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;
223 #ifdef HAVE_LRU_RESIZE_SUPPORT
224 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
225 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
227 #ifdef CONFIG_FS_POSIX_ACL
228 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
231 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
232 /* flag mdc connection as lightweight, only used for test
233 * purpose, use with care */
234 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
236 data->ocd_ibits_known = MDS_INODELOCK_FULL;
237 data->ocd_version = LUSTRE_VERSION_CODE;
239 if (sb->s_flags & MS_RDONLY)
240 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
241 if (sbi->ll_flags & LL_SBI_USER_XATTR)
242 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
244 #ifdef HAVE_MS_FLOCK_LOCK
245 /* force vfs to use lustre handler for flock() calls - bug 10743 */
246 sb->s_flags |= MS_FLOCK_LOCK;
248 #ifdef MS_HAS_NEW_AOPS
249 sb->s_flags |= MS_HAS_NEW_AOPS;
252 if (sbi->ll_flags & LL_SBI_FLOCK)
253 sbi->ll_fop = &ll_file_operations_flock;
254 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
255 sbi->ll_fop = &ll_file_operations;
257 sbi->ll_fop = &ll_file_operations_noflock;
260 data->ocd_connect_flags |= OBD_CONNECT_REAL;
261 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
262 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
264 data->ocd_brw_size = MD_MAX_BRW_SIZE;
266 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
268 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
269 "recovery, of which this client is not a "
270 "part. Please wait for recovery to complete,"
271 " abort, or time out.\n", md);
274 CERROR("cannot connect to %s: rc = %d\n", md, err);
278 sbi->ll_md_exp->exp_connect_data = *data;
280 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
281 LUSTRE_SEQ_METADATA);
283 CERROR("%s: Can't init metadata layer FID infrastructure, "
284 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
288 /* For mount, we only need fs info from MDT0, and also in DNE, it
289 * can make sure the client can be mounted as long as MDT0 is
291 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
292 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
293 OBD_STATFS_FOR_MDT0);
295 GOTO(out_md_fid, err);
297 /* This needs to be after statfs to ensure connect has finished.
298 * Note that "data" does NOT contain the valid connect reply.
299 * If connecting to a 1.8 server there will be no LMV device, so
300 * we can access the MDC export directly and exp_connect_flags will
301 * be non-zero, but if accessing an upgraded 2.1 server it will
302 * have the correct flags filled in.
303 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
304 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
305 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
306 valid != CLIENT_CONNECT_MDT_REQD) {
309 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
310 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
311 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
312 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
313 "feature(s) needed for correct operation "
314 "of this client (%s). Please upgrade "
315 "server or downgrade client.\n",
316 sbi->ll_md_exp->exp_obd->obd_name, buf);
317 OBD_FREE(buf, PAGE_CACHE_SIZE);
318 GOTO(out_md_fid, err = -EPROTO);
321 size = sizeof(*data);
322 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
323 KEY_CONN_DATA, &size, data);
325 CERROR("%s: Get connect data failed: rc = %d\n",
326 sbi->ll_md_exp->exp_obd->obd_name, err);
327 GOTO(out_md_fid, err);
330 LASSERT(osfs->os_bsize);
331 sb->s_blocksize = osfs->os_bsize;
332 sb->s_blocksize_bits = log2(osfs->os_bsize);
333 sb->s_magic = LL_SUPER_MAGIC;
334 sb->s_maxbytes = MAX_LFS_FILESIZE;
335 sbi->ll_namelen = osfs->os_namelen;
337 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
338 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
339 LCONSOLE_INFO("Disabling user_xattr feature because "
340 "it is not supported on the server\n");
341 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
344 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
346 sb->s_flags |= MS_POSIXACL;
348 sbi->ll_flags |= LL_SBI_ACL;
350 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
352 sb->s_flags &= ~MS_POSIXACL;
354 sbi->ll_flags &= ~LL_SBI_ACL;
357 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
358 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
359 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
360 LCONSOLE_INFO("client is set as remote by default.\n");
363 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
364 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
365 LCONSOLE_INFO("client claims to be remote, but server "
366 "rejected, forced to be local.\n");
370 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
371 sbi->ll_flags |= LL_SBI_64BIT_HASH;
373 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
374 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
376 sbi->ll_md_brw_pages = 1;
378 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
379 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
381 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
382 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
383 LCONSOLE_INFO("%s: disabling xattr cache due to "
384 "unknown maximum xattr size.\n", dt);
386 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
387 sbi->ll_xattr_cache_enabled = 1;
391 obd = class_name2obd(dt);
393 CERROR("DT %s: not setup or attached\n", dt);
394 GOTO(out_md_fid, err = -ENODEV);
397 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
398 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
399 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
400 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
401 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
402 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
403 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
404 OBD_CONNECT_MAXBYTES |
405 OBD_CONNECT_EINPROGRESS |
406 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
407 OBD_CONNECT_LAYOUTLOCK |
408 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
410 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
411 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
412 * disabled by default, because it can still be enabled on the
413 * fly via /proc. As a consequence, we still need to come to an
414 * agreement on the supported algorithms at connect time */
415 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
417 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
418 data->ocd_cksum_types = OBD_CKSUM_ADLER;
420 data->ocd_cksum_types = cksum_types_supported_client();
423 #ifdef HAVE_LRU_RESIZE_SUPPORT
424 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
426 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
427 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
429 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
430 "ocd_grant: %d\n", data->ocd_connect_flags,
431 data->ocd_version, data->ocd_grant);
433 obd->obd_upcall.onu_owner = &sbi->ll_lco;
434 obd->obd_upcall.onu_upcall = cl_ocd_update;
436 data->ocd_brw_size = DT_MAX_BRW_SIZE;
438 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
441 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
442 "recovery, of which this client is not a "
443 "part. Please wait for recovery to "
444 "complete, abort, or time out.\n", dt);
447 CERROR("%s: Cannot connect to %s: rc = %d\n",
448 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
452 sbi->ll_dt_exp->exp_connect_data = *data;
454 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
455 LUSTRE_SEQ_METADATA);
457 CERROR("%s: Can't init data layer FID infrastructure, "
458 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
462 mutex_lock(&sbi->ll_lco.lco_lock);
463 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
464 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
465 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
466 mutex_unlock(&sbi->ll_lco.lco_lock);
468 fid_zero(&sbi->ll_root_fid);
469 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
471 CERROR("cannot mds_connect: rc = %d\n", err);
472 GOTO(out_lock_cn_cb, err);
474 if (!fid_is_sane(&sbi->ll_root_fid)) {
475 CERROR("%s: Invalid root fid "DFID" during mount\n",
476 sbi->ll_md_exp->exp_obd->obd_name,
477 PFID(&sbi->ll_root_fid));
478 GOTO(out_lock_cn_cb, err = -EINVAL);
480 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
482 sb->s_op = &lustre_super_operations;
483 #if THREAD_SIZE >= 8192 /*b=17630*/
484 sb->s_export_op = &lustre_export_operations;
488 * XXX: move this to after cbd setup? */
489 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
490 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
491 valid |= OBD_MD_FLRMTPERM;
492 else if (sbi->ll_flags & LL_SBI_ACL)
493 valid |= OBD_MD_FLACL;
495 OBD_ALLOC_PTR(op_data);
497 GOTO(out_lock_cn_cb, err = -ENOMEM);
499 op_data->op_fid1 = sbi->ll_root_fid;
500 op_data->op_mode = 0;
501 op_data->op_valid = valid;
503 err = md_getattr(sbi->ll_md_exp, op_data, &request);
505 OBD_FREE_PTR(op_data);
507 CERROR("%s: md_getattr failed for root: rc = %d\n",
508 sbi->ll_md_exp->exp_obd->obd_name, err);
509 GOTO(out_lock_cn_cb, err);
512 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
513 sbi->ll_md_exp, &lmd);
515 CERROR("failed to understand root inode md: rc = %d\n", err);
516 ptlrpc_req_finished(request);
517 GOTO(out_lock_cn_cb, err);
520 LASSERT(fid_is_sane(&sbi->ll_root_fid));
521 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
522 sbi->ll_flags & LL_SBI_32BIT_API),
524 md_free_lustre_md(sbi->ll_md_exp, &lmd);
525 ptlrpc_req_finished(request);
528 #ifdef CONFIG_FS_POSIX_ACL
530 posix_acl_release(lmd.posix_acl);
531 lmd.posix_acl = NULL;
534 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
536 CERROR("lustre_lite: bad iget4 for root\n");
540 #ifdef CONFIG_FS_POSIX_ACL
541 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
542 rct_init(&sbi->ll_rct);
543 et_init(&sbi->ll_et);
547 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
548 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
549 KEY_CHECKSUM, sizeof(checksum), &checksum,
553 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
554 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
555 sbi->ll_cache, NULL);
557 sb->s_root = d_make_root(root);
558 if (sb->s_root == NULL) {
559 CERROR("%s: can't make root dentry\n",
560 ll_get_fsname(sb, NULL, 0));
561 GOTO(out_root, err = -ENOMEM);
563 #ifdef HAVE_DCACHE_LOCK
564 sb->s_root->d_op = &ll_d_ops;
567 sbi->ll_sdev_orig = sb->s_dev;
569 /* We set sb->s_dev equal on all lustre clients in order to support
570 * NFS export clustering. NFSD requires that the FSID be the same
572 /* s_dev is also used in lt_compare() to compare two fs, but that is
573 * only a node-local comparison. */
574 uuid = obd_get_uuid(sbi->ll_md_exp);
576 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
582 if (proc_lustre_fs_root != NULL) {
583 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
586 CERROR("%s: could not register mount in lprocfs: "
587 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
597 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
599 obd_disconnect(sbi->ll_dt_exp);
600 sbi->ll_dt_exp = NULL;
602 obd_fid_fini(sbi->ll_md_exp->exp_obd);
604 obd_disconnect(sbi->ll_md_exp);
605 sbi->ll_md_exp = NULL;
614 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
618 size = sizeof(*lmmsize);
619 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
620 KEY_MAX_EASIZE, &size, lmmsize);
622 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
623 sbi->ll_dt_exp->exp_obd->obd_name, rc);
628 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
629 KEY_MAX_EASIZE, &size, lmmsize);
631 CERROR("Get max mdsize error rc %d\n", rc);
637 * Get the value of the default_easize parameter.
639 * \see client_obd::cl_default_mds_easize
641 * \param[in] sbi superblock info for this filesystem
642 * \param[out] lmmsize pointer to storage location for value
644 * \retval 0 on success
645 * \retval negative negated errno on failure
647 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
652 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
653 KEY_DEFAULT_EASIZE, &size, lmmsize);
655 CERROR("Get default mdsize error rc %d\n", rc);
661 * Set the default_easize parameter to the given value.
663 * \see client_obd::cl_default_mds_easize
665 * \param[in] sbi superblock info for this filesystem
666 * \param[in] lmmsize the size to set
668 * \retval 0 on success
669 * \retval negative negated errno on failure
671 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
675 if (lmmsize < sizeof(struct lov_mds_md) ||
676 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
679 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
680 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
681 sizeof(int), &lmmsize, NULL);
686 static void ll_dump_inode(struct inode *inode)
688 struct ll_d_hlist_node *tmp;
689 int dentry_count = 0;
691 LASSERT(inode != NULL);
693 ll_d_hlist_for_each(tmp, &inode->i_dentry)
696 CERROR("%s: inode %p dump: dev=%s fid="DFID
697 " mode=%o count=%u, %d dentries\n",
698 ll_get_fsname(inode->i_sb, NULL, 0), inode,
699 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
700 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
703 void lustre_dump_dentry(struct dentry *dentry, int recur)
705 struct list_head *tmp;
708 LASSERT(dentry != NULL);
710 list_for_each(tmp, &dentry->d_subdirs)
713 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
714 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
715 dentry->d_name.len, dentry->d_name.name,
716 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
717 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
718 dentry->d_flags, dentry->d_fsdata, subdirs);
719 if (dentry->d_inode != NULL)
720 ll_dump_inode(dentry->d_inode);
725 list_for_each(tmp, &dentry->d_subdirs) {
726 struct dentry *d = list_entry(tmp, struct dentry, d_child);
727 lustre_dump_dentry(d, recur - 1);
731 static void client_common_put_super(struct super_block *sb)
733 struct ll_sb_info *sbi = ll_s2sbi(sb);
736 #ifdef CONFIG_FS_POSIX_ACL
737 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
738 et_fini(&sbi->ll_et);
739 rct_fini(&sbi->ll_rct);
745 list_del(&sbi->ll_conn_chain);
747 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
748 obd_disconnect(sbi->ll_dt_exp);
749 sbi->ll_dt_exp = NULL;
751 lprocfs_unregister_mountpoint(sbi);
753 obd_fid_fini(sbi->ll_md_exp->exp_obd);
754 obd_disconnect(sbi->ll_md_exp);
755 sbi->ll_md_exp = NULL;
760 void ll_kill_super(struct super_block *sb)
762 struct ll_sb_info *sbi;
766 if (!(sb->s_flags & MS_ACTIVE))
770 /* we need restore s_dev from changed for clustred NFS before put_super
771 * because new kernels have cached s_dev and change sb->s_dev in
772 * put_super not affected real removing devices */
774 sb->s_dev = sbi->ll_sdev_orig;
775 sbi->ll_umounting = 1;
777 /* wait running statahead threads to quit */
778 while (atomic_read(&sbi->ll_sa_running) > 0) {
779 set_current_state(TASK_UNINTERRUPTIBLE);
780 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
787 static inline int ll_set_opt(const char *opt, char *data, int fl)
789 if (strncmp(opt, data, strlen(opt)) != 0)
795 /* non-client-specific mount options are parsed in lmd_parse */
796 static int ll_options(char *options, int *flags)
799 char *s1 = options, *s2;
805 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
808 CDEBUG(D_SUPER, "next opt=%s\n", s1);
809 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
814 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
819 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
824 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
829 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
834 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
839 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
844 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
849 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
855 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
860 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
865 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
870 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
875 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
880 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
885 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
890 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
895 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
900 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
906 s2 = strchr(s1, ',');
914 void ll_lli_init(struct ll_inode_info *lli)
916 lli->lli_inode_magic = LLI_INODE_MAGIC;
918 spin_lock_init(&lli->lli_lock);
919 lli->lli_posix_acl = NULL;
920 lli->lli_remote_perms = NULL;
921 mutex_init(&lli->lli_rmtperm_mutex);
922 /* Do not set lli_fid, it has been initialized already. */
923 fid_zero(&lli->lli_pfid);
924 atomic_set(&lli->lli_open_count, 0);
925 lli->lli_rmtperm_time = 0;
926 lli->lli_mds_read_och = NULL;
927 lli->lli_mds_write_och = NULL;
928 lli->lli_mds_exec_och = NULL;
929 lli->lli_open_fd_read_count = 0;
930 lli->lli_open_fd_write_count = 0;
931 lli->lli_open_fd_exec_count = 0;
932 mutex_init(&lli->lli_och_mutex);
933 spin_lock_init(&lli->lli_agl_lock);
934 spin_lock_init(&lli->lli_layout_lock);
935 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
936 lli->lli_clob = NULL;
938 init_rwsem(&lli->lli_xattrs_list_rwsem);
939 mutex_init(&lli->lli_xattrs_enq_lock);
941 LASSERT(lli->lli_vfs_inode.i_mode != 0);
942 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
943 mutex_init(&lli->lli_readdir_mutex);
944 lli->lli_opendir_key = NULL;
946 spin_lock_init(&lli->lli_sa_lock);
947 lli->lli_opendir_pid = 0;
948 lli->lli_sa_enabled = 0;
949 lli->lli_def_stripe_offset = -1;
951 mutex_init(&lli->lli_size_mutex);
952 lli->lli_symlink_name = NULL;
953 init_rwsem(&lli->lli_trunc_sem);
954 range_lock_tree_init(&lli->lli_write_tree);
955 init_rwsem(&lli->lli_glimpse_sem);
956 lli->lli_glimpse_time = 0;
957 INIT_LIST_HEAD(&lli->lli_agl_list);
958 lli->lli_agl_index = 0;
959 lli->lli_async_rc = 0;
961 mutex_init(&lli->lli_layout_mutex);
964 static inline int ll_bdi_register(struct backing_dev_info *bdi)
966 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
968 bdi->name = "lustre";
969 return bdi_register(bdi, NULL, "lustre-%d",
970 atomic_inc_return(&ll_bdi_num));
973 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
975 struct lustre_profile *lprof = NULL;
976 struct lustre_sb_info *lsi = s2lsi(sb);
977 struct ll_sb_info *sbi;
978 char *dt = NULL, *md = NULL;
979 char *profilenm = get_profile_name(sb);
980 struct config_llog_instance *cfg;
981 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
982 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
986 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
992 try_module_get(THIS_MODULE);
994 /* client additional sb info */
995 lsi->lsi_llsbi = sbi = ll_init_sbi();
997 module_put(THIS_MODULE);
1002 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1004 GOTO(out_free, err);
1006 err = bdi_init(&lsi->lsi_bdi);
1008 GOTO(out_free, err);
1009 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1010 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1011 err = ll_bdi_register(&lsi->lsi_bdi);
1013 GOTO(out_free, err);
1015 sb->s_bdi = &lsi->lsi_bdi;
1016 #ifndef HAVE_DCACHE_LOCK
1017 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1018 sb->s_d_op = &ll_d_ops;
1021 /* Generate a string unique to this super, in case some joker tries
1022 to mount the same fs at two mount points.
1023 Use the address of the super itself.*/
1024 cfg->cfg_instance = sb;
1025 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1026 cfg->cfg_callback = class_config_llog_handler;
1027 /* set up client obds */
1028 err = lustre_process_log(sb, profilenm, cfg);
1030 GOTO(out_free, err);
1032 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1033 lprof = class_get_profile(profilenm);
1034 if (lprof == NULL) {
1035 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1036 " read from the MGS. Does that filesystem "
1037 "exist?\n", profilenm);
1038 GOTO(out_free, err = -EINVAL);
1040 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1041 lprof->lp_md, lprof->lp_dt);
1043 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1045 GOTO(out_free, err = -ENOMEM);
1046 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1048 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1050 GOTO(out_free, err = -ENOMEM);
1051 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1053 /* connections, registrations, sb setup */
1054 err = client_common_fill_super(sb, md, dt, mnt);
1056 GOTO(out_free, err);
1058 sbi->ll_client_common_fill_super_succeeded = 1;
1062 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1064 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1067 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1068 LCONSOLE_WARN("Mounted %s\n", profilenm);
1072 } /* ll_fill_super */
1074 void ll_put_super(struct super_block *sb)
1076 struct config_llog_instance cfg, params_cfg;
1077 struct obd_device *obd;
1078 struct lustre_sb_info *lsi = s2lsi(sb);
1079 struct ll_sb_info *sbi = ll_s2sbi(sb);
1080 char *profilenm = get_profile_name(sb);
1082 int next, force = 1, rc = 0;
1085 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1087 cfg.cfg_instance = sb;
1088 lustre_end_log(sb, profilenm, &cfg);
1090 params_cfg.cfg_instance = sb;
1091 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1093 if (sbi->ll_md_exp) {
1094 obd = class_exp2obd(sbi->ll_md_exp);
1096 force = obd->obd_force;
1099 /* Wait for unstable pages to be committed to stable storage */
1101 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1102 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1103 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1107 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1108 if (force == 0 && rc != -EINTR)
1109 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1112 /* We need to set force before the lov_disconnect in
1113 lustre_common_put_super, since l_d cleans up osc's as well. */
1116 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1118 obd->obd_force = force;
1122 if (sbi->ll_client_common_fill_super_succeeded) {
1123 /* Only if client_common_fill_super succeeded */
1124 client_common_put_super(sb);
1128 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1129 class_manual_cleanup(obd);
1132 if (sbi->ll_flags & LL_SBI_VERBOSE)
1133 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1136 class_del_profile(profilenm);
1138 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1139 bdi_destroy(&lsi->lsi_bdi);
1140 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1144 lsi->lsi_llsbi = NULL;
1146 lustre_common_put_super(sb);
1148 cl_env_cache_purge(~0);
1150 module_put(THIS_MODULE);
1153 } /* client_put_super */
1155 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1157 struct inode *inode = NULL;
1159 /* NOTE: we depend on atomic igrab() -bzzz */
1160 lock_res_and_lock(lock);
1161 if (lock->l_resource->lr_lvb_inode) {
1162 struct ll_inode_info * lli;
1163 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1164 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1165 inode = igrab(lock->l_resource->lr_lvb_inode);
1167 inode = lock->l_resource->lr_lvb_inode;
1168 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1169 D_WARNING, lock, "lr_lvb_inode %p is "
1170 "bogus: magic %08x",
1171 lock->l_resource->lr_lvb_inode,
1172 lli->lli_inode_magic);
1176 unlock_res_and_lock(lock);
1180 static void ll_dir_clear_lsm_md(struct inode *inode)
1182 struct ll_inode_info *lli = ll_i2info(inode);
1184 LASSERT(S_ISDIR(inode->i_mode));
1186 if (lli->lli_lsm_md != NULL) {
1187 lmv_free_memmd(lli->lli_lsm_md);
1188 lli->lli_lsm_md = NULL;
1192 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1193 const struct lu_fid *fid,
1194 struct lustre_md *md)
1196 struct ll_sb_info *sbi = ll_s2sbi(sb);
1197 struct mdt_body *body = md->body;
1198 struct inode *inode;
1202 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1203 inode = iget_locked(sb, ino);
1204 if (inode == NULL) {
1205 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1206 ll_get_fsname(sb, NULL, 0), PFID(fid));
1207 RETURN(ERR_PTR(-ENOENT));
1210 if (inode->i_state & I_NEW) {
1211 struct ll_inode_info *lli = ll_i2info(inode);
1212 struct lmv_stripe_md *lsm = md->lmv;
1214 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1215 (body->mbo_mode & S_IFMT);
1216 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1219 LTIME_S(inode->i_mtime) = 0;
1220 LTIME_S(inode->i_atime) = 0;
1221 LTIME_S(inode->i_ctime) = 0;
1224 /* initializing backing dev info. */
1225 inode->i_mapping->backing_dev_info =
1226 &s2lsi(inode->i_sb)->lsi_bdi;
1227 inode->i_op = &ll_dir_inode_operations;
1228 inode->i_fop = &ll_dir_operations;
1229 lli->lli_fid = *fid;
1232 LASSERT(lsm != NULL);
1233 /* master object FID */
1234 lli->lli_pfid = body->mbo_fid1;
1235 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1236 lli, PFID(fid), PFID(&lli->lli_pfid));
1237 unlock_new_inode(inode);
1243 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1246 struct lmv_stripe_md *lsm = md->lmv;
1249 LASSERT(lsm != NULL);
1250 /* XXX sigh, this lsm_root initialization should be in
1251 * LMV layer, but it needs ll_iget right now, so we
1252 * put this here right now. */
1253 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1254 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1255 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1256 /* Unfortunately ll_iget will call ll_update_inode,
1257 * where the initialization of slave inode is slightly
1258 * different, so it reset lsm_md to NULL to avoid
1259 * initializing lsm for slave inode. */
1260 /* For migrating inode, master stripe and master object will
1261 * be same, so we only need assign this inode */
1262 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1263 lsm->lsm_md_oinfo[i].lmo_root = inode;
1265 lsm->lsm_md_oinfo[i].lmo_root =
1266 ll_iget_anon_dir(inode->i_sb, fid, md);
1268 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1269 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1271 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1279 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1280 const struct lmv_stripe_md *lsm_md2)
1282 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1283 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1284 lsm_md1->lsm_md_master_mdt_index ==
1285 lsm_md2->lsm_md_master_mdt_index &&
1286 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1287 lsm_md1->lsm_md_layout_version ==
1288 lsm_md2->lsm_md_layout_version &&
1289 strcmp(lsm_md1->lsm_md_pool_name,
1290 lsm_md2->lsm_md_pool_name) == 0;
1293 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1295 struct ll_inode_info *lli = ll_i2info(inode);
1296 struct lmv_stripe_md *lsm = md->lmv;
1300 LASSERT(S_ISDIR(inode->i_mode));
1301 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1302 PFID(ll_inode2fid(inode)));
1304 /* no striped information from request. */
1306 if (lli->lli_lsm_md == NULL) {
1308 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1309 LMV_HASH_FLAG_MIGRATION) {
1310 /* migration is done, the temporay MIGRATE layout has
1312 CDEBUG(D_INODE, DFID" finish migration.\n",
1313 PFID(ll_inode2fid(inode)));
1314 lmv_free_memmd(lli->lli_lsm_md);
1315 lli->lli_lsm_md = NULL;
1318 /* The lustre_md from req does not include stripeEA,
1319 * see ll_md_setattr */
1324 /* set the directory layout */
1325 if (lli->lli_lsm_md == NULL) {
1327 rc = ll_init_lsm_md(inode, md);
1331 lli->lli_lsm_md = lsm;
1332 /* set lsm_md to NULL, so the following free lustre_md
1333 * will not free this lsm */
1335 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1336 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1340 /* Compare the old and new stripe information */
1341 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1342 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1345 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1346 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1347 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1348 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1349 inode, lsm, old_lsm,
1350 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1351 lsm->lsm_md_stripe_count,
1352 old_lsm->lsm_md_stripe_count,
1353 lsm->lsm_md_master_mdt_index,
1354 old_lsm->lsm_md_master_mdt_index,
1355 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1356 lsm->lsm_md_layout_version,
1357 old_lsm->lsm_md_layout_version,
1358 lsm->lsm_md_pool_name,
1359 old_lsm->lsm_md_pool_name);
1361 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1362 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1363 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1364 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1367 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1368 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1369 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1370 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1379 void ll_clear_inode(struct inode *inode)
1381 struct ll_inode_info *lli = ll_i2info(inode);
1382 struct ll_sb_info *sbi = ll_i2sbi(inode);
1385 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1386 PFID(ll_inode2fid(inode)), inode);
1388 if (S_ISDIR(inode->i_mode)) {
1389 /* these should have been cleared in ll_file_release */
1390 LASSERT(lli->lli_opendir_key == NULL);
1391 LASSERT(lli->lli_sai == NULL);
1392 LASSERT(lli->lli_opendir_pid == 0);
1395 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1397 LASSERT(!lli->lli_open_fd_write_count);
1398 LASSERT(!lli->lli_open_fd_read_count);
1399 LASSERT(!lli->lli_open_fd_exec_count);
1401 if (lli->lli_mds_write_och)
1402 ll_md_real_close(inode, FMODE_WRITE);
1403 if (lli->lli_mds_exec_och)
1404 ll_md_real_close(inode, FMODE_EXEC);
1405 if (lli->lli_mds_read_och)
1406 ll_md_real_close(inode, FMODE_READ);
1408 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1409 OBD_FREE(lli->lli_symlink_name,
1410 strlen(lli->lli_symlink_name) + 1);
1411 lli->lli_symlink_name = NULL;
1414 ll_xattr_cache_destroy(inode);
1416 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1417 LASSERT(lli->lli_posix_acl == NULL);
1418 if (lli->lli_remote_perms) {
1419 free_rmtperm_hash(lli->lli_remote_perms);
1420 lli->lli_remote_perms = NULL;
1423 #ifdef CONFIG_FS_POSIX_ACL
1424 else if (lli->lli_posix_acl) {
1425 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1426 LASSERT(lli->lli_remote_perms == NULL);
1427 posix_acl_release(lli->lli_posix_acl);
1428 lli->lli_posix_acl = NULL;
1431 lli->lli_inode_magic = LLI_INODE_DEAD;
1433 if (S_ISDIR(inode->i_mode))
1434 ll_dir_clear_lsm_md(inode);
1435 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1436 LASSERT(list_empty(&lli->lli_agl_list));
1439 * XXX This has to be done before lsm is freed below, because
1440 * cl_object still uses inode lsm.
1442 cl_inode_fini(inode);
1447 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1449 struct lustre_md md;
1450 struct inode *inode = dentry->d_inode;
1451 struct ll_sb_info *sbi = ll_i2sbi(inode);
1452 struct ptlrpc_request *request = NULL;
1456 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1457 LUSTRE_OPC_ANY, NULL);
1458 if (IS_ERR(op_data))
1459 RETURN(PTR_ERR(op_data));
1461 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1463 ptlrpc_req_finished(request);
1464 if (rc == -ENOENT) {
1466 /* Unlinked special device node? Or just a race?
1467 * Pretend we done everything. */
1468 if (!S_ISREG(inode->i_mode) &&
1469 !S_ISDIR(inode->i_mode)) {
1470 ia_valid = op_data->op_attr.ia_valid;
1471 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1472 rc = simple_setattr(dentry, &op_data->op_attr);
1473 op_data->op_attr.ia_valid = ia_valid;
1475 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1476 CERROR("md_setattr fails: rc = %d\n", rc);
1481 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1482 sbi->ll_md_exp, &md);
1484 ptlrpc_req_finished(request);
1488 ia_valid = op_data->op_attr.ia_valid;
1489 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1490 * cache is not cleared yet. */
1491 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1492 rc = simple_setattr(dentry, &op_data->op_attr);
1493 op_data->op_attr.ia_valid = ia_valid;
1495 rc = ll_update_inode(inode, &md);
1496 ptlrpc_req_finished(request);
1501 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1502 * object(s) determine the file size and mtime. Otherwise, the MDS will
1503 * keep these values until such a time that objects are allocated for it.
1504 * We do the MDS operations first, as it is checking permissions for us.
1505 * We don't to the MDS RPC if there is nothing that we want to store there,
1506 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1507 * going to do an RPC anyways.
1509 * If we are doing a truncate, we will send the mtime and ctime updates
1510 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1511 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1514 * In case of HSMimport, we only set attr on MDS.
1516 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1518 struct inode *inode = dentry->d_inode;
1519 struct ll_inode_info *lli = ll_i2info(inode);
1520 struct md_op_data *op_data = NULL;
1521 bool file_is_released = false;
1525 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1526 "valid %x, hsm_import %d\n",
1527 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1528 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1531 if (attr->ia_valid & ATTR_SIZE) {
1532 /* Check new size against VFS/VM file size limit and rlimit */
1533 rc = inode_newsize_ok(inode, attr->ia_size);
1537 /* The maximum Lustre file size is variable, based on the
1538 * OST maximum object size and number of stripes. This
1539 * needs another check in addition to the VFS check above. */
1540 if (attr->ia_size > ll_file_maxbytes(inode)) {
1541 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1542 PFID(&lli->lli_fid), attr->ia_size,
1543 ll_file_maxbytes(inode));
1547 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1550 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1551 if (attr->ia_valid & TIMES_SET_FLAGS) {
1552 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1553 !cfs_capable(CFS_CAP_FOWNER))
1557 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1558 if (attr->ia_valid & ATTR_CTIME) {
1559 attr->ia_ctime = CFS_CURRENT_TIME;
1560 attr->ia_valid |= ATTR_CTIME_SET;
1562 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1563 (attr->ia_valid & ATTR_ATIME)) {
1564 attr->ia_atime = CFS_CURRENT_TIME;
1565 attr->ia_valid |= ATTR_ATIME_SET;
1567 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1568 (attr->ia_valid & ATTR_MTIME)) {
1569 attr->ia_mtime = CFS_CURRENT_TIME;
1570 attr->ia_valid |= ATTR_MTIME_SET;
1573 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1574 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1575 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1576 cfs_time_current_sec());
1578 /* We always do an MDS RPC, even if we're only changing the size;
1579 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1581 OBD_ALLOC_PTR(op_data);
1582 if (op_data == NULL)
1585 if (!S_ISDIR(inode->i_mode)) {
1586 if (attr->ia_valid & ATTR_SIZE)
1587 inode_dio_write_done(inode);
1588 mutex_unlock(&inode->i_mutex);
1591 /* truncate on a released file must failed with -ENODATA,
1592 * so size must not be set on MDS for released file
1593 * but other attributes must be set
1595 if (S_ISREG(inode->i_mode)) {
1596 struct cl_layout cl = {
1597 .cl_is_released = false,
1603 rc = ll_layout_refresh(inode, &gen);
1607 /* XXX: the only place we need to know the layout type,
1608 * this will be removed by a later patch. -Jinshan */
1609 env = cl_env_get(&refcheck);
1611 GOTO(out, rc = PTR_ERR(env));
1613 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1614 cl_env_put(env, &refcheck);
1618 file_is_released = cl.cl_is_released;
1620 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1621 if (file_is_released) {
1622 rc = ll_layout_restore(inode, 0, attr->ia_size);
1626 file_is_released = false;
1627 ll_layout_refresh(inode, &gen);
1630 /* If we are changing file size, file content is
1631 * modified, flag it. */
1632 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1633 spin_lock(&lli->lli_lock);
1634 lli->lli_flags |= LLIF_DATA_MODIFIED;
1635 spin_unlock(&lli->lli_lock);
1636 op_data->op_bias |= MDS_DATA_MODIFIED;
1640 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1642 rc = ll_md_setattr(dentry, op_data);
1646 /* RPC to MDT is sent, cancel data modification flag */
1647 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1648 spin_lock(&lli->lli_lock);
1649 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1650 spin_unlock(&lli->lli_lock);
1653 if (!S_ISREG(inode->i_mode) || file_is_released)
1656 if (attr->ia_valid & (ATTR_SIZE |
1657 ATTR_ATIME | ATTR_ATIME_SET |
1658 ATTR_MTIME | ATTR_MTIME_SET)) {
1659 /* For truncate and utimes sending attributes to OSTs, setting
1660 * mtime/atime to the past will be performed under PW [0:EOF]
1661 * extent lock (new_size:EOF for truncate). It may seem
1662 * excessive to send mtime/atime updates to OSTs when not
1663 * setting times to past, but it is necessary due to possible
1664 * time de-synchronization between MDT inode and OST objects */
1665 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1669 if (op_data != NULL)
1670 ll_finish_md_op_data(op_data);
1672 if (!S_ISDIR(inode->i_mode)) {
1673 mutex_lock(&inode->i_mutex);
1674 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1675 inode_dio_wait(inode);
1678 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1679 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1684 int ll_setattr(struct dentry *de, struct iattr *attr)
1686 int mode = de->d_inode->i_mode;
1688 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1689 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1690 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1692 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1693 (ATTR_SIZE|ATTR_MODE)) &&
1694 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1695 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1696 !(attr->ia_mode & S_ISGID))))
1697 attr->ia_valid |= ATTR_FORCE;
1699 if ((attr->ia_valid & ATTR_MODE) &&
1701 !(attr->ia_mode & S_ISUID) &&
1702 !(attr->ia_valid & ATTR_KILL_SUID))
1703 attr->ia_valid |= ATTR_KILL_SUID;
1705 if ((attr->ia_valid & ATTR_MODE) &&
1706 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1707 !(attr->ia_mode & S_ISGID) &&
1708 !(attr->ia_valid & ATTR_KILL_SGID))
1709 attr->ia_valid |= ATTR_KILL_SGID;
1711 return ll_setattr_raw(de, attr, false);
1714 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1715 __u64 max_age, __u32 flags)
1717 struct ll_sb_info *sbi = ll_s2sbi(sb);
1718 struct obd_statfs obd_osfs;
1722 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1724 CERROR("md_statfs fails: rc = %d\n", rc);
1728 osfs->os_type = sb->s_magic;
1730 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1731 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1733 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1734 flags |= OBD_STATFS_NODELAY;
1736 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1738 CERROR("obd_statfs fails: rc = %d\n", rc);
1742 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1743 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1746 osfs->os_bsize = obd_osfs.os_bsize;
1747 osfs->os_blocks = obd_osfs.os_blocks;
1748 osfs->os_bfree = obd_osfs.os_bfree;
1749 osfs->os_bavail = obd_osfs.os_bavail;
1751 /* If we don't have as many objects free on the OST as inodes
1752 * on the MDS, we reduce the total number of inodes to
1753 * compensate, so that the "inodes in use" number is correct.
1755 if (obd_osfs.os_ffree < osfs->os_ffree) {
1756 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1758 osfs->os_ffree = obd_osfs.os_ffree;
1763 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1765 struct super_block *sb = de->d_sb;
1766 struct obd_statfs osfs;
1767 __u64 fsid = huge_encode_dev(sb->s_dev);
1770 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1771 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1773 /* Some amount of caching on the client is allowed */
1774 rc = ll_statfs_internal(sb, &osfs,
1775 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1780 statfs_unpack(sfs, &osfs);
1782 /* We need to downshift for all 32-bit kernels, because we can't
1783 * tell if the kernel is being called via sys_statfs64() or not.
1784 * Stop before overflowing f_bsize - in which case it is better
1785 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1786 if (sizeof(long) < 8) {
1787 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1790 osfs.os_blocks >>= 1;
1791 osfs.os_bfree >>= 1;
1792 osfs.os_bavail >>= 1;
1796 sfs->f_blocks = osfs.os_blocks;
1797 sfs->f_bfree = osfs.os_bfree;
1798 sfs->f_bavail = osfs.os_bavail;
1799 sfs->f_fsid.val[0] = (__u32)fsid;
1800 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1804 void ll_inode_size_lock(struct inode *inode)
1806 struct ll_inode_info *lli;
1808 LASSERT(!S_ISDIR(inode->i_mode));
1810 lli = ll_i2info(inode);
1811 mutex_lock(&lli->lli_size_mutex);
1814 void ll_inode_size_unlock(struct inode *inode)
1816 struct ll_inode_info *lli;
1818 lli = ll_i2info(inode);
1819 mutex_unlock(&lli->lli_size_mutex);
1822 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1824 struct ll_inode_info *lli = ll_i2info(inode);
1825 struct mdt_body *body = md->body;
1826 struct ll_sb_info *sbi = ll_i2sbi(inode);
1828 if (body->mbo_valid & OBD_MD_FLEASIZE)
1829 cl_file_inode_init(inode, md);
1831 if (S_ISDIR(inode->i_mode)) {
1834 rc = ll_update_lsm_md(inode, md);
1839 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1840 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1841 ll_update_remote_perm(inode, md->remote_perm);
1843 #ifdef CONFIG_FS_POSIX_ACL
1844 else if (body->mbo_valid & OBD_MD_FLACL) {
1845 spin_lock(&lli->lli_lock);
1846 if (lli->lli_posix_acl)
1847 posix_acl_release(lli->lli_posix_acl);
1848 lli->lli_posix_acl = md->posix_acl;
1849 spin_unlock(&lli->lli_lock);
1852 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1853 sbi->ll_flags & LL_SBI_32BIT_API);
1854 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1856 if (body->mbo_valid & OBD_MD_FLATIME) {
1857 if (body->mbo_atime > LTIME_S(inode->i_atime))
1858 LTIME_S(inode->i_atime) = body->mbo_atime;
1859 lli->lli_atime = body->mbo_atime;
1862 if (body->mbo_valid & OBD_MD_FLMTIME) {
1863 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1864 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1865 "to "LPU64"\n", inode->i_ino,
1866 LTIME_S(inode->i_mtime), body->mbo_mtime);
1867 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1869 lli->lli_mtime = body->mbo_mtime;
1872 if (body->mbo_valid & OBD_MD_FLCTIME) {
1873 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1874 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1875 lli->lli_ctime = body->mbo_ctime;
1878 if (body->mbo_valid & OBD_MD_FLMODE)
1879 inode->i_mode = (inode->i_mode & S_IFMT) |
1880 (body->mbo_mode & ~S_IFMT);
1882 if (body->mbo_valid & OBD_MD_FLTYPE)
1883 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1884 (body->mbo_mode & S_IFMT);
1886 LASSERT(inode->i_mode != 0);
1887 if (S_ISREG(inode->i_mode))
1888 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1889 LL_MAX_BLKSIZE_BITS);
1891 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1893 if (body->mbo_valid & OBD_MD_FLUID)
1894 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1895 if (body->mbo_valid & OBD_MD_FLGID)
1896 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1897 if (body->mbo_valid & OBD_MD_FLFLAGS)
1898 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1899 if (body->mbo_valid & OBD_MD_FLNLINK)
1900 set_nlink(inode, body->mbo_nlink);
1901 if (body->mbo_valid & OBD_MD_FLRDEV)
1902 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1904 if (body->mbo_valid & OBD_MD_FLID) {
1905 /* FID shouldn't be changed! */
1906 if (fid_is_sane(&lli->lli_fid)) {
1907 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1908 "Trying to change FID "DFID
1909 " to the "DFID", inode "DFID"(%p)\n",
1910 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1911 PFID(ll_inode2fid(inode)), inode);
1913 lli->lli_fid = body->mbo_fid1;
1917 LASSERT(fid_seq(&lli->lli_fid) != 0);
1919 if (body->mbo_valid & OBD_MD_FLSIZE) {
1920 i_size_write(inode, body->mbo_size);
1922 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1923 PFID(ll_inode2fid(inode)),
1924 (unsigned long long)body->mbo_size);
1926 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1927 inode->i_blocks = body->mbo_blocks;
1930 if (body->mbo_valid & OBD_MD_TSTATE) {
1931 if (body->mbo_t_state & MS_RESTORE)
1932 lli->lli_flags |= LLIF_FILE_RESTORING;
1938 int ll_read_inode2(struct inode *inode, void *opaque)
1940 struct lustre_md *md = opaque;
1941 struct ll_inode_info *lli = ll_i2info(inode);
1945 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1946 PFID(&lli->lli_fid), inode);
1948 /* Core attributes from the MDS first. This is a new inode, and
1949 * the VFS doesn't zero times in the core inode so we have to do
1950 * it ourselves. They will be overwritten by either MDS or OST
1951 * attributes - we just need to make sure they aren't newer. */
1952 LTIME_S(inode->i_mtime) = 0;
1953 LTIME_S(inode->i_atime) = 0;
1954 LTIME_S(inode->i_ctime) = 0;
1956 rc = ll_update_inode(inode, md);
1960 /* OIDEBUG(inode); */
1962 /* initializing backing dev info. */
1963 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1966 if (S_ISREG(inode->i_mode)) {
1967 struct ll_sb_info *sbi = ll_i2sbi(inode);
1968 inode->i_op = &ll_file_inode_operations;
1969 inode->i_fop = sbi->ll_fop;
1970 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1972 } else if (S_ISDIR(inode->i_mode)) {
1973 inode->i_op = &ll_dir_inode_operations;
1974 inode->i_fop = &ll_dir_operations;
1976 } else if (S_ISLNK(inode->i_mode)) {
1977 inode->i_op = &ll_fast_symlink_inode_operations;
1980 inode->i_op = &ll_special_inode_operations;
1982 init_special_inode(inode, inode->i_mode,
1991 void ll_delete_inode(struct inode *inode)
1993 struct ll_inode_info *lli = ll_i2info(inode);
1996 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1997 /* It is last chance to write out dirty pages,
1998 * otherwise we may lose data while umount */
1999 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2001 truncate_inode_pages_final(&inode->i_data);
2003 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2004 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2005 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2007 #ifdef HAVE_SBOPS_EVICT_INODE
2008 ll_clear_inode(inode);
2015 int ll_iocontrol(struct inode *inode, struct file *file,
2016 unsigned int cmd, unsigned long arg)
2018 struct ll_sb_info *sbi = ll_i2sbi(inode);
2019 struct ptlrpc_request *req = NULL;
2024 case FSFILT_IOC_GETFLAGS: {
2025 struct mdt_body *body;
2026 struct md_op_data *op_data;
2028 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2029 0, 0, LUSTRE_OPC_ANY,
2031 if (IS_ERR(op_data))
2032 RETURN(PTR_ERR(op_data));
2034 op_data->op_valid = OBD_MD_FLFLAGS;
2035 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2036 ll_finish_md_op_data(op_data);
2038 CERROR("%s: failure inode "DFID": rc = %d\n",
2039 sbi->ll_md_exp->exp_obd->obd_name,
2040 PFID(ll_inode2fid(inode)), rc);
2044 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2046 flags = body->mbo_flags;
2048 ptlrpc_req_finished(req);
2050 RETURN(put_user(flags, (int __user *)arg));
2052 case FSFILT_IOC_SETFLAGS: {
2054 struct md_op_data *op_data;
2055 struct cl_object *obj;
2057 if (get_user(flags, (int __user *)arg))
2060 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2061 LUSTRE_OPC_ANY, NULL);
2062 if (IS_ERR(op_data))
2063 RETURN(PTR_ERR(op_data));
2065 op_data->op_attr_flags = flags;
2066 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2067 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2068 ll_finish_md_op_data(op_data);
2069 ptlrpc_req_finished(req);
2073 inode->i_flags = ll_ext_to_inode_flags(flags);
2075 obj = ll_i2info(inode)->lli_clob;
2079 OBD_ALLOC_PTR(attr);
2083 attr->ia_valid = ATTR_ATTR_FLAG;
2084 rc = cl_setattr_ost(obj, attr, flags);
2096 int ll_flush_ctx(struct inode *inode)
2098 struct ll_sb_info *sbi = ll_i2sbi(inode);
2100 CDEBUG(D_SEC, "flush context for user %d\n",
2101 from_kuid(&init_user_ns, current_uid()));
2103 obd_set_info_async(NULL, sbi->ll_md_exp,
2104 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2106 obd_set_info_async(NULL, sbi->ll_dt_exp,
2107 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2112 /* umount -f client means force down, don't save state */
2113 void ll_umount_begin(struct super_block *sb)
2115 struct ll_sb_info *sbi = ll_s2sbi(sb);
2116 struct obd_device *obd;
2117 struct obd_ioctl_data *ioc_data;
2120 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2121 sb->s_count, atomic_read(&sb->s_active));
2123 obd = class_exp2obd(sbi->ll_md_exp);
2125 CERROR("Invalid MDC connection handle "LPX64"\n",
2126 sbi->ll_md_exp->exp_handle.h_cookie);
2132 obd = class_exp2obd(sbi->ll_dt_exp);
2134 CERROR("Invalid LOV connection handle "LPX64"\n",
2135 sbi->ll_dt_exp->exp_handle.h_cookie);
2141 OBD_ALLOC_PTR(ioc_data);
2143 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2144 sizeof *ioc_data, ioc_data, NULL);
2146 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2147 sizeof *ioc_data, ioc_data, NULL);
2149 OBD_FREE_PTR(ioc_data);
2152 /* Really, we'd like to wait until there are no requests outstanding,
2153 * and then continue. For now, we just invalidate the requests,
2154 * schedule() and sleep one second if needed, and hope.
2160 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2162 struct ll_sb_info *sbi = ll_s2sbi(sb);
2163 char *profilenm = get_profile_name(sb);
2167 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2168 read_only = *flags & MS_RDONLY;
2169 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2170 sizeof(KEY_READ_ONLY),
2171 KEY_READ_ONLY, sizeof(read_only),
2174 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2175 profilenm, read_only ?
2176 "read-only" : "read-write", err);
2181 sb->s_flags |= MS_RDONLY;
2183 sb->s_flags &= ~MS_RDONLY;
2185 if (sbi->ll_flags & LL_SBI_VERBOSE)
2186 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2187 read_only ? "read-only" : "read-write");
2193 * Cleanup the open handle that is cached on MDT-side.
2195 * For open case, the client side open handling thread may hit error
2196 * after the MDT grant the open. Under such case, the client should
2197 * send close RPC to the MDT as cleanup; otherwise, the open handle
2198 * on the MDT will be leaked there until the client umount or evicted.
2200 * In further, if someone unlinked the file, because the open handle
2201 * holds the reference on such file/object, then it will block the
2202 * subsequent threads that want to locate such object via FID.
2204 * \param[in] sb super block for this file-system
2205 * \param[in] open_req pointer to the original open request
2207 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2209 struct mdt_body *body;
2210 struct md_op_data *op_data;
2211 struct ptlrpc_request *close_req = NULL;
2212 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2215 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2216 OBD_ALLOC_PTR(op_data);
2217 if (op_data == NULL) {
2218 CWARN("%s: cannot allocate op_data to release open handle for "
2220 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2225 op_data->op_fid1 = body->mbo_fid1;
2226 op_data->op_handle = body->mbo_handle;
2227 op_data->op_mod_time = cfs_time_current_sec();
2228 md_close(exp, op_data, NULL, &close_req);
2229 ptlrpc_req_finished(close_req);
2230 ll_finish_md_op_data(op_data);
2235 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2236 struct super_block *sb, struct lookup_intent *it)
2238 struct ll_sb_info *sbi = NULL;
2239 struct lustre_md md = { NULL };
2243 LASSERT(*inode || sb);
2244 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2245 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2246 sbi->ll_md_exp, &md);
2251 rc = ll_update_inode(*inode, &md);
2255 LASSERT(sb != NULL);
2258 * At this point server returns to client's same fid as client
2259 * generated for creating. So using ->fid1 is okay here.
2261 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2263 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2264 sbi->ll_flags & LL_SBI_32BIT_API),
2266 if (IS_ERR(*inode)) {
2267 #ifdef CONFIG_FS_POSIX_ACL
2269 posix_acl_release(md.posix_acl);
2270 md.posix_acl = NULL;
2273 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2275 CERROR("new_inode -fatal: rc %d\n", rc);
2280 /* Handling piggyback layout lock.
2281 * Layout lock can be piggybacked by getattr and open request.
2282 * The lsm can be applied to inode only if it comes with a layout lock
2283 * otherwise correct layout may be overwritten, for example:
2284 * 1. proc1: mdt returns a lsm but not granting layout
2285 * 2. layout was changed by another client
2286 * 3. proc2: refresh layout and layout lock granted
2287 * 4. proc1: to apply a stale layout */
2288 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2289 struct lustre_handle lockh;
2290 struct ldlm_lock *lock;
2292 lockh.cookie = it->d.lustre.it_lock_handle;
2293 lock = ldlm_handle2lock(&lockh);
2294 LASSERT(lock != NULL);
2295 if (ldlm_has_layout(lock)) {
2296 struct cl_object_conf conf;
2298 memset(&conf, 0, sizeof(conf));
2299 conf.coc_opc = OBJECT_CONF_SET;
2300 conf.coc_inode = *inode;
2301 conf.coc_lock = lock;
2302 conf.u.coc_layout = md.layout;
2303 (void)ll_layout_conf(*inode, &conf);
2305 LDLM_LOCK_PUT(lock);
2311 md_free_lustre_md(sbi->ll_md_exp, &md);
2314 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2315 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2320 int ll_obd_statfs(struct inode *inode, void __user *arg)
2322 struct ll_sb_info *sbi = NULL;
2323 struct obd_export *exp;
2325 struct obd_ioctl_data *data = NULL;
2327 __u32 __user flags; /* not user, but obd_iocontrol is abused */
2330 if (!inode || !(sbi = ll_i2sbi(inode)))
2331 GOTO(out_statfs, rc = -EINVAL);
2333 rc = obd_ioctl_getdata(&buf, &len, arg);
2335 GOTO(out_statfs, rc);
2338 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2339 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2340 GOTO(out_statfs, rc = -EINVAL);
2342 if (data->ioc_inllen1 != sizeof(__u32) ||
2343 data->ioc_inllen2 != sizeof(__u32) ||
2344 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2345 data->ioc_plen2 != sizeof(struct obd_uuid))
2346 GOTO(out_statfs, rc = -EINVAL);
2348 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2349 if (type & LL_STATFS_LMV)
2350 exp = sbi->ll_md_exp;
2351 else if (type & LL_STATFS_LOV)
2352 exp = sbi->ll_dt_exp;
2354 GOTO(out_statfs, rc = -ENODEV);
2356 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2357 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2359 GOTO(out_statfs, rc);
2362 obd_ioctl_freedata(buf, len);
2366 int ll_process_config(struct lustre_cfg *lcfg)
2368 struct super_block *sb;
2373 /* The instance name contains the sb: lustre-client-aacfe000 */
2374 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2375 if (!ptr || !*(++ptr))
2377 if (sscanf(ptr, "%lx", &x) != 1)
2379 sb = (struct super_block *)x;
2380 /* This better be a real Lustre superblock! */
2381 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2383 /* Note we have not called client_common_fill_super yet, so
2384 proc fns must be able to handle that! */
2385 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2392 /* this function prepares md_op_data hint for passing it down to MD stack. */
2393 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2394 struct inode *i1, struct inode *i2,
2395 const char *name, size_t namelen,
2396 __u32 mode, __u32 opc, void *data)
2398 LASSERT(i1 != NULL);
2401 /* Do not reuse namelen for something else. */
2403 return ERR_PTR(-EINVAL);
2405 if (namelen > ll_i2sbi(i1)->ll_namelen)
2406 return ERR_PTR(-ENAMETOOLONG);
2408 if (!lu_name_is_valid_2(name, namelen))
2409 return ERR_PTR(-EINVAL);
2412 if (op_data == NULL)
2413 OBD_ALLOC_PTR(op_data);
2415 if (op_data == NULL)
2416 return ERR_PTR(-ENOMEM);
2418 ll_i2gids(op_data->op_suppgids, i1, i2);
2419 op_data->op_fid1 = *ll_inode2fid(i1);
2420 op_data->op_default_stripe_offset = -1;
2421 if (S_ISDIR(i1->i_mode)) {
2422 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2423 if (opc == LUSTRE_OPC_MKDIR)
2424 op_data->op_default_stripe_offset =
2425 ll_i2info(i1)->lli_def_stripe_offset;
2429 op_data->op_fid2 = *ll_inode2fid(i2);
2430 if (S_ISDIR(i2->i_mode))
2431 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2433 fid_zero(&op_data->op_fid2);
2436 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2437 op_data->op_cli_flags |= CLI_HASH64;
2439 if (ll_need_32bit_api(ll_i2sbi(i1)))
2440 op_data->op_cli_flags |= CLI_API32;
2442 op_data->op_name = name;
2443 op_data->op_namelen = namelen;
2444 op_data->op_mode = mode;
2445 op_data->op_mod_time = cfs_time_current_sec();
2446 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2447 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2448 op_data->op_cap = cfs_curproc_cap_pack();
2449 op_data->op_bias = 0;
2450 op_data->op_cli_flags = 0;
2451 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2452 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2453 op_data->op_bias |= MDS_CREATE_VOLATILE;
2455 op_data->op_mds = 0;
2457 op_data->op_data = data;
2459 /* When called by ll_setattr_raw, file is i1. */
2460 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2461 op_data->op_bias |= MDS_DATA_MODIFIED;
2466 void ll_finish_md_op_data(struct md_op_data *op_data)
2468 OBD_FREE_PTR(op_data);
2471 #ifdef HAVE_SUPEROPS_USE_DENTRY
2472 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2474 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2477 struct ll_sb_info *sbi;
2479 #ifdef HAVE_SUPEROPS_USE_DENTRY
2480 LASSERT((seq != NULL) && (dentry != NULL));
2481 sbi = ll_s2sbi(dentry->d_sb);
2483 LASSERT((seq != NULL) && (vfs != NULL));
2484 sbi = ll_s2sbi(vfs->mnt_sb);
2487 if (sbi->ll_flags & LL_SBI_NOLCK)
2488 seq_puts(seq, ",nolock");
2490 if (sbi->ll_flags & LL_SBI_FLOCK)
2491 seq_puts(seq, ",flock");
2493 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2494 seq_puts(seq, ",localflock");
2496 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2497 seq_puts(seq, ",user_xattr");
2499 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2500 seq_puts(seq, ",lazystatfs");
2502 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2503 seq_puts(seq, ",user_fid2path");
2509 * Get obd name by cmd, and copy out to user space
2511 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2513 struct ll_sb_info *sbi = ll_i2sbi(inode);
2514 struct obd_device *obd;
2517 if (cmd == OBD_IOC_GETDTNAME)
2518 obd = class_exp2obd(sbi->ll_dt_exp);
2519 else if (cmd == OBD_IOC_GETMDNAME)
2520 obd = class_exp2obd(sbi->ll_md_exp);
2527 if (copy_to_user((void __user *)arg, obd->obd_name,
2528 strlen(obd->obd_name) + 1))
2535 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2536 * fsname will be returned in this buffer; otherwise, a static buffer will be
2537 * used to store the fsname and returned to caller.
2539 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2541 static char fsname_static[MTI_NAME_MAXLEN];
2542 struct lustre_sb_info *lsi = s2lsi(sb);
2547 /* this means the caller wants to use static buffer
2548 * and it doesn't care about race. Usually this is
2549 * in error reporting path */
2550 buf = fsname_static;
2551 buflen = sizeof(fsname_static);
2554 len = strlen(lsi->lsi_lmd->lmd_profile);
2555 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2556 if (ptr && (strcmp(ptr, "-client") == 0))
2559 if (unlikely(len >= buflen))
2561 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2567 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2574 p.mnt = current->fs->root.mnt;
2576 path = d_path(&p, buf, bufsize);
2581 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2583 char *buf, *path = NULL;
2584 struct dentry *dentry = NULL;
2585 struct inode *inode = page->mapping->host;
2587 /* this can be called inside spin lock so use GFP_ATOMIC. */
2588 buf = (char *)__get_free_page(GFP_ATOMIC);
2590 dentry = d_find_alias(page->mapping->host);
2592 path = ll_d_path(dentry, buf, PAGE_SIZE);
2596 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2597 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2598 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2599 PFID(ll_inode2fid(inode)),
2600 (path && !IS_ERR(path)) ? path : "", ioret);
2606 free_page((unsigned long)buf);
2609 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2610 struct lov_user_md **kbuf)
2612 struct lov_user_md lum;
2616 if (copy_from_user(&lum, md, sizeof(lum)))
2619 lum_size = ll_lov_user_md_size(&lum);
2623 OBD_ALLOC(*kbuf, lum_size);
2627 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2628 OBD_FREE(*kbuf, lum_size);
2636 * Compute llite root squash state after a change of root squash
2637 * configuration setting or add/remove of a lnet nid
2639 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2641 struct root_squash_info *squash = &sbi->ll_squash;
2644 lnet_process_id_t id;
2646 /* Update norootsquash flag */
2647 down_write(&squash->rsi_sem);
2648 if (list_empty(&squash->rsi_nosquash_nids))
2649 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2651 /* Do not apply root squash as soon as one of our NIDs is
2652 * in the nosquash_nids list */
2655 while (LNetGetId(i++, &id) != -ENOENT) {
2656 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2658 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2664 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2666 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2668 up_write(&squash->rsi_sem);
2672 * Parse linkea content to extract information about a given hardlink
2674 * \param[in] ldata - Initialized linkea data
2675 * \param[in] linkno - Link identifier
2676 * \param[out] parent_fid - The entry's parent FID
2677 * \param[out] ln - Entry name destination buffer
2679 * \retval 0 on success
2680 * \retval Appropriate negative error code on failure
2682 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2683 struct lu_fid *parent_fid, struct lu_name *ln)
2689 rc = linkea_init(ldata);
2693 if (linkno >= ldata->ld_leh->leh_reccount)
2694 /* beyond last link */
2697 linkea_first_entry(ldata);
2698 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2699 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2704 linkea_next_entry(ldata);
2714 * Get parent FID and name of an identified link. Operation is performed for
2715 * a given link number, letting the caller iterate over linkno to list one or
2716 * all links of an entry.
2718 * \param[in] file - File descriptor against which to perform the operation
2719 * \param[in,out] arg - User-filled structure containing the linkno to operate
2720 * on and the available size. It is eventually filled with
2721 * the requested information or left untouched on error
2723 * \retval - 0 on success
2724 * \retval - Appropriate negative error code on failure
2726 int ll_getparent(struct file *file, struct getparent __user *arg)
2728 struct dentry *dentry = file->f_path.dentry;
2729 struct inode *inode = dentry->d_inode;
2730 struct linkea_data *ldata;
2731 struct lu_buf buf = LU_BUF_NULL;
2733 struct lu_fid parent_fid;
2740 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2741 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2744 if (get_user(name_size, &arg->gp_name_size))
2747 if (get_user(linkno, &arg->gp_linkno))
2750 if (name_size > PATH_MAX)
2753 OBD_ALLOC(ldata, sizeof(*ldata));
2757 rc = linkea_data_new(ldata, &buf);
2759 GOTO(ldata_free, rc);
2761 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2765 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2769 if (ln.ln_namelen >= name_size)
2770 GOTO(lb_free, rc = -EOVERFLOW);
2772 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2773 GOTO(lb_free, rc = -EFAULT);
2775 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2776 GOTO(lb_free, rc = -EFAULT);
2778 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2779 GOTO(lb_free, rc = -EFAULT);
2784 OBD_FREE(ldata, sizeof(*ldata));