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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/types.h>
44 #include "../include/lustre/lustre_ioctl.h"
45 #include "../include/lustre_ha.h"
46 #include "../include/lustre_dlm.h"
47 #include "../include/lprocfs_status.h"
48 #include "../include/lustre_disk.h"
49 #include "../include/lustre_param.h"
50 #include "../include/lustre_log.h"
51 #include "../include/cl_object.h"
52 #include "../include/obd_cksum.h"
53 #include "llite_internal.h"
55 struct kmem_cache *ll_file_data_slab;
56 struct dentry *llite_root;
57 struct kset *llite_kset;
60 #define log2(n) ffz(~(n))
63 static struct ll_sb_info *ll_init_sbi(struct super_block *sb)
65 struct ll_sb_info *sbi = NULL;
67 unsigned long lru_page_max;
72 sbi = kzalloc(sizeof(*sbi), GFP_NOFS);
76 spin_lock_init(&sbi->ll_lock);
77 mutex_init(&sbi->ll_lco.lco_lock);
78 spin_lock_init(&sbi->ll_pp_extent_lock);
79 spin_lock_init(&sbi->ll_process_lock);
80 sbi->ll_rw_stats_on = 0;
83 pages = si.totalram - si.totalhigh;
84 lru_page_max = pages / 2;
86 sbi->ll_cache = cl_cache_init(lru_page_max);
92 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
93 SBI_DEFAULT_READAHEAD_MAX);
94 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
95 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
96 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
98 ll_generate_random_uuid(uuid);
99 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
100 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
102 sbi->ll_flags |= LL_SBI_VERBOSE;
103 sbi->ll_flags |= LL_SBI_CHECKSUM;
105 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
106 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
108 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
109 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
111 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
115 /* metadata statahead is enabled by default */
116 sbi->ll_sa_max = LL_SA_RPC_DEF;
117 atomic_set(&sbi->ll_sa_total, 0);
118 atomic_set(&sbi->ll_sa_wrong, 0);
119 atomic_set(&sbi->ll_sa_running, 0);
120 atomic_set(&sbi->ll_agl_total, 0);
121 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
124 sbi->ll_squash.rsi_uid = 0;
125 sbi->ll_squash.rsi_gid = 0;
126 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
127 init_rwsem(&sbi->ll_squash.rsi_sem);
134 static void ll_free_sbi(struct super_block *sb)
136 struct ll_sb_info *sbi = ll_s2sbi(sb);
139 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
140 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
141 cl_cache_decref(sbi->ll_cache);
142 sbi->ll_cache = NULL;
148 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
149 struct vfsmount *mnt)
151 struct inode *root = NULL;
152 struct ll_sb_info *sbi = ll_s2sbi(sb);
153 struct obd_device *obd;
154 struct obd_statfs *osfs = NULL;
155 struct ptlrpc_request *request = NULL;
156 struct obd_connect_data *data = NULL;
157 struct obd_uuid *uuid;
158 struct md_op_data *op_data;
159 struct lustre_md lmd;
161 int size, err, checksum;
163 obd = class_name2obd(md);
165 CERROR("MD %s: not setup or attached\n", md);
169 data = kzalloc(sizeof(*data), GFP_NOFS);
173 osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
179 /* indicate the features supported by this client */
180 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
181 OBD_CONNECT_ATTRFID |
182 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
183 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
184 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
185 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
186 OBD_CONNECT_64BITHASH |
187 OBD_CONNECT_EINPROGRESS |
188 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
189 OBD_CONNECT_LAYOUTLOCK |
190 OBD_CONNECT_PINGLESS |
191 OBD_CONNECT_MAX_EASIZE |
192 OBD_CONNECT_FLOCK_DEAD |
193 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
194 OBD_CONNECT_OPEN_BY_FID |
195 OBD_CONNECT_DIR_STRIPE |
196 OBD_CONNECT_BULK_MBITS;
198 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
199 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
200 #ifdef CONFIG_FS_POSIX_ACL
201 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
204 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
205 /* flag mdc connection as lightweight, only used for test
206 * purpose, use with care
208 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
210 data->ocd_ibits_known = MDS_INODELOCK_FULL;
211 data->ocd_version = LUSTRE_VERSION_CODE;
213 if (sb->s_flags & MS_RDONLY)
214 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
215 if (sbi->ll_flags & LL_SBI_USER_XATTR)
216 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
218 if (sbi->ll_flags & LL_SBI_FLOCK)
219 sbi->ll_fop = &ll_file_operations_flock;
220 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
221 sbi->ll_fop = &ll_file_operations;
223 sbi->ll_fop = &ll_file_operations_noflock;
226 data->ocd_connect_flags |= OBD_CONNECT_REAL;
228 /* always ping even if server suppress_pings */
229 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
230 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
232 data->ocd_brw_size = MD_MAX_BRW_SIZE;
234 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
237 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing recovery, of which this client is not a part. Please wait for recovery to complete, abort, or time out.\n",
241 CERROR("cannot connect to %s: rc = %d\n", md, err);
245 sbi->ll_md_exp->exp_connect_data = *data;
247 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
248 LUSTRE_SEQ_METADATA);
250 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
251 sbi->ll_md_exp->exp_obd->obd_name, err);
255 /* For mount, we only need fs info from MDT0, and also in DNE, it
256 * can make sure the client can be mounted as long as MDT0 is
259 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
260 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
261 OBD_STATFS_FOR_MDT0);
265 /* This needs to be after statfs to ensure connect has finished.
266 * Note that "data" does NOT contain the valid connect reply.
267 * If connecting to a 1.8 server there will be no LMV device, so
268 * we can access the MDC export directly and exp_connect_flags will
269 * be non-zero, but if accessing an upgraded 2.1 server it will
270 * have the correct flags filled in.
271 * XXX: fill in the LMV exp_connect_flags from MDC(s).
273 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
274 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
275 valid != CLIENT_CONNECT_MDT_REQD) {
278 buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
283 obd_connect_flags2str(buf, PAGE_SIZE,
284 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
285 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support feature(s) needed for correct operation of this client (%s). Please upgrade server or downgrade client.\n",
286 sbi->ll_md_exp->exp_obd->obd_name, buf);
292 size = sizeof(*data);
293 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
294 KEY_CONN_DATA, &size, data);
296 CERROR("%s: Get connect data failed: rc = %d\n",
297 sbi->ll_md_exp->exp_obd->obd_name, err);
301 LASSERT(osfs->os_bsize);
302 sb->s_blocksize = osfs->os_bsize;
303 sb->s_blocksize_bits = log2(osfs->os_bsize);
304 sb->s_magic = LL_SUPER_MAGIC;
305 sb->s_maxbytes = MAX_LFS_FILESIZE;
306 sbi->ll_namelen = osfs->os_namelen;
307 sbi->ll_mnt.mnt = current->fs->root.mnt;
309 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
310 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
311 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
312 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
315 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
316 sb->s_flags |= MS_POSIXACL;
317 sbi->ll_flags |= LL_SBI_ACL;
319 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
320 sb->s_flags &= ~MS_POSIXACL;
321 sbi->ll_flags &= ~LL_SBI_ACL;
324 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
325 sbi->ll_flags |= LL_SBI_64BIT_HASH;
327 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
328 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_SHIFT;
330 sbi->ll_md_brw_pages = 1;
332 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
333 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
335 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
336 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
338 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
341 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
342 sbi->ll_xattr_cache_enabled = 1;
346 obd = class_name2obd(dt);
348 CERROR("DT %s: not setup or attached\n", dt);
353 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
354 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
355 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
356 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
357 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
358 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
359 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
360 OBD_CONNECT_EINPROGRESS |
361 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
362 OBD_CONNECT_LAYOUTLOCK |
363 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
364 OBD_CONNECT_BULK_MBITS;
366 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
367 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
368 * disabled by default, because it can still be enabled on the
369 * fly via /sys. As a consequence, we still need to come to an
370 * agreement on the supported algorithms at connect time
372 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
374 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
375 data->ocd_cksum_types = OBD_CKSUM_ADLER;
377 data->ocd_cksum_types = cksum_types_supported_client();
380 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
382 /* always ping even if server suppress_pings */
383 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
384 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
386 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
387 data->ocd_connect_flags,
388 data->ocd_version, data->ocd_grant);
390 obd->obd_upcall.onu_owner = &sbi->ll_lco;
391 obd->obd_upcall.onu_upcall = cl_ocd_update;
393 data->ocd_brw_size = DT_MAX_BRW_SIZE;
395 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
398 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing recovery, of which this client is not a part. Please wait for recovery to complete, abort, or time out.\n",
402 CERROR("%s: Cannot connect to %s: rc = %d\n",
403 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
407 sbi->ll_dt_exp->exp_connect_data = *data;
409 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
410 LUSTRE_SEQ_METADATA);
412 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
413 sbi->ll_dt_exp->exp_obd->obd_name, err);
417 mutex_lock(&sbi->ll_lco.lco_lock);
418 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
419 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
420 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
421 mutex_unlock(&sbi->ll_lco.lco_lock);
423 fid_zero(&sbi->ll_root_fid);
424 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
426 CERROR("cannot mds_connect: rc = %d\n", err);
429 if (!fid_is_sane(&sbi->ll_root_fid)) {
430 CERROR("%s: Invalid root fid "DFID" during mount\n",
431 sbi->ll_md_exp->exp_obd->obd_name,
432 PFID(&sbi->ll_root_fid));
436 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
438 sb->s_op = &lustre_super_operations;
439 sb->s_xattr = ll_xattr_handlers;
440 #if THREAD_SIZE >= 8192 /*b=17630*/
441 sb->s_export_op = &lustre_export_operations;
445 * XXX: move this to after cbd setup?
447 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
448 if (sbi->ll_flags & LL_SBI_ACL)
449 valid |= OBD_MD_FLACL;
451 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
457 op_data->op_fid1 = sbi->ll_root_fid;
458 op_data->op_mode = 0;
459 op_data->op_valid = valid;
461 err = md_getattr(sbi->ll_md_exp, op_data, &request);
464 CERROR("%s: md_getattr failed for root: rc = %d\n",
465 sbi->ll_md_exp->exp_obd->obd_name, err);
469 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
470 sbi->ll_md_exp, &lmd);
472 CERROR("failed to understand root inode md: rc = %d\n", err);
473 ptlrpc_req_finished(request);
477 LASSERT(fid_is_sane(&sbi->ll_root_fid));
478 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
479 sbi->ll_flags & LL_SBI_32BIT_API),
481 md_free_lustre_md(sbi->ll_md_exp, &lmd);
482 ptlrpc_req_finished(request);
485 #ifdef CONFIG_FS_POSIX_ACL
487 posix_acl_release(lmd.posix_acl);
488 lmd.posix_acl = NULL;
492 CERROR("lustre_lite: bad iget4 for root\n");
496 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
497 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
498 KEY_CHECKSUM, sizeof(checksum), &checksum,
501 CERROR("%s: Set checksum failed: rc = %d\n",
502 sbi->ll_dt_exp->exp_obd->obd_name, err);
507 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
508 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
509 sbi->ll_cache, NULL);
511 CERROR("%s: Set cache_set failed: rc = %d\n",
512 sbi->ll_dt_exp->exp_obd->obd_name, err);
516 sb->s_root = d_make_root(root);
518 CERROR("%s: can't make root dentry\n",
519 ll_get_fsname(sb, NULL, 0));
524 sbi->ll_sdev_orig = sb->s_dev;
526 /* We set sb->s_dev equal on all lustre clients in order to support
527 * NFS export clustering. NFSD requires that the FSID be the same
530 /* s_dev is also used in lt_compare() to compare two fs, but that is
531 * only a node-local comparison.
533 uuid = obd_get_uuid(sbi->ll_md_exp);
535 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
536 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
543 err = ldebugfs_register_mountpoint(llite_root, sb, dt, md);
545 CERROR("%s: could not register mount in debugfs: "
546 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
555 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
557 obd_disconnect(sbi->ll_dt_exp);
558 sbi->ll_dt_exp = NULL;
560 obd_fid_fini(sbi->ll_md_exp->exp_obd);
562 obd_disconnect(sbi->ll_md_exp);
563 sbi->ll_md_exp = NULL;
570 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
574 size = sizeof(*lmmsize);
575 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
576 KEY_MAX_EASIZE, &size, lmmsize);
578 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
579 sbi->ll_dt_exp->exp_obd->obd_name, rc);
584 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
585 KEY_MAX_EASIZE, &size, lmmsize);
587 CERROR("Get max mdsize error rc %d\n", rc);
593 * Get the value of the default_easize parameter.
595 * \see client_obd::cl_default_mds_easize
597 * \param[in] sbi superblock info for this filesystem
598 * \param[out] lmmsize pointer to storage location for value
600 * \retval 0 on success
601 * \retval negative negated errno on failure
603 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
608 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
609 KEY_DEFAULT_EASIZE, &size, lmmsize);
611 CERROR("Get default mdsize error rc %d\n", rc);
617 * Set the default_easize parameter to the given value.
619 * \see client_obd::cl_default_mds_easize
621 * \param[in] sbi superblock info for this filesystem
622 * \param[in] lmmsize the size to set
624 * \retval 0 on success
625 * \retval negative negated errno on failure
627 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
629 if (lmmsize < sizeof(struct lov_mds_md) ||
630 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
633 return obd_set_info_async(NULL, sbi->ll_md_exp,
634 sizeof(KEY_DEFAULT_EASIZE),
636 sizeof(int), &lmmsize, NULL);
639 static void client_common_put_super(struct super_block *sb)
641 struct ll_sb_info *sbi = ll_s2sbi(sb);
645 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
646 obd_disconnect(sbi->ll_dt_exp);
647 sbi->ll_dt_exp = NULL;
649 ldebugfs_unregister_mountpoint(sbi);
651 obd_fid_fini(sbi->ll_md_exp->exp_obd);
652 obd_disconnect(sbi->ll_md_exp);
653 sbi->ll_md_exp = NULL;
656 void ll_kill_super(struct super_block *sb)
658 struct ll_sb_info *sbi;
661 if (!(sb->s_flags & MS_ACTIVE))
665 /* we need to restore s_dev from changed for clustered NFS before
666 * put_super because new kernels have cached s_dev and change sb->s_dev
667 * in put_super not affected real removing devices
670 sb->s_dev = sbi->ll_sdev_orig;
671 sbi->ll_umounting = 1;
673 /* wait running statahead threads to quit */
674 while (atomic_read(&sbi->ll_sa_running) > 0) {
675 set_current_state(TASK_UNINTERRUPTIBLE);
676 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
681 static inline int ll_set_opt(const char *opt, char *data, int fl)
683 if (strncmp(opt, data, strlen(opt)) != 0)
689 /* non-client-specific mount options are parsed in lmd_parse */
690 static int ll_options(char *options, int *flags)
693 char *s1 = options, *s2;
698 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
701 CDEBUG(D_SUPER, "next opt=%s\n", s1);
702 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
707 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
712 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
717 tmp = ll_set_opt("noflock", s1,
718 LL_SBI_FLOCK | LL_SBI_LOCALFLOCK);
723 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
728 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
733 tmp = ll_set_opt("context", s1, 1);
736 tmp = ll_set_opt("fscontext", s1, 1);
739 tmp = ll_set_opt("defcontext", s1, 1);
742 tmp = ll_set_opt("rootcontext", s1, 1);
745 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
750 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
756 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
761 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
766 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
771 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
776 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
781 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
786 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
791 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
796 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
801 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
806 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
812 s2 = strchr(s1, ',');
820 void ll_lli_init(struct ll_inode_info *lli)
822 lli->lli_inode_magic = LLI_INODE_MAGIC;
824 spin_lock_init(&lli->lli_lock);
825 lli->lli_posix_acl = NULL;
826 /* Do not set lli_fid, it has been initialized already. */
827 fid_zero(&lli->lli_pfid);
828 lli->lli_mds_read_och = NULL;
829 lli->lli_mds_write_och = NULL;
830 lli->lli_mds_exec_och = NULL;
831 lli->lli_open_fd_read_count = 0;
832 lli->lli_open_fd_write_count = 0;
833 lli->lli_open_fd_exec_count = 0;
834 mutex_init(&lli->lli_och_mutex);
835 spin_lock_init(&lli->lli_agl_lock);
836 spin_lock_init(&lli->lli_layout_lock);
837 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
838 lli->lli_clob = NULL;
840 init_rwsem(&lli->lli_xattrs_list_rwsem);
841 mutex_init(&lli->lli_xattrs_enq_lock);
843 LASSERT(lli->lli_vfs_inode.i_mode != 0);
844 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
845 mutex_init(&lli->lli_readdir_mutex);
846 lli->lli_opendir_key = NULL;
848 spin_lock_init(&lli->lli_sa_lock);
849 lli->lli_opendir_pid = 0;
850 lli->lli_sa_enabled = 0;
851 lli->lli_def_stripe_offset = -1;
853 mutex_init(&lli->lli_size_mutex);
854 lli->lli_symlink_name = NULL;
855 init_rwsem(&lli->lli_trunc_sem);
856 range_lock_tree_init(&lli->lli_write_tree);
857 init_rwsem(&lli->lli_glimpse_sem);
858 lli->lli_glimpse_time = 0;
859 INIT_LIST_HEAD(&lli->lli_agl_list);
860 lli->lli_agl_index = 0;
861 lli->lli_async_rc = 0;
863 mutex_init(&lli->lli_layout_mutex);
866 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
868 struct lustre_profile *lprof = NULL;
869 struct lustre_sb_info *lsi = s2lsi(sb);
870 struct ll_sb_info *sbi;
871 char *dt = NULL, *md = NULL;
872 char *profilenm = get_profile_name(sb);
873 struct config_llog_instance *cfg;
875 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
877 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
879 cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
883 try_module_get(THIS_MODULE);
885 /* client additional sb info */
886 sbi = ll_init_sbi(sb);
887 lsi->lsi_llsbi = sbi;
889 module_put(THIS_MODULE);
894 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
898 err = super_setup_bdi_name(sb, "lustre-%d",
899 atomic_inc_return(&ll_bdi_num));
903 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
904 sb->s_d_op = &ll_d_ops;
906 /* Generate a string unique to this super, in case some joker tries
907 * to mount the same fs at two mount points.
908 * Use the address of the super itself.
910 cfg->cfg_instance = sb;
911 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
912 cfg->cfg_callback = class_config_llog_handler;
913 /* set up client obds */
914 err = lustre_process_log(sb, profilenm, cfg);
918 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
919 lprof = class_get_profile(profilenm);
921 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS. Does that filesystem exist?\n",
926 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
927 lprof->lp_md, lprof->lp_dt);
929 dt = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
935 md = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_md, cfg->cfg_instance);
941 /* connections, registrations, sb setup */
942 err = client_common_fill_super(sb, md, dt, mnt);
944 sbi->ll_client_common_fill_super_succeeded = 1;
950 class_put_profile(lprof);
953 else if (sbi->ll_flags & LL_SBI_VERBOSE)
954 LCONSOLE_WARN("Mounted %s\n", profilenm);
958 } /* ll_fill_super */
960 void ll_put_super(struct super_block *sb)
962 struct config_llog_instance cfg, params_cfg;
963 struct obd_device *obd;
964 struct lustre_sb_info *lsi = s2lsi(sb);
965 struct ll_sb_info *sbi = ll_s2sbi(sb);
966 char *profilenm = get_profile_name(sb);
967 int next, force = 1, rc = 0;
970 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
972 cfg.cfg_instance = sb;
973 lustre_end_log(sb, profilenm, &cfg);
975 params_cfg.cfg_instance = sb;
976 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
978 if (sbi->ll_md_exp) {
979 obd = class_exp2obd(sbi->ll_md_exp);
981 force = obd->obd_force;
984 /* Wait for unstable pages to be committed to stable storage */
986 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
988 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
989 !atomic_long_read(&sbi->ll_cache->ccc_unstable_nr),
993 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
994 if (!force && rc != -EINTR)
995 LASSERTF(!ccc_count, "count: %li\n", ccc_count);
997 /* We need to set force before the lov_disconnect in
998 * lustre_common_put_super, since l_d cleans up osc's as well.
1002 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1004 obd->obd_force = force;
1008 if (sbi->ll_client_common_fill_super_succeeded) {
1009 /* Only if client_common_fill_super succeeded */
1010 client_common_put_super(sb);
1014 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1015 class_manual_cleanup(obd);
1017 if (sbi->ll_flags & LL_SBI_VERBOSE)
1018 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1021 class_del_profile(profilenm);
1024 lsi->lsi_llsbi = NULL;
1026 lustre_common_put_super(sb);
1028 cl_env_cache_purge(~0);
1030 module_put(THIS_MODULE);
1031 } /* client_put_super */
1033 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1035 struct inode *inode = NULL;
1037 /* NOTE: we depend on atomic igrab() -bzzz */
1038 lock_res_and_lock(lock);
1039 if (lock->l_resource->lr_lvb_inode) {
1040 struct ll_inode_info *lli;
1042 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1043 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1044 inode = igrab(lock->l_resource->lr_lvb_inode);
1046 inode = lock->l_resource->lr_lvb_inode;
1047 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1048 D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
1049 lock->l_resource->lr_lvb_inode,
1050 lli->lli_inode_magic);
1054 unlock_res_and_lock(lock);
1058 void ll_dir_clear_lsm_md(struct inode *inode)
1060 struct ll_inode_info *lli = ll_i2info(inode);
1062 LASSERT(S_ISDIR(inode->i_mode));
1064 if (lli->lli_lsm_md) {
1065 lmv_free_memmd(lli->lli_lsm_md);
1066 lli->lli_lsm_md = NULL;
1070 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1071 const struct lu_fid *fid,
1072 struct lustre_md *md)
1074 struct ll_sb_info *sbi = ll_s2sbi(sb);
1075 struct mdt_body *body = md->body;
1076 struct inode *inode;
1079 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1080 inode = iget_locked(sb, ino);
1082 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1083 ll_get_fsname(sb, NULL, 0), PFID(fid));
1084 return ERR_PTR(-ENOENT);
1087 if (inode->i_state & I_NEW) {
1088 struct ll_inode_info *lli = ll_i2info(inode);
1089 struct lmv_stripe_md *lsm = md->lmv;
1091 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1092 (body->mbo_mode & S_IFMT);
1093 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1096 LTIME_S(inode->i_mtime) = 0;
1097 LTIME_S(inode->i_atime) = 0;
1098 LTIME_S(inode->i_ctime) = 0;
1101 inode->i_op = &ll_dir_inode_operations;
1102 inode->i_fop = &ll_dir_operations;
1103 lli->lli_fid = *fid;
1107 /* master object FID */
1108 lli->lli_pfid = body->mbo_fid1;
1109 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1110 lli, PFID(fid), PFID(&lli->lli_pfid));
1111 unlock_new_inode(inode);
1117 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1119 struct lmv_stripe_md *lsm = md->lmv;
1125 * XXX sigh, this lsm_root initialization should be in
1126 * LMV layer, but it needs ll_iget right now, so we
1127 * put this here right now.
1129 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1130 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1131 LASSERT(!lsm->lsm_md_oinfo[i].lmo_root);
1132 /* Unfortunately ll_iget will call ll_update_inode,
1133 * where the initialization of slave inode is slightly
1134 * different, so it reset lsm_md to NULL to avoid
1135 * initializing lsm for slave inode.
1137 /* For migrating inode, master stripe and master object will
1138 * be same, so we only need assign this inode
1140 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && !i)
1141 lsm->lsm_md_oinfo[i].lmo_root = inode;
1143 lsm->lsm_md_oinfo[i].lmo_root =
1144 ll_iget_anon_dir(inode->i_sb, fid, md);
1145 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1146 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1148 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1156 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1157 const struct lmv_stripe_md *lsm_md2)
1159 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1160 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1161 lsm_md1->lsm_md_master_mdt_index ==
1162 lsm_md2->lsm_md_master_mdt_index &&
1163 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1164 lsm_md1->lsm_md_layout_version ==
1165 lsm_md2->lsm_md_layout_version &&
1166 !strcmp(lsm_md1->lsm_md_pool_name,
1167 lsm_md2->lsm_md_pool_name);
1170 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1172 struct ll_inode_info *lli = ll_i2info(inode);
1173 struct lmv_stripe_md *lsm = md->lmv;
1176 LASSERT(S_ISDIR(inode->i_mode));
1177 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1178 PFID(ll_inode2fid(inode)));
1180 /* no striped information from request. */
1182 if (!lli->lli_lsm_md) {
1184 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1185 LMV_HASH_FLAG_MIGRATION) {
1187 * migration is done, the temporay MIGRATE layout has
1190 CDEBUG(D_INODE, DFID" finish migration.\n",
1191 PFID(ll_inode2fid(inode)));
1192 lmv_free_memmd(lli->lli_lsm_md);
1193 lli->lli_lsm_md = NULL;
1197 * The lustre_md from req does not include stripeEA,
1204 /* set the directory layout */
1205 if (!lli->lli_lsm_md) {
1206 struct cl_attr *attr;
1208 rc = ll_init_lsm_md(inode, md);
1213 * set lsm_md to NULL, so the following free lustre_md
1214 * will not free this lsm
1217 lli->lli_lsm_md = lsm;
1219 attr = kzalloc(sizeof(*attr), GFP_NOFS);
1223 /* validate the lsm */
1224 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1225 ll_md_blocking_ast);
1231 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1232 md->body->mbo_nlink = attr->cat_nlink;
1233 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1234 md->body->mbo_size = attr->cat_size;
1235 if (md->body->mbo_valid & OBD_MD_FLATIME)
1236 md->body->mbo_atime = attr->cat_atime;
1237 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1238 md->body->mbo_ctime = attr->cat_ctime;
1239 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1240 md->body->mbo_mtime = attr->cat_mtime;
1244 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1245 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1249 /* Compare the old and new stripe information */
1250 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1251 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1254 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p) magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1255 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1256 inode, lsm, old_lsm,
1257 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1258 lsm->lsm_md_stripe_count,
1259 old_lsm->lsm_md_stripe_count,
1260 lsm->lsm_md_master_mdt_index,
1261 old_lsm->lsm_md_master_mdt_index,
1262 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1263 lsm->lsm_md_layout_version,
1264 old_lsm->lsm_md_layout_version,
1265 lsm->lsm_md_pool_name,
1266 old_lsm->lsm_md_pool_name);
1268 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1269 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1270 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1271 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1274 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1275 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1276 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1277 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1286 void ll_clear_inode(struct inode *inode)
1288 struct ll_inode_info *lli = ll_i2info(inode);
1289 struct ll_sb_info *sbi = ll_i2sbi(inode);
1291 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1292 PFID(ll_inode2fid(inode)), inode);
1294 if (S_ISDIR(inode->i_mode)) {
1295 /* these should have been cleared in ll_file_release */
1296 LASSERT(!lli->lli_opendir_key);
1297 LASSERT(!lli->lli_sai);
1298 LASSERT(lli->lli_opendir_pid == 0);
1301 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1303 LASSERT(!lli->lli_open_fd_write_count);
1304 LASSERT(!lli->lli_open_fd_read_count);
1305 LASSERT(!lli->lli_open_fd_exec_count);
1307 if (lli->lli_mds_write_och)
1308 ll_md_real_close(inode, FMODE_WRITE);
1309 if (lli->lli_mds_exec_och)
1310 ll_md_real_close(inode, FMODE_EXEC);
1311 if (lli->lli_mds_read_och)
1312 ll_md_real_close(inode, FMODE_READ);
1314 if (S_ISLNK(inode->i_mode)) {
1315 kfree(lli->lli_symlink_name);
1316 lli->lli_symlink_name = NULL;
1319 ll_xattr_cache_destroy(inode);
1321 #ifdef CONFIG_FS_POSIX_ACL
1322 if (lli->lli_posix_acl) {
1323 posix_acl_release(lli->lli_posix_acl);
1324 lli->lli_posix_acl = NULL;
1327 lli->lli_inode_magic = LLI_INODE_DEAD;
1329 if (S_ISDIR(inode->i_mode))
1330 ll_dir_clear_lsm_md(inode);
1331 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1332 LASSERT(list_empty(&lli->lli_agl_list));
1335 * XXX This has to be done before lsm is freed below, because
1336 * cl_object still uses inode lsm.
1338 cl_inode_fini(inode);
1341 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
1343 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1345 struct lustre_md md;
1346 struct inode *inode = d_inode(dentry);
1347 struct ll_sb_info *sbi = ll_i2sbi(inode);
1348 struct ptlrpc_request *request = NULL;
1351 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1352 LUSTRE_OPC_ANY, NULL);
1353 if (IS_ERR(op_data))
1354 return PTR_ERR(op_data);
1356 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1358 ptlrpc_req_finished(request);
1359 if (rc == -ENOENT) {
1361 /* Unlinked special device node? Or just a race?
1362 * Pretend we did everything.
1364 if (!S_ISREG(inode->i_mode) &&
1365 !S_ISDIR(inode->i_mode)) {
1366 ia_valid = op_data->op_attr.ia_valid;
1367 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1368 rc = simple_setattr(dentry, &op_data->op_attr);
1369 op_data->op_attr.ia_valid = ia_valid;
1371 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1372 CERROR("md_setattr fails: rc = %d\n", rc);
1377 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1378 sbi->ll_md_exp, &md);
1380 ptlrpc_req_finished(request);
1384 ia_valid = op_data->op_attr.ia_valid;
1385 /* inode size will be in cl_setattr_ost, can't do it now since dirty
1386 * cache is not cleared yet.
1388 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1389 if (S_ISREG(inode->i_mode))
1391 rc = simple_setattr(dentry, &op_data->op_attr);
1392 if (S_ISREG(inode->i_mode))
1393 inode_unlock(inode);
1394 op_data->op_attr.ia_valid = ia_valid;
1396 rc = ll_update_inode(inode, &md);
1397 ptlrpc_req_finished(request);
1402 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1403 * object(s) determine the file size and mtime. Otherwise, the MDS will
1404 * keep these values until such a time that objects are allocated for it.
1405 * We do the MDS operations first, as it is checking permissions for us.
1406 * We don't to the MDS RPC if there is nothing that we want to store there,
1407 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1408 * going to do an RPC anyways.
1410 * If we are doing a truncate, we will send the mtime and ctime updates
1411 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1412 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1415 * In case of HSMimport, we only set attr on MDS.
1417 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1419 struct inode *inode = d_inode(dentry);
1420 struct ll_inode_info *lli = ll_i2info(inode);
1421 struct md_op_data *op_data = NULL;
1424 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, valid %x, hsm_import %d\n",
1425 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid), inode,
1426 i_size_read(inode), attr->ia_size, attr->ia_valid, hsm_import);
1428 if (attr->ia_valid & ATTR_SIZE) {
1429 /* Check new size against VFS/VM file size limit and rlimit */
1430 rc = inode_newsize_ok(inode, attr->ia_size);
1434 /* The maximum Lustre file size is variable, based on the
1435 * OST maximum object size and number of stripes. This
1436 * needs another check in addition to the VFS check above.
1438 if (attr->ia_size > ll_file_maxbytes(inode)) {
1439 CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1440 PFID(&lli->lli_fid), attr->ia_size,
1441 ll_file_maxbytes(inode));
1445 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1448 /* POSIX: check before ATTR_*TIME_SET set (from setattr_prepare) */
1449 if (attr->ia_valid & TIMES_SET_FLAGS) {
1450 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1451 !capable(CFS_CAP_FOWNER))
1455 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1456 if (attr->ia_valid & ATTR_CTIME) {
1457 attr->ia_ctime = CURRENT_TIME;
1458 attr->ia_valid |= ATTR_CTIME_SET;
1460 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1461 (attr->ia_valid & ATTR_ATIME)) {
1462 attr->ia_atime = CURRENT_TIME;
1463 attr->ia_valid |= ATTR_ATIME_SET;
1465 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1466 (attr->ia_valid & ATTR_MTIME)) {
1467 attr->ia_mtime = CURRENT_TIME;
1468 attr->ia_valid |= ATTR_MTIME_SET;
1471 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1472 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1473 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1474 (s64)ktime_get_real_seconds());
1476 if (S_ISREG(inode->i_mode))
1477 inode_unlock(inode);
1480 * We always do an MDS RPC, even if we're only changing the size;
1481 * only the MDS knows whether truncate() should fail with -ETXTBUSY
1483 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1489 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1491 * If we are changing file size, file content is
1492 * modified, flag it.
1494 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1495 op_data->op_bias |= MDS_DATA_MODIFIED;
1496 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
1499 op_data->op_attr = *attr;
1501 rc = ll_md_setattr(dentry, op_data);
1505 if (!S_ISREG(inode->i_mode) || hsm_import) {
1510 if (attr->ia_valid & (ATTR_SIZE |
1511 ATTR_ATIME | ATTR_ATIME_SET |
1512 ATTR_MTIME | ATTR_MTIME_SET)) {
1513 /* For truncate and utimes sending attributes to OSTs, setting
1514 * mtime/atime to the past will be performed under PW [0:EOF]
1515 * extent lock (new_size:EOF for truncate). It may seem
1516 * excessive to send mtime/atime updates to OSTs when not
1517 * setting times to past, but it is necessary due to possible
1518 * time de-synchronization between MDT inode and OST objects
1520 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, attr, 0);
1524 * If the file was restored, it needs to set dirty flag.
1526 * We've already sent MDS_DATA_MODIFIED flag in
1527 * ll_md_setattr() for truncate. However, the MDT refuses to
1528 * set the HS_DIRTY flag on released files, so we have to set
1529 * it again if the file has been restored. Please check how
1530 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1532 * Please notice that if the file is not released, the previous
1533 * MDS_DATA_MODIFIED has taken effect and usually
1534 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1535 * This way we can save an RPC for common open + trunc
1538 if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
1539 struct hsm_state_set hss = {
1540 .hss_valid = HSS_SETMASK,
1541 .hss_setmask = HS_DIRTY,
1545 rc2 = ll_hsm_state_set(inode, &hss);
1547 * truncate and write can happen at the same time, so that
1548 * the file can be set modified even though the file is not
1549 * restored from released state, and ll_hsm_state_set() is
1550 * not applicable for the file, and rc2 < 0 is normal in this
1554 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1555 PFID(ll_inode2fid(inode)), rc2);
1560 ll_finish_md_op_data(op_data);
1562 if (S_ISREG(inode->i_mode)) {
1564 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1565 inode_dio_wait(inode);
1568 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1569 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1574 int ll_setattr(struct dentry *de, struct iattr *attr)
1576 int mode = d_inode(de)->i_mode;
1578 if ((attr->ia_valid & (ATTR_CTIME | ATTR_SIZE | ATTR_MODE)) ==
1579 (ATTR_CTIME | ATTR_SIZE | ATTR_MODE))
1580 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1582 if (((attr->ia_valid & (ATTR_MODE | ATTR_FORCE | ATTR_SIZE)) ==
1583 (ATTR_SIZE | ATTR_MODE)) &&
1584 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1585 (((mode & (S_ISGID | 0010)) == (S_ISGID | 0010)) &&
1586 !(attr->ia_mode & S_ISGID))))
1587 attr->ia_valid |= ATTR_FORCE;
1589 if ((attr->ia_valid & ATTR_MODE) &&
1591 !(attr->ia_mode & S_ISUID) &&
1592 !(attr->ia_valid & ATTR_KILL_SUID))
1593 attr->ia_valid |= ATTR_KILL_SUID;
1595 if ((attr->ia_valid & ATTR_MODE) &&
1596 ((mode & (S_ISGID | 0010)) == (S_ISGID | 0010)) &&
1597 !(attr->ia_mode & S_ISGID) &&
1598 !(attr->ia_valid & ATTR_KILL_SGID))
1599 attr->ia_valid |= ATTR_KILL_SGID;
1601 return ll_setattr_raw(de, attr, false);
1604 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1605 __u64 max_age, __u32 flags)
1607 struct ll_sb_info *sbi = ll_s2sbi(sb);
1608 struct obd_statfs obd_osfs;
1611 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1613 CERROR("md_statfs fails: rc = %d\n", rc);
1617 osfs->os_type = sb->s_magic;
1619 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1620 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1623 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1624 flags |= OBD_STATFS_NODELAY;
1626 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1628 CERROR("obd_statfs fails: rc = %d\n", rc);
1632 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1633 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1636 osfs->os_bsize = obd_osfs.os_bsize;
1637 osfs->os_blocks = obd_osfs.os_blocks;
1638 osfs->os_bfree = obd_osfs.os_bfree;
1639 osfs->os_bavail = obd_osfs.os_bavail;
1641 /* If we don't have as many objects free on the OST as inodes
1642 * on the MDS, we reduce the total number of inodes to
1643 * compensate, so that the "inodes in use" number is correct.
1645 if (obd_osfs.os_ffree < osfs->os_ffree) {
1646 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1648 osfs->os_ffree = obd_osfs.os_ffree;
1654 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1656 struct super_block *sb = de->d_sb;
1657 struct obd_statfs osfs;
1660 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1661 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1663 /* Some amount of caching on the client is allowed */
1664 rc = ll_statfs_internal(sb, &osfs,
1665 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1670 statfs_unpack(sfs, &osfs);
1672 /* We need to downshift for all 32-bit kernels, because we can't
1673 * tell if the kernel is being called via sys_statfs64() or not.
1674 * Stop before overflowing f_bsize - in which case it is better
1675 * to just risk EOVERFLOW if caller is using old sys_statfs().
1677 if (sizeof(long) < 8) {
1678 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1681 osfs.os_blocks >>= 1;
1682 osfs.os_bfree >>= 1;
1683 osfs.os_bavail >>= 1;
1687 sfs->f_blocks = osfs.os_blocks;
1688 sfs->f_bfree = osfs.os_bfree;
1689 sfs->f_bavail = osfs.os_bavail;
1690 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1694 void ll_inode_size_lock(struct inode *inode)
1696 struct ll_inode_info *lli;
1698 LASSERT(!S_ISDIR(inode->i_mode));
1700 lli = ll_i2info(inode);
1701 mutex_lock(&lli->lli_size_mutex);
1704 void ll_inode_size_unlock(struct inode *inode)
1706 struct ll_inode_info *lli;
1708 lli = ll_i2info(inode);
1709 mutex_unlock(&lli->lli_size_mutex);
1712 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1714 struct ll_inode_info *lli = ll_i2info(inode);
1715 struct mdt_body *body = md->body;
1716 struct ll_sb_info *sbi = ll_i2sbi(inode);
1718 if (body->mbo_valid & OBD_MD_FLEASIZE)
1719 cl_file_inode_init(inode, md);
1721 if (S_ISDIR(inode->i_mode)) {
1724 rc = ll_update_lsm_md(inode, md);
1729 #ifdef CONFIG_FS_POSIX_ACL
1730 if (body->mbo_valid & OBD_MD_FLACL) {
1731 spin_lock(&lli->lli_lock);
1732 if (lli->lli_posix_acl)
1733 posix_acl_release(lli->lli_posix_acl);
1734 lli->lli_posix_acl = md->posix_acl;
1735 spin_unlock(&lli->lli_lock);
1738 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1739 sbi->ll_flags & LL_SBI_32BIT_API);
1740 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1742 if (body->mbo_valid & OBD_MD_FLATIME) {
1743 if (body->mbo_atime > LTIME_S(inode->i_atime))
1744 LTIME_S(inode->i_atime) = body->mbo_atime;
1745 lli->lli_atime = body->mbo_atime;
1747 if (body->mbo_valid & OBD_MD_FLMTIME) {
1748 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1749 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1750 inode->i_ino, LTIME_S(inode->i_mtime),
1752 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1754 lli->lli_mtime = body->mbo_mtime;
1756 if (body->mbo_valid & OBD_MD_FLCTIME) {
1757 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1758 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1759 lli->lli_ctime = body->mbo_ctime;
1761 if (body->mbo_valid & OBD_MD_FLMODE)
1762 inode->i_mode = (inode->i_mode & S_IFMT) |
1763 (body->mbo_mode & ~S_IFMT);
1764 if (body->mbo_valid & OBD_MD_FLTYPE)
1765 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1766 (body->mbo_mode & S_IFMT);
1767 LASSERT(inode->i_mode != 0);
1768 if (S_ISREG(inode->i_mode))
1769 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1770 LL_MAX_BLKSIZE_BITS);
1772 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1773 if (body->mbo_valid & OBD_MD_FLUID)
1774 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1775 if (body->mbo_valid & OBD_MD_FLGID)
1776 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1777 if (body->mbo_valid & OBD_MD_FLFLAGS)
1778 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1779 if (body->mbo_valid & OBD_MD_FLNLINK)
1780 set_nlink(inode, body->mbo_nlink);
1781 if (body->mbo_valid & OBD_MD_FLRDEV)
1782 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1784 if (body->mbo_valid & OBD_MD_FLID) {
1785 /* FID shouldn't be changed! */
1786 if (fid_is_sane(&lli->lli_fid)) {
1787 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1788 "Trying to change FID "DFID" to the "DFID", inode "DFID"(%p)\n",
1789 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1790 PFID(ll_inode2fid(inode)), inode);
1792 lli->lli_fid = body->mbo_fid1;
1796 LASSERT(fid_seq(&lli->lli_fid) != 0);
1798 if (body->mbo_valid & OBD_MD_FLSIZE) {
1799 i_size_write(inode, body->mbo_size);
1801 CDEBUG(D_VFSTRACE, "inode=" DFID ", updating i_size %llu\n",
1802 PFID(ll_inode2fid(inode)),
1803 (unsigned long long)body->mbo_size);
1805 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1806 inode->i_blocks = body->mbo_blocks;
1809 if (body->mbo_valid & OBD_MD_TSTATE) {
1810 if (body->mbo_t_state & MS_RESTORE)
1811 set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
1817 int ll_read_inode2(struct inode *inode, void *opaque)
1819 struct lustre_md *md = opaque;
1820 struct ll_inode_info *lli = ll_i2info(inode);
1823 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1824 PFID(&lli->lli_fid), inode);
1826 /* Core attributes from the MDS first. This is a new inode, and
1827 * the VFS doesn't zero times in the core inode so we have to do
1828 * it ourselves. They will be overwritten by either MDS or OST
1829 * attributes - we just need to make sure they aren't newer.
1831 LTIME_S(inode->i_mtime) = 0;
1832 LTIME_S(inode->i_atime) = 0;
1833 LTIME_S(inode->i_ctime) = 0;
1835 rc = ll_update_inode(inode, md);
1839 /* OIDEBUG(inode); */
1841 if (S_ISREG(inode->i_mode)) {
1842 struct ll_sb_info *sbi = ll_i2sbi(inode);
1844 inode->i_op = &ll_file_inode_operations;
1845 inode->i_fop = sbi->ll_fop;
1846 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1847 } else if (S_ISDIR(inode->i_mode)) {
1848 inode->i_op = &ll_dir_inode_operations;
1849 inode->i_fop = &ll_dir_operations;
1850 } else if (S_ISLNK(inode->i_mode)) {
1851 inode->i_op = &ll_fast_symlink_inode_operations;
1853 inode->i_op = &ll_special_inode_operations;
1855 init_special_inode(inode, inode->i_mode,
1862 void ll_delete_inode(struct inode *inode)
1864 struct ll_inode_info *lli = ll_i2info(inode);
1866 if (S_ISREG(inode->i_mode) && lli->lli_clob)
1867 /* discard all dirty pages before truncating them, required by
1868 * osc_extent implementation at LU-1030.
1870 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1873 truncate_inode_pages_final(&inode->i_data);
1875 LASSERTF(!inode->i_data.nrpages,
1876 "inode=" DFID "(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1877 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
1879 ll_clear_inode(inode);
1883 int ll_iocontrol(struct inode *inode, struct file *file,
1884 unsigned int cmd, unsigned long arg)
1886 struct ll_sb_info *sbi = ll_i2sbi(inode);
1887 struct ptlrpc_request *req = NULL;
1891 case FSFILT_IOC_GETFLAGS: {
1892 struct mdt_body *body;
1893 struct md_op_data *op_data;
1895 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1896 0, 0, LUSTRE_OPC_ANY,
1898 if (IS_ERR(op_data))
1899 return PTR_ERR(op_data);
1901 op_data->op_valid = OBD_MD_FLFLAGS;
1902 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1903 ll_finish_md_op_data(op_data);
1905 CERROR("%s: failure inode "DFID": rc = %d\n",
1906 sbi->ll_md_exp->exp_obd->obd_name,
1907 PFID(ll_inode2fid(inode)), rc);
1911 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1913 flags = body->mbo_flags;
1915 ptlrpc_req_finished(req);
1917 return put_user(flags, (int __user *)arg);
1919 case FSFILT_IOC_SETFLAGS: {
1920 struct md_op_data *op_data;
1921 struct cl_object *obj;
1924 if (get_user(flags, (int __user *)arg))
1927 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1928 LUSTRE_OPC_ANY, NULL);
1929 if (IS_ERR(op_data))
1930 return PTR_ERR(op_data);
1932 op_data->op_attr_flags = flags;
1933 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1934 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
1935 ll_finish_md_op_data(op_data);
1936 ptlrpc_req_finished(req);
1940 inode->i_flags = ll_ext_to_inode_flags(flags);
1942 obj = ll_i2info(inode)->lli_clob;
1946 attr = kzalloc(sizeof(*attr), GFP_NOFS);
1950 attr->ia_valid = ATTR_ATTR_FLAG;
1951 rc = cl_setattr_ost(obj, attr, flags);
1962 int ll_flush_ctx(struct inode *inode)
1964 struct ll_sb_info *sbi = ll_i2sbi(inode);
1966 CDEBUG(D_SEC, "flush context for user %d\n",
1967 from_kuid(&init_user_ns, current_uid()));
1969 obd_set_info_async(NULL, sbi->ll_md_exp,
1970 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1972 obd_set_info_async(NULL, sbi->ll_dt_exp,
1973 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1978 /* umount -f client means force down, don't save state */
1979 void ll_umount_begin(struct super_block *sb)
1981 struct ll_sb_info *sbi = ll_s2sbi(sb);
1982 struct obd_device *obd;
1983 struct obd_ioctl_data *ioc_data;
1984 wait_queue_head_t waitq;
1985 struct l_wait_info lwi;
1987 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1988 sb->s_count, atomic_read(&sb->s_active));
1990 obd = class_exp2obd(sbi->ll_md_exp);
1992 CERROR("Invalid MDC connection handle %#llx\n",
1993 sbi->ll_md_exp->exp_handle.h_cookie);
1998 obd = class_exp2obd(sbi->ll_dt_exp);
2000 CERROR("Invalid LOV connection handle %#llx\n",
2001 sbi->ll_dt_exp->exp_handle.h_cookie);
2006 ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
2008 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2009 sizeof(*ioc_data), ioc_data, NULL);
2011 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2012 sizeof(*ioc_data), ioc_data, NULL);
2017 /* Really, we'd like to wait until there are no requests outstanding,
2018 * and then continue. For now, we just periodically checking for vfs
2019 * to decrement mnt_cnt and hope to finish it within 10sec.
2021 init_waitqueue_head(&waitq);
2022 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2023 cfs_time_seconds(1), NULL, NULL);
2024 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2029 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2031 struct ll_sb_info *sbi = ll_s2sbi(sb);
2032 char *profilenm = get_profile_name(sb);
2036 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2037 read_only = *flags & MS_RDONLY;
2038 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2039 sizeof(KEY_READ_ONLY),
2040 KEY_READ_ONLY, sizeof(read_only),
2043 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2044 profilenm, read_only ?
2045 "read-only" : "read-write", err);
2050 sb->s_flags |= MS_RDONLY;
2052 sb->s_flags &= ~MS_RDONLY;
2054 if (sbi->ll_flags & LL_SBI_VERBOSE)
2055 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2056 read_only ? "read-only" : "read-write");
2062 * Cleanup the open handle that is cached on MDT-side.
2064 * For open case, the client side open handling thread may hit error
2065 * after the MDT grant the open. Under such case, the client should
2066 * send close RPC to the MDT as cleanup; otherwise, the open handle
2067 * on the MDT will be leaked there until the client umount or evicted.
2069 * In further, if someone unlinked the file, because the open handle
2070 * holds the reference on such file/object, then it will block the
2071 * subsequent threads that want to locate such object via FID.
2073 * \param[in] sb super block for this file-system
2074 * \param[in] open_req pointer to the original open request
2076 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2078 struct mdt_body *body;
2079 struct md_op_data *op_data;
2080 struct ptlrpc_request *close_req = NULL;
2081 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2083 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2084 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2088 op_data->op_fid1 = body->mbo_fid1;
2089 op_data->op_handle = body->mbo_handle;
2090 op_data->op_mod_time = get_seconds();
2091 md_close(exp, op_data, NULL, &close_req);
2092 ptlrpc_req_finished(close_req);
2093 ll_finish_md_op_data(op_data);
2096 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2097 struct super_block *sb, struct lookup_intent *it)
2099 struct ll_sb_info *sbi = NULL;
2100 struct lustre_md md = { NULL };
2103 LASSERT(*inode || sb);
2104 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2105 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2106 sbi->ll_md_exp, &md);
2111 rc = ll_update_inode(*inode, &md);
2118 * At this point server returns to client's same fid as client
2119 * generated for creating. So using ->fid1 is okay here.
2121 if (!fid_is_sane(&md.body->mbo_fid1)) {
2122 CERROR("%s: Fid is insane " DFID "\n",
2123 ll_get_fsname(sb, NULL, 0),
2124 PFID(&md.body->mbo_fid1));
2129 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2130 sbi->ll_flags & LL_SBI_32BIT_API),
2132 if (IS_ERR(*inode)) {
2133 #ifdef CONFIG_FS_POSIX_ACL
2135 posix_acl_release(md.posix_acl);
2136 md.posix_acl = NULL;
2140 CERROR("new_inode -fatal: rc %d\n", rc);
2145 /* Handling piggyback layout lock.
2146 * Layout lock can be piggybacked by getattr and open request.
2147 * The lsm can be applied to inode only if it comes with a layout lock
2148 * otherwise correct layout may be overwritten, for example:
2149 * 1. proc1: mdt returns a lsm but not granting layout
2150 * 2. layout was changed by another client
2151 * 3. proc2: refresh layout and layout lock granted
2152 * 4. proc1: to apply a stale layout
2154 if (it && it->it_lock_mode != 0) {
2155 struct lustre_handle lockh;
2156 struct ldlm_lock *lock;
2158 lockh.cookie = it->it_lock_handle;
2159 lock = ldlm_handle2lock(&lockh);
2161 if (ldlm_has_layout(lock)) {
2162 struct cl_object_conf conf;
2164 memset(&conf, 0, sizeof(conf));
2165 conf.coc_opc = OBJECT_CONF_SET;
2166 conf.coc_inode = *inode;
2167 conf.coc_lock = lock;
2168 conf.u.coc_layout = md.layout;
2169 (void)ll_layout_conf(*inode, &conf);
2171 LDLM_LOCK_PUT(lock);
2175 md_free_lustre_md(sbi->ll_md_exp, &md);
2177 if (rc != 0 && it && it->it_op & IT_OPEN)
2178 ll_open_cleanup(sb ? sb : (*inode)->i_sb, req);
2183 int ll_obd_statfs(struct inode *inode, void __user *arg)
2185 struct ll_sb_info *sbi = NULL;
2186 struct obd_export *exp;
2188 struct obd_ioctl_data *data = NULL;
2197 sbi = ll_i2sbi(inode);
2203 rc = obd_ioctl_getdata(&buf, &len, arg);
2208 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2209 !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2214 if (data->ioc_inllen1 != sizeof(__u32) ||
2215 data->ioc_inllen2 != sizeof(__u32) ||
2216 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2217 data->ioc_plen2 != sizeof(struct obd_uuid)) {
2222 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2223 if (type & LL_STATFS_LMV) {
2224 exp = sbi->ll_md_exp;
2225 } else if (type & LL_STATFS_LOV) {
2226 exp = sbi->ll_dt_exp;
2232 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2237 obd_ioctl_freedata(buf, len);
2241 int ll_process_config(struct lustre_cfg *lcfg)
2245 struct lprocfs_static_vars lvars;
2249 lprocfs_llite_init_vars(&lvars);
2251 /* The instance name contains the sb: lustre-client-aacfe000 */
2252 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2253 if (!ptr || !*(++ptr))
2255 rc = kstrtoul(ptr, 16, &x);
2259 /* This better be a real Lustre superblock! */
2260 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2262 /* Note we have not called client_common_fill_super yet, so
2263 * proc fns must be able to handle that!
2265 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2272 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2273 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2274 struct inode *i1, struct inode *i2,
2275 const char *name, size_t namelen,
2276 u32 mode, __u32 opc, void *data)
2279 /* Do not reuse namelen for something else. */
2281 return ERR_PTR(-EINVAL);
2283 if (namelen > ll_i2sbi(i1)->ll_namelen)
2284 return ERR_PTR(-ENAMETOOLONG);
2286 if (!lu_name_is_valid_2(name, namelen))
2287 return ERR_PTR(-EINVAL);
2291 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2294 return ERR_PTR(-ENOMEM);
2296 ll_i2gids(op_data->op_suppgids, i1, i2);
2297 op_data->op_fid1 = *ll_inode2fid(i1);
2298 op_data->op_default_stripe_offset = -1;
2299 if (S_ISDIR(i1->i_mode)) {
2300 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2301 if (opc == LUSTRE_OPC_MKDIR)
2302 op_data->op_default_stripe_offset =
2303 ll_i2info(i1)->lli_def_stripe_offset;
2307 op_data->op_fid2 = *ll_inode2fid(i2);
2308 if (S_ISDIR(i2->i_mode))
2309 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2311 fid_zero(&op_data->op_fid2);
2314 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2315 op_data->op_cli_flags |= CLI_HASH64;
2317 if (ll_need_32bit_api(ll_i2sbi(i1)))
2318 op_data->op_cli_flags |= CLI_API32;
2320 op_data->op_name = name;
2321 op_data->op_namelen = namelen;
2322 op_data->op_mode = mode;
2323 op_data->op_mod_time = ktime_get_real_seconds();
2324 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2325 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2326 op_data->op_cap = cfs_curproc_cap_pack();
2327 if ((opc == LUSTRE_OPC_CREATE) && name &&
2328 filename_is_volatile(name, namelen, &op_data->op_mds))
2329 op_data->op_bias |= MDS_CREATE_VOLATILE;
2331 op_data->op_mds = 0;
2332 op_data->op_data = data;
2337 void ll_finish_md_op_data(struct md_op_data *op_data)
2342 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2344 struct ll_sb_info *sbi;
2346 LASSERT(seq && dentry);
2347 sbi = ll_s2sbi(dentry->d_sb);
2349 if (sbi->ll_flags & LL_SBI_NOLCK)
2350 seq_puts(seq, ",nolock");
2352 if (sbi->ll_flags & LL_SBI_FLOCK)
2353 seq_puts(seq, ",flock");
2355 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2356 seq_puts(seq, ",localflock");
2358 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2359 seq_puts(seq, ",user_xattr");
2361 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2362 seq_puts(seq, ",lazystatfs");
2364 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2365 seq_puts(seq, ",user_fid2path");
2367 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2368 seq_puts(seq, ",always_ping");
2374 * Get obd name by cmd, and copy out to user space
2376 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2378 struct ll_sb_info *sbi = ll_i2sbi(inode);
2379 struct obd_device *obd;
2381 if (cmd == OBD_IOC_GETDTNAME)
2382 obd = class_exp2obd(sbi->ll_dt_exp);
2383 else if (cmd == OBD_IOC_GETMDNAME)
2384 obd = class_exp2obd(sbi->ll_md_exp);
2391 if (copy_to_user((void __user *)arg, obd->obd_name,
2392 strlen(obd->obd_name) + 1))
2399 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2400 * fsname will be returned in this buffer; otherwise, a static buffer will be
2401 * used to store the fsname and returned to caller.
2403 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2405 static char fsname_static[MTI_NAME_MAXLEN];
2406 struct lustre_sb_info *lsi = s2lsi(sb);
2411 /* this means the caller wants to use static buffer
2412 * and it doesn't care about race. Usually this is
2413 * in error reporting path
2415 buf = fsname_static;
2416 buflen = sizeof(fsname_static);
2419 len = strlen(lsi->lsi_lmd->lmd_profile);
2420 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2421 if (ptr && (strcmp(ptr, "-client") == 0))
2424 if (unlikely(len >= buflen))
2426 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2432 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2434 char *buf, *path = NULL;
2435 struct dentry *dentry = NULL;
2436 struct vvp_object *obj = cl_inode2vvp(page->mapping->host);
2438 /* this can be called inside spin lock so use GFP_ATOMIC. */
2439 buf = (char *)__get_free_page(GFP_ATOMIC);
2441 dentry = d_find_alias(page->mapping->host);
2443 path = dentry_path_raw(dentry, buf, PAGE_SIZE);
2447 "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2448 ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2449 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2450 PFID(&obj->vob_header.coh_lu.loh_fid),
2451 (path && !IS_ERR(path)) ? path : "", ioret);
2457 free_page((unsigned long)buf);
2460 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2461 struct lov_user_md **kbuf)
2463 struct lov_user_md lum;
2466 if (copy_from_user(&lum, md, sizeof(lum))) {
2471 lum_size = ll_lov_user_md_size(&lum);
2475 *kbuf = kzalloc(lum_size, GFP_NOFS);
2481 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2491 * Compute llite root squash state after a change of root squash
2492 * configuration setting or add/remove of a lnet nid
2494 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2496 struct root_squash_info *squash = &sbi->ll_squash;
2497 struct lnet_process_id id;
2501 /* Update norootsquash flag */
2502 down_write(&squash->rsi_sem);
2503 if (list_empty(&squash->rsi_nosquash_nids)) {
2504 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2507 * Do not apply root squash as soon as one of our NIDs is
2508 * in the nosquash_nids list
2513 while (LNetGetId(i++, &id) != -ENOENT) {
2514 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2516 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2522 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2524 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2526 up_write(&squash->rsi_sem);
2530 * Parse linkea content to extract information about a given hardlink
2532 * \param[in] ldata - Initialized linkea data
2533 * \param[in] linkno - Link identifier
2534 * \param[out] parent_fid - The entry's parent FID
2535 * \param[in] size - Entry name destination buffer
2537 * \retval 0 on success
2538 * \retval Appropriate negative error code on failure
2540 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2541 struct lu_fid *parent_fid, struct lu_name *ln)
2546 rc = linkea_init(ldata);
2550 if (linkno >= ldata->ld_leh->leh_reccount)
2551 /* beyond last link */
2554 linkea_first_entry(ldata);
2555 for (idx = 0; ldata->ld_lee; idx++) {
2556 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2561 linkea_next_entry(ldata);
2571 * Get parent FID and name of an identified link. Operation is performed for
2572 * a given link number, letting the caller iterate over linkno to list one or
2573 * all links of an entry.
2575 * \param[in] file - File descriptor against which to perform the operation
2576 * \param[in,out] arg - User-filled structure containing the linkno to operate
2577 * on and the available size. It is eventually filled with
2578 * the requested information or left untouched on error
2580 * \retval - 0 on success
2581 * \retval - Appropriate negative error code on failure
2583 int ll_getparent(struct file *file, struct getparent __user *arg)
2585 struct inode *inode = file_inode(file);
2586 struct linkea_data *ldata;
2587 struct lu_fid parent_fid;
2588 struct lu_buf buf = {
2597 if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
2598 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2601 if (get_user(name_size, &arg->gp_name_size))
2604 if (get_user(linkno, &arg->gp_linkno))
2607 if (name_size > PATH_MAX)
2610 ldata = kzalloc(sizeof(*ldata), GFP_NOFS);
2614 rc = linkea_data_new(ldata, &buf);
2618 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2619 buf.lb_len, OBD_MD_FLXATTR);
2623 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2627 if (ln.ln_namelen >= name_size) {
2632 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid))) {
2637 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen)) {
2642 if (put_user('\0', arg->gp_name + ln.ln_namelen)) {