Linux Audio

Check our new training course

Loading...
v6.9.4
   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
  33static DEFINE_SPINLOCK(ceph_fsc_lock);
  34static LIST_HEAD(ceph_fsc_list);
  35
  36/*
  37 * Ceph superblock operations
  38 *
  39 * Handle the basics of mounting, unmounting.
  40 */
  41
  42/*
  43 * super ops
  44 */
  45static void ceph_put_super(struct super_block *s)
  46{
  47	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
  48
  49	doutc(fsc->client, "begin\n");
  50	ceph_fscrypt_free_dummy_policy(fsc);
  51	ceph_mdsc_close_sessions(fsc->mdsc);
  52	doutc(fsc->client, "done\n");
  53}
  54
  55static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  56{
  57	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
  58	struct ceph_mon_client *monc = &fsc->client->monc;
  59	struct ceph_statfs st;
  60	int i, err;
 
  61	u64 data_pool;
  62
  63	doutc(fsc->client, "begin\n");
  64	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  65		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  66	} else {
  67		data_pool = CEPH_NOPOOL;
  68	}
  69
  70	err = ceph_monc_do_statfs(monc, data_pool, &st);
 
  71	if (err < 0)
  72		return err;
  73
  74	/* fill in kstatfs */
  75	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  76
  77	/*
  78	 * Express utilization in terms of large blocks to avoid
  79	 * overflow on 32-bit machines.
 
 
 
 
 
  80	 */
 
  81	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  82
  83	/*
  84	 * By default use root quota for stats; fallback to overall filesystem
  85	 * usage if using 'noquotadf' mount option or if the root dir doesn't
  86	 * have max_bytes quota set.
  87	 */
  88	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  89	    !ceph_quota_update_statfs(fsc, buf)) {
  90		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  91		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  92		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  93	}
  94
  95	/*
  96	 * NOTE: for the time being, we make bsize == frsize to humor
  97	 * not-yet-ancient versions of glibc that are broken.
  98	 * Someday, we will probably want to report a real block
  99	 * size...  whatever that may mean for a network file system!
 100	 */
 101	buf->f_bsize = buf->f_frsize;
 102
 103	buf->f_files = le64_to_cpu(st.num_objects);
 104	buf->f_ffree = -1;
 105	buf->f_namelen = NAME_MAX;
 106
 107	/* Must convert the fsid, for consistent values across arches */
 108	buf->f_fsid.val[0] = 0;
 109	mutex_lock(&monc->mutex);
 110	for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
 111		buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
 112	mutex_unlock(&monc->mutex);
 113
 114	/* fold the fs_cluster_id into the upper bits */
 115	buf->f_fsid.val[1] = monc->fs_cluster_id;
 116
 117	doutc(fsc->client, "done\n");
 118	return 0;
 119}
 120
 
 121static int ceph_sync_fs(struct super_block *sb, int wait)
 122{
 123	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
 124	struct ceph_client *cl = fsc->client;
 125
 126	if (!wait) {
 127		doutc(cl, "(non-blocking)\n");
 128		ceph_flush_dirty_caps(fsc->mdsc);
 129		doutc(cl, "(non-blocking) done\n");
 130		return 0;
 131	}
 132
 133	doutc(cl, "(blocking)\n");
 134	ceph_osdc_sync(&fsc->client->osdc);
 135	ceph_mdsc_sync(fsc->mdsc);
 136	doutc(cl, "(blocking) done\n");
 137	return 0;
 138}
 139
 140/*
 141 * mount options
 142 */
 143enum {
 144	Opt_wsize,
 145	Opt_rsize,
 146	Opt_rasize,
 147	Opt_caps_wanted_delay_min,
 148	Opt_caps_wanted_delay_max,
 149	Opt_caps_max,
 150	Opt_readdir_max_entries,
 151	Opt_readdir_max_bytes,
 152	Opt_congestion_kb,
 
 153	/* int args above */
 154	Opt_snapdirname,
 155	Opt_mds_namespace,
 156	Opt_recover_session,
 157	Opt_source,
 158	Opt_mon_addr,
 159	Opt_test_dummy_encryption,
 160	/* string args above */
 161	Opt_dirstat,
 
 162	Opt_rbytes,
 
 163	Opt_asyncreaddir,
 
 164	Opt_dcache,
 
 165	Opt_ino32,
 
 166	Opt_fscache,
 
 167	Opt_poolperm,
 
 168	Opt_require_active_mds,
 
 
 169	Opt_acl,
 
 
 170	Opt_quotadf,
 171	Opt_copyfrom,
 172	Opt_wsync,
 173	Opt_pagecache,
 174	Opt_sparseread,
 175};
 176
 177enum ceph_recover_session_mode {
 178	ceph_recover_session_no,
 179	ceph_recover_session_clean
 180};
 181
 182static const struct constant_table ceph_param_recover[] = {
 183	{ "no",		ceph_recover_session_no },
 184	{ "clean",	ceph_recover_session_clean },
 185	{}
 186};
 187
 188static const struct fs_parameter_spec ceph_mount_parameters[] = {
 189	fsparam_flag_no ("acl",				Opt_acl),
 190	fsparam_flag_no ("asyncreaddir",		Opt_asyncreaddir),
 191	fsparam_s32	("caps_max",			Opt_caps_max),
 192	fsparam_u32	("caps_wanted_delay_max",	Opt_caps_wanted_delay_max),
 193	fsparam_u32	("caps_wanted_delay_min",	Opt_caps_wanted_delay_min),
 194	fsparam_u32	("write_congestion_kb",		Opt_congestion_kb),
 195	fsparam_flag_no ("copyfrom",			Opt_copyfrom),
 196	fsparam_flag_no ("dcache",			Opt_dcache),
 197	fsparam_flag_no ("dirstat",			Opt_dirstat),
 198	fsparam_flag_no	("fsc",				Opt_fscache), // fsc|nofsc
 199	fsparam_string	("fsc",				Opt_fscache), // fsc=...
 200	fsparam_flag_no ("ino32",			Opt_ino32),
 201	fsparam_string	("mds_namespace",		Opt_mds_namespace),
 202	fsparam_string	("mon_addr",			Opt_mon_addr),
 203	fsparam_flag_no ("poolperm",			Opt_poolperm),
 204	fsparam_flag_no ("quotadf",			Opt_quotadf),
 205	fsparam_u32	("rasize",			Opt_rasize),
 206	fsparam_flag_no ("rbytes",			Opt_rbytes),
 207	fsparam_u32	("readdir_max_bytes",		Opt_readdir_max_bytes),
 208	fsparam_u32	("readdir_max_entries",		Opt_readdir_max_entries),
 209	fsparam_enum	("recover_session",		Opt_recover_session, ceph_param_recover),
 210	fsparam_flag_no ("require_active_mds",		Opt_require_active_mds),
 211	fsparam_u32	("rsize",			Opt_rsize),
 212	fsparam_string	("snapdirname",			Opt_snapdirname),
 213	fsparam_string	("source",			Opt_source),
 214	fsparam_flag	("test_dummy_encryption",	Opt_test_dummy_encryption),
 215	fsparam_string	("test_dummy_encryption",	Opt_test_dummy_encryption),
 216	fsparam_u32	("wsize",			Opt_wsize),
 217	fsparam_flag_no	("wsync",			Opt_wsync),
 218	fsparam_flag_no	("pagecache",			Opt_pagecache),
 219	fsparam_flag_no	("sparseread",			Opt_sparseread),
 220	{}
 221};
 222
 223struct ceph_parse_opts_ctx {
 224	struct ceph_options		*copts;
 225	struct ceph_mount_options	*opts;
 226};
 227
 228/*
 229 * Remove adjacent slashes and then the trailing slash, unless it is
 230 * the only remaining character.
 231 *
 232 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 233 */
 234static void canonicalize_path(char *path)
 235{
 236	int i, j = 0;
 237
 238	for (i = 0; path[i] != '\0'; i++) {
 239		if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 240			path[j++] = path[i];
 241	}
 242
 243	if (j > 1 && path[j - 1] == '/')
 244		j--;
 245	path[j] = '\0';
 246}
 247
 248/*
 249 * Check if the mds namespace in ceph_mount_options matches
 250 * the passed in namespace string. First time match (when
 251 * ->mds_namespace is NULL) is treated specially, since
 252 * ->mds_namespace needs to be initialized by the caller.
 253 */
 254static int namespace_equals(struct ceph_mount_options *fsopt,
 255			    const char *namespace, size_t len)
 256{
 257	return !(fsopt->mds_namespace &&
 258		 (strlen(fsopt->mds_namespace) != len ||
 259		  strncmp(fsopt->mds_namespace, namespace, len)));
 260}
 261
 262static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
 263				 struct fs_context *fc)
 264{
 265	int r;
 266	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 267	struct ceph_mount_options *fsopt = pctx->opts;
 268
 269	if (*dev_name_end != ':')
 270		return invalfc(fc, "separator ':' missing in source");
 271
 272	r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
 273			       pctx->copts, fc->log.log, ',');
 274	if (r)
 275		return r;
 276
 277	fsopt->new_dev_syntax = false;
 278	return 0;
 279}
 280
 281static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
 282				 struct fs_context *fc)
 283{
 284	size_t len;
 285	struct ceph_fsid fsid;
 286	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 287	struct ceph_mount_options *fsopt = pctx->opts;
 288	char *fsid_start, *fs_name_start;
 289
 290	if (*dev_name_end != '=') {
 291		dout("separator '=' missing in source");
 292		return -EINVAL;
 293	}
 294
 295	fsid_start = strchr(dev_name, '@');
 296	if (!fsid_start)
 297		return invalfc(fc, "missing cluster fsid");
 298	++fsid_start; /* start of cluster fsid */
 299
 300	fs_name_start = strchr(fsid_start, '.');
 301	if (!fs_name_start)
 302		return invalfc(fc, "missing file system name");
 303
 304	if (ceph_parse_fsid(fsid_start, &fsid))
 305		return invalfc(fc, "Invalid FSID");
 306
 307	++fs_name_start; /* start of file system name */
 308	len = dev_name_end - fs_name_start;
 309
 310	if (!namespace_equals(fsopt, fs_name_start, len))
 311		return invalfc(fc, "Mismatching mds_namespace");
 312	kfree(fsopt->mds_namespace);
 313	fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
 314	if (!fsopt->mds_namespace)
 315		return -ENOMEM;
 316	dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
 317
 318	fsopt->new_dev_syntax = true;
 319	return 0;
 320}
 321
 322/*
 323 * Parse the source parameter for new device format. Distinguish the device
 324 * spec from the path. Try parsing new device format and fallback to old
 325 * format if needed.
 326 *
 327 * New device syntax will looks like:
 328 *     <device_spec>=/<path>
 329 * where
 330 *     <device_spec> is name@fsid.fsname
 331 *     <path> is optional, but if present must begin with '/'
 332 * (monitor addresses are passed via mount option)
 333 *
 334 * Old device syntax is:
 335 *     <server_spec>[,<server_spec>...]:[<path>]
 336 * where
 337 *     <server_spec> is <ip>[:<port>]
 338 *     <path> is optional, but if present must begin with '/'
 339 */
 340static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
 341{
 342	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 343	struct ceph_mount_options *fsopt = pctx->opts;
 344	char *dev_name = param->string, *dev_name_end;
 345	int ret;
 346
 347	dout("'%s'\n", dev_name);
 348	if (!dev_name || !*dev_name)
 349		return invalfc(fc, "Empty source");
 350
 351	dev_name_end = strchr(dev_name, '/');
 352	if (dev_name_end) {
 353		/*
 354		 * The server_path will include the whole chars from userland
 355		 * including the leading '/'.
 356		 */
 357		kfree(fsopt->server_path);
 358		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 359		if (!fsopt->server_path)
 360			return -ENOMEM;
 361
 362		canonicalize_path(fsopt->server_path);
 363	} else {
 364		dev_name_end = dev_name + strlen(dev_name);
 365	}
 366
 367	dev_name_end--;		/* back up to separator */
 368	if (dev_name_end < dev_name)
 369		return invalfc(fc, "Path missing in source");
 370
 371	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 372	if (fsopt->server_path)
 373		dout("server path '%s'\n", fsopt->server_path);
 374
 375	dout("trying new device syntax");
 376	ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
 377	if (ret) {
 378		if (ret != -EINVAL)
 379			return ret;
 380		dout("trying old device syntax");
 381		ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
 382		if (ret)
 383			return ret;
 
 
 
 
 
 
 
 384	}
 385
 386	fc->source = param->string;
 387	param->string = NULL;
 388	return 0;
 389}
 390
 391static int ceph_parse_mon_addr(struct fs_parameter *param,
 392			       struct fs_context *fc)
 393{
 394	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 395	struct ceph_mount_options *fsopt = pctx->opts;
 396
 397	kfree(fsopt->mon_addr);
 398	fsopt->mon_addr = param->string;
 399	param->string = NULL;
 400
 401	return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
 402				  pctx->copts, fc->log.log, '/');
 403}
 404
 405static int ceph_parse_mount_param(struct fs_context *fc,
 406				  struct fs_parameter *param)
 407{
 408	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 409	struct ceph_mount_options *fsopt = pctx->opts;
 410	struct fs_parse_result result;
 411	unsigned int mode;
 412	int token, ret;
 413
 414	ret = ceph_parse_param(param, pctx->copts, fc->log.log);
 415	if (ret != -ENOPARAM)
 416		return ret;
 417
 418	token = fs_parse(fc, ceph_mount_parameters, param, &result);
 419	dout("%s: fs_parse '%s' token %d\n",__func__, param->key, token);
 420	if (token < 0)
 421		return token;
 422
 423	switch (token) {
 424	case Opt_snapdirname:
 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", 0, 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, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
 965	if (!ceph_wb_pagevec_pool)
 966		goto bad_pagevec_pool;
 967
 968	return 0;
 969
 970bad_pagevec_pool:
 971	kmem_cache_destroy(ceph_mds_request_cachep);
 972bad_mds_req:
 973	kmem_cache_destroy(ceph_dir_file_cachep);
 974bad_dir_file:
 975	kmem_cache_destroy(ceph_file_cachep);
 976bad_file:
 977	kmem_cache_destroy(ceph_dentry_cachep);
 978bad_dentry:
 979	kmem_cache_destroy(ceph_cap_flush_cachep);
 980bad_cap_flush:
 981	kmem_cache_destroy(ceph_cap_snap_cachep);
 982bad_cap_snap:
 983	kmem_cache_destroy(ceph_cap_cachep);
 984bad_cap:
 985	kmem_cache_destroy(ceph_inode_cachep);
 986	return error;
 987}
 988
 989static void destroy_caches(void)
 990{
 991	/*
 992	 * Make sure all delayed rcu free inodes are flushed before we
 993	 * destroy cache.
 994	 */
 995	rcu_barrier();
 996
 997	kmem_cache_destroy(ceph_inode_cachep);
 998	kmem_cache_destroy(ceph_cap_cachep);
 999	kmem_cache_destroy(ceph_cap_snap_cachep);
1000	kmem_cache_destroy(ceph_cap_flush_cachep);
1001	kmem_cache_destroy(ceph_dentry_cachep);
1002	kmem_cache_destroy(ceph_file_cachep);
1003	kmem_cache_destroy(ceph_dir_file_cachep);
1004	kmem_cache_destroy(ceph_mds_request_cachep);
1005	mempool_destroy(ceph_wb_pagevec_pool);
1006}
1007
1008static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1009{
1010	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1011	ceph_mdsc_force_umount(fsc->mdsc);
1012	fsc->filp_gen++; // invalidate open files
1013}
1014
 
1015/*
1016 * ceph_umount_begin - initiate forced umount.  Tear down the
1017 * mount, skipping steps that may hang while waiting for server(s).
1018 */
1019void ceph_umount_begin(struct super_block *sb)
1020{
1021	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1022
1023	doutc(fsc->client, "starting forced umount\n");
1024	if (!fsc)
1025		return;
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	= generic_delete_inode,
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, started);
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	s->s_d_op = &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
1219	ceph_fscrypt_set_ops(s);
1220
1221	ret = set_anon_super_fc(s, fc);
1222	if (ret != 0)
1223		fsc->sb = NULL;
 
 
 
 
 
 
1224	return ret;
1225}
1226
1227/*
1228 * share superblock if same fs AND options
1229 */
1230static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1231{
1232	struct ceph_fs_client *new = fc->s_fs_info;
1233	struct ceph_mount_options *fsopt = new->mount_options;
1234	struct ceph_options *opt = new->client->options;
1235	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1236	struct ceph_client *cl = fsc->client;
1237
1238	doutc(cl, "%p\n", sb);
1239
1240	if (compare_mount_options(fsopt, opt, fsc)) {
1241		doutc(cl, "monitor(s)/mount options don't match\n");
1242		return 0;
1243	}
1244	if ((opt->flags & CEPH_OPT_FSID) &&
1245	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1246		doutc(cl, "fsid doesn't match\n");
1247		return 0;
1248	}
1249	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1250		doutc(cl, "flags differ\n");
1251		return 0;
1252	}
1253
1254	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1255		doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1256		return 0;
1257	}
1258
1259	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1260		doutc(cl, "client has been forcibly unmounted\n");
1261		return 0;
1262	}
1263
1264	return 1;
1265}
1266
1267/*
1268 * construct our own bdi so we can control readahead, etc.
1269 */
1270static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1271
1272static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1273{
1274	int err;
1275
1276	err = super_setup_bdi_name(sb, "ceph-%ld",
1277				   atomic_long_inc_return(&bdi_seq));
1278	if (err)
1279		return err;
1280
1281	/* set ra_pages based on rasize mount option? */
1282	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1283
1284	/* set io_pages based on max osd read size */
1285	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1286
1287	return 0;
1288}
1289
1290static int ceph_get_tree(struct fs_context *fc)
 
1291{
1292	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1293	struct ceph_mount_options *fsopt = pctx->opts;
1294	struct super_block *sb;
1295	struct ceph_fs_client *fsc;
1296	struct dentry *res;
1297	int (*compare_super)(struct super_block *, struct fs_context *) =
1298		ceph_compare_super;
1299	int err;
 
 
 
1300
1301	dout("ceph_get_tree\n");
1302
1303	if (!fc->source)
1304		return invalfc(fc, "No source");
1305	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1306		return invalfc(fc, "No monitor address");
 
 
 
 
1307
1308	/* create client (which we may/may not use) */
1309	fsc = create_fs_client(pctx->opts, pctx->copts);
1310	pctx->opts = NULL;
1311	pctx->copts = NULL;
1312	if (IS_ERR(fsc)) {
1313		err = PTR_ERR(fsc);
 
 
1314		goto out_final;
1315	}
1316
1317	err = ceph_mdsc_init(fsc);
1318	if (err < 0)
 
1319		goto out;
 
1320
1321	if (ceph_test_opt(fsc->client, NOSHARE))
1322		compare_super = NULL;
1323
1324	fc->s_fs_info = fsc;
1325	sb = sget_fc(fc, compare_super, ceph_set_super);
1326	fc->s_fs_info = NULL;
1327	if (IS_ERR(sb)) {
1328		err = PTR_ERR(sb);
1329		goto out;
1330	}
1331
1332	if (ceph_sb_to_fs_client(sb) != fsc) {
 
1333		destroy_fs_client(fsc);
1334		fsc = ceph_sb_to_fs_client(sb);
1335		dout("get_sb got existing client %p\n", fsc);
1336	} else {
1337		dout("get_sb using new client %p\n", fsc);
1338		err = ceph_setup_bdi(sb, fsc);
1339		if (err < 0)
 
1340			goto out_splat;
 
1341	}
1342
1343	res = ceph_real_mount(fsc, fc);
1344	if (IS_ERR(res)) {
1345		err = PTR_ERR(res);
1346		goto out_splat;
1347	}
1348
1349	doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1350		    d_inode(res), ceph_vinop(d_inode(res)));
1351	fc->root = fsc->sb->s_root;
1352	return 0;
1353
1354out_splat:
1355	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1356		pr_info("No mds server is up or the cluster is laggy\n");
1357		err = -EHOSTUNREACH;
1358	}
1359
1360	ceph_mdsc_close_sessions(fsc->mdsc);
1361	deactivate_locked_super(sb);
1362	goto out_final;
1363
1364out:
 
1365	destroy_fs_client(fsc);
1366out_final:
1367	dout("ceph_get_tree fail %d\n", err);
1368	return err;
1369}
1370
1371static void ceph_free_fc(struct fs_context *fc)
1372{
1373	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1374
1375	if (pctx) {
1376		destroy_mount_options(pctx->opts);
1377		ceph_destroy_options(pctx->copts);
1378		kfree(pctx);
1379	}
1380}
1381
1382static int ceph_reconfigure_fc(struct fs_context *fc)
1383{
1384	int err;
1385	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1386	struct ceph_mount_options *fsopt = pctx->opts;
1387	struct super_block *sb = fc->root->d_sb;
1388	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1389
1390	err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1391	if (err)
1392		return err;
1393
1394	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1395		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1396	else
1397		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1398
1399	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1400		ceph_set_mount_opt(fsc, SPARSEREAD);
1401	else
1402		ceph_clear_mount_opt(fsc, SPARSEREAD);
1403
1404	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1405		kfree(fsc->mount_options->mon_addr);
1406		fsc->mount_options->mon_addr = fsopt->mon_addr;
1407		fsopt->mon_addr = NULL;
1408		pr_notice_client(fsc->client,
1409			"monitor addresses recorded, but not used for reconnection");
1410	}
1411
1412	sync_filesystem(sb);
1413	return 0;
1414}
1415
1416static const struct fs_context_operations ceph_context_ops = {
1417	.free		= ceph_free_fc,
1418	.parse_param	= ceph_parse_mount_param,
1419	.get_tree	= ceph_get_tree,
1420	.reconfigure	= ceph_reconfigure_fc,
1421};
1422
1423/*
1424 * Set up the filesystem mount context.
1425 */
1426static int ceph_init_fs_context(struct fs_context *fc)
1427{
1428	struct ceph_parse_opts_ctx *pctx;
1429	struct ceph_mount_options *fsopt;
1430
1431	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1432	if (!pctx)
1433		return -ENOMEM;
1434
1435	pctx->copts = ceph_alloc_options();
1436	if (!pctx->copts)
1437		goto nomem;
1438
1439	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1440	if (!pctx->opts)
1441		goto nomem;
1442
1443	fsopt = pctx->opts;
1444	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1445
1446	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1447	fsopt->rsize = CEPH_MAX_READ_SIZE;
1448	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1449	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1450	if (!fsopt->snapdir_name)
1451		goto nomem;
1452
1453	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1454	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1455	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1456	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1457	fsopt->congestion_kb = default_congestion_kb();
1458
1459#ifdef CONFIG_CEPH_FS_POSIX_ACL
1460	fc->sb_flags |= SB_POSIXACL;
1461#endif
1462
1463	fc->fs_private = pctx;
1464	fc->ops = &ceph_context_ops;
1465	return 0;
1466
1467nomem:
1468	destroy_mount_options(pctx->opts);
1469	ceph_destroy_options(pctx->copts);
1470	kfree(pctx);
1471	return -ENOMEM;
1472}
1473
1474/*
1475 * Return true if it successfully increases the blocker counter,
1476 * or false if the mdsc is in stopping and flushed state.
1477 */
1478static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1479{
1480	spin_lock(&mdsc->stopping_lock);
1481	if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1482		spin_unlock(&mdsc->stopping_lock);
1483		return false;
1484	}
1485	atomic_inc(&mdsc->stopping_blockers);
1486	spin_unlock(&mdsc->stopping_lock);
1487	return true;
1488}
1489
1490static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1491{
1492	spin_lock(&mdsc->stopping_lock);
1493	if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1494	    mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1495		complete_all(&mdsc->stopping_waiter);
1496	spin_unlock(&mdsc->stopping_lock);
1497}
1498
1499/* For metadata IO requests */
1500bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1501				   struct ceph_mds_session *session)
1502{
1503	mutex_lock(&session->s_mutex);
1504	inc_session_sequence(session);
1505	mutex_unlock(&session->s_mutex);
1506
1507	return __inc_stopping_blocker(mdsc);
1508}
1509
1510void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1511{
1512	__dec_stopping_blocker(mdsc);
1513}
1514
1515/* For data IO requests */
1516bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1517{
1518	return __inc_stopping_blocker(mdsc);
1519}
1520
1521void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1522{
1523	__dec_stopping_blocker(mdsc);
1524}
1525
1526static void ceph_kill_sb(struct super_block *s)
1527{
1528	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1529	struct ceph_client *cl = fsc->client;
1530	struct ceph_mds_client *mdsc = fsc->mdsc;
1531	bool wait;
1532
1533	doutc(cl, "%p\n", s);
1534
1535	ceph_mdsc_pre_umount(mdsc);
1536	flush_fs_workqueues(fsc);
1537
1538	/*
1539	 * Though the kill_anon_super() will finally trigger the
1540	 * sync_filesystem() anyway, we still need to do it here and
1541	 * then bump the stage of shutdown. This will allow us to
1542	 * drop any further message, which will increase the inodes'
1543	 * i_count reference counters but makes no sense any more,
1544	 * from MDSs.
1545	 *
1546	 * Without this when evicting the inodes it may fail in the
1547	 * kill_anon_super(), which will trigger a warning when
1548	 * destroying the fscrypt keyring and then possibly trigger
1549	 * a further crash in ceph module when the iput() tries to
1550	 * evict the inodes later.
1551	 */
1552	sync_filesystem(s);
1553
1554	spin_lock(&mdsc->stopping_lock);
1555	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1556	wait = !!atomic_read(&mdsc->stopping_blockers);
1557	spin_unlock(&mdsc->stopping_lock);
1558
1559	if (wait && atomic_read(&mdsc->stopping_blockers)) {
1560		long timeleft = wait_for_completion_killable_timeout(
1561					&mdsc->stopping_waiter,
1562					fsc->client->options->mount_timeout);
1563		if (!timeleft) /* timed out */
1564			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1565		else if (timeleft < 0) /* killed */
1566			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1567	}
1568
1569	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1570	kill_anon_super(s);
1571
1572	fsc->client->extra_mon_dispatch = NULL;
1573	ceph_fs_debugfs_cleanup(fsc);
1574
1575	ceph_fscache_unregister_fs(fsc);
1576
 
 
1577	destroy_fs_client(fsc);
 
1578}
1579
1580static struct file_system_type ceph_fs_type = {
1581	.owner		= THIS_MODULE,
1582	.name		= "ceph",
1583	.init_fs_context = ceph_init_fs_context,
1584	.kill_sb	= ceph_kill_sb,
1585	.fs_flags	= FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1586};
1587MODULE_ALIAS_FS("ceph");
1588
1589int ceph_force_reconnect(struct super_block *sb)
1590{
1591	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1592	int err = 0;
1593
1594	fsc->mount_state = CEPH_MOUNT_RECOVER;
1595	__ceph_umount_begin(fsc);
1596
1597	/* Make sure all page caches get invalidated.
1598	 * see remove_session_caps_cb() */
1599	flush_workqueue(fsc->inode_wq);
1600
1601	/* In case that we were blocklisted. This also reset
1602	 * all mon/osd connections */
1603	ceph_reset_client_addr(fsc->client);
1604
1605	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1606
1607	fsc->blocklisted = false;
1608	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1609
1610	if (sb->s_root) {
1611		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1612					CEPH_STAT_CAP_INODE, true);
1613	}
1614	return err;
1615}
1616
1617static int __init init_ceph(void)
1618{
1619	int ret = init_caches();
1620	if (ret)
1621		goto out;
1622
1623	ceph_flock_init();
 
1624	ret = register_filesystem(&ceph_fs_type);
1625	if (ret)
1626		goto out_caches;
1627
1628	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1629
1630	return 0;
1631
1632out_caches:
 
1633	destroy_caches();
1634out:
1635	return ret;
1636}
1637
1638static void __exit exit_ceph(void)
1639{
1640	dout("exit_ceph\n");
1641	unregister_filesystem(&ceph_fs_type);
 
1642	destroy_caches();
1643}
1644
1645static int param_set_metrics(const char *val, const struct kernel_param *kp)
1646{
1647	struct ceph_fs_client *fsc;
1648	int ret;
1649
1650	ret = param_set_bool(val, kp);
1651	if (ret) {
1652		pr_err("Failed to parse sending metrics switch value '%s'\n",
1653		       val);
1654		return ret;
1655	} else if (!disable_send_metrics) {
1656		// wake up all the mds clients
1657		spin_lock(&ceph_fsc_lock);
1658		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1659			metric_schedule_delayed(&fsc->mdsc->metric);
1660		}
1661		spin_unlock(&ceph_fsc_lock);
1662	}
1663
1664	return 0;
1665}
1666
1667static const struct kernel_param_ops param_ops_metrics = {
1668	.set = param_set_metrics,
1669	.get = param_get_bool,
1670};
1671
1672bool disable_send_metrics = false;
1673module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1674MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1675
1676/* for both v1 and v2 syntax */
1677static bool mount_support = true;
1678static const struct kernel_param_ops param_ops_mount_syntax = {
1679	.get = param_get_bool,
1680};
1681module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1682module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1683
1684bool enable_unsafe_idmap = false;
1685module_param(enable_unsafe_idmap, bool, 0644);
1686MODULE_PARM_DESC(enable_unsafe_idmap,
1687		 "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1688
1689module_init(init_ceph);
1690module_exit(exit_ceph);
1691
1692MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1693MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1694MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1695MODULE_DESCRIPTION("Ceph filesystem for Linux");
1696MODULE_LICENSE("GPL");
v4.17
 
   1
   2#include <linux/ceph/ceph_debug.h>
   3
   4#include <linux/backing-dev.h>
   5#include <linux/ctype.h>
   6#include <linux/fs.h>
   7#include <linux/inet.h>
   8#include <linux/in6.h>
   9#include <linux/module.h>
  10#include <linux/mount.h>
  11#include <linux/parser.h>
 
  12#include <linux/sched.h>
  13#include <linux/seq_file.h>
  14#include <linux/slab.h>
  15#include <linux/statfs.h>
  16#include <linux/string.h>
  17
  18#include "super.h"
  19#include "mds_client.h"
  20#include "cache.h"
 
  21
  22#include <linux/ceph/ceph_features.h>
  23#include <linux/ceph/decode.h>
  24#include <linux/ceph/mon_client.h>
  25#include <linux/ceph/auth.h>
  26#include <linux/ceph/debugfs.h>
  27
 
 
 
 
 
  28/*
  29 * Ceph superblock operations
  30 *
  31 * Handle the basics of mounting, unmounting.
  32 */
  33
  34/*
  35 * super ops
  36 */
  37static void ceph_put_super(struct super_block *s)
  38{
  39	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
  40
  41	dout("put_super\n");
 
  42	ceph_mdsc_close_sessions(fsc->mdsc);
 
  43}
  44
  45static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  46{
  47	struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
  48	struct ceph_monmap *monmap = fsc->client->monc.monmap;
  49	struct ceph_statfs st;
  50	u64 fsid;
  51	int err;
  52	u64 data_pool;
  53
 
  54	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  55		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  56	} else {
  57		data_pool = CEPH_NOPOOL;
  58	}
  59
  60	dout("statfs\n");
  61	err = ceph_monc_do_statfs(&fsc->client->monc, data_pool, &st);
  62	if (err < 0)
  63		return err;
  64
  65	/* fill in kstatfs */
  66	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  67
  68	/*
  69	 * express utilization in terms of large blocks to avoid
  70	 * overflow on 32-bit machines.
  71	 *
  72	 * NOTE: for the time being, we make bsize == frsize to humor
  73	 * not-yet-ancient versions of glibc that are broken.
  74	 * Someday, we will probably want to report a real block
  75	 * size...  whatever that may mean for a network file system!
  76	 */
  77	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
  78	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  79
  80	/*
  81	 * By default use root quota for stats; fallback to overall filesystem
  82	 * usage if using 'noquotadf' mount option or if the root dir doesn't
  83	 * have max_bytes quota set.
  84	 */
  85	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  86	    !ceph_quota_update_statfs(fsc, buf)) {
  87		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  88		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  89		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  90	}
  91
 
 
 
 
 
 
 
 
  92	buf->f_files = le64_to_cpu(st.num_objects);
  93	buf->f_ffree = -1;
  94	buf->f_namelen = NAME_MAX;
  95
  96	/* Must convert the fsid, for consistent values across arches */
  97	fsid = le64_to_cpu(*(__le64 *)(&monmap->fsid)) ^
  98	       le64_to_cpu(*((__le64 *)&monmap->fsid + 1));
  99	buf->f_fsid.val[0] = fsid & 0xffffffff;
 100	buf->f_fsid.val[1] = fsid >> 32;
 
 101
 
 
 
 
 102	return 0;
 103}
 104
 105
 106static int ceph_sync_fs(struct super_block *sb, int wait)
 107{
 108	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 
 109
 110	if (!wait) {
 111		dout("sync_fs (non-blocking)\n");
 112		ceph_flush_dirty_caps(fsc->mdsc);
 113		dout("sync_fs (non-blocking) done\n");
 114		return 0;
 115	}
 116
 117	dout("sync_fs (blocking)\n");
 118	ceph_osdc_sync(&fsc->client->osdc);
 119	ceph_mdsc_sync(fsc->mdsc);
 120	dout("sync_fs (blocking) done\n");
 121	return 0;
 122}
 123
 124/*
 125 * mount options
 126 */
 127enum {
 128	Opt_wsize,
 129	Opt_rsize,
 130	Opt_rasize,
 131	Opt_caps_wanted_delay_min,
 132	Opt_caps_wanted_delay_max,
 
 133	Opt_readdir_max_entries,
 134	Opt_readdir_max_bytes,
 135	Opt_congestion_kb,
 136	Opt_last_int,
 137	/* int args above */
 138	Opt_snapdirname,
 139	Opt_mds_namespace,
 140	Opt_fscache_uniq,
 141	Opt_last_string,
 
 
 142	/* string args above */
 143	Opt_dirstat,
 144	Opt_nodirstat,
 145	Opt_rbytes,
 146	Opt_norbytes,
 147	Opt_asyncreaddir,
 148	Opt_noasyncreaddir,
 149	Opt_dcache,
 150	Opt_nodcache,
 151	Opt_ino32,
 152	Opt_noino32,
 153	Opt_fscache,
 154	Opt_nofscache,
 155	Opt_poolperm,
 156	Opt_nopoolperm,
 157	Opt_require_active_mds,
 158	Opt_norequire_active_mds,
 159#ifdef CONFIG_CEPH_FS_POSIX_ACL
 160	Opt_acl,
 161#endif
 162	Opt_noacl,
 163	Opt_quotadf,
 164	Opt_noquotadf,
 
 
 
 
 
 
 
 
 165};
 166
 167static match_table_t fsopt_tokens = {
 168	{Opt_wsize, "wsize=%d"},
 169	{Opt_rsize, "rsize=%d"},
 170	{Opt_rasize, "rasize=%d"},
 171	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 172	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 173	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
 174	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 175	{Opt_congestion_kb, "write_congestion_kb=%d"},
 176	/* int args above */
 177	{Opt_snapdirname, "snapdirname=%s"},
 178	{Opt_mds_namespace, "mds_namespace=%s"},
 179	{Opt_fscache_uniq, "fsc=%s"},
 180	/* string args above */
 181	{Opt_dirstat, "dirstat"},
 182	{Opt_nodirstat, "nodirstat"},
 183	{Opt_rbytes, "rbytes"},
 184	{Opt_norbytes, "norbytes"},
 185	{Opt_asyncreaddir, "asyncreaddir"},
 186	{Opt_noasyncreaddir, "noasyncreaddir"},
 187	{Opt_dcache, "dcache"},
 188	{Opt_nodcache, "nodcache"},
 189	{Opt_ino32, "ino32"},
 190	{Opt_noino32, "noino32"},
 191	{Opt_fscache, "fsc"},
 192	{Opt_nofscache, "nofsc"},
 193	{Opt_poolperm, "poolperm"},
 194	{Opt_nopoolperm, "nopoolperm"},
 195	{Opt_require_active_mds, "require_active_mds"},
 196	{Opt_norequire_active_mds, "norequire_active_mds"},
 197#ifdef CONFIG_CEPH_FS_POSIX_ACL
 198	{Opt_acl, "acl"},
 199#endif
 200	{Opt_noacl, "noacl"},
 201	{Opt_quotadf, "quotadf"},
 202	{Opt_noquotadf, "noquotadf"},
 203	{-1, NULL}
 
 
 
 
 
 
 
 204};
 205
 206static int parse_fsopt_token(char *c, void *private)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 207{
 208	struct ceph_mount_options *fsopt = private;
 209	substring_t argstr[MAX_OPT_ARGS];
 210	int token, intval, ret;
 
 
 211
 212	token = match_token((char *)c, fsopt_tokens, argstr);
 213	if (token < 0)
 214		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 215
 216	if (token < Opt_last_int) {
 217		ret = match_int(&argstr[0], &intval);
 218		if (ret < 0) {
 219			pr_err("bad mount option arg (not int) "
 220			       "at '%s'\n", c);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 221			return ret;
 222		}
 223		dout("got int token %d val %d\n", token, intval);
 224	} else if (token > Opt_last_int && token < Opt_last_string) {
 225		dout("got string token %d val %s\n", token,
 226		     argstr[0].from);
 227	} else {
 228		dout("got token %d\n", token);
 229	}
 230
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 231	switch (token) {
 232	case Opt_snapdirname:
 233		kfree(fsopt->snapdir_name);
 234		fsopt->snapdir_name = kstrndup(argstr[0].from,
 235					       argstr[0].to-argstr[0].from,
 236					       GFP_KERNEL);
 237		if (!fsopt->snapdir_name)
 238			return -ENOMEM;
 239		break;
 240	case Opt_mds_namespace:
 
 
 241		kfree(fsopt->mds_namespace);
 242		fsopt->mds_namespace = kstrndup(argstr[0].from,
 243						argstr[0].to-argstr[0].from,
 244						GFP_KERNEL);
 245		if (!fsopt->mds_namespace)
 246			return -ENOMEM;
 247		break;
 248	case Opt_fscache_uniq:
 249		kfree(fsopt->fscache_uniq);
 250		fsopt->fscache_uniq = kstrndup(argstr[0].from,
 251					       argstr[0].to-argstr[0].from,
 252					       GFP_KERNEL);
 253		if (!fsopt->fscache_uniq)
 254			return -ENOMEM;
 255		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 256		break;
 257		/* misc */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 258	case Opt_wsize:
 259		if (intval < PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
 260			return -EINVAL;
 261		fsopt->wsize = ALIGN(intval, PAGE_SIZE);
 
 262		break;
 263	case Opt_rsize:
 264		if (intval < PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
 265			return -EINVAL;
 266		fsopt->rsize = ALIGN(intval, PAGE_SIZE);
 
 267		break;
 268	case Opt_rasize:
 269		if (intval < 0)
 270			return -EINVAL;
 271		fsopt->rasize = ALIGN(intval + PAGE_SIZE - 1, PAGE_SIZE);
 272		break;
 273	case Opt_caps_wanted_delay_min:
 274		if (intval < 1)
 275			return -EINVAL;
 276		fsopt->caps_wanted_delay_min = intval;
 277		break;
 278	case Opt_caps_wanted_delay_max:
 279		if (intval < 1)
 280			return -EINVAL;
 281		fsopt->caps_wanted_delay_max = intval;
 
 
 
 
 
 282		break;
 283	case Opt_readdir_max_entries:
 284		if (intval < 1)
 285			return -EINVAL;
 286		fsopt->max_readdir = intval;
 287		break;
 288	case Opt_readdir_max_bytes:
 289		if (intval < PAGE_SIZE && intval != 0)
 290			return -EINVAL;
 291		fsopt->max_readdir_bytes = intval;
 292		break;
 293	case Opt_congestion_kb:
 294		if (intval < 1024) /* at least 1M */
 295			return -EINVAL;
 296		fsopt->congestion_kb = intval;
 297		break;
 298	case Opt_dirstat:
 299		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 300		break;
 301	case Opt_nodirstat:
 302		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 303		break;
 304	case Opt_rbytes:
 305		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 306		break;
 307	case Opt_norbytes:
 308		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 309		break;
 310	case Opt_asyncreaddir:
 311		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 312		break;
 313	case Opt_noasyncreaddir:
 314		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 315		break;
 316	case Opt_dcache:
 317		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 318		break;
 319	case Opt_nodcache:
 320		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 321		break;
 322	case Opt_ino32:
 323		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 324		break;
 325	case Opt_noino32:
 326		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 327		break;
 
 328	case Opt_fscache:
 329		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 330		kfree(fsopt->fscache_uniq);
 331		fsopt->fscache_uniq = NULL;
 332		break;
 333	case Opt_nofscache:
 334		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 335		kfree(fsopt->fscache_uniq);
 336		fsopt->fscache_uniq = NULL;
 
 
 
 
 
 
 
 337		break;
 
 
 
 338	case Opt_poolperm:
 339		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 340		break;
 341	case Opt_nopoolperm:
 342		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 343		break;
 344	case Opt_require_active_mds:
 345		fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 346		break;
 347	case Opt_norequire_active_mds:
 348		fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 349		break;
 350	case Opt_quotadf:
 351		fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
 352		break;
 353	case Opt_noquotadf:
 354		fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
 
 
 
 
 
 
 355		break;
 
 
 356#ifdef CONFIG_CEPH_FS_POSIX_ACL
 357	case Opt_acl:
 358		fsopt->sb_flags |= SB_POSIXACL;
 
 
 
 
 
 359		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 360#endif
 361	case Opt_noacl:
 362		fsopt->sb_flags &= ~SB_POSIXACL;
 363		break;
 364	default:
 365		BUG_ON(token);
 366	}
 367	return 0;
 
 
 
 368}
 369
 370static void destroy_mount_options(struct ceph_mount_options *args)
 371{
 372	dout("destroy_mount_options %p\n", args);
 
 
 
 373	kfree(args->snapdir_name);
 374	kfree(args->mds_namespace);
 375	kfree(args->server_path);
 376	kfree(args->fscache_uniq);
 
 
 377	kfree(args);
 378}
 379
 380static int strcmp_null(const char *s1, const char *s2)
 381{
 382	if (!s1 && !s2)
 383		return 0;
 384	if (s1 && !s2)
 385		return -1;
 386	if (!s1 && s2)
 387		return 1;
 388	return strcmp(s1, s2);
 389}
 390
 391static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 392				 struct ceph_options *new_opt,
 393				 struct ceph_fs_client *fsc)
 394{
 395	struct ceph_mount_options *fsopt1 = new_fsopt;
 396	struct ceph_mount_options *fsopt2 = fsc->mount_options;
 397	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 398	int ret;
 399
 400	ret = memcmp(fsopt1, fsopt2, ofs);
 401	if (ret)
 402		return ret;
 403
 404	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 405	if (ret)
 406		return ret;
 
 407	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 408	if (ret)
 409		return ret;
 
 410	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 411	if (ret)
 412		return ret;
 
 413	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 414	if (ret)
 415		return ret;
 416
 
 
 
 
 417	return ceph_compare_options(new_opt, fsc->client);
 418}
 419
 420static int parse_mount_options(struct ceph_mount_options **pfsopt,
 421			       struct ceph_options **popt,
 422			       int flags, char *options,
 423			       const char *dev_name)
 424{
 425	struct ceph_mount_options *fsopt;
 426	const char *dev_name_end;
 427	int err;
 428
 429	if (!dev_name || !*dev_name)
 430		return -EINVAL;
 431
 432	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
 433	if (!fsopt)
 434		return -ENOMEM;
 435
 436	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
 437
 438	fsopt->sb_flags = flags;
 439	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
 440
 441	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
 442	fsopt->rsize = CEPH_MAX_READ_SIZE;
 443	fsopt->rasize = CEPH_RASIZE_DEFAULT;
 444	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 445	if (!fsopt->snapdir_name) {
 446		err = -ENOMEM;
 447		goto out;
 448	}
 449
 450	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 451	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 452	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 453	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 454	fsopt->congestion_kb = default_congestion_kb();
 455
 456	/*
 457	 * Distinguish the server list from the path in "dev_name".
 458	 * Internally we do not include the leading '/' in the path.
 459	 *
 460	 * "dev_name" will look like:
 461	 *     <server_spec>[,<server_spec>...]:[<path>]
 462	 * where
 463	 *     <server_spec> is <ip>[:<port>]
 464	 *     <path> is optional, but if present must begin with '/'
 465	 */
 466	dev_name_end = strchr(dev_name, '/');
 467	if (dev_name_end) {
 468		if (strlen(dev_name_end) > 1) {
 469			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 470			if (!fsopt->server_path) {
 471				err = -ENOMEM;
 472				goto out;
 473			}
 474		}
 475	} else {
 476		dev_name_end = dev_name + strlen(dev_name);
 477	}
 478	err = -EINVAL;
 479	dev_name_end--;		/* back up to ':' separator */
 480	if (dev_name_end < dev_name || *dev_name_end != ':') {
 481		pr_err("device name is missing path (no : separator in %s)\n",
 482				dev_name);
 483		goto out;
 484	}
 485	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 486	if (fsopt->server_path)
 487		dout("server path '%s'\n", fsopt->server_path);
 488
 489	*popt = ceph_parse_options(options, dev_name, dev_name_end,
 490				 parse_fsopt_token, (void *)fsopt);
 491	if (IS_ERR(*popt)) {
 492		err = PTR_ERR(*popt);
 493		goto out;
 494	}
 495
 496	/* success */
 497	*pfsopt = fsopt;
 498	return 0;
 499
 500out:
 501	destroy_mount_options(fsopt);
 502	return err;
 503}
 504
 505/**
 506 * ceph_show_options - Show mount options in /proc/mounts
 507 * @m: seq_file to write to
 508 * @root: root of that (sub)tree
 509 */
 510static int ceph_show_options(struct seq_file *m, struct dentry *root)
 511{
 512	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 513	struct ceph_mount_options *fsopt = fsc->mount_options;
 514	size_t pos;
 515	int ret;
 516
 517	/* a comma between MNT/MS and client options */
 518	seq_putc(m, ',');
 519	pos = m->count;
 520
 521	ret = ceph_print_client_options(m, fsc->client);
 522	if (ret)
 523		return ret;
 524
 525	/* retract our comma if no client options */
 526	if (m->count == pos)
 527		m->count--;
 528
 529	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 530		seq_puts(m, ",dirstat");
 531	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 532		seq_puts(m, ",rbytes");
 533	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 534		seq_puts(m, ",noasyncreaddir");
 535	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 536		seq_puts(m, ",nodcache");
 
 
 537	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 538		seq_show_option(m, "fsc", fsopt->fscache_uniq);
 539	}
 540	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 541		seq_puts(m, ",nopoolperm");
 542	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
 543		seq_puts(m, ",noquotadf");
 544
 545#ifdef CONFIG_CEPH_FS_POSIX_ACL
 546	if (fsopt->sb_flags & SB_POSIXACL)
 547		seq_puts(m, ",acl");
 548	else
 549		seq_puts(m, ",noacl");
 550#endif
 551
 552	if (fsopt->mds_namespace)
 
 
 
 
 553		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
 554	if (fsopt->wsize)
 555		seq_printf(m, ",wsize=%d", fsopt->wsize);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 556	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 557		seq_printf(m, ",rsize=%d", fsopt->rsize);
 558	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 559		seq_printf(m, ",rasize=%d", fsopt->rasize);
 560	if (fsopt->congestion_kb != default_congestion_kb())
 561		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 
 
 562	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 563		seq_printf(m, ",caps_wanted_delay_min=%d",
 564			 fsopt->caps_wanted_delay_min);
 565	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 566		seq_printf(m, ",caps_wanted_delay_max=%d",
 567			   fsopt->caps_wanted_delay_max);
 568	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 569		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 570	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 571		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 572	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 573		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 574
 575	return 0;
 576}
 577
 578/*
 579 * handle any mon messages the standard library doesn't understand.
 580 * return error if we don't either.
 581 */
 582static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 583{
 584	struct ceph_fs_client *fsc = client->private;
 585	int type = le16_to_cpu(msg->hdr.type);
 586
 587	switch (type) {
 588	case CEPH_MSG_MDS_MAP:
 589		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 590		return 0;
 591	case CEPH_MSG_FS_MAP_USER:
 592		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 593		return 0;
 594	default:
 595		return -1;
 596	}
 597}
 598
 599/*
 600 * create a new fs client
 
 
 601 */
 602static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 603					struct ceph_options *opt)
 604{
 605	struct ceph_fs_client *fsc;
 606	int page_count;
 607	size_t size;
 608	int err = -ENOMEM;
 609
 610	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 611	if (!fsc)
 612		return ERR_PTR(-ENOMEM);
 
 
 613
 614	fsc->client = ceph_create_client(opt, fsc);
 615	if (IS_ERR(fsc->client)) {
 616		err = PTR_ERR(fsc->client);
 617		goto fail;
 618	}
 
 
 619	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 
 620
 621	if (!fsopt->mds_namespace) {
 622		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 623				   0, true);
 624	} else {
 625		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 626				   0, false);
 627	}
 628
 629	fsc->mount_options = fsopt;
 630
 631	fsc->sb = NULL;
 632	fsc->mount_state = CEPH_MOUNT_MOUNTING;
 
 
 633
 634	atomic_long_set(&fsc->writeback_count, 0);
 
 635
 636	err = -ENOMEM;
 637	/*
 638	 * The number of concurrent works can be high but they don't need
 639	 * to be processed in parallel, limit concurrency.
 640	 */
 641	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 642	if (!fsc->wb_wq)
 643		goto fail_client;
 644	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 645	if (!fsc->pg_inv_wq)
 646		goto fail_wb_wq;
 647	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 648	if (!fsc->trunc_wq)
 649		goto fail_pg_inv_wq;
 650
 651	/* set up mempools */
 652	err = -ENOMEM;
 653	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 654	size = sizeof (struct page *) * (page_count ? page_count : 1);
 655	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 656	if (!fsc->wb_pagevec_pool)
 657		goto fail_trunc_wq;
 658
 659	/* caps */
 660	fsc->min_caps = fsopt->max_readdir;
 661
 662	return fsc;
 663
 664fail_trunc_wq:
 665	destroy_workqueue(fsc->trunc_wq);
 666fail_pg_inv_wq:
 667	destroy_workqueue(fsc->pg_inv_wq);
 668fail_wb_wq:
 669	destroy_workqueue(fsc->wb_wq);
 670fail_client:
 671	ceph_destroy_client(fsc->client);
 672fail:
 673	kfree(fsc);
 
 
 
 674	return ERR_PTR(err);
 675}
 676
 
 
 
 
 
 
 677static void destroy_fs_client(struct ceph_fs_client *fsc)
 678{
 679	dout("destroy_fs_client %p\n", fsc);
 680
 681	destroy_workqueue(fsc->wb_wq);
 682	destroy_workqueue(fsc->pg_inv_wq);
 683	destroy_workqueue(fsc->trunc_wq);
 684
 685	mempool_destroy(fsc->wb_pagevec_pool);
 
 
 686
 687	destroy_mount_options(fsc->mount_options);
 688
 689	ceph_destroy_client(fsc->client);
 690
 691	kfree(fsc);
 692	dout("destroy_fs_client %p done\n", fsc);
 693}
 694
 695/*
 696 * caches
 697 */
 698struct kmem_cache *ceph_inode_cachep;
 699struct kmem_cache *ceph_cap_cachep;
 
 700struct kmem_cache *ceph_cap_flush_cachep;
 701struct kmem_cache *ceph_dentry_cachep;
 702struct kmem_cache *ceph_file_cachep;
 703struct kmem_cache *ceph_dir_file_cachep;
 
 
 704
 705static void ceph_inode_init_once(void *foo)
 706{
 707	struct ceph_inode_info *ci = foo;
 708	inode_init_once(&ci->vfs_inode);
 709}
 710
 711static int __init init_caches(void)
 712{
 713	int error = -ENOMEM;
 714
 715	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 716				      sizeof(struct ceph_inode_info),
 717				      __alignof__(struct ceph_inode_info),
 718				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 719				      SLAB_ACCOUNT, ceph_inode_init_once);
 720	if (!ceph_inode_cachep)
 721		return -ENOMEM;
 722
 723	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
 724	if (!ceph_cap_cachep)
 725		goto bad_cap;
 
 
 
 726	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 727					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 728	if (!ceph_cap_flush_cachep)
 729		goto bad_cap_flush;
 730
 731	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 732					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 733	if (!ceph_dentry_cachep)
 734		goto bad_dentry;
 735
 736	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 737	if (!ceph_file_cachep)
 738		goto bad_file;
 739
 740	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
 741	if (!ceph_dir_file_cachep)
 742		goto bad_dir_file;
 743
 744	error = ceph_fscache_register();
 745	if (error)
 746		goto bad_fscache;
 
 
 
 
 747
 748	return 0;
 749
 750bad_fscache:
 
 
 751	kmem_cache_destroy(ceph_dir_file_cachep);
 752bad_dir_file:
 753	kmem_cache_destroy(ceph_file_cachep);
 754bad_file:
 755	kmem_cache_destroy(ceph_dentry_cachep);
 756bad_dentry:
 757	kmem_cache_destroy(ceph_cap_flush_cachep);
 758bad_cap_flush:
 
 
 759	kmem_cache_destroy(ceph_cap_cachep);
 760bad_cap:
 761	kmem_cache_destroy(ceph_inode_cachep);
 762	return error;
 763}
 764
 765static void destroy_caches(void)
 766{
 767	/*
 768	 * Make sure all delayed rcu free inodes are flushed before we
 769	 * destroy cache.
 770	 */
 771	rcu_barrier();
 772
 773	kmem_cache_destroy(ceph_inode_cachep);
 774	kmem_cache_destroy(ceph_cap_cachep);
 
 775	kmem_cache_destroy(ceph_cap_flush_cachep);
 776	kmem_cache_destroy(ceph_dentry_cachep);
 777	kmem_cache_destroy(ceph_file_cachep);
 778	kmem_cache_destroy(ceph_dir_file_cachep);
 
 
 
 779
 780	ceph_fscache_unregister();
 
 
 
 
 781}
 782
 783
 784/*
 785 * ceph_umount_begin - initiate forced umount.  Tear down down the
 786 * mount, skipping steps that may hang while waiting for server(s).
 787 */
 788static void ceph_umount_begin(struct super_block *sb)
 789{
 790	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 791
 792	dout("ceph_umount_begin - starting forced umount\n");
 793	if (!fsc)
 794		return;
 795	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 796	ceph_mdsc_force_umount(fsc->mdsc);
 797	return;
 798}
 799
 800static const struct super_operations ceph_super_ops = {
 801	.alloc_inode	= ceph_alloc_inode,
 802	.destroy_inode	= ceph_destroy_inode,
 803	.write_inode    = ceph_write_inode,
 804	.drop_inode	= ceph_drop_inode,
 
 805	.sync_fs        = ceph_sync_fs,
 806	.put_super	= ceph_put_super,
 807	.show_options   = ceph_show_options,
 808	.statfs		= ceph_statfs,
 809	.umount_begin   = ceph_umount_begin,
 810};
 811
 812/*
 813 * Bootstrap mount by opening the root directory.  Note the mount
 814 * @started time from caller, and time out if this takes too long.
 815 */
 816static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 817				       const char *path,
 818				       unsigned long started)
 819{
 
 820	struct ceph_mds_client *mdsc = fsc->mdsc;
 821	struct ceph_mds_request *req = NULL;
 822	int err;
 823	struct dentry *root;
 824
 825	/* open dir */
 826	dout("open_root_inode opening '%s'\n", path);
 827	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 828	if (IS_ERR(req))
 829		return ERR_CAST(req);
 830	req->r_path1 = kstrdup(path, GFP_NOFS);
 831	if (!req->r_path1) {
 832		root = ERR_PTR(-ENOMEM);
 833		goto out;
 834	}
 835
 836	req->r_ino1.ino = CEPH_INO_ROOT;
 837	req->r_ino1.snap = CEPH_NOSNAP;
 838	req->r_started = started;
 839	req->r_timeout = fsc->client->options->mount_timeout;
 840	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 841	req->r_num_caps = 2;
 842	err = ceph_mdsc_do_request(mdsc, NULL, req);
 843	if (err == 0) {
 844		struct inode *inode = req->r_target_inode;
 845		req->r_target_inode = NULL;
 846		dout("open_root_inode success\n");
 847		root = d_make_root(inode);
 848		if (!root) {
 849			root = ERR_PTR(-ENOMEM);
 850			goto out;
 851		}
 852		dout("open_root_inode success, root dentry is %p\n", root);
 853	} else {
 854		root = ERR_PTR(err);
 855	}
 856out:
 857	ceph_mdsc_put_request(req);
 858	return root;
 859}
 860
 
 
 
 
 
 
 861
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 862
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 863
 864/*
 865 * mount: join the ceph cluster, and open root directory.
 866 */
 867static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
 
 868{
 
 869	int err;
 870	unsigned long started = jiffies;  /* note the start time */
 871	struct dentry *root;
 872
 873	dout("mount start %p\n", fsc);
 874	mutex_lock(&fsc->client->mount_mutex);
 875
 876	if (!fsc->sb->s_root) {
 877		const char *path;
 
 
 878		err = __ceph_open_session(fsc->client, started);
 879		if (err < 0)
 880			goto out;
 881
 882		/* setup fscache */
 883		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
 884			err = ceph_fscache_register_fs(fsc);
 885			if (err < 0)
 886				goto out;
 887		}
 888
 889		if (!fsc->mount_options->server_path) {
 890			path = "";
 891			dout("mount opening path \\t\n");
 892		} else {
 893			path = fsc->mount_options->server_path + 1;
 894			dout("mount opening path %s\n", path);
 895		}
 896
 897		err = ceph_fs_debugfs_init(fsc);
 898		if (err < 0)
 899			goto out;
 900
 901		root = open_root_dentry(fsc, path, started);
 902		if (IS_ERR(root)) {
 903			err = PTR_ERR(root);
 904			goto out;
 905		}
 906		fsc->sb->s_root = dget(root);
 907	} else {
 908		root = dget(fsc->sb->s_root);
 909	}
 910
 911	fsc->mount_state = CEPH_MOUNT_MOUNTED;
 912	dout("mount success\n");
 913	mutex_unlock(&fsc->client->mount_mutex);
 914	return root;
 915
 916out:
 917	mutex_unlock(&fsc->client->mount_mutex);
 
 918	return ERR_PTR(err);
 919}
 920
 921static int ceph_set_super(struct super_block *s, void *data)
 922{
 923	struct ceph_fs_client *fsc = data;
 
 924	int ret;
 925
 926	dout("set_super %p data %p\n", s, data);
 927
 928	s->s_flags = fsc->mount_options->sb_flags;
 929	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 930
 931	s->s_xattr = ceph_xattr_handlers;
 932	s->s_fs_info = fsc;
 933	fsc->sb = s;
 
 934
 935	s->s_op = &ceph_super_ops;
 936	s->s_d_op = &ceph_dentry_ops;
 937	s->s_export_op = &ceph_export_ops;
 938
 939	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 
 
 
 940
 941	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 
 
 942	if (ret != 0)
 943		goto fail;
 944
 945	return ret;
 946
 947fail:
 948	s->s_fs_info = NULL;
 949	fsc->sb = NULL;
 950	return ret;
 951}
 952
 953/*
 954 * share superblock if same fs AND options
 955 */
 956static int ceph_compare_super(struct super_block *sb, void *data)
 957{
 958	struct ceph_fs_client *new = data;
 959	struct ceph_mount_options *fsopt = new->mount_options;
 960	struct ceph_options *opt = new->client->options;
 961	struct ceph_fs_client *other = ceph_sb_to_client(sb);
 
 962
 963	dout("ceph_compare_super %p\n", sb);
 964
 965	if (compare_mount_options(fsopt, opt, other)) {
 966		dout("monitor(s)/mount options don't match\n");
 967		return 0;
 968	}
 969	if ((opt->flags & CEPH_OPT_FSID) &&
 970	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
 971		dout("fsid doesn't match\n");
 
 
 
 
 
 
 
 
 
 972		return 0;
 973	}
 974	if (fsopt->sb_flags != other->mount_options->sb_flags) {
 975		dout("flags differ\n");
 
 976		return 0;
 977	}
 
 978	return 1;
 979}
 980
 981/*
 982 * construct our own bdi so we can control readahead, etc.
 983 */
 984static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 985
 986static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
 987{
 988	int err;
 989
 990	err = super_setup_bdi_name(sb, "ceph-%ld",
 991				   atomic_long_inc_return(&bdi_seq));
 992	if (err)
 993		return err;
 994
 995	/* set ra_pages based on rasize mount option? */
 996	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
 997
 998	/* set io_pages based on max osd read size */
 999	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1000
1001	return 0;
1002}
1003
1004static struct dentry *ceph_mount(struct file_system_type *fs_type,
1005		       int flags, const char *dev_name, void *data)
1006{
 
 
1007	struct super_block *sb;
1008	struct ceph_fs_client *fsc;
1009	struct dentry *res;
 
 
1010	int err;
1011	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1012	struct ceph_mount_options *fsopt = NULL;
1013	struct ceph_options *opt = NULL;
1014
1015	dout("ceph_mount\n");
1016
1017#ifdef CONFIG_CEPH_FS_POSIX_ACL
1018	flags |= SB_POSIXACL;
1019#endif
1020	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1021	if (err < 0) {
1022		res = ERR_PTR(err);
1023		goto out_final;
1024	}
1025
1026	/* create client (which we may/may not use) */
1027	fsc = create_fs_client(fsopt, opt);
 
 
1028	if (IS_ERR(fsc)) {
1029		res = ERR_CAST(fsc);
1030		destroy_mount_options(fsopt);
1031		ceph_destroy_options(opt);
1032		goto out_final;
1033	}
1034
1035	err = ceph_mdsc_init(fsc);
1036	if (err < 0) {
1037		res = ERR_PTR(err);
1038		goto out;
1039	}
1040
1041	if (ceph_test_opt(fsc->client, NOSHARE))
1042		compare_super = NULL;
1043	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
 
 
 
1044	if (IS_ERR(sb)) {
1045		res = ERR_CAST(sb);
1046		goto out;
1047	}
1048
1049	if (ceph_sb_to_client(sb) != fsc) {
1050		ceph_mdsc_destroy(fsc);
1051		destroy_fs_client(fsc);
1052		fsc = ceph_sb_to_client(sb);
1053		dout("get_sb got existing client %p\n", fsc);
1054	} else {
1055		dout("get_sb using new client %p\n", fsc);
1056		err = ceph_setup_bdi(sb, fsc);
1057		if (err < 0) {
1058			res = ERR_PTR(err);
1059			goto out_splat;
1060		}
1061	}
1062
1063	res = ceph_real_mount(fsc);
1064	if (IS_ERR(res))
 
1065		goto out_splat;
1066	dout("root %p inode %p ino %llx.%llx\n", res,
1067	     d_inode(res), ceph_vinop(d_inode(res)));
1068	return res;
 
 
 
1069
1070out_splat:
 
 
 
 
 
1071	ceph_mdsc_close_sessions(fsc->mdsc);
1072	deactivate_locked_super(sb);
1073	goto out_final;
1074
1075out:
1076	ceph_mdsc_destroy(fsc);
1077	destroy_fs_client(fsc);
1078out_final:
1079	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1080	return res;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1081}
1082
1083static void ceph_kill_sb(struct super_block *s)
1084{
1085	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1086	dev_t dev = s->s_dev;
 
 
1087
1088	dout("kill_sb %p\n", s);
1089
1090	ceph_mdsc_pre_umount(fsc->mdsc);
1091	generic_shutdown_super(s);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1092
1093	fsc->client->extra_mon_dispatch = NULL;
1094	ceph_fs_debugfs_cleanup(fsc);
1095
1096	ceph_fscache_unregister_fs(fsc);
1097
1098	ceph_mdsc_destroy(fsc);
1099
1100	destroy_fs_client(fsc);
1101	free_anon_bdev(dev);
1102}
1103
1104static struct file_system_type ceph_fs_type = {
1105	.owner		= THIS_MODULE,
1106	.name		= "ceph",
1107	.mount		= ceph_mount,
1108	.kill_sb	= ceph_kill_sb,
1109	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1110};
1111MODULE_ALIAS_FS("ceph");
1112
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1113static int __init init_ceph(void)
1114{
1115	int ret = init_caches();
1116	if (ret)
1117		goto out;
1118
1119	ceph_flock_init();
1120	ceph_xattr_init();
1121	ret = register_filesystem(&ceph_fs_type);
1122	if (ret)
1123		goto out_xattr;
1124
1125	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1126
1127	return 0;
1128
1129out_xattr:
1130	ceph_xattr_exit();
1131	destroy_caches();
1132out:
1133	return ret;
1134}
1135
1136static void __exit exit_ceph(void)
1137{
1138	dout("exit_ceph\n");
1139	unregister_filesystem(&ceph_fs_type);
1140	ceph_xattr_exit();
1141	destroy_caches();
1142}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1143
1144module_init(init_ceph);
1145module_exit(exit_ceph);
1146
1147MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1148MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1149MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1150MODULE_DESCRIPTION("Ceph filesystem for Linux");
1151MODULE_LICENSE("GPL");