4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 cfs_mem_cache_t *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 DEFINE_SPINLOCK(ll_sb_lock);
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) cfs_ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
83 OBD_ALLOC(sbi, sizeof(*sbi));
87 cfs_spin_lock_init(&sbi->ll_lock);
88 cfs_mutex_init(&sbi->ll_lco.lco_lock);
89 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
90 cfs_spin_lock_init(&sbi->ll_process_lock);
91 sbi->ll_rw_stats_on = 0;
94 pages = si.totalram - si.totalhigh;
95 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
96 #ifdef HAVE_BGL_SUPPORT
97 sbi->ll_async_page_max = pages / 4;
99 sbi->ll_async_page_max = pages / 2;
102 sbi->ll_async_page_max = (pages / 4) * 3;
105 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
106 SBI_DEFAULT_READAHEAD_MAX);
107 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
108 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
109 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
110 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
111 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
113 ll_generate_random_uuid(uuid);
114 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
115 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
117 cfs_spin_lock(&ll_sb_lock);
118 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
119 cfs_spin_unlock(&ll_sb_lock);
121 sbi->ll_flags |= LL_SBI_VERBOSE;
122 #ifdef ENABLE_CHECKSUM
123 sbi->ll_flags |= LL_SBI_CHECKSUM;
126 #ifdef HAVE_LRU_RESIZE_SUPPORT
127 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
130 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
131 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
133 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
137 /* metadata statahead is enabled by default */
138 sbi->ll_sa_max = LL_SA_RPC_DEF;
139 cfs_atomic_set(&sbi->ll_sa_total, 0);
140 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
141 cfs_atomic_set(&sbi->ll_agl_total, 0);
142 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
147 void ll_free_sbi(struct super_block *sb)
149 struct ll_sb_info *sbi = ll_s2sbi(sb);
153 cfs_spin_lock(&ll_sb_lock);
154 cfs_list_del(&sbi->ll_list);
155 cfs_spin_unlock(&ll_sb_lock);
156 OBD_FREE(sbi, sizeof(*sbi));
161 static struct dentry_operations ll_d_root_ops = {
162 .d_compare = ll_dcompare,
163 .d_revalidate = ll_revalidate_nd,
166 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
167 struct vfsmount *mnt)
169 struct inode *root = 0;
170 struct ll_sb_info *sbi = ll_s2sbi(sb);
171 struct obd_device *obd;
172 struct obd_capa *oc = NULL;
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 if (proc_lustre_fs_root) {
200 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
203 CERROR("could not register mount in /proc/fs/lustre\n");
206 /* indicate the features supported by this client */
207 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
208 OBD_CONNECT_ATTRFID |
209 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
210 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
211 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
212 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
213 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
214 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
215 OBD_CONNECT_EINPROGRESS |
216 OBD_CONNECT_JOBSTATS;
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;
229 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
230 /* flag mdc connection as lightweight, only used for test
231 * purpose, use with care */
232 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
234 data->ocd_ibits_known = MDS_INODELOCK_FULL;
235 data->ocd_version = LUSTRE_VERSION_CODE;
237 if (sb->s_flags & MS_RDONLY)
238 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
239 if (sbi->ll_flags & LL_SBI_USER_XATTR)
240 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
242 #ifdef HAVE_MS_FLOCK_LOCK
243 /* force vfs to use lustre handler for flock() calls - bug 10743 */
244 sb->s_flags |= MS_FLOCK_LOCK;
246 #ifdef MS_HAS_NEW_AOPS
247 sb->s_flags |= MS_HAS_NEW_AOPS;
250 if (sbi->ll_flags & LL_SBI_FLOCK)
251 sbi->ll_fop = &ll_file_operations_flock;
252 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
253 sbi->ll_fop = &ll_file_operations;
255 sbi->ll_fop = &ll_file_operations_noflock;
258 data->ocd_connect_flags |= OBD_CONNECT_REAL;
259 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
260 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
262 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
264 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
266 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
267 "recovery, of which this client is not a "
268 "part. Please wait for recovery to complete,"
269 " abort, or time out.\n", md);
272 CERROR("cannot connect to %s: rc = %d\n", md, err);
276 err = obd_fid_init(sbi->ll_md_exp);
278 CERROR("Can't init metadata layer FID infrastructure, "
283 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
284 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
286 GOTO(out_md_fid, err);
288 /* This needs to be after statfs to ensure connect has finished.
289 * Note that "data" does NOT contain the valid connect reply.
290 * If connecting to a 1.8 server there will be no LMV device, so
291 * we can access the MDC export directly and exp_connect_flags will
292 * be non-zero, but if accessing an upgraded 2.1 server it will
293 * have the correct flags filled in.
294 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
295 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
296 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
297 valid != CLIENT_CONNECT_MDT_REQD) {
300 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
301 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
302 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
303 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
304 "feature(s) needed for correct operation "
305 "of this client (%s). Please upgrade "
306 "server or downgrade client.\n",
307 sbi->ll_md_exp->exp_obd->obd_name, buf);
308 OBD_FREE(buf, CFS_PAGE_SIZE);
309 GOTO(out_md, err = -EPROTO);
312 size = sizeof(*data);
313 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
314 KEY_CONN_DATA, &size, data, NULL);
316 CERROR("Get connect data failed: %d \n", err);
320 LASSERT(osfs->os_bsize);
321 sb->s_blocksize = osfs->os_bsize;
322 sb->s_blocksize_bits = log2(osfs->os_bsize);
323 sb->s_magic = LL_SUPER_MAGIC;
324 sb->s_maxbytes = MAX_LFS_FILESIZE;
325 sbi->ll_namelen = osfs->os_namelen;
326 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
328 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
329 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
330 LCONSOLE_INFO("Disabling user_xattr feature because "
331 "it is not supported on the server\n");
332 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
335 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
337 sb->s_flags |= MS_POSIXACL;
339 sbi->ll_flags |= LL_SBI_ACL;
341 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
343 sb->s_flags &= ~MS_POSIXACL;
345 sbi->ll_flags &= ~LL_SBI_ACL;
348 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
349 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
350 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
351 LCONSOLE_INFO("client is set as remote by default.\n");
354 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
355 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
356 LCONSOLE_INFO("client claims to be remote, but server "
357 "rejected, forced to be local.\n");
361 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
362 LCONSOLE_INFO("client enabled MDS capability!\n");
363 sbi->ll_flags |= LL_SBI_MDS_CAPA;
366 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
367 LCONSOLE_INFO("client enabled OSS capability!\n");
368 sbi->ll_flags |= LL_SBI_OSS_CAPA;
371 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
372 sbi->ll_flags |= LL_SBI_64BIT_HASH;
374 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
375 sbi->ll_md_brw_size = data->ocd_brw_size;
377 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
379 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
380 LCONSOLE_INFO("Layout lock feature supported.\n");
381 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
384 obd = class_name2obd(dt);
386 CERROR("DT %s: not setup or attached\n", dt);
387 GOTO(out_md_fid, err = -ENODEV);
390 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
391 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
392 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
393 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
394 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
395 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
396 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
397 OBD_CONNECT_MAXBYTES |
398 OBD_CONNECT_EINPROGRESS |
399 OBD_CONNECT_JOBSTATS;
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_client();
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_lock(&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_unlock(&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 /*b=17630*/
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(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
542 KEY_CHECKSUM, sizeof(checksum), &checksum,
546 sb->s_root = d_alloc_root(root);
547 #ifdef HAVE_DCACHE_LOCK
548 sb->s_root->d_op = &ll_d_root_ops;
550 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
551 d_set_d_op(sb->s_root, &ll_d_root_ops);
552 sb->s_d_op = &ll_d_ops;
555 sbi->ll_sdev_orig = sb->s_dev;
557 /* We set sb->s_dev equal on all lustre clients in order to support
558 * NFS export clustering. NFSD requires that the FSID be the same
560 /* s_dev is also used in lt_compare() to compare two fs, but that is
561 * only a node-local comparison. */
562 uuid = obd_get_uuid(sbi->ll_md_exp);
564 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
576 obd_fid_fini(sbi->ll_dt_exp);
578 obd_disconnect(sbi->ll_dt_exp);
579 sbi->ll_dt_exp = NULL;
581 obd_fid_fini(sbi->ll_md_exp);
583 obd_disconnect(sbi->ll_md_exp);
584 sbi->ll_md_exp = NULL;
590 lprocfs_unregister_mountpoint(sbi);
594 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
598 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
600 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
601 KEY_MAX_EASIZE, &size, lmmsize, NULL);
603 CERROR("Get max mdsize error rc %d \n", rc);
608 void ll_dump_inode(struct inode *inode)
610 struct list_head *tmp;
611 int dentry_count = 0;
613 LASSERT(inode != NULL);
615 list_for_each(tmp, &inode->i_dentry)
618 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
619 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
620 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
623 void lustre_dump_dentry(struct dentry *dentry, int recur)
625 struct list_head *tmp;
628 LASSERT(dentry != NULL);
630 list_for_each(tmp, &dentry->d_subdirs)
633 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
634 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
635 dentry->d_name.len, dentry->d_name.name,
636 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
637 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
638 dentry->d_flags, dentry->d_fsdata, subdirs);
639 if (dentry->d_inode != NULL)
640 ll_dump_inode(dentry->d_inode);
645 list_for_each(tmp, &dentry->d_subdirs) {
646 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
647 lustre_dump_dentry(d, recur - 1);
651 void client_common_put_super(struct super_block *sb)
653 struct ll_sb_info *sbi = ll_s2sbi(sb);
656 #ifdef CONFIG_FS_POSIX_ACL
657 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
658 et_fini(&sbi->ll_et);
659 rct_fini(&sbi->ll_rct);
663 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
665 ll_close_thread_shutdown(sbi->ll_lcq);
669 cfs_list_del(&sbi->ll_conn_chain);
671 obd_fid_fini(sbi->ll_dt_exp);
672 obd_disconnect(sbi->ll_dt_exp);
673 sbi->ll_dt_exp = NULL;
675 lprocfs_unregister_mountpoint(sbi);
677 obd_fid_fini(sbi->ll_md_exp);
678 obd_disconnect(sbi->ll_md_exp);
679 sbi->ll_md_exp = NULL;
684 void ll_kill_super(struct super_block *sb)
686 struct ll_sb_info *sbi;
691 if (!(sb->s_flags & MS_ACTIVE))
695 /* we need restore s_dev from changed for clustred NFS before put_super
696 * because new kernels have cached s_dev and change sb->s_dev in
697 * put_super not affected real removing devices */
699 sb->s_dev = sbi->ll_sdev_orig;
703 char *ll_read_opt(const char *opt, char *data)
709 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
710 if (strncmp(opt, data, strlen(opt)))
712 if ((value = strchr(data, '=')) == NULL)
716 OBD_ALLOC(retval, strlen(value) + 1);
718 CERROR("out of memory!\n");
722 memcpy(retval, value, strlen(value)+1);
723 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
727 static inline int ll_set_opt(const char *opt, char *data, int fl)
729 if (strncmp(opt, data, strlen(opt)) != 0)
735 /* non-client-specific mount options are parsed in lmd_parse */
736 static int ll_options(char *options, int *flags)
739 char *s1 = options, *s2;
745 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
748 CDEBUG(D_SUPER, "next opt=%s\n", s1);
749 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
754 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
759 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
764 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
769 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
774 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
779 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
780 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
782 /* Ignore deprecated mount option. The client will
783 * always try to mount with ACL support, whether this
784 * is used depends on whether server supports it. */
785 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
786 "mount option 'acl'.\n");
789 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
791 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
792 "mount option 'noacl'.\n");
796 #warning "{no}acl options have been deprecated since 1.8, please remove them"
798 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
803 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
808 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
814 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
819 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
824 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
829 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
834 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
839 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
844 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
849 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
854 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
859 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
864 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
870 s2 = strchr(s1, ',');
878 void ll_lli_init(struct ll_inode_info *lli)
880 lli->lli_inode_magic = LLI_INODE_MAGIC;
882 lli->lli_ioepoch = 0;
883 lli->lli_maxbytes = MAX_LFS_FILESIZE;
884 cfs_spin_lock_init(&lli->lli_lock);
885 lli->lli_posix_acl = NULL;
886 lli->lli_remote_perms = NULL;
887 cfs_mutex_init(&lli->lli_rmtperm_mutex);
888 /* Do not set lli_fid, it has been initialized already. */
889 fid_zero(&lli->lli_pfid);
890 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
891 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
892 cfs_atomic_set(&lli->lli_open_count, 0);
893 lli->lli_mds_capa = NULL;
894 lli->lli_rmtperm_time = 0;
895 lli->lli_pending_och = NULL;
896 lli->lli_mds_read_och = NULL;
897 lli->lli_mds_write_och = NULL;
898 lli->lli_mds_exec_och = NULL;
899 lli->lli_open_fd_read_count = 0;
900 lli->lli_open_fd_write_count = 0;
901 lli->lli_open_fd_exec_count = 0;
902 cfs_mutex_init(&lli->lli_och_mutex);
903 cfs_spin_lock_init(&lli->lli_agl_lock);
904 lli->lli_has_smd = false;
905 lli->lli_clob = NULL;
907 LASSERT(lli->lli_vfs_inode.i_mode != 0);
908 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
909 cfs_mutex_init(&lli->lli_readdir_mutex);
910 lli->lli_opendir_key = NULL;
912 lli->lli_def_acl = NULL;
913 cfs_spin_lock_init(&lli->lli_sa_lock);
914 lli->lli_opendir_pid = 0;
916 cfs_sema_init(&lli->lli_size_sem, 1);
917 lli->lli_size_sem_owner = NULL;
918 lli->lli_symlink_name = NULL;
919 cfs_init_rwsem(&lli->lli_trunc_sem);
920 cfs_mutex_init(&lli->lli_write_mutex);
921 cfs_init_rwsem(&lli->lli_glimpse_sem);
922 lli->lli_glimpse_time = 0;
923 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
924 lli->lli_agl_index = 0;
925 lli->lli_async_rc = 0;
927 cfs_mutex_init(&lli->lli_layout_mutex);
930 static inline int ll_bdi_register(struct backing_dev_info *bdi)
932 #ifdef HAVE_BDI_REGISTER
933 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
936 bdi->name = "lustre";
938 return bdi_register(bdi, NULL, "lustre-%d",
939 atomic_inc_return(&ll_bdi_num));
945 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
947 struct lustre_profile *lprof = NULL;
948 struct lustre_sb_info *lsi = s2lsi(sb);
949 struct ll_sb_info *sbi;
950 char *dt = NULL, *md = NULL;
951 char *profilenm = get_profile_name(sb);
952 struct config_llog_instance *cfg;
953 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
954 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
958 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
966 /* client additional sb info */
967 lsi->lsi_llsbi = sbi = ll_init_sbi();
969 cfs_module_put(THIS_MODULE);
974 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
978 err = ll_bdi_init(&lsi->lsi_bdi);
981 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
982 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
983 err = ll_bdi_register(&lsi->lsi_bdi);
988 sb->s_bdi = &lsi->lsi_bdi;
991 /* Generate a string unique to this super, in case some joker tries
992 to mount the same fs at two mount points.
993 Use the address of the super itself.*/
994 cfg->cfg_instance = sb;
995 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
996 cfg->cfg_callback = class_config_llog_handler;
997 /* set up client obds */
998 err = lustre_process_log(sb, profilenm, cfg);
1000 CERROR("Unable to process log: %d\n", err);
1001 GOTO(out_free, err);
1004 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1005 lprof = class_get_profile(profilenm);
1006 if (lprof == NULL) {
1007 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1008 " read from the MGS. Does that filesystem "
1009 "exist?\n", profilenm);
1010 GOTO(out_free, err = -EINVAL);
1012 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1013 lprof->lp_md, lprof->lp_dt);
1015 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1017 GOTO(out_free, err = -ENOMEM);
1018 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1020 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1022 GOTO(out_free, err = -ENOMEM);
1023 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1025 /* connections, registrations, sb setup */
1026 err = client_common_fill_super(sb, md, dt, mnt);
1030 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1032 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1035 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1036 LCONSOLE_WARN("Mounted %s\n", profilenm);
1040 } /* ll_fill_super */
1043 void lu_context_keys_dump(void);
1045 void ll_put_super(struct super_block *sb)
1047 struct config_llog_instance cfg;
1048 struct obd_device *obd;
1049 struct lustre_sb_info *lsi = s2lsi(sb);
1050 struct ll_sb_info *sbi = ll_s2sbi(sb);
1051 char *profilenm = get_profile_name(sb);
1052 int force = 1, next;
1055 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1057 ll_print_capa_stat(sbi);
1059 cfg.cfg_instance = sb;
1060 lustre_end_log(sb, profilenm, &cfg);
1062 if (sbi->ll_md_exp) {
1063 obd = class_exp2obd(sbi->ll_md_exp);
1065 force = obd->obd_force;
1068 /* We need to set force before the lov_disconnect in
1069 lustre_common_put_super, since l_d cleans up osc's as well. */
1072 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1074 obd->obd_force = force;
1079 /* Only if client_common_fill_super succeeded */
1080 client_common_put_super(sb);
1084 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1085 class_manual_cleanup(obd);
1088 if (sbi->ll_flags & LL_SBI_VERBOSE)
1089 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1092 class_del_profile(profilenm);
1094 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1095 ll_bdi_destroy(&lsi->lsi_bdi);
1096 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1100 lsi->lsi_llsbi = NULL;
1102 lustre_common_put_super(sb);
1104 cl_env_cache_purge(~0);
1106 cfs_module_put(THIS_MODULE);
1109 } /* client_put_super */
1111 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1113 struct inode *inode = NULL;
1114 /* NOTE: we depend on atomic igrab() -bzzz */
1115 lock_res_and_lock(lock);
1116 if (lock->l_ast_data) {
1117 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1118 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1119 inode = igrab(lock->l_ast_data);
1121 inode = lock->l_ast_data;
1122 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1123 D_WARNING, lock, "l_ast_data %p is "
1124 "bogus: magic %08x", lock->l_ast_data,
1125 lli->lli_inode_magic);
1129 unlock_res_and_lock(lock);
1133 static int null_if_equal(struct ldlm_lock *lock, void *data)
1135 if (data == lock->l_ast_data) {
1136 lock->l_ast_data = NULL;
1138 if (lock->l_req_mode != lock->l_granted_mode)
1139 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1142 return LDLM_ITER_CONTINUE;
1145 void ll_clear_inode(struct inode *inode)
1147 struct ll_inode_info *lli = ll_i2info(inode);
1148 struct ll_sb_info *sbi = ll_i2sbi(inode);
1151 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1152 inode->i_generation, inode);
1154 if (S_ISDIR(inode->i_mode)) {
1155 /* these should have been cleared in ll_file_release */
1156 LASSERT(lli->lli_opendir_key == NULL);
1157 LASSERT(lli->lli_sai == NULL);
1158 LASSERT(lli->lli_opendir_pid == 0);
1161 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1162 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1163 null_if_equal, inode);
1165 LASSERT(!lli->lli_open_fd_write_count);
1166 LASSERT(!lli->lli_open_fd_read_count);
1167 LASSERT(!lli->lli_open_fd_exec_count);
1169 if (lli->lli_mds_write_och)
1170 ll_md_real_close(inode, FMODE_WRITE);
1171 if (lli->lli_mds_exec_och)
1172 ll_md_real_close(inode, FMODE_EXEC);
1173 if (lli->lli_mds_read_och)
1174 ll_md_real_close(inode, FMODE_READ);
1176 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1177 OBD_FREE(lli->lli_symlink_name,
1178 strlen(lli->lli_symlink_name) + 1);
1179 lli->lli_symlink_name = NULL;
1182 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1183 LASSERT(lli->lli_posix_acl == NULL);
1184 if (lli->lli_remote_perms) {
1185 free_rmtperm_hash(lli->lli_remote_perms);
1186 lli->lli_remote_perms = NULL;
1189 #ifdef CONFIG_FS_POSIX_ACL
1190 else if (lli->lli_posix_acl) {
1191 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1192 LASSERT(lli->lli_remote_perms == NULL);
1193 posix_acl_release(lli->lli_posix_acl);
1194 lli->lli_posix_acl = NULL;
1197 lli->lli_inode_magic = LLI_INODE_DEAD;
1199 ll_clear_inode_capas(inode);
1200 if (!S_ISDIR(inode->i_mode))
1201 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1204 * XXX This has to be done before lsm is freed below, because
1205 * cl_object still uses inode lsm.
1207 cl_inode_fini(inode);
1208 lli->lli_has_smd = false;
1213 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1214 struct md_open_data **mod)
1216 struct lustre_md md;
1217 struct inode *inode = dentry->d_inode;
1218 struct ll_sb_info *sbi = ll_i2sbi(inode);
1219 struct ptlrpc_request *request = NULL;
1223 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1224 LUSTRE_OPC_ANY, NULL);
1225 if (IS_ERR(op_data))
1226 RETURN(PTR_ERR(op_data));
1228 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1231 ptlrpc_req_finished(request);
1232 if (rc == -ENOENT) {
1234 /* Unlinked special device node? Or just a race?
1235 * Pretend we done everything. */
1236 if (!S_ISREG(inode->i_mode) &&
1237 !S_ISDIR(inode->i_mode)) {
1238 ia_valid = op_data->op_attr.ia_valid;
1239 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1240 rc = simple_setattr(dentry, &op_data->op_attr);
1241 op_data->op_attr.ia_valid = ia_valid;
1243 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1244 CERROR("md_setattr fails: rc = %d\n", rc);
1249 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1250 sbi->ll_md_exp, &md);
1252 ptlrpc_req_finished(request);
1256 /* We want to adjust timestamps.
1257 * If there is at least some data in file, we cleared ATTR_SIZE
1258 * to avoid update size, otherwise it is important to do.(SOM case)
1260 ia_valid = op_data->op_attr.ia_valid;
1261 /* Since we set ATTR_*_SET flags above, and already done permission
1262 * check, So don't let inode_change_ok() check it again. */
1263 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1264 rc = simple_setattr(dentry, &op_data->op_attr);
1265 op_data->op_attr.ia_valid = ia_valid;
1267 /* Extract epoch data if obtained. */
1268 op_data->op_handle = md.body->handle;
1269 op_data->op_ioepoch = md.body->ioepoch;
1271 ll_update_inode(inode, &md);
1272 ptlrpc_req_finished(request);
1277 /* Close IO epoch and send Size-on-MDS attribute update. */
1278 static int ll_setattr_done_writing(struct inode *inode,
1279 struct md_op_data *op_data,
1280 struct md_open_data *mod)
1282 struct ll_inode_info *lli = ll_i2info(inode);
1286 LASSERT(op_data != NULL);
1287 if (!S_ISREG(inode->i_mode))
1290 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1291 op_data->op_ioepoch, PFID(&lli->lli_fid));
1293 op_data->op_flags = MF_EPOCH_CLOSE;
1294 ll_done_writing_attr(inode, op_data);
1295 ll_pack_inode2opdata(inode, op_data, NULL);
1297 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1298 if (rc == -EAGAIN) {
1299 /* MDS has instructed us to obtain Size-on-MDS attribute
1300 * from OSTs and send setattr to back to MDS. */
1301 rc = ll_som_update(inode, op_data);
1303 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1309 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1311 struct obd_capa *capa;
1314 if (attr->ia_valid & ATTR_SIZE)
1315 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1317 capa = ll_mdscapa_get(inode);
1319 rc = cl_setattr_ost(inode, attr, capa);
1321 if (attr->ia_valid & ATTR_SIZE)
1322 ll_truncate_free_capa(capa);
1329 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1331 * inode_newsize_ok - may this inode be truncated to a given size
1332 * @inode: the inode to be truncated
1333 * @offset: the new size to assign to the inode
1334 * @Returns: 0 on success, -ve errno on failure
1336 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1337 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1338 * when necessary. Caller must not proceed with inode size change if failure is
1339 * returned. @inode must be a file (not directory), with appropriate
1340 * permissions to allow truncate (inode_newsize_ok does NOT check these
1343 * inode_newsize_ok must be called with i_mutex held.
1345 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1347 if (inode->i_size < offset) {
1348 unsigned long limit;
1350 limit = rlimit(RLIMIT_FSIZE);
1351 if (limit != RLIM_INFINITY && offset > limit)
1353 if (offset > inode->i_sb->s_maxbytes)
1357 * truncation of in-use swapfiles is disallowed - it would
1358 * cause subsequent swapout to scribble on the now-freed
1361 if (IS_SWAPFILE(inode))
1367 send_sig(SIGXFSZ, current, 0);
1373 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1374 * object(s) determine the file size and mtime. Otherwise, the MDS will
1375 * keep these values until such a time that objects are allocated for it.
1376 * We do the MDS operations first, as it is checking permissions for us.
1377 * We don't to the MDS RPC if there is nothing that we want to store there,
1378 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1379 * going to do an RPC anyways.
1381 * If we are doing a truncate, we will send the mtime and ctime updates
1382 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1383 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1386 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1388 struct lov_stripe_md *lsm;
1389 struct inode *inode = dentry->d_inode;
1390 struct ll_inode_info *lli = ll_i2info(inode);
1391 struct md_op_data *op_data = NULL;
1392 struct md_open_data *mod = NULL;
1393 int ia_valid = attr->ia_valid;
1394 int rc = 0, rc1 = 0;
1397 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1400 if (ia_valid & ATTR_SIZE) {
1401 /* Check new size against VFS/VM file size limit and rlimit */
1402 rc = inode_newsize_ok(inode, attr->ia_size);
1406 /* The maximum Lustre file size is variable, based on the
1407 * OST maximum object size and number of stripes. This
1408 * needs another check in addition to the VFS check above. */
1409 if (attr->ia_size > ll_file_maxbytes(inode)) {
1410 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1411 PFID(&lli->lli_fid), attr->ia_size,
1412 ll_file_maxbytes(inode));
1416 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1419 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1420 if (ia_valid & TIMES_SET_FLAGS) {
1421 if (cfs_curproc_fsuid() != inode->i_uid &&
1422 !cfs_capable(CFS_CAP_FOWNER))
1426 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1427 if (attr->ia_valid & ATTR_CTIME) {
1428 attr->ia_ctime = CFS_CURRENT_TIME;
1429 attr->ia_valid |= ATTR_CTIME_SET;
1431 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1432 attr->ia_atime = CFS_CURRENT_TIME;
1433 attr->ia_valid |= ATTR_ATIME_SET;
1435 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1436 attr->ia_mtime = CFS_CURRENT_TIME;
1437 attr->ia_valid |= ATTR_MTIME_SET;
1440 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1441 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1442 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1443 cfs_time_current_sec());
1445 /* We always do an MDS RPC, even if we're only changing the size;
1446 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1448 OBD_ALLOC_PTR(op_data);
1449 if (op_data == NULL)
1452 if (!S_ISDIR(inode->i_mode)) {
1453 if (ia_valid & ATTR_SIZE)
1454 UP_WRITE_I_ALLOC_SEM(inode);
1455 mutex_unlock(&inode->i_mutex);
1456 cfs_down_write(&lli->lli_trunc_sem);
1457 mutex_lock(&inode->i_mutex);
1458 if (ia_valid & ATTR_SIZE)
1459 DOWN_WRITE_I_ALLOC_SEM(inode);
1462 /* We need a steady stripe configuration for setattr to avoid
1464 lsm = ccc_inode_lsm_get(inode);
1466 /* NB: ATTR_SIZE will only be set after this point if the size
1467 * resides on the MDS, ie, this file has no objects. */
1469 attr->ia_valid &= ~ATTR_SIZE;
1471 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1473 /* Open epoch for truncate. */
1474 if (exp_connect_som(ll_i2mdexp(inode)) &&
1475 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1476 op_data->op_flags = MF_EPOCH_OPEN;
1478 rc = ll_md_setattr(dentry, op_data, &mod);
1482 ll_ioepoch_open(lli, op_data->op_ioepoch);
1483 if (lsm == NULL || !S_ISREG(inode->i_mode)) {
1484 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1488 if (ia_valid & ATTR_SIZE)
1489 attr->ia_valid |= ATTR_SIZE;
1490 if (ia_valid & (ATTR_SIZE |
1491 ATTR_ATIME | ATTR_ATIME_SET |
1492 ATTR_MTIME | ATTR_MTIME_SET))
1493 /* For truncate and utimes sending attributes to OSTs, setting
1494 * mtime/atime to the past will be performed under PW [0:EOF]
1495 * extent lock (new_size:EOF for truncate). It may seem
1496 * excessive to send mtime/atime updates to OSTs when not
1497 * setting times to past, but it is necessary due to possible
1498 * time de-synchronization between MDT inode and OST objects */
1499 rc = ll_setattr_ost(inode, attr);
1502 ccc_inode_lsm_put(inode, lsm);
1504 if (op_data->op_ioepoch) {
1505 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1509 ll_finish_md_op_data(op_data);
1511 if (!S_ISDIR(inode->i_mode))
1512 cfs_up_write(&lli->lli_trunc_sem);
1514 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1515 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1520 int ll_setattr(struct dentry *de, struct iattr *attr)
1522 int mode = de->d_inode->i_mode;
1524 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1525 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1526 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1528 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1529 (ATTR_SIZE|ATTR_MODE)) &&
1530 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1531 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1532 !(attr->ia_mode & S_ISGID))))
1533 attr->ia_valid |= ATTR_FORCE;
1535 if ((mode & S_ISUID) &&
1536 !(attr->ia_mode & S_ISUID) &&
1537 !(attr->ia_valid & ATTR_KILL_SUID))
1538 attr->ia_valid |= ATTR_KILL_SUID;
1540 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1541 !(attr->ia_mode & S_ISGID) &&
1542 !(attr->ia_valid & ATTR_KILL_SGID))
1543 attr->ia_valid |= ATTR_KILL_SGID;
1545 return ll_setattr_raw(de, attr);
1548 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1549 __u64 max_age, __u32 flags)
1551 struct ll_sb_info *sbi = ll_s2sbi(sb);
1552 struct obd_statfs obd_osfs;
1556 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1558 CERROR("md_statfs fails: rc = %d\n", rc);
1562 osfs->os_type = sb->s_magic;
1564 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1565 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1567 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1568 flags |= OBD_STATFS_NODELAY;
1570 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1572 CERROR("obd_statfs fails: rc = %d\n", rc);
1576 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1577 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1580 osfs->os_bsize = obd_osfs.os_bsize;
1581 osfs->os_blocks = obd_osfs.os_blocks;
1582 osfs->os_bfree = obd_osfs.os_bfree;
1583 osfs->os_bavail = obd_osfs.os_bavail;
1585 /* If we don't have as many objects free on the OST as inodes
1586 * on the MDS, we reduce the total number of inodes to
1587 * compensate, so that the "inodes in use" number is correct.
1589 if (obd_osfs.os_ffree < osfs->os_ffree) {
1590 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1592 osfs->os_ffree = obd_osfs.os_ffree;
1597 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1599 struct super_block *sb = de->d_sb;
1600 struct obd_statfs osfs;
1603 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1604 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1606 /* Some amount of caching on the client is allowed */
1607 rc = ll_statfs_internal(sb, &osfs,
1608 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1613 statfs_unpack(sfs, &osfs);
1615 /* We need to downshift for all 32-bit kernels, because we can't
1616 * tell if the kernel is being called via sys_statfs64() or not.
1617 * Stop before overflowing f_bsize - in which case it is better
1618 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1619 if (sizeof(long) < 8) {
1620 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1623 osfs.os_blocks >>= 1;
1624 osfs.os_bfree >>= 1;
1625 osfs.os_bavail >>= 1;
1629 sfs->f_blocks = osfs.os_blocks;
1630 sfs->f_bfree = osfs.os_bfree;
1631 sfs->f_bavail = osfs.os_bavail;
1636 void ll_inode_size_lock(struct inode *inode)
1638 struct ll_inode_info *lli;
1640 LASSERT(!S_ISDIR(inode->i_mode));
1642 lli = ll_i2info(inode);
1643 LASSERT(lli->lli_size_sem_owner != current);
1644 cfs_down(&lli->lli_size_sem);
1645 LASSERT(lli->lli_size_sem_owner == NULL);
1646 lli->lli_size_sem_owner = current;
1649 void ll_inode_size_unlock(struct inode *inode)
1651 struct ll_inode_info *lli;
1653 lli = ll_i2info(inode);
1654 LASSERT(lli->lli_size_sem_owner == current);
1655 lli->lli_size_sem_owner = NULL;
1656 cfs_up(&lli->lli_size_sem);
1659 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1661 struct ll_inode_info *lli = ll_i2info(inode);
1662 struct mdt_body *body = md->body;
1663 struct lov_stripe_md *lsm = md->lsm;
1664 struct ll_sb_info *sbi = ll_i2sbi(inode);
1666 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1668 LASSERT(S_ISREG(inode->i_mode));
1669 cfs_mutex_lock(&lli->lli_och_mutex);
1670 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1671 lsm, inode->i_ino, inode->i_generation, inode);
1672 /* cl_file_inode_init must go before lli_has_smd or a race
1673 * is possible where client thinks the file has stripes,
1674 * but lov raid0 is not setup yet and parallel e.g.
1675 * glimpse would try to use uninitialized lov */
1676 if (cl_file_inode_init(inode, md) == 0)
1677 lli->lli_has_smd = true;
1678 cfs_mutex_unlock(&lli->lli_och_mutex);
1680 lli->lli_maxbytes = lsm->lsm_maxbytes;
1681 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1682 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1683 if (md->lsm != NULL)
1684 obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1687 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1688 if (body->valid & OBD_MD_FLRMTPERM)
1689 ll_update_remote_perm(inode, md->remote_perm);
1691 #ifdef CONFIG_FS_POSIX_ACL
1692 else if (body->valid & OBD_MD_FLACL) {
1693 cfs_spin_lock(&lli->lli_lock);
1694 if (lli->lli_posix_acl)
1695 posix_acl_release(lli->lli_posix_acl);
1696 lli->lli_posix_acl = md->posix_acl;
1697 cfs_spin_unlock(&lli->lli_lock);
1700 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1701 inode->i_generation = cl_fid_build_gen(&body->fid1);
1703 if (body->valid & OBD_MD_FLATIME) {
1704 if (body->atime > LTIME_S(inode->i_atime))
1705 LTIME_S(inode->i_atime) = body->atime;
1706 lli->lli_lvb.lvb_atime = body->atime;
1708 if (body->valid & OBD_MD_FLMTIME) {
1709 if (body->mtime > LTIME_S(inode->i_mtime)) {
1710 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1711 "to "LPU64"\n", inode->i_ino,
1712 LTIME_S(inode->i_mtime), body->mtime);
1713 LTIME_S(inode->i_mtime) = body->mtime;
1715 lli->lli_lvb.lvb_mtime = body->mtime;
1717 if (body->valid & OBD_MD_FLCTIME) {
1718 if (body->ctime > LTIME_S(inode->i_ctime))
1719 LTIME_S(inode->i_ctime) = body->ctime;
1720 lli->lli_lvb.lvb_ctime = body->ctime;
1722 if (body->valid & OBD_MD_FLMODE)
1723 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1724 if (body->valid & OBD_MD_FLTYPE)
1725 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1726 LASSERT(inode->i_mode != 0);
1727 if (S_ISREG(inode->i_mode)) {
1728 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1730 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1732 if (body->valid & OBD_MD_FLUID)
1733 inode->i_uid = body->uid;
1734 if (body->valid & OBD_MD_FLGID)
1735 inode->i_gid = body->gid;
1736 if (body->valid & OBD_MD_FLFLAGS)
1737 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1738 if (body->valid & OBD_MD_FLNLINK)
1739 inode->i_nlink = body->nlink;
1740 if (body->valid & OBD_MD_FLRDEV)
1741 inode->i_rdev = old_decode_dev(body->rdev);
1743 if (body->valid & OBD_MD_FLID) {
1744 /* FID shouldn't be changed! */
1745 if (fid_is_sane(&lli->lli_fid)) {
1746 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1747 "Trying to change FID "DFID
1748 " to the "DFID", inode %lu/%u(%p)\n",
1749 PFID(&lli->lli_fid), PFID(&body->fid1),
1750 inode->i_ino, inode->i_generation, inode);
1752 lli->lli_fid = body->fid1;
1755 LASSERT(fid_seq(&lli->lli_fid) != 0);
1757 if (body->valid & OBD_MD_FLSIZE) {
1758 if (exp_connect_som(ll_i2mdexp(inode)) &&
1759 S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1760 struct lustre_handle lockh;
1763 /* As it is possible a blocking ast has been processed
1764 * by this time, we need to check there is an UPDATE
1765 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1767 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1770 if (lli->lli_flags & (LLIF_DONE_WRITING |
1771 LLIF_EPOCH_PENDING |
1773 CERROR("ino %lu flags %u still has "
1774 "size authority! do not trust "
1775 "the size got from MDS\n",
1776 inode->i_ino, lli->lli_flags);
1778 /* Use old size assignment to avoid
1779 * deadlock bz14138 & bz14326 */
1780 i_size_write(inode, body->size);
1781 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1783 ldlm_lock_decref(&lockh, mode);
1786 /* Use old size assignment to avoid
1787 * deadlock bz14138 & bz14326 */
1788 i_size_write(inode, body->size);
1790 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1791 inode->i_ino, (unsigned long long)body->size);
1794 if (body->valid & OBD_MD_FLBLOCKS)
1795 inode->i_blocks = body->blocks;
1798 if (body->valid & OBD_MD_FLMDSCAPA) {
1799 LASSERT(md->mds_capa);
1800 ll_add_capa(inode, md->mds_capa);
1802 if (body->valid & OBD_MD_FLOSSCAPA) {
1803 LASSERT(md->oss_capa);
1804 ll_add_capa(inode, md->oss_capa);
1808 void ll_read_inode2(struct inode *inode, void *opaque)
1810 struct lustre_md *md = opaque;
1811 struct ll_inode_info *lli = ll_i2info(inode);
1814 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1815 PFID(&lli->lli_fid), inode);
1817 LASSERT(!lli->lli_has_smd);
1819 /* Core attributes from the MDS first. This is a new inode, and
1820 * the VFS doesn't zero times in the core inode so we have to do
1821 * it ourselves. They will be overwritten by either MDS or OST
1822 * attributes - we just need to make sure they aren't newer. */
1823 LTIME_S(inode->i_mtime) = 0;
1824 LTIME_S(inode->i_atime) = 0;
1825 LTIME_S(inode->i_ctime) = 0;
1827 ll_update_inode(inode, md);
1829 /* OIDEBUG(inode); */
1831 /* initializing backing dev info. */
1832 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1835 if (S_ISREG(inode->i_mode)) {
1836 struct ll_sb_info *sbi = ll_i2sbi(inode);
1837 inode->i_op = &ll_file_inode_operations;
1838 inode->i_fop = sbi->ll_fop;
1839 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1841 } else if (S_ISDIR(inode->i_mode)) {
1842 inode->i_op = &ll_dir_inode_operations;
1843 inode->i_fop = &ll_dir_operations;
1845 } else if (S_ISLNK(inode->i_mode)) {
1846 inode->i_op = &ll_fast_symlink_inode_operations;
1849 inode->i_op = &ll_special_inode_operations;
1851 init_special_inode(inode, inode->i_mode,
1852 kdev_t_to_nr(inode->i_rdev));
1858 void ll_delete_inode(struct inode *inode)
1860 struct cl_inode_info *lli = cl_i2info(inode);
1861 struct ll_sb_info *sbi = ll_i2sbi(inode);
1865 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1867 CERROR("fid_delete() failed, rc %d\n", rc);
1869 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1870 /* discard all dirty pages before truncating them, required by
1871 * osc_extent implementation at LU-1030. */
1872 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1874 truncate_inode_pages(&inode->i_data, 0);
1876 /* Workaround for LU-118 */
1877 if (inode->i_data.nrpages) {
1878 TREE_READ_LOCK_IRQ(&inode->i_data);
1879 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1880 LASSERTF(inode->i_data.nrpages == 0,
1881 "inode=%lu/%u(%p) nrpages=%lu, see "
1882 "http://jira.whamcloud.com/browse/LU-118\n",
1883 inode->i_ino, inode->i_generation, inode,
1884 inode->i_data.nrpages);
1886 /* Workaround end */
1888 #ifdef HAVE_SBOPS_EVICT_INODE
1889 ll_clear_inode(inode);
1890 end_writeback(inode);
1898 int ll_iocontrol(struct inode *inode, struct file *file,
1899 unsigned int cmd, unsigned long arg)
1901 struct ll_sb_info *sbi = ll_i2sbi(inode);
1902 struct ptlrpc_request *req = NULL;
1907 case FSFILT_IOC_GETFLAGS: {
1908 struct mdt_body *body;
1909 struct md_op_data *op_data;
1911 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1912 0, 0, LUSTRE_OPC_ANY,
1914 if (IS_ERR(op_data))
1915 RETURN(PTR_ERR(op_data));
1917 op_data->op_valid = OBD_MD_FLFLAGS;
1918 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1919 ll_finish_md_op_data(op_data);
1921 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1925 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1927 flags = body->flags;
1929 ptlrpc_req_finished(req);
1931 RETURN(put_user(flags, (int *)arg));
1933 case FSFILT_IOC_SETFLAGS: {
1934 struct lov_stripe_md *lsm;
1935 struct obd_info oinfo = { { { 0 } } };
1936 struct md_op_data *op_data;
1938 if (get_user(flags, (int *)arg))
1941 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1942 LUSTRE_OPC_ANY, NULL);
1943 if (IS_ERR(op_data))
1944 RETURN(PTR_ERR(op_data));
1946 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1947 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1948 rc = md_setattr(sbi->ll_md_exp, op_data,
1949 NULL, 0, NULL, 0, &req, NULL);
1950 ll_finish_md_op_data(op_data);
1951 ptlrpc_req_finished(req);
1955 OBDO_ALLOC(oinfo.oi_oa);
1959 lsm = ccc_inode_lsm_get(inode);
1961 inode->i_flags = ll_ext_to_inode_flags(flags);
1962 OBDO_FREE(oinfo.oi_oa);
1967 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1968 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1969 oinfo.oi_oa->o_flags = flags;
1970 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1972 oinfo.oi_capa = ll_mdscapa_get(inode);
1973 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1974 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1975 capa_put(oinfo.oi_capa);
1976 OBDO_FREE(oinfo.oi_oa);
1977 ccc_inode_lsm_put(inode, lsm);
1979 if (rc && rc != -EPERM && rc != -EACCES)
1980 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1991 int ll_flush_ctx(struct inode *inode)
1993 struct ll_sb_info *sbi = ll_i2sbi(inode);
1995 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1997 obd_set_info_async(NULL, sbi->ll_md_exp,
1998 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2000 obd_set_info_async(NULL, sbi->ll_dt_exp,
2001 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2006 /* umount -f client means force down, don't save state */
2007 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2008 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2010 struct super_block *sb = vfsmnt->mnt_sb;
2012 void ll_umount_begin(struct super_block *sb)
2015 struct ll_sb_info *sbi = ll_s2sbi(sb);
2016 struct obd_device *obd;
2017 struct obd_ioctl_data *ioc_data;
2020 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2021 if (!(flags & MNT_FORCE)) {
2027 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2028 sb->s_count, atomic_read(&sb->s_active));
2030 obd = class_exp2obd(sbi->ll_md_exp);
2032 CERROR("Invalid MDC connection handle "LPX64"\n",
2033 sbi->ll_md_exp->exp_handle.h_cookie);
2039 obd = class_exp2obd(sbi->ll_dt_exp);
2041 CERROR("Invalid LOV connection handle "LPX64"\n",
2042 sbi->ll_dt_exp->exp_handle.h_cookie);
2048 OBD_ALLOC_PTR(ioc_data);
2050 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2051 sizeof ioc_data, ioc_data, NULL);
2053 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2054 sizeof ioc_data, ioc_data, NULL);
2056 OBD_FREE_PTR(ioc_data);
2060 /* Really, we'd like to wait until there are no requests outstanding,
2061 * and then continue. For now, we just invalidate the requests,
2062 * schedule() and sleep one second if needed, and hope.
2065 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2066 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2067 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2068 cfs_time_seconds(1));
2069 if (atomic_read(&vfsmnt->mnt_count) > 2)
2070 LCONSOLE_WARN("Mount still busy with %d refs! You "
2071 "may try to umount it a bit later\n",
2072 atomic_read(&vfsmnt->mnt_count));
2079 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2081 struct ll_sb_info *sbi = ll_s2sbi(sb);
2082 char *profilenm = get_profile_name(sb);
2086 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2087 read_only = *flags & MS_RDONLY;
2088 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2089 sizeof(KEY_READ_ONLY),
2090 KEY_READ_ONLY, sizeof(read_only),
2093 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2094 profilenm, read_only ?
2095 "read-only" : "read-write", err);
2100 sb->s_flags |= MS_RDONLY;
2102 sb->s_flags &= ~MS_RDONLY;
2104 if (sbi->ll_flags & LL_SBI_VERBOSE)
2105 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2106 read_only ? "read-only" : "read-write");
2111 int ll_prep_inode(struct inode **inode,
2112 struct ptlrpc_request *req,
2113 struct super_block *sb)
2115 struct ll_sb_info *sbi = NULL;
2116 struct lustre_md md;
2121 LASSERT(*inode || sb);
2122 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2123 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2124 sbi->ll_md_exp, &md);
2129 ll_update_inode(*inode, &md);
2131 LASSERT(sb != NULL);
2134 * At this point server returns to client's same fid as client
2135 * generated for creating. So using ->fid1 is okay here.
2137 LASSERT(fid_is_sane(&md.body->fid1));
2139 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2140 if (*inode == NULL || IS_ERR(*inode)) {
2142 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2143 #ifdef CONFIG_FS_POSIX_ACL
2145 posix_acl_release(md.posix_acl);
2146 md.posix_acl = NULL;
2149 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2151 CERROR("new_inode -fatal: rc %d\n", rc);
2156 /* sanity check for LAYOUT lock. */
2157 ibits = MDS_INODELOCK_LAYOUT;
2158 if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2159 md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2160 char *fsname = ll_get_fsname(*inode);
2161 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2162 fsname, PFID(ll_inode2fid(*inode)),
2165 OBD_FREE(fsname, MGS_PARAM_MAXLEN);
2169 md_free_lustre_md(sbi->ll_md_exp, &md);
2173 int ll_obd_statfs(struct inode *inode, void *arg)
2175 struct ll_sb_info *sbi = NULL;
2176 struct obd_export *exp;
2178 struct obd_ioctl_data *data = NULL;
2183 if (!inode || !(sbi = ll_i2sbi(inode)))
2184 GOTO(out_statfs, rc = -EINVAL);
2186 rc = obd_ioctl_getdata(&buf, &len, arg);
2188 GOTO(out_statfs, rc);
2191 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2192 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2193 GOTO(out_statfs, rc = -EINVAL);
2195 if (data->ioc_inllen1 != sizeof(__u32) ||
2196 data->ioc_inllen2 != sizeof(__u32) ||
2197 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2198 data->ioc_plen2 != sizeof(struct obd_uuid))
2199 GOTO(out_statfs, rc = -EINVAL);
2201 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2202 if (type & LL_STATFS_LMV)
2203 exp = sbi->ll_md_exp;
2204 else if (type & LL_STATFS_LOV)
2205 exp = sbi->ll_dt_exp;
2207 GOTO(out_statfs, rc = -ENODEV);
2209 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2210 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2212 GOTO(out_statfs, rc);
2215 obd_ioctl_freedata(buf, len);
2219 int ll_process_config(struct lustre_cfg *lcfg)
2223 struct lprocfs_static_vars lvars;
2227 lprocfs_llite_init_vars(&lvars);
2229 /* The instance name contains the sb: lustre-client-aacfe000 */
2230 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2231 if (!ptr || !*(++ptr))
2233 if (sscanf(ptr, "%lx", &x) != 1)
2236 /* This better be a real Lustre superblock! */
2237 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2239 /* Note we have not called client_common_fill_super yet, so
2240 proc fns must be able to handle that! */
2241 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2248 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2249 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2250 struct inode *i1, struct inode *i2,
2251 const char *name, int namelen,
2252 int mode, __u32 opc, void *data)
2254 LASSERT(i1 != NULL);
2256 if (namelen > ll_i2sbi(i1)->ll_namelen)
2257 return ERR_PTR(-ENAMETOOLONG);
2259 if (op_data == NULL)
2260 OBD_ALLOC_PTR(op_data);
2262 if (op_data == NULL)
2263 return ERR_PTR(-ENOMEM);
2265 ll_i2gids(op_data->op_suppgids, i1, i2);
2266 op_data->op_fid1 = *ll_inode2fid(i1);
2267 op_data->op_capa1 = ll_mdscapa_get(i1);
2270 op_data->op_fid2 = *ll_inode2fid(i2);
2271 op_data->op_capa2 = ll_mdscapa_get(i2);
2273 fid_zero(&op_data->op_fid2);
2274 op_data->op_capa2 = NULL;
2277 op_data->op_name = name;
2278 op_data->op_namelen = namelen;
2279 op_data->op_mode = mode;
2280 op_data->op_mod_time = cfs_time_current_sec();
2281 op_data->op_fsuid = cfs_curproc_fsuid();
2282 op_data->op_fsgid = cfs_curproc_fsgid();
2283 op_data->op_cap = cfs_curproc_cap_pack();
2284 op_data->op_bias = MDS_CHECK_SPLIT;
2285 op_data->op_opc = opc;
2286 op_data->op_mds = 0;
2287 op_data->op_data = data;
2289 /* If the file is being opened after mknod() (normally due to NFS)
2290 * try to use the default stripe data from parent directory for
2291 * allocating OST objects. Try to pass the parent FID to MDS. */
2292 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2293 !ll_i2info(i2)->lli_has_smd) {
2294 struct ll_inode_info *lli = ll_i2info(i2);
2296 cfs_spin_lock(&lli->lli_lock);
2297 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2298 op_data->op_fid1 = lli->lli_pfid;
2299 cfs_spin_unlock(&lli->lli_lock);
2300 /** We ignore parent's capability temporary. */
2306 void ll_finish_md_op_data(struct md_op_data *op_data)
2308 capa_put(op_data->op_capa1);
2309 capa_put(op_data->op_capa2);
2310 OBD_FREE_PTR(op_data);
2313 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2315 struct ll_sb_info *sbi;
2317 LASSERT((seq != NULL) && (vfs != NULL));
2318 sbi = ll_s2sbi(vfs->mnt_sb);
2320 if (sbi->ll_flags & LL_SBI_NOLCK)
2321 seq_puts(seq, ",nolock");
2323 if (sbi->ll_flags & LL_SBI_FLOCK)
2324 seq_puts(seq, ",flock");
2326 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2327 seq_puts(seq, ",localflock");
2329 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2330 seq_puts(seq, ",user_xattr");
2332 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2333 seq_puts(seq, ",lazystatfs");
2335 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2336 seq_puts(seq, ",user_fid2path");
2342 * Get obd name by cmd, and copy out to user space
2344 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2346 struct ll_sb_info *sbi = ll_i2sbi(inode);
2347 struct obd_device *obd;
2350 if (cmd == OBD_IOC_GETDTNAME)
2351 obd = class_exp2obd(sbi->ll_dt_exp);
2352 else if (cmd == OBD_IOC_GETMDNAME)
2353 obd = class_exp2obd(sbi->ll_md_exp);
2360 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2361 strlen(obd->obd_name) + 1))