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, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/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) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
84 OBD_ALLOC(sbi, sizeof(*sbi));
88 spin_lock_init(&sbi->ll_lock);
89 mutex_init(&sbi->ll_lco.lco_lock);
90 spin_lock_init(&sbi->ll_pp_extent_lock);
91 spin_lock_init(&sbi->ll_process_lock);
92 sbi->ll_rw_stats_on = 0;
95 pages = si.totalram - si.totalhigh;
96 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
97 lru_page_max = pages / 2;
99 lru_page_max = (pages / 4) * 3;
102 /* initialize lru data */
103 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
104 sbi->ll_cache.ccc_lru_max = lru_page_max;
105 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
106 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
107 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
109 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
110 SBI_DEFAULT_READAHEAD_MAX);
111 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
112 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
113 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
114 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
115 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
117 ll_generate_random_uuid(uuid);
118 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
119 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
121 spin_lock(&ll_sb_lock);
122 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
123 spin_unlock(&ll_sb_lock);
125 sbi->ll_flags |= LL_SBI_VERBOSE;
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 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
137 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);
145 cfs_atomic_set(&sbi->ll_agl_total, 0);
146 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151 void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 spin_lock(&ll_sb_lock);
158 cfs_list_del(&sbi->ll_list);
159 spin_unlock(&ll_sb_lock);
160 OBD_FREE(sbi, sizeof(*sbi));
165 static struct dentry_operations ll_d_root_ops = {
166 .d_compare = ll_dcompare,
167 .d_revalidate = ll_revalidate_nd,
170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
171 struct vfsmount *mnt)
173 struct inode *root = 0;
174 struct ll_sb_info *sbi = ll_s2sbi(sb);
175 struct obd_device *obd;
176 struct obd_capa *oc = NULL;
177 struct obd_statfs *osfs = NULL;
178 struct ptlrpc_request *request = NULL;
179 struct obd_connect_data *data = NULL;
180 struct obd_uuid *uuid;
181 struct md_op_data *op_data;
182 struct lustre_md lmd;
184 int size, err, checksum;
187 obd = class_name2obd(md);
189 CERROR("MD %s: not setup or attached\n", md);
203 if (proc_lustre_fs_root) {
204 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
207 CERROR("could not register mount in /proc/fs/lustre\n");
210 /* indicate the features supported by this client */
211 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
212 OBD_CONNECT_ATTRFID |
213 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
214 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
215 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
216 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
217 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
218 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
219 OBD_CONNECT_EINPROGRESS |
220 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
221 OBD_CONNECT_LAYOUTLOCK;
223 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
224 data->ocd_connect_flags |= OBD_CONNECT_SOM;
226 #ifdef HAVE_LRU_RESIZE_SUPPORT
227 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
228 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
230 #ifdef CONFIG_FS_POSIX_ACL
231 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
234 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
235 /* flag mdc connection as lightweight, only used for test
236 * purpose, use with care */
237 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
239 data->ocd_ibits_known = MDS_INODELOCK_FULL;
240 data->ocd_version = LUSTRE_VERSION_CODE;
242 if (sb->s_flags & MS_RDONLY)
243 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
244 if (sbi->ll_flags & LL_SBI_USER_XATTR)
245 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
247 #ifdef HAVE_MS_FLOCK_LOCK
248 /* force vfs to use lustre handler for flock() calls - bug 10743 */
249 sb->s_flags |= MS_FLOCK_LOCK;
251 #ifdef MS_HAS_NEW_AOPS
252 sb->s_flags |= MS_HAS_NEW_AOPS;
255 if (sbi->ll_flags & LL_SBI_FLOCK)
256 sbi->ll_fop = &ll_file_operations_flock;
257 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
258 sbi->ll_fop = &ll_file_operations;
260 sbi->ll_fop = &ll_file_operations_noflock;
263 data->ocd_connect_flags |= OBD_CONNECT_REAL;
264 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
265 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
267 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
269 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
271 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
272 "recovery, of which this client is not a "
273 "part. Please wait for recovery to complete,"
274 " abort, or time out.\n", md);
277 CERROR("cannot connect to %s: rc = %d\n", md, err);
281 /* For mount, we only need fs info from MDT0, and also in DNE, it
282 * can make sure the client can be mounted as long as MDT0 is
284 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
285 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
286 OBD_STATFS_FOR_MDT0);
290 /* This needs to be after statfs to ensure connect has finished.
291 * Note that "data" does NOT contain the valid connect reply.
292 * If connecting to a 1.8 server there will be no LMV device, so
293 * we can access the MDC export directly and exp_connect_flags will
294 * be non-zero, but if accessing an upgraded 2.1 server it will
295 * have the correct flags filled in.
296 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
297 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
298 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
299 valid != CLIENT_CONNECT_MDT_REQD) {
302 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
303 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
304 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
305 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
306 "feature(s) needed for correct operation "
307 "of this client (%s). Please upgrade "
308 "server or downgrade client.\n",
309 sbi->ll_md_exp->exp_obd->obd_name, buf);
310 OBD_FREE(buf, CFS_PAGE_SIZE);
311 GOTO(out_md, err = -EPROTO);
314 size = sizeof(*data);
315 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
316 KEY_CONN_DATA, &size, data, NULL);
318 CERROR("%s: Get connect data failed: rc = %d\n",
319 sbi->ll_md_exp->exp_obd->obd_name, err);
323 LASSERT(osfs->os_bsize);
324 sb->s_blocksize = osfs->os_bsize;
325 sb->s_blocksize_bits = log2(osfs->os_bsize);
326 sb->s_magic = LL_SUPER_MAGIC;
327 sb->s_maxbytes = MAX_LFS_FILESIZE;
328 sbi->ll_namelen = osfs->os_namelen;
329 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
331 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
332 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
333 LCONSOLE_INFO("Disabling user_xattr feature because "
334 "it is not supported on the server\n");
335 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
338 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
340 sb->s_flags |= MS_POSIXACL;
342 sbi->ll_flags |= LL_SBI_ACL;
344 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
346 sb->s_flags &= ~MS_POSIXACL;
348 sbi->ll_flags &= ~LL_SBI_ACL;
351 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
352 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
353 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
354 LCONSOLE_INFO("client is set as remote by default.\n");
357 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
358 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
359 LCONSOLE_INFO("client claims to be remote, but server "
360 "rejected, forced to be local.\n");
364 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
365 LCONSOLE_INFO("client enabled MDS capability!\n");
366 sbi->ll_flags |= LL_SBI_MDS_CAPA;
369 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
370 LCONSOLE_INFO("client enabled OSS capability!\n");
371 sbi->ll_flags |= LL_SBI_OSS_CAPA;
374 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
375 sbi->ll_flags |= LL_SBI_64BIT_HASH;
377 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
378 sbi->ll_md_brw_size = data->ocd_brw_size;
380 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
382 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
383 LCONSOLE_INFO("Layout lock feature supported.\n");
384 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
387 obd = class_name2obd(dt);
389 CERROR("DT %s: not setup or attached\n", dt);
390 GOTO(out_md, err = -ENODEV);
393 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
394 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
395 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
396 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
397 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
398 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
399 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
400 OBD_CONNECT_MAXBYTES |
401 OBD_CONNECT_EINPROGRESS |
402 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
403 OBD_CONNECT_LAYOUTLOCK;
405 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
406 data->ocd_connect_flags |= OBD_CONNECT_SOM;
408 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
409 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
410 * disabled by default, because it can still be enabled on the
411 * fly via /proc. As a consequence, we still need to come to an
412 * agreement on the supported algorithms at connect time */
413 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
415 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
416 data->ocd_cksum_types = OBD_CKSUM_ADLER;
418 data->ocd_cksum_types = cksum_types_supported_client();
421 #ifdef HAVE_LRU_RESIZE_SUPPORT
422 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
424 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
425 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
427 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
428 "ocd_grant: %d\n", data->ocd_connect_flags,
429 data->ocd_version, data->ocd_grant);
431 obd->obd_upcall.onu_owner = &sbi->ll_lco;
432 obd->obd_upcall.onu_upcall = cl_ocd_update;
434 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
436 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
439 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
440 "recovery, of which this client is not a "
441 "part. Please wait for recovery to "
442 "complete, abort, or time out.\n", dt);
445 CERROR("%s: Cannot connect to %s: rc = %d\n",
446 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
450 mutex_lock(&sbi->ll_lco.lco_lock);
451 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
452 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
453 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
454 mutex_unlock(&sbi->ll_lco.lco_lock);
456 fid_zero(&sbi->ll_root_fid);
457 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
459 CERROR("cannot mds_connect: rc = %d\n", err);
462 if (!fid_is_sane(&sbi->ll_root_fid)) {
463 CERROR("%s: Invalid root fid "DFID" during mount\n",
464 sbi->ll_md_exp->exp_obd->obd_name,
465 PFID(&sbi->ll_root_fid));
466 GOTO(out_dt, err = -EINVAL);
468 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
470 sb->s_op = &lustre_super_operations;
471 #if THREAD_SIZE >= 8192 /*b=17630*/
472 sb->s_export_op = &lustre_export_operations;
476 * XXX: move this to after cbd setup? */
477 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
478 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
479 valid |= OBD_MD_FLRMTPERM;
480 else if (sbi->ll_flags & LL_SBI_ACL)
481 valid |= OBD_MD_FLACL;
483 OBD_ALLOC_PTR(op_data);
485 GOTO(out_dt, err = -ENOMEM);
487 op_data->op_fid1 = sbi->ll_root_fid;
488 op_data->op_mode = 0;
489 op_data->op_capa1 = oc;
490 op_data->op_valid = valid;
492 err = md_getattr(sbi->ll_md_exp, op_data, &request);
495 OBD_FREE_PTR(op_data);
497 CERROR("%s: md_getattr failed for root: rc = %d\n",
498 sbi->ll_md_exp->exp_obd->obd_name, err);
502 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
503 sbi->ll_md_exp, &lmd);
505 CERROR("failed to understand root inode md: rc = %d\n", err);
506 ptlrpc_req_finished(request);
510 LASSERT(fid_is_sane(&sbi->ll_root_fid));
511 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
512 md_free_lustre_md(sbi->ll_md_exp, &lmd);
513 ptlrpc_req_finished(request);
515 if (root == NULL || IS_ERR(root)) {
517 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
518 #ifdef CONFIG_FS_POSIX_ACL
520 posix_acl_release(lmd.posix_acl);
521 lmd.posix_acl = NULL;
524 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
526 CERROR("lustre_lite: bad iget4 for root\n");
530 err = ll_close_thread_start(&sbi->ll_lcq);
532 CERROR("cannot start close thread: rc %d\n", err);
536 #ifdef CONFIG_FS_POSIX_ACL
537 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
538 rct_init(&sbi->ll_rct);
539 et_init(&sbi->ll_et);
543 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
544 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
545 KEY_CHECKSUM, sizeof(checksum), &checksum,
549 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
550 KEY_CACHE_SET, sizeof(sbi->ll_cache),
551 &sbi->ll_cache, NULL);
553 sb->s_root = d_make_root(root);
554 if (sb->s_root == NULL) {
555 CERROR("%s: can't make root dentry\n",
556 ll_get_fsname(sb, NULL, 0));
557 GOTO(out_root, err = -ENOMEM);
560 #ifdef HAVE_DCACHE_LOCK
561 sb->s_root->d_op = &ll_d_root_ops;
563 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
564 d_set_d_op(sb->s_root, &ll_d_root_ops);
565 sb->s_d_op = &ll_d_ops;
568 sbi->ll_sdev_orig = sb->s_dev;
570 /* We set sb->s_dev equal on all lustre clients in order to support
571 * NFS export clustering. NFSD requires that the FSID be the same
573 /* s_dev is also used in lt_compare() to compare two fs, but that is
574 * only a node-local comparison. */
575 uuid = obd_get_uuid(sbi->ll_md_exp);
577 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
589 obd_disconnect(sbi->ll_dt_exp);
590 sbi->ll_dt_exp = NULL;
591 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
592 obd_zombie_barrier();
594 obd_disconnect(sbi->ll_md_exp);
595 sbi->ll_md_exp = NULL;
601 lprocfs_unregister_mountpoint(sbi);
605 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
609 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
611 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
612 KEY_MAX_EASIZE, &size, lmmsize, NULL);
614 CERROR("Get max mdsize error rc %d \n", rc);
619 void ll_dump_inode(struct inode *inode)
621 struct ll_d_hlist_node *tmp;
622 int dentry_count = 0;
624 LASSERT(inode != NULL);
626 ll_d_hlist_for_each(tmp, &inode->i_dentry)
629 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
630 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
631 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
634 void lustre_dump_dentry(struct dentry *dentry, int recur)
636 struct list_head *tmp;
639 LASSERT(dentry != NULL);
641 list_for_each(tmp, &dentry->d_subdirs)
644 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
645 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
646 dentry->d_name.len, dentry->d_name.name,
647 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
648 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
649 dentry->d_flags, dentry->d_fsdata, subdirs);
650 if (dentry->d_inode != NULL)
651 ll_dump_inode(dentry->d_inode);
656 list_for_each(tmp, &dentry->d_subdirs) {
657 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
658 lustre_dump_dentry(d, recur - 1);
662 void client_common_put_super(struct super_block *sb)
664 struct ll_sb_info *sbi = ll_s2sbi(sb);
667 #ifdef CONFIG_FS_POSIX_ACL
668 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
669 et_fini(&sbi->ll_et);
670 rct_fini(&sbi->ll_rct);
674 ll_close_thread_shutdown(sbi->ll_lcq);
678 cfs_list_del(&sbi->ll_conn_chain);
680 obd_disconnect(sbi->ll_dt_exp);
681 sbi->ll_dt_exp = NULL;
682 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
684 obd_zombie_barrier();
686 lprocfs_unregister_mountpoint(sbi);
688 obd_disconnect(sbi->ll_md_exp);
689 sbi->ll_md_exp = NULL;
694 void ll_kill_super(struct super_block *sb)
696 struct ll_sb_info *sbi;
701 if (!(sb->s_flags & MS_ACTIVE))
705 /* we need restore s_dev from changed for clustred NFS before put_super
706 * because new kernels have cached s_dev and change sb->s_dev in
707 * put_super not affected real removing devices */
709 sb->s_dev = sbi->ll_sdev_orig;
713 char *ll_read_opt(const char *opt, char *data)
719 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
720 if (strncmp(opt, data, strlen(opt)))
722 if ((value = strchr(data, '=')) == NULL)
726 OBD_ALLOC(retval, strlen(value) + 1);
728 CERROR("out of memory!\n");
732 memcpy(retval, value, strlen(value)+1);
733 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
737 static inline int ll_set_opt(const char *opt, char *data, int fl)
739 if (strncmp(opt, data, strlen(opt)) != 0)
745 /* non-client-specific mount options are parsed in lmd_parse */
746 static int ll_options(char *options, int *flags)
749 char *s1 = options, *s2;
755 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
758 CDEBUG(D_SUPER, "next opt=%s\n", s1);
759 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
764 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
769 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
774 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
779 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
784 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
789 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
790 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
792 /* Ignore deprecated mount option. The client will
793 * always try to mount with ACL support, whether this
794 * is used depends on whether server supports it. */
795 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
796 "mount option 'acl'.\n");
799 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
801 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
802 "mount option 'noacl'.\n");
806 #warning "{no}acl options have been deprecated since 1.8, please remove them"
808 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
813 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
818 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
824 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
829 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
834 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
839 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
844 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
849 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
854 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
859 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
864 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
869 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
874 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
880 s2 = strchr(s1, ',');
888 void ll_lli_init(struct ll_inode_info *lli)
890 lli->lli_inode_magic = LLI_INODE_MAGIC;
892 lli->lli_ioepoch = 0;
893 lli->lli_maxbytes = MAX_LFS_FILESIZE;
894 spin_lock_init(&lli->lli_lock);
895 lli->lli_posix_acl = NULL;
896 lli->lli_remote_perms = NULL;
897 mutex_init(&lli->lli_rmtperm_mutex);
898 /* Do not set lli_fid, it has been initialized already. */
899 fid_zero(&lli->lli_pfid);
900 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
901 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
902 cfs_atomic_set(&lli->lli_open_count, 0);
903 lli->lli_mds_capa = NULL;
904 lli->lli_rmtperm_time = 0;
905 lli->lli_pending_och = NULL;
906 lli->lli_mds_read_och = NULL;
907 lli->lli_mds_write_och = NULL;
908 lli->lli_mds_exec_och = NULL;
909 lli->lli_open_fd_read_count = 0;
910 lli->lli_open_fd_write_count = 0;
911 lli->lli_open_fd_exec_count = 0;
912 mutex_init(&lli->lli_och_mutex);
913 spin_lock_init(&lli->lli_agl_lock);
914 lli->lli_has_smd = false;
915 lli->lli_layout_gen = LL_LAYOUT_GEN_ZERO;
916 lli->lli_clob = NULL;
918 LASSERT(lli->lli_vfs_inode.i_mode != 0);
919 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
920 mutex_init(&lli->lli_readdir_mutex);
921 lli->lli_opendir_key = NULL;
923 lli->lli_def_acl = NULL;
924 spin_lock_init(&lli->lli_sa_lock);
925 lli->lli_opendir_pid = 0;
927 sema_init(&lli->lli_size_sem, 1);
928 lli->lli_size_sem_owner = NULL;
929 lli->lli_symlink_name = NULL;
930 init_rwsem(&lli->lli_trunc_sem);
931 mutex_init(&lli->lli_write_mutex);
932 init_rwsem(&lli->lli_glimpse_sem);
933 lli->lli_glimpse_time = 0;
934 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
935 lli->lli_agl_index = 0;
936 lli->lli_async_rc = 0;
937 lli->lli_volatile = false;
939 mutex_init(&lli->lli_layout_mutex);
942 static inline int ll_bdi_register(struct backing_dev_info *bdi)
944 #ifdef HAVE_BDI_REGISTER
945 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
948 bdi->name = "lustre";
950 return bdi_register(bdi, NULL, "lustre-%d",
951 atomic_inc_return(&ll_bdi_num));
957 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
959 struct lustre_profile *lprof = NULL;
960 struct lustre_sb_info *lsi = s2lsi(sb);
961 struct ll_sb_info *sbi;
962 char *dt = NULL, *md = NULL;
963 char *profilenm = get_profile_name(sb);
964 struct config_llog_instance *cfg;
965 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
966 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
970 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
978 /* client additional sb info */
979 lsi->lsi_llsbi = sbi = ll_init_sbi();
981 cfs_module_put(THIS_MODULE);
986 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
990 err = bdi_init(&lsi->lsi_bdi);
993 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
994 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
995 err = ll_bdi_register(&lsi->lsi_bdi);
1000 sb->s_bdi = &lsi->lsi_bdi;
1003 /* Generate a string unique to this super, in case some joker tries
1004 to mount the same fs at two mount points.
1005 Use the address of the super itself.*/
1006 cfg->cfg_instance = sb;
1007 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1008 cfg->cfg_callback = class_config_llog_handler;
1009 /* set up client obds */
1010 err = lustre_process_log(sb, profilenm, cfg);
1012 CERROR("Unable to process log: %d\n", err);
1013 GOTO(out_free, err);
1016 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1017 lprof = class_get_profile(profilenm);
1018 if (lprof == NULL) {
1019 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1020 " read from the MGS. Does that filesystem "
1021 "exist?\n", profilenm);
1022 GOTO(out_free, err = -EINVAL);
1024 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1025 lprof->lp_md, lprof->lp_dt);
1027 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1029 GOTO(out_free, err = -ENOMEM);
1030 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1032 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1034 GOTO(out_free, err = -ENOMEM);
1035 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1037 /* connections, registrations, sb setup */
1038 err = client_common_fill_super(sb, md, dt, mnt);
1042 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1044 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1047 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1048 LCONSOLE_WARN("Mounted %s\n", profilenm);
1052 } /* ll_fill_super */
1055 void lu_context_keys_dump(void);
1057 void ll_put_super(struct super_block *sb)
1059 struct config_llog_instance cfg;
1060 struct obd_device *obd;
1061 struct lustre_sb_info *lsi = s2lsi(sb);
1062 struct ll_sb_info *sbi = ll_s2sbi(sb);
1063 char *profilenm = get_profile_name(sb);
1064 int force = 1, next;
1067 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1069 ll_print_capa_stat(sbi);
1071 cfg.cfg_instance = sb;
1072 lustre_end_log(sb, profilenm, &cfg);
1074 if (sbi->ll_md_exp) {
1075 obd = class_exp2obd(sbi->ll_md_exp);
1077 force = obd->obd_force;
1080 /* We need to set force before the lov_disconnect in
1081 lustre_common_put_super, since l_d cleans up osc's as well. */
1084 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1086 obd->obd_force = force;
1091 /* Only if client_common_fill_super succeeded */
1092 client_common_put_super(sb);
1096 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1097 class_manual_cleanup(obd);
1100 if (sbi->ll_flags & LL_SBI_VERBOSE)
1101 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1104 class_del_profile(profilenm);
1106 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1107 bdi_destroy(&lsi->lsi_bdi);
1108 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1112 lsi->lsi_llsbi = NULL;
1114 lustre_common_put_super(sb);
1116 cl_env_cache_purge(~0);
1118 cfs_module_put(THIS_MODULE);
1121 } /* client_put_super */
1123 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1125 struct inode *inode = NULL;
1126 /* NOTE: we depend on atomic igrab() -bzzz */
1127 lock_res_and_lock(lock);
1128 if (lock->l_ast_data) {
1129 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1130 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1131 inode = igrab(lock->l_ast_data);
1133 inode = lock->l_ast_data;
1134 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1135 D_WARNING, lock, "l_ast_data %p is "
1136 "bogus: magic %08x", lock->l_ast_data,
1137 lli->lli_inode_magic);
1141 unlock_res_and_lock(lock);
1145 static int null_if_equal(struct ldlm_lock *lock, void *data)
1147 if (data == lock->l_ast_data) {
1148 lock->l_ast_data = NULL;
1150 if (lock->l_req_mode != lock->l_granted_mode)
1151 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1154 return LDLM_ITER_CONTINUE;
1157 void ll_clear_inode(struct inode *inode)
1159 struct ll_inode_info *lli = ll_i2info(inode);
1160 struct ll_sb_info *sbi = ll_i2sbi(inode);
1163 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1164 inode->i_generation, inode);
1166 if (S_ISDIR(inode->i_mode)) {
1167 /* these should have been cleared in ll_file_release */
1168 LASSERT(lli->lli_opendir_key == NULL);
1169 LASSERT(lli->lli_sai == NULL);
1170 LASSERT(lli->lli_opendir_pid == 0);
1173 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1174 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1175 null_if_equal, inode);
1177 LASSERT(!lli->lli_open_fd_write_count);
1178 LASSERT(!lli->lli_open_fd_read_count);
1179 LASSERT(!lli->lli_open_fd_exec_count);
1181 if (lli->lli_mds_write_och)
1182 ll_md_real_close(inode, FMODE_WRITE);
1183 if (lli->lli_mds_exec_och)
1184 ll_md_real_close(inode, FMODE_EXEC);
1185 if (lli->lli_mds_read_och)
1186 ll_md_real_close(inode, FMODE_READ);
1188 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1189 OBD_FREE(lli->lli_symlink_name,
1190 strlen(lli->lli_symlink_name) + 1);
1191 lli->lli_symlink_name = NULL;
1194 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1195 LASSERT(lli->lli_posix_acl == NULL);
1196 if (lli->lli_remote_perms) {
1197 free_rmtperm_hash(lli->lli_remote_perms);
1198 lli->lli_remote_perms = NULL;
1201 #ifdef CONFIG_FS_POSIX_ACL
1202 else if (lli->lli_posix_acl) {
1203 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1204 LASSERT(lli->lli_remote_perms == NULL);
1205 posix_acl_release(lli->lli_posix_acl);
1206 lli->lli_posix_acl = NULL;
1209 lli->lli_inode_magic = LLI_INODE_DEAD;
1211 ll_clear_inode_capas(inode);
1212 if (!S_ISDIR(inode->i_mode))
1213 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1216 * XXX This has to be done before lsm is freed below, because
1217 * cl_object still uses inode lsm.
1219 cl_inode_fini(inode);
1220 lli->lli_has_smd = false;
1225 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1226 struct md_open_data **mod)
1228 struct lustre_md md;
1229 struct inode *inode = dentry->d_inode;
1230 struct ll_sb_info *sbi = ll_i2sbi(inode);
1231 struct ptlrpc_request *request = NULL;
1235 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1236 LUSTRE_OPC_ANY, NULL);
1237 if (IS_ERR(op_data))
1238 RETURN(PTR_ERR(op_data));
1240 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1243 ptlrpc_req_finished(request);
1244 if (rc == -ENOENT) {
1246 /* Unlinked special device node? Or just a race?
1247 * Pretend we done everything. */
1248 if (!S_ISREG(inode->i_mode) &&
1249 !S_ISDIR(inode->i_mode)) {
1250 ia_valid = op_data->op_attr.ia_valid;
1251 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1252 rc = simple_setattr(dentry, &op_data->op_attr);
1253 op_data->op_attr.ia_valid = ia_valid;
1255 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1256 CERROR("md_setattr fails: rc = %d\n", rc);
1261 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1262 sbi->ll_md_exp, &md);
1264 ptlrpc_req_finished(request);
1268 /* We want to adjust timestamps.
1269 * If there is at least some data in file, we cleared ATTR_SIZE
1270 * to avoid update size, otherwise it is important to do.(SOM case)
1272 ia_valid = op_data->op_attr.ia_valid;
1273 /* Since we set ATTR_*_SET flags above, and already done permission
1274 * check, So don't let inode_change_ok() check it again. */
1275 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1276 rc = simple_setattr(dentry, &op_data->op_attr);
1277 op_data->op_attr.ia_valid = ia_valid;
1279 /* Extract epoch data if obtained. */
1280 op_data->op_handle = md.body->handle;
1281 op_data->op_ioepoch = md.body->ioepoch;
1283 ll_update_inode(inode, &md);
1284 ptlrpc_req_finished(request);
1289 /* Close IO epoch and send Size-on-MDS attribute update. */
1290 static int ll_setattr_done_writing(struct inode *inode,
1291 struct md_op_data *op_data,
1292 struct md_open_data *mod)
1294 struct ll_inode_info *lli = ll_i2info(inode);
1298 LASSERT(op_data != NULL);
1299 if (!S_ISREG(inode->i_mode))
1302 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1303 op_data->op_ioepoch, PFID(&lli->lli_fid));
1305 op_data->op_flags = MF_EPOCH_CLOSE;
1306 ll_done_writing_attr(inode, op_data);
1307 ll_pack_inode2opdata(inode, op_data, NULL);
1309 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1310 if (rc == -EAGAIN) {
1311 /* MDS has instructed us to obtain Size-on-MDS attribute
1312 * from OSTs and send setattr to back to MDS. */
1313 rc = ll_som_update(inode, op_data);
1315 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1321 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1323 struct obd_capa *capa;
1326 if (attr->ia_valid & ATTR_SIZE)
1327 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1329 capa = ll_mdscapa_get(inode);
1331 rc = cl_setattr_ost(inode, attr, capa);
1333 if (attr->ia_valid & ATTR_SIZE)
1334 ll_truncate_free_capa(capa);
1341 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1343 * inode_newsize_ok - may this inode be truncated to a given size
1344 * @inode: the inode to be truncated
1345 * @offset: the new size to assign to the inode
1346 * @Returns: 0 on success, -ve errno on failure
1348 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1349 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1350 * when necessary. Caller must not proceed with inode size change if failure is
1351 * returned. @inode must be a file (not directory), with appropriate
1352 * permissions to allow truncate (inode_newsize_ok does NOT check these
1355 * inode_newsize_ok must be called with i_mutex held.
1357 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1359 if (inode->i_size < offset) {
1360 unsigned long limit;
1362 limit = rlimit(RLIMIT_FSIZE);
1363 if (limit != RLIM_INFINITY && offset > limit)
1365 if (offset > inode->i_sb->s_maxbytes)
1369 * truncation of in-use swapfiles is disallowed - it would
1370 * cause subsequent swapout to scribble on the now-freed
1373 if (IS_SWAPFILE(inode))
1379 send_sig(SIGXFSZ, current, 0);
1385 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1386 * object(s) determine the file size and mtime. Otherwise, the MDS will
1387 * keep these values until such a time that objects are allocated for it.
1388 * We do the MDS operations first, as it is checking permissions for us.
1389 * We don't to the MDS RPC if there is nothing that we want to store there,
1390 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1391 * going to do an RPC anyways.
1393 * If we are doing a truncate, we will send the mtime and ctime updates
1394 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1395 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1398 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1400 struct lov_stripe_md *lsm;
1401 struct inode *inode = dentry->d_inode;
1402 struct ll_inode_info *lli = ll_i2info(inode);
1403 struct md_op_data *op_data = NULL;
1404 struct md_open_data *mod = NULL;
1405 int ia_valid = attr->ia_valid;
1406 int rc = 0, rc1 = 0;
1409 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1410 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1411 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1414 if (ia_valid & ATTR_SIZE) {
1415 /* Check new size against VFS/VM file size limit and rlimit */
1416 rc = inode_newsize_ok(inode, attr->ia_size);
1420 /* The maximum Lustre file size is variable, based on the
1421 * OST maximum object size and number of stripes. This
1422 * needs another check in addition to the VFS check above. */
1423 if (attr->ia_size > ll_file_maxbytes(inode)) {
1424 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1425 PFID(&lli->lli_fid), attr->ia_size,
1426 ll_file_maxbytes(inode));
1430 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1433 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1434 if (ia_valid & TIMES_SET_FLAGS) {
1435 if (cfs_curproc_fsuid() != inode->i_uid &&
1436 !cfs_capable(CFS_CAP_FOWNER))
1440 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1441 if (attr->ia_valid & ATTR_CTIME) {
1442 attr->ia_ctime = CFS_CURRENT_TIME;
1443 attr->ia_valid |= ATTR_CTIME_SET;
1445 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1446 attr->ia_atime = CFS_CURRENT_TIME;
1447 attr->ia_valid |= ATTR_ATIME_SET;
1449 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1450 attr->ia_mtime = CFS_CURRENT_TIME;
1451 attr->ia_valid |= ATTR_MTIME_SET;
1454 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1455 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1456 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1457 cfs_time_current_sec());
1459 /* If we are changing file size, file content is modified, flag it. */
1460 if (attr->ia_valid & ATTR_SIZE) {
1461 spin_lock(&lli->lli_lock);
1462 lli->lli_flags |= LLIF_DATA_MODIFIED;
1463 spin_unlock(&lli->lli_lock);
1466 /* We always do an MDS RPC, even if we're only changing the size;
1467 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1469 OBD_ALLOC_PTR(op_data);
1470 if (op_data == NULL)
1473 if (!S_ISDIR(inode->i_mode)) {
1474 if (ia_valid & ATTR_SIZE)
1475 inode_dio_write_done(inode);
1476 mutex_unlock(&inode->i_mutex);
1477 down_write(&lli->lli_trunc_sem);
1480 /* We need a steady stripe configuration for setattr to avoid
1482 lsm = ccc_inode_lsm_get(inode);
1484 /* NB: ATTR_SIZE will only be set after this point if the size
1485 * resides on the MDS, ie, this file has no objects. */
1487 attr->ia_valid &= ~ATTR_SIZE;
1488 /* can't call ll_setattr_ost() while holding a refcount of lsm */
1489 ccc_inode_lsm_put(inode, lsm);
1491 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1493 /* Open epoch for truncate. */
1494 if (exp_connect_som(ll_i2mdexp(inode)) &&
1495 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1496 op_data->op_flags = MF_EPOCH_OPEN;
1498 rc = ll_md_setattr(dentry, op_data, &mod);
1502 /* RPC to MDT is sent, cancel data modification flag */
1503 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1504 spin_lock(&lli->lli_lock);
1505 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1506 spin_unlock(&lli->lli_lock);
1509 ll_ioepoch_open(lli, op_data->op_ioepoch);
1510 if (!S_ISREG(inode->i_mode))
1513 if (ia_valid & ATTR_SIZE)
1514 attr->ia_valid |= ATTR_SIZE;
1515 if (ia_valid & (ATTR_SIZE |
1516 ATTR_ATIME | ATTR_ATIME_SET |
1517 ATTR_MTIME | ATTR_MTIME_SET))
1518 /* For truncate and utimes sending attributes to OSTs, setting
1519 * mtime/atime to the past will be performed under PW [0:EOF]
1520 * extent lock (new_size:EOF for truncate). It may seem
1521 * excessive to send mtime/atime updates to OSTs when not
1522 * setting times to past, but it is necessary due to possible
1523 * time de-synchronization between MDT inode and OST objects */
1524 rc = ll_setattr_ost(inode, attr);
1528 if (op_data->op_ioepoch) {
1529 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1533 ll_finish_md_op_data(op_data);
1535 if (!S_ISDIR(inode->i_mode)) {
1536 up_write(&lli->lli_trunc_sem);
1537 mutex_lock(&inode->i_mutex);
1538 if (ia_valid & ATTR_SIZE)
1539 inode_dio_wait(inode);
1542 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1543 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1548 int ll_setattr(struct dentry *de, struct iattr *attr)
1550 int mode = de->d_inode->i_mode;
1552 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1553 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1554 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1556 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1557 (ATTR_SIZE|ATTR_MODE)) &&
1558 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1559 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1560 !(attr->ia_mode & S_ISGID))))
1561 attr->ia_valid |= ATTR_FORCE;
1563 if ((mode & S_ISUID) &&
1564 !(attr->ia_mode & S_ISUID) &&
1565 !(attr->ia_valid & ATTR_KILL_SUID))
1566 attr->ia_valid |= ATTR_KILL_SUID;
1568 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1569 !(attr->ia_mode & S_ISGID) &&
1570 !(attr->ia_valid & ATTR_KILL_SGID))
1571 attr->ia_valid |= ATTR_KILL_SGID;
1573 return ll_setattr_raw(de, attr);
1576 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1577 __u64 max_age, __u32 flags)
1579 struct ll_sb_info *sbi = ll_s2sbi(sb);
1580 struct obd_statfs obd_osfs;
1584 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1586 CERROR("md_statfs fails: rc = %d\n", rc);
1590 osfs->os_type = sb->s_magic;
1592 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1593 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1595 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1596 flags |= OBD_STATFS_NODELAY;
1598 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1600 CERROR("obd_statfs fails: rc = %d\n", rc);
1604 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1605 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1608 osfs->os_bsize = obd_osfs.os_bsize;
1609 osfs->os_blocks = obd_osfs.os_blocks;
1610 osfs->os_bfree = obd_osfs.os_bfree;
1611 osfs->os_bavail = obd_osfs.os_bavail;
1613 /* If we don't have as many objects free on the OST as inodes
1614 * on the MDS, we reduce the total number of inodes to
1615 * compensate, so that the "inodes in use" number is correct.
1617 if (obd_osfs.os_ffree < osfs->os_ffree) {
1618 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1620 osfs->os_ffree = obd_osfs.os_ffree;
1625 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1627 struct super_block *sb = de->d_sb;
1628 struct obd_statfs osfs;
1631 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1632 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1634 /* Some amount of caching on the client is allowed */
1635 rc = ll_statfs_internal(sb, &osfs,
1636 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1641 statfs_unpack(sfs, &osfs);
1643 /* We need to downshift for all 32-bit kernels, because we can't
1644 * tell if the kernel is being called via sys_statfs64() or not.
1645 * Stop before overflowing f_bsize - in which case it is better
1646 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1647 if (sizeof(long) < 8) {
1648 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1651 osfs.os_blocks >>= 1;
1652 osfs.os_bfree >>= 1;
1653 osfs.os_bavail >>= 1;
1657 sfs->f_blocks = osfs.os_blocks;
1658 sfs->f_bfree = osfs.os_bfree;
1659 sfs->f_bavail = osfs.os_bavail;
1664 void ll_inode_size_lock(struct inode *inode)
1666 struct ll_inode_info *lli;
1668 LASSERT(!S_ISDIR(inode->i_mode));
1670 lli = ll_i2info(inode);
1671 LASSERT(lli->lli_size_sem_owner != current);
1672 down(&lli->lli_size_sem);
1673 LASSERT(lli->lli_size_sem_owner == NULL);
1674 lli->lli_size_sem_owner = current;
1677 void ll_inode_size_unlock(struct inode *inode)
1679 struct ll_inode_info *lli;
1681 lli = ll_i2info(inode);
1682 LASSERT(lli->lli_size_sem_owner == current);
1683 lli->lli_size_sem_owner = NULL;
1684 up(&lli->lli_size_sem);
1687 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1689 struct ll_inode_info *lli = ll_i2info(inode);
1690 struct mdt_body *body = md->body;
1691 struct lov_stripe_md *lsm = md->lsm;
1692 struct ll_sb_info *sbi = ll_i2sbi(inode);
1694 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1696 lli->lli_maxbytes = lsm->lsm_maxbytes;
1697 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1698 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1701 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1702 if (body->valid & OBD_MD_FLRMTPERM)
1703 ll_update_remote_perm(inode, md->remote_perm);
1705 #ifdef CONFIG_FS_POSIX_ACL
1706 else if (body->valid & OBD_MD_FLACL) {
1707 spin_lock(&lli->lli_lock);
1708 if (lli->lli_posix_acl)
1709 posix_acl_release(lli->lli_posix_acl);
1710 lli->lli_posix_acl = md->posix_acl;
1711 spin_unlock(&lli->lli_lock);
1714 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1715 inode->i_generation = cl_fid_build_gen(&body->fid1);
1717 if (body->valid & OBD_MD_FLATIME) {
1718 if (body->atime > LTIME_S(inode->i_atime))
1719 LTIME_S(inode->i_atime) = body->atime;
1720 lli->lli_lvb.lvb_atime = body->atime;
1722 if (body->valid & OBD_MD_FLMTIME) {
1723 if (body->mtime > LTIME_S(inode->i_mtime)) {
1724 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1725 "to "LPU64"\n", inode->i_ino,
1726 LTIME_S(inode->i_mtime), body->mtime);
1727 LTIME_S(inode->i_mtime) = body->mtime;
1729 lli->lli_lvb.lvb_mtime = body->mtime;
1731 if (body->valid & OBD_MD_FLCTIME) {
1732 if (body->ctime > LTIME_S(inode->i_ctime))
1733 LTIME_S(inode->i_ctime) = body->ctime;
1734 lli->lli_lvb.lvb_ctime = body->ctime;
1736 if (body->valid & OBD_MD_FLMODE)
1737 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1738 if (body->valid & OBD_MD_FLTYPE)
1739 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1740 LASSERT(inode->i_mode != 0);
1741 if (S_ISREG(inode->i_mode)) {
1742 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1744 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1746 if (body->valid & OBD_MD_FLUID)
1747 inode->i_uid = body->uid;
1748 if (body->valid & OBD_MD_FLGID)
1749 inode->i_gid = body->gid;
1750 if (body->valid & OBD_MD_FLFLAGS)
1751 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1752 if (body->valid & OBD_MD_FLNLINK)
1753 set_nlink(inode, body->nlink);
1754 if (body->valid & OBD_MD_FLRDEV)
1755 inode->i_rdev = old_decode_dev(body->rdev);
1757 if (body->valid & OBD_MD_FLID) {
1758 /* FID shouldn't be changed! */
1759 if (fid_is_sane(&lli->lli_fid)) {
1760 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1761 "Trying to change FID "DFID
1762 " to the "DFID", inode %lu/%u(%p)\n",
1763 PFID(&lli->lli_fid), PFID(&body->fid1),
1764 inode->i_ino, inode->i_generation, inode);
1766 lli->lli_fid = body->fid1;
1769 LASSERT(fid_seq(&lli->lli_fid) != 0);
1771 if (body->valid & OBD_MD_FLSIZE) {
1772 if (exp_connect_som(ll_i2mdexp(inode)) &&
1773 S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1774 struct lustre_handle lockh;
1777 /* As it is possible a blocking ast has been processed
1778 * by this time, we need to check there is an UPDATE
1779 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1781 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1782 &lockh, LDLM_FL_CBPENDING);
1784 if (lli->lli_flags & (LLIF_DONE_WRITING |
1785 LLIF_EPOCH_PENDING |
1787 CERROR("ino %lu flags %u still has "
1788 "size authority! do not trust "
1789 "the size got from MDS\n",
1790 inode->i_ino, lli->lli_flags);
1792 /* Use old size assignment to avoid
1793 * deadlock bz14138 & bz14326 */
1794 i_size_write(inode, body->size);
1795 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1797 ldlm_lock_decref(&lockh, mode);
1800 /* Use old size assignment to avoid
1801 * deadlock bz14138 & bz14326 */
1802 i_size_write(inode, body->size);
1804 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1805 inode->i_ino, (unsigned long long)body->size);
1808 if (body->valid & OBD_MD_FLBLOCKS)
1809 inode->i_blocks = body->blocks;
1812 if (body->valid & OBD_MD_FLMDSCAPA) {
1813 LASSERT(md->mds_capa);
1814 ll_add_capa(inode, md->mds_capa);
1816 if (body->valid & OBD_MD_FLOSSCAPA) {
1817 LASSERT(md->oss_capa);
1818 ll_add_capa(inode, md->oss_capa);
1822 void ll_read_inode2(struct inode *inode, void *opaque)
1824 struct lustre_md *md = opaque;
1825 struct ll_inode_info *lli = ll_i2info(inode);
1828 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1829 PFID(&lli->lli_fid), inode);
1831 LASSERT(!lli->lli_has_smd);
1833 /* Core attributes from the MDS first. This is a new inode, and
1834 * the VFS doesn't zero times in the core inode so we have to do
1835 * it ourselves. They will be overwritten by either MDS or OST
1836 * attributes - we just need to make sure they aren't newer. */
1837 LTIME_S(inode->i_mtime) = 0;
1838 LTIME_S(inode->i_atime) = 0;
1839 LTIME_S(inode->i_ctime) = 0;
1841 ll_update_inode(inode, md);
1843 /* OIDEBUG(inode); */
1845 /* initializing backing dev info. */
1846 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1849 if (S_ISREG(inode->i_mode)) {
1850 struct ll_sb_info *sbi = ll_i2sbi(inode);
1851 inode->i_op = &ll_file_inode_operations;
1852 inode->i_fop = sbi->ll_fop;
1853 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1855 } else if (S_ISDIR(inode->i_mode)) {
1856 inode->i_op = &ll_dir_inode_operations;
1857 inode->i_fop = &ll_dir_operations;
1859 } else if (S_ISLNK(inode->i_mode)) {
1860 inode->i_op = &ll_fast_symlink_inode_operations;
1863 inode->i_op = &ll_special_inode_operations;
1865 init_special_inode(inode, inode->i_mode,
1872 void ll_delete_inode(struct inode *inode)
1874 struct cl_inode_info *lli = cl_i2info(inode);
1877 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1878 /* discard all dirty pages before truncating them, required by
1879 * osc_extent implementation at LU-1030. */
1880 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1882 truncate_inode_pages(&inode->i_data, 0);
1884 /* Workaround for LU-118 */
1885 if (inode->i_data.nrpages) {
1886 TREE_READ_LOCK_IRQ(&inode->i_data);
1887 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1888 LASSERTF(inode->i_data.nrpages == 0,
1889 "inode=%lu/%u(%p) nrpages=%lu, see "
1890 "http://jira.whamcloud.com/browse/LU-118\n",
1891 inode->i_ino, inode->i_generation, inode,
1892 inode->i_data.nrpages);
1894 /* Workaround end */
1896 #ifdef HAVE_SBOPS_EVICT_INODE
1897 ll_clear_inode(inode);
1904 int ll_iocontrol(struct inode *inode, struct file *file,
1905 unsigned int cmd, unsigned long arg)
1907 struct ll_sb_info *sbi = ll_i2sbi(inode);
1908 struct ptlrpc_request *req = NULL;
1913 case FSFILT_IOC_GETFLAGS: {
1914 struct mdt_body *body;
1915 struct md_op_data *op_data;
1917 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1918 0, 0, LUSTRE_OPC_ANY,
1920 if (IS_ERR(op_data))
1921 RETURN(PTR_ERR(op_data));
1923 op_data->op_valid = OBD_MD_FLFLAGS;
1924 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1925 ll_finish_md_op_data(op_data);
1927 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1931 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1933 flags = body->flags;
1935 ptlrpc_req_finished(req);
1937 RETURN(put_user(flags, (int *)arg));
1939 case FSFILT_IOC_SETFLAGS: {
1940 struct lov_stripe_md *lsm;
1941 struct obd_info oinfo = { { { 0 } } };
1942 struct md_op_data *op_data;
1944 if (get_user(flags, (int *)arg))
1947 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1948 LUSTRE_OPC_ANY, NULL);
1949 if (IS_ERR(op_data))
1950 RETURN(PTR_ERR(op_data));
1952 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1953 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1954 rc = md_setattr(sbi->ll_md_exp, op_data,
1955 NULL, 0, NULL, 0, &req, NULL);
1956 ll_finish_md_op_data(op_data);
1957 ptlrpc_req_finished(req);
1961 inode->i_flags = ll_ext_to_inode_flags(flags);
1963 lsm = ccc_inode_lsm_get(inode);
1967 OBDO_ALLOC(oinfo.oi_oa);
1969 ccc_inode_lsm_put(inode, lsm);
1973 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1974 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1975 oinfo.oi_oa->o_flags = flags;
1976 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1978 oinfo.oi_capa = ll_mdscapa_get(inode);
1979 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1980 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1981 capa_put(oinfo.oi_capa);
1982 OBDO_FREE(oinfo.oi_oa);
1983 ccc_inode_lsm_put(inode, lsm);
1985 if (rc && rc != -EPERM && rc != -EACCES)
1986 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1997 int ll_flush_ctx(struct inode *inode)
1999 struct ll_sb_info *sbi = ll_i2sbi(inode);
2001 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2003 obd_set_info_async(NULL, sbi->ll_md_exp,
2004 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2006 obd_set_info_async(NULL, sbi->ll_dt_exp,
2007 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2012 /* umount -f client means force down, don't save state */
2013 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2014 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2016 struct super_block *sb = vfsmnt->mnt_sb;
2018 void ll_umount_begin(struct super_block *sb)
2021 struct ll_sb_info *sbi = ll_s2sbi(sb);
2022 struct obd_device *obd;
2023 struct obd_ioctl_data *ioc_data;
2026 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2027 if (!(flags & MNT_FORCE)) {
2033 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2034 sb->s_count, atomic_read(&sb->s_active));
2036 obd = class_exp2obd(sbi->ll_md_exp);
2038 CERROR("Invalid MDC connection handle "LPX64"\n",
2039 sbi->ll_md_exp->exp_handle.h_cookie);
2045 obd = class_exp2obd(sbi->ll_dt_exp);
2047 CERROR("Invalid LOV connection handle "LPX64"\n",
2048 sbi->ll_dt_exp->exp_handle.h_cookie);
2054 OBD_ALLOC_PTR(ioc_data);
2056 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2057 sizeof ioc_data, ioc_data, NULL);
2059 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2060 sizeof ioc_data, ioc_data, NULL);
2062 OBD_FREE_PTR(ioc_data);
2066 /* Really, we'd like to wait until there are no requests outstanding,
2067 * and then continue. For now, we just invalidate the requests,
2068 * schedule() and sleep one second if needed, and hope.
2071 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2072 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2073 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2074 cfs_time_seconds(1));
2075 if (atomic_read(&vfsmnt->mnt_count) > 2)
2076 LCONSOLE_WARN("Mount still busy with %d refs! You "
2077 "may try to umount it a bit later\n",
2078 atomic_read(&vfsmnt->mnt_count));
2085 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2087 struct ll_sb_info *sbi = ll_s2sbi(sb);
2088 char *profilenm = get_profile_name(sb);
2092 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2093 read_only = *flags & MS_RDONLY;
2094 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2095 sizeof(KEY_READ_ONLY),
2096 KEY_READ_ONLY, sizeof(read_only),
2099 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2100 profilenm, read_only ?
2101 "read-only" : "read-write", err);
2106 sb->s_flags |= MS_RDONLY;
2108 sb->s_flags &= ~MS_RDONLY;
2110 if (sbi->ll_flags & LL_SBI_VERBOSE)
2111 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2112 read_only ? "read-only" : "read-write");
2117 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2118 struct super_block *sb, struct lookup_intent *it)
2120 struct ll_sb_info *sbi = NULL;
2121 struct lustre_md md;
2125 LASSERT(*inode || sb);
2126 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2127 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2128 sbi->ll_md_exp, &md);
2133 ll_update_inode(*inode, &md);
2135 LASSERT(sb != NULL);
2138 * At this point server returns to client's same fid as client
2139 * generated for creating. So using ->fid1 is okay here.
2141 LASSERT(fid_is_sane(&md.body->fid1));
2143 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2144 if (*inode == NULL || IS_ERR(*inode)) {
2145 #ifdef CONFIG_FS_POSIX_ACL
2147 posix_acl_release(md.posix_acl);
2148 md.posix_acl = NULL;
2151 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2153 CERROR("new_inode -fatal: rc %d\n", rc);
2158 /* Handling piggyback layout lock.
2159 * Layout lock can be piggybacked by getattr and open request.
2160 * The lsm can be applied to inode only if it comes with a layout lock
2161 * otherwise correct layout may be overwritten, for example:
2162 * 1. proc1: mdt returns a lsm but not granting layout
2163 * 2. layout was changed by another client
2164 * 3. proc2: refresh layout and layout lock granted
2165 * 4. proc1: to apply a stale layout */
2166 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2167 struct lustre_handle lockh;
2168 struct ldlm_lock *lock;
2170 lockh.cookie = it->d.lustre.it_lock_handle;
2171 lock = ldlm_handle2lock(&lockh);
2172 LASSERT(lock != NULL);
2173 if (ldlm_has_layout(lock)) {
2174 struct cl_object_conf conf;
2176 memset(&conf, 0, sizeof(conf));
2177 conf.coc_opc = OBJECT_CONF_SET;
2178 conf.coc_inode = *inode;
2179 conf.coc_lock = lock;
2180 conf.u.coc_md = &md;
2181 (void)ll_layout_conf(*inode, &conf);
2183 LDLM_LOCK_PUT(lock);
2188 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2189 md_free_lustre_md(sbi->ll_md_exp, &md);
2193 int ll_obd_statfs(struct inode *inode, void *arg)
2195 struct ll_sb_info *sbi = NULL;
2196 struct obd_export *exp;
2198 struct obd_ioctl_data *data = NULL;
2203 if (!inode || !(sbi = ll_i2sbi(inode)))
2204 GOTO(out_statfs, rc = -EINVAL);
2206 rc = obd_ioctl_getdata(&buf, &len, arg);
2208 GOTO(out_statfs, rc);
2211 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2212 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2213 GOTO(out_statfs, rc = -EINVAL);
2215 if (data->ioc_inllen1 != sizeof(__u32) ||
2216 data->ioc_inllen2 != sizeof(__u32) ||
2217 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2218 data->ioc_plen2 != sizeof(struct obd_uuid))
2219 GOTO(out_statfs, rc = -EINVAL);
2221 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2222 if (type & LL_STATFS_LMV)
2223 exp = sbi->ll_md_exp;
2224 else if (type & LL_STATFS_LOV)
2225 exp = sbi->ll_dt_exp;
2227 GOTO(out_statfs, rc = -ENODEV);
2229 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2230 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2232 GOTO(out_statfs, rc);
2235 obd_ioctl_freedata(buf, len);
2239 int ll_process_config(struct lustre_cfg *lcfg)
2243 struct lprocfs_static_vars lvars;
2247 lprocfs_llite_init_vars(&lvars);
2249 /* The instance name contains the sb: lustre-client-aacfe000 */
2250 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2251 if (!ptr || !*(++ptr))
2253 if (sscanf(ptr, "%lx", &x) != 1)
2256 /* This better be a real Lustre superblock! */
2257 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2259 /* Note we have not called client_common_fill_super yet, so
2260 proc fns must be able to handle that! */
2261 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2268 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2269 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2270 struct inode *i1, struct inode *i2,
2271 const char *name, int namelen,
2272 int mode, __u32 opc, void *data)
2274 LASSERT(i1 != NULL);
2276 if (namelen > ll_i2sbi(i1)->ll_namelen)
2277 return ERR_PTR(-ENAMETOOLONG);
2279 if (op_data == NULL)
2280 OBD_ALLOC_PTR(op_data);
2282 if (op_data == NULL)
2283 return ERR_PTR(-ENOMEM);
2285 ll_i2gids(op_data->op_suppgids, i1, i2);
2286 op_data->op_fid1 = *ll_inode2fid(i1);
2287 op_data->op_capa1 = ll_mdscapa_get(i1);
2290 op_data->op_fid2 = *ll_inode2fid(i2);
2291 op_data->op_capa2 = ll_mdscapa_get(i2);
2293 fid_zero(&op_data->op_fid2);
2294 op_data->op_capa2 = NULL;
2297 op_data->op_name = name;
2298 op_data->op_namelen = namelen;
2299 op_data->op_mode = mode;
2300 op_data->op_mod_time = cfs_time_current_sec();
2301 op_data->op_fsuid = cfs_curproc_fsuid();
2302 op_data->op_fsgid = cfs_curproc_fsgid();
2303 op_data->op_cap = cfs_curproc_cap_pack();
2304 op_data->op_bias = 0;
2305 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2306 filename_is_volatile(name, namelen, NULL))
2307 op_data->op_bias |= MDS_CREATE_VOLATILE;
2308 op_data->op_opc = opc;
2309 op_data->op_mds = 0;
2310 op_data->op_data = data;
2312 /* If the file is being opened after mknod() (normally due to NFS)
2313 * try to use the default stripe data from parent directory for
2314 * allocating OST objects. Try to pass the parent FID to MDS. */
2315 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2316 !ll_i2info(i2)->lli_has_smd) {
2317 struct ll_inode_info *lli = ll_i2info(i2);
2319 spin_lock(&lli->lli_lock);
2320 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2321 op_data->op_fid1 = lli->lli_pfid;
2322 spin_unlock(&lli->lli_lock);
2323 /** We ignore parent's capability temporary. */
2326 /* When called by ll_setattr_raw, file is i1. */
2327 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2328 op_data->op_bias |= MDS_DATA_MODIFIED;
2333 void ll_finish_md_op_data(struct md_op_data *op_data)
2335 capa_put(op_data->op_capa1);
2336 capa_put(op_data->op_capa2);
2337 OBD_FREE_PTR(op_data);
2340 #ifdef HAVE_SUPEROPS_USE_DENTRY
2341 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2343 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2346 struct ll_sb_info *sbi;
2348 #ifdef HAVE_SUPEROPS_USE_DENTRY
2349 LASSERT((seq != NULL) && (dentry != NULL));
2350 sbi = ll_s2sbi(dentry->d_sb);
2352 LASSERT((seq != NULL) && (vfs != NULL));
2353 sbi = ll_s2sbi(vfs->mnt_sb);
2356 if (sbi->ll_flags & LL_SBI_NOLCK)
2357 seq_puts(seq, ",nolock");
2359 if (sbi->ll_flags & LL_SBI_FLOCK)
2360 seq_puts(seq, ",flock");
2362 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2363 seq_puts(seq, ",localflock");
2365 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2366 seq_puts(seq, ",user_xattr");
2368 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2369 seq_puts(seq, ",lazystatfs");
2371 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2372 seq_puts(seq, ",user_fid2path");
2378 * Get obd name by cmd, and copy out to user space
2380 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2382 struct ll_sb_info *sbi = ll_i2sbi(inode);
2383 struct obd_device *obd;
2386 if (cmd == OBD_IOC_GETDTNAME)
2387 obd = class_exp2obd(sbi->ll_dt_exp);
2388 else if (cmd == OBD_IOC_GETMDNAME)
2389 obd = class_exp2obd(sbi->ll_md_exp);
2396 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2397 strlen(obd->obd_name) + 1))
2404 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2405 * fsname will be returned in this buffer; otherwise, a static buffer will be
2406 * used to store the fsname and returned to caller.
2408 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2410 static char fsname_static[MTI_NAME_MAXLEN];
2411 struct lustre_sb_info *lsi = s2lsi(sb);
2416 /* this means the caller wants to use static buffer
2417 * and it doesn't care about race. Usually this is
2418 * in error reporting path */
2419 buf = fsname_static;
2420 buflen = sizeof(fsname_static);
2423 len = strlen(lsi->lsi_lmd->lmd_profile);
2424 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2425 if (ptr && (strcmp(ptr, "-client") == 0))
2428 if (unlikely(len >= buflen))
2430 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2436 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2440 #ifdef HAVE_FS_STRUCT_USE_PATH
2444 p.mnt = current->fs->root.mnt;
2446 path = d_path(&p, buf, bufsize);
2449 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2455 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2457 char *buf, *path = NULL;
2458 struct dentry *dentry = NULL;
2459 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2461 /* this can be called inside spin lock so use GFP_ATOMIC. */
2462 buf = (char *)__get_free_page(GFP_ATOMIC);
2464 dentry = d_find_alias(page->mapping->host);
2466 path = ll_d_path(dentry, buf, PAGE_SIZE);
2469 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2470 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2471 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2472 PFID(&obj->cob_header.coh_lu.loh_fid),
2473 (path && !IS_ERR(path)) ? path : "", ioret);
2479 free_page((unsigned long)buf);