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 #ifdef HAVE_BGL_SUPPORT
98 lru_page_max = pages / 4;
100 lru_page_max = pages / 2;
103 lru_page_max = (pages / 4) * 3;
106 /* initialize lru data */
107 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
108 sbi->ll_cache.ccc_lru_max = lru_page_max;
109 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
110 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
111 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
113 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
114 SBI_DEFAULT_READAHEAD_MAX);
115 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
116 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
117 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
118 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
119 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
121 ll_generate_random_uuid(uuid);
122 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
123 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
125 spin_lock(&ll_sb_lock);
126 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
127 spin_unlock(&ll_sb_lock);
129 sbi->ll_flags |= LL_SBI_VERBOSE;
130 #ifdef ENABLE_CHECKSUM
131 sbi->ll_flags |= LL_SBI_CHECKSUM;
134 #ifdef HAVE_LRU_RESIZE_SUPPORT
135 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
138 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
139 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
141 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
145 /* metadata statahead is enabled by default */
146 sbi->ll_sa_max = LL_SA_RPC_DEF;
147 cfs_atomic_set(&sbi->ll_sa_total, 0);
148 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
149 cfs_atomic_set(&sbi->ll_agl_total, 0);
150 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
155 void ll_free_sbi(struct super_block *sb)
157 struct ll_sb_info *sbi = ll_s2sbi(sb);
161 spin_lock(&ll_sb_lock);
162 cfs_list_del(&sbi->ll_list);
163 spin_unlock(&ll_sb_lock);
164 OBD_FREE(sbi, sizeof(*sbi));
169 static struct dentry_operations ll_d_root_ops = {
170 .d_compare = ll_dcompare,
171 .d_revalidate = ll_revalidate_nd,
174 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
175 struct vfsmount *mnt)
177 struct inode *root = 0;
178 struct ll_sb_info *sbi = ll_s2sbi(sb);
179 struct obd_device *obd;
180 struct obd_capa *oc = NULL;
181 struct obd_statfs *osfs = NULL;
182 struct ptlrpc_request *request = NULL;
183 struct obd_connect_data *data = NULL;
184 struct obd_uuid *uuid;
185 struct md_op_data *op_data;
186 struct lustre_md lmd;
188 int size, err, checksum;
191 obd = class_name2obd(md);
193 CERROR("MD %s: not setup or attached\n", md);
207 if (proc_lustre_fs_root) {
208 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
211 CERROR("could not register mount in /proc/fs/lustre\n");
214 /* indicate the features supported by this client */
215 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
216 OBD_CONNECT_ATTRFID |
217 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
218 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
219 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
220 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
221 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
222 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
223 OBD_CONNECT_EINPROGRESS |
224 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
225 OBD_CONNECT_LAYOUTLOCK;
227 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
228 data->ocd_connect_flags |= OBD_CONNECT_SOM;
230 #ifdef HAVE_LRU_RESIZE_SUPPORT
231 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
232 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
234 #ifdef CONFIG_FS_POSIX_ACL
235 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
238 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
239 /* flag mdc connection as lightweight, only used for test
240 * purpose, use with care */
241 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
243 data->ocd_ibits_known = MDS_INODELOCK_FULL;
244 data->ocd_version = LUSTRE_VERSION_CODE;
246 if (sb->s_flags & MS_RDONLY)
247 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
248 if (sbi->ll_flags & LL_SBI_USER_XATTR)
249 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
251 #ifdef HAVE_MS_FLOCK_LOCK
252 /* force vfs to use lustre handler for flock() calls - bug 10743 */
253 sb->s_flags |= MS_FLOCK_LOCK;
255 #ifdef MS_HAS_NEW_AOPS
256 sb->s_flags |= MS_HAS_NEW_AOPS;
259 if (sbi->ll_flags & LL_SBI_FLOCK)
260 sbi->ll_fop = &ll_file_operations_flock;
261 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
262 sbi->ll_fop = &ll_file_operations;
264 sbi->ll_fop = &ll_file_operations_noflock;
267 data->ocd_connect_flags |= OBD_CONNECT_REAL;
268 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
269 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
271 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
273 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
275 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
276 "recovery, of which this client is not a "
277 "part. Please wait for recovery to complete,"
278 " abort, or time out.\n", md);
281 CERROR("cannot connect to %s: rc = %d\n", md, err);
285 /* For mount, we only need fs info from MDT0, and also in DNE, it
286 * can make sure the client can be mounted as long as MDT0 is
288 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
289 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
290 OBD_STATFS_FOR_MDT0);
294 /* This needs to be after statfs to ensure connect has finished.
295 * Note that "data" does NOT contain the valid connect reply.
296 * If connecting to a 1.8 server there will be no LMV device, so
297 * we can access the MDC export directly and exp_connect_flags will
298 * be non-zero, but if accessing an upgraded 2.1 server it will
299 * have the correct flags filled in.
300 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
301 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
302 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
303 valid != CLIENT_CONNECT_MDT_REQD) {
306 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
307 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
308 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
309 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
310 "feature(s) needed for correct operation "
311 "of this client (%s). Please upgrade "
312 "server or downgrade client.\n",
313 sbi->ll_md_exp->exp_obd->obd_name, buf);
314 OBD_FREE(buf, CFS_PAGE_SIZE);
315 GOTO(out_md, err = -EPROTO);
318 size = sizeof(*data);
319 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
320 KEY_CONN_DATA, &size, data, NULL);
322 CERROR("%s: Get connect data failed: rc = %d\n",
323 sbi->ll_md_exp->exp_obd->obd_name, err);
327 LASSERT(osfs->os_bsize);
328 sb->s_blocksize = osfs->os_bsize;
329 sb->s_blocksize_bits = log2(osfs->os_bsize);
330 sb->s_magic = LL_SUPER_MAGIC;
331 sb->s_maxbytes = MAX_LFS_FILESIZE;
332 sbi->ll_namelen = osfs->os_namelen;
333 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
335 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
336 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
337 LCONSOLE_INFO("Disabling user_xattr feature because "
338 "it is not supported on the server\n");
339 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
342 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
344 sb->s_flags |= MS_POSIXACL;
346 sbi->ll_flags |= LL_SBI_ACL;
348 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
350 sb->s_flags &= ~MS_POSIXACL;
352 sbi->ll_flags &= ~LL_SBI_ACL;
355 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
356 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
357 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
358 LCONSOLE_INFO("client is set as remote by default.\n");
361 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
362 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
363 LCONSOLE_INFO("client claims to be remote, but server "
364 "rejected, forced to be local.\n");
368 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
369 LCONSOLE_INFO("client enabled MDS capability!\n");
370 sbi->ll_flags |= LL_SBI_MDS_CAPA;
373 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
374 LCONSOLE_INFO("client enabled OSS capability!\n");
375 sbi->ll_flags |= LL_SBI_OSS_CAPA;
378 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
379 sbi->ll_flags |= LL_SBI_64BIT_HASH;
381 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
382 sbi->ll_md_brw_size = data->ocd_brw_size;
384 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
386 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
387 LCONSOLE_INFO("Layout lock feature supported.\n");
388 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
391 obd = class_name2obd(dt);
393 CERROR("DT %s: not setup or attached\n", dt);
394 GOTO(out_md, err = -ENODEV);
397 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
398 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
399 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
400 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
401 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
402 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
403 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
404 OBD_CONNECT_MAXBYTES |
405 OBD_CONNECT_EINPROGRESS |
406 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
407 OBD_CONNECT_LAYOUTLOCK;
409 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
410 data->ocd_connect_flags |= OBD_CONNECT_SOM;
412 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
413 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
414 * disabled by default, because it can still be enabled on the
415 * fly via /proc. As a consequence, we still need to come to an
416 * agreement on the supported algorithms at connect time */
417 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
419 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
420 data->ocd_cksum_types = OBD_CKSUM_ADLER;
422 data->ocd_cksum_types = cksum_types_supported_client();
425 #ifdef HAVE_LRU_RESIZE_SUPPORT
426 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
428 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
429 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
431 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
432 "ocd_grant: %d\n", data->ocd_connect_flags,
433 data->ocd_version, data->ocd_grant);
435 obd->obd_upcall.onu_owner = &sbi->ll_lco;
436 obd->obd_upcall.onu_upcall = cl_ocd_update;
438 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
440 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
443 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
444 "recovery, of which this client is not a "
445 "part. Please wait for recovery to "
446 "complete, abort, or time out.\n", dt);
449 CERROR("%s: Cannot connect to %s: rc = %d\n",
450 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
454 mutex_lock(&sbi->ll_lco.lco_lock);
455 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
456 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
457 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
458 mutex_unlock(&sbi->ll_lco.lco_lock);
460 fid_zero(&sbi->ll_root_fid);
461 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
463 CERROR("cannot mds_connect: rc = %d\n", err);
466 if (!fid_is_sane(&sbi->ll_root_fid)) {
467 CERROR("%s: Invalid root fid "DFID" during mount\n",
468 sbi->ll_md_exp->exp_obd->obd_name,
469 PFID(&sbi->ll_root_fid));
470 GOTO(out_dt, err = -EINVAL);
472 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
474 sb->s_op = &lustre_super_operations;
475 #if THREAD_SIZE >= 8192 /*b=17630*/
476 sb->s_export_op = &lustre_export_operations;
480 * XXX: move this to after cbd setup? */
481 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
482 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
483 valid |= OBD_MD_FLRMTPERM;
484 else if (sbi->ll_flags & LL_SBI_ACL)
485 valid |= OBD_MD_FLACL;
487 OBD_ALLOC_PTR(op_data);
489 GOTO(out_dt, err = -ENOMEM);
491 op_data->op_fid1 = sbi->ll_root_fid;
492 op_data->op_mode = 0;
493 op_data->op_capa1 = oc;
494 op_data->op_valid = valid;
496 err = md_getattr(sbi->ll_md_exp, op_data, &request);
499 OBD_FREE_PTR(op_data);
501 CERROR("%s: md_getattr failed for root: rc = %d\n",
502 sbi->ll_md_exp->exp_obd->obd_name, err);
506 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
507 sbi->ll_md_exp, &lmd);
509 CERROR("failed to understand root inode md: rc = %d\n", err);
510 ptlrpc_req_finished(request);
514 LASSERT(fid_is_sane(&sbi->ll_root_fid));
515 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
516 md_free_lustre_md(sbi->ll_md_exp, &lmd);
517 ptlrpc_req_finished(request);
519 if (root == NULL || IS_ERR(root)) {
521 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
522 #ifdef CONFIG_FS_POSIX_ACL
524 posix_acl_release(lmd.posix_acl);
525 lmd.posix_acl = NULL;
528 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
530 CERROR("lustre_lite: bad iget4 for root\n");
534 err = ll_close_thread_start(&sbi->ll_lcq);
536 CERROR("cannot start close thread: rc %d\n", err);
540 #ifdef CONFIG_FS_POSIX_ACL
541 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
542 rct_init(&sbi->ll_rct);
543 et_init(&sbi->ll_et);
547 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
548 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
549 KEY_CHECKSUM, sizeof(checksum), &checksum,
553 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
554 KEY_CACHE_SET, sizeof(sbi->ll_cache),
555 &sbi->ll_cache, NULL);
557 sb->s_root = d_make_root(root);
558 if (sb->s_root == NULL) {
559 CERROR("%s: can't make root dentry\n",
560 ll_get_fsname(sb, NULL, 0));
561 GOTO(out_root, err = -ENOMEM);
564 #ifdef HAVE_DCACHE_LOCK
565 sb->s_root->d_op = &ll_d_root_ops;
567 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
568 d_set_d_op(sb->s_root, &ll_d_root_ops);
569 sb->s_d_op = &ll_d_ops;
572 sbi->ll_sdev_orig = sb->s_dev;
574 /* We set sb->s_dev equal on all lustre clients in order to support
575 * NFS export clustering. NFSD requires that the FSID be the same
577 /* s_dev is also used in lt_compare() to compare two fs, but that is
578 * only a node-local comparison. */
579 uuid = obd_get_uuid(sbi->ll_md_exp);
581 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
593 obd_disconnect(sbi->ll_dt_exp);
594 sbi->ll_dt_exp = NULL;
595 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
596 obd_zombie_barrier();
598 obd_disconnect(sbi->ll_md_exp);
599 sbi->ll_md_exp = NULL;
605 lprocfs_unregister_mountpoint(sbi);
609 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
613 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
615 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
616 KEY_MAX_EASIZE, &size, lmmsize, NULL);
618 CERROR("Get max mdsize error rc %d \n", rc);
623 void ll_dump_inode(struct inode *inode)
625 struct ll_d_hlist_node *tmp;
626 int dentry_count = 0;
628 LASSERT(inode != NULL);
630 ll_d_hlist_for_each(tmp, &inode->i_dentry)
633 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
634 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
635 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
638 void lustre_dump_dentry(struct dentry *dentry, int recur)
640 struct list_head *tmp;
643 LASSERT(dentry != NULL);
645 list_for_each(tmp, &dentry->d_subdirs)
648 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
649 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
650 dentry->d_name.len, dentry->d_name.name,
651 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
652 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
653 dentry->d_flags, dentry->d_fsdata, subdirs);
654 if (dentry->d_inode != NULL)
655 ll_dump_inode(dentry->d_inode);
660 list_for_each(tmp, &dentry->d_subdirs) {
661 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
662 lustre_dump_dentry(d, recur - 1);
666 void client_common_put_super(struct super_block *sb)
668 struct ll_sb_info *sbi = ll_s2sbi(sb);
671 #ifdef CONFIG_FS_POSIX_ACL
672 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
673 et_fini(&sbi->ll_et);
674 rct_fini(&sbi->ll_rct);
678 ll_close_thread_shutdown(sbi->ll_lcq);
682 cfs_list_del(&sbi->ll_conn_chain);
684 obd_disconnect(sbi->ll_dt_exp);
685 sbi->ll_dt_exp = NULL;
686 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
688 obd_zombie_barrier();
690 lprocfs_unregister_mountpoint(sbi);
692 obd_disconnect(sbi->ll_md_exp);
693 sbi->ll_md_exp = NULL;
698 void ll_kill_super(struct super_block *sb)
700 struct ll_sb_info *sbi;
705 if (!(sb->s_flags & MS_ACTIVE))
709 /* we need restore s_dev from changed for clustred NFS before put_super
710 * because new kernels have cached s_dev and change sb->s_dev in
711 * put_super not affected real removing devices */
713 sb->s_dev = sbi->ll_sdev_orig;
717 char *ll_read_opt(const char *opt, char *data)
723 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
724 if (strncmp(opt, data, strlen(opt)))
726 if ((value = strchr(data, '=')) == NULL)
730 OBD_ALLOC(retval, strlen(value) + 1);
732 CERROR("out of memory!\n");
736 memcpy(retval, value, strlen(value)+1);
737 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
741 static inline int ll_set_opt(const char *opt, char *data, int fl)
743 if (strncmp(opt, data, strlen(opt)) != 0)
749 /* non-client-specific mount options are parsed in lmd_parse */
750 static int ll_options(char *options, int *flags)
753 char *s1 = options, *s2;
759 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
762 CDEBUG(D_SUPER, "next opt=%s\n", s1);
763 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
768 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
773 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
778 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
783 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
788 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
793 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
794 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
796 /* Ignore deprecated mount option. The client will
797 * always try to mount with ACL support, whether this
798 * is used depends on whether server supports it. */
799 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
800 "mount option 'acl'.\n");
803 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
805 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
806 "mount option 'noacl'.\n");
810 #warning "{no}acl options have been deprecated since 1.8, please remove them"
812 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
817 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
822 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
828 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
833 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
838 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
843 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
848 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
853 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
858 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
863 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
868 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
873 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
878 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
884 s2 = strchr(s1, ',');
892 void ll_lli_init(struct ll_inode_info *lli)
894 lli->lli_inode_magic = LLI_INODE_MAGIC;
896 lli->lli_ioepoch = 0;
897 lli->lli_maxbytes = MAX_LFS_FILESIZE;
898 spin_lock_init(&lli->lli_lock);
899 lli->lli_posix_acl = NULL;
900 lli->lli_remote_perms = NULL;
901 mutex_init(&lli->lli_rmtperm_mutex);
902 /* Do not set lli_fid, it has been initialized already. */
903 fid_zero(&lli->lli_pfid);
904 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
905 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
906 cfs_atomic_set(&lli->lli_open_count, 0);
907 lli->lli_mds_capa = NULL;
908 lli->lli_rmtperm_time = 0;
909 lli->lli_pending_och = NULL;
910 lli->lli_mds_read_och = NULL;
911 lli->lli_mds_write_och = NULL;
912 lli->lli_mds_exec_och = NULL;
913 lli->lli_open_fd_read_count = 0;
914 lli->lli_open_fd_write_count = 0;
915 lli->lli_open_fd_exec_count = 0;
916 mutex_init(&lli->lli_och_mutex);
917 spin_lock_init(&lli->lli_agl_lock);
918 lli->lli_has_smd = false;
919 lli->lli_layout_gen = LL_LAYOUT_GEN_ZERO;
920 lli->lli_clob = NULL;
922 LASSERT(lli->lli_vfs_inode.i_mode != 0);
923 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
924 mutex_init(&lli->lli_readdir_mutex);
925 lli->lli_opendir_key = NULL;
927 lli->lli_def_acl = NULL;
928 spin_lock_init(&lli->lli_sa_lock);
929 lli->lli_opendir_pid = 0;
931 sema_init(&lli->lli_size_sem, 1);
932 lli->lli_size_sem_owner = NULL;
933 lli->lli_symlink_name = NULL;
934 init_rwsem(&lli->lli_trunc_sem);
935 mutex_init(&lli->lli_write_mutex);
936 init_rwsem(&lli->lli_glimpse_sem);
937 lli->lli_glimpse_time = 0;
938 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
939 lli->lli_agl_index = 0;
940 lli->lli_async_rc = 0;
942 mutex_init(&lli->lli_layout_mutex);
945 static inline int ll_bdi_register(struct backing_dev_info *bdi)
947 #ifdef HAVE_BDI_REGISTER
948 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
951 bdi->name = "lustre";
953 return bdi_register(bdi, NULL, "lustre-%d",
954 atomic_inc_return(&ll_bdi_num));
960 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
962 struct lustre_profile *lprof = NULL;
963 struct lustre_sb_info *lsi = s2lsi(sb);
964 struct ll_sb_info *sbi;
965 char *dt = NULL, *md = NULL;
966 char *profilenm = get_profile_name(sb);
967 struct config_llog_instance *cfg;
968 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
969 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
973 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
981 /* client additional sb info */
982 lsi->lsi_llsbi = sbi = ll_init_sbi();
984 cfs_module_put(THIS_MODULE);
989 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
993 err = bdi_init(&lsi->lsi_bdi);
996 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
997 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
998 err = ll_bdi_register(&lsi->lsi_bdi);
1000 GOTO(out_free, err);
1003 sb->s_bdi = &lsi->lsi_bdi;
1006 /* Generate a string unique to this super, in case some joker tries
1007 to mount the same fs at two mount points.
1008 Use the address of the super itself.*/
1009 cfg->cfg_instance = sb;
1010 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1011 cfg->cfg_callback = class_config_llog_handler;
1012 /* set up client obds */
1013 err = lustre_process_log(sb, profilenm, cfg);
1015 CERROR("Unable to process log: %d\n", err);
1016 GOTO(out_free, err);
1019 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1020 lprof = class_get_profile(profilenm);
1021 if (lprof == NULL) {
1022 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1023 " read from the MGS. Does that filesystem "
1024 "exist?\n", profilenm);
1025 GOTO(out_free, err = -EINVAL);
1027 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1028 lprof->lp_md, lprof->lp_dt);
1030 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1032 GOTO(out_free, err = -ENOMEM);
1033 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1035 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1037 GOTO(out_free, err = -ENOMEM);
1038 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1040 /* connections, registrations, sb setup */
1041 err = client_common_fill_super(sb, md, dt, mnt);
1045 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1047 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1050 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1051 LCONSOLE_WARN("Mounted %s\n", profilenm);
1055 } /* ll_fill_super */
1058 void lu_context_keys_dump(void);
1060 void ll_put_super(struct super_block *sb)
1062 struct config_llog_instance cfg;
1063 struct obd_device *obd;
1064 struct lustre_sb_info *lsi = s2lsi(sb);
1065 struct ll_sb_info *sbi = ll_s2sbi(sb);
1066 char *profilenm = get_profile_name(sb);
1067 int force = 1, next;
1070 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1072 ll_print_capa_stat(sbi);
1074 cfg.cfg_instance = sb;
1075 lustre_end_log(sb, profilenm, &cfg);
1077 if (sbi->ll_md_exp) {
1078 obd = class_exp2obd(sbi->ll_md_exp);
1080 force = obd->obd_force;
1083 /* We need to set force before the lov_disconnect in
1084 lustre_common_put_super, since l_d cleans up osc's as well. */
1087 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1089 obd->obd_force = force;
1094 /* Only if client_common_fill_super succeeded */
1095 client_common_put_super(sb);
1099 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1100 class_manual_cleanup(obd);
1103 if (sbi->ll_flags & LL_SBI_VERBOSE)
1104 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1107 class_del_profile(profilenm);
1109 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1110 bdi_destroy(&lsi->lsi_bdi);
1111 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1115 lsi->lsi_llsbi = NULL;
1117 lustre_common_put_super(sb);
1119 cl_env_cache_purge(~0);
1121 cfs_module_put(THIS_MODULE);
1124 } /* client_put_super */
1126 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1128 struct inode *inode = NULL;
1129 /* NOTE: we depend on atomic igrab() -bzzz */
1130 lock_res_and_lock(lock);
1131 if (lock->l_ast_data) {
1132 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1133 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1134 inode = igrab(lock->l_ast_data);
1136 inode = lock->l_ast_data;
1137 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1138 D_WARNING, lock, "l_ast_data %p is "
1139 "bogus: magic %08x", lock->l_ast_data,
1140 lli->lli_inode_magic);
1144 unlock_res_and_lock(lock);
1148 static int null_if_equal(struct ldlm_lock *lock, void *data)
1150 if (data == lock->l_ast_data) {
1151 lock->l_ast_data = NULL;
1153 if (lock->l_req_mode != lock->l_granted_mode)
1154 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1157 return LDLM_ITER_CONTINUE;
1160 void ll_clear_inode(struct inode *inode)
1162 struct ll_inode_info *lli = ll_i2info(inode);
1163 struct ll_sb_info *sbi = ll_i2sbi(inode);
1166 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1167 inode->i_generation, inode);
1169 if (S_ISDIR(inode->i_mode)) {
1170 /* these should have been cleared in ll_file_release */
1171 LASSERT(lli->lli_opendir_key == NULL);
1172 LASSERT(lli->lli_sai == NULL);
1173 LASSERT(lli->lli_opendir_pid == 0);
1176 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1177 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1178 null_if_equal, inode);
1180 LASSERT(!lli->lli_open_fd_write_count);
1181 LASSERT(!lli->lli_open_fd_read_count);
1182 LASSERT(!lli->lli_open_fd_exec_count);
1184 if (lli->lli_mds_write_och)
1185 ll_md_real_close(inode, FMODE_WRITE);
1186 if (lli->lli_mds_exec_och)
1187 ll_md_real_close(inode, FMODE_EXEC);
1188 if (lli->lli_mds_read_och)
1189 ll_md_real_close(inode, FMODE_READ);
1191 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1192 OBD_FREE(lli->lli_symlink_name,
1193 strlen(lli->lli_symlink_name) + 1);
1194 lli->lli_symlink_name = NULL;
1197 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1198 LASSERT(lli->lli_posix_acl == NULL);
1199 if (lli->lli_remote_perms) {
1200 free_rmtperm_hash(lli->lli_remote_perms);
1201 lli->lli_remote_perms = NULL;
1204 #ifdef CONFIG_FS_POSIX_ACL
1205 else if (lli->lli_posix_acl) {
1206 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1207 LASSERT(lli->lli_remote_perms == NULL);
1208 posix_acl_release(lli->lli_posix_acl);
1209 lli->lli_posix_acl = NULL;
1212 lli->lli_inode_magic = LLI_INODE_DEAD;
1214 ll_clear_inode_capas(inode);
1215 if (!S_ISDIR(inode->i_mode))
1216 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1219 * XXX This has to be done before lsm is freed below, because
1220 * cl_object still uses inode lsm.
1222 cl_inode_fini(inode);
1223 lli->lli_has_smd = false;
1228 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1229 struct md_open_data **mod)
1231 struct lustre_md md;
1232 struct inode *inode = dentry->d_inode;
1233 struct ll_sb_info *sbi = ll_i2sbi(inode);
1234 struct ptlrpc_request *request = NULL;
1238 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1239 LUSTRE_OPC_ANY, NULL);
1240 if (IS_ERR(op_data))
1241 RETURN(PTR_ERR(op_data));
1243 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1246 ptlrpc_req_finished(request);
1247 if (rc == -ENOENT) {
1249 /* Unlinked special device node? Or just a race?
1250 * Pretend we done everything. */
1251 if (!S_ISREG(inode->i_mode) &&
1252 !S_ISDIR(inode->i_mode)) {
1253 ia_valid = op_data->op_attr.ia_valid;
1254 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1255 rc = simple_setattr(dentry, &op_data->op_attr);
1256 op_data->op_attr.ia_valid = ia_valid;
1258 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1259 CERROR("md_setattr fails: rc = %d\n", rc);
1264 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1265 sbi->ll_md_exp, &md);
1267 ptlrpc_req_finished(request);
1271 /* We want to adjust timestamps.
1272 * If there is at least some data in file, we cleared ATTR_SIZE
1273 * to avoid update size, otherwise it is important to do.(SOM case)
1275 ia_valid = op_data->op_attr.ia_valid;
1276 /* Since we set ATTR_*_SET flags above, and already done permission
1277 * check, So don't let inode_change_ok() check it again. */
1278 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1279 rc = simple_setattr(dentry, &op_data->op_attr);
1280 op_data->op_attr.ia_valid = ia_valid;
1282 /* Extract epoch data if obtained. */
1283 op_data->op_handle = md.body->handle;
1284 op_data->op_ioepoch = md.body->ioepoch;
1286 ll_update_inode(inode, &md);
1287 ptlrpc_req_finished(request);
1292 /* Close IO epoch and send Size-on-MDS attribute update. */
1293 static int ll_setattr_done_writing(struct inode *inode,
1294 struct md_op_data *op_data,
1295 struct md_open_data *mod)
1297 struct ll_inode_info *lli = ll_i2info(inode);
1301 LASSERT(op_data != NULL);
1302 if (!S_ISREG(inode->i_mode))
1305 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1306 op_data->op_ioepoch, PFID(&lli->lli_fid));
1308 op_data->op_flags = MF_EPOCH_CLOSE;
1309 ll_done_writing_attr(inode, op_data);
1310 ll_pack_inode2opdata(inode, op_data, NULL);
1312 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1313 if (rc == -EAGAIN) {
1314 /* MDS has instructed us to obtain Size-on-MDS attribute
1315 * from OSTs and send setattr to back to MDS. */
1316 rc = ll_som_update(inode, op_data);
1318 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1324 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1326 struct obd_capa *capa;
1329 if (attr->ia_valid & ATTR_SIZE)
1330 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1332 capa = ll_mdscapa_get(inode);
1334 rc = cl_setattr_ost(inode, attr, capa);
1336 if (attr->ia_valid & ATTR_SIZE)
1337 ll_truncate_free_capa(capa);
1344 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1346 * inode_newsize_ok - may this inode be truncated to a given size
1347 * @inode: the inode to be truncated
1348 * @offset: the new size to assign to the inode
1349 * @Returns: 0 on success, -ve errno on failure
1351 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1352 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1353 * when necessary. Caller must not proceed with inode size change if failure is
1354 * returned. @inode must be a file (not directory), with appropriate
1355 * permissions to allow truncate (inode_newsize_ok does NOT check these
1358 * inode_newsize_ok must be called with i_mutex held.
1360 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1362 if (inode->i_size < offset) {
1363 unsigned long limit;
1365 limit = rlimit(RLIMIT_FSIZE);
1366 if (limit != RLIM_INFINITY && offset > limit)
1368 if (offset > inode->i_sb->s_maxbytes)
1372 * truncation of in-use swapfiles is disallowed - it would
1373 * cause subsequent swapout to scribble on the now-freed
1376 if (IS_SWAPFILE(inode))
1382 send_sig(SIGXFSZ, current, 0);
1388 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1389 * object(s) determine the file size and mtime. Otherwise, the MDS will
1390 * keep these values until such a time that objects are allocated for it.
1391 * We do the MDS operations first, as it is checking permissions for us.
1392 * We don't to the MDS RPC if there is nothing that we want to store there,
1393 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1394 * going to do an RPC anyways.
1396 * If we are doing a truncate, we will send the mtime and ctime updates
1397 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1398 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1401 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1403 struct lov_stripe_md *lsm;
1404 struct inode *inode = dentry->d_inode;
1405 struct ll_inode_info *lli = ll_i2info(inode);
1406 struct md_op_data *op_data = NULL;
1407 struct md_open_data *mod = NULL;
1408 int ia_valid = attr->ia_valid;
1409 int rc = 0, rc1 = 0;
1412 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1413 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1414 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1417 if (ia_valid & ATTR_SIZE) {
1418 /* Check new size against VFS/VM file size limit and rlimit */
1419 rc = inode_newsize_ok(inode, attr->ia_size);
1423 /* The maximum Lustre file size is variable, based on the
1424 * OST maximum object size and number of stripes. This
1425 * needs another check in addition to the VFS check above. */
1426 if (attr->ia_size > ll_file_maxbytes(inode)) {
1427 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1428 PFID(&lli->lli_fid), attr->ia_size,
1429 ll_file_maxbytes(inode));
1433 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1436 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1437 if (ia_valid & TIMES_SET_FLAGS) {
1438 if (cfs_curproc_fsuid() != inode->i_uid &&
1439 !cfs_capable(CFS_CAP_FOWNER))
1443 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1444 if (attr->ia_valid & ATTR_CTIME) {
1445 attr->ia_ctime = CFS_CURRENT_TIME;
1446 attr->ia_valid |= ATTR_CTIME_SET;
1448 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1449 attr->ia_atime = CFS_CURRENT_TIME;
1450 attr->ia_valid |= ATTR_ATIME_SET;
1452 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1453 attr->ia_mtime = CFS_CURRENT_TIME;
1454 attr->ia_valid |= ATTR_MTIME_SET;
1457 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1458 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1459 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1460 cfs_time_current_sec());
1462 /* If we are changing file size, file content is modified, flag it. */
1463 if (attr->ia_valid & ATTR_SIZE) {
1464 spin_lock(&lli->lli_lock);
1465 lli->lli_flags |= LLIF_DATA_MODIFIED;
1466 spin_unlock(&lli->lli_lock);
1469 /* We always do an MDS RPC, even if we're only changing the size;
1470 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1472 OBD_ALLOC_PTR(op_data);
1473 if (op_data == NULL)
1476 if (!S_ISDIR(inode->i_mode)) {
1477 if (ia_valid & ATTR_SIZE)
1478 inode_dio_write_done(inode);
1479 mutex_unlock(&inode->i_mutex);
1480 down_write(&lli->lli_trunc_sem);
1481 mutex_lock(&inode->i_mutex);
1482 if (ia_valid & ATTR_SIZE)
1483 inode_dio_wait(inode);
1486 /* We need a steady stripe configuration for setattr to avoid
1488 lsm = ccc_inode_lsm_get(inode);
1490 /* NB: ATTR_SIZE will only be set after this point if the size
1491 * resides on the MDS, ie, this file has no objects. */
1493 attr->ia_valid &= ~ATTR_SIZE;
1494 /* can't call ll_setattr_ost() while holding a refcount of lsm */
1495 ccc_inode_lsm_put(inode, lsm);
1497 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1499 /* Open epoch for truncate. */
1500 if (exp_connect_som(ll_i2mdexp(inode)) &&
1501 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1502 op_data->op_flags = MF_EPOCH_OPEN;
1504 rc = ll_md_setattr(dentry, op_data, &mod);
1508 /* RPC to MDT is sent, cancel data modification flag */
1509 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1510 spin_lock(&lli->lli_lock);
1511 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1512 spin_unlock(&lli->lli_lock);
1515 ll_ioepoch_open(lli, op_data->op_ioepoch);
1516 if (!S_ISREG(inode->i_mode))
1519 if (ia_valid & ATTR_SIZE)
1520 attr->ia_valid |= ATTR_SIZE;
1521 if (ia_valid & (ATTR_SIZE |
1522 ATTR_ATIME | ATTR_ATIME_SET |
1523 ATTR_MTIME | ATTR_MTIME_SET))
1524 /* For truncate and utimes sending attributes to OSTs, setting
1525 * mtime/atime to the past will be performed under PW [0:EOF]
1526 * extent lock (new_size:EOF for truncate). It may seem
1527 * excessive to send mtime/atime updates to OSTs when not
1528 * setting times to past, but it is necessary due to possible
1529 * time de-synchronization between MDT inode and OST objects */
1530 rc = ll_setattr_ost(inode, attr);
1534 if (op_data->op_ioepoch) {
1535 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1539 ll_finish_md_op_data(op_data);
1541 if (!S_ISDIR(inode->i_mode))
1542 up_write(&lli->lli_trunc_sem);
1544 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1545 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1550 int ll_setattr(struct dentry *de, struct iattr *attr)
1552 int mode = de->d_inode->i_mode;
1554 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1555 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1556 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1558 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1559 (ATTR_SIZE|ATTR_MODE)) &&
1560 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1561 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1562 !(attr->ia_mode & S_ISGID))))
1563 attr->ia_valid |= ATTR_FORCE;
1565 if ((mode & S_ISUID) &&
1566 !(attr->ia_mode & S_ISUID) &&
1567 !(attr->ia_valid & ATTR_KILL_SUID))
1568 attr->ia_valid |= ATTR_KILL_SUID;
1570 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1571 !(attr->ia_mode & S_ISGID) &&
1572 !(attr->ia_valid & ATTR_KILL_SGID))
1573 attr->ia_valid |= ATTR_KILL_SGID;
1575 return ll_setattr_raw(de, attr);
1578 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1579 __u64 max_age, __u32 flags)
1581 struct ll_sb_info *sbi = ll_s2sbi(sb);
1582 struct obd_statfs obd_osfs;
1586 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1588 CERROR("md_statfs fails: rc = %d\n", rc);
1592 osfs->os_type = sb->s_magic;
1594 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1595 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1597 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1598 flags |= OBD_STATFS_NODELAY;
1600 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1602 CERROR("obd_statfs fails: rc = %d\n", rc);
1606 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1607 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1610 osfs->os_bsize = obd_osfs.os_bsize;
1611 osfs->os_blocks = obd_osfs.os_blocks;
1612 osfs->os_bfree = obd_osfs.os_bfree;
1613 osfs->os_bavail = obd_osfs.os_bavail;
1615 /* If we don't have as many objects free on the OST as inodes
1616 * on the MDS, we reduce the total number of inodes to
1617 * compensate, so that the "inodes in use" number is correct.
1619 if (obd_osfs.os_ffree < osfs->os_ffree) {
1620 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1622 osfs->os_ffree = obd_osfs.os_ffree;
1627 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1629 struct super_block *sb = de->d_sb;
1630 struct obd_statfs osfs;
1633 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1634 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1636 /* Some amount of caching on the client is allowed */
1637 rc = ll_statfs_internal(sb, &osfs,
1638 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1643 statfs_unpack(sfs, &osfs);
1645 /* We need to downshift for all 32-bit kernels, because we can't
1646 * tell if the kernel is being called via sys_statfs64() or not.
1647 * Stop before overflowing f_bsize - in which case it is better
1648 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1649 if (sizeof(long) < 8) {
1650 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1653 osfs.os_blocks >>= 1;
1654 osfs.os_bfree >>= 1;
1655 osfs.os_bavail >>= 1;
1659 sfs->f_blocks = osfs.os_blocks;
1660 sfs->f_bfree = osfs.os_bfree;
1661 sfs->f_bavail = osfs.os_bavail;
1666 void ll_inode_size_lock(struct inode *inode)
1668 struct ll_inode_info *lli;
1670 LASSERT(!S_ISDIR(inode->i_mode));
1672 lli = ll_i2info(inode);
1673 LASSERT(lli->lli_size_sem_owner != current);
1674 down(&lli->lli_size_sem);
1675 LASSERT(lli->lli_size_sem_owner == NULL);
1676 lli->lli_size_sem_owner = current;
1679 void ll_inode_size_unlock(struct inode *inode)
1681 struct ll_inode_info *lli;
1683 lli = ll_i2info(inode);
1684 LASSERT(lli->lli_size_sem_owner == current);
1685 lli->lli_size_sem_owner = NULL;
1686 up(&lli->lli_size_sem);
1689 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1691 struct ll_inode_info *lli = ll_i2info(inode);
1692 struct mdt_body *body = md->body;
1693 struct lov_stripe_md *lsm = md->lsm;
1694 struct ll_sb_info *sbi = ll_i2sbi(inode);
1696 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1698 lli->lli_maxbytes = lsm->lsm_maxbytes;
1699 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1700 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1703 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1704 if (body->valid & OBD_MD_FLRMTPERM)
1705 ll_update_remote_perm(inode, md->remote_perm);
1707 #ifdef CONFIG_FS_POSIX_ACL
1708 else if (body->valid & OBD_MD_FLACL) {
1709 spin_lock(&lli->lli_lock);
1710 if (lli->lli_posix_acl)
1711 posix_acl_release(lli->lli_posix_acl);
1712 lli->lli_posix_acl = md->posix_acl;
1713 spin_unlock(&lli->lli_lock);
1716 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1717 inode->i_generation = cl_fid_build_gen(&body->fid1);
1719 if (body->valid & OBD_MD_FLATIME) {
1720 if (body->atime > LTIME_S(inode->i_atime))
1721 LTIME_S(inode->i_atime) = body->atime;
1722 lli->lli_lvb.lvb_atime = body->atime;
1724 if (body->valid & OBD_MD_FLMTIME) {
1725 if (body->mtime > LTIME_S(inode->i_mtime)) {
1726 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1727 "to "LPU64"\n", inode->i_ino,
1728 LTIME_S(inode->i_mtime), body->mtime);
1729 LTIME_S(inode->i_mtime) = body->mtime;
1731 lli->lli_lvb.lvb_mtime = body->mtime;
1733 if (body->valid & OBD_MD_FLCTIME) {
1734 if (body->ctime > LTIME_S(inode->i_ctime))
1735 LTIME_S(inode->i_ctime) = body->ctime;
1736 lli->lli_lvb.lvb_ctime = body->ctime;
1738 if (body->valid & OBD_MD_FLMODE)
1739 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1740 if (body->valid & OBD_MD_FLTYPE)
1741 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1742 LASSERT(inode->i_mode != 0);
1743 if (S_ISREG(inode->i_mode)) {
1744 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1746 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1748 if (body->valid & OBD_MD_FLUID)
1749 inode->i_uid = body->uid;
1750 if (body->valid & OBD_MD_FLGID)
1751 inode->i_gid = body->gid;
1752 if (body->valid & OBD_MD_FLFLAGS)
1753 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1754 if (body->valid & OBD_MD_FLNLINK)
1755 set_nlink(inode, body->nlink);
1756 if (body->valid & OBD_MD_FLRDEV)
1757 inode->i_rdev = old_decode_dev(body->rdev);
1759 if (body->valid & OBD_MD_FLID) {
1760 /* FID shouldn't be changed! */
1761 if (fid_is_sane(&lli->lli_fid)) {
1762 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1763 "Trying to change FID "DFID
1764 " to the "DFID", inode %lu/%u(%p)\n",
1765 PFID(&lli->lli_fid), PFID(&body->fid1),
1766 inode->i_ino, inode->i_generation, inode);
1768 lli->lli_fid = body->fid1;
1771 LASSERT(fid_seq(&lli->lli_fid) != 0);
1773 if (body->valid & OBD_MD_FLSIZE) {
1774 if (exp_connect_som(ll_i2mdexp(inode)) &&
1775 S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1776 struct lustre_handle lockh;
1779 /* As it is possible a blocking ast has been processed
1780 * by this time, we need to check there is an UPDATE
1781 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1783 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1784 &lockh, LDLM_FL_CBPENDING);
1786 if (lli->lli_flags & (LLIF_DONE_WRITING |
1787 LLIF_EPOCH_PENDING |
1789 CERROR("ino %lu flags %u still has "
1790 "size authority! do not trust "
1791 "the size got from MDS\n",
1792 inode->i_ino, lli->lli_flags);
1794 /* Use old size assignment to avoid
1795 * deadlock bz14138 & bz14326 */
1796 i_size_write(inode, body->size);
1797 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1799 ldlm_lock_decref(&lockh, mode);
1802 /* Use old size assignment to avoid
1803 * deadlock bz14138 & bz14326 */
1804 i_size_write(inode, body->size);
1806 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1807 inode->i_ino, (unsigned long long)body->size);
1810 if (body->valid & OBD_MD_FLBLOCKS)
1811 inode->i_blocks = body->blocks;
1814 if (body->valid & OBD_MD_FLMDSCAPA) {
1815 LASSERT(md->mds_capa);
1816 ll_add_capa(inode, md->mds_capa);
1818 if (body->valid & OBD_MD_FLOSSCAPA) {
1819 LASSERT(md->oss_capa);
1820 ll_add_capa(inode, md->oss_capa);
1824 void ll_read_inode2(struct inode *inode, void *opaque)
1826 struct lustre_md *md = opaque;
1827 struct ll_inode_info *lli = ll_i2info(inode);
1830 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1831 PFID(&lli->lli_fid), inode);
1833 LASSERT(!lli->lli_has_smd);
1835 /* Core attributes from the MDS first. This is a new inode, and
1836 * the VFS doesn't zero times in the core inode so we have to do
1837 * it ourselves. They will be overwritten by either MDS or OST
1838 * attributes - we just need to make sure they aren't newer. */
1839 LTIME_S(inode->i_mtime) = 0;
1840 LTIME_S(inode->i_atime) = 0;
1841 LTIME_S(inode->i_ctime) = 0;
1843 ll_update_inode(inode, md);
1845 /* OIDEBUG(inode); */
1847 /* initializing backing dev info. */
1848 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1851 if (S_ISREG(inode->i_mode)) {
1852 struct ll_sb_info *sbi = ll_i2sbi(inode);
1853 inode->i_op = &ll_file_inode_operations;
1854 inode->i_fop = sbi->ll_fop;
1855 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1857 } else if (S_ISDIR(inode->i_mode)) {
1858 inode->i_op = &ll_dir_inode_operations;
1859 inode->i_fop = &ll_dir_operations;
1861 } else if (S_ISLNK(inode->i_mode)) {
1862 inode->i_op = &ll_fast_symlink_inode_operations;
1865 inode->i_op = &ll_special_inode_operations;
1867 init_special_inode(inode, inode->i_mode,
1874 void ll_delete_inode(struct inode *inode)
1876 struct cl_inode_info *lli = cl_i2info(inode);
1879 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1880 /* discard all dirty pages before truncating them, required by
1881 * osc_extent implementation at LU-1030. */
1882 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1884 truncate_inode_pages(&inode->i_data, 0);
1886 /* Workaround for LU-118 */
1887 if (inode->i_data.nrpages) {
1888 TREE_READ_LOCK_IRQ(&inode->i_data);
1889 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1890 LASSERTF(inode->i_data.nrpages == 0,
1891 "inode=%lu/%u(%p) nrpages=%lu, see "
1892 "http://jira.whamcloud.com/browse/LU-118\n",
1893 inode->i_ino, inode->i_generation, inode,
1894 inode->i_data.nrpages);
1896 /* Workaround end */
1898 #ifdef HAVE_SBOPS_EVICT_INODE
1899 ll_clear_inode(inode);
1906 int ll_iocontrol(struct inode *inode, struct file *file,
1907 unsigned int cmd, unsigned long arg)
1909 struct ll_sb_info *sbi = ll_i2sbi(inode);
1910 struct ptlrpc_request *req = NULL;
1915 case FSFILT_IOC_GETFLAGS: {
1916 struct mdt_body *body;
1917 struct md_op_data *op_data;
1919 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1920 0, 0, LUSTRE_OPC_ANY,
1922 if (IS_ERR(op_data))
1923 RETURN(PTR_ERR(op_data));
1925 op_data->op_valid = OBD_MD_FLFLAGS;
1926 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1927 ll_finish_md_op_data(op_data);
1929 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1933 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1935 flags = body->flags;
1937 ptlrpc_req_finished(req);
1939 RETURN(put_user(flags, (int *)arg));
1941 case FSFILT_IOC_SETFLAGS: {
1942 struct lov_stripe_md *lsm;
1943 struct obd_info oinfo = { { { 0 } } };
1944 struct md_op_data *op_data;
1946 if (get_user(flags, (int *)arg))
1949 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1950 LUSTRE_OPC_ANY, NULL);
1951 if (IS_ERR(op_data))
1952 RETURN(PTR_ERR(op_data));
1954 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1955 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1956 rc = md_setattr(sbi->ll_md_exp, op_data,
1957 NULL, 0, NULL, 0, &req, NULL);
1958 ll_finish_md_op_data(op_data);
1959 ptlrpc_req_finished(req);
1963 inode->i_flags = ll_ext_to_inode_flags(flags);
1965 lsm = ccc_inode_lsm_get(inode);
1969 OBDO_ALLOC(oinfo.oi_oa);
1971 ccc_inode_lsm_put(inode, lsm);
1975 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1976 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1977 oinfo.oi_oa->o_flags = flags;
1978 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1980 oinfo.oi_capa = ll_mdscapa_get(inode);
1981 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1982 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1983 capa_put(oinfo.oi_capa);
1984 OBDO_FREE(oinfo.oi_oa);
1985 ccc_inode_lsm_put(inode, lsm);
1987 if (rc && rc != -EPERM && rc != -EACCES)
1988 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1999 int ll_flush_ctx(struct inode *inode)
2001 struct ll_sb_info *sbi = ll_i2sbi(inode);
2003 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2005 obd_set_info_async(NULL, sbi->ll_md_exp,
2006 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2008 obd_set_info_async(NULL, sbi->ll_dt_exp,
2009 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2014 /* umount -f client means force down, don't save state */
2015 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2016 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2018 struct super_block *sb = vfsmnt->mnt_sb;
2020 void ll_umount_begin(struct super_block *sb)
2023 struct ll_sb_info *sbi = ll_s2sbi(sb);
2024 struct obd_device *obd;
2025 struct obd_ioctl_data *ioc_data;
2028 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2029 if (!(flags & MNT_FORCE)) {
2035 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2036 sb->s_count, atomic_read(&sb->s_active));
2038 obd = class_exp2obd(sbi->ll_md_exp);
2040 CERROR("Invalid MDC connection handle "LPX64"\n",
2041 sbi->ll_md_exp->exp_handle.h_cookie);
2047 obd = class_exp2obd(sbi->ll_dt_exp);
2049 CERROR("Invalid LOV connection handle "LPX64"\n",
2050 sbi->ll_dt_exp->exp_handle.h_cookie);
2056 OBD_ALLOC_PTR(ioc_data);
2058 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2059 sizeof ioc_data, ioc_data, NULL);
2061 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2062 sizeof ioc_data, ioc_data, NULL);
2064 OBD_FREE_PTR(ioc_data);
2068 /* Really, we'd like to wait until there are no requests outstanding,
2069 * and then continue. For now, we just invalidate the requests,
2070 * schedule() and sleep one second if needed, and hope.
2073 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2074 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2075 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2076 cfs_time_seconds(1));
2077 if (atomic_read(&vfsmnt->mnt_count) > 2)
2078 LCONSOLE_WARN("Mount still busy with %d refs! You "
2079 "may try to umount it a bit later\n",
2080 atomic_read(&vfsmnt->mnt_count));
2087 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2089 struct ll_sb_info *sbi = ll_s2sbi(sb);
2090 char *profilenm = get_profile_name(sb);
2094 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2095 read_only = *flags & MS_RDONLY;
2096 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2097 sizeof(KEY_READ_ONLY),
2098 KEY_READ_ONLY, sizeof(read_only),
2101 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2102 profilenm, read_only ?
2103 "read-only" : "read-write", err);
2108 sb->s_flags |= MS_RDONLY;
2110 sb->s_flags &= ~MS_RDONLY;
2112 if (sbi->ll_flags & LL_SBI_VERBOSE)
2113 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2114 read_only ? "read-only" : "read-write");
2119 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2120 struct super_block *sb, struct lookup_intent *it)
2122 struct ll_sb_info *sbi = NULL;
2123 struct lustre_md md;
2127 LASSERT(*inode || sb);
2128 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2129 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2130 sbi->ll_md_exp, &md);
2135 ll_update_inode(*inode, &md);
2137 LASSERT(sb != NULL);
2140 * At this point server returns to client's same fid as client
2141 * generated for creating. So using ->fid1 is okay here.
2143 LASSERT(fid_is_sane(&md.body->fid1));
2145 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2146 if (*inode == NULL || IS_ERR(*inode)) {
2147 #ifdef CONFIG_FS_POSIX_ACL
2149 posix_acl_release(md.posix_acl);
2150 md.posix_acl = NULL;
2153 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2155 CERROR("new_inode -fatal: rc %d\n", rc);
2160 /* Handling piggyback layout lock.
2161 * Layout lock can be piggybacked by getattr and open request.
2162 * The lsm can be applied to inode only if it comes with a layout lock
2163 * otherwise correct layout may be overwritten, for example:
2164 * 1. proc1: mdt returns a lsm but not granting layout
2165 * 2. layout was changed by another client
2166 * 3. proc2: refresh layout and layout lock granted
2167 * 4. proc1: to apply a stale layout */
2168 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2169 struct lustre_handle lockh;
2170 struct ldlm_lock *lock;
2172 lockh.cookie = it->d.lustre.it_lock_handle;
2173 lock = ldlm_handle2lock(&lockh);
2174 LASSERT(lock != NULL);
2175 if (ldlm_has_layout(lock)) {
2176 struct cl_object_conf conf;
2178 memset(&conf, 0, sizeof(conf));
2179 conf.coc_opc = OBJECT_CONF_SET;
2180 conf.coc_inode = *inode;
2181 conf.coc_lock = lock;
2182 conf.u.coc_md = &md;
2183 (void)ll_layout_conf(*inode, &conf);
2185 LDLM_LOCK_PUT(lock);
2190 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2191 md_free_lustre_md(sbi->ll_md_exp, &md);
2195 int ll_obd_statfs(struct inode *inode, void *arg)
2197 struct ll_sb_info *sbi = NULL;
2198 struct obd_export *exp;
2200 struct obd_ioctl_data *data = NULL;
2205 if (!inode || !(sbi = ll_i2sbi(inode)))
2206 GOTO(out_statfs, rc = -EINVAL);
2208 rc = obd_ioctl_getdata(&buf, &len, arg);
2210 GOTO(out_statfs, rc);
2213 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2214 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2215 GOTO(out_statfs, rc = -EINVAL);
2217 if (data->ioc_inllen1 != sizeof(__u32) ||
2218 data->ioc_inllen2 != sizeof(__u32) ||
2219 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2220 data->ioc_plen2 != sizeof(struct obd_uuid))
2221 GOTO(out_statfs, rc = -EINVAL);
2223 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2224 if (type & LL_STATFS_LMV)
2225 exp = sbi->ll_md_exp;
2226 else if (type & LL_STATFS_LOV)
2227 exp = sbi->ll_dt_exp;
2229 GOTO(out_statfs, rc = -ENODEV);
2231 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2232 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2234 GOTO(out_statfs, rc);
2237 obd_ioctl_freedata(buf, len);
2241 int ll_process_config(struct lustre_cfg *lcfg)
2245 struct lprocfs_static_vars lvars;
2249 lprocfs_llite_init_vars(&lvars);
2251 /* The instance name contains the sb: lustre-client-aacfe000 */
2252 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2253 if (!ptr || !*(++ptr))
2255 if (sscanf(ptr, "%lx", &x) != 1)
2258 /* This better be a real Lustre superblock! */
2259 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2261 /* Note we have not called client_common_fill_super yet, so
2262 proc fns must be able to handle that! */
2263 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2270 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2271 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2272 struct inode *i1, struct inode *i2,
2273 const char *name, int namelen,
2274 int mode, __u32 opc, void *data)
2276 LASSERT(i1 != NULL);
2278 if (namelen > ll_i2sbi(i1)->ll_namelen)
2279 return ERR_PTR(-ENAMETOOLONG);
2281 if (op_data == NULL)
2282 OBD_ALLOC_PTR(op_data);
2284 if (op_data == NULL)
2285 return ERR_PTR(-ENOMEM);
2287 ll_i2gids(op_data->op_suppgids, i1, i2);
2288 op_data->op_fid1 = *ll_inode2fid(i1);
2289 op_data->op_capa1 = ll_mdscapa_get(i1);
2292 op_data->op_fid2 = *ll_inode2fid(i2);
2293 op_data->op_capa2 = ll_mdscapa_get(i2);
2295 fid_zero(&op_data->op_fid2);
2296 op_data->op_capa2 = NULL;
2299 op_data->op_name = name;
2300 op_data->op_namelen = namelen;
2301 op_data->op_mode = mode;
2302 op_data->op_mod_time = cfs_time_current_sec();
2303 op_data->op_fsuid = cfs_curproc_fsuid();
2304 op_data->op_fsgid = cfs_curproc_fsgid();
2305 op_data->op_cap = cfs_curproc_cap_pack();
2306 op_data->op_bias = 0;
2307 op_data->op_opc = opc;
2308 op_data->op_mds = 0;
2309 op_data->op_data = data;
2311 /* If the file is being opened after mknod() (normally due to NFS)
2312 * try to use the default stripe data from parent directory for
2313 * allocating OST objects. Try to pass the parent FID to MDS. */
2314 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2315 !ll_i2info(i2)->lli_has_smd) {
2316 struct ll_inode_info *lli = ll_i2info(i2);
2318 spin_lock(&lli->lli_lock);
2319 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2320 op_data->op_fid1 = lli->lli_pfid;
2321 spin_unlock(&lli->lli_lock);
2322 /** We ignore parent's capability temporary. */
2325 /* When called by ll_setattr_raw, file is i1. */
2326 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2327 op_data->op_bias |= MDS_DATA_MODIFIED;
2332 void ll_finish_md_op_data(struct md_op_data *op_data)
2334 capa_put(op_data->op_capa1);
2335 capa_put(op_data->op_capa2);
2336 OBD_FREE_PTR(op_data);
2339 #ifdef HAVE_SUPEROPS_USE_DENTRY
2340 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2342 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2345 struct ll_sb_info *sbi;
2347 #ifdef HAVE_SUPEROPS_USE_DENTRY
2348 LASSERT((seq != NULL) && (dentry != NULL));
2349 sbi = ll_s2sbi(dentry->d_sb);
2351 LASSERT((seq != NULL) && (vfs != NULL));
2352 sbi = ll_s2sbi(vfs->mnt_sb);
2355 if (sbi->ll_flags & LL_SBI_NOLCK)
2356 seq_puts(seq, ",nolock");
2358 if (sbi->ll_flags & LL_SBI_FLOCK)
2359 seq_puts(seq, ",flock");
2361 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2362 seq_puts(seq, ",localflock");
2364 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2365 seq_puts(seq, ",user_xattr");
2367 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2368 seq_puts(seq, ",lazystatfs");
2370 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2371 seq_puts(seq, ",user_fid2path");
2377 * Get obd name by cmd, and copy out to user space
2379 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2381 struct ll_sb_info *sbi = ll_i2sbi(inode);
2382 struct obd_device *obd;
2385 if (cmd == OBD_IOC_GETDTNAME)
2386 obd = class_exp2obd(sbi->ll_dt_exp);
2387 else if (cmd == OBD_IOC_GETMDNAME)
2388 obd = class_exp2obd(sbi->ll_md_exp);
2395 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2396 strlen(obd->obd_name) + 1))
2403 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2404 * fsname will be returned in this buffer; otherwise, a static buffer will be
2405 * used to store the fsname and returned to caller.
2407 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2409 static char fsname_static[MTI_NAME_MAXLEN];
2410 struct lustre_sb_info *lsi = s2lsi(sb);
2415 /* this means the caller wants to use static buffer
2416 * and it doesn't care about race. Usually this is
2417 * in error reporting path */
2418 buf = fsname_static;
2419 buflen = sizeof(fsname_static);
2422 len = strlen(lsi->lsi_lmd->lmd_profile);
2423 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2424 if (ptr && (strcmp(ptr, "-client") == 0))
2427 if (unlikely(len >= buflen))
2429 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2435 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2439 #ifdef HAVE_FS_STRUCT_USE_PATH
2443 p.mnt = current->fs->root.mnt;
2445 path = d_path(&p, buf, bufsize);
2448 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2454 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2456 char *buf, *path = NULL;
2457 struct dentry *dentry = NULL;
2458 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2460 /* this can be called inside spin lock so use GFP_ATOMIC. */
2461 buf = (char *)__get_free_page(GFP_ATOMIC);
2463 dentry = d_find_alias(page->mapping->host);
2465 path = ll_d_path(dentry, buf, PAGE_SIZE);
2468 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2469 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2470 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2471 PFID(&obj->cob_header.coh_lu.loh_fid),
2472 (path && !IS_ERR(path)) ? path : "", ioret);
2478 free_page((unsigned long)buf);