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;
67 #define log2(n) ffz(~(n))
70 static struct ll_sb_info *ll_init_sbi(void)
72 struct ll_sb_info *sbi = NULL;
74 unsigned long lru_page_max;
84 spin_lock_init(&sbi->ll_lock);
85 mutex_init(&sbi->ll_lco.lco_lock);
86 spin_lock_init(&sbi->ll_pp_extent_lock);
87 spin_lock_init(&sbi->ll_process_lock);
88 sbi->ll_rw_stats_on = 0;
91 pages = si.totalram - si.totalhigh;
92 lru_page_max = pages / 2;
94 /* initialize ll_cache data */
95 sbi->ll_cache = cl_cache_init(lru_page_max);
96 if (sbi->ll_cache == NULL) {
97 OBD_FREE(sbi, sizeof(*sbi));
101 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
102 SBI_DEFAULT_READAHEAD_MAX);
103 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
104 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
105 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
106 INIT_LIST_HEAD(&sbi->ll_conn_chain);
107 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
109 ll_generate_random_uuid(uuid);
110 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
111 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
113 sbi->ll_flags |= LL_SBI_VERBOSE;
114 #ifdef ENABLE_CHECKSUM
115 sbi->ll_flags |= LL_SBI_CHECKSUM;
118 #ifdef HAVE_LRU_RESIZE_SUPPORT
119 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
122 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
123 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
129 /* metadata statahead is enabled by default */
130 sbi->ll_sa_max = LL_SA_RPC_DEF;
131 atomic_set(&sbi->ll_sa_total, 0);
132 atomic_set(&sbi->ll_sa_wrong, 0);
133 atomic_set(&sbi->ll_sa_running, 0);
134 atomic_set(&sbi->ll_agl_total, 0);
135 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
138 sbi->ll_squash.rsi_uid = 0;
139 sbi->ll_squash.rsi_gid = 0;
140 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
141 init_rwsem(&sbi->ll_squash.rsi_sem);
146 static void ll_free_sbi(struct super_block *sb)
148 struct ll_sb_info *sbi = ll_s2sbi(sb);
152 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
153 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
154 if (sbi->ll_cache != NULL) {
155 cl_cache_decref(sbi->ll_cache);
156 sbi->ll_cache = NULL;
158 OBD_FREE(sbi, sizeof(*sbi));
163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
164 struct vfsmount *mnt)
166 struct inode *root = NULL;
167 struct ll_sb_info *sbi = ll_s2sbi(sb);
168 struct obd_device *obd;
169 struct obd_statfs *osfs = NULL;
170 struct ptlrpc_request *request = NULL;
171 struct obd_connect_data *data = NULL;
172 struct obd_uuid *uuid;
173 struct md_op_data *op_data;
174 struct lustre_md lmd;
176 int size, err, checksum;
179 obd = class_name2obd(md);
181 CERROR("MD %s: not setup or attached\n", md);
195 /* indicate the features supported by this client */
196 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
197 OBD_CONNECT_ATTRFID |
198 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
199 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
200 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
201 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
202 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
203 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
204 OBD_CONNECT_EINPROGRESS |
205 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
206 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
207 OBD_CONNECT_MAX_EASIZE |
208 OBD_CONNECT_FLOCK_DEAD |
209 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
210 OBD_CONNECT_OPEN_BY_FID |
211 OBD_CONNECT_DIR_STRIPE |
212 OBD_CONNECT_BULK_MBITS;
214 #ifdef HAVE_LRU_RESIZE_SUPPORT
215 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
216 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
218 #ifdef CONFIG_FS_POSIX_ACL
219 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
222 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
223 /* flag mdc connection as lightweight, only used for test
224 * purpose, use with care */
225 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
227 data->ocd_ibits_known = MDS_INODELOCK_FULL;
228 data->ocd_version = LUSTRE_VERSION_CODE;
230 if (sb->s_flags & MS_RDONLY)
231 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
232 if (sbi->ll_flags & LL_SBI_USER_XATTR)
233 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
235 #ifdef HAVE_MS_FLOCK_LOCK
236 /* force vfs to use lustre handler for flock() calls - bug 10743 */
237 sb->s_flags |= MS_FLOCK_LOCK;
239 #ifdef MS_HAS_NEW_AOPS
240 sb->s_flags |= MS_HAS_NEW_AOPS;
243 if (sbi->ll_flags & LL_SBI_FLOCK)
244 sbi->ll_fop = &ll_file_operations_flock;
245 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
246 sbi->ll_fop = &ll_file_operations;
248 sbi->ll_fop = &ll_file_operations_noflock;
251 data->ocd_connect_flags |= OBD_CONNECT_REAL;
252 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
253 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
255 data->ocd_brw_size = MD_MAX_BRW_SIZE;
257 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
259 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
260 "recovery, of which this client is not a "
261 "part. Please wait for recovery to complete,"
262 " abort, or time out.\n", md);
265 CERROR("cannot connect to %s: rc = %d\n", md, err);
269 sbi->ll_md_exp->exp_connect_data = *data;
271 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
272 LUSTRE_SEQ_METADATA);
274 CERROR("%s: Can't init metadata layer FID infrastructure, "
275 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
279 /* For mount, we only need fs info from MDT0, and also in DNE, it
280 * can make sure the client can be mounted as long as MDT0 is
282 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
283 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
284 OBD_STATFS_FOR_MDT0);
286 GOTO(out_md_fid, err);
288 /* This needs to be after statfs to ensure connect has finished.
289 * Note that "data" does NOT contain the valid connect reply.
290 * If connecting to a 1.8 server there will be no LMV device, so
291 * we can access the MDC export directly and exp_connect_flags will
292 * be non-zero, but if accessing an upgraded 2.1 server it will
293 * have the correct flags filled in.
294 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
295 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
296 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
297 valid != CLIENT_CONNECT_MDT_REQD) {
300 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
301 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
302 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
303 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
304 "feature(s) needed for correct operation "
305 "of this client (%s). Please upgrade "
306 "server or downgrade client.\n",
307 sbi->ll_md_exp->exp_obd->obd_name, buf);
308 OBD_FREE(buf, PAGE_CACHE_SIZE);
309 GOTO(out_md_fid, err = -EPROTO);
312 size = sizeof(*data);
313 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
314 KEY_CONN_DATA, &size, data);
316 CERROR("%s: Get connect data failed: rc = %d\n",
317 sbi->ll_md_exp->exp_obd->obd_name, err);
318 GOTO(out_md_fid, err);
321 LASSERT(osfs->os_bsize);
322 sb->s_blocksize = osfs->os_bsize;
323 sb->s_blocksize_bits = log2(osfs->os_bsize);
324 sb->s_magic = LL_SUPER_MAGIC;
325 sb->s_maxbytes = MAX_LFS_FILESIZE;
326 sbi->ll_namelen = osfs->os_namelen;
328 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
329 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
330 LCONSOLE_INFO("Disabling user_xattr feature because "
331 "it is not supported on the server\n");
332 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
335 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
337 sb->s_flags |= MS_POSIXACL;
339 sbi->ll_flags |= LL_SBI_ACL;
341 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
343 sb->s_flags &= ~MS_POSIXACL;
345 sbi->ll_flags &= ~LL_SBI_ACL;
348 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
349 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
350 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
351 LCONSOLE_INFO("client is set as remote by default.\n");
354 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
355 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
356 LCONSOLE_INFO("client claims to be remote, but server "
357 "rejected, forced to be local.\n");
361 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
362 sbi->ll_flags |= LL_SBI_64BIT_HASH;
364 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
365 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
367 sbi->ll_md_brw_pages = 1;
369 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
370 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
372 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
373 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
374 LCONSOLE_INFO("%s: disabling xattr cache due to "
375 "unknown maximum xattr size.\n", dt);
377 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
378 sbi->ll_xattr_cache_enabled = 1;
382 obd = class_name2obd(dt);
384 CERROR("DT %s: not setup or attached\n", dt);
385 GOTO(out_md_fid, err = -ENODEV);
388 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
389 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
390 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
391 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
392 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
393 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
394 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
395 OBD_CONNECT_MAXBYTES |
396 OBD_CONNECT_EINPROGRESS |
397 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
398 OBD_CONNECT_LAYOUTLOCK |
399 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
400 OBD_CONNECT_BULK_MBITS;
402 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
403 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
404 * disabled by default, because it can still be enabled on the
405 * fly via /proc. As a consequence, we still need to come to an
406 * agreement on the supported algorithms at connect time */
407 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
409 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
410 data->ocd_cksum_types = OBD_CKSUM_ADLER;
412 data->ocd_cksum_types = cksum_types_supported_client();
415 #ifdef HAVE_LRU_RESIZE_SUPPORT
416 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
418 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
419 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
421 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
422 "ocd_grant: %d\n", data->ocd_connect_flags,
423 data->ocd_version, data->ocd_grant);
425 obd->obd_upcall.onu_owner = &sbi->ll_lco;
426 obd->obd_upcall.onu_upcall = cl_ocd_update;
428 data->ocd_brw_size = DT_MAX_BRW_SIZE;
430 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
433 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
434 "recovery, of which this client is not a "
435 "part. Please wait for recovery to "
436 "complete, abort, or time out.\n", dt);
439 CERROR("%s: Cannot connect to %s: rc = %d\n",
440 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
444 sbi->ll_dt_exp->exp_connect_data = *data;
446 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
447 LUSTRE_SEQ_METADATA);
449 CERROR("%s: Can't init data layer FID infrastructure, "
450 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
454 mutex_lock(&sbi->ll_lco.lco_lock);
455 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
456 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
457 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
458 mutex_unlock(&sbi->ll_lco.lco_lock);
460 fid_zero(&sbi->ll_root_fid);
461 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
463 CERROR("cannot mds_connect: rc = %d\n", err);
464 GOTO(out_lock_cn_cb, err);
466 if (!fid_is_sane(&sbi->ll_root_fid)) {
467 CERROR("%s: Invalid root fid "DFID" during mount\n",
468 sbi->ll_md_exp->exp_obd->obd_name,
469 PFID(&sbi->ll_root_fid));
470 GOTO(out_lock_cn_cb, err = -EINVAL);
472 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
474 sb->s_op = &lustre_super_operations;
475 #if THREAD_SIZE >= 8192 /*b=17630*/
476 sb->s_export_op = &lustre_export_operations;
480 * XXX: move this to after cbd setup? */
481 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
482 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
483 valid |= OBD_MD_FLRMTPERM;
484 else if (sbi->ll_flags & LL_SBI_ACL)
485 valid |= OBD_MD_FLACL;
487 OBD_ALLOC_PTR(op_data);
489 GOTO(out_lock_cn_cb, err = -ENOMEM);
491 op_data->op_fid1 = sbi->ll_root_fid;
492 op_data->op_mode = 0;
493 op_data->op_valid = valid;
495 err = md_getattr(sbi->ll_md_exp, op_data, &request);
497 OBD_FREE_PTR(op_data);
499 CERROR("%s: md_getattr failed for root: rc = %d\n",
500 sbi->ll_md_exp->exp_obd->obd_name, err);
501 GOTO(out_lock_cn_cb, err);
504 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
505 sbi->ll_md_exp, &lmd);
507 CERROR("failed to understand root inode md: rc = %d\n", err);
508 ptlrpc_req_finished(request);
509 GOTO(out_lock_cn_cb, err);
512 LASSERT(fid_is_sane(&sbi->ll_root_fid));
513 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
514 sbi->ll_flags & LL_SBI_32BIT_API),
516 md_free_lustre_md(sbi->ll_md_exp, &lmd);
517 ptlrpc_req_finished(request);
520 #ifdef CONFIG_FS_POSIX_ACL
522 posix_acl_release(lmd.posix_acl);
523 lmd.posix_acl = NULL;
526 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
528 CERROR("lustre_lite: bad iget4 for root\n");
532 #ifdef CONFIG_FS_POSIX_ACL
533 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
534 rct_init(&sbi->ll_rct);
535 et_init(&sbi->ll_et);
539 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
540 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
541 KEY_CHECKSUM, sizeof(checksum), &checksum,
545 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
546 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
547 sbi->ll_cache, NULL);
549 sb->s_root = d_make_root(root);
550 if (sb->s_root == NULL) {
551 CERROR("%s: can't make root dentry\n",
552 ll_get_fsname(sb, NULL, 0));
553 GOTO(out_root, err = -ENOMEM);
555 #ifdef HAVE_DCACHE_LOCK
556 sb->s_root->d_op = &ll_d_ops;
559 sbi->ll_sdev_orig = sb->s_dev;
561 /* We set sb->s_dev equal on all lustre clients in order to support
562 * NFS export clustering. NFSD requires that the FSID be the same
564 /* s_dev is also used in lt_compare() to compare two fs, but that is
565 * only a node-local comparison. */
566 uuid = obd_get_uuid(sbi->ll_md_exp);
568 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
574 if (proc_lustre_fs_root != NULL) {
575 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
578 CERROR("%s: could not register mount in lprocfs: "
579 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
589 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
591 obd_disconnect(sbi->ll_dt_exp);
592 sbi->ll_dt_exp = NULL;
594 obd_fid_fini(sbi->ll_md_exp->exp_obd);
596 obd_disconnect(sbi->ll_md_exp);
597 sbi->ll_md_exp = NULL;
606 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
610 size = sizeof(*lmmsize);
611 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
612 KEY_MAX_EASIZE, &size, lmmsize);
614 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
615 sbi->ll_dt_exp->exp_obd->obd_name, rc);
620 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
621 KEY_MAX_EASIZE, &size, lmmsize);
623 CERROR("Get max mdsize error rc %d\n", rc);
629 * Get the value of the default_easize parameter.
631 * \see client_obd::cl_default_mds_easize
633 * \param[in] sbi superblock info for this filesystem
634 * \param[out] lmmsize pointer to storage location for value
636 * \retval 0 on success
637 * \retval negative negated errno on failure
639 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
644 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
645 KEY_DEFAULT_EASIZE, &size, lmmsize);
647 CERROR("Get default mdsize error rc %d\n", rc);
653 * Set the default_easize parameter to the given value.
655 * \see client_obd::cl_default_mds_easize
657 * \param[in] sbi superblock info for this filesystem
658 * \param[in] lmmsize the size to set
660 * \retval 0 on success
661 * \retval negative negated errno on failure
663 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
667 if (lmmsize < sizeof(struct lov_mds_md) ||
668 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
671 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
672 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
673 sizeof(int), &lmmsize, NULL);
678 static void ll_dump_inode(struct inode *inode)
680 struct ll_d_hlist_node *tmp;
681 int dentry_count = 0;
683 LASSERT(inode != NULL);
685 ll_d_hlist_for_each(tmp, &inode->i_dentry)
688 CERROR("%s: inode %p dump: dev=%s fid="DFID
689 " mode=%o count=%u, %d dentries\n",
690 ll_get_fsname(inode->i_sb, NULL, 0), inode,
691 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
692 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
695 void lustre_dump_dentry(struct dentry *dentry, int recur)
697 struct list_head *tmp;
700 LASSERT(dentry != NULL);
702 list_for_each(tmp, &dentry->d_subdirs)
705 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
706 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
707 dentry->d_name.len, dentry->d_name.name,
708 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
709 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
710 dentry->d_flags, dentry->d_fsdata, subdirs);
711 if (dentry->d_inode != NULL)
712 ll_dump_inode(dentry->d_inode);
717 list_for_each(tmp, &dentry->d_subdirs) {
718 struct dentry *d = list_entry(tmp, struct dentry, d_child);
719 lustre_dump_dentry(d, recur - 1);
723 static void client_common_put_super(struct super_block *sb)
725 struct ll_sb_info *sbi = ll_s2sbi(sb);
728 #ifdef CONFIG_FS_POSIX_ACL
729 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
730 et_fini(&sbi->ll_et);
731 rct_fini(&sbi->ll_rct);
737 list_del(&sbi->ll_conn_chain);
739 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
740 obd_disconnect(sbi->ll_dt_exp);
741 sbi->ll_dt_exp = NULL;
743 lprocfs_unregister_mountpoint(sbi);
745 obd_fid_fini(sbi->ll_md_exp->exp_obd);
746 obd_disconnect(sbi->ll_md_exp);
747 sbi->ll_md_exp = NULL;
752 void ll_kill_super(struct super_block *sb)
754 struct ll_sb_info *sbi;
758 if (!(sb->s_flags & MS_ACTIVE))
762 /* we need restore s_dev from changed for clustred NFS before put_super
763 * because new kernels have cached s_dev and change sb->s_dev in
764 * put_super not affected real removing devices */
766 sb->s_dev = sbi->ll_sdev_orig;
767 sbi->ll_umounting = 1;
769 /* wait running statahead threads to quit */
770 while (atomic_read(&sbi->ll_sa_running) > 0) {
771 set_current_state(TASK_UNINTERRUPTIBLE);
772 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
779 static inline int ll_set_opt(const char *opt, char *data, int fl)
781 if (strncmp(opt, data, strlen(opt)) != 0)
787 /* non-client-specific mount options are parsed in lmd_parse */
788 static int ll_options(char *options, int *flags)
791 char *s1 = options, *s2;
797 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
800 CDEBUG(D_SUPER, "next opt=%s\n", s1);
801 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
806 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
811 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
816 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
821 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
826 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
831 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
836 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
841 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
847 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
852 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
857 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
862 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
867 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
872 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
877 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
882 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
887 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
892 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
898 s2 = strchr(s1, ',');
906 void ll_lli_init(struct ll_inode_info *lli)
908 lli->lli_inode_magic = LLI_INODE_MAGIC;
910 spin_lock_init(&lli->lli_lock);
911 lli->lli_posix_acl = NULL;
912 lli->lli_remote_perms = NULL;
913 mutex_init(&lli->lli_rmtperm_mutex);
914 /* Do not set lli_fid, it has been initialized already. */
915 fid_zero(&lli->lli_pfid);
916 atomic_set(&lli->lli_open_count, 0);
917 lli->lli_rmtperm_time = 0;
918 lli->lli_mds_read_och = NULL;
919 lli->lli_mds_write_och = NULL;
920 lli->lli_mds_exec_och = NULL;
921 lli->lli_open_fd_read_count = 0;
922 lli->lli_open_fd_write_count = 0;
923 lli->lli_open_fd_exec_count = 0;
924 mutex_init(&lli->lli_och_mutex);
925 spin_lock_init(&lli->lli_agl_lock);
926 spin_lock_init(&lli->lli_layout_lock);
927 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
928 lli->lli_clob = NULL;
930 init_rwsem(&lli->lli_xattrs_list_rwsem);
931 mutex_init(&lli->lli_xattrs_enq_lock);
933 LASSERT(lli->lli_vfs_inode.i_mode != 0);
934 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
935 mutex_init(&lli->lli_readdir_mutex);
936 lli->lli_opendir_key = NULL;
938 spin_lock_init(&lli->lli_sa_lock);
939 lli->lli_opendir_pid = 0;
940 lli->lli_sa_enabled = 0;
941 lli->lli_def_stripe_offset = -1;
943 mutex_init(&lli->lli_size_mutex);
944 lli->lli_symlink_name = NULL;
945 init_rwsem(&lli->lli_trunc_sem);
946 range_lock_tree_init(&lli->lli_write_tree);
947 init_rwsem(&lli->lli_glimpse_sem);
948 lli->lli_glimpse_time = 0;
949 INIT_LIST_HEAD(&lli->lli_agl_list);
950 lli->lli_agl_index = 0;
951 lli->lli_async_rc = 0;
953 mutex_init(&lli->lli_layout_mutex);
956 static inline int ll_bdi_register(struct backing_dev_info *bdi)
958 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
960 bdi->name = "lustre";
961 return bdi_register(bdi, NULL, "lustre-%d",
962 atomic_inc_return(&ll_bdi_num));
965 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
967 struct lustre_profile *lprof = NULL;
968 struct lustre_sb_info *lsi = s2lsi(sb);
969 struct ll_sb_info *sbi;
970 char *dt = NULL, *md = NULL;
971 char *profilenm = get_profile_name(sb);
972 struct config_llog_instance *cfg;
973 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
974 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
978 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
984 try_module_get(THIS_MODULE);
986 /* client additional sb info */
987 lsi->lsi_llsbi = sbi = ll_init_sbi();
989 module_put(THIS_MODULE);
994 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
998 err = bdi_init(&lsi->lsi_bdi);
1000 GOTO(out_free, err);
1001 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1002 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1003 err = ll_bdi_register(&lsi->lsi_bdi);
1005 GOTO(out_free, err);
1007 sb->s_bdi = &lsi->lsi_bdi;
1008 #ifndef HAVE_DCACHE_LOCK
1009 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1010 sb->s_d_op = &ll_d_ops;
1013 /* Generate a string unique to this super, in case some joker tries
1014 to mount the same fs at two mount points.
1015 Use the address of the super itself.*/
1016 cfg->cfg_instance = sb;
1017 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1018 cfg->cfg_callback = class_config_llog_handler;
1019 /* set up client obds */
1020 err = lustre_process_log(sb, profilenm, cfg);
1022 GOTO(out_free, err);
1024 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1025 lprof = class_get_profile(profilenm);
1026 if (lprof == NULL) {
1027 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1028 " read from the MGS. Does that filesystem "
1029 "exist?\n", profilenm);
1030 GOTO(out_free, err = -EINVAL);
1032 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1033 lprof->lp_md, lprof->lp_dt);
1035 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1037 GOTO(out_free, err = -ENOMEM);
1038 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1040 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1042 GOTO(out_free, err = -ENOMEM);
1043 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1045 /* connections, registrations, sb setup */
1046 err = client_common_fill_super(sb, md, dt, mnt);
1048 GOTO(out_free, err);
1050 sbi->ll_client_common_fill_super_succeeded = 1;
1054 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1056 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1059 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1060 LCONSOLE_WARN("Mounted %s\n", profilenm);
1064 } /* ll_fill_super */
1066 void ll_put_super(struct super_block *sb)
1068 struct config_llog_instance cfg, params_cfg;
1069 struct obd_device *obd;
1070 struct lustre_sb_info *lsi = s2lsi(sb);
1071 struct ll_sb_info *sbi = ll_s2sbi(sb);
1072 char *profilenm = get_profile_name(sb);
1074 int next, force = 1, rc = 0;
1077 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1079 cfg.cfg_instance = sb;
1080 lustre_end_log(sb, profilenm, &cfg);
1082 params_cfg.cfg_instance = sb;
1083 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1085 if (sbi->ll_md_exp) {
1086 obd = class_exp2obd(sbi->ll_md_exp);
1088 force = obd->obd_force;
1091 /* Wait for unstable pages to be committed to stable storage */
1093 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1094 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1095 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1099 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1100 if (force == 0 && rc != -EINTR)
1101 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1104 /* We need to set force before the lov_disconnect in
1105 lustre_common_put_super, since l_d cleans up osc's as well. */
1108 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1110 obd->obd_force = force;
1114 if (sbi->ll_client_common_fill_super_succeeded) {
1115 /* Only if client_common_fill_super succeeded */
1116 client_common_put_super(sb);
1120 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1121 class_manual_cleanup(obd);
1124 if (sbi->ll_flags & LL_SBI_VERBOSE)
1125 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1128 class_del_profile(profilenm);
1130 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1131 bdi_destroy(&lsi->lsi_bdi);
1132 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1136 lsi->lsi_llsbi = NULL;
1138 lustre_common_put_super(sb);
1140 cl_env_cache_purge(~0);
1142 module_put(THIS_MODULE);
1145 } /* client_put_super */
1147 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1149 struct inode *inode = NULL;
1151 /* NOTE: we depend on atomic igrab() -bzzz */
1152 lock_res_and_lock(lock);
1153 if (lock->l_resource->lr_lvb_inode) {
1154 struct ll_inode_info * lli;
1155 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1156 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1157 inode = igrab(lock->l_resource->lr_lvb_inode);
1159 inode = lock->l_resource->lr_lvb_inode;
1160 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1161 D_WARNING, lock, "lr_lvb_inode %p is "
1162 "bogus: magic %08x",
1163 lock->l_resource->lr_lvb_inode,
1164 lli->lli_inode_magic);
1168 unlock_res_and_lock(lock);
1172 static void ll_dir_clear_lsm_md(struct inode *inode)
1174 struct ll_inode_info *lli = ll_i2info(inode);
1176 LASSERT(S_ISDIR(inode->i_mode));
1178 if (lli->lli_lsm_md != NULL) {
1179 lmv_free_memmd(lli->lli_lsm_md);
1180 lli->lli_lsm_md = NULL;
1184 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1185 const struct lu_fid *fid,
1186 struct lustre_md *md)
1188 struct ll_sb_info *sbi = ll_s2sbi(sb);
1189 struct mdt_body *body = md->body;
1190 struct inode *inode;
1194 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1195 inode = iget_locked(sb, ino);
1196 if (inode == NULL) {
1197 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1198 ll_get_fsname(sb, NULL, 0), PFID(fid));
1199 RETURN(ERR_PTR(-ENOENT));
1202 if (inode->i_state & I_NEW) {
1203 struct ll_inode_info *lli = ll_i2info(inode);
1204 struct lmv_stripe_md *lsm = md->lmv;
1206 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1207 (body->mbo_mode & S_IFMT);
1208 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1211 LTIME_S(inode->i_mtime) = 0;
1212 LTIME_S(inode->i_atime) = 0;
1213 LTIME_S(inode->i_ctime) = 0;
1216 /* initializing backing dev info. */
1217 inode->i_mapping->backing_dev_info =
1218 &s2lsi(inode->i_sb)->lsi_bdi;
1219 inode->i_op = &ll_dir_inode_operations;
1220 inode->i_fop = &ll_dir_operations;
1221 lli->lli_fid = *fid;
1224 LASSERT(lsm != NULL);
1225 /* master object FID */
1226 lli->lli_pfid = body->mbo_fid1;
1227 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1228 lli, PFID(fid), PFID(&lli->lli_pfid));
1229 unlock_new_inode(inode);
1235 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1238 struct lmv_stripe_md *lsm = md->lmv;
1241 LASSERT(lsm != NULL);
1242 /* XXX sigh, this lsm_root initialization should be in
1243 * LMV layer, but it needs ll_iget right now, so we
1244 * put this here right now. */
1245 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1246 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1247 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1248 /* Unfortunately ll_iget will call ll_update_inode,
1249 * where the initialization of slave inode is slightly
1250 * different, so it reset lsm_md to NULL to avoid
1251 * initializing lsm for slave inode. */
1252 /* For migrating inode, master stripe and master object will
1253 * be same, so we only need assign this inode */
1254 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1255 lsm->lsm_md_oinfo[i].lmo_root = inode;
1257 lsm->lsm_md_oinfo[i].lmo_root =
1258 ll_iget_anon_dir(inode->i_sb, fid, md);
1260 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1261 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1263 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1271 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1272 const struct lmv_stripe_md *lsm_md2)
1274 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1275 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1276 lsm_md1->lsm_md_master_mdt_index ==
1277 lsm_md2->lsm_md_master_mdt_index &&
1278 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1279 lsm_md1->lsm_md_layout_version ==
1280 lsm_md2->lsm_md_layout_version &&
1281 strcmp(lsm_md1->lsm_md_pool_name,
1282 lsm_md2->lsm_md_pool_name) == 0;
1285 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1287 struct ll_inode_info *lli = ll_i2info(inode);
1288 struct lmv_stripe_md *lsm = md->lmv;
1292 LASSERT(S_ISDIR(inode->i_mode));
1293 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1294 PFID(ll_inode2fid(inode)));
1296 /* no striped information from request. */
1298 if (lli->lli_lsm_md == NULL) {
1300 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1301 LMV_HASH_FLAG_MIGRATION) {
1302 /* migration is done, the temporay MIGRATE layout has
1304 CDEBUG(D_INODE, DFID" finish migration.\n",
1305 PFID(ll_inode2fid(inode)));
1306 lmv_free_memmd(lli->lli_lsm_md);
1307 lli->lli_lsm_md = NULL;
1310 /* The lustre_md from req does not include stripeEA,
1311 * see ll_md_setattr */
1316 /* set the directory layout */
1317 if (lli->lli_lsm_md == NULL) {
1318 struct cl_attr *attr;
1320 rc = ll_init_lsm_md(inode, md);
1324 lli->lli_lsm_md = lsm;
1326 OBD_ALLOC_PTR(attr);
1330 /* validate the lsm */
1331 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1332 ll_md_blocking_ast);
1338 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1339 md->body->mbo_nlink = attr->cat_nlink;
1340 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1341 md->body->mbo_size = attr->cat_size;
1342 if (md->body->mbo_valid & OBD_MD_FLATIME)
1343 md->body->mbo_atime = attr->cat_atime;
1344 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1345 md->body->mbo_ctime = attr->cat_ctime;
1346 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1347 md->body->mbo_mtime = attr->cat_mtime;
1352 /* set lsm_md to NULL, so the following free lustre_md
1353 * will not free this lsm */
1355 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1356 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1360 /* Compare the old and new stripe information */
1361 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1362 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1365 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1366 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1367 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1368 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1369 inode, lsm, old_lsm,
1370 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1371 lsm->lsm_md_stripe_count,
1372 old_lsm->lsm_md_stripe_count,
1373 lsm->lsm_md_master_mdt_index,
1374 old_lsm->lsm_md_master_mdt_index,
1375 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1376 lsm->lsm_md_layout_version,
1377 old_lsm->lsm_md_layout_version,
1378 lsm->lsm_md_pool_name,
1379 old_lsm->lsm_md_pool_name);
1381 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1382 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1383 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1384 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1387 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1388 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1389 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1390 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1399 void ll_clear_inode(struct inode *inode)
1401 struct ll_inode_info *lli = ll_i2info(inode);
1402 struct ll_sb_info *sbi = ll_i2sbi(inode);
1405 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1406 PFID(ll_inode2fid(inode)), inode);
1408 if (S_ISDIR(inode->i_mode)) {
1409 /* these should have been cleared in ll_file_release */
1410 LASSERT(lli->lli_opendir_key == NULL);
1411 LASSERT(lli->lli_sai == NULL);
1412 LASSERT(lli->lli_opendir_pid == 0);
1415 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1417 LASSERT(!lli->lli_open_fd_write_count);
1418 LASSERT(!lli->lli_open_fd_read_count);
1419 LASSERT(!lli->lli_open_fd_exec_count);
1421 if (lli->lli_mds_write_och)
1422 ll_md_real_close(inode, FMODE_WRITE);
1423 if (lli->lli_mds_exec_och)
1424 ll_md_real_close(inode, FMODE_EXEC);
1425 if (lli->lli_mds_read_och)
1426 ll_md_real_close(inode, FMODE_READ);
1428 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1429 OBD_FREE(lli->lli_symlink_name,
1430 strlen(lli->lli_symlink_name) + 1);
1431 lli->lli_symlink_name = NULL;
1434 ll_xattr_cache_destroy(inode);
1436 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1437 LASSERT(lli->lli_posix_acl == NULL);
1438 if (lli->lli_remote_perms) {
1439 free_rmtperm_hash(lli->lli_remote_perms);
1440 lli->lli_remote_perms = NULL;
1443 #ifdef CONFIG_FS_POSIX_ACL
1444 else if (lli->lli_posix_acl) {
1445 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1446 LASSERT(lli->lli_remote_perms == NULL);
1447 posix_acl_release(lli->lli_posix_acl);
1448 lli->lli_posix_acl = NULL;
1451 lli->lli_inode_magic = LLI_INODE_DEAD;
1453 if (S_ISDIR(inode->i_mode))
1454 ll_dir_clear_lsm_md(inode);
1455 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1456 LASSERT(list_empty(&lli->lli_agl_list));
1459 * XXX This has to be done before lsm is freed below, because
1460 * cl_object still uses inode lsm.
1462 cl_inode_fini(inode);
1467 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1469 struct lustre_md md;
1470 struct inode *inode = dentry->d_inode;
1471 struct ll_sb_info *sbi = ll_i2sbi(inode);
1472 struct ptlrpc_request *request = NULL;
1476 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1477 LUSTRE_OPC_ANY, NULL);
1478 if (IS_ERR(op_data))
1479 RETURN(PTR_ERR(op_data));
1481 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1483 ptlrpc_req_finished(request);
1484 if (rc == -ENOENT) {
1486 /* Unlinked special device node? Or just a race?
1487 * Pretend we done everything. */
1488 if (!S_ISREG(inode->i_mode) &&
1489 !S_ISDIR(inode->i_mode)) {
1490 ia_valid = op_data->op_attr.ia_valid;
1491 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1492 rc = simple_setattr(dentry, &op_data->op_attr);
1493 op_data->op_attr.ia_valid = ia_valid;
1495 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1496 CERROR("md_setattr fails: rc = %d\n", rc);
1501 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1502 sbi->ll_md_exp, &md);
1504 ptlrpc_req_finished(request);
1508 ia_valid = op_data->op_attr.ia_valid;
1509 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1510 * cache is not cleared yet. */
1511 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1512 if (S_ISREG(inode->i_mode))
1513 mutex_lock(&inode->i_mutex);
1514 rc = simple_setattr(dentry, &op_data->op_attr);
1515 if (S_ISREG(inode->i_mode))
1516 mutex_unlock(&inode->i_mutex);
1517 op_data->op_attr.ia_valid = ia_valid;
1519 rc = ll_update_inode(inode, &md);
1520 ptlrpc_req_finished(request);
1525 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1526 * object(s) determine the file size and mtime. Otherwise, the MDS will
1527 * keep these values until such a time that objects are allocated for it.
1528 * We do the MDS operations first, as it is checking permissions for us.
1529 * We don't to the MDS RPC if there is nothing that we want to store there,
1530 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1531 * going to do an RPC anyways.
1533 * If we are doing a truncate, we will send the mtime and ctime updates
1534 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1535 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1538 * In case of HSMimport, we only set attr on MDS.
1540 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1542 struct inode *inode = dentry->d_inode;
1543 struct ll_inode_info *lli = ll_i2info(inode);
1544 struct md_op_data *op_data = NULL;
1548 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1549 "valid %x, hsm_import %d\n",
1550 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1551 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1554 if (attr->ia_valid & ATTR_SIZE) {
1555 /* Check new size against VFS/VM file size limit and rlimit */
1556 rc = inode_newsize_ok(inode, attr->ia_size);
1560 /* The maximum Lustre file size is variable, based on the
1561 * OST maximum object size and number of stripes. This
1562 * needs another check in addition to the VFS check above. */
1563 if (attr->ia_size > ll_file_maxbytes(inode)) {
1564 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1565 PFID(&lli->lli_fid), attr->ia_size,
1566 ll_file_maxbytes(inode));
1570 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1573 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1574 if (attr->ia_valid & TIMES_SET_FLAGS) {
1575 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1576 !cfs_capable(CFS_CAP_FOWNER))
1580 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1581 if (attr->ia_valid & ATTR_CTIME) {
1582 attr->ia_ctime = CFS_CURRENT_TIME;
1583 attr->ia_valid |= ATTR_CTIME_SET;
1585 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1586 (attr->ia_valid & ATTR_ATIME)) {
1587 attr->ia_atime = CFS_CURRENT_TIME;
1588 attr->ia_valid |= ATTR_ATIME_SET;
1590 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1591 (attr->ia_valid & ATTR_MTIME)) {
1592 attr->ia_mtime = CFS_CURRENT_TIME;
1593 attr->ia_valid |= ATTR_MTIME_SET;
1596 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1597 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1598 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1599 cfs_time_current_sec());
1601 if (S_ISREG(inode->i_mode)) {
1602 if (attr->ia_valid & ATTR_SIZE)
1603 inode_dio_write_done(inode);
1604 mutex_unlock(&inode->i_mutex);
1607 /* We always do an MDS RPC, even if we're only changing the size;
1608 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1610 OBD_ALLOC_PTR(op_data);
1611 if (op_data == NULL)
1612 GOTO(out, rc = -ENOMEM);
1614 op_data->op_attr = *attr;
1616 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1617 /* If we are changing file size, file content is
1618 * modified, flag it. */
1619 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1620 op_data->op_bias |= MDS_DATA_MODIFIED;
1621 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1624 rc = ll_md_setattr(dentry, op_data);
1628 if (!S_ISREG(inode->i_mode) || hsm_import)
1631 if (attr->ia_valid & (ATTR_SIZE |
1632 ATTR_ATIME | ATTR_ATIME_SET |
1633 ATTR_MTIME | ATTR_MTIME_SET)) {
1634 /* For truncate and utimes sending attributes to OSTs, setting
1635 * mtime/atime to the past will be performed under PW [0:EOF]
1636 * extent lock (new_size:EOF for truncate). It may seem
1637 * excessive to send mtime/atime updates to OSTs when not
1638 * setting times to past, but it is necessary due to possible
1639 * time de-synchronization between MDT inode and OST objects */
1640 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1643 /* If the file was restored, it needs to set dirty flag.
1645 * We've already sent MDS_DATA_MODIFIED flag in
1646 * ll_md_setattr() for truncate. However, the MDT refuses to
1647 * set the HS_DIRTY flag on released files, so we have to set
1648 * it again if the file has been restored. Please check how
1649 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1651 * Please notice that if the file is not released, the previous
1652 * MDS_DATA_MODIFIED has taken effect and usually
1653 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1654 * This way we can save an RPC for common open + trunc
1656 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1657 struct hsm_state_set hss = {
1658 .hss_valid = HSS_SETMASK,
1659 .hss_setmask = HS_DIRTY,
1663 rc2 = ll_hsm_state_set(inode, &hss);
1664 /* truncate and write can happen at the same time, so that
1665 * the file can be set modified even though the file is not
1666 * restored from released state, and ll_hsm_state_set() is
1667 * not applicable for the file, and rc2 < 0 is normal in this
1670 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1671 PFID(ll_inode2fid(inode)), rc2);
1676 if (op_data != NULL)
1677 ll_finish_md_op_data(op_data);
1679 if (S_ISREG(inode->i_mode)) {
1680 mutex_lock(&inode->i_mutex);
1681 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1682 inode_dio_wait(inode);
1685 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1686 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1691 int ll_setattr(struct dentry *de, struct iattr *attr)
1693 int mode = de->d_inode->i_mode;
1695 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1696 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1697 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1699 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1700 (ATTR_SIZE|ATTR_MODE)) &&
1701 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1702 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1703 !(attr->ia_mode & S_ISGID))))
1704 attr->ia_valid |= ATTR_FORCE;
1706 if ((attr->ia_valid & ATTR_MODE) &&
1708 !(attr->ia_mode & S_ISUID) &&
1709 !(attr->ia_valid & ATTR_KILL_SUID))
1710 attr->ia_valid |= ATTR_KILL_SUID;
1712 if ((attr->ia_valid & ATTR_MODE) &&
1713 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1714 !(attr->ia_mode & S_ISGID) &&
1715 !(attr->ia_valid & ATTR_KILL_SGID))
1716 attr->ia_valid |= ATTR_KILL_SGID;
1718 return ll_setattr_raw(de, attr, false);
1721 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1722 __u64 max_age, __u32 flags)
1724 struct ll_sb_info *sbi = ll_s2sbi(sb);
1725 struct obd_statfs obd_osfs;
1729 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1731 CERROR("md_statfs fails: rc = %d\n", rc);
1735 osfs->os_type = sb->s_magic;
1737 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1738 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1740 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1741 flags |= OBD_STATFS_NODELAY;
1743 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1745 CERROR("obd_statfs fails: rc = %d\n", rc);
1749 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1750 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1753 osfs->os_bsize = obd_osfs.os_bsize;
1754 osfs->os_blocks = obd_osfs.os_blocks;
1755 osfs->os_bfree = obd_osfs.os_bfree;
1756 osfs->os_bavail = obd_osfs.os_bavail;
1758 /* If we don't have as many objects free on the OST as inodes
1759 * on the MDS, we reduce the total number of inodes to
1760 * compensate, so that the "inodes in use" number is correct.
1762 if (obd_osfs.os_ffree < osfs->os_ffree) {
1763 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1765 osfs->os_ffree = obd_osfs.os_ffree;
1770 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1772 struct super_block *sb = de->d_sb;
1773 struct obd_statfs osfs;
1774 __u64 fsid = huge_encode_dev(sb->s_dev);
1777 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1778 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1780 /* Some amount of caching on the client is allowed */
1781 rc = ll_statfs_internal(sb, &osfs,
1782 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1787 statfs_unpack(sfs, &osfs);
1789 /* We need to downshift for all 32-bit kernels, because we can't
1790 * tell if the kernel is being called via sys_statfs64() or not.
1791 * Stop before overflowing f_bsize - in which case it is better
1792 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1793 if (sizeof(long) < 8) {
1794 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1797 osfs.os_blocks >>= 1;
1798 osfs.os_bfree >>= 1;
1799 osfs.os_bavail >>= 1;
1803 sfs->f_blocks = osfs.os_blocks;
1804 sfs->f_bfree = osfs.os_bfree;
1805 sfs->f_bavail = osfs.os_bavail;
1806 sfs->f_fsid.val[0] = (__u32)fsid;
1807 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1811 void ll_inode_size_lock(struct inode *inode)
1813 struct ll_inode_info *lli;
1815 LASSERT(!S_ISDIR(inode->i_mode));
1817 lli = ll_i2info(inode);
1818 mutex_lock(&lli->lli_size_mutex);
1821 void ll_inode_size_unlock(struct inode *inode)
1823 struct ll_inode_info *lli;
1825 lli = ll_i2info(inode);
1826 mutex_unlock(&lli->lli_size_mutex);
1829 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1831 struct ll_inode_info *lli = ll_i2info(inode);
1832 struct mdt_body *body = md->body;
1833 struct ll_sb_info *sbi = ll_i2sbi(inode);
1835 if (body->mbo_valid & OBD_MD_FLEASIZE)
1836 cl_file_inode_init(inode, md);
1838 if (S_ISDIR(inode->i_mode)) {
1841 rc = ll_update_lsm_md(inode, md);
1846 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1847 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1848 ll_update_remote_perm(inode, md->remote_perm);
1850 #ifdef CONFIG_FS_POSIX_ACL
1851 else if (body->mbo_valid & OBD_MD_FLACL) {
1852 spin_lock(&lli->lli_lock);
1853 if (lli->lli_posix_acl)
1854 posix_acl_release(lli->lli_posix_acl);
1855 lli->lli_posix_acl = md->posix_acl;
1856 spin_unlock(&lli->lli_lock);
1859 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1860 sbi->ll_flags & LL_SBI_32BIT_API);
1861 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1863 if (body->mbo_valid & OBD_MD_FLATIME) {
1864 if (body->mbo_atime > LTIME_S(inode->i_atime))
1865 LTIME_S(inode->i_atime) = body->mbo_atime;
1866 lli->lli_atime = body->mbo_atime;
1869 if (body->mbo_valid & OBD_MD_FLMTIME) {
1870 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1871 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1872 "to "LPU64"\n", inode->i_ino,
1873 LTIME_S(inode->i_mtime), body->mbo_mtime);
1874 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1876 lli->lli_mtime = body->mbo_mtime;
1879 if (body->mbo_valid & OBD_MD_FLCTIME) {
1880 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1881 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1882 lli->lli_ctime = body->mbo_ctime;
1885 if (body->mbo_valid & OBD_MD_FLMODE)
1886 inode->i_mode = (inode->i_mode & S_IFMT) |
1887 (body->mbo_mode & ~S_IFMT);
1889 if (body->mbo_valid & OBD_MD_FLTYPE)
1890 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1891 (body->mbo_mode & S_IFMT);
1893 LASSERT(inode->i_mode != 0);
1894 if (S_ISREG(inode->i_mode))
1895 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1896 LL_MAX_BLKSIZE_BITS);
1898 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1900 if (body->mbo_valid & OBD_MD_FLUID)
1901 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1902 if (body->mbo_valid & OBD_MD_FLGID)
1903 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1904 if (body->mbo_valid & OBD_MD_FLFLAGS)
1905 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1906 if (body->mbo_valid & OBD_MD_FLNLINK)
1907 set_nlink(inode, body->mbo_nlink);
1908 if (body->mbo_valid & OBD_MD_FLRDEV)
1909 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1911 if (body->mbo_valid & OBD_MD_FLID) {
1912 /* FID shouldn't be changed! */
1913 if (fid_is_sane(&lli->lli_fid)) {
1914 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1915 "Trying to change FID "DFID
1916 " to the "DFID", inode "DFID"(%p)\n",
1917 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1918 PFID(ll_inode2fid(inode)), inode);
1920 lli->lli_fid = body->mbo_fid1;
1924 LASSERT(fid_seq(&lli->lli_fid) != 0);
1926 if (body->mbo_valid & OBD_MD_FLSIZE) {
1927 i_size_write(inode, body->mbo_size);
1929 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1930 PFID(ll_inode2fid(inode)),
1931 (unsigned long long)body->mbo_size);
1933 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1934 inode->i_blocks = body->mbo_blocks;
1937 if (body->mbo_valid & OBD_MD_TSTATE) {
1938 /* Set LLIF_FILE_RESTORING if restore ongoing and
1939 * clear it when done to ensure to start again
1940 * glimpsing updated attrs
1942 if (body->mbo_t_state & MS_RESTORE)
1943 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1945 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1951 int ll_read_inode2(struct inode *inode, void *opaque)
1953 struct lustre_md *md = opaque;
1954 struct ll_inode_info *lli = ll_i2info(inode);
1958 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1959 PFID(&lli->lli_fid), inode);
1961 /* Core attributes from the MDS first. This is a new inode, and
1962 * the VFS doesn't zero times in the core inode so we have to do
1963 * it ourselves. They will be overwritten by either MDS or OST
1964 * attributes - we just need to make sure they aren't newer. */
1965 LTIME_S(inode->i_mtime) = 0;
1966 LTIME_S(inode->i_atime) = 0;
1967 LTIME_S(inode->i_ctime) = 0;
1969 rc = ll_update_inode(inode, md);
1973 /* OIDEBUG(inode); */
1975 /* initializing backing dev info. */
1976 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1979 if (S_ISREG(inode->i_mode)) {
1980 struct ll_sb_info *sbi = ll_i2sbi(inode);
1981 inode->i_op = &ll_file_inode_operations;
1982 inode->i_fop = sbi->ll_fop;
1983 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1985 } else if (S_ISDIR(inode->i_mode)) {
1986 inode->i_op = &ll_dir_inode_operations;
1987 inode->i_fop = &ll_dir_operations;
1989 } else if (S_ISLNK(inode->i_mode)) {
1990 inode->i_op = &ll_fast_symlink_inode_operations;
1993 inode->i_op = &ll_special_inode_operations;
1995 init_special_inode(inode, inode->i_mode,
2004 void ll_delete_inode(struct inode *inode)
2006 struct ll_inode_info *lli = ll_i2info(inode);
2009 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2010 /* It is last chance to write out dirty pages,
2011 * otherwise we may lose data while umount */
2012 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2014 truncate_inode_pages_final(&inode->i_data);
2016 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2017 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2018 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2020 #ifdef HAVE_SBOPS_EVICT_INODE
2021 ll_clear_inode(inode);
2028 int ll_iocontrol(struct inode *inode, struct file *file,
2029 unsigned int cmd, unsigned long arg)
2031 struct ll_sb_info *sbi = ll_i2sbi(inode);
2032 struct ptlrpc_request *req = NULL;
2037 case FSFILT_IOC_GETFLAGS: {
2038 struct mdt_body *body;
2039 struct md_op_data *op_data;
2041 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2042 0, 0, LUSTRE_OPC_ANY,
2044 if (IS_ERR(op_data))
2045 RETURN(PTR_ERR(op_data));
2047 op_data->op_valid = OBD_MD_FLFLAGS;
2048 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2049 ll_finish_md_op_data(op_data);
2051 CERROR("%s: failure inode "DFID": rc = %d\n",
2052 sbi->ll_md_exp->exp_obd->obd_name,
2053 PFID(ll_inode2fid(inode)), rc);
2057 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2059 flags = body->mbo_flags;
2061 ptlrpc_req_finished(req);
2063 RETURN(put_user(flags, (int __user *)arg));
2065 case FSFILT_IOC_SETFLAGS: {
2067 struct md_op_data *op_data;
2068 struct cl_object *obj;
2070 if (get_user(flags, (int __user *)arg))
2073 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2074 LUSTRE_OPC_ANY, NULL);
2075 if (IS_ERR(op_data))
2076 RETURN(PTR_ERR(op_data));
2078 op_data->op_attr_flags = flags;
2079 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2080 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2081 ll_finish_md_op_data(op_data);
2082 ptlrpc_req_finished(req);
2086 inode->i_flags = ll_ext_to_inode_flags(flags);
2088 obj = ll_i2info(inode)->lli_clob;
2092 OBD_ALLOC_PTR(attr);
2096 attr->ia_valid = ATTR_ATTR_FLAG;
2097 rc = cl_setattr_ost(obj, attr, flags);
2109 int ll_flush_ctx(struct inode *inode)
2111 struct ll_sb_info *sbi = ll_i2sbi(inode);
2113 CDEBUG(D_SEC, "flush context for user %d\n",
2114 from_kuid(&init_user_ns, current_uid()));
2116 obd_set_info_async(NULL, sbi->ll_md_exp,
2117 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2119 obd_set_info_async(NULL, sbi->ll_dt_exp,
2120 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2125 /* umount -f client means force down, don't save state */
2126 void ll_umount_begin(struct super_block *sb)
2128 struct ll_sb_info *sbi = ll_s2sbi(sb);
2129 struct obd_device *obd;
2130 struct obd_ioctl_data *ioc_data;
2133 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2134 sb->s_count, atomic_read(&sb->s_active));
2136 obd = class_exp2obd(sbi->ll_md_exp);
2138 CERROR("Invalid MDC connection handle "LPX64"\n",
2139 sbi->ll_md_exp->exp_handle.h_cookie);
2145 obd = class_exp2obd(sbi->ll_dt_exp);
2147 CERROR("Invalid LOV connection handle "LPX64"\n",
2148 sbi->ll_dt_exp->exp_handle.h_cookie);
2154 OBD_ALLOC_PTR(ioc_data);
2156 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2157 sizeof *ioc_data, ioc_data, NULL);
2159 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2160 sizeof *ioc_data, ioc_data, NULL);
2162 OBD_FREE_PTR(ioc_data);
2165 /* Really, we'd like to wait until there are no requests outstanding,
2166 * and then continue. For now, we just invalidate the requests,
2167 * schedule() and sleep one second if needed, and hope.
2173 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2175 struct ll_sb_info *sbi = ll_s2sbi(sb);
2176 char *profilenm = get_profile_name(sb);
2180 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2181 read_only = *flags & MS_RDONLY;
2182 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2183 sizeof(KEY_READ_ONLY),
2184 KEY_READ_ONLY, sizeof(read_only),
2187 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2188 profilenm, read_only ?
2189 "read-only" : "read-write", err);
2194 sb->s_flags |= MS_RDONLY;
2196 sb->s_flags &= ~MS_RDONLY;
2198 if (sbi->ll_flags & LL_SBI_VERBOSE)
2199 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2200 read_only ? "read-only" : "read-write");
2206 * Cleanup the open handle that is cached on MDT-side.
2208 * For open case, the client side open handling thread may hit error
2209 * after the MDT grant the open. Under such case, the client should
2210 * send close RPC to the MDT as cleanup; otherwise, the open handle
2211 * on the MDT will be leaked there until the client umount or evicted.
2213 * In further, if someone unlinked the file, because the open handle
2214 * holds the reference on such file/object, then it will block the
2215 * subsequent threads that want to locate such object via FID.
2217 * \param[in] sb super block for this file-system
2218 * \param[in] open_req pointer to the original open request
2220 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2222 struct mdt_body *body;
2223 struct md_op_data *op_data;
2224 struct ptlrpc_request *close_req = NULL;
2225 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2228 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2229 OBD_ALLOC_PTR(op_data);
2230 if (op_data == NULL) {
2231 CWARN("%s: cannot allocate op_data to release open handle for "
2233 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2238 op_data->op_fid1 = body->mbo_fid1;
2239 op_data->op_handle = body->mbo_handle;
2240 op_data->op_mod_time = cfs_time_current_sec();
2241 md_close(exp, op_data, NULL, &close_req);
2242 ptlrpc_req_finished(close_req);
2243 ll_finish_md_op_data(op_data);
2248 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2249 struct super_block *sb, struct lookup_intent *it)
2251 struct ll_sb_info *sbi = NULL;
2252 struct lustre_md md = { NULL };
2256 LASSERT(*inode || sb);
2257 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2258 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2259 sbi->ll_md_exp, &md);
2264 rc = ll_update_inode(*inode, &md);
2268 LASSERT(sb != NULL);
2271 * At this point server returns to client's same fid as client
2272 * generated for creating. So using ->fid1 is okay here.
2274 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2276 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2277 sbi->ll_flags & LL_SBI_32BIT_API),
2279 if (IS_ERR(*inode)) {
2280 #ifdef CONFIG_FS_POSIX_ACL
2282 posix_acl_release(md.posix_acl);
2283 md.posix_acl = NULL;
2286 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2288 CERROR("new_inode -fatal: rc %d\n", rc);
2293 /* Handling piggyback layout lock.
2294 * Layout lock can be piggybacked by getattr and open request.
2295 * The lsm can be applied to inode only if it comes with a layout lock
2296 * otherwise correct layout may be overwritten, for example:
2297 * 1. proc1: mdt returns a lsm but not granting layout
2298 * 2. layout was changed by another client
2299 * 3. proc2: refresh layout and layout lock granted
2300 * 4. proc1: to apply a stale layout */
2301 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2302 struct lustre_handle lockh;
2303 struct ldlm_lock *lock;
2305 lockh.cookie = it->d.lustre.it_lock_handle;
2306 lock = ldlm_handle2lock(&lockh);
2307 LASSERT(lock != NULL);
2308 if (ldlm_has_layout(lock)) {
2309 struct cl_object_conf conf;
2311 memset(&conf, 0, sizeof(conf));
2312 conf.coc_opc = OBJECT_CONF_SET;
2313 conf.coc_inode = *inode;
2314 conf.coc_lock = lock;
2315 conf.u.coc_layout = md.layout;
2316 (void)ll_layout_conf(*inode, &conf);
2318 LDLM_LOCK_PUT(lock);
2324 md_free_lustre_md(sbi->ll_md_exp, &md);
2327 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2328 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2333 int ll_obd_statfs(struct inode *inode, void __user *arg)
2335 struct ll_sb_info *sbi = NULL;
2336 struct obd_export *exp;
2338 struct obd_ioctl_data *data = NULL;
2340 __u32 __user flags; /* not user, but obd_iocontrol is abused */
2343 if (!inode || !(sbi = ll_i2sbi(inode)))
2344 GOTO(out_statfs, rc = -EINVAL);
2346 rc = obd_ioctl_getdata(&buf, &len, arg);
2348 GOTO(out_statfs, rc);
2351 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2352 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2353 GOTO(out_statfs, rc = -EINVAL);
2355 if (data->ioc_inllen1 != sizeof(__u32) ||
2356 data->ioc_inllen2 != sizeof(__u32) ||
2357 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2358 data->ioc_plen2 != sizeof(struct obd_uuid))
2359 GOTO(out_statfs, rc = -EINVAL);
2361 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2362 if (type & LL_STATFS_LMV)
2363 exp = sbi->ll_md_exp;
2364 else if (type & LL_STATFS_LOV)
2365 exp = sbi->ll_dt_exp;
2367 GOTO(out_statfs, rc = -ENODEV);
2369 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2370 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2372 GOTO(out_statfs, rc);
2375 obd_ioctl_freedata(buf, len);
2379 int ll_process_config(struct lustre_cfg *lcfg)
2381 struct super_block *sb;
2386 /* The instance name contains the sb: lustre-client-aacfe000 */
2387 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2388 if (!ptr || !*(++ptr))
2390 if (sscanf(ptr, "%lx", &x) != 1)
2392 sb = (struct super_block *)x;
2393 /* This better be a real Lustre superblock! */
2394 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2396 /* Note we have not called client_common_fill_super yet, so
2397 proc fns must be able to handle that! */
2398 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2405 /* this function prepares md_op_data hint for passing it down to MD stack. */
2406 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2407 struct inode *i1, struct inode *i2,
2408 const char *name, size_t namelen,
2409 __u32 mode, __u32 opc, void *data)
2411 LASSERT(i1 != NULL);
2414 /* Do not reuse namelen for something else. */
2416 return ERR_PTR(-EINVAL);
2418 if (namelen > ll_i2sbi(i1)->ll_namelen)
2419 return ERR_PTR(-ENAMETOOLONG);
2421 if (!lu_name_is_valid_2(name, namelen))
2422 return ERR_PTR(-EINVAL);
2425 if (op_data == NULL)
2426 OBD_ALLOC_PTR(op_data);
2428 if (op_data == NULL)
2429 return ERR_PTR(-ENOMEM);
2431 ll_i2gids(op_data->op_suppgids, i1, i2);
2432 op_data->op_fid1 = *ll_inode2fid(i1);
2433 op_data->op_default_stripe_offset = -1;
2434 if (S_ISDIR(i1->i_mode)) {
2435 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2436 if (opc == LUSTRE_OPC_MKDIR)
2437 op_data->op_default_stripe_offset =
2438 ll_i2info(i1)->lli_def_stripe_offset;
2442 op_data->op_fid2 = *ll_inode2fid(i2);
2443 if (S_ISDIR(i2->i_mode))
2444 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2446 fid_zero(&op_data->op_fid2);
2449 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2450 op_data->op_cli_flags |= CLI_HASH64;
2452 if (ll_need_32bit_api(ll_i2sbi(i1)))
2453 op_data->op_cli_flags |= CLI_API32;
2455 op_data->op_name = name;
2456 op_data->op_namelen = namelen;
2457 op_data->op_mode = mode;
2458 op_data->op_mod_time = cfs_time_current_sec();
2459 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2460 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2461 op_data->op_cap = cfs_curproc_cap_pack();
2462 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2463 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2464 op_data->op_bias |= MDS_CREATE_VOLATILE;
2466 op_data->op_mds = 0;
2468 op_data->op_data = data;
2473 void ll_finish_md_op_data(struct md_op_data *op_data)
2475 OBD_FREE_PTR(op_data);
2478 #ifdef HAVE_SUPEROPS_USE_DENTRY
2479 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2481 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2484 struct ll_sb_info *sbi;
2486 #ifdef HAVE_SUPEROPS_USE_DENTRY
2487 LASSERT((seq != NULL) && (dentry != NULL));
2488 sbi = ll_s2sbi(dentry->d_sb);
2490 LASSERT((seq != NULL) && (vfs != NULL));
2491 sbi = ll_s2sbi(vfs->mnt_sb);
2494 if (sbi->ll_flags & LL_SBI_NOLCK)
2495 seq_puts(seq, ",nolock");
2497 if (sbi->ll_flags & LL_SBI_FLOCK)
2498 seq_puts(seq, ",flock");
2500 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2501 seq_puts(seq, ",localflock");
2503 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2504 seq_puts(seq, ",user_xattr");
2506 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2507 seq_puts(seq, ",lazystatfs");
2509 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2510 seq_puts(seq, ",user_fid2path");
2516 * Get obd name by cmd, and copy out to user space
2518 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2520 struct ll_sb_info *sbi = ll_i2sbi(inode);
2521 struct obd_device *obd;
2524 if (cmd == OBD_IOC_GETDTNAME)
2525 obd = class_exp2obd(sbi->ll_dt_exp);
2526 else if (cmd == OBD_IOC_GETMDNAME)
2527 obd = class_exp2obd(sbi->ll_md_exp);
2534 if (copy_to_user((void __user *)arg, obd->obd_name,
2535 strlen(obd->obd_name) + 1))
2542 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2543 * fsname will be returned in this buffer; otherwise, a static buffer will be
2544 * used to store the fsname and returned to caller.
2546 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2548 static char fsname_static[MTI_NAME_MAXLEN];
2549 struct lustre_sb_info *lsi = s2lsi(sb);
2554 /* this means the caller wants to use static buffer
2555 * and it doesn't care about race. Usually this is
2556 * in error reporting path */
2557 buf = fsname_static;
2558 buflen = sizeof(fsname_static);
2561 len = strlen(lsi->lsi_lmd->lmd_profile);
2562 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2563 if (ptr && (strcmp(ptr, "-client") == 0))
2566 if (unlikely(len >= buflen))
2568 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2574 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2581 p.mnt = current->fs->root.mnt;
2583 path = d_path(&p, buf, bufsize);
2588 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2590 char *buf, *path = NULL;
2591 struct dentry *dentry = NULL;
2592 struct inode *inode = page->mapping->host;
2594 /* this can be called inside spin lock so use GFP_ATOMIC. */
2595 buf = (char *)__get_free_page(GFP_ATOMIC);
2597 dentry = d_find_alias(page->mapping->host);
2599 path = ll_d_path(dentry, buf, PAGE_SIZE);
2603 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2604 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2605 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2606 PFID(ll_inode2fid(inode)),
2607 (path && !IS_ERR(path)) ? path : "", ioret);
2613 free_page((unsigned long)buf);
2616 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2617 struct lov_user_md **kbuf)
2619 struct lov_user_md lum;
2623 if (copy_from_user(&lum, md, sizeof(lum)))
2626 lum_size = ll_lov_user_md_size(&lum);
2630 OBD_ALLOC(*kbuf, lum_size);
2634 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2635 OBD_FREE(*kbuf, lum_size);
2643 * Compute llite root squash state after a change of root squash
2644 * configuration setting or add/remove of a lnet nid
2646 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2648 struct root_squash_info *squash = &sbi->ll_squash;
2651 lnet_process_id_t id;
2653 /* Update norootsquash flag */
2654 down_write(&squash->rsi_sem);
2655 if (list_empty(&squash->rsi_nosquash_nids))
2656 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2658 /* Do not apply root squash as soon as one of our NIDs is
2659 * in the nosquash_nids list */
2662 while (LNetGetId(i++, &id) != -ENOENT) {
2663 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2665 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2671 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2673 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2675 up_write(&squash->rsi_sem);
2679 * Parse linkea content to extract information about a given hardlink
2681 * \param[in] ldata - Initialized linkea data
2682 * \param[in] linkno - Link identifier
2683 * \param[out] parent_fid - The entry's parent FID
2684 * \param[out] ln - Entry name destination buffer
2686 * \retval 0 on success
2687 * \retval Appropriate negative error code on failure
2689 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2690 struct lu_fid *parent_fid, struct lu_name *ln)
2696 rc = linkea_init(ldata);
2700 if (linkno >= ldata->ld_leh->leh_reccount)
2701 /* beyond last link */
2704 linkea_first_entry(ldata);
2705 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2706 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2711 linkea_next_entry(ldata);
2721 * Get parent FID and name of an identified link. Operation is performed for
2722 * a given link number, letting the caller iterate over linkno to list one or
2723 * all links of an entry.
2725 * \param[in] file - File descriptor against which to perform the operation
2726 * \param[in,out] arg - User-filled structure containing the linkno to operate
2727 * on and the available size. It is eventually filled with
2728 * the requested information or left untouched on error
2730 * \retval - 0 on success
2731 * \retval - Appropriate negative error code on failure
2733 int ll_getparent(struct file *file, struct getparent __user *arg)
2735 struct dentry *dentry = file->f_path.dentry;
2736 struct inode *inode = dentry->d_inode;
2737 struct linkea_data *ldata;
2738 struct lu_buf buf = LU_BUF_NULL;
2740 struct lu_fid parent_fid;
2747 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2748 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2751 if (get_user(name_size, &arg->gp_name_size))
2754 if (get_user(linkno, &arg->gp_linkno))
2757 if (name_size > PATH_MAX)
2760 OBD_ALLOC(ldata, sizeof(*ldata));
2764 rc = linkea_data_new(ldata, &buf);
2766 GOTO(ldata_free, rc);
2768 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2772 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2776 if (ln.ln_namelen >= name_size)
2777 GOTO(lb_free, rc = -EOVERFLOW);
2779 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2780 GOTO(lb_free, rc = -EFAULT);
2782 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2783 GOTO(lb_free, rc = -EFAULT);
2785 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2786 GOTO(lb_free, rc = -EFAULT);
2791 OBD_FREE(ldata, sizeof(*ldata));