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.
33 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/llite/llite_lib.c
41 * Lustre Light Super operations
44 #define DEBUG_SUBSYSTEM S_LLITE
46 #include <linux/module.h>
47 #include <linux/types.h>
48 #include <linux/version.h>
51 #include <lustre_lite.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 cfs_mem_cache_t *ll_file_data_slab;
64 CFS_LIST_HEAD(ll_super_blocks);
65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
67 #ifndef MS_HAS_NEW_AOPS
68 extern struct address_space_operations ll_aops;
69 extern struct address_space_operations ll_dir_aops;
71 extern struct address_space_operations_ext ll_aops;
72 extern struct address_space_operations_ext ll_dir_aops;
76 #define log2(n) cfs_ffz(~(n))
79 static struct ll_sb_info *ll_init_sbi(void)
81 struct ll_sb_info *sbi = NULL;
88 OBD_ALLOC(sbi, sizeof(*sbi));
92 cfs_spin_lock_init(&sbi->ll_lock);
93 cfs_init_mutex(&sbi->ll_lco.lco_lock);
94 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
95 cfs_spin_lock_init(&sbi->ll_process_lock);
96 sbi->ll_rw_stats_on = 0;
99 pages = si.totalram - si.totalhigh;
100 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
101 #ifdef HAVE_BGL_SUPPORT
102 sbi->ll_async_page_max = pages / 4;
104 sbi->ll_async_page_max = pages / 2;
107 sbi->ll_async_page_max = (pages / 4) * 3;
110 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
111 SBI_DEFAULT_READAHEAD_MAX);
112 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
113 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
114 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
115 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
116 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
118 ll_generate_random_uuid(uuid);
119 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
120 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
122 cfs_spin_lock(&ll_sb_lock);
123 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
124 cfs_spin_unlock(&ll_sb_lock);
126 #ifdef ENABLE_CHECKSUM
127 sbi->ll_flags |= LL_SBI_CHECKSUM;
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
134 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
135 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
137 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
141 /* metadata statahead is enabled by default */
142 sbi->ll_sa_max = LL_SA_RPC_DEF;
143 cfs_atomic_set(&sbi->ll_sa_total, 0);
144 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
149 void ll_free_sbi(struct super_block *sb)
151 struct ll_sb_info *sbi = ll_s2sbi(sb);
155 cfs_spin_lock(&ll_sb_lock);
156 cfs_list_del(&sbi->ll_list);
157 cfs_spin_unlock(&ll_sb_lock);
158 OBD_FREE(sbi, sizeof(*sbi));
163 static struct dentry_operations ll_d_root_ops = {
164 .d_compare = ll_dcompare,
165 .d_revalidate = ll_revalidate_nd,
168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
169 struct vfsmount *mnt)
171 struct inode *root = 0;
172 struct ll_sb_info *sbi = ll_s2sbi(sb);
173 struct obd_device *obd;
174 struct obd_capa *oc = NULL;
175 struct obd_statfs *osfs = NULL;
176 struct ptlrpc_request *request = NULL;
177 struct obd_connect_data *data = NULL;
178 struct obd_uuid *uuid;
179 struct md_op_data *op_data;
180 struct lustre_md lmd;
182 int size, err, checksum;
185 obd = class_name2obd(md);
187 CERROR("MD %s: not setup or attached\n", md);
201 if (proc_lustre_fs_root) {
202 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
205 CERROR("could not register mount in /proc/fs/lustre\n");
208 /* indicate the features supported by this client */
209 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
210 OBD_CONNECT_ATTRFID |
211 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
212 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
213 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
214 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
215 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
216 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH;
218 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
219 data->ocd_connect_flags |= OBD_CONNECT_SOM;
221 #ifdef HAVE_LRU_RESIZE_SUPPORT
222 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
223 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
225 #ifdef CONFIG_FS_POSIX_ACL
226 data->ocd_connect_flags |= OBD_CONNECT_ACL;
228 data->ocd_ibits_known = MDS_INODELOCK_FULL;
229 data->ocd_version = LUSTRE_VERSION_CODE;
231 if (sb->s_flags & MS_RDONLY)
232 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
233 if (sbi->ll_flags & LL_SBI_USER_XATTR)
234 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
236 #ifdef HAVE_MS_FLOCK_LOCK
237 /* force vfs to use lustre handler for flock() calls - bug 10743 */
238 sb->s_flags |= MS_FLOCK_LOCK;
240 #ifdef MS_HAS_NEW_AOPS
241 sb->s_flags |= MS_HAS_NEW_AOPS;
244 if (sbi->ll_flags & LL_SBI_FLOCK)
245 sbi->ll_fop = &ll_file_operations_flock;
246 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
247 sbi->ll_fop = &ll_file_operations;
249 sbi->ll_fop = &ll_file_operations_noflock;
252 data->ocd_connect_flags |= OBD_CONNECT_REAL;
253 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
254 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
256 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
258 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
260 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
261 "recovery, of which this client is not a "
262 "part. Please wait for recovery to complete,"
263 " abort, or time out.\n", md);
266 CERROR("cannot connect to %s: rc = %d\n", md, err);
270 err = obd_fid_init(sbi->ll_md_exp);
272 CERROR("Can't init metadata layer FID infrastructure, "
277 err = obd_statfs(obd, osfs,
278 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
280 GOTO(out_md_fid, err);
282 /* This needs to be after statfs to ensure connect has finished.
283 * Note that "data" does NOT contain the valid connect reply.
284 * If connecting to a 1.8 server there will be no LMV device, so
285 * we can access the MDC export directly and exp_connect_flags will
286 * be non-zero, but if accessing an upgraded 2.1 server it will
287 * have the correct flags filled in.
288 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
289 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
290 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
291 valid != CLIENT_CONNECT_MDT_REQD) {
294 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
295 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
296 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
297 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
298 "feature(s) needed for correct operation "
299 "of this client (%s). Please upgrade "
300 "server or downgrade client.\n",
301 sbi->ll_md_exp->exp_obd->obd_name, buf);
302 OBD_FREE(buf, CFS_PAGE_SIZE);
303 GOTO(out_md, err = -EPROTO);
306 size = sizeof(*data);
307 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
308 KEY_CONN_DATA, &size, data, NULL);
310 CERROR("Get connect data failed: %d \n", err);
314 LASSERT(osfs->os_bsize);
315 sb->s_blocksize = osfs->os_bsize;
316 sb->s_blocksize_bits = log2(osfs->os_bsize);
317 sb->s_magic = LL_SUPER_MAGIC;
319 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
320 * retval = in_file->f_op->sendfile(...);
322 * retval = -EOVERFLOW;
324 * it will check if *ppos is greater than max. However, max equals to
325 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
326 * has been defined as a signed long long integer in linux kernel. */
327 #if BITS_PER_LONG == 64
328 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
330 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
332 sbi->ll_namelen = osfs->os_namelen;
333 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
335 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
336 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
337 LCONSOLE_INFO("Disabling user_xattr feature because "
338 "it is not supported on the server\n");
339 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
342 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
344 sb->s_flags |= MS_POSIXACL;
346 sbi->ll_flags |= LL_SBI_ACL;
348 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
350 sb->s_flags &= ~MS_POSIXACL;
352 sbi->ll_flags &= ~LL_SBI_ACL;
355 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
356 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
357 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
358 LCONSOLE_INFO("client is set as remote by default.\n");
361 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
362 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
363 LCONSOLE_INFO("client claims to be remote, but server "
364 "rejected, forced to be local.\n");
368 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
369 LCONSOLE_INFO("client enabled MDS capability!\n");
370 sbi->ll_flags |= LL_SBI_MDS_CAPA;
373 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
374 LCONSOLE_INFO("client enabled OSS capability!\n");
375 sbi->ll_flags |= LL_SBI_OSS_CAPA;
378 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
379 sbi->ll_flags |= LL_SBI_64BIT_HASH;
381 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
382 sbi->ll_md_brw_size = data->ocd_brw_size;
384 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
386 obd = class_name2obd(dt);
388 CERROR("DT %s: not setup or attached\n", dt);
389 GOTO(out_md_fid, err = -ENODEV);
392 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
393 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
394 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
395 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
396 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
397 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
398 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
399 OBD_CONNECT_MAXBYTES;
401 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
402 data->ocd_connect_flags |= OBD_CONNECT_SOM;
404 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
405 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
406 * disabled by default, because it can still be enabled on the
407 * fly via /proc. As a consequence, we still need to come to an
408 * agreement on the supported algorithms at connect time */
409 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
411 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
412 data->ocd_cksum_types = OBD_CKSUM_ADLER;
414 data->ocd_cksum_types = cksum_types_supported();
417 #ifdef HAVE_LRU_RESIZE_SUPPORT
418 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
420 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
421 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
423 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
424 "ocd_grant: %d\n", data->ocd_connect_flags,
425 data->ocd_version, data->ocd_grant);
427 obd->obd_upcall.onu_owner = &sbi->ll_lco;
428 obd->obd_upcall.onu_upcall = cl_ocd_update;
430 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
432 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
434 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
435 "recovery, of which this client is not a "
436 "part. Please wait for recovery to "
437 "complete, abort, or time out.\n", dt);
438 GOTO(out_md_fid, err);
440 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
441 GOTO(out_md_fid, err);
444 err = obd_fid_init(sbi->ll_dt_exp);
446 CERROR("Can't init data layer FID infrastructure, "
451 cfs_mutex_down(&sbi->ll_lco.lco_lock);
452 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
453 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
454 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
455 cfs_mutex_up(&sbi->ll_lco.lco_lock);
457 fid_zero(&sbi->ll_root_fid);
458 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
460 CERROR("cannot mds_connect: rc = %d\n", err);
461 GOTO(out_lock_cn_cb, err);
463 if (!fid_is_sane(&sbi->ll_root_fid)) {
464 CERROR("Invalid root fid during mount\n");
465 GOTO(out_lock_cn_cb, err = -EINVAL);
467 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
469 sb->s_op = &lustre_super_operations;
470 #if THREAD_SIZE >= 8192
471 sb->s_export_op = &lustre_export_operations;
475 * XXX: move this to after cbd setup? */
476 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
477 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
478 valid |= OBD_MD_FLRMTPERM;
479 else if (sbi->ll_flags & LL_SBI_ACL)
480 valid |= OBD_MD_FLACL;
482 OBD_ALLOC_PTR(op_data);
484 GOTO(out_lock_cn_cb, err = -ENOMEM);
486 op_data->op_fid1 = sbi->ll_root_fid;
487 op_data->op_mode = 0;
488 op_data->op_capa1 = oc;
489 op_data->op_valid = valid;
491 err = md_getattr(sbi->ll_md_exp, op_data, &request);
494 OBD_FREE_PTR(op_data);
496 CERROR("md_getattr failed for root: rc = %d\n", err);
497 GOTO(out_lock_cn_cb, err);
499 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
500 sbi->ll_md_exp, &lmd);
502 CERROR("failed to understand root inode md: rc = %d\n", err);
503 ptlrpc_req_finished (request);
504 GOTO(out_lock_cn_cb, err);
507 LASSERT(fid_is_sane(&sbi->ll_root_fid));
508 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
509 md_free_lustre_md(sbi->ll_md_exp, &lmd);
510 ptlrpc_req_finished(request);
512 if (root == NULL || IS_ERR(root)) {
514 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
515 #ifdef CONFIG_FS_POSIX_ACL
517 posix_acl_release(lmd.posix_acl);
518 lmd.posix_acl = NULL;
521 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
523 CERROR("lustre_lite: bad iget4 for root\n");
527 err = ll_close_thread_start(&sbi->ll_lcq);
529 CERROR("cannot start close thread: rc %d\n", err);
533 #ifdef CONFIG_FS_POSIX_ACL
534 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
535 rct_init(&sbi->ll_rct);
536 et_init(&sbi->ll_et);
540 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
541 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
542 KEY_CHECKSUM, sizeof(checksum), &checksum,
546 sb->s_root = d_alloc_root(root);
547 sb->s_root->d_op = &ll_d_root_ops;
549 sbi->ll_sdev_orig = sb->s_dev;
551 /* We set sb->s_dev equal on all lustre clients in order to support
552 * NFS export clustering. NFSD requires that the FSID be the same
554 /* s_dev is also used in lt_compare() to compare two fs, but that is
555 * only a node-local comparison. */
556 uuid = obd_get_uuid(sbi->ll_md_exp);
558 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
571 obd_fid_fini(sbi->ll_dt_exp);
573 obd_disconnect(sbi->ll_dt_exp);
574 sbi->ll_dt_exp = NULL;
576 obd_fid_fini(sbi->ll_md_exp);
578 obd_disconnect(sbi->ll_md_exp);
579 sbi->ll_md_exp = NULL;
585 lprocfs_unregister_mountpoint(sbi);
589 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
593 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
595 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
596 KEY_MAX_EASIZE, &size, lmmsize, NULL);
598 CERROR("Get max mdsize error rc %d \n", rc);
603 void ll_dump_inode(struct inode *inode)
605 struct list_head *tmp;
606 int dentry_count = 0;
608 LASSERT(inode != NULL);
610 list_for_each(tmp, &inode->i_dentry)
613 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
614 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
615 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
618 void lustre_dump_dentry(struct dentry *dentry, int recur)
620 struct list_head *tmp;
623 LASSERT(dentry != NULL);
625 list_for_each(tmp, &dentry->d_subdirs)
628 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
629 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
630 dentry->d_name.len, dentry->d_name.name,
631 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
632 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
633 dentry->d_flags, dentry->d_fsdata, subdirs);
634 if (dentry->d_inode != NULL)
635 ll_dump_inode(dentry->d_inode);
640 list_for_each(tmp, &dentry->d_subdirs) {
641 struct dentry *d = list_entry(tmp, struct dentry, d_child);
642 lustre_dump_dentry(d, recur - 1);
646 void client_common_put_super(struct super_block *sb)
648 struct ll_sb_info *sbi = ll_s2sbi(sb);
651 #ifdef CONFIG_FS_POSIX_ACL
652 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
653 et_fini(&sbi->ll_et);
654 rct_fini(&sbi->ll_rct);
658 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
660 ll_close_thread_shutdown(sbi->ll_lcq);
664 cfs_list_del(&sbi->ll_conn_chain);
666 obd_fid_fini(sbi->ll_dt_exp);
667 obd_disconnect(sbi->ll_dt_exp);
668 sbi->ll_dt_exp = NULL;
670 lprocfs_unregister_mountpoint(sbi);
672 obd_fid_fini(sbi->ll_md_exp);
673 obd_disconnect(sbi->ll_md_exp);
674 sbi->ll_md_exp = NULL;
679 void ll_kill_super(struct super_block *sb)
681 struct ll_sb_info *sbi;
686 if (!(sb->s_flags & MS_ACTIVE))
690 /* we need restore s_dev from changed for clustred NFS before put_super
691 * because new kernels have cached s_dev and change sb->s_dev in
692 * put_super not affected real removing devices */
694 sb->s_dev = sbi->ll_sdev_orig;
698 char *ll_read_opt(const char *opt, char *data)
704 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
705 if (strncmp(opt, data, strlen(opt)))
707 if ((value = strchr(data, '=')) == NULL)
711 OBD_ALLOC(retval, strlen(value) + 1);
713 CERROR("out of memory!\n");
717 memcpy(retval, value, strlen(value)+1);
718 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
722 static inline int ll_set_opt(const char *opt, char *data, int fl)
724 if (strncmp(opt, data, strlen(opt)) != 0)
730 /* non-client-specific mount options are parsed in lmd_parse */
731 static int ll_options(char *options, int *flags)
734 char *s1 = options, *s2;
740 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
743 CDEBUG(D_SUPER, "next opt=%s\n", s1);
744 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
749 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
754 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
759 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
764 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
769 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
774 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
775 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
777 /* Ignore deprecated mount option. The client will
778 * always try to mount with ACL support, whether this
779 * is used depends on whether server supports it. */
780 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
781 "mount option 'acl'.\n");
784 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
786 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
787 "mount option 'noacl'.\n");
791 #warning "{no}acl options have been deprecated since 1.8, please remove them"
793 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
799 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
804 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
809 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
814 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
819 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
824 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
829 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
834 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
840 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
846 s2 = strchr(s1, ',');
854 void ll_lli_init(struct ll_inode_info *lli)
856 lli->lli_inode_magic = LLI_INODE_MAGIC;
857 cfs_sema_init(&lli->lli_size_sem, 1);
858 cfs_sema_init(&lli->lli_write_sem, 1);
859 cfs_init_rwsem(&lli->lli_trunc_sem);
861 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
862 cfs_spin_lock_init(&lli->lli_lock);
863 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
864 lli->lli_inode_magic = LLI_INODE_MAGIC;
865 cfs_sema_init(&lli->lli_och_sem, 1);
866 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
867 lli->lli_mds_exec_och = NULL;
868 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
869 lli->lli_open_fd_exec_count = 0;
870 CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
871 lli->lli_remote_perms = NULL;
872 lli->lli_rmtperm_utime = 0;
873 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
874 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
875 cfs_spin_lock_init(&lli->lli_sa_lock);
876 lli->lli_clob = NULL;
877 cfs_sema_init(&lli->lli_readdir_sem, 1);
878 fid_zero(&lli->lli_pfid);
881 static inline int ll_bdi_register(struct backing_dev_info *bdi)
883 #ifdef HAVE_BDI_REGISTER
884 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
887 bdi->name = "lustre";
889 return bdi_register(bdi, NULL, "lustre-%d",
890 atomic_inc_return(&ll_bdi_num));
896 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
898 struct lustre_profile *lprof = NULL;
899 struct lustre_sb_info *lsi = s2lsi(sb);
900 struct ll_sb_info *sbi;
901 char *dt = NULL, *md = NULL;
902 char *profilenm = get_profile_name(sb);
903 struct config_llog_instance *cfg;
904 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
905 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
909 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
917 /* client additional sb info */
918 lsi->lsi_llsbi = sbi = ll_init_sbi();
920 cfs_module_put(THIS_MODULE);
925 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
929 err = ll_bdi_init(&lsi->lsi_bdi);
932 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
933 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
934 err = ll_bdi_register(&lsi->lsi_bdi);
939 sb->s_bdi = &lsi->lsi_bdi;
942 /* Generate a string unique to this super, in case some joker tries
943 to mount the same fs at two mount points.
944 Use the address of the super itself.*/
945 cfg->cfg_instance = sb;
946 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
948 /* set up client obds */
949 err = lustre_process_log(sb, profilenm, cfg);
951 CERROR("Unable to process log: %d\n", err);
955 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
956 lprof = class_get_profile(profilenm);
958 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
959 " read from the MGS. Does that filesystem "
960 "exist?\n", profilenm);
961 GOTO(out_free, err = -EINVAL);
963 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
964 lprof->lp_md, lprof->lp_dt);
966 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
968 GOTO(out_free, err = -ENOMEM);
969 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
971 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
973 GOTO(out_free, err = -ENOMEM);
974 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
976 /* connections, registrations, sb setup */
977 err = client_common_fill_super(sb, md, dt, mnt);
981 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
983 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
987 LCONSOLE_WARN("Client %s has started\n", profilenm);
991 } /* ll_fill_super */
994 void lu_context_keys_dump(void);
996 void ll_put_super(struct super_block *sb)
998 struct config_llog_instance cfg;
999 struct obd_device *obd;
1000 struct lustre_sb_info *lsi = s2lsi(sb);
1001 struct ll_sb_info *sbi = ll_s2sbi(sb);
1002 char *profilenm = get_profile_name(sb);
1003 int force = 1, next;
1006 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1008 ll_print_capa_stat(sbi);
1010 cfg.cfg_instance = sb;
1011 lustre_end_log(sb, profilenm, &cfg);
1013 if (sbi->ll_md_exp) {
1014 obd = class_exp2obd(sbi->ll_md_exp);
1016 force = obd->obd_force;
1019 /* We need to set force before the lov_disconnect in
1020 lustre_common_put_super, since l_d cleans up osc's as well. */
1023 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1025 obd->obd_force = force;
1030 /* Only if client_common_fill_super succeeded */
1031 client_common_put_super(sb);
1035 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1036 class_manual_cleanup(obd);
1040 class_del_profile(profilenm);
1042 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1043 ll_bdi_destroy(&lsi->lsi_bdi);
1044 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1048 lsi->lsi_llsbi = NULL;
1050 lustre_common_put_super(sb);
1052 cl_env_cache_purge(~0);
1054 LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1056 cfs_module_put(THIS_MODULE);
1059 } /* client_put_super */
1061 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1063 struct inode *inode = NULL;
1064 /* NOTE: we depend on atomic igrab() -bzzz */
1065 lock_res_and_lock(lock);
1066 if (lock->l_ast_data) {
1067 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1068 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1069 inode = igrab(lock->l_ast_data);
1071 inode = lock->l_ast_data;
1072 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1073 D_WARNING, lock, "l_ast_data %p is "
1074 "bogus: magic %08x", lock->l_ast_data,
1075 lli->lli_inode_magic);
1079 unlock_res_and_lock(lock);
1083 static int null_if_equal(struct ldlm_lock *lock, void *data)
1085 if (data == lock->l_ast_data) {
1086 lock->l_ast_data = NULL;
1088 if (lock->l_req_mode != lock->l_granted_mode)
1089 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1092 return LDLM_ITER_CONTINUE;
1095 void ll_clear_inode(struct inode *inode)
1097 struct ll_inode_info *lli = ll_i2info(inode);
1098 struct ll_sb_info *sbi = ll_i2sbi(inode);
1101 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1102 inode->i_generation, inode);
1104 if (S_ISDIR(inode->i_mode)) {
1105 /* these should have been cleared in ll_file_release */
1106 LASSERT(lli->lli_sai == NULL);
1107 LASSERT(lli->lli_opendir_key == NULL);
1108 LASSERT(lli->lli_opendir_pid == 0);
1111 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1112 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1113 null_if_equal, inode);
1115 LASSERT(!lli->lli_open_fd_write_count);
1116 LASSERT(!lli->lli_open_fd_read_count);
1117 LASSERT(!lli->lli_open_fd_exec_count);
1119 if (lli->lli_mds_write_och)
1120 ll_md_real_close(inode, FMODE_WRITE);
1121 if (lli->lli_mds_exec_och)
1122 ll_md_real_close(inode, FMODE_EXEC);
1123 if (lli->lli_mds_read_och)
1124 ll_md_real_close(inode, FMODE_READ);
1126 if (lli->lli_symlink_name) {
1127 OBD_FREE(lli->lli_symlink_name,
1128 strlen(lli->lli_symlink_name) + 1);
1129 lli->lli_symlink_name = NULL;
1132 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1133 LASSERT(lli->lli_posix_acl == NULL);
1134 if (lli->lli_remote_perms) {
1135 free_rmtperm_hash(lli->lli_remote_perms);
1136 lli->lli_remote_perms = NULL;
1139 #ifdef CONFIG_FS_POSIX_ACL
1140 else if (lli->lli_posix_acl) {
1141 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1142 LASSERT(lli->lli_remote_perms == NULL);
1143 posix_acl_release(lli->lli_posix_acl);
1144 lli->lli_posix_acl = NULL;
1147 lli->lli_inode_magic = LLI_INODE_DEAD;
1149 ll_clear_inode_capas(inode);
1151 * XXX This has to be done before lsm is freed below, because
1152 * cl_object still uses inode lsm.
1154 cl_inode_fini(inode);
1157 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1158 lli->lli_smd = NULL;
1165 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1166 struct md_open_data **mod)
1168 struct lustre_md md;
1169 struct ll_sb_info *sbi = ll_i2sbi(inode);
1170 struct ptlrpc_request *request = NULL;
1174 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1175 LUSTRE_OPC_ANY, NULL);
1176 if (IS_ERR(op_data))
1177 RETURN(PTR_ERR(op_data));
1179 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1182 ptlrpc_req_finished(request);
1183 if (rc == -ENOENT) {
1185 /* Unlinked special device node? Or just a race?
1186 * Pretend we done everything. */
1187 if (!S_ISREG(inode->i_mode) &&
1188 !S_ISDIR(inode->i_mode))
1189 rc = inode_setattr(inode, &op_data->op_attr);
1190 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1191 CERROR("md_setattr fails: rc = %d\n", rc);
1196 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1197 sbi->ll_md_exp, &md);
1199 ptlrpc_req_finished(request);
1203 /* We call inode_setattr to adjust timestamps.
1204 * If there is at least some data in file, we cleared ATTR_SIZE
1205 * above to avoid invoking vmtruncate, otherwise it is important
1206 * to call vmtruncate in inode_setattr to update inode->i_size
1208 rc = inode_setattr(inode, &op_data->op_attr);
1210 /* Extract epoch data if obtained. */
1211 op_data->op_handle = md.body->handle;
1212 op_data->op_ioepoch = md.body->ioepoch;
1214 ll_update_inode(inode, &md);
1215 ptlrpc_req_finished(request);
1220 /* Close IO epoch and send Size-on-MDS attribute update. */
1221 static int ll_setattr_done_writing(struct inode *inode,
1222 struct md_op_data *op_data,
1223 struct md_open_data *mod)
1225 struct ll_inode_info *lli = ll_i2info(inode);
1229 LASSERT(op_data != NULL);
1230 if (!S_ISREG(inode->i_mode))
1233 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1234 op_data->op_ioepoch, PFID(&lli->lli_fid));
1236 op_data->op_flags = MF_EPOCH_CLOSE;
1237 ll_done_writing_attr(inode, op_data);
1238 ll_pack_inode2opdata(inode, op_data, NULL);
1240 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1241 if (rc == -EAGAIN) {
1242 /* MDS has instructed us to obtain Size-on-MDS attribute
1243 * from OSTs and send setattr to back to MDS. */
1244 rc = ll_som_update(inode, op_data);
1246 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1252 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1254 struct obd_capa *capa;
1257 if (attr->ia_valid & ATTR_SIZE)
1258 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1260 capa = ll_mdscapa_get(inode);
1262 rc = cl_setattr_ost(inode, attr, capa);
1264 if (attr->ia_valid & ATTR_SIZE)
1265 ll_truncate_free_capa(capa);
1272 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1273 * object(s) determine the file size and mtime. Otherwise, the MDS will
1274 * keep these values until such a time that objects are allocated for it.
1275 * We do the MDS operations first, as it is checking permissions for us.
1276 * We don't to the MDS RPC if there is nothing that we want to store there,
1277 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1278 * going to do an RPC anyways.
1280 * If we are doing a truncate, we will send the mtime and ctime updates
1281 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1282 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1285 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1287 struct ll_inode_info *lli = ll_i2info(inode);
1288 struct lov_stripe_md *lsm = lli->lli_smd;
1289 struct md_op_data *op_data = NULL;
1290 struct md_open_data *mod = NULL;
1291 int ia_valid = attr->ia_valid;
1292 int rc = 0, rc1 = 0;
1295 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1297 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1299 if (ia_valid & ATTR_SIZE) {
1300 if (attr->ia_size > ll_file_maxbytes(inode)) {
1301 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1302 attr->ia_size, ll_file_maxbytes(inode));
1306 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1309 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1310 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1311 if (cfs_curproc_fsuid() != inode->i_uid &&
1312 !cfs_capable(CFS_CAP_FOWNER))
1316 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1317 if (attr->ia_valid & ATTR_CTIME) {
1318 attr->ia_ctime = CFS_CURRENT_TIME;
1319 attr->ia_valid |= ATTR_CTIME_SET;
1321 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1322 attr->ia_atime = CFS_CURRENT_TIME;
1323 attr->ia_valid |= ATTR_ATIME_SET;
1325 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1326 attr->ia_mtime = CFS_CURRENT_TIME;
1327 attr->ia_valid |= ATTR_MTIME_SET;
1330 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1331 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1332 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1333 cfs_time_current_sec());
1335 /* NB: ATTR_SIZE will only be set after this point if the size
1336 * resides on the MDS, ie, this file has no objects. */
1338 attr->ia_valid &= ~ATTR_SIZE;
1340 /* We always do an MDS RPC, even if we're only changing the size;
1341 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1343 OBD_ALLOC_PTR(op_data);
1344 if (op_data == NULL)
1347 UNLOCK_INODE_MUTEX(inode);
1348 if (ia_valid & ATTR_SIZE)
1349 UP_WRITE_I_ALLOC_SEM(inode);
1350 cfs_down_write(&lli->lli_trunc_sem);
1351 LOCK_INODE_MUTEX(inode);
1352 if (ia_valid & ATTR_SIZE)
1353 DOWN_WRITE_I_ALLOC_SEM(inode);
1355 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1357 /* Open epoch for truncate. */
1358 if (exp_connect_som(ll_i2mdexp(inode)) &&
1359 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1360 op_data->op_flags = MF_EPOCH_OPEN;
1362 rc = ll_md_setattr(inode, op_data, &mod);
1366 ll_ioepoch_open(lli, op_data->op_ioepoch);
1367 if (!lsm || !S_ISREG(inode->i_mode)) {
1368 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1372 if (ia_valid & ATTR_SIZE)
1373 attr->ia_valid |= ATTR_SIZE;
1374 if (ia_valid & (ATTR_SIZE |
1375 ATTR_ATIME | ATTR_ATIME_SET |
1376 ATTR_MTIME | ATTR_MTIME_SET))
1377 /* on truncate and utimes send attributes to osts, setting
1378 * mtime/atime to past will be performed under PW 0:EOF extent
1379 * lock (new_size:EOF for truncate)
1380 * it may seem excessive to send mtime/atime updates to osts
1381 * when not setting times to past, but it is necessary due to
1382 * possible time de-synchronization */
1383 rc = ll_setattr_ost(inode, attr);
1387 if (op_data->op_ioepoch)
1388 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1389 ll_finish_md_op_data(op_data);
1391 cfs_up_write(&lli->lli_trunc_sem);
1392 return rc ? rc : rc1;
1395 int ll_setattr(struct dentry *de, struct iattr *attr)
1397 int mode = de->d_inode->i_mode;
1399 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1400 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1401 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1403 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1404 (ATTR_SIZE|ATTR_MODE)) &&
1405 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1406 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1407 !(attr->ia_mode & S_ISGID))))
1408 attr->ia_valid |= ATTR_FORCE;
1410 if ((mode & S_ISUID) &&
1411 !(attr->ia_mode & S_ISUID) &&
1412 !(attr->ia_valid & ATTR_KILL_SUID))
1413 attr->ia_valid |= ATTR_KILL_SUID;
1415 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1416 !(attr->ia_mode & S_ISGID) &&
1417 !(attr->ia_valid & ATTR_KILL_SGID))
1418 attr->ia_valid |= ATTR_KILL_SGID;
1420 return ll_setattr_raw(de->d_inode, attr);
1423 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1424 __u64 max_age, __u32 flags)
1426 struct ll_sb_info *sbi = ll_s2sbi(sb);
1427 struct obd_statfs obd_osfs;
1431 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1433 CERROR("md_statfs fails: rc = %d\n", rc);
1437 osfs->os_type = sb->s_magic;
1439 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1440 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1442 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1443 flags |= OBD_STATFS_NODELAY;
1445 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1446 &obd_osfs, max_age, flags);
1448 CERROR("obd_statfs fails: rc = %d\n", rc);
1452 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1453 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1456 osfs->os_bsize = obd_osfs.os_bsize;
1457 osfs->os_blocks = obd_osfs.os_blocks;
1458 osfs->os_bfree = obd_osfs.os_bfree;
1459 osfs->os_bavail = obd_osfs.os_bavail;
1461 /* If we don't have as many objects free on the OST as inodes
1462 * on the MDS, we reduce the total number of inodes to
1463 * compensate, so that the "inodes in use" number is correct.
1465 if (obd_osfs.os_ffree < osfs->os_ffree) {
1466 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1468 osfs->os_ffree = obd_osfs.os_ffree;
1473 #ifndef HAVE_STATFS_DENTRY_PARAM
1474 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1477 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1479 struct super_block *sb = de->d_sb;
1481 struct obd_statfs osfs;
1484 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1485 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1487 /* Some amount of caching on the client is allowed */
1488 rc = ll_statfs_internal(sb, &osfs,
1489 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1494 statfs_unpack(sfs, &osfs);
1496 /* We need to downshift for all 32-bit kernels, because we can't
1497 * tell if the kernel is being called via sys_statfs64() or not.
1498 * Stop before overflowing f_bsize - in which case it is better
1499 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1500 if (sizeof(long) < 8) {
1501 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1504 osfs.os_blocks >>= 1;
1505 osfs.os_bfree >>= 1;
1506 osfs.os_bavail >>= 1;
1510 sfs->f_blocks = osfs.os_blocks;
1511 sfs->f_bfree = osfs.os_bfree;
1512 sfs->f_bavail = osfs.os_bavail;
1517 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1519 struct ll_inode_info *lli;
1520 struct lov_stripe_md *lsm;
1522 lli = ll_i2info(inode);
1523 LASSERT(lli->lli_size_sem_owner != current);
1524 cfs_down(&lli->lli_size_sem);
1525 LASSERT(lli->lli_size_sem_owner == NULL);
1526 lli->lli_size_sem_owner = current;
1528 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1531 lov_stripe_lock(lsm);
1534 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1536 struct ll_inode_info *lli;
1537 struct lov_stripe_md *lsm;
1539 lli = ll_i2info(inode);
1541 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1544 lov_stripe_unlock(lsm);
1545 LASSERT(lli->lli_size_sem_owner == current);
1546 lli->lli_size_sem_owner = NULL;
1547 cfs_up(&lli->lli_size_sem);
1550 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1552 struct ll_inode_info *lli = ll_i2info(inode);
1553 struct mdt_body *body = md->body;
1554 struct lov_stripe_md *lsm = md->lsm;
1555 struct ll_sb_info *sbi = ll_i2sbi(inode);
1557 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1559 cfs_down(&lli->lli_och_sem);
1560 if (lli->lli_smd == NULL) {
1561 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1562 lsm->lsm_magic != LOV_MAGIC_V3) {
1563 dump_lsm(D_ERROR, lsm);
1566 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1567 lsm, inode->i_ino, inode->i_generation, inode);
1568 /* cl_inode_init must go before lli_smd or a race is
1569 * possible where client thinks the file has stripes,
1570 * but lov raid0 is not setup yet and parallel e.g.
1571 * glimpse would try to use uninitialized lov */
1572 cl_inode_init(inode, md);
1573 cfs_spin_lock(&lli->lli_lock);
1575 cfs_spin_unlock(&lli->lli_lock);
1576 cfs_up(&lli->lli_och_sem);
1577 lli->lli_maxbytes = lsm->lsm_maxbytes;
1578 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1579 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1581 cfs_up(&lli->lli_och_sem);
1582 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1583 lli->lli_smd->lsm_stripe_count ==
1584 lsm->lsm_stripe_count);
1585 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1586 CERROR("lsm mismatch for inode %ld\n",
1588 CERROR("lli_smd:\n");
1589 dump_lsm(D_ERROR, lli->lli_smd);
1591 dump_lsm(D_ERROR, lsm);
1595 if (lli->lli_smd != lsm)
1596 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1599 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1600 if (body->valid & OBD_MD_FLRMTPERM)
1601 ll_update_remote_perm(inode, md->remote_perm);
1603 #ifdef CONFIG_FS_POSIX_ACL
1604 else if (body->valid & OBD_MD_FLACL) {
1605 cfs_spin_lock(&lli->lli_lock);
1606 if (lli->lli_posix_acl)
1607 posix_acl_release(lli->lli_posix_acl);
1608 lli->lli_posix_acl = md->posix_acl;
1609 cfs_spin_unlock(&lli->lli_lock);
1612 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1613 inode->i_generation = cl_fid_build_gen(&body->fid1);
1615 if (body->valid & OBD_MD_FLATIME) {
1616 if (body->atime > LTIME_S(inode->i_atime))
1617 LTIME_S(inode->i_atime) = body->atime;
1618 lli->lli_lvb.lvb_atime = body->atime;
1620 if (body->valid & OBD_MD_FLMTIME) {
1621 if (body->mtime > LTIME_S(inode->i_mtime)) {
1622 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1623 "to "LPU64"\n", inode->i_ino,
1624 LTIME_S(inode->i_mtime), body->mtime);
1625 LTIME_S(inode->i_mtime) = body->mtime;
1627 lli->lli_lvb.lvb_mtime = body->mtime;
1629 if (body->valid & OBD_MD_FLCTIME) {
1630 if (body->ctime > LTIME_S(inode->i_ctime))
1631 LTIME_S(inode->i_ctime) = body->ctime;
1632 lli->lli_lvb.lvb_ctime = body->ctime;
1634 if (body->valid & OBD_MD_FLMODE)
1635 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1636 if (body->valid & OBD_MD_FLTYPE)
1637 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1638 LASSERT(inode->i_mode != 0);
1639 if (S_ISREG(inode->i_mode)) {
1640 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1642 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1644 #ifdef HAVE_INODE_BLKSIZE
1645 inode->i_blksize = 1<<inode->i_blkbits;
1647 if (body->valid & OBD_MD_FLUID)
1648 inode->i_uid = body->uid;
1649 if (body->valid & OBD_MD_FLGID)
1650 inode->i_gid = body->gid;
1651 if (body->valid & OBD_MD_FLFLAGS)
1652 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1653 if (body->valid & OBD_MD_FLNLINK)
1654 inode->i_nlink = body->nlink;
1655 if (body->valid & OBD_MD_FLRDEV)
1656 inode->i_rdev = old_decode_dev(body->rdev);
1658 if (body->valid & OBD_MD_FLID) {
1659 /* FID shouldn't be changed! */
1660 if (fid_is_sane(&lli->lli_fid)) {
1661 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1662 "Trying to change FID "DFID
1663 " to the "DFID", inode %lu/%u(%p)\n",
1664 PFID(&lli->lli_fid), PFID(&body->fid1),
1665 inode->i_ino, inode->i_generation, inode);
1667 lli->lli_fid = body->fid1;
1670 LASSERT(fid_seq(&lli->lli_fid) != 0);
1672 if (body->valid & OBD_MD_FLSIZE) {
1673 if (exp_connect_som(ll_i2mdexp(inode)) &&
1674 S_ISREG(inode->i_mode) && lli->lli_smd) {
1675 struct lustre_handle lockh;
1678 /* As it is possible a blocking ast has been processed
1679 * by this time, we need to check there is an UPDATE
1680 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1682 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1685 if (lli->lli_flags & (LLIF_DONE_WRITING |
1686 LLIF_EPOCH_PENDING |
1688 CERROR("ino %lu flags %lu still has "
1689 "size authority! do not trust "
1690 "the size got from MDS\n",
1691 inode->i_ino, lli->lli_flags);
1693 /* Use old size assignment to avoid
1694 * deadlock bz14138 & bz14326 */
1695 inode->i_size = body->size;
1696 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1698 ldlm_lock_decref(&lockh, mode);
1701 /* Use old size assignment to avoid
1702 * deadlock bz14138 & bz14326 */
1703 inode->i_size = body->size;
1705 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1706 inode->i_ino, (unsigned long long)body->size);
1709 if (body->valid & OBD_MD_FLBLOCKS)
1710 inode->i_blocks = body->blocks;
1713 if (body->valid & OBD_MD_FLMDSCAPA) {
1714 LASSERT(md->mds_capa);
1715 ll_add_capa(inode, md->mds_capa);
1717 if (body->valid & OBD_MD_FLOSSCAPA) {
1718 LASSERT(md->oss_capa);
1719 ll_add_capa(inode, md->oss_capa);
1723 void ll_read_inode2(struct inode *inode, void *opaque)
1725 struct lustre_md *md = opaque;
1726 struct ll_inode_info *lli = ll_i2info(inode);
1729 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1730 PFID(&lli->lli_fid), inode);
1734 LASSERT(!lli->lli_smd);
1736 /* Core attributes from the MDS first. This is a new inode, and
1737 * the VFS doesn't zero times in the core inode so we have to do
1738 * it ourselves. They will be overwritten by either MDS or OST
1739 * attributes - we just need to make sure they aren't newer. */
1740 LTIME_S(inode->i_mtime) = 0;
1741 LTIME_S(inode->i_atime) = 0;
1742 LTIME_S(inode->i_ctime) = 0;
1744 ll_update_inode(inode, md);
1746 /* OIDEBUG(inode); */
1748 /* initializing backing dev info. */
1749 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1752 if (S_ISREG(inode->i_mode)) {
1753 struct ll_sb_info *sbi = ll_i2sbi(inode);
1754 inode->i_op = &ll_file_inode_operations;
1755 inode->i_fop = sbi->ll_fop;
1756 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1758 } else if (S_ISDIR(inode->i_mode)) {
1759 inode->i_op = &ll_dir_inode_operations;
1760 inode->i_fop = &ll_dir_operations;
1761 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1763 } else if (S_ISLNK(inode->i_mode)) {
1764 inode->i_op = &ll_fast_symlink_inode_operations;
1767 inode->i_op = &ll_special_inode_operations;
1769 init_special_inode(inode, inode->i_mode,
1770 kdev_t_to_nr(inode->i_rdev));
1776 void ll_delete_inode(struct inode *inode)
1778 struct ll_sb_info *sbi = ll_i2sbi(inode);
1782 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1784 CERROR("fid_delete() failed, rc %d\n", rc);
1786 truncate_inode_pages(&inode->i_data, 0);
1788 /* Workaround for LU-118 */
1789 if (inode->i_data.nrpages) {
1790 TREE_READ_LOCK_IRQ(&inode->i_data);
1791 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1792 LASSERTF(inode->i_data.nrpages == 0,
1793 "inode=%lu/%u(%p) nrpages=%lu, see "
1794 "http://jira.whamcloud.com/browse/LU-118\n",
1795 inode->i_ino, inode->i_generation, inode,
1796 inode->i_data.nrpages);
1798 /* Workaround end */
1800 #ifdef HAVE_SBOPS_EVICT_INODE
1801 ll_clear_inode(inode);
1802 end_writeback(inode);
1810 int ll_iocontrol(struct inode *inode, struct file *file,
1811 unsigned int cmd, unsigned long arg)
1813 struct ll_sb_info *sbi = ll_i2sbi(inode);
1814 struct ptlrpc_request *req = NULL;
1819 case FSFILT_IOC_GETFLAGS: {
1820 struct mdt_body *body;
1821 struct md_op_data *op_data;
1823 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1824 0, 0, LUSTRE_OPC_ANY,
1826 if (IS_ERR(op_data))
1827 RETURN(PTR_ERR(op_data));
1829 op_data->op_valid = OBD_MD_FLFLAGS;
1830 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1831 ll_finish_md_op_data(op_data);
1833 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1837 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1839 flags = body->flags;
1841 ptlrpc_req_finished(req);
1843 RETURN(put_user(flags, (int *)arg));
1845 case FSFILT_IOC_SETFLAGS: {
1846 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1847 struct obd_info oinfo = { { { 0 } } };
1848 struct md_op_data *op_data;
1850 if (get_user(flags, (int *)arg))
1854 OBDO_ALLOC(oinfo.oi_oa);
1858 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1859 LUSTRE_OPC_ANY, NULL);
1860 if (IS_ERR(op_data))
1861 RETURN(PTR_ERR(op_data));
1863 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1864 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1865 rc = md_setattr(sbi->ll_md_exp, op_data,
1866 NULL, 0, NULL, 0, &req, NULL);
1867 ll_finish_md_op_data(op_data);
1868 ptlrpc_req_finished(req);
1870 OBDO_FREE(oinfo.oi_oa);
1875 OBDO_FREE(oinfo.oi_oa);
1876 GOTO(update_cache, rc);
1879 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1880 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1881 oinfo.oi_oa->o_flags = flags;
1882 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1884 oinfo.oi_capa = ll_mdscapa_get(inode);
1885 obdo_from_inode(oinfo.oi_oa, inode,
1886 &ll_i2info(inode)->lli_fid, 0);
1887 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1888 capa_put(oinfo.oi_capa);
1889 OBDO_FREE(oinfo.oi_oa);
1891 if (rc != -EPERM && rc != -EACCES)
1892 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1898 inode->i_flags = ll_ext_to_inode_flags(flags);
1908 int ll_flush_ctx(struct inode *inode)
1910 struct ll_sb_info *sbi = ll_i2sbi(inode);
1912 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1914 obd_set_info_async(sbi->ll_md_exp,
1915 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1917 obd_set_info_async(sbi->ll_dt_exp,
1918 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1923 /* umount -f client means force down, don't save state */
1924 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1925 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1927 struct super_block *sb = vfsmnt->mnt_sb;
1929 void ll_umount_begin(struct super_block *sb)
1932 struct lustre_sb_info *lsi = s2lsi(sb);
1933 struct ll_sb_info *sbi = ll_s2sbi(sb);
1934 struct obd_device *obd;
1935 struct obd_ioctl_data *ioc_data;
1938 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1939 if (!(flags & MNT_FORCE)) {
1945 /* Tell the MGC we got umount -f */
1946 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1948 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1949 sb->s_count, atomic_read(&sb->s_active));
1951 obd = class_exp2obd(sbi->ll_md_exp);
1953 CERROR("Invalid MDC connection handle "LPX64"\n",
1954 sbi->ll_md_exp->exp_handle.h_cookie);
1960 obd = class_exp2obd(sbi->ll_dt_exp);
1962 CERROR("Invalid LOV connection handle "LPX64"\n",
1963 sbi->ll_dt_exp->exp_handle.h_cookie);
1969 OBD_ALLOC_PTR(ioc_data);
1971 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1972 sizeof ioc_data, ioc_data, NULL);
1974 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1975 sizeof ioc_data, ioc_data, NULL);
1977 OBD_FREE_PTR(ioc_data);
1981 /* Really, we'd like to wait until there are no requests outstanding,
1982 * and then continue. For now, we just invalidate the requests,
1983 * schedule() and sleep one second if needed, and hope.
1986 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1987 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1988 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1989 cfs_time_seconds(1));
1990 if (atomic_read(&vfsmnt->mnt_count) > 2)
1991 LCONSOLE_WARN("Mount still busy with %d refs! You "
1992 "may try to umount it a bit later\n",
1993 atomic_read(&vfsmnt->mnt_count));
2000 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2002 struct ll_sb_info *sbi = ll_s2sbi(sb);
2006 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2007 read_only = *flags & MS_RDONLY;
2008 err = obd_set_info_async(sbi->ll_md_exp,
2009 sizeof(KEY_READ_ONLY),
2010 KEY_READ_ONLY, sizeof(read_only),
2013 CERROR("Failed to change the read-only flag during "
2014 "remount: %d\n", err);
2019 sb->s_flags |= MS_RDONLY;
2021 sb->s_flags &= ~MS_RDONLY;
2026 int ll_prep_inode(struct inode **inode,
2027 struct ptlrpc_request *req,
2028 struct super_block *sb)
2030 struct ll_sb_info *sbi = NULL;
2031 struct lustre_md md;
2035 LASSERT(*inode || sb);
2036 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2037 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2038 sbi->ll_md_exp, &md);
2043 ll_update_inode(*inode, &md);
2045 LASSERT(sb != NULL);
2048 * At this point server returns to client's same fid as client
2049 * generated for creating. So using ->fid1 is okay here.
2051 LASSERT(fid_is_sane(&md.body->fid1));
2053 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2054 if (*inode == NULL || IS_ERR(*inode)) {
2056 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2057 #ifdef CONFIG_FS_POSIX_ACL
2059 posix_acl_release(md.posix_acl);
2060 md.posix_acl = NULL;
2063 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2065 CERROR("new_inode -fatal: rc %d\n", rc);
2071 md_free_lustre_md(sbi->ll_md_exp, &md);
2075 int ll_obd_statfs(struct inode *inode, void *arg)
2077 struct ll_sb_info *sbi = NULL;
2078 struct obd_export *exp;
2080 struct obd_ioctl_data *data = NULL;
2084 if (!inode || !(sbi = ll_i2sbi(inode)))
2085 GOTO(out_statfs, rc = -EINVAL);
2087 rc = obd_ioctl_getdata(&buf, &len, arg);
2089 GOTO(out_statfs, rc);
2092 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2093 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2094 GOTO(out_statfs, rc = -EINVAL);
2096 if (data->ioc_inllen1 != sizeof(__u32) ||
2097 data->ioc_inllen2 != sizeof(__u32) ||
2098 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2099 data->ioc_plen2 != sizeof(struct obd_uuid))
2100 GOTO(out_statfs, rc = -EINVAL);
2102 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2103 if (type == LL_STATFS_MDC)
2104 exp = sbi->ll_md_exp;
2105 else if (type == LL_STATFS_LOV)
2106 exp = sbi->ll_dt_exp;
2108 GOTO(out_statfs, rc = -ENODEV);
2110 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2112 GOTO(out_statfs, rc);
2115 obd_ioctl_freedata(buf, len);
2119 int ll_process_config(struct lustre_cfg *lcfg)
2123 struct lprocfs_static_vars lvars;
2127 lprocfs_llite_init_vars(&lvars);
2129 /* The instance name contains the sb: lustre-client-aacfe000 */
2130 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2131 if (!ptr || !*(++ptr))
2133 if (sscanf(ptr, "%lx", &x) != 1)
2136 /* This better be a real Lustre superblock! */
2137 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2139 /* Note we have not called client_common_fill_super yet, so
2140 proc fns must be able to handle that! */
2141 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2148 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2149 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2150 struct inode *i1, struct inode *i2,
2151 const char *name, int namelen,
2152 int mode, __u32 opc, void *data)
2154 LASSERT(i1 != NULL);
2156 if (namelen > ll_i2sbi(i1)->ll_namelen)
2157 return ERR_PTR(-ENAMETOOLONG);
2159 if (op_data == NULL)
2160 OBD_ALLOC_PTR(op_data);
2162 if (op_data == NULL)
2163 return ERR_PTR(-ENOMEM);
2165 ll_i2gids(op_data->op_suppgids, i1, i2);
2166 op_data->op_fid1 = *ll_inode2fid(i1);
2167 op_data->op_capa1 = ll_mdscapa_get(i1);
2170 op_data->op_fid2 = *ll_inode2fid(i2);
2171 op_data->op_capa2 = ll_mdscapa_get(i2);
2173 fid_zero(&op_data->op_fid2);
2174 op_data->op_capa2 = NULL;
2177 op_data->op_name = name;
2178 op_data->op_namelen = namelen;
2179 op_data->op_mode = mode;
2180 op_data->op_mod_time = cfs_time_current_sec();
2181 op_data->op_fsuid = cfs_curproc_fsuid();
2182 op_data->op_fsgid = cfs_curproc_fsgid();
2183 op_data->op_cap = cfs_curproc_cap_pack();
2184 op_data->op_bias = MDS_CHECK_SPLIT;
2185 op_data->op_opc = opc;
2186 op_data->op_mds = 0;
2187 op_data->op_data = data;
2189 /* If the file is being opened after mknod() (normally due to NFS)
2190 * try to use the default stripe data from parent directory for
2191 * allocating OST objects. Try to pass the parent FID to MDS. */
2192 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2193 ll_i2info(i2)->lli_smd == NULL) {
2194 struct ll_inode_info *lli = ll_i2info(i2);
2196 cfs_spin_lock(&lli->lli_lock);
2197 if (likely(lli->lli_smd == NULL &&
2198 !fid_is_zero(&lli->lli_pfid)))
2199 op_data->op_fid1 = lli->lli_pfid;
2200 cfs_spin_unlock(&lli->lli_lock);
2201 /** We ignore parent's capability temporary. */
2207 void ll_finish_md_op_data(struct md_op_data *op_data)
2209 capa_put(op_data->op_capa1);
2210 capa_put(op_data->op_capa2);
2211 OBD_FREE_PTR(op_data);
2214 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2216 struct ll_sb_info *sbi;
2218 LASSERT((seq != NULL) && (vfs != NULL));
2219 sbi = ll_s2sbi(vfs->mnt_sb);
2221 if (sbi->ll_flags & LL_SBI_NOLCK)
2222 seq_puts(seq, ",nolock");
2224 if (sbi->ll_flags & LL_SBI_FLOCK)
2225 seq_puts(seq, ",flock");
2227 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2228 seq_puts(seq, ",localflock");
2230 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2231 seq_puts(seq, ",user_xattr");
2233 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2234 seq_puts(seq, ",lazystatfs");