4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/statfs.h>
45 #include <linux/types.h>
46 #include <linux/version.h>
48 #include <linux/user_namespace.h>
49 #ifdef HAVE_UIDGID_HEADER
50 # include <linux/uidgid.h>
53 #include <lustre_ioctl.h>
54 #include <lustre_ha.h>
55 #include <lustre_dlm.h>
56 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <lustre_param.h>
59 #include <lustre_log.h>
60 #include <cl_object.h>
61 #include <obd_cksum.h>
62 #include "llite_internal.h"
64 struct kmem_cache *ll_file_data_slab;
66 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
67 static DEFINE_SPINLOCK(ll_sb_lock);
70 #define log2(n) ffz(~(n))
73 static struct ll_sb_info *ll_init_sbi(void)
75 struct ll_sb_info *sbi = NULL;
77 unsigned long lru_page_max;
87 spin_lock_init(&sbi->ll_lock);
88 mutex_init(&sbi->ll_lco.lco_lock);
89 spin_lock_init(&sbi->ll_pp_extent_lock);
90 spin_lock_init(&sbi->ll_process_lock);
91 sbi->ll_rw_stats_on = 0;
94 pages = si.totalram - si.totalhigh;
95 lru_page_max = pages / 2;
97 /* initialize ll_cache data */
98 sbi->ll_cache = cl_cache_init(lru_page_max);
99 if (sbi->ll_cache == NULL) {
100 OBD_FREE(sbi, sizeof(*sbi));
104 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
105 SBI_DEFAULT_READAHEAD_MAX);
106 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
107 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
108 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
109 INIT_LIST_HEAD(&sbi->ll_conn_chain);
110 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
112 ll_generate_random_uuid(uuid);
113 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
114 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
116 spin_lock(&ll_sb_lock);
117 list_add_tail(&sbi->ll_list, &ll_super_blocks);
118 spin_unlock(&ll_sb_lock);
120 sbi->ll_flags |= LL_SBI_VERBOSE;
121 #ifdef ENABLE_CHECKSUM
122 sbi->ll_flags |= LL_SBI_CHECKSUM;
125 #ifdef HAVE_LRU_RESIZE_SUPPORT
126 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
129 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
130 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
132 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
136 /* metadata statahead is enabled by default */
137 sbi->ll_sa_max = LL_SA_RPC_DEF;
138 atomic_set(&sbi->ll_sa_total, 0);
139 atomic_set(&sbi->ll_sa_wrong, 0);
140 atomic_set(&sbi->ll_sa_running, 0);
141 atomic_set(&sbi->ll_agl_total, 0);
142 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
145 sbi->ll_squash.rsi_uid = 0;
146 sbi->ll_squash.rsi_gid = 0;
147 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
148 init_rwsem(&sbi->ll_squash.rsi_sem);
153 static void ll_free_sbi(struct super_block *sb)
155 struct ll_sb_info *sbi = ll_s2sbi(sb);
159 spin_lock(&ll_sb_lock);
160 list_del(&sbi->ll_list);
161 spin_unlock(&ll_sb_lock);
162 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
163 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
164 if (sbi->ll_cache != NULL) {
165 cl_cache_decref(sbi->ll_cache);
166 sbi->ll_cache = NULL;
168 OBD_FREE(sbi, sizeof(*sbi));
173 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
174 struct vfsmount *mnt)
176 struct inode *root = NULL;
177 struct ll_sb_info *sbi = ll_s2sbi(sb);
178 struct obd_device *obd;
179 struct obd_statfs *osfs = NULL;
180 struct ptlrpc_request *request = NULL;
181 struct obd_connect_data *data = NULL;
182 struct obd_uuid *uuid;
183 struct md_op_data *op_data;
184 struct lustre_md lmd;
186 int size, err, checksum;
189 obd = class_name2obd(md);
191 CERROR("MD %s: not setup or attached\n", md);
205 /* indicate the features supported by this client */
206 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
207 OBD_CONNECT_ATTRFID |
208 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
209 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
210 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
211 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
212 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
213 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
214 OBD_CONNECT_EINPROGRESS |
215 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
216 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
217 OBD_CONNECT_MAX_EASIZE |
218 OBD_CONNECT_FLOCK_DEAD |
219 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
220 OBD_CONNECT_OPEN_BY_FID |
221 OBD_CONNECT_DIR_STRIPE |
222 OBD_CONNECT_BULK_MBITS;
224 #ifdef HAVE_LRU_RESIZE_SUPPORT
225 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
226 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
228 #ifdef CONFIG_FS_POSIX_ACL
229 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
232 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
233 /* flag mdc connection as lightweight, only used for test
234 * purpose, use with care */
235 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
237 data->ocd_ibits_known = MDS_INODELOCK_FULL;
238 data->ocd_version = LUSTRE_VERSION_CODE;
240 if (sb->s_flags & MS_RDONLY)
241 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
242 if (sbi->ll_flags & LL_SBI_USER_XATTR)
243 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
245 #ifdef HAVE_MS_FLOCK_LOCK
246 /* force vfs to use lustre handler for flock() calls - bug 10743 */
247 sb->s_flags |= MS_FLOCK_LOCK;
249 #ifdef MS_HAS_NEW_AOPS
250 sb->s_flags |= MS_HAS_NEW_AOPS;
253 if (sbi->ll_flags & LL_SBI_FLOCK)
254 sbi->ll_fop = &ll_file_operations_flock;
255 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
256 sbi->ll_fop = &ll_file_operations;
258 sbi->ll_fop = &ll_file_operations_noflock;
261 data->ocd_connect_flags |= OBD_CONNECT_REAL;
262 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
263 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
265 data->ocd_brw_size = MD_MAX_BRW_SIZE;
267 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
269 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
270 "recovery, of which this client is not a "
271 "part. Please wait for recovery to complete,"
272 " abort, or time out.\n", md);
275 CERROR("cannot connect to %s: rc = %d\n", md, err);
279 sbi->ll_md_exp->exp_connect_data = *data;
281 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
282 LUSTRE_SEQ_METADATA);
284 CERROR("%s: Can't init metadata layer FID infrastructure, "
285 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
289 /* For mount, we only need fs info from MDT0, and also in DNE, it
290 * can make sure the client can be mounted as long as MDT0 is
292 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
293 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
294 OBD_STATFS_FOR_MDT0);
296 GOTO(out_md_fid, err);
298 /* This needs to be after statfs to ensure connect has finished.
299 * Note that "data" does NOT contain the valid connect reply.
300 * If connecting to a 1.8 server there will be no LMV device, so
301 * we can access the MDC export directly and exp_connect_flags will
302 * be non-zero, but if accessing an upgraded 2.1 server it will
303 * have the correct flags filled in.
304 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
305 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
306 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
307 valid != CLIENT_CONNECT_MDT_REQD) {
310 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
311 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
312 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
313 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
314 "feature(s) needed for correct operation "
315 "of this client (%s). Please upgrade "
316 "server or downgrade client.\n",
317 sbi->ll_md_exp->exp_obd->obd_name, buf);
318 OBD_FREE(buf, PAGE_CACHE_SIZE);
319 GOTO(out_md_fid, err = -EPROTO);
322 size = sizeof(*data);
323 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
324 KEY_CONN_DATA, &size, data);
326 CERROR("%s: Get connect data failed: rc = %d\n",
327 sbi->ll_md_exp->exp_obd->obd_name, err);
328 GOTO(out_md_fid, err);
331 LASSERT(osfs->os_bsize);
332 sb->s_blocksize = osfs->os_bsize;
333 sb->s_blocksize_bits = log2(osfs->os_bsize);
334 sb->s_magic = LL_SUPER_MAGIC;
335 sb->s_maxbytes = MAX_LFS_FILESIZE;
336 sbi->ll_namelen = osfs->os_namelen;
338 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
339 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
340 LCONSOLE_INFO("Disabling user_xattr feature because "
341 "it is not supported on the server\n");
342 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
345 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
347 sb->s_flags |= MS_POSIXACL;
349 sbi->ll_flags |= LL_SBI_ACL;
351 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
353 sb->s_flags &= ~MS_POSIXACL;
355 sbi->ll_flags &= ~LL_SBI_ACL;
358 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
359 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
360 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
361 LCONSOLE_INFO("client is set as remote by default.\n");
364 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
365 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
366 LCONSOLE_INFO("client claims to be remote, but server "
367 "rejected, forced to be local.\n");
371 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
372 sbi->ll_flags |= LL_SBI_64BIT_HASH;
374 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
375 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
377 sbi->ll_md_brw_pages = 1;
379 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
380 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
382 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
383 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
384 LCONSOLE_INFO("%s: disabling xattr cache due to "
385 "unknown maximum xattr size.\n", dt);
387 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
388 sbi->ll_xattr_cache_enabled = 1;
392 obd = class_name2obd(dt);
394 CERROR("DT %s: not setup or attached\n", dt);
395 GOTO(out_md_fid, err = -ENODEV);
398 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
399 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
400 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
401 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
402 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
403 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
404 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
405 OBD_CONNECT_MAXBYTES |
406 OBD_CONNECT_EINPROGRESS |
407 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
408 OBD_CONNECT_LAYOUTLOCK |
409 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
410 OBD_CONNECT_BULK_MBITS;
412 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
413 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
414 * disabled by default, because it can still be enabled on the
415 * fly via /proc. As a consequence, we still need to come to an
416 * agreement on the supported algorithms at connect time */
417 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
419 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
420 data->ocd_cksum_types = OBD_CKSUM_ADLER;
422 data->ocd_cksum_types = cksum_types_supported_client();
425 #ifdef HAVE_LRU_RESIZE_SUPPORT
426 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
428 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
429 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
431 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
432 "ocd_grant: %d\n", data->ocd_connect_flags,
433 data->ocd_version, data->ocd_grant);
435 obd->obd_upcall.onu_owner = &sbi->ll_lco;
436 obd->obd_upcall.onu_upcall = cl_ocd_update;
438 data->ocd_brw_size = DT_MAX_BRW_SIZE;
440 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
443 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
444 "recovery, of which this client is not a "
445 "part. Please wait for recovery to "
446 "complete, abort, or time out.\n", dt);
449 CERROR("%s: Cannot connect to %s: rc = %d\n",
450 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
454 sbi->ll_dt_exp->exp_connect_data = *data;
456 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
457 LUSTRE_SEQ_METADATA);
459 CERROR("%s: Can't init data layer FID infrastructure, "
460 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
464 mutex_lock(&sbi->ll_lco.lco_lock);
465 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
466 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
467 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
468 mutex_unlock(&sbi->ll_lco.lco_lock);
470 fid_zero(&sbi->ll_root_fid);
471 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
473 CERROR("cannot mds_connect: rc = %d\n", err);
474 GOTO(out_lock_cn_cb, err);
476 if (!fid_is_sane(&sbi->ll_root_fid)) {
477 CERROR("%s: Invalid root fid "DFID" during mount\n",
478 sbi->ll_md_exp->exp_obd->obd_name,
479 PFID(&sbi->ll_root_fid));
480 GOTO(out_lock_cn_cb, err = -EINVAL);
482 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
484 sb->s_op = &lustre_super_operations;
485 #if THREAD_SIZE >= 8192 /*b=17630*/
486 sb->s_export_op = &lustre_export_operations;
490 * XXX: move this to after cbd setup? */
491 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
492 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
493 valid |= OBD_MD_FLRMTPERM;
494 else if (sbi->ll_flags & LL_SBI_ACL)
495 valid |= OBD_MD_FLACL;
497 OBD_ALLOC_PTR(op_data);
499 GOTO(out_lock_cn_cb, err = -ENOMEM);
501 op_data->op_fid1 = sbi->ll_root_fid;
502 op_data->op_mode = 0;
503 op_data->op_valid = valid;
505 err = md_getattr(sbi->ll_md_exp, op_data, &request);
507 OBD_FREE_PTR(op_data);
509 CERROR("%s: md_getattr failed for root: rc = %d\n",
510 sbi->ll_md_exp->exp_obd->obd_name, err);
511 GOTO(out_lock_cn_cb, err);
514 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
515 sbi->ll_md_exp, &lmd);
517 CERROR("failed to understand root inode md: rc = %d\n", err);
518 ptlrpc_req_finished(request);
519 GOTO(out_lock_cn_cb, err);
522 LASSERT(fid_is_sane(&sbi->ll_root_fid));
523 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
524 sbi->ll_flags & LL_SBI_32BIT_API),
526 md_free_lustre_md(sbi->ll_md_exp, &lmd);
527 ptlrpc_req_finished(request);
530 #ifdef CONFIG_FS_POSIX_ACL
532 posix_acl_release(lmd.posix_acl);
533 lmd.posix_acl = NULL;
536 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
538 CERROR("lustre_lite: bad iget4 for root\n");
542 #ifdef CONFIG_FS_POSIX_ACL
543 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
544 rct_init(&sbi->ll_rct);
545 et_init(&sbi->ll_et);
549 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
550 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
551 KEY_CHECKSUM, sizeof(checksum), &checksum,
555 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
556 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
557 sbi->ll_cache, NULL);
559 sb->s_root = d_make_root(root);
560 if (sb->s_root == NULL) {
561 CERROR("%s: can't make root dentry\n",
562 ll_get_fsname(sb, NULL, 0));
563 GOTO(out_root, err = -ENOMEM);
565 #ifdef HAVE_DCACHE_LOCK
566 sb->s_root->d_op = &ll_d_ops;
569 sbi->ll_sdev_orig = sb->s_dev;
571 /* We set sb->s_dev equal on all lustre clients in order to support
572 * NFS export clustering. NFSD requires that the FSID be the same
574 /* s_dev is also used in lt_compare() to compare two fs, but that is
575 * only a node-local comparison. */
576 uuid = obd_get_uuid(sbi->ll_md_exp);
578 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
584 if (proc_lustre_fs_root != NULL) {
585 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
588 CERROR("%s: could not register mount in lprocfs: "
589 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
599 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
601 obd_disconnect(sbi->ll_dt_exp);
602 sbi->ll_dt_exp = NULL;
604 obd_fid_fini(sbi->ll_md_exp->exp_obd);
606 obd_disconnect(sbi->ll_md_exp);
607 sbi->ll_md_exp = NULL;
616 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
620 size = sizeof(*lmmsize);
621 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
622 KEY_MAX_EASIZE, &size, lmmsize);
624 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
625 sbi->ll_dt_exp->exp_obd->obd_name, rc);
630 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
631 KEY_MAX_EASIZE, &size, lmmsize);
633 CERROR("Get max mdsize error rc %d\n", rc);
639 * Get the value of the default_easize parameter.
641 * \see client_obd::cl_default_mds_easize
643 * \param[in] sbi superblock info for this filesystem
644 * \param[out] lmmsize pointer to storage location for value
646 * \retval 0 on success
647 * \retval negative negated errno on failure
649 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
654 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
655 KEY_DEFAULT_EASIZE, &size, lmmsize);
657 CERROR("Get default mdsize error rc %d\n", rc);
663 * Set the default_easize parameter to the given value.
665 * \see client_obd::cl_default_mds_easize
667 * \param[in] sbi superblock info for this filesystem
668 * \param[in] lmmsize the size to set
670 * \retval 0 on success
671 * \retval negative negated errno on failure
673 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
677 if (lmmsize < sizeof(struct lov_mds_md) ||
678 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
681 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
682 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
683 sizeof(int), &lmmsize, NULL);
688 static void ll_dump_inode(struct inode *inode)
690 struct ll_d_hlist_node *tmp;
691 int dentry_count = 0;
693 LASSERT(inode != NULL);
695 ll_d_hlist_for_each(tmp, &inode->i_dentry)
698 CERROR("%s: inode %p dump: dev=%s fid="DFID
699 " mode=%o count=%u, %d dentries\n",
700 ll_get_fsname(inode->i_sb, NULL, 0), inode,
701 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
702 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
705 void lustre_dump_dentry(struct dentry *dentry, int recur)
707 struct list_head *tmp;
710 LASSERT(dentry != NULL);
712 list_for_each(tmp, &dentry->d_subdirs)
715 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
716 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
717 dentry->d_name.len, dentry->d_name.name,
718 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
719 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
720 dentry->d_flags, dentry->d_fsdata, subdirs);
721 if (dentry->d_inode != NULL)
722 ll_dump_inode(dentry->d_inode);
727 list_for_each(tmp, &dentry->d_subdirs) {
728 struct dentry *d = list_entry(tmp, struct dentry, d_child);
729 lustre_dump_dentry(d, recur - 1);
733 static void client_common_put_super(struct super_block *sb)
735 struct ll_sb_info *sbi = ll_s2sbi(sb);
738 #ifdef CONFIG_FS_POSIX_ACL
739 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
740 et_fini(&sbi->ll_et);
741 rct_fini(&sbi->ll_rct);
747 list_del(&sbi->ll_conn_chain);
749 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
750 obd_disconnect(sbi->ll_dt_exp);
751 sbi->ll_dt_exp = NULL;
753 lprocfs_unregister_mountpoint(sbi);
755 obd_fid_fini(sbi->ll_md_exp->exp_obd);
756 obd_disconnect(sbi->ll_md_exp);
757 sbi->ll_md_exp = NULL;
762 void ll_kill_super(struct super_block *sb)
764 struct ll_sb_info *sbi;
768 if (!(sb->s_flags & MS_ACTIVE))
772 /* we need restore s_dev from changed for clustred NFS before put_super
773 * because new kernels have cached s_dev and change sb->s_dev in
774 * put_super not affected real removing devices */
776 sb->s_dev = sbi->ll_sdev_orig;
777 sbi->ll_umounting = 1;
779 /* wait running statahead threads to quit */
780 while (atomic_read(&sbi->ll_sa_running) > 0) {
781 set_current_state(TASK_UNINTERRUPTIBLE);
782 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
789 static inline int ll_set_opt(const char *opt, char *data, int fl)
791 if (strncmp(opt, data, strlen(opt)) != 0)
797 /* non-client-specific mount options are parsed in lmd_parse */
798 static int ll_options(char *options, int *flags)
801 char *s1 = options, *s2;
807 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
810 CDEBUG(D_SUPER, "next opt=%s\n", s1);
811 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
816 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
821 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
826 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
831 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
836 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
841 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
846 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
851 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
857 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
862 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
867 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
872 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
877 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
882 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
887 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
892 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
897 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
902 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
908 s2 = strchr(s1, ',');
916 void ll_lli_init(struct ll_inode_info *lli)
918 lli->lli_inode_magic = LLI_INODE_MAGIC;
920 spin_lock_init(&lli->lli_lock);
921 lli->lli_posix_acl = NULL;
922 lli->lli_remote_perms = NULL;
923 mutex_init(&lli->lli_rmtperm_mutex);
924 /* Do not set lli_fid, it has been initialized already. */
925 fid_zero(&lli->lli_pfid);
926 atomic_set(&lli->lli_open_count, 0);
927 lli->lli_rmtperm_time = 0;
928 lli->lli_mds_read_och = NULL;
929 lli->lli_mds_write_och = NULL;
930 lli->lli_mds_exec_och = NULL;
931 lli->lli_open_fd_read_count = 0;
932 lli->lli_open_fd_write_count = 0;
933 lli->lli_open_fd_exec_count = 0;
934 mutex_init(&lli->lli_och_mutex);
935 spin_lock_init(&lli->lli_agl_lock);
936 spin_lock_init(&lli->lli_layout_lock);
937 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
938 lli->lli_clob = NULL;
940 init_rwsem(&lli->lli_xattrs_list_rwsem);
941 mutex_init(&lli->lli_xattrs_enq_lock);
943 LASSERT(lli->lli_vfs_inode.i_mode != 0);
944 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
945 mutex_init(&lli->lli_readdir_mutex);
946 lli->lli_opendir_key = NULL;
948 spin_lock_init(&lli->lli_sa_lock);
949 lli->lli_opendir_pid = 0;
950 lli->lli_sa_enabled = 0;
951 lli->lli_def_stripe_offset = -1;
953 mutex_init(&lli->lli_size_mutex);
954 lli->lli_symlink_name = NULL;
955 init_rwsem(&lli->lli_trunc_sem);
956 range_lock_tree_init(&lli->lli_write_tree);
957 init_rwsem(&lli->lli_glimpse_sem);
958 lli->lli_glimpse_time = 0;
959 INIT_LIST_HEAD(&lli->lli_agl_list);
960 lli->lli_agl_index = 0;
961 lli->lli_async_rc = 0;
963 mutex_init(&lli->lli_layout_mutex);
966 static inline int ll_bdi_register(struct backing_dev_info *bdi)
968 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
970 bdi->name = "lustre";
971 return bdi_register(bdi, NULL, "lustre-%d",
972 atomic_inc_return(&ll_bdi_num));
975 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
977 struct lustre_profile *lprof = NULL;
978 struct lustre_sb_info *lsi = s2lsi(sb);
979 struct ll_sb_info *sbi;
980 char *dt = NULL, *md = NULL;
981 char *profilenm = get_profile_name(sb);
982 struct config_llog_instance *cfg;
983 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
984 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
988 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
994 try_module_get(THIS_MODULE);
996 /* client additional sb info */
997 lsi->lsi_llsbi = sbi = ll_init_sbi();
999 module_put(THIS_MODULE);
1004 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1006 GOTO(out_free, err);
1008 err = bdi_init(&lsi->lsi_bdi);
1010 GOTO(out_free, err);
1011 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1012 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1013 err = ll_bdi_register(&lsi->lsi_bdi);
1015 GOTO(out_free, err);
1017 sb->s_bdi = &lsi->lsi_bdi;
1018 #ifndef HAVE_DCACHE_LOCK
1019 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1020 sb->s_d_op = &ll_d_ops;
1023 /* Generate a string unique to this super, in case some joker tries
1024 to mount the same fs at two mount points.
1025 Use the address of the super itself.*/
1026 cfg->cfg_instance = sb;
1027 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1028 cfg->cfg_callback = class_config_llog_handler;
1029 /* set up client obds */
1030 err = lustre_process_log(sb, profilenm, cfg);
1032 GOTO(out_free, err);
1034 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1035 lprof = class_get_profile(profilenm);
1036 if (lprof == NULL) {
1037 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1038 " read from the MGS. Does that filesystem "
1039 "exist?\n", profilenm);
1040 GOTO(out_free, err = -EINVAL);
1042 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1043 lprof->lp_md, lprof->lp_dt);
1045 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1047 GOTO(out_free, err = -ENOMEM);
1048 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1050 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1052 GOTO(out_free, err = -ENOMEM);
1053 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1055 /* connections, registrations, sb setup */
1056 err = client_common_fill_super(sb, md, dt, mnt);
1058 GOTO(out_free, err);
1060 sbi->ll_client_common_fill_super_succeeded = 1;
1064 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1066 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1069 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1070 LCONSOLE_WARN("Mounted %s\n", profilenm);
1074 } /* ll_fill_super */
1076 void ll_put_super(struct super_block *sb)
1078 struct config_llog_instance cfg, params_cfg;
1079 struct obd_device *obd;
1080 struct lustre_sb_info *lsi = s2lsi(sb);
1081 struct ll_sb_info *sbi = ll_s2sbi(sb);
1082 char *profilenm = get_profile_name(sb);
1084 int next, force = 1, rc = 0;
1087 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1089 cfg.cfg_instance = sb;
1090 lustre_end_log(sb, profilenm, &cfg);
1092 params_cfg.cfg_instance = sb;
1093 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1095 if (sbi->ll_md_exp) {
1096 obd = class_exp2obd(sbi->ll_md_exp);
1098 force = obd->obd_force;
1101 /* Wait for unstable pages to be committed to stable storage */
1103 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1104 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1105 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1109 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1110 if (force == 0 && rc != -EINTR)
1111 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1114 /* We need to set force before the lov_disconnect in
1115 lustre_common_put_super, since l_d cleans up osc's as well. */
1118 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1120 obd->obd_force = force;
1124 if (sbi->ll_client_common_fill_super_succeeded) {
1125 /* Only if client_common_fill_super succeeded */
1126 client_common_put_super(sb);
1130 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1131 class_manual_cleanup(obd);
1134 if (sbi->ll_flags & LL_SBI_VERBOSE)
1135 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1138 class_del_profile(profilenm);
1140 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1141 bdi_destroy(&lsi->lsi_bdi);
1142 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1146 lsi->lsi_llsbi = NULL;
1148 lustre_common_put_super(sb);
1150 cl_env_cache_purge(~0);
1152 module_put(THIS_MODULE);
1155 } /* client_put_super */
1157 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1159 struct inode *inode = NULL;
1161 /* NOTE: we depend on atomic igrab() -bzzz */
1162 lock_res_and_lock(lock);
1163 if (lock->l_resource->lr_lvb_inode) {
1164 struct ll_inode_info * lli;
1165 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1166 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1167 inode = igrab(lock->l_resource->lr_lvb_inode);
1169 inode = lock->l_resource->lr_lvb_inode;
1170 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1171 D_WARNING, lock, "lr_lvb_inode %p is "
1172 "bogus: magic %08x",
1173 lock->l_resource->lr_lvb_inode,
1174 lli->lli_inode_magic);
1178 unlock_res_and_lock(lock);
1182 static void ll_dir_clear_lsm_md(struct inode *inode)
1184 struct ll_inode_info *lli = ll_i2info(inode);
1186 LASSERT(S_ISDIR(inode->i_mode));
1188 if (lli->lli_lsm_md != NULL) {
1189 lmv_free_memmd(lli->lli_lsm_md);
1190 lli->lli_lsm_md = NULL;
1194 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1195 const struct lu_fid *fid,
1196 struct lustre_md *md)
1198 struct ll_sb_info *sbi = ll_s2sbi(sb);
1199 struct mdt_body *body = md->body;
1200 struct inode *inode;
1204 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1205 inode = iget_locked(sb, ino);
1206 if (inode == NULL) {
1207 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1208 ll_get_fsname(sb, NULL, 0), PFID(fid));
1209 RETURN(ERR_PTR(-ENOENT));
1212 if (inode->i_state & I_NEW) {
1213 struct ll_inode_info *lli = ll_i2info(inode);
1214 struct lmv_stripe_md *lsm = md->lmv;
1216 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1217 (body->mbo_mode & S_IFMT);
1218 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1221 LTIME_S(inode->i_mtime) = 0;
1222 LTIME_S(inode->i_atime) = 0;
1223 LTIME_S(inode->i_ctime) = 0;
1226 /* initializing backing dev info. */
1227 inode->i_mapping->backing_dev_info =
1228 &s2lsi(inode->i_sb)->lsi_bdi;
1229 inode->i_op = &ll_dir_inode_operations;
1230 inode->i_fop = &ll_dir_operations;
1231 lli->lli_fid = *fid;
1234 LASSERT(lsm != NULL);
1235 /* master object FID */
1236 lli->lli_pfid = body->mbo_fid1;
1237 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1238 lli, PFID(fid), PFID(&lli->lli_pfid));
1239 unlock_new_inode(inode);
1245 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1248 struct lmv_stripe_md *lsm = md->lmv;
1251 LASSERT(lsm != NULL);
1252 /* XXX sigh, this lsm_root initialization should be in
1253 * LMV layer, but it needs ll_iget right now, so we
1254 * put this here right now. */
1255 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1256 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1257 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1258 /* Unfortunately ll_iget will call ll_update_inode,
1259 * where the initialization of slave inode is slightly
1260 * different, so it reset lsm_md to NULL to avoid
1261 * initializing lsm for slave inode. */
1262 /* For migrating inode, master stripe and master object will
1263 * be same, so we only need assign this inode */
1264 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1265 lsm->lsm_md_oinfo[i].lmo_root = inode;
1267 lsm->lsm_md_oinfo[i].lmo_root =
1268 ll_iget_anon_dir(inode->i_sb, fid, md);
1270 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1271 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1273 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1281 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1282 const struct lmv_stripe_md *lsm_md2)
1284 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1285 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1286 lsm_md1->lsm_md_master_mdt_index ==
1287 lsm_md2->lsm_md_master_mdt_index &&
1288 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1289 lsm_md1->lsm_md_layout_version ==
1290 lsm_md2->lsm_md_layout_version &&
1291 strcmp(lsm_md1->lsm_md_pool_name,
1292 lsm_md2->lsm_md_pool_name) == 0;
1295 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1297 struct ll_inode_info *lli = ll_i2info(inode);
1298 struct lmv_stripe_md *lsm = md->lmv;
1302 LASSERT(S_ISDIR(inode->i_mode));
1303 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1304 PFID(ll_inode2fid(inode)));
1306 /* no striped information from request. */
1308 if (lli->lli_lsm_md == NULL) {
1310 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1311 LMV_HASH_FLAG_MIGRATION) {
1312 /* migration is done, the temporay MIGRATE layout has
1314 CDEBUG(D_INODE, DFID" finish migration.\n",
1315 PFID(ll_inode2fid(inode)));
1316 lmv_free_memmd(lli->lli_lsm_md);
1317 lli->lli_lsm_md = NULL;
1320 /* The lustre_md from req does not include stripeEA,
1321 * see ll_md_setattr */
1326 /* set the directory layout */
1327 if (lli->lli_lsm_md == NULL) {
1329 rc = ll_init_lsm_md(inode, md);
1333 lli->lli_lsm_md = lsm;
1334 /* set lsm_md to NULL, so the following free lustre_md
1335 * will not free this lsm */
1337 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1338 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1342 /* Compare the old and new stripe information */
1343 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1344 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1347 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1348 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1349 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1350 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1351 inode, lsm, old_lsm,
1352 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1353 lsm->lsm_md_stripe_count,
1354 old_lsm->lsm_md_stripe_count,
1355 lsm->lsm_md_master_mdt_index,
1356 old_lsm->lsm_md_master_mdt_index,
1357 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1358 lsm->lsm_md_layout_version,
1359 old_lsm->lsm_md_layout_version,
1360 lsm->lsm_md_pool_name,
1361 old_lsm->lsm_md_pool_name);
1363 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1364 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1365 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1366 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1369 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1370 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1371 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1372 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1381 void ll_clear_inode(struct inode *inode)
1383 struct ll_inode_info *lli = ll_i2info(inode);
1384 struct ll_sb_info *sbi = ll_i2sbi(inode);
1387 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1388 PFID(ll_inode2fid(inode)), inode);
1390 if (S_ISDIR(inode->i_mode)) {
1391 /* these should have been cleared in ll_file_release */
1392 LASSERT(lli->lli_opendir_key == NULL);
1393 LASSERT(lli->lli_sai == NULL);
1394 LASSERT(lli->lli_opendir_pid == 0);
1397 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1399 LASSERT(!lli->lli_open_fd_write_count);
1400 LASSERT(!lli->lli_open_fd_read_count);
1401 LASSERT(!lli->lli_open_fd_exec_count);
1403 if (lli->lli_mds_write_och)
1404 ll_md_real_close(inode, FMODE_WRITE);
1405 if (lli->lli_mds_exec_och)
1406 ll_md_real_close(inode, FMODE_EXEC);
1407 if (lli->lli_mds_read_och)
1408 ll_md_real_close(inode, FMODE_READ);
1410 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1411 OBD_FREE(lli->lli_symlink_name,
1412 strlen(lli->lli_symlink_name) + 1);
1413 lli->lli_symlink_name = NULL;
1416 ll_xattr_cache_destroy(inode);
1418 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1419 LASSERT(lli->lli_posix_acl == NULL);
1420 if (lli->lli_remote_perms) {
1421 free_rmtperm_hash(lli->lli_remote_perms);
1422 lli->lli_remote_perms = NULL;
1425 #ifdef CONFIG_FS_POSIX_ACL
1426 else if (lli->lli_posix_acl) {
1427 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1428 LASSERT(lli->lli_remote_perms == NULL);
1429 posix_acl_release(lli->lli_posix_acl);
1430 lli->lli_posix_acl = NULL;
1433 lli->lli_inode_magic = LLI_INODE_DEAD;
1435 if (S_ISDIR(inode->i_mode))
1436 ll_dir_clear_lsm_md(inode);
1437 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1438 LASSERT(list_empty(&lli->lli_agl_list));
1441 * XXX This has to be done before lsm is freed below, because
1442 * cl_object still uses inode lsm.
1444 cl_inode_fini(inode);
1449 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1451 struct lustre_md md;
1452 struct inode *inode = dentry->d_inode;
1453 struct ll_sb_info *sbi = ll_i2sbi(inode);
1454 struct ptlrpc_request *request = NULL;
1458 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1459 LUSTRE_OPC_ANY, NULL);
1460 if (IS_ERR(op_data))
1461 RETURN(PTR_ERR(op_data));
1463 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1465 ptlrpc_req_finished(request);
1466 if (rc == -ENOENT) {
1468 /* Unlinked special device node? Or just a race?
1469 * Pretend we done everything. */
1470 if (!S_ISREG(inode->i_mode) &&
1471 !S_ISDIR(inode->i_mode)) {
1472 ia_valid = op_data->op_attr.ia_valid;
1473 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1474 rc = simple_setattr(dentry, &op_data->op_attr);
1475 op_data->op_attr.ia_valid = ia_valid;
1477 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1478 CERROR("md_setattr fails: rc = %d\n", rc);
1483 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1484 sbi->ll_md_exp, &md);
1486 ptlrpc_req_finished(request);
1490 ia_valid = op_data->op_attr.ia_valid;
1491 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1492 * cache is not cleared yet. */
1493 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1494 if (S_ISREG(inode->i_mode))
1495 mutex_lock(&inode->i_mutex);
1496 rc = simple_setattr(dentry, &op_data->op_attr);
1497 if (S_ISREG(inode->i_mode))
1498 mutex_unlock(&inode->i_mutex);
1499 op_data->op_attr.ia_valid = ia_valid;
1501 rc = ll_update_inode(inode, &md);
1502 ptlrpc_req_finished(request);
1507 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1508 * object(s) determine the file size and mtime. Otherwise, the MDS will
1509 * keep these values until such a time that objects are allocated for it.
1510 * We do the MDS operations first, as it is checking permissions for us.
1511 * We don't to the MDS RPC if there is nothing that we want to store there,
1512 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1513 * going to do an RPC anyways.
1515 * If we are doing a truncate, we will send the mtime and ctime updates
1516 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1517 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1520 * In case of HSMimport, we only set attr on MDS.
1522 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1524 struct inode *inode = dentry->d_inode;
1525 struct ll_inode_info *lli = ll_i2info(inode);
1526 struct md_op_data *op_data = NULL;
1530 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1531 "valid %x, hsm_import %d\n",
1532 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1533 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1536 if (attr->ia_valid & ATTR_SIZE) {
1537 /* Check new size against VFS/VM file size limit and rlimit */
1538 rc = inode_newsize_ok(inode, attr->ia_size);
1542 /* The maximum Lustre file size is variable, based on the
1543 * OST maximum object size and number of stripes. This
1544 * needs another check in addition to the VFS check above. */
1545 if (attr->ia_size > ll_file_maxbytes(inode)) {
1546 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1547 PFID(&lli->lli_fid), attr->ia_size,
1548 ll_file_maxbytes(inode));
1552 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1555 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1556 if (attr->ia_valid & TIMES_SET_FLAGS) {
1557 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1558 !cfs_capable(CFS_CAP_FOWNER))
1562 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1563 if (attr->ia_valid & ATTR_CTIME) {
1564 attr->ia_ctime = CFS_CURRENT_TIME;
1565 attr->ia_valid |= ATTR_CTIME_SET;
1567 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1568 (attr->ia_valid & ATTR_ATIME)) {
1569 attr->ia_atime = CFS_CURRENT_TIME;
1570 attr->ia_valid |= ATTR_ATIME_SET;
1572 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1573 (attr->ia_valid & ATTR_MTIME)) {
1574 attr->ia_mtime = CFS_CURRENT_TIME;
1575 attr->ia_valid |= ATTR_MTIME_SET;
1578 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1579 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1580 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1581 cfs_time_current_sec());
1583 if (S_ISREG(inode->i_mode)) {
1584 if (attr->ia_valid & ATTR_SIZE)
1585 inode_dio_write_done(inode);
1586 mutex_unlock(&inode->i_mutex);
1589 /* We always do an MDS RPC, even if we're only changing the size;
1590 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1592 OBD_ALLOC_PTR(op_data);
1593 if (op_data == NULL)
1594 GOTO(out, rc = -ENOMEM);
1596 op_data->op_attr = *attr;
1598 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1599 /* If we are changing file size, file content is
1600 * modified, flag it. */
1601 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1602 op_data->op_bias |= MDS_DATA_MODIFIED;
1605 rc = ll_md_setattr(dentry, op_data);
1609 if (!S_ISREG(inode->i_mode) || hsm_import)
1612 if (attr->ia_valid & (ATTR_SIZE |
1613 ATTR_ATIME | ATTR_ATIME_SET |
1614 ATTR_MTIME | ATTR_MTIME_SET)) {
1615 /* For truncate and utimes sending attributes to OSTs, setting
1616 * mtime/atime to the past will be performed under PW [0:EOF]
1617 * extent lock (new_size:EOF for truncate). It may seem
1618 * excessive to send mtime/atime updates to OSTs when not
1619 * setting times to past, but it is necessary due to possible
1620 * time de-synchronization between MDT inode and OST objects */
1621 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1624 /* If the file was restored, it needs to set dirty flag.
1626 * We've already sent MDS_DATA_MODIFIED flag in
1627 * ll_md_setattr() for truncate. However, the MDT refuses to
1628 * set the HS_DIRTY flag on released files, so we have to set
1629 * it again if the file has been restored. Please check how
1630 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1632 * Please notice that if the file is not released, the previous
1633 * MDS_DATA_MODIFIED has taken effect and usually
1634 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1635 * This way we can save an RPC for common open + trunc
1637 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1638 struct hsm_state_set hss = {
1639 .hss_valid = HSS_SETMASK,
1640 .hss_setmask = HS_DIRTY,
1644 rc2 = ll_hsm_state_set(inode, &hss);
1646 CERROR(DFID "HSM set dirty failed: rc2 = %d\n",
1647 PFID(ll_inode2fid(inode)), rc2);
1652 if (op_data != NULL)
1653 ll_finish_md_op_data(op_data);
1655 if (S_ISREG(inode->i_mode)) {
1656 mutex_lock(&inode->i_mutex);
1657 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1658 inode_dio_wait(inode);
1661 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1662 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1667 int ll_setattr(struct dentry *de, struct iattr *attr)
1669 int mode = de->d_inode->i_mode;
1671 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1672 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1673 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1675 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1676 (ATTR_SIZE|ATTR_MODE)) &&
1677 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1678 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1679 !(attr->ia_mode & S_ISGID))))
1680 attr->ia_valid |= ATTR_FORCE;
1682 if ((attr->ia_valid & ATTR_MODE) &&
1684 !(attr->ia_mode & S_ISUID) &&
1685 !(attr->ia_valid & ATTR_KILL_SUID))
1686 attr->ia_valid |= ATTR_KILL_SUID;
1688 if ((attr->ia_valid & ATTR_MODE) &&
1689 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1690 !(attr->ia_mode & S_ISGID) &&
1691 !(attr->ia_valid & ATTR_KILL_SGID))
1692 attr->ia_valid |= ATTR_KILL_SGID;
1694 return ll_setattr_raw(de, attr, false);
1697 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1698 __u64 max_age, __u32 flags)
1700 struct ll_sb_info *sbi = ll_s2sbi(sb);
1701 struct obd_statfs obd_osfs;
1705 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1707 CERROR("md_statfs fails: rc = %d\n", rc);
1711 osfs->os_type = sb->s_magic;
1713 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1714 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1716 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1717 flags |= OBD_STATFS_NODELAY;
1719 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1721 CERROR("obd_statfs fails: rc = %d\n", rc);
1725 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1726 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1729 osfs->os_bsize = obd_osfs.os_bsize;
1730 osfs->os_blocks = obd_osfs.os_blocks;
1731 osfs->os_bfree = obd_osfs.os_bfree;
1732 osfs->os_bavail = obd_osfs.os_bavail;
1734 /* If we don't have as many objects free on the OST as inodes
1735 * on the MDS, we reduce the total number of inodes to
1736 * compensate, so that the "inodes in use" number is correct.
1738 if (obd_osfs.os_ffree < osfs->os_ffree) {
1739 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1741 osfs->os_ffree = obd_osfs.os_ffree;
1746 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1748 struct super_block *sb = de->d_sb;
1749 struct obd_statfs osfs;
1750 __u64 fsid = huge_encode_dev(sb->s_dev);
1753 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1754 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1756 /* Some amount of caching on the client is allowed */
1757 rc = ll_statfs_internal(sb, &osfs,
1758 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1763 statfs_unpack(sfs, &osfs);
1765 /* We need to downshift for all 32-bit kernels, because we can't
1766 * tell if the kernel is being called via sys_statfs64() or not.
1767 * Stop before overflowing f_bsize - in which case it is better
1768 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1769 if (sizeof(long) < 8) {
1770 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1773 osfs.os_blocks >>= 1;
1774 osfs.os_bfree >>= 1;
1775 osfs.os_bavail >>= 1;
1779 sfs->f_blocks = osfs.os_blocks;
1780 sfs->f_bfree = osfs.os_bfree;
1781 sfs->f_bavail = osfs.os_bavail;
1782 sfs->f_fsid.val[0] = (__u32)fsid;
1783 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1787 void ll_inode_size_lock(struct inode *inode)
1789 struct ll_inode_info *lli;
1791 LASSERT(!S_ISDIR(inode->i_mode));
1793 lli = ll_i2info(inode);
1794 mutex_lock(&lli->lli_size_mutex);
1797 void ll_inode_size_unlock(struct inode *inode)
1799 struct ll_inode_info *lli;
1801 lli = ll_i2info(inode);
1802 mutex_unlock(&lli->lli_size_mutex);
1805 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1807 struct ll_inode_info *lli = ll_i2info(inode);
1808 struct mdt_body *body = md->body;
1809 struct ll_sb_info *sbi = ll_i2sbi(inode);
1811 if (body->mbo_valid & OBD_MD_FLEASIZE)
1812 cl_file_inode_init(inode, md);
1814 if (S_ISDIR(inode->i_mode)) {
1817 rc = ll_update_lsm_md(inode, md);
1822 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1823 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1824 ll_update_remote_perm(inode, md->remote_perm);
1826 #ifdef CONFIG_FS_POSIX_ACL
1827 else if (body->mbo_valid & OBD_MD_FLACL) {
1828 spin_lock(&lli->lli_lock);
1829 if (lli->lli_posix_acl)
1830 posix_acl_release(lli->lli_posix_acl);
1831 lli->lli_posix_acl = md->posix_acl;
1832 spin_unlock(&lli->lli_lock);
1835 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1836 sbi->ll_flags & LL_SBI_32BIT_API);
1837 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1839 if (body->mbo_valid & OBD_MD_FLATIME) {
1840 if (body->mbo_atime > LTIME_S(inode->i_atime))
1841 LTIME_S(inode->i_atime) = body->mbo_atime;
1842 lli->lli_atime = body->mbo_atime;
1845 if (body->mbo_valid & OBD_MD_FLMTIME) {
1846 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1847 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1848 "to "LPU64"\n", inode->i_ino,
1849 LTIME_S(inode->i_mtime), body->mbo_mtime);
1850 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1852 lli->lli_mtime = body->mbo_mtime;
1855 if (body->mbo_valid & OBD_MD_FLCTIME) {
1856 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1857 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1858 lli->lli_ctime = body->mbo_ctime;
1861 if (body->mbo_valid & OBD_MD_FLMODE)
1862 inode->i_mode = (inode->i_mode & S_IFMT) |
1863 (body->mbo_mode & ~S_IFMT);
1865 if (body->mbo_valid & OBD_MD_FLTYPE)
1866 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1867 (body->mbo_mode & S_IFMT);
1869 LASSERT(inode->i_mode != 0);
1870 if (S_ISREG(inode->i_mode))
1871 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1872 LL_MAX_BLKSIZE_BITS);
1874 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1876 if (body->mbo_valid & OBD_MD_FLUID)
1877 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1878 if (body->mbo_valid & OBD_MD_FLGID)
1879 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1880 if (body->mbo_valid & OBD_MD_FLFLAGS)
1881 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1882 if (body->mbo_valid & OBD_MD_FLNLINK)
1883 set_nlink(inode, body->mbo_nlink);
1884 if (body->mbo_valid & OBD_MD_FLRDEV)
1885 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1887 if (body->mbo_valid & OBD_MD_FLID) {
1888 /* FID shouldn't be changed! */
1889 if (fid_is_sane(&lli->lli_fid)) {
1890 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1891 "Trying to change FID "DFID
1892 " to the "DFID", inode "DFID"(%p)\n",
1893 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1894 PFID(ll_inode2fid(inode)), inode);
1896 lli->lli_fid = body->mbo_fid1;
1900 LASSERT(fid_seq(&lli->lli_fid) != 0);
1902 if (body->mbo_valid & OBD_MD_FLSIZE) {
1903 i_size_write(inode, body->mbo_size);
1905 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1906 PFID(ll_inode2fid(inode)),
1907 (unsigned long long)body->mbo_size);
1909 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1910 inode->i_blocks = body->mbo_blocks;
1913 if (body->mbo_valid & OBD_MD_TSTATE) {
1914 if (body->mbo_t_state & MS_RESTORE)
1915 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1921 int ll_read_inode2(struct inode *inode, void *opaque)
1923 struct lustre_md *md = opaque;
1924 struct ll_inode_info *lli = ll_i2info(inode);
1928 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1929 PFID(&lli->lli_fid), inode);
1931 /* Core attributes from the MDS first. This is a new inode, and
1932 * the VFS doesn't zero times in the core inode so we have to do
1933 * it ourselves. They will be overwritten by either MDS or OST
1934 * attributes - we just need to make sure they aren't newer. */
1935 LTIME_S(inode->i_mtime) = 0;
1936 LTIME_S(inode->i_atime) = 0;
1937 LTIME_S(inode->i_ctime) = 0;
1939 rc = ll_update_inode(inode, md);
1943 /* OIDEBUG(inode); */
1945 /* initializing backing dev info. */
1946 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1949 if (S_ISREG(inode->i_mode)) {
1950 struct ll_sb_info *sbi = ll_i2sbi(inode);
1951 inode->i_op = &ll_file_inode_operations;
1952 inode->i_fop = sbi->ll_fop;
1953 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1955 } else if (S_ISDIR(inode->i_mode)) {
1956 inode->i_op = &ll_dir_inode_operations;
1957 inode->i_fop = &ll_dir_operations;
1959 } else if (S_ISLNK(inode->i_mode)) {
1960 inode->i_op = &ll_fast_symlink_inode_operations;
1963 inode->i_op = &ll_special_inode_operations;
1965 init_special_inode(inode, inode->i_mode,
1974 void ll_delete_inode(struct inode *inode)
1976 struct ll_inode_info *lli = ll_i2info(inode);
1979 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1980 /* It is last chance to write out dirty pages,
1981 * otherwise we may lose data while umount */
1982 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
1984 truncate_inode_pages_final(&inode->i_data);
1986 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
1987 "see https://jira.hpdd.intel.com/browse/LU-118\n",
1988 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
1990 #ifdef HAVE_SBOPS_EVICT_INODE
1991 ll_clear_inode(inode);
1998 int ll_iocontrol(struct inode *inode, struct file *file,
1999 unsigned int cmd, unsigned long arg)
2001 struct ll_sb_info *sbi = ll_i2sbi(inode);
2002 struct ptlrpc_request *req = NULL;
2007 case FSFILT_IOC_GETFLAGS: {
2008 struct mdt_body *body;
2009 struct md_op_data *op_data;
2011 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2012 0, 0, LUSTRE_OPC_ANY,
2014 if (IS_ERR(op_data))
2015 RETURN(PTR_ERR(op_data));
2017 op_data->op_valid = OBD_MD_FLFLAGS;
2018 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2019 ll_finish_md_op_data(op_data);
2021 CERROR("%s: failure inode "DFID": rc = %d\n",
2022 sbi->ll_md_exp->exp_obd->obd_name,
2023 PFID(ll_inode2fid(inode)), rc);
2027 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2029 flags = body->mbo_flags;
2031 ptlrpc_req_finished(req);
2033 RETURN(put_user(flags, (int __user *)arg));
2035 case FSFILT_IOC_SETFLAGS: {
2037 struct md_op_data *op_data;
2038 struct cl_object *obj;
2040 if (get_user(flags, (int __user *)arg))
2043 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2044 LUSTRE_OPC_ANY, NULL);
2045 if (IS_ERR(op_data))
2046 RETURN(PTR_ERR(op_data));
2048 op_data->op_attr_flags = flags;
2049 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2050 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2051 ll_finish_md_op_data(op_data);
2052 ptlrpc_req_finished(req);
2056 inode->i_flags = ll_ext_to_inode_flags(flags);
2058 obj = ll_i2info(inode)->lli_clob;
2062 OBD_ALLOC_PTR(attr);
2066 attr->ia_valid = ATTR_ATTR_FLAG;
2067 rc = cl_setattr_ost(obj, attr, flags);
2079 int ll_flush_ctx(struct inode *inode)
2081 struct ll_sb_info *sbi = ll_i2sbi(inode);
2083 CDEBUG(D_SEC, "flush context for user %d\n",
2084 from_kuid(&init_user_ns, current_uid()));
2086 obd_set_info_async(NULL, sbi->ll_md_exp,
2087 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2089 obd_set_info_async(NULL, sbi->ll_dt_exp,
2090 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2095 /* umount -f client means force down, don't save state */
2096 void ll_umount_begin(struct super_block *sb)
2098 struct ll_sb_info *sbi = ll_s2sbi(sb);
2099 struct obd_device *obd;
2100 struct obd_ioctl_data *ioc_data;
2103 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2104 sb->s_count, atomic_read(&sb->s_active));
2106 obd = class_exp2obd(sbi->ll_md_exp);
2108 CERROR("Invalid MDC connection handle "LPX64"\n",
2109 sbi->ll_md_exp->exp_handle.h_cookie);
2115 obd = class_exp2obd(sbi->ll_dt_exp);
2117 CERROR("Invalid LOV connection handle "LPX64"\n",
2118 sbi->ll_dt_exp->exp_handle.h_cookie);
2124 OBD_ALLOC_PTR(ioc_data);
2126 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2127 sizeof *ioc_data, ioc_data, NULL);
2129 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2130 sizeof *ioc_data, ioc_data, NULL);
2132 OBD_FREE_PTR(ioc_data);
2135 /* Really, we'd like to wait until there are no requests outstanding,
2136 * and then continue. For now, we just invalidate the requests,
2137 * schedule() and sleep one second if needed, and hope.
2143 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2145 struct ll_sb_info *sbi = ll_s2sbi(sb);
2146 char *profilenm = get_profile_name(sb);
2150 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2151 read_only = *flags & MS_RDONLY;
2152 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2153 sizeof(KEY_READ_ONLY),
2154 KEY_READ_ONLY, sizeof(read_only),
2157 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2158 profilenm, read_only ?
2159 "read-only" : "read-write", err);
2164 sb->s_flags |= MS_RDONLY;
2166 sb->s_flags &= ~MS_RDONLY;
2168 if (sbi->ll_flags & LL_SBI_VERBOSE)
2169 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2170 read_only ? "read-only" : "read-write");
2176 * Cleanup the open handle that is cached on MDT-side.
2178 * For open case, the client side open handling thread may hit error
2179 * after the MDT grant the open. Under such case, the client should
2180 * send close RPC to the MDT as cleanup; otherwise, the open handle
2181 * on the MDT will be leaked there until the client umount or evicted.
2183 * In further, if someone unlinked the file, because the open handle
2184 * holds the reference on such file/object, then it will block the
2185 * subsequent threads that want to locate such object via FID.
2187 * \param[in] sb super block for this file-system
2188 * \param[in] open_req pointer to the original open request
2190 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2192 struct mdt_body *body;
2193 struct md_op_data *op_data;
2194 struct ptlrpc_request *close_req = NULL;
2195 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2198 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2199 OBD_ALLOC_PTR(op_data);
2200 if (op_data == NULL) {
2201 CWARN("%s: cannot allocate op_data to release open handle for "
2203 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2208 op_data->op_fid1 = body->mbo_fid1;
2209 op_data->op_handle = body->mbo_handle;
2210 op_data->op_mod_time = cfs_time_current_sec();
2211 md_close(exp, op_data, NULL, &close_req);
2212 ptlrpc_req_finished(close_req);
2213 ll_finish_md_op_data(op_data);
2218 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2219 struct super_block *sb, struct lookup_intent *it)
2221 struct ll_sb_info *sbi = NULL;
2222 struct lustre_md md = { NULL };
2226 LASSERT(*inode || sb);
2227 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2228 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2229 sbi->ll_md_exp, &md);
2234 rc = ll_update_inode(*inode, &md);
2238 LASSERT(sb != NULL);
2241 * At this point server returns to client's same fid as client
2242 * generated for creating. So using ->fid1 is okay here.
2244 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2246 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2247 sbi->ll_flags & LL_SBI_32BIT_API),
2249 if (IS_ERR(*inode)) {
2250 #ifdef CONFIG_FS_POSIX_ACL
2252 posix_acl_release(md.posix_acl);
2253 md.posix_acl = NULL;
2256 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2258 CERROR("new_inode -fatal: rc %d\n", rc);
2263 /* Handling piggyback layout lock.
2264 * Layout lock can be piggybacked by getattr and open request.
2265 * The lsm can be applied to inode only if it comes with a layout lock
2266 * otherwise correct layout may be overwritten, for example:
2267 * 1. proc1: mdt returns a lsm but not granting layout
2268 * 2. layout was changed by another client
2269 * 3. proc2: refresh layout and layout lock granted
2270 * 4. proc1: to apply a stale layout */
2271 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2272 struct lustre_handle lockh;
2273 struct ldlm_lock *lock;
2275 lockh.cookie = it->d.lustre.it_lock_handle;
2276 lock = ldlm_handle2lock(&lockh);
2277 LASSERT(lock != NULL);
2278 if (ldlm_has_layout(lock)) {
2279 struct cl_object_conf conf;
2281 memset(&conf, 0, sizeof(conf));
2282 conf.coc_opc = OBJECT_CONF_SET;
2283 conf.coc_inode = *inode;
2284 conf.coc_lock = lock;
2285 conf.u.coc_layout = md.layout;
2286 (void)ll_layout_conf(*inode, &conf);
2288 LDLM_LOCK_PUT(lock);
2294 md_free_lustre_md(sbi->ll_md_exp, &md);
2297 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2298 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2303 int ll_obd_statfs(struct inode *inode, void __user *arg)
2305 struct ll_sb_info *sbi = NULL;
2306 struct obd_export *exp;
2308 struct obd_ioctl_data *data = NULL;
2310 __u32 __user flags; /* not user, but obd_iocontrol is abused */
2313 if (!inode || !(sbi = ll_i2sbi(inode)))
2314 GOTO(out_statfs, rc = -EINVAL);
2316 rc = obd_ioctl_getdata(&buf, &len, arg);
2318 GOTO(out_statfs, rc);
2321 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2322 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2323 GOTO(out_statfs, rc = -EINVAL);
2325 if (data->ioc_inllen1 != sizeof(__u32) ||
2326 data->ioc_inllen2 != sizeof(__u32) ||
2327 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2328 data->ioc_plen2 != sizeof(struct obd_uuid))
2329 GOTO(out_statfs, rc = -EINVAL);
2331 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2332 if (type & LL_STATFS_LMV)
2333 exp = sbi->ll_md_exp;
2334 else if (type & LL_STATFS_LOV)
2335 exp = sbi->ll_dt_exp;
2337 GOTO(out_statfs, rc = -ENODEV);
2339 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2340 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2342 GOTO(out_statfs, rc);
2345 obd_ioctl_freedata(buf, len);
2349 int ll_process_config(struct lustre_cfg *lcfg)
2351 struct super_block *sb;
2356 /* The instance name contains the sb: lustre-client-aacfe000 */
2357 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2358 if (!ptr || !*(++ptr))
2360 if (sscanf(ptr, "%lx", &x) != 1)
2362 sb = (struct super_block *)x;
2363 /* This better be a real Lustre superblock! */
2364 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2366 /* Note we have not called client_common_fill_super yet, so
2367 proc fns must be able to handle that! */
2368 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2375 /* this function prepares md_op_data hint for passing it down to MD stack. */
2376 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2377 struct inode *i1, struct inode *i2,
2378 const char *name, size_t namelen,
2379 __u32 mode, __u32 opc, void *data)
2381 LASSERT(i1 != NULL);
2384 /* Do not reuse namelen for something else. */
2386 return ERR_PTR(-EINVAL);
2388 if (namelen > ll_i2sbi(i1)->ll_namelen)
2389 return ERR_PTR(-ENAMETOOLONG);
2391 if (!lu_name_is_valid_2(name, namelen))
2392 return ERR_PTR(-EINVAL);
2395 if (op_data == NULL)
2396 OBD_ALLOC_PTR(op_data);
2398 if (op_data == NULL)
2399 return ERR_PTR(-ENOMEM);
2401 ll_i2gids(op_data->op_suppgids, i1, i2);
2402 op_data->op_fid1 = *ll_inode2fid(i1);
2403 op_data->op_default_stripe_offset = -1;
2404 if (S_ISDIR(i1->i_mode)) {
2405 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2406 if (opc == LUSTRE_OPC_MKDIR)
2407 op_data->op_default_stripe_offset =
2408 ll_i2info(i1)->lli_def_stripe_offset;
2412 op_data->op_fid2 = *ll_inode2fid(i2);
2413 if (S_ISDIR(i2->i_mode))
2414 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2416 fid_zero(&op_data->op_fid2);
2419 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2420 op_data->op_cli_flags |= CLI_HASH64;
2422 if (ll_need_32bit_api(ll_i2sbi(i1)))
2423 op_data->op_cli_flags |= CLI_API32;
2425 op_data->op_name = name;
2426 op_data->op_namelen = namelen;
2427 op_data->op_mode = mode;
2428 op_data->op_mod_time = cfs_time_current_sec();
2429 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2430 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2431 op_data->op_cap = cfs_curproc_cap_pack();
2432 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2433 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2434 op_data->op_bias |= MDS_CREATE_VOLATILE;
2436 op_data->op_mds = 0;
2438 op_data->op_data = data;
2443 void ll_finish_md_op_data(struct md_op_data *op_data)
2445 OBD_FREE_PTR(op_data);
2448 #ifdef HAVE_SUPEROPS_USE_DENTRY
2449 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2451 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2454 struct ll_sb_info *sbi;
2456 #ifdef HAVE_SUPEROPS_USE_DENTRY
2457 LASSERT((seq != NULL) && (dentry != NULL));
2458 sbi = ll_s2sbi(dentry->d_sb);
2460 LASSERT((seq != NULL) && (vfs != NULL));
2461 sbi = ll_s2sbi(vfs->mnt_sb);
2464 if (sbi->ll_flags & LL_SBI_NOLCK)
2465 seq_puts(seq, ",nolock");
2467 if (sbi->ll_flags & LL_SBI_FLOCK)
2468 seq_puts(seq, ",flock");
2470 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2471 seq_puts(seq, ",localflock");
2473 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2474 seq_puts(seq, ",user_xattr");
2476 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2477 seq_puts(seq, ",lazystatfs");
2479 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2480 seq_puts(seq, ",user_fid2path");
2486 * Get obd name by cmd, and copy out to user space
2488 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2490 struct ll_sb_info *sbi = ll_i2sbi(inode);
2491 struct obd_device *obd;
2494 if (cmd == OBD_IOC_GETDTNAME)
2495 obd = class_exp2obd(sbi->ll_dt_exp);
2496 else if (cmd == OBD_IOC_GETMDNAME)
2497 obd = class_exp2obd(sbi->ll_md_exp);
2504 if (copy_to_user((void __user *)arg, obd->obd_name,
2505 strlen(obd->obd_name) + 1))
2512 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2513 * fsname will be returned in this buffer; otherwise, a static buffer will be
2514 * used to store the fsname and returned to caller.
2516 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2518 static char fsname_static[MTI_NAME_MAXLEN];
2519 struct lustre_sb_info *lsi = s2lsi(sb);
2524 /* this means the caller wants to use static buffer
2525 * and it doesn't care about race. Usually this is
2526 * in error reporting path */
2527 buf = fsname_static;
2528 buflen = sizeof(fsname_static);
2531 len = strlen(lsi->lsi_lmd->lmd_profile);
2532 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2533 if (ptr && (strcmp(ptr, "-client") == 0))
2536 if (unlikely(len >= buflen))
2538 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2544 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2551 p.mnt = current->fs->root.mnt;
2553 path = d_path(&p, buf, bufsize);
2558 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2560 char *buf, *path = NULL;
2561 struct dentry *dentry = NULL;
2562 struct inode *inode = page->mapping->host;
2564 /* this can be called inside spin lock so use GFP_ATOMIC. */
2565 buf = (char *)__get_free_page(GFP_ATOMIC);
2567 dentry = d_find_alias(page->mapping->host);
2569 path = ll_d_path(dentry, buf, PAGE_SIZE);
2573 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2574 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2575 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2576 PFID(ll_inode2fid(inode)),
2577 (path && !IS_ERR(path)) ? path : "", ioret);
2583 free_page((unsigned long)buf);
2586 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2587 struct lov_user_md **kbuf)
2589 struct lov_user_md lum;
2593 if (copy_from_user(&lum, md, sizeof(lum)))
2596 lum_size = ll_lov_user_md_size(&lum);
2600 OBD_ALLOC(*kbuf, lum_size);
2604 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2605 OBD_FREE(*kbuf, lum_size);
2613 * Compute llite root squash state after a change of root squash
2614 * configuration setting or add/remove of a lnet nid
2616 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2618 struct root_squash_info *squash = &sbi->ll_squash;
2621 lnet_process_id_t id;
2623 /* Update norootsquash flag */
2624 down_write(&squash->rsi_sem);
2625 if (list_empty(&squash->rsi_nosquash_nids))
2626 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2628 /* Do not apply root squash as soon as one of our NIDs is
2629 * in the nosquash_nids list */
2632 while (LNetGetId(i++, &id) != -ENOENT) {
2633 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2635 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2641 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2643 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2645 up_write(&squash->rsi_sem);
2649 * Parse linkea content to extract information about a given hardlink
2651 * \param[in] ldata - Initialized linkea data
2652 * \param[in] linkno - Link identifier
2653 * \param[out] parent_fid - The entry's parent FID
2654 * \param[out] ln - Entry name destination buffer
2656 * \retval 0 on success
2657 * \retval Appropriate negative error code on failure
2659 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2660 struct lu_fid *parent_fid, struct lu_name *ln)
2666 rc = linkea_init(ldata);
2670 if (linkno >= ldata->ld_leh->leh_reccount)
2671 /* beyond last link */
2674 linkea_first_entry(ldata);
2675 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2676 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2681 linkea_next_entry(ldata);
2691 * Get parent FID and name of an identified link. Operation is performed for
2692 * a given link number, letting the caller iterate over linkno to list one or
2693 * all links of an entry.
2695 * \param[in] file - File descriptor against which to perform the operation
2696 * \param[in,out] arg - User-filled structure containing the linkno to operate
2697 * on and the available size. It is eventually filled with
2698 * the requested information or left untouched on error
2700 * \retval - 0 on success
2701 * \retval - Appropriate negative error code on failure
2703 int ll_getparent(struct file *file, struct getparent __user *arg)
2705 struct dentry *dentry = file->f_path.dentry;
2706 struct inode *inode = dentry->d_inode;
2707 struct linkea_data *ldata;
2708 struct lu_buf buf = LU_BUF_NULL;
2710 struct lu_fid parent_fid;
2717 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2718 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2721 if (get_user(name_size, &arg->gp_name_size))
2724 if (get_user(linkno, &arg->gp_linkno))
2727 if (name_size > PATH_MAX)
2730 OBD_ALLOC(ldata, sizeof(*ldata));
2734 rc = linkea_data_new(ldata, &buf);
2736 GOTO(ldata_free, rc);
2738 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2742 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2746 if (ln.ln_namelen >= name_size)
2747 GOTO(lb_free, rc = -EOVERFLOW);
2749 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2750 GOTO(lb_free, rc = -EFAULT);
2752 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2753 GOTO(lb_free, rc = -EFAULT);
2755 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2756 GOTO(lb_free, rc = -EFAULT);
2761 OBD_FREE(ldata, sizeof(*ldata));