1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/llite/llite_lib.c
40 * Lustre Light Super operations
43 #define DEBUG_SUBSYSTEM S_LLITE
45 #include <linux/module.h>
46 #include <linux/types.h>
47 #include <linux/version.h>
50 #include <lustre_lite.h>
51 #include <lustre_ha.h>
52 #include <lustre_dlm.h>
53 #include <lprocfs_status.h>
54 #include <lustre_disk.h>
55 #include <lustre_param.h>
56 #include <lustre_log.h>
57 #include <cl_object.h>
58 #include <obd_cksum.h>
59 #include "llite_internal.h"
61 cfs_mem_cache_t *ll_file_data_slab;
63 CFS_LIST_HEAD(ll_super_blocks);
64 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
66 #ifndef MS_HAS_NEW_AOPS
67 extern struct address_space_operations ll_aops;
68 extern struct address_space_operations ll_dir_aops;
70 extern struct address_space_operations_ext ll_aops;
71 extern struct address_space_operations_ext ll_dir_aops;
75 #define log2(n) cfs_ffz(~(n))
78 static struct ll_sb_info *ll_init_sbi(void)
80 struct ll_sb_info *sbi = NULL;
87 OBD_ALLOC(sbi, sizeof(*sbi));
91 cfs_spin_lock_init(&sbi->ll_lock);
92 cfs_init_mutex(&sbi->ll_lco.lco_lock);
93 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
94 cfs_spin_lock_init(&sbi->ll_process_lock);
95 sbi->ll_rw_stats_on = 0;
98 pages = si.totalram - si.totalhigh;
99 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
100 #ifdef HAVE_BGL_SUPPORT
101 sbi->ll_async_page_max = pages / 4;
103 sbi->ll_async_page_max = pages / 2;
106 sbi->ll_async_page_max = (pages / 4) * 3;
109 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
110 SBI_DEFAULT_READAHEAD_MAX);
111 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
112 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
113 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
114 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
115 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
117 ll_generate_random_uuid(uuid);
118 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
119 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
121 cfs_spin_lock(&ll_sb_lock);
122 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
123 cfs_spin_unlock(&ll_sb_lock);
125 #ifdef ENABLE_CHECKSUM
126 sbi->ll_flags |= LL_SBI_CHECKSUM;
129 #ifdef HAVE_LRU_RESIZE_SUPPORT
130 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
133 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
134 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
136 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
140 /* metadata statahead is enabled by default */
141 sbi->ll_sa_max = LL_SA_RPC_DEF;
142 cfs_atomic_set(&sbi->ll_sa_total, 0);
143 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
144 cfs_atomic_set(&sbi->ll_agl_total, 0);
145 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
150 void ll_free_sbi(struct super_block *sb)
152 struct ll_sb_info *sbi = ll_s2sbi(sb);
156 cfs_spin_lock(&ll_sb_lock);
157 cfs_list_del(&sbi->ll_list);
158 cfs_spin_unlock(&ll_sb_lock);
159 OBD_FREE(sbi, sizeof(*sbi));
164 static struct dentry_operations ll_d_root_ops = {
165 .d_compare = ll_dcompare,
166 .d_revalidate = ll_revalidate_nd,
169 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
170 struct vfsmount *mnt)
172 struct inode *root = 0;
173 struct ll_sb_info *sbi = ll_s2sbi(sb);
174 struct obd_device *obd;
175 struct obd_capa *oc = NULL;
176 struct obd_statfs *osfs = NULL;
177 struct ptlrpc_request *request = NULL;
178 struct obd_connect_data *data = NULL;
179 struct obd_uuid *uuid;
180 struct md_op_data *op_data;
181 struct lustre_md lmd;
183 int size, err, checksum;
186 obd = class_name2obd(md);
188 CERROR("MD %s: not setup or attached\n", md);
202 if (proc_lustre_fs_root) {
203 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
206 CERROR("could not register mount in /proc/fs/lustre\n");
209 /* indicate the features supported by this client */
210 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
211 OBD_CONNECT_ATTRFID |
212 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
213 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
214 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
215 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
216 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
217 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH;
219 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
220 data->ocd_connect_flags |= OBD_CONNECT_SOM;
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
226 #ifdef CONFIG_FS_POSIX_ACL
227 data->ocd_connect_flags |= OBD_CONNECT_ACL;
229 data->ocd_ibits_known = MDS_INODELOCK_FULL;
230 data->ocd_version = LUSTRE_VERSION_CODE;
232 if (sb->s_flags & MS_RDONLY)
233 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
234 if (sbi->ll_flags & LL_SBI_USER_XATTR)
235 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
237 #ifdef HAVE_MS_FLOCK_LOCK
238 /* force vfs to use lustre handler for flock() calls - bug 10743 */
239 sb->s_flags |= MS_FLOCK_LOCK;
241 #ifdef MS_HAS_NEW_AOPS
242 sb->s_flags |= MS_HAS_NEW_AOPS;
245 if (sbi->ll_flags & LL_SBI_FLOCK)
246 sbi->ll_fop = &ll_file_operations_flock;
247 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
248 sbi->ll_fop = &ll_file_operations;
250 sbi->ll_fop = &ll_file_operations_noflock;
253 data->ocd_connect_flags |= OBD_CONNECT_REAL;
254 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
255 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
257 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
259 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
261 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
262 "recovery, of which this client is not a "
263 "part. Please wait for recovery to complete,"
264 " abort, or time out.\n", md);
267 CERROR("cannot connect to %s: rc = %d\n", md, err);
271 err = obd_fid_init(sbi->ll_md_exp);
273 CERROR("Can't init metadata layer FID infrastructure, "
278 err = obd_statfs(obd, osfs,
279 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
281 GOTO(out_md_fid, err);
283 /* This needs to be after statfs to ensure connect has finished.
284 * Note that "data" does NOT contain the valid connect reply.
285 * If connecting to a 1.8 server there will be no LMV device, so
286 * we can access the MDC export directly and exp_connect_flags will
287 * be non-zero, but if accessing an upgraded 2.1 server it will
288 * have the correct flags filled in.
289 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
290 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
291 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
292 valid != CLIENT_CONNECT_MDT_REQD) {
295 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
296 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
297 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
298 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
299 "feature(s) needed for correct operation "
300 "of this client (%s). Please upgrade "
301 "server or downgrade client.\n",
302 sbi->ll_md_exp->exp_obd->obd_name, buf);
303 OBD_FREE(buf, CFS_PAGE_SIZE);
304 GOTO(out_md, err = -EPROTO);
307 size = sizeof(*data);
308 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
309 KEY_CONN_DATA, &size, data, NULL);
311 CERROR("Get connect data failed: %d \n", err);
315 LASSERT(osfs->os_bsize);
316 sb->s_blocksize = osfs->os_bsize;
317 sb->s_blocksize_bits = log2(osfs->os_bsize);
318 sb->s_magic = LL_SUPER_MAGIC;
320 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
321 * retval = in_file->f_op->sendfile(...);
323 * retval = -EOVERFLOW;
325 * it will check if *ppos is greater than max. However, max equals to
326 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
327 * has been defined as a signed long long integer in linux kernel. */
328 #if BITS_PER_LONG == 64
329 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
331 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
333 sbi->ll_namelen = osfs->os_namelen;
334 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
336 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
337 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
338 LCONSOLE_INFO("Disabling user_xattr feature because "
339 "it is not supported on the server\n");
340 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
343 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
345 sb->s_flags |= MS_POSIXACL;
347 sbi->ll_flags |= LL_SBI_ACL;
349 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
351 sb->s_flags &= ~MS_POSIXACL;
353 sbi->ll_flags &= ~LL_SBI_ACL;
356 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
357 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
358 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
359 LCONSOLE_INFO("client is set as remote by default.\n");
362 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
363 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
364 LCONSOLE_INFO("client claims to be remote, but server "
365 "rejected, forced to be local.\n");
369 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
370 LCONSOLE_INFO("client enabled MDS capability!\n");
371 sbi->ll_flags |= LL_SBI_MDS_CAPA;
374 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
375 LCONSOLE_INFO("client enabled OSS capability!\n");
376 sbi->ll_flags |= LL_SBI_OSS_CAPA;
379 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
380 sbi->ll_flags |= LL_SBI_64BIT_HASH;
382 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
383 sbi->ll_md_brw_size = data->ocd_brw_size;
385 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
387 obd = class_name2obd(dt);
389 CERROR("DT %s: not setup or attached\n", dt);
390 GOTO(out_md_fid, err = -ENODEV);
393 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
394 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
395 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
396 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
397 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
398 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
399 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
400 OBD_CONNECT_MAXBYTES;
402 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
403 data->ocd_connect_flags |= OBD_CONNECT_SOM;
405 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
406 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
407 * disabled by default, because it can still be enabled on the
408 * fly via /proc. As a consequence, we still need to come to an
409 * agreement on the supported algorithms at connect time */
410 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
412 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
413 data->ocd_cksum_types = OBD_CKSUM_ADLER;
415 data->ocd_cksum_types = cksum_types_supported();
418 #ifdef HAVE_LRU_RESIZE_SUPPORT
419 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
421 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
422 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
424 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
425 "ocd_grant: %d\n", data->ocd_connect_flags,
426 data->ocd_version, data->ocd_grant);
428 obd->obd_upcall.onu_owner = &sbi->ll_lco;
429 obd->obd_upcall.onu_upcall = cl_ocd_update;
431 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
433 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
435 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
436 "recovery, of which this client is not a "
437 "part. Please wait for recovery to "
438 "complete, abort, or time out.\n", dt);
439 GOTO(out_md_fid, err);
441 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
442 GOTO(out_md_fid, err);
445 err = obd_fid_init(sbi->ll_dt_exp);
447 CERROR("Can't init data layer FID infrastructure, "
452 cfs_mutex_down(&sbi->ll_lco.lco_lock);
453 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
454 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
455 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
456 cfs_mutex_up(&sbi->ll_lco.lco_lock);
458 fid_zero(&sbi->ll_root_fid);
459 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
461 CERROR("cannot mds_connect: rc = %d\n", err);
462 GOTO(out_lock_cn_cb, err);
464 if (!fid_is_sane(&sbi->ll_root_fid)) {
465 CERROR("Invalid root fid during mount\n");
466 GOTO(out_lock_cn_cb, err = -EINVAL);
468 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
470 sb->s_op = &lustre_super_operations;
471 #if THREAD_SIZE >= 8192
472 sb->s_export_op = &lustre_export_operations;
476 * XXX: move this to after cbd setup? */
477 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
478 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
479 valid |= OBD_MD_FLRMTPERM;
480 else if (sbi->ll_flags & LL_SBI_ACL)
481 valid |= OBD_MD_FLACL;
483 OBD_ALLOC_PTR(op_data);
485 GOTO(out_lock_cn_cb, err = -ENOMEM);
487 op_data->op_fid1 = sbi->ll_root_fid;
488 op_data->op_mode = 0;
489 op_data->op_capa1 = oc;
490 op_data->op_valid = valid;
492 err = md_getattr(sbi->ll_md_exp, op_data, &request);
495 OBD_FREE_PTR(op_data);
497 CERROR("md_getattr failed for root: rc = %d\n", err);
498 GOTO(out_lock_cn_cb, err);
500 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
501 sbi->ll_md_exp, &lmd);
503 CERROR("failed to understand root inode md: rc = %d\n", err);
504 ptlrpc_req_finished (request);
505 GOTO(out_lock_cn_cb, err);
508 LASSERT(fid_is_sane(&sbi->ll_root_fid));
509 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
510 md_free_lustre_md(sbi->ll_md_exp, &lmd);
511 ptlrpc_req_finished(request);
513 if (root == NULL || IS_ERR(root)) {
515 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
516 #ifdef CONFIG_FS_POSIX_ACL
518 posix_acl_release(lmd.posix_acl);
519 lmd.posix_acl = NULL;
522 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
524 CERROR("lustre_lite: bad iget4 for root\n");
528 err = ll_close_thread_start(&sbi->ll_lcq);
530 CERROR("cannot start close thread: rc %d\n", err);
534 #ifdef CONFIG_FS_POSIX_ACL
535 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
536 rct_init(&sbi->ll_rct);
537 et_init(&sbi->ll_et);
541 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
542 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
543 KEY_CHECKSUM, sizeof(checksum), &checksum,
547 sb->s_root = d_alloc_root(root);
548 sb->s_root->d_op = &ll_d_root_ops;
550 sbi->ll_sdev_orig = sb->s_dev;
552 /* We set sb->s_dev equal on all lustre clients in order to support
553 * NFS export clustering. NFSD requires that the FSID be the same
555 /* s_dev is also used in lt_compare() to compare two fs, but that is
556 * only a node-local comparison. */
557 uuid = obd_get_uuid(sbi->ll_md_exp);
559 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
572 obd_fid_fini(sbi->ll_dt_exp);
574 obd_disconnect(sbi->ll_dt_exp);
575 sbi->ll_dt_exp = NULL;
577 obd_fid_fini(sbi->ll_md_exp);
579 obd_disconnect(sbi->ll_md_exp);
580 sbi->ll_md_exp = NULL;
586 lprocfs_unregister_mountpoint(sbi);
590 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
594 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
596 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
597 KEY_MAX_EASIZE, &size, lmmsize, NULL);
599 CERROR("Get max mdsize error rc %d \n", rc);
604 void ll_dump_inode(struct inode *inode)
606 struct list_head *tmp;
607 int dentry_count = 0;
609 LASSERT(inode != NULL);
611 list_for_each(tmp, &inode->i_dentry)
614 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
615 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
616 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
619 void lustre_dump_dentry(struct dentry *dentry, int recur)
621 struct list_head *tmp;
624 LASSERT(dentry != NULL);
626 list_for_each(tmp, &dentry->d_subdirs)
629 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
630 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
631 dentry->d_name.len, dentry->d_name.name,
632 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
633 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
634 dentry->d_flags, dentry->d_fsdata, subdirs);
635 if (dentry->d_inode != NULL)
636 ll_dump_inode(dentry->d_inode);
641 list_for_each(tmp, &dentry->d_subdirs) {
642 struct dentry *d = list_entry(tmp, struct dentry, d_child);
643 lustre_dump_dentry(d, recur - 1);
647 void client_common_put_super(struct super_block *sb)
649 struct ll_sb_info *sbi = ll_s2sbi(sb);
652 #ifdef CONFIG_FS_POSIX_ACL
653 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
654 et_fini(&sbi->ll_et);
655 rct_fini(&sbi->ll_rct);
659 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
661 ll_close_thread_shutdown(sbi->ll_lcq);
665 cfs_list_del(&sbi->ll_conn_chain);
667 obd_fid_fini(sbi->ll_dt_exp);
668 obd_disconnect(sbi->ll_dt_exp);
669 sbi->ll_dt_exp = NULL;
671 lprocfs_unregister_mountpoint(sbi);
673 obd_fid_fini(sbi->ll_md_exp);
674 obd_disconnect(sbi->ll_md_exp);
675 sbi->ll_md_exp = NULL;
680 void ll_kill_super(struct super_block *sb)
682 struct ll_sb_info *sbi;
687 if (!(sb->s_flags & MS_ACTIVE))
691 /* we need restore s_dev from changed for clustred NFS before put_super
692 * because new kernels have cached s_dev and change sb->s_dev in
693 * put_super not affected real removing devices */
695 sb->s_dev = sbi->ll_sdev_orig;
699 char *ll_read_opt(const char *opt, char *data)
705 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
706 if (strncmp(opt, data, strlen(opt)))
708 if ((value = strchr(data, '=')) == NULL)
712 OBD_ALLOC(retval, strlen(value) + 1);
714 CERROR("out of memory!\n");
718 memcpy(retval, value, strlen(value)+1);
719 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
723 static inline int ll_set_opt(const char *opt, char *data, int fl)
725 if (strncmp(opt, data, strlen(opt)) != 0)
731 /* non-client-specific mount options are parsed in lmd_parse */
732 static int ll_options(char *options, int *flags)
735 char *s1 = options, *s2;
741 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
744 CDEBUG(D_SUPER, "next opt=%s\n", s1);
745 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
750 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
755 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
760 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
765 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
770 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
775 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
776 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
778 /* Ignore deprecated mount option. The client will
779 * always try to mount with ACL support, whether this
780 * is used depends on whether server supports it. */
781 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
782 "mount option 'acl'.\n");
785 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
787 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
788 "mount option 'noacl'.\n");
792 #warning "{no}acl options have been deprecated since 1.8, please remove them"
794 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
800 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
805 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
810 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
815 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
820 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
825 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
830 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
835 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
841 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
847 s2 = strchr(s1, ',');
855 void ll_lli_init(struct ll_inode_info *lli)
857 lli->lli_inode_magic = LLI_INODE_MAGIC;
859 lli->lli_ioepoch = 0;
860 cfs_spin_lock_init(&lli->lli_lock);
861 lli->lli_posix_acl = NULL;
862 lli->lli_remote_perms = NULL;
863 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
864 /* Do not set lli_fid, it has been initialized already. */
865 fid_zero(&lli->lli_pfid);
866 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
867 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
868 cfs_atomic_set(&lli->lli_open_count, 0);
869 lli->lli_mds_capa = NULL;
870 lli->lli_rmtperm_time = 0;
871 lli->lli_pending_och = NULL;
872 lli->lli_mds_read_och = NULL;
873 lli->lli_mds_write_och = NULL;
874 lli->lli_mds_exec_och = NULL;
875 lli->lli_open_fd_read_count = 0;
876 lli->lli_open_fd_write_count = 0;
877 lli->lli_open_fd_exec_count = 0;
878 cfs_sema_init(&lli->lli_och_sem, 1);
879 cfs_spin_lock_init(&lli->lli_agl_lock);
881 lli->lli_clob = NULL;
883 LASSERT(lli->lli_vfs_inode.i_mode != 0);
884 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
885 cfs_sema_init(&lli->lli_readdir_sem, 1);
886 lli->lli_opendir_key = NULL;
889 lli->lli_def_acl = NULL;
890 cfs_spin_lock_init(&lli->lli_sa_lock);
891 lli->lli_opendir_pid = 0;
893 cfs_sema_init(&lli->lli_size_sem, 1);
894 lli->lli_size_sem_owner = NULL;
895 lli->lli_symlink_name = NULL;
896 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
897 cfs_init_rwsem(&lli->lli_trunc_sem);
898 cfs_sema_init(&lli->lli_write_sem, 1);
899 lli->lli_async_rc = 0;
900 lli->lli_write_rc = 0;
901 cfs_init_rwsem(&lli->lli_glimpse_sem);
902 lli->lli_glimpse_time = 0;
903 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
904 lli->lli_agl_index = 0;
908 static inline int ll_bdi_register(struct backing_dev_info *bdi)
910 #ifdef HAVE_BDI_REGISTER
911 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
914 bdi->name = "lustre";
916 return bdi_register(bdi, NULL, "lustre-%d",
917 atomic_inc_return(&ll_bdi_num));
923 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
925 struct lustre_profile *lprof = NULL;
926 struct lustre_sb_info *lsi = s2lsi(sb);
927 struct ll_sb_info *sbi;
928 char *dt = NULL, *md = NULL;
929 char *profilenm = get_profile_name(sb);
930 struct config_llog_instance *cfg;
931 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
932 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
936 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
944 /* client additional sb info */
945 lsi->lsi_llsbi = sbi = ll_init_sbi();
947 cfs_module_put(THIS_MODULE);
952 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
956 err = ll_bdi_init(&lsi->lsi_bdi);
959 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
960 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
961 err = ll_bdi_register(&lsi->lsi_bdi);
966 sb->s_bdi = &lsi->lsi_bdi;
969 /* Generate a string unique to this super, in case some joker tries
970 to mount the same fs at two mount points.
971 Use the address of the super itself.*/
972 cfg->cfg_instance = sb;
973 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
975 /* set up client obds */
976 err = lustre_process_log(sb, profilenm, cfg);
978 CERROR("Unable to process log: %d\n", err);
982 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
983 lprof = class_get_profile(profilenm);
985 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
986 " read from the MGS. Does that filesystem "
987 "exist?\n", profilenm);
988 GOTO(out_free, err = -EINVAL);
990 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
991 lprof->lp_md, lprof->lp_dt);
993 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
995 GOTO(out_free, err = -ENOMEM);
996 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
998 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1000 GOTO(out_free, err = -ENOMEM);
1001 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1003 /* connections, registrations, sb setup */
1004 err = client_common_fill_super(sb, md, dt, mnt);
1008 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1010 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1014 LCONSOLE_WARN("Client %s has started\n", profilenm);
1018 } /* ll_fill_super */
1021 void lu_context_keys_dump(void);
1023 void ll_put_super(struct super_block *sb)
1025 struct config_llog_instance cfg;
1026 struct obd_device *obd;
1027 struct lustre_sb_info *lsi = s2lsi(sb);
1028 struct ll_sb_info *sbi = ll_s2sbi(sb);
1029 char *profilenm = get_profile_name(sb);
1030 int force = 1, next;
1033 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1035 ll_print_capa_stat(sbi);
1037 cfg.cfg_instance = sb;
1038 lustre_end_log(sb, profilenm, &cfg);
1040 if (sbi->ll_md_exp) {
1041 obd = class_exp2obd(sbi->ll_md_exp);
1043 force = obd->obd_force;
1046 /* We need to set force before the lov_disconnect in
1047 lustre_common_put_super, since l_d cleans up osc's as well. */
1050 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1052 obd->obd_force = force;
1057 /* Only if client_common_fill_super succeeded */
1058 client_common_put_super(sb);
1062 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1063 class_manual_cleanup(obd);
1067 class_del_profile(profilenm);
1069 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1070 ll_bdi_destroy(&lsi->lsi_bdi);
1071 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1075 lsi->lsi_llsbi = NULL;
1077 lustre_common_put_super(sb);
1079 cl_env_cache_purge(~0);
1081 LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1083 cfs_module_put(THIS_MODULE);
1086 } /* client_put_super */
1088 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1090 struct inode *inode = NULL;
1091 /* NOTE: we depend on atomic igrab() -bzzz */
1092 lock_res_and_lock(lock);
1093 if (lock->l_ast_data) {
1094 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1095 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1096 inode = igrab(lock->l_ast_data);
1098 inode = lock->l_ast_data;
1099 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1100 D_WARNING, lock, "l_ast_data %p is "
1101 "bogus: magic %08x", lock->l_ast_data,
1102 lli->lli_inode_magic);
1106 unlock_res_and_lock(lock);
1110 static int null_if_equal(struct ldlm_lock *lock, void *data)
1112 if (data == lock->l_ast_data) {
1113 lock->l_ast_data = NULL;
1115 if (lock->l_req_mode != lock->l_granted_mode)
1116 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1119 return LDLM_ITER_CONTINUE;
1122 void ll_clear_inode(struct inode *inode)
1124 struct ll_inode_info *lli = ll_i2info(inode);
1125 struct ll_sb_info *sbi = ll_i2sbi(inode);
1128 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1129 inode->i_generation, inode);
1131 if (S_ISDIR(inode->i_mode)) {
1132 /* these should have been cleared in ll_file_release */
1133 LASSERT(lli->lli_opendir_key == NULL);
1134 LASSERT(lli->lli_sai == NULL);
1135 LASSERT(lli->lli_opendir_pid == 0);
1138 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1139 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1140 null_if_equal, inode);
1142 LASSERT(!lli->lli_open_fd_write_count);
1143 LASSERT(!lli->lli_open_fd_read_count);
1144 LASSERT(!lli->lli_open_fd_exec_count);
1146 if (lli->lli_mds_write_och)
1147 ll_md_real_close(inode, FMODE_WRITE);
1148 if (lli->lli_mds_exec_och)
1149 ll_md_real_close(inode, FMODE_EXEC);
1150 if (lli->lli_mds_read_och)
1151 ll_md_real_close(inode, FMODE_READ);
1153 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1154 OBD_FREE(lli->lli_symlink_name,
1155 strlen(lli->lli_symlink_name) + 1);
1156 lli->lli_symlink_name = NULL;
1159 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1160 LASSERT(lli->lli_posix_acl == NULL);
1161 if (lli->lli_remote_perms) {
1162 free_rmtperm_hash(lli->lli_remote_perms);
1163 lli->lli_remote_perms = NULL;
1166 #ifdef CONFIG_FS_POSIX_ACL
1167 else if (lli->lli_posix_acl) {
1168 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1169 LASSERT(lli->lli_remote_perms == NULL);
1170 posix_acl_release(lli->lli_posix_acl);
1171 lli->lli_posix_acl = NULL;
1174 lli->lli_inode_magic = LLI_INODE_DEAD;
1176 ll_clear_inode_capas(inode);
1177 if (!S_ISDIR(inode->i_mode))
1178 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1181 * XXX This has to be done before lsm is freed below, because
1182 * cl_object still uses inode lsm.
1184 cl_inode_fini(inode);
1187 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1188 lli->lli_smd = NULL;
1195 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1196 struct md_open_data **mod)
1198 struct lustre_md md;
1199 struct inode *inode = dentry->d_inode;
1200 struct ll_sb_info *sbi = ll_i2sbi(inode);
1201 struct ptlrpc_request *request = NULL;
1205 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1206 LUSTRE_OPC_ANY, NULL);
1207 if (IS_ERR(op_data))
1208 RETURN(PTR_ERR(op_data));
1210 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1213 ptlrpc_req_finished(request);
1214 if (rc == -ENOENT) {
1216 /* Unlinked special device node? Or just a race?
1217 * Pretend we done everything. */
1218 if (!S_ISREG(inode->i_mode) &&
1219 !S_ISDIR(inode->i_mode)) {
1220 ia_valid = op_data->op_attr.ia_valid;
1221 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1222 rc = simple_setattr(dentry, &op_data->op_attr);
1223 op_data->op_attr.ia_valid = ia_valid;
1225 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1226 CERROR("md_setattr fails: rc = %d\n", rc);
1231 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1232 sbi->ll_md_exp, &md);
1234 ptlrpc_req_finished(request);
1238 /* We want to adjust timestamps.
1239 * If there is at least some data in file, we cleared ATTR_SIZE
1240 * to avoid update size, otherwise it is important to do.(SOM case)
1242 ia_valid = op_data->op_attr.ia_valid;
1243 /* Since we set ATTR_*_SET flags above, and already done permission
1244 * check, So don't let inode_change_ok() check it again. */
1245 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1246 rc = simple_setattr(dentry, &op_data->op_attr);
1247 op_data->op_attr.ia_valid = ia_valid;
1249 /* Extract epoch data if obtained. */
1250 op_data->op_handle = md.body->handle;
1251 op_data->op_ioepoch = md.body->ioepoch;
1253 ll_update_inode(inode, &md);
1254 ptlrpc_req_finished(request);
1259 /* Close IO epoch and send Size-on-MDS attribute update. */
1260 static int ll_setattr_done_writing(struct inode *inode,
1261 struct md_op_data *op_data,
1262 struct md_open_data *mod)
1264 struct ll_inode_info *lli = ll_i2info(inode);
1268 LASSERT(op_data != NULL);
1269 if (!S_ISREG(inode->i_mode))
1272 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1273 op_data->op_ioepoch, PFID(&lli->lli_fid));
1275 op_data->op_flags = MF_EPOCH_CLOSE;
1276 ll_done_writing_attr(inode, op_data);
1277 ll_pack_inode2opdata(inode, op_data, NULL);
1279 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1280 if (rc == -EAGAIN) {
1281 /* MDS has instructed us to obtain Size-on-MDS attribute
1282 * from OSTs and send setattr to back to MDS. */
1283 rc = ll_som_update(inode, op_data);
1285 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1291 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1293 struct obd_capa *capa;
1296 if (attr->ia_valid & ATTR_SIZE)
1297 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1299 capa = ll_mdscapa_get(inode);
1301 rc = cl_setattr_ost(inode, attr, capa);
1303 if (attr->ia_valid & ATTR_SIZE)
1304 ll_truncate_free_capa(capa);
1311 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1312 * object(s) determine the file size and mtime. Otherwise, the MDS will
1313 * keep these values until such a time that objects are allocated for it.
1314 * We do the MDS operations first, as it is checking permissions for us.
1315 * We don't to the MDS RPC if there is nothing that we want to store there,
1316 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1317 * going to do an RPC anyways.
1319 * If we are doing a truncate, we will send the mtime and ctime updates
1320 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1321 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1324 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1326 struct inode *inode = dentry->d_inode;
1327 struct ll_inode_info *lli = ll_i2info(inode);
1328 struct lov_stripe_md *lsm = lli->lli_smd;
1329 struct md_op_data *op_data = NULL;
1330 struct md_open_data *mod = NULL;
1331 int ia_valid = attr->ia_valid;
1332 int rc = 0, rc1 = 0;
1335 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1337 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1339 if (ia_valid & ATTR_SIZE) {
1340 if (attr->ia_size > ll_file_maxbytes(inode)) {
1341 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1342 attr->ia_size, ll_file_maxbytes(inode));
1346 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1349 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1350 if (ia_valid & TIMES_SET_FLAGS) {
1351 if (cfs_curproc_fsuid() != inode->i_uid &&
1352 !cfs_capable(CFS_CAP_FOWNER))
1356 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1357 if (attr->ia_valid & ATTR_CTIME) {
1358 attr->ia_ctime = CFS_CURRENT_TIME;
1359 attr->ia_valid |= ATTR_CTIME_SET;
1361 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1362 attr->ia_atime = CFS_CURRENT_TIME;
1363 attr->ia_valid |= ATTR_ATIME_SET;
1365 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1366 attr->ia_mtime = CFS_CURRENT_TIME;
1367 attr->ia_valid |= ATTR_MTIME_SET;
1370 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1371 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1372 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1373 cfs_time_current_sec());
1375 /* NB: ATTR_SIZE will only be set after this point if the size
1376 * resides on the MDS, ie, this file has no objects. */
1378 attr->ia_valid &= ~ATTR_SIZE;
1380 /* We always do an MDS RPC, even if we're only changing the size;
1381 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1383 OBD_ALLOC_PTR(op_data);
1384 if (op_data == NULL)
1387 if (!S_ISDIR(inode->i_mode)) {
1388 if (ia_valid & ATTR_SIZE)
1389 UP_WRITE_I_ALLOC_SEM(inode);
1390 UNLOCK_INODE_MUTEX(inode);
1391 cfs_down_write(&lli->lli_trunc_sem);
1392 LOCK_INODE_MUTEX(inode);
1393 if (ia_valid & ATTR_SIZE)
1394 DOWN_WRITE_I_ALLOC_SEM(inode);
1397 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1399 /* Open epoch for truncate. */
1400 if (exp_connect_som(ll_i2mdexp(inode)) &&
1401 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1402 op_data->op_flags = MF_EPOCH_OPEN;
1404 rc = ll_md_setattr(dentry, op_data, &mod);
1408 ll_ioepoch_open(lli, op_data->op_ioepoch);
1409 if (!lsm || !S_ISREG(inode->i_mode)) {
1410 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1414 if (ia_valid & ATTR_SIZE)
1415 attr->ia_valid |= ATTR_SIZE;
1416 if (ia_valid & (ATTR_SIZE |
1417 ATTR_ATIME | ATTR_ATIME_SET |
1418 ATTR_MTIME | ATTR_MTIME_SET))
1419 /* on truncate and utimes send attributes to osts, setting
1420 * mtime/atime to past will be performed under PW 0:EOF extent
1421 * lock (new_size:EOF for truncate)
1422 * it may seem excessive to send mtime/atime updates to osts
1423 * when not setting times to past, but it is necessary due to
1424 * possible time de-synchronization */
1425 rc = ll_setattr_ost(inode, attr);
1429 if (op_data->op_ioepoch)
1430 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1431 ll_finish_md_op_data(op_data);
1433 if (!S_ISDIR(inode->i_mode))
1434 cfs_up_write(&lli->lli_trunc_sem);
1435 return rc ? rc : rc1;
1438 int ll_setattr(struct dentry *de, struct iattr *attr)
1440 int mode = de->d_inode->i_mode;
1442 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1443 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1444 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1446 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1447 (ATTR_SIZE|ATTR_MODE)) &&
1448 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1449 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1450 !(attr->ia_mode & S_ISGID))))
1451 attr->ia_valid |= ATTR_FORCE;
1453 if ((mode & S_ISUID) &&
1454 !(attr->ia_mode & S_ISUID) &&
1455 !(attr->ia_valid & ATTR_KILL_SUID))
1456 attr->ia_valid |= ATTR_KILL_SUID;
1458 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1459 !(attr->ia_mode & S_ISGID) &&
1460 !(attr->ia_valid & ATTR_KILL_SGID))
1461 attr->ia_valid |= ATTR_KILL_SGID;
1463 return ll_setattr_raw(de, attr);
1466 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1467 __u64 max_age, __u32 flags)
1469 struct ll_sb_info *sbi = ll_s2sbi(sb);
1470 struct obd_statfs obd_osfs;
1474 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1476 CERROR("md_statfs fails: rc = %d\n", rc);
1480 osfs->os_type = sb->s_magic;
1482 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1483 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1485 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1486 flags |= OBD_STATFS_NODELAY;
1488 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1489 &obd_osfs, max_age, flags);
1491 CERROR("obd_statfs fails: rc = %d\n", rc);
1495 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1496 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1499 osfs->os_bsize = obd_osfs.os_bsize;
1500 osfs->os_blocks = obd_osfs.os_blocks;
1501 osfs->os_bfree = obd_osfs.os_bfree;
1502 osfs->os_bavail = obd_osfs.os_bavail;
1504 /* If we don't have as many objects free on the OST as inodes
1505 * on the MDS, we reduce the total number of inodes to
1506 * compensate, so that the "inodes in use" number is correct.
1508 if (obd_osfs.os_ffree < osfs->os_ffree) {
1509 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1511 osfs->os_ffree = obd_osfs.os_ffree;
1516 #ifndef HAVE_STATFS_DENTRY_PARAM
1517 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1520 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1522 struct super_block *sb = de->d_sb;
1524 struct obd_statfs osfs;
1527 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1528 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1530 /* Some amount of caching on the client is allowed */
1531 rc = ll_statfs_internal(sb, &osfs,
1532 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1537 statfs_unpack(sfs, &osfs);
1539 /* We need to downshift for all 32-bit kernels, because we can't
1540 * tell if the kernel is being called via sys_statfs64() or not.
1541 * Stop before overflowing f_bsize - in which case it is better
1542 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1543 if (sizeof(long) < 8) {
1544 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1547 osfs.os_blocks >>= 1;
1548 osfs.os_bfree >>= 1;
1549 osfs.os_bavail >>= 1;
1553 sfs->f_blocks = osfs.os_blocks;
1554 sfs->f_bfree = osfs.os_bfree;
1555 sfs->f_bavail = osfs.os_bavail;
1560 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1562 struct ll_inode_info *lli;
1563 struct lov_stripe_md *lsm;
1565 LASSERT(!S_ISDIR(inode->i_mode));
1567 lli = ll_i2info(inode);
1568 LASSERT(lli->lli_size_sem_owner != current);
1569 cfs_down(&lli->lli_size_sem);
1570 LASSERT(lli->lli_size_sem_owner == NULL);
1571 lli->lli_size_sem_owner = current;
1573 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1576 lov_stripe_lock(lsm);
1579 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1581 struct ll_inode_info *lli;
1582 struct lov_stripe_md *lsm;
1584 lli = ll_i2info(inode);
1586 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1589 lov_stripe_unlock(lsm);
1590 LASSERT(lli->lli_size_sem_owner == current);
1591 lli->lli_size_sem_owner = NULL;
1592 cfs_up(&lli->lli_size_sem);
1595 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1597 struct ll_inode_info *lli = ll_i2info(inode);
1598 struct mdt_body *body = md->body;
1599 struct lov_stripe_md *lsm = md->lsm;
1600 struct ll_sb_info *sbi = ll_i2sbi(inode);
1602 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1604 LASSERT(S_ISREG(inode->i_mode));
1606 cfs_down(&lli->lli_och_sem);
1607 if (lli->lli_smd == NULL) {
1608 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1609 lsm->lsm_magic != LOV_MAGIC_V3) {
1610 dump_lsm(D_ERROR, lsm);
1613 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1614 lsm, inode->i_ino, inode->i_generation, inode);
1615 /* cl_file_inode_init must go before lli_smd or a race
1616 * is possible where client thinks the file has stripes,
1617 * but lov raid0 is not setup yet and parallel e.g.
1618 * glimpse would try to use uninitialized lov */
1619 cl_file_inode_init(inode, md);
1620 cfs_spin_lock(&lli->lli_lock);
1622 cfs_spin_unlock(&lli->lli_lock);
1623 cfs_up(&lli->lli_och_sem);
1624 lli->lli_maxbytes = lsm->lsm_maxbytes;
1625 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1626 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1628 cfs_up(&lli->lli_och_sem);
1629 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1630 lli->lli_smd->lsm_stripe_count ==
1631 lsm->lsm_stripe_count);
1632 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1633 CERROR("lsm mismatch for inode %ld\n",
1635 CERROR("lli_smd:\n");
1636 dump_lsm(D_ERROR, lli->lli_smd);
1638 dump_lsm(D_ERROR, lsm);
1642 if (lli->lli_smd != lsm)
1643 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1646 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1647 if (body->valid & OBD_MD_FLRMTPERM)
1648 ll_update_remote_perm(inode, md->remote_perm);
1650 #ifdef CONFIG_FS_POSIX_ACL
1651 else if (body->valid & OBD_MD_FLACL) {
1652 cfs_spin_lock(&lli->lli_lock);
1653 if (lli->lli_posix_acl)
1654 posix_acl_release(lli->lli_posix_acl);
1655 lli->lli_posix_acl = md->posix_acl;
1656 cfs_spin_unlock(&lli->lli_lock);
1659 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1660 inode->i_generation = cl_fid_build_gen(&body->fid1);
1662 if (body->valid & OBD_MD_FLATIME) {
1663 if (body->atime > LTIME_S(inode->i_atime))
1664 LTIME_S(inode->i_atime) = body->atime;
1665 lli->lli_lvb.lvb_atime = body->atime;
1667 if (body->valid & OBD_MD_FLMTIME) {
1668 if (body->mtime > LTIME_S(inode->i_mtime)) {
1669 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1670 "to "LPU64"\n", inode->i_ino,
1671 LTIME_S(inode->i_mtime), body->mtime);
1672 LTIME_S(inode->i_mtime) = body->mtime;
1674 lli->lli_lvb.lvb_mtime = body->mtime;
1676 if (body->valid & OBD_MD_FLCTIME) {
1677 if (body->ctime > LTIME_S(inode->i_ctime))
1678 LTIME_S(inode->i_ctime) = body->ctime;
1679 lli->lli_lvb.lvb_ctime = body->ctime;
1681 if (body->valid & OBD_MD_FLMODE)
1682 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1683 if (body->valid & OBD_MD_FLTYPE)
1684 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1685 LASSERT(inode->i_mode != 0);
1686 if (S_ISREG(inode->i_mode)) {
1687 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1689 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1691 #ifdef HAVE_INODE_BLKSIZE
1692 inode->i_blksize = 1<<inode->i_blkbits;
1694 if (body->valid & OBD_MD_FLUID)
1695 inode->i_uid = body->uid;
1696 if (body->valid & OBD_MD_FLGID)
1697 inode->i_gid = body->gid;
1698 if (body->valid & OBD_MD_FLFLAGS)
1699 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1700 if (body->valid & OBD_MD_FLNLINK)
1701 inode->i_nlink = body->nlink;
1702 if (body->valid & OBD_MD_FLRDEV)
1703 inode->i_rdev = old_decode_dev(body->rdev);
1705 if (body->valid & OBD_MD_FLID) {
1706 /* FID shouldn't be changed! */
1707 if (fid_is_sane(&lli->lli_fid)) {
1708 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1709 "Trying to change FID "DFID
1710 " to the "DFID", inode %lu/%u(%p)\n",
1711 PFID(&lli->lli_fid), PFID(&body->fid1),
1712 inode->i_ino, inode->i_generation, inode);
1714 lli->lli_fid = body->fid1;
1717 LASSERT(fid_seq(&lli->lli_fid) != 0);
1719 if (body->valid & OBD_MD_FLSIZE) {
1720 if (exp_connect_som(ll_i2mdexp(inode)) &&
1721 S_ISREG(inode->i_mode) && lli->lli_smd) {
1722 struct lustre_handle lockh;
1725 /* As it is possible a blocking ast has been processed
1726 * by this time, we need to check there is an UPDATE
1727 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1729 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1732 if (lli->lli_flags & (LLIF_DONE_WRITING |
1733 LLIF_EPOCH_PENDING |
1735 CERROR("ino %lu flags %u still has "
1736 "size authority! do not trust "
1737 "the size got from MDS\n",
1738 inode->i_ino, lli->lli_flags);
1740 /* Use old size assignment to avoid
1741 * deadlock bz14138 & bz14326 */
1742 i_size_write(inode, body->size);
1743 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1745 ldlm_lock_decref(&lockh, mode);
1748 /* Use old size assignment to avoid
1749 * deadlock bz14138 & bz14326 */
1750 i_size_write(inode, body->size);
1752 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1753 inode->i_ino, (unsigned long long)body->size);
1756 if (body->valid & OBD_MD_FLBLOCKS)
1757 inode->i_blocks = body->blocks;
1760 if (body->valid & OBD_MD_FLMDSCAPA) {
1761 LASSERT(md->mds_capa);
1762 ll_add_capa(inode, md->mds_capa);
1764 if (body->valid & OBD_MD_FLOSSCAPA) {
1765 LASSERT(md->oss_capa);
1766 ll_add_capa(inode, md->oss_capa);
1770 void ll_read_inode2(struct inode *inode, void *opaque)
1772 struct lustre_md *md = opaque;
1773 struct ll_inode_info *lli = ll_i2info(inode);
1776 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1777 PFID(&lli->lli_fid), inode);
1779 LASSERT(!lli->lli_smd);
1781 /* Core attributes from the MDS first. This is a new inode, and
1782 * the VFS doesn't zero times in the core inode so we have to do
1783 * it ourselves. They will be overwritten by either MDS or OST
1784 * attributes - we just need to make sure they aren't newer. */
1785 LTIME_S(inode->i_mtime) = 0;
1786 LTIME_S(inode->i_atime) = 0;
1787 LTIME_S(inode->i_ctime) = 0;
1789 ll_update_inode(inode, md);
1791 /* OIDEBUG(inode); */
1793 /* initializing backing dev info. */
1794 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1797 if (S_ISREG(inode->i_mode)) {
1798 struct ll_sb_info *sbi = ll_i2sbi(inode);
1799 inode->i_op = &ll_file_inode_operations;
1800 inode->i_fop = sbi->ll_fop;
1801 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1803 } else if (S_ISDIR(inode->i_mode)) {
1804 inode->i_op = &ll_dir_inode_operations;
1805 inode->i_fop = &ll_dir_operations;
1806 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1808 } else if (S_ISLNK(inode->i_mode)) {
1809 inode->i_op = &ll_fast_symlink_inode_operations;
1812 inode->i_op = &ll_special_inode_operations;
1814 init_special_inode(inode, inode->i_mode,
1815 kdev_t_to_nr(inode->i_rdev));
1821 void ll_delete_inode(struct inode *inode)
1823 struct ll_sb_info *sbi = ll_i2sbi(inode);
1827 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1829 CERROR("fid_delete() failed, rc %d\n", rc);
1831 truncate_inode_pages(&inode->i_data, 0);
1833 /* Workaround for LU-118 */
1834 if (inode->i_data.nrpages) {
1835 TREE_READ_LOCK_IRQ(&inode->i_data);
1836 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1837 LASSERTF(inode->i_data.nrpages == 0,
1838 "inode=%lu/%u(%p) nrpages=%lu, see "
1839 "http://jira.whamcloud.com/browse/LU-118\n",
1840 inode->i_ino, inode->i_generation, inode,
1841 inode->i_data.nrpages);
1843 /* Workaround end */
1845 #ifdef HAVE_SBOPS_EVICT_INODE
1846 ll_clear_inode(inode);
1847 end_writeback(inode);
1855 int ll_iocontrol(struct inode *inode, struct file *file,
1856 unsigned int cmd, unsigned long arg)
1858 struct ll_sb_info *sbi = ll_i2sbi(inode);
1859 struct ptlrpc_request *req = NULL;
1864 case FSFILT_IOC_GETFLAGS: {
1865 struct mdt_body *body;
1866 struct md_op_data *op_data;
1868 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1869 0, 0, LUSTRE_OPC_ANY,
1871 if (IS_ERR(op_data))
1872 RETURN(PTR_ERR(op_data));
1874 op_data->op_valid = OBD_MD_FLFLAGS;
1875 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1876 ll_finish_md_op_data(op_data);
1878 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1882 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1884 flags = body->flags;
1886 ptlrpc_req_finished(req);
1888 RETURN(put_user(flags, (int *)arg));
1890 case FSFILT_IOC_SETFLAGS: {
1891 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1892 struct obd_info oinfo = { { { 0 } } };
1893 struct md_op_data *op_data;
1895 if (get_user(flags, (int *)arg))
1899 OBDO_ALLOC(oinfo.oi_oa);
1903 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1904 LUSTRE_OPC_ANY, NULL);
1905 if (IS_ERR(op_data))
1906 RETURN(PTR_ERR(op_data));
1908 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1909 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1910 rc = md_setattr(sbi->ll_md_exp, op_data,
1911 NULL, 0, NULL, 0, &req, NULL);
1912 ll_finish_md_op_data(op_data);
1913 ptlrpc_req_finished(req);
1915 OBDO_FREE(oinfo.oi_oa);
1920 OBDO_FREE(oinfo.oi_oa);
1921 GOTO(update_cache, rc);
1924 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1925 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1926 oinfo.oi_oa->o_flags = flags;
1927 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1929 oinfo.oi_capa = ll_mdscapa_get(inode);
1930 obdo_from_inode(oinfo.oi_oa, inode,
1931 &ll_i2info(inode)->lli_fid, 0);
1932 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1933 capa_put(oinfo.oi_capa);
1934 OBDO_FREE(oinfo.oi_oa);
1936 if (rc != -EPERM && rc != -EACCES)
1937 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1943 inode->i_flags = ll_ext_to_inode_flags(flags);
1953 int ll_flush_ctx(struct inode *inode)
1955 struct ll_sb_info *sbi = ll_i2sbi(inode);
1957 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1959 obd_set_info_async(sbi->ll_md_exp,
1960 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1962 obd_set_info_async(sbi->ll_dt_exp,
1963 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1968 /* umount -f client means force down, don't save state */
1969 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1970 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1972 struct super_block *sb = vfsmnt->mnt_sb;
1974 void ll_umount_begin(struct super_block *sb)
1977 struct lustre_sb_info *lsi = s2lsi(sb);
1978 struct ll_sb_info *sbi = ll_s2sbi(sb);
1979 struct obd_device *obd;
1980 struct obd_ioctl_data *ioc_data;
1983 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1984 if (!(flags & MNT_FORCE)) {
1990 /* Tell the MGC we got umount -f */
1991 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1993 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1994 sb->s_count, atomic_read(&sb->s_active));
1996 obd = class_exp2obd(sbi->ll_md_exp);
1998 CERROR("Invalid MDC connection handle "LPX64"\n",
1999 sbi->ll_md_exp->exp_handle.h_cookie);
2005 obd = class_exp2obd(sbi->ll_dt_exp);
2007 CERROR("Invalid LOV connection handle "LPX64"\n",
2008 sbi->ll_dt_exp->exp_handle.h_cookie);
2014 OBD_ALLOC_PTR(ioc_data);
2016 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2017 sizeof ioc_data, ioc_data, NULL);
2019 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2020 sizeof ioc_data, ioc_data, NULL);
2022 OBD_FREE_PTR(ioc_data);
2026 /* Really, we'd like to wait until there are no requests outstanding,
2027 * and then continue. For now, we just invalidate the requests,
2028 * schedule() and sleep one second if needed, and hope.
2031 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2032 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2033 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2034 cfs_time_seconds(1));
2035 if (atomic_read(&vfsmnt->mnt_count) > 2)
2036 LCONSOLE_WARN("Mount still busy with %d refs! You "
2037 "may try to umount it a bit later\n",
2038 atomic_read(&vfsmnt->mnt_count));
2045 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2047 struct ll_sb_info *sbi = ll_s2sbi(sb);
2051 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2052 read_only = *flags & MS_RDONLY;
2053 err = obd_set_info_async(sbi->ll_md_exp,
2054 sizeof(KEY_READ_ONLY),
2055 KEY_READ_ONLY, sizeof(read_only),
2058 CERROR("Failed to change the read-only flag during "
2059 "remount: %d\n", err);
2064 sb->s_flags |= MS_RDONLY;
2066 sb->s_flags &= ~MS_RDONLY;
2071 int ll_prep_inode(struct inode **inode,
2072 struct ptlrpc_request *req,
2073 struct super_block *sb)
2075 struct ll_sb_info *sbi = NULL;
2076 struct lustre_md md;
2080 LASSERT(*inode || sb);
2081 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2082 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2083 sbi->ll_md_exp, &md);
2088 ll_update_inode(*inode, &md);
2090 LASSERT(sb != NULL);
2093 * At this point server returns to client's same fid as client
2094 * generated for creating. So using ->fid1 is okay here.
2096 LASSERT(fid_is_sane(&md.body->fid1));
2098 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2099 if (*inode == NULL || IS_ERR(*inode)) {
2101 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2102 #ifdef CONFIG_FS_POSIX_ACL
2104 posix_acl_release(md.posix_acl);
2105 md.posix_acl = NULL;
2108 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2110 CERROR("new_inode -fatal: rc %d\n", rc);
2116 md_free_lustre_md(sbi->ll_md_exp, &md);
2120 int ll_obd_statfs(struct inode *inode, void *arg)
2122 struct ll_sb_info *sbi = NULL;
2123 struct obd_export *exp;
2125 struct obd_ioctl_data *data = NULL;
2129 if (!inode || !(sbi = ll_i2sbi(inode)))
2130 GOTO(out_statfs, rc = -EINVAL);
2132 rc = obd_ioctl_getdata(&buf, &len, arg);
2134 GOTO(out_statfs, rc);
2137 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2138 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2139 GOTO(out_statfs, rc = -EINVAL);
2141 if (data->ioc_inllen1 != sizeof(__u32) ||
2142 data->ioc_inllen2 != sizeof(__u32) ||
2143 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2144 data->ioc_plen2 != sizeof(struct obd_uuid))
2145 GOTO(out_statfs, rc = -EINVAL);
2147 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2148 if (type == LL_STATFS_LMV)
2149 exp = sbi->ll_md_exp;
2150 else if (type == LL_STATFS_LOV)
2151 exp = sbi->ll_dt_exp;
2153 GOTO(out_statfs, rc = -ENODEV);
2155 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2157 GOTO(out_statfs, rc);
2160 obd_ioctl_freedata(buf, len);
2164 int ll_process_config(struct lustre_cfg *lcfg)
2168 struct lprocfs_static_vars lvars;
2172 lprocfs_llite_init_vars(&lvars);
2174 /* The instance name contains the sb: lustre-client-aacfe000 */
2175 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2176 if (!ptr || !*(++ptr))
2178 if (sscanf(ptr, "%lx", &x) != 1)
2181 /* This better be a real Lustre superblock! */
2182 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2184 /* Note we have not called client_common_fill_super yet, so
2185 proc fns must be able to handle that! */
2186 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2193 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2194 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2195 struct inode *i1, struct inode *i2,
2196 const char *name, int namelen,
2197 int mode, __u32 opc, void *data)
2199 LASSERT(i1 != NULL);
2201 if (namelen > ll_i2sbi(i1)->ll_namelen)
2202 return ERR_PTR(-ENAMETOOLONG);
2204 if (op_data == NULL)
2205 OBD_ALLOC_PTR(op_data);
2207 if (op_data == NULL)
2208 return ERR_PTR(-ENOMEM);
2210 ll_i2gids(op_data->op_suppgids, i1, i2);
2211 op_data->op_fid1 = *ll_inode2fid(i1);
2212 op_data->op_capa1 = ll_mdscapa_get(i1);
2215 op_data->op_fid2 = *ll_inode2fid(i2);
2216 op_data->op_capa2 = ll_mdscapa_get(i2);
2218 fid_zero(&op_data->op_fid2);
2219 op_data->op_capa2 = NULL;
2222 op_data->op_name = name;
2223 op_data->op_namelen = namelen;
2224 op_data->op_mode = mode;
2225 op_data->op_mod_time = cfs_time_current_sec();
2226 op_data->op_fsuid = cfs_curproc_fsuid();
2227 op_data->op_fsgid = cfs_curproc_fsgid();
2228 op_data->op_cap = cfs_curproc_cap_pack();
2229 op_data->op_bias = MDS_CHECK_SPLIT;
2230 op_data->op_opc = opc;
2231 op_data->op_mds = 0;
2232 op_data->op_data = data;
2234 /* If the file is being opened after mknod() (normally due to NFS)
2235 * try to use the default stripe data from parent directory for
2236 * allocating OST objects. Try to pass the parent FID to MDS. */
2237 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2238 ll_i2info(i2)->lli_smd == NULL) {
2239 struct ll_inode_info *lli = ll_i2info(i2);
2241 cfs_spin_lock(&lli->lli_lock);
2242 if (likely(lli->lli_smd == NULL &&
2243 !fid_is_zero(&lli->lli_pfid)))
2244 op_data->op_fid1 = lli->lli_pfid;
2245 cfs_spin_unlock(&lli->lli_lock);
2246 /** We ignore parent's capability temporary. */
2252 void ll_finish_md_op_data(struct md_op_data *op_data)
2254 capa_put(op_data->op_capa1);
2255 capa_put(op_data->op_capa2);
2256 OBD_FREE_PTR(op_data);
2259 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2261 struct ll_sb_info *sbi;
2263 LASSERT((seq != NULL) && (vfs != NULL));
2264 sbi = ll_s2sbi(vfs->mnt_sb);
2266 if (sbi->ll_flags & LL_SBI_NOLCK)
2267 seq_puts(seq, ",nolock");
2269 if (sbi->ll_flags & LL_SBI_FLOCK)
2270 seq_puts(seq, ",flock");
2272 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2273 seq_puts(seq, ",localflock");
2275 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2276 seq_puts(seq, ",user_xattr");
2278 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2279 seq_puts(seq, ",lazystatfs");
2285 * Get obd name by cmd, and copy out to user space
2287 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2289 struct ll_sb_info *sbi = ll_i2sbi(inode);
2290 struct obd_device *obd;
2293 if (cmd == OBD_IOC_GETDTNAME)
2294 obd = class_exp2obd(sbi->ll_dt_exp);
2295 else if (cmd == OBD_IOC_GETMDNAME)
2296 obd = class_exp2obd(sbi->ll_md_exp);
2303 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2304 strlen(obd->obd_name) + 1))