						&fsopt->dummy_enc_policy);
 600		if (ret == -EINVAL) {
 601			warnfc(fc, "Value of option \"%s\" is unrecognized",
 602			       param->key);
 603		} else if (ret == -EEXIST) {
 604			warnfc(fc, "Conflicting test_dummy_encryption options");
 605			ret = -EINVAL;
 606		}
 607#else
 608		warnfc(fc,
 609		       "FS encryption not supported: test_dummy_encryption mount option ignored");
 610#endif
 611		break;
 612	default:
 613		BUG();
 614	}
 615	return 0;
 616
 617out_of_range:
 618	return invalfc(fc, "%s out of range", param->key);
 619}
 620
 621static void destroy_mount_options(struct ceph_mount_options *args)
 622{
 623	dout("destroy_mount_options %p\n", args);
 624	if (!args)
 625		return;
 626
 627	kfree(args->snapdir_name);
 628	kfree(args->mds_namespace);
 629	kfree(args->server_path);
 630	kfree(args->fscache_uniq);
 631	kfree(args->mon_addr);
 632	fscrypt_free_dummy_policy(&args->dummy_enc_policy);
 633	kfree(args);
 634}
 635
 636static int strcmp_null(const char *s1, const char *s2)
 637{
 638	if (!s1 && !s2)
 639		return 0;
 640	if (s1 && !s2)
 641		return -1;
 642	if (!s1 && s2)
 643		return 1;
 644	return strcmp(s1, s2);
 645}
 646
 647static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 648				 struct ceph_options *new_opt,
 649				 struct ceph_fs_client *fsc)
 650{
 651	struct ceph_mount_options *fsopt1 = new_fsopt;
 652	struct ceph_mount_options *fsopt2 = fsc->mount_options;
 653	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 654	int ret;
 655
 656	ret = memcmp(fsopt1, fsopt2, ofs);
 657	if (ret)
 658		return ret;
 659
 660	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 661	if (ret)
 662		return ret;
 663
 664	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 665	if (ret)
 666		return ret;
 667
 668	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 669	if (ret)
 670		return ret;
 671
 672	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 673	if (ret)
 674		return ret;
 675
 676	ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
 677	if (ret)
 678		return ret;
 679
 680	return ceph_compare_options(new_opt, fsc->client);
 681}
 682
 683/**
 684 * ceph_show_options - Show mount options in /proc/mounts
 685 * @m: seq_file to write to
 686 * @root: root of that (sub)tree
 687 */
 688static int ceph_show_options(struct seq_file *m, struct dentry *root)
 689{
 690	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(root->d_sb);
 691	struct ceph_mount_options *fsopt = fsc->mount_options;
 692	size_t pos;
 693	int ret;
 694
 695	/* a comma between MNT/MS and client options */
 696	seq_putc(m, ',');
 697	pos = m->count;
 698
 699	ret = ceph_print_client_options(m, fsc->client, false);
 700	if (ret)
 701		return ret;
 702
 703	/* retract our comma if no client options */
 704	if (m->count == pos)
 705		m->count--;
 706
 707	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 708		seq_puts(m, ",dirstat");
 709	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 710		seq_puts(m, ",rbytes");
 711	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 712		seq_puts(m, ",noasyncreaddir");
 713	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 714		seq_puts(m, ",nodcache");
 715	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
 716		seq_puts(m, ",ino32");
 717	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 718		seq_show_option(m, "fsc", fsopt->fscache_uniq);
 719	}
 720	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 721		seq_puts(m, ",nopoolperm");
 722	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
 723		seq_puts(m, ",noquotadf");
 724
 725#ifdef CONFIG_CEPH_FS_POSIX_ACL
 726	if (root->d_sb->s_flags & SB_POSIXACL)
 727		seq_puts(m, ",acl");
 728	else
 729		seq_puts(m, ",noacl");
 730#endif
 731
 732	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
 733		seq_puts(m, ",copyfrom");
 734
 735	/* dump mds_namespace when old device syntax is in use */
 736	if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
 737		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
 738
 739	if (fsopt->mon_addr)
 740		seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
 741
 742	if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
 743		seq_show_option(m, "recover_session", "clean");
 744
 745	if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
 746		seq_puts(m, ",wsync");
 747	if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
 748		seq_puts(m, ",nopagecache");
 749	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
 750		seq_puts(m, ",sparseread");
 751
 752	fscrypt_show_test_dummy_encryption(m, ',', root->d_sb);
 753
 754	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
 755		seq_printf(m, ",wsize=%u", fsopt->wsize);
 756	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 757		seq_printf(m, ",rsize=%u", fsopt->rsize);
 758	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 759		seq_printf(m, ",rasize=%u", fsopt->rasize);
 760	if (fsopt->congestion_kb != default_congestion_kb())
 761		seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
 762	if (fsopt->caps_max)
 763		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
 764	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 765		seq_printf(m, ",caps_wanted_delay_min=%u",
 766			 fsopt->caps_wanted_delay_min);
 767	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 768		seq_printf(m, ",caps_wanted_delay_max=%u",
 769			   fsopt->caps_wanted_delay_max);
 770	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 771		seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
 772	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 773		seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
 774	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 775		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 776
 777	return 0;
 778}
 779
 780/*
 781 * handle any mon messages the standard library doesn't understand.
 782 * return error if we don't either.
 783 */
 784static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 785{
 786	struct ceph_fs_client *fsc = client->private;
 787	int type = le16_to_cpu(msg->hdr.type);
 788
 789	switch (type) {
 790	case CEPH_MSG_MDS_MAP:
 791		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 792		return 0;
 793	case CEPH_MSG_FS_MAP_USER:
 794		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 795		return 0;
 796	default:
 797		return -1;
 798	}
 799}
 800
 801/*
 802 * create a new fs client
 803 *
 804 * Success or not, this function consumes @fsopt and @opt.
 805 */
 806static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 807					struct ceph_options *opt)
 808{
 809	struct ceph_fs_client *fsc;
 810	int err;
 811
 812	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 813	if (!fsc) {
 814		err = -ENOMEM;
 815		goto fail;
 816	}
 817
 818	fsc->client = ceph_create_client(opt, fsc);
 819	if (IS_ERR(fsc->client)) {
 820		err = PTR_ERR(fsc->client);
 821		goto fail;
 822	}
 823	opt = NULL; /* fsc->client now owns this */
 824
 825	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 826	ceph_set_opt(fsc->client, ABORT_ON_FULL);
 827
 828	if (!fsopt->mds_namespace) {
 829		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 830				   0, true);
 831	} else {
 832		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 833				   0, false);
 834	}
 835
 836	fsc->mount_options = fsopt;
 837
 838	fsc->sb = NULL;
 839	fsc->mount_state = CEPH_MOUNT_MOUNTING;
 840	fsc->filp_gen = 1;
 841	fsc->have_copy_from2 = true;
 842
 843	atomic_long_set(&fsc->writeback_count, 0);
 844	fsc->write_congested = false;
 845
 846	err = -ENOMEM;
 847	/*
 848	 * The number of concurrent works can be high but they don't need
 849	 * to be processed in parallel, limit concurrency.
 850	 */
 851	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
 852	if (!fsc->inode_wq)
 853		goto fail_client;
 854	fsc->cap_wq = alloc_workqueue("ceph-cap", WQ_PERCPU, 1);
 855	if (!fsc->cap_wq)
 856		goto fail_inode_wq;
 857
 858	hash_init(fsc->async_unlink_conflict);
 859	spin_lock_init(&fsc->async_unlink_conflict_lock);
 860
 861	spin_lock(&ceph_fsc_lock);
 862	list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
 863	spin_unlock(&ceph_fsc_lock);
 864
 865	return fsc;
 866
 867fail_inode_wq:
 868	destroy_workqueue(fsc->inode_wq);
 869fail_client:
 870	ceph_destroy_client(fsc->client);
 871fail:
 872	kfree(fsc);
 873	if (opt)
 874		ceph_destroy_options(opt);
 875	destroy_mount_options(fsopt);
 876	return ERR_PTR(err);
 877}
 878
 879static void flush_fs_workqueues(struct ceph_fs_client *fsc)
 880{
 881	flush_workqueue(fsc->inode_wq);
 882	flush_workqueue(fsc->cap_wq);
 883}
 884
 885static void destroy_fs_client(struct ceph_fs_client *fsc)
 886{
 887	doutc(fsc->client, "%p\n", fsc);
 888
 889	spin_lock(&ceph_fsc_lock);
 890	list_del(&fsc->metric_wakeup);
 891	spin_unlock(&ceph_fsc_lock);
 892
 893	ceph_mdsc_destroy(fsc);
 894	destroy_workqueue(fsc->inode_wq);
 895	destroy_workqueue(fsc->cap_wq);
 896
 897	destroy_mount_options(fsc->mount_options);
 898
 899	ceph_destroy_client(fsc->client);
 900
 901	kfree(fsc);
 902	dout("%s: %p done\n", __func__, fsc);
 903}
 904
 905/*
 906 * caches
 907 */
 908struct kmem_cache *ceph_inode_cachep;
 909struct kmem_cache *ceph_cap_cachep;
 910struct kmem_cache *ceph_cap_snap_cachep;
 911struct kmem_cache *ceph_cap_flush_cachep;
 912struct kmem_cache *ceph_dentry_cachep;
 913struct kmem_cache *ceph_file_cachep;
 914struct kmem_cache *ceph_dir_file_cachep;
 915struct kmem_cache *ceph_mds_request_cachep;
 916mempool_t *ceph_wb_pagevec_pool;
 917
 918static void ceph_inode_init_once(void *foo)
 919{
 920	struct ceph_inode_info *ci = foo;
 921	inode_init_once(&ci->netfs.inode);
 922}
 923
 924static int __init init_caches(void)
 925{
 926	int error = -ENOMEM;
 927
 928	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 929				      sizeof(struct ceph_inode_info),
 930				      __alignof__(struct ceph_inode_info),
 931				      SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
 932				      ceph_inode_init_once);
 933	if (!ceph_inode_cachep)
 934		return -ENOMEM;
 935
 936	ceph_cap_cachep = KMEM_CACHE(ceph_cap, 0);
 937	if (!ceph_cap_cachep)
 938		goto bad_cap;
 939	ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, 0);
 940	if (!ceph_cap_snap_cachep)
 941		goto bad_cap_snap;
 942	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 943					   SLAB_RECLAIM_ACCOUNT);
 944	if (!ceph_cap_flush_cachep)
 945		goto bad_cap_flush;
 946
 947	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 948					SLAB_RECLAIM_ACCOUNT);
 949	if (!ceph_dentry_cachep)
 950		goto bad_dentry;
 951
 952	ceph_file_cachep = KMEM_CACHE(ceph_file_info, 0);
 953	if (!ceph_file_cachep)
 954		goto bad_file;
 955
 956	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, 0);
 957	if (!ceph_dir_file_cachep)
 958		goto bad_dir_file;
 959
 960	ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, 0);
 961	if (!ceph_mds_request_cachep)
 962		goto bad_mds_req;
 963
 964	ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10,
 965	    (CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT) * sizeof(struct page *));
 966	if (!ceph_wb_pagevec_pool)
 967		goto bad_pagevec_pool;
 968
 969	return 0;
 970
 971bad_pagevec_pool:
 972	kmem_cache_destroy(ceph_mds_request_cachep);
 973bad_mds_req:
 974	kmem_cache_destroy(ceph_dir_file_cachep);
 975bad_dir_file:
 976	kmem_cache_destroy(ceph_file_cachep);
 977bad_file:
 978	kmem_cache_destroy(ceph_dentry_cachep);
 979bad_dentry:
 980	kmem_cache_destroy(ceph_cap_flush_cachep);
 981bad_cap_flush:
 982	kmem_cache_destroy(ceph_cap_snap_cachep);
 983bad_cap_snap:
 984	kmem_cache_destroy(ceph_cap_cachep);
 985bad_cap:
 986	kmem_cache_destroy(ceph_inode_cachep);
 987	return error;
 988}
 989
 990static void destroy_caches(void)
 991{
 992	/*
 993	 * Make sure all delayed rcu free inodes are flushed before we
 994	 * destroy cache.
 995	 */
 996	rcu_barrier();
 997
 998	kmem_cache_destroy(ceph_inode_cachep);
 999	kmem_cache_destroy(ceph_cap_cachep);
1000	kmem_cache_destroy(ceph_cap_snap_cachep);
1001	kmem_cache_destroy(ceph_cap_flush_cachep);
1002	kmem_cache_destroy(ceph_dentry_cachep);
1003	kmem_cache_destroy(ceph_file_cachep);
1004	kmem_cache_destroy(ceph_dir_file_cachep);
1005	kmem_cache_destroy(ceph_mds_request_cachep);
1006	mempool_destroy(ceph_wb_pagevec_pool);
1007}
1008
1009static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1010{
1011	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1012	ceph_mdsc_force_umount(fsc->mdsc);
1013	fsc->filp_gen++; // invalidate open files
1014}
1015
1016/*
1017 * ceph_umount_begin - initiate forced umount.  Tear down the
1018 * mount, skipping steps that may hang while waiting for server(s).
1019 */
1020void ceph_umount_begin(struct super_block *sb)
1021{
1022	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1023
1024	doutc(fsc->client, "starting forced umount\n");
1025
1026	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
1027	__ceph_umount_begin(fsc);
1028}
1029
1030static const struct super_operations ceph_super_ops = {
1031	.alloc_inode	= ceph_alloc_inode,
1032	.free_inode	= ceph_free_inode,
1033	.write_inode    = ceph_write_inode,
1034	.drop_inode	= inode_just_drop,
1035	.evict_inode	= ceph_evict_inode,
1036	.sync_fs        = ceph_sync_fs,
1037	.put_super	= ceph_put_super,
1038	.show_options   = ceph_show_options,
1039	.statfs		= ceph_statfs,
1040	.umount_begin   = ceph_umount_begin,
1041};
1042
1043/*
1044 * Bootstrap mount by opening the root directory.  Note the mount
1045 * @started time from caller, and time out if this takes too long.
1046 */
1047static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1048				       const char *path,
1049				       unsigned long started)
1050{
1051	struct ceph_client *cl = fsc->client;
1052	struct ceph_mds_client *mdsc = fsc->mdsc;
1053	struct ceph_mds_request *req = NULL;
1054	int err;
1055	struct dentry *root;
1056
1057	/* open dir */
1058	doutc(cl, "opening '%s'\n", path);
1059	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1060	if (IS_ERR(req))
1061		return ERR_CAST(req);
1062	req->r_path1 = kstrdup(path, GFP_NOFS);
1063	if (!req->r_path1) {
1064		root = ERR_PTR(-ENOMEM);
1065		goto out;
1066	}
1067
1068	req->r_ino1.ino = CEPH_INO_ROOT;
1069	req->r_ino1.snap = CEPH_NOSNAP;
1070	req->r_started = started;
1071	req->r_timeout = fsc->client->options->mount_timeout;
1072	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1073	req->r_num_caps = 2;
1074	err = ceph_mdsc_do_request(mdsc, NULL, req);
1075	if (err == 0) {
1076		struct inode *inode = req->r_target_inode;
1077		req->r_target_inode = NULL;
1078		doutc(cl, "success\n");
1079		root = d_make_root(inode);
1080		if (!root) {
1081			root = ERR_PTR(-ENOMEM);
1082			goto out;
1083		}
1084		doutc(cl, "success, root dentry is %p\n", root);
1085	} else {
1086		root = ERR_PTR(err);
1087	}
1088out:
1089	ceph_mdsc_put_request(req);
1090	return root;
1091}
1092
1093#ifdef CONFIG_FS_ENCRYPTION
1094static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1095					    struct fs_context *fc,
1096					    struct ceph_mount_options *fsopt)
1097{
1098	struct ceph_fs_client *fsc = sb->s_fs_info;
1099
1100	if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1101		return 0;
1102
1103	/* No changing encryption context on remount. */
1104	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1105	    !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1106		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1107						 &fsc->fsc_dummy_enc_policy))
1108			return 0;
1109		errorfc(fc, "Can't set test_dummy_encryption on remount");
1110		return -EINVAL;
1111	}
1112
1113	/* Also make sure fsopt doesn't contain a conflicting value. */
1114	if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1115		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1116						 &fsc->fsc_dummy_enc_policy))
1117			return 0;
1118		errorfc(fc, "Conflicting test_dummy_encryption options");
1119		return -EINVAL;
1120	}
1121
1122	fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1123	memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1124
1125	warnfc(fc, "test_dummy_encryption mode enabled");
1126	return 0;
1127}
1128#else
1129static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1130					    struct fs_context *fc,
1131					    struct ceph_mount_options *fsopt)
1132{
1133	return 0;
1134}
1135#endif
1136
1137/*
1138 * mount: join the ceph cluster, and open root directory.
1139 */
1140static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1141				      struct fs_context *fc)
1142{
1143	struct ceph_client *cl = fsc->client;
1144	int err;
1145	unsigned long started = jiffies;  /* note the start time */
1146	struct dentry *root;
1147
1148	doutc(cl, "mount start %p\n", fsc);
1149	mutex_lock(&fsc->client->mount_mutex);
1150
1151	if (!fsc->sb->s_root) {
1152		const char *path = fsc->mount_options->server_path ?
1153				     fsc->mount_options->server_path + 1 : "";
1154
1155		err = __ceph_open_session(fsc->client);
1156		if (err < 0)
1157			goto out;
1158
1159		/* setup fscache */
1160		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1161			err = ceph_fscache_register_fs(fsc, fc);
1162			if (err < 0)
1163				goto out;
1164		}
1165
1166		err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1167						       fsc->mount_options);
1168		if (err)
1169			goto out;
1170
1171		doutc(cl, "mount opening path '%s'\n", path);
1172
1173		ceph_fs_debugfs_init(fsc);
1174
1175		root = open_root_dentry(fsc, path, started);
1176		if (IS_ERR(root)) {
1177			err = PTR_ERR(root);
1178			goto out;
1179		}
1180		fsc->sb->s_root = dget(root);
1181	} else {
1182		root = dget(fsc->sb->s_root);
1183	}
1184
1185	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1186	doutc(cl, "mount success\n");
1187	mutex_unlock(&fsc->client->mount_mutex);
1188	return root;
1189
1190out:
1191	mutex_unlock(&fsc->client->mount_mutex);
1192	ceph_fscrypt_free_dummy_policy(fsc);
1193	return ERR_PTR(err);
1194}
1195
1196static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1197{
1198	struct ceph_fs_client *fsc = s->s_fs_info;
1199	struct ceph_client *cl = fsc->client;
1200	int ret;
1201
1202	doutc(cl, "%p\n", s);
1203
1204	s->s_maxbytes = MAX_LFS_FILESIZE;
1205
1206	s->s_xattr = ceph_xattr_handlers;
1207	fsc->sb = s;
1208	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1209
1210	s->s_op = &ceph_super_ops;
1211	set_default_d_op(s, &ceph_dentry_ops);
1212	s->s_export_op = &ceph_export_ops;
1213
1214	s->s_time_gran = 1;
1215	s->s_time_min = 0;
1216	s->s_time_max = U32_MAX;
1217	s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1218	s->s_magic = CEPH_SUPER_MAGIC;
1219
1220	ceph_fscrypt_set_ops(s);
1221
1222	ret = set_anon_super_fc(s, fc);
1223	if (ret != 0)
1224		fsc->sb = NULL;
1225	return ret;
1226}
1227
1228/*
1229 * share superblock if same fs AND options
1230 */
1231static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1232{
1233	struct ceph_fs_client *new = fc->s_fs_info;
1234	struct ceph_mount_options *fsopt = new->mount_options;
1235	struct ceph_options *opt = new->client->options;
1236	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1237	struct ceph_client *cl = fsc->client;
1238
1239	doutc(cl, "%p\n", sb);
1240
1241	if (compare_mount_options(fsopt, opt, fsc)) {
1242		doutc(cl, "monitor(s)/mount options don't match\n");
1243		return 0;
1244	}
1245	if ((opt->flags & CEPH_OPT_FSID) &&
1246	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1247		doutc(cl, "fsid doesn't match\n");
1248		return 0;
1249	}
1250	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1251		doutc(cl, "flags differ\n");
1252		return 0;
1253	}
1254
1255	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1256		doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1257		return 0;
1258	}
1259
1260	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1261		doutc(cl, "client has been forcibly unmounted\n");
1262		return 0;
1263	}
1264
1265	return 1;
1266}
1267
1268/*
1269 * construct our own bdi so we can control readahead, etc.
1270 */
1271static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1272
1273static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1274{
1275	int err;
1276
1277	err = super_setup_bdi_name(sb, "ceph-%ld",
1278				   atomic_long_inc_return(&bdi_seq));
1279	if (err)
1280		return err;
1281
1282	/* set ra_pages based on rasize mount option? */
1283	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1284
1285	/* set io_pages based on max osd read size */
1286	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1287
1288	return 0;
1289}
1290
1291static int ceph_get_tree(struct fs_context *fc)
1292{
1293	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1294	struct ceph_mount_options *fsopt = pctx->opts;
1295	struct super_block *sb;
1296	struct ceph_fs_client *fsc;
1297	struct dentry *res;
1298	int (*compare_super)(struct super_block *, struct fs_context *) =
1299		ceph_compare_super;
1300	int err;
1301
1302	dout("ceph_get_tree\n");
1303
1304	if (!fc->source)
1305		return invalfc(fc, "No source");
1306	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1307		return invalfc(fc, "No monitor address");
1308
1309	/* create client (which we may/may not use) */
1310	fsc = create_fs_client(pctx->opts, pctx->copts);
1311	pctx->opts = NULL;
1312	pctx->copts = NULL;
1313	if (IS_ERR(fsc)) {
1314		err = PTR_ERR(fsc);
1315		goto out_final;
1316	}
1317
1318	err = ceph_mdsc_init(fsc);
1319	if (err < 0)
1320		goto out;
1321
1322	if (ceph_test_opt(fsc->client, NOSHARE))
1323		compare_super = NULL;
1324
1325	fc->s_fs_info = fsc;
1326	sb = sget_fc(fc, compare_super, ceph_set_super);
1327	fc->s_fs_info = NULL;
1328	if (IS_ERR(sb)) {
1329		err = PTR_ERR(sb);
1330		goto out;
1331	}
1332
1333	if (ceph_sb_to_fs_client(sb) != fsc) {
1334		destroy_fs_client(fsc);
1335		fsc = ceph_sb_to_fs_client(sb);
1336		dout("get_sb got existing client %p\n", fsc);
1337	} else {
1338		dout("get_sb using new client %p\n", fsc);
1339		err = ceph_setup_bdi(sb, fsc);
1340		if (err < 0)
1341			goto out_splat;
1342	}
1343
1344	res = ceph_real_mount(fsc, fc);
1345	if (IS_ERR(res)) {
1346		err = PTR_ERR(res);
1347		goto out_splat;
1348	}
1349
1350	doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1351		    d_inode(res), ceph_vinop(d_inode(res)));
1352	fc->root = fsc->sb->s_root;
1353	return 0;
1354
1355out_splat:
1356	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1357		pr_info("No mds server is up or the cluster is laggy\n");
1358		err = -EHOSTUNREACH;
1359	}
1360
1361	ceph_mdsc_close_sessions(fsc->mdsc);
1362	deactivate_locked_super(sb);
1363	goto out_final;
1364
1365out:
1366	destroy_fs_client(fsc);
1367out_final:
1368	dout("ceph_get_tree fail %d\n", err);
1369	return err;
1370}
1371
1372static void ceph_free_fc(struct fs_context *fc)
1373{
1374	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1375
1376	if (pctx) {
1377		destroy_mount_options(pctx->opts);
1378		ceph_destroy_options(pctx->copts);
1379		kfree(pctx);
1380	}
1381}
1382
1383static int ceph_reconfigure_fc(struct fs_context *fc)
1384{
1385	int err;
1386	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1387	struct ceph_mount_options *fsopt = pctx->opts;
1388	struct super_block *sb = fc->root->d_sb;
1389	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1390
1391	err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1392	if (err)
1393		return err;
1394
1395	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1396		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1397	else
1398		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1399
1400	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1401		ceph_set_mount_opt(fsc, SPARSEREAD);
1402	else
1403		ceph_clear_mount_opt(fsc, SPARSEREAD);
1404
1405	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1406		kfree(fsc->mount_options->mon_addr);
1407		fsc->mount_options->mon_addr = fsopt->mon_addr;
1408		fsopt->mon_addr = NULL;
1409		pr_notice_client(fsc->client,
1410			"monitor addresses recorded, but not used for reconnection");
1411	}
1412
1413	sync_filesystem(sb);
1414	return 0;
1415}
1416
1417static const struct fs_context_operations ceph_context_ops = {
1418	.free		= ceph_free_fc,
1419	.parse_param	= ceph_parse_mount_param,
1420	.get_tree	= ceph_get_tree,
1421	.reconfigure	= ceph_reconfigure_fc,
1422};
1423
1424/*
1425 * Set up the filesystem mount context.
1426 */
1427static int ceph_init_fs_context(struct fs_context *fc)
1428{
1429	struct ceph_parse_opts_ctx *pctx;
1430	struct ceph_mount_options *fsopt;
1431
1432	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1433	if (!pctx)
1434		return -ENOMEM;
1435
1436	pctx->copts = ceph_alloc_options();
1437	if (!pctx->copts)
1438		goto nomem;
1439
1440	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1441	if (!pctx->opts)
1442		goto nomem;
1443
1444	fsopt = pctx->opts;
1445	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1446
1447	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1448	fsopt->rsize = CEPH_MAX_READ_SIZE;
1449	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1450	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1451	if (!fsopt->snapdir_name)
1452		goto nomem;
1453
1454	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1455	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1456	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1457	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1458	fsopt->congestion_kb = default_congestion_kb();
1459
1460#ifdef CONFIG_CEPH_FS_POSIX_ACL
1461	fc->sb_flags |= SB_POSIXACL;
1462#endif
1463
1464	fc->fs_private = pctx;
1465	fc->ops = &ceph_context_ops;
1466	return 0;
1467
1468nomem:
1469	destroy_mount_options(pctx->opts);
1470	ceph_destroy_options(pctx->copts);
1471	kfree(pctx);
1472	return -ENOMEM;
1473}
1474
1475/*
1476 * Return true if it successfully increases the blocker counter,
1477 * or false if the mdsc is in stopping and flushed state.
1478 */
1479static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1480{
1481	spin_lock(&mdsc->stopping_lock);
1482	if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1483		spin_unlock(&mdsc->stopping_lock);
1484		return false;
1485	}
1486	atomic_inc(&mdsc->stopping_blockers);
1487	spin_unlock(&mdsc->stopping_lock);
1488	return true;
1489}
1490
1491static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1492{
1493	spin_lock(&mdsc->stopping_lock);
1494	if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1495	    mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1496		complete_all(&mdsc->stopping_waiter);
1497	spin_unlock(&mdsc->stopping_lock);
1498}
1499
1500/* For metadata IO requests */
1501bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1502				   struct ceph_mds_session *session)
1503{
1504	mutex_lock(&session->s_mutex);
1505	inc_session_sequence(session);
1506	mutex_unlock(&session->s_mutex);
1507
1508	return __inc_stopping_blocker(mdsc);
1509}
1510
1511void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1512{
1513	__dec_stopping_blocker(mdsc);
1514}
1515
1516/* For data IO requests */
1517bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1518{
1519	return __inc_stopping_blocker(mdsc);
1520}
1521
1522void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1523{
1524	__dec_stopping_blocker(mdsc);
1525}
1526
1527static void ceph_kill_sb(struct super_block *s)
1528{
1529	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1530	struct ceph_client *cl = fsc->client;
1531	struct ceph_mds_client *mdsc = fsc->mdsc;
1532	bool wait;
1533
1534	doutc(cl, "%p\n", s);
1535
1536	ceph_mdsc_pre_umount(mdsc);
1537	flush_fs_workqueues(fsc);
1538
1539	/*
1540	 * Though the kill_anon_super() will finally trigger the
1541	 * sync_filesystem() anyway, we still need to do it here and
1542	 * then bump the stage of shutdown. This will allow us to
1543	 * drop any further message, which will increase the inodes'
1544	 * i_count reference counters but makes no sense any more,
1545	 * from MDSs.
1546	 *
1547	 * Without this when evicting the inodes it may fail in the
1548	 * kill_anon_super(), which will trigger a warning when
1549	 * destroying the fscrypt keyring and then possibly trigger
1550	 * a further crash in ceph module when the iput() tries to
1551	 * evict the inodes later.
1552	 */
1553	sync_filesystem(s);
1554
1555	if (atomic64_read(&mdsc->dirty_folios) > 0) {
1556		wait_queue_head_t *wq = &mdsc->flush_end_wq;
1557		long timeleft = wait_event_killable_timeout(*wq,
1558					atomic64_read(&mdsc->dirty_folios) <= 0,
1559					fsc->client->options->mount_timeout);
1560		if (!timeleft) /* timed out */
1561			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1562		else if (timeleft < 0) /* killed */
1563			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1564	}
1565
1566	spin_lock(&mdsc->stopping_lock);
1567	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1568	wait = !!atomic_read(&mdsc->stopping_blockers);
1569	spin_unlock(&mdsc->stopping_lock);
1570
1571	if (wait && atomic_read(&mdsc->stopping_blockers)) {
1572		long timeleft = wait_for_completion_killable_timeout(
1573					&mdsc->stopping_waiter,
1574					fsc->client->options->mount_timeout);
1575		if (!timeleft) /* timed out */
1576			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1577		else if (timeleft < 0) /* killed */
1578			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1579	}
1580
1581	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1582	kill_anon_super(s);
1583
1584	fsc->client->extra_mon_dispatch = NULL;
1585	ceph_fs_debugfs_cleanup(fsc);
1586
1587	ceph_fscache_unregister_fs(fsc);
1588
1589	destroy_fs_client(fsc);
1590}
1591
1592static struct file_system_type ceph_fs_type = {
1593	.owner		= THIS_MODULE,
1594	.name		= "ceph",
1595	.init_fs_context = ceph_init_fs_context,
1596	.kill_sb	= ceph_kill_sb,
1597	.fs_flags	= FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1598};
1599MODULE_ALIAS_FS("ceph");
1600
1601int ceph_force_reconnect(struct super_block *sb)
1602{
1603	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1604	int err = 0;
1605
1606	fsc->mount_state = CEPH_MOUNT_RECOVER;
1607	__ceph_umount_begin(fsc);
1608
1609	/* Make sure all page caches get invalidated.
1610	 * see remove_session_caps_cb() */
1611	flush_workqueue(fsc->inode_wq);
1612
1613	/* In case that we were blocklisted. This also reset
1614	 * all mon/osd connections */
1615	ceph_reset_client_addr(fsc->client);
1616
1617	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1618
1619	fsc->blocklisted = false;
1620	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1621
1622	if (sb->s_root) {
1623		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1624					CEPH_STAT_CAP_INODE, true);
1625	}
1626	return err;
1627}
1628
1629static int __init init_ceph(void)
1630{
1631	int ret = init_caches();
1632	if (ret)
1633		goto out;
1634
1635	ceph_flock_init();
1636	ret = register_filesystem(&ceph_fs_type);
1637	if (ret)
1638		goto out_caches;
1639
1640	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1641
1642	return 0;
1643
1644out_caches:
1645	destroy_caches();
1646out:
1647	return ret;
1648}
1649
1650static void __exit exit_ceph(void)
1651{
1652	dout("exit_ceph\n");
1653	unregister_filesystem(&ceph_fs_type);
1654	destroy_caches();
1655}
1656
1657static int param_set_metrics(const char *val, const struct kernel_param *kp)
1658{
1659	struct ceph_fs_client *fsc;
1660	int ret;
1661
1662	ret = param_set_bool(val, kp);
1663	if (ret) {
1664		pr_err("Failed to parse sending metrics switch value '%s'\n",
1665		       val);
1666		return ret;
1667	} else if (!disable_send_metrics) {
1668		// wake up all the mds clients
1669		spin_lock(&ceph_fsc_lock);
1670		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1671			metric_schedule_delayed(&fsc->mdsc->metric);
1672		}
1673		spin_unlock(&ceph_fsc_lock);
1674	}
1675
1676	return 0;
1677}
1678
1679static const struct kernel_param_ops param_ops_metrics = {
1680	.set = param_set_metrics,
1681	.get = param_get_bool,
1682};
1683
1684bool disable_send_metrics = false;
1685module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1686MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1687
1688/* for both v1 and v2 syntax */
1689static bool mount_support = true;
1690static const struct kernel_param_ops param_ops_mount_syntax = {
1691	.get = param_get_bool,
1692};
1693module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1694module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1695
1696bool enable_unsafe_idmap = false;
1697module_param(enable_unsafe_idmap, bool, 0644);
1698MODULE_PARM_DESC(enable_unsafe_idmap,
1699		 "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1700
1701module_init(init_ceph);
1702module_exit(exit_ceph);
1703
1704MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1705MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1706MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1707MODULE_DESCRIPTION("Ceph filesystem for Linux");
1708MODULE_LICENSE("GPL");