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 | OBD_CONNECT_PINGLESS;
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 = MD_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 sbi->ll_md_exp->exp_connect_data = *data;
283 /* For mount, we only need fs info from MDT0, and also in DNE, it
284 * can make sure the client can be mounted as long as MDT0 is
286 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
287 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
288 OBD_STATFS_FOR_MDT0);
292 /* This needs to be after statfs to ensure connect has finished.
293 * Note that "data" does NOT contain the valid connect reply.
294 * If connecting to a 1.8 server there will be no LMV device, so
295 * we can access the MDC export directly and exp_connect_flags will
296 * be non-zero, but if accessing an upgraded 2.1 server it will
297 * have the correct flags filled in.
298 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
299 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
300 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
301 valid != CLIENT_CONNECT_MDT_REQD) {
304 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
305 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
306 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
307 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
308 "feature(s) needed for correct operation "
309 "of this client (%s). Please upgrade "
310 "server or downgrade client.\n",
311 sbi->ll_md_exp->exp_obd->obd_name, buf);
312 OBD_FREE(buf, CFS_PAGE_SIZE);
313 GOTO(out_md, err = -EPROTO);
316 size = sizeof(*data);
317 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
318 KEY_CONN_DATA, &size, data, NULL);
320 CERROR("%s: Get connect data failed: rc = %d\n",
321 sbi->ll_md_exp->exp_obd->obd_name, err);
325 LASSERT(osfs->os_bsize);
326 sb->s_blocksize = osfs->os_bsize;
327 sb->s_blocksize_bits = log2(osfs->os_bsize);
328 sb->s_magic = LL_SUPER_MAGIC;
329 sb->s_maxbytes = MAX_LFS_FILESIZE;
330 sbi->ll_namelen = osfs->os_namelen;
331 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
333 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
334 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
335 LCONSOLE_INFO("Disabling user_xattr feature because "
336 "it is not supported on the server\n");
337 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
340 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
342 sb->s_flags |= MS_POSIXACL;
344 sbi->ll_flags |= LL_SBI_ACL;
346 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
348 sb->s_flags &= ~MS_POSIXACL;
350 sbi->ll_flags &= ~LL_SBI_ACL;
353 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
354 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
355 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
356 LCONSOLE_INFO("client is set as remote by default.\n");
359 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
360 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
361 LCONSOLE_INFO("client claims to be remote, but server "
362 "rejected, forced to be local.\n");
366 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
367 LCONSOLE_INFO("client enabled MDS capability!\n");
368 sbi->ll_flags |= LL_SBI_MDS_CAPA;
371 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
372 LCONSOLE_INFO("client enabled OSS capability!\n");
373 sbi->ll_flags |= LL_SBI_OSS_CAPA;
376 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
377 sbi->ll_flags |= LL_SBI_64BIT_HASH;
379 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
380 sbi->ll_md_brw_size = data->ocd_brw_size;
382 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
384 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
385 LCONSOLE_INFO("Layout lock feature supported.\n");
386 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
389 obd = class_name2obd(dt);
391 CERROR("DT %s: not setup or attached\n", dt);
392 GOTO(out_md, err = -ENODEV);
395 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
396 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
397 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
398 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
399 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
400 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
401 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
402 OBD_CONNECT_MAXBYTES |
403 OBD_CONNECT_EINPROGRESS |
404 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
405 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
407 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
408 data->ocd_connect_flags |= OBD_CONNECT_SOM;
410 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
411 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
412 * disabled by default, because it can still be enabled on the
413 * fly via /proc. As a consequence, we still need to come to an
414 * agreement on the supported algorithms at connect time */
415 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
417 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
418 data->ocd_cksum_types = OBD_CKSUM_ADLER;
420 data->ocd_cksum_types = cksum_types_supported_client();
423 #ifdef HAVE_LRU_RESIZE_SUPPORT
424 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
426 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
427 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
429 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
430 "ocd_grant: %d\n", data->ocd_connect_flags,
431 data->ocd_version, data->ocd_grant);
433 obd->obd_upcall.onu_owner = &sbi->ll_lco;
434 obd->obd_upcall.onu_upcall = cl_ocd_update;
436 data->ocd_brw_size = DT_MAX_BRW_SIZE;
438 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
441 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
442 "recovery, of which this client is not a "
443 "part. Please wait for recovery to "
444 "complete, abort, or time out.\n", dt);
447 CERROR("%s: Cannot connect to %s: rc = %d\n",
448 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
452 sbi->ll_dt_exp->exp_connect_data = *data;
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_NONE;
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;
941 lli->lli_volatile = false;
943 mutex_init(&lli->lli_layout_mutex);
946 static inline int ll_bdi_register(struct backing_dev_info *bdi)
948 #ifdef HAVE_BDI_REGISTER
949 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
952 bdi->name = "lustre";
954 return bdi_register(bdi, NULL, "lustre-%d",
955 atomic_inc_return(&ll_bdi_num));
961 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
963 struct lustre_profile *lprof = NULL;
964 struct lustre_sb_info *lsi = s2lsi(sb);
965 struct ll_sb_info *sbi;
966 char *dt = NULL, *md = NULL;
967 char *profilenm = get_profile_name(sb);
968 struct config_llog_instance *cfg;
969 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
970 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
974 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
982 /* client additional sb info */
983 lsi->lsi_llsbi = sbi = ll_init_sbi();
985 cfs_module_put(THIS_MODULE);
990 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
994 err = bdi_init(&lsi->lsi_bdi);
997 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
998 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
999 err = ll_bdi_register(&lsi->lsi_bdi);
1001 GOTO(out_free, err);
1004 sb->s_bdi = &lsi->lsi_bdi;
1007 /* Generate a string unique to this super, in case some joker tries
1008 to mount the same fs at two mount points.
1009 Use the address of the super itself.*/
1010 cfg->cfg_instance = sb;
1011 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1012 cfg->cfg_callback = class_config_llog_handler;
1013 /* set up client obds */
1014 err = lustre_process_log(sb, profilenm, cfg);
1016 CERROR("Unable to process log: %d\n", err);
1017 GOTO(out_free, err);
1020 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1021 lprof = class_get_profile(profilenm);
1022 if (lprof == NULL) {
1023 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1024 " read from the MGS. Does that filesystem "
1025 "exist?\n", profilenm);
1026 GOTO(out_free, err = -EINVAL);
1028 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1029 lprof->lp_md, lprof->lp_dt);
1031 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1033 GOTO(out_free, err = -ENOMEM);
1034 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1036 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1038 GOTO(out_free, err = -ENOMEM);
1039 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1041 /* connections, registrations, sb setup */
1042 err = client_common_fill_super(sb, md, dt, mnt);
1046 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1048 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1051 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1052 LCONSOLE_WARN("Mounted %s\n", profilenm);
1056 } /* ll_fill_super */
1059 void lu_context_keys_dump(void);
1061 void ll_put_super(struct super_block *sb)
1063 struct config_llog_instance cfg;
1064 struct obd_device *obd;
1065 struct lustre_sb_info *lsi = s2lsi(sb);
1066 struct ll_sb_info *sbi = ll_s2sbi(sb);
1067 char *profilenm = get_profile_name(sb);
1068 int force = 1, next;
1071 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1073 ll_print_capa_stat(sbi);
1075 cfg.cfg_instance = sb;
1076 lustre_end_log(sb, profilenm, &cfg);
1078 if (sbi->ll_md_exp) {
1079 obd = class_exp2obd(sbi->ll_md_exp);
1081 force = obd->obd_force;
1084 /* We need to set force before the lov_disconnect in
1085 lustre_common_put_super, since l_d cleans up osc's as well. */
1088 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1090 obd->obd_force = force;
1095 /* Only if client_common_fill_super succeeded */
1096 client_common_put_super(sb);
1100 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1101 class_manual_cleanup(obd);
1104 if (sbi->ll_flags & LL_SBI_VERBOSE)
1105 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1108 class_del_profile(profilenm);
1110 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1111 bdi_destroy(&lsi->lsi_bdi);
1112 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1116 lsi->lsi_llsbi = NULL;
1118 lustre_common_put_super(sb);
1120 cl_env_cache_purge(~0);
1122 cfs_module_put(THIS_MODULE);
1125 } /* client_put_super */
1127 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1129 struct inode *inode = NULL;
1130 /* NOTE: we depend on atomic igrab() -bzzz */
1131 lock_res_and_lock(lock);
1132 if (lock->l_ast_data) {
1133 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1134 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1135 inode = igrab(lock->l_ast_data);
1137 inode = lock->l_ast_data;
1138 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1139 D_WARNING, lock, "l_ast_data %p is "
1140 "bogus: magic %08x", lock->l_ast_data,
1141 lli->lli_inode_magic);
1145 unlock_res_and_lock(lock);
1149 static int null_if_equal(struct ldlm_lock *lock, void *data)
1151 if (data == lock->l_ast_data) {
1152 lock->l_ast_data = NULL;
1154 if (lock->l_req_mode != lock->l_granted_mode)
1155 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1158 return LDLM_ITER_CONTINUE;
1161 void ll_clear_inode(struct inode *inode)
1163 struct ll_inode_info *lli = ll_i2info(inode);
1164 struct ll_sb_info *sbi = ll_i2sbi(inode);
1167 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1168 inode->i_generation, inode);
1170 if (S_ISDIR(inode->i_mode)) {
1171 /* these should have been cleared in ll_file_release */
1172 LASSERT(lli->lli_opendir_key == NULL);
1173 LASSERT(lli->lli_sai == NULL);
1174 LASSERT(lli->lli_opendir_pid == 0);
1177 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1178 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1179 null_if_equal, inode);
1181 LASSERT(!lli->lli_open_fd_write_count);
1182 LASSERT(!lli->lli_open_fd_read_count);
1183 LASSERT(!lli->lli_open_fd_exec_count);
1185 if (lli->lli_mds_write_och)
1186 ll_md_real_close(inode, FMODE_WRITE);
1187 if (lli->lli_mds_exec_och)
1188 ll_md_real_close(inode, FMODE_EXEC);
1189 if (lli->lli_mds_read_och)
1190 ll_md_real_close(inode, FMODE_READ);
1192 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1193 OBD_FREE(lli->lli_symlink_name,
1194 strlen(lli->lli_symlink_name) + 1);
1195 lli->lli_symlink_name = NULL;
1198 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1199 LASSERT(lli->lli_posix_acl == NULL);
1200 if (lli->lli_remote_perms) {
1201 free_rmtperm_hash(lli->lli_remote_perms);
1202 lli->lli_remote_perms = NULL;
1205 #ifdef CONFIG_FS_POSIX_ACL
1206 else if (lli->lli_posix_acl) {
1207 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1208 LASSERT(lli->lli_remote_perms == NULL);
1209 posix_acl_release(lli->lli_posix_acl);
1210 lli->lli_posix_acl = NULL;
1213 lli->lli_inode_magic = LLI_INODE_DEAD;
1215 ll_clear_inode_capas(inode);
1216 if (!S_ISDIR(inode->i_mode))
1217 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1220 * XXX This has to be done before lsm is freed below, because
1221 * cl_object still uses inode lsm.
1223 cl_inode_fini(inode);
1224 lli->lli_has_smd = false;
1229 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1230 struct md_open_data **mod)
1232 struct lustre_md md;
1233 struct inode *inode = dentry->d_inode;
1234 struct ll_sb_info *sbi = ll_i2sbi(inode);
1235 struct ptlrpc_request *request = NULL;
1239 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1240 LUSTRE_OPC_ANY, NULL);
1241 if (IS_ERR(op_data))
1242 RETURN(PTR_ERR(op_data));
1244 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1247 ptlrpc_req_finished(request);
1248 if (rc == -ENOENT) {
1250 /* Unlinked special device node? Or just a race?
1251 * Pretend we done everything. */
1252 if (!S_ISREG(inode->i_mode) &&
1253 !S_ISDIR(inode->i_mode)) {
1254 ia_valid = op_data->op_attr.ia_valid;
1255 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1256 rc = simple_setattr(dentry, &op_data->op_attr);
1257 op_data->op_attr.ia_valid = ia_valid;
1259 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1260 CERROR("md_setattr fails: rc = %d\n", rc);
1265 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1266 sbi->ll_md_exp, &md);
1268 ptlrpc_req_finished(request);
1272 /* We want to adjust timestamps.
1273 * If there is at least some data in file, we cleared ATTR_SIZE
1274 * to avoid update size, otherwise it is important to do.(SOM case)
1276 ia_valid = op_data->op_attr.ia_valid;
1277 /* Since we set ATTR_*_SET flags above, and already done permission
1278 * check, So don't let inode_change_ok() check it again. */
1279 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1280 rc = simple_setattr(dentry, &op_data->op_attr);
1281 op_data->op_attr.ia_valid = ia_valid;
1283 /* Extract epoch data if obtained. */
1284 op_data->op_handle = md.body->handle;
1285 op_data->op_ioepoch = md.body->ioepoch;
1287 ll_update_inode(inode, &md);
1288 ptlrpc_req_finished(request);
1293 /* Close IO epoch and send Size-on-MDS attribute update. */
1294 static int ll_setattr_done_writing(struct inode *inode,
1295 struct md_op_data *op_data,
1296 struct md_open_data *mod)
1298 struct ll_inode_info *lli = ll_i2info(inode);
1302 LASSERT(op_data != NULL);
1303 if (!S_ISREG(inode->i_mode))
1306 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1307 op_data->op_ioepoch, PFID(&lli->lli_fid));
1309 op_data->op_flags = MF_EPOCH_CLOSE;
1310 ll_done_writing_attr(inode, op_data);
1311 ll_pack_inode2opdata(inode, op_data, NULL);
1313 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1314 if (rc == -EAGAIN) {
1315 /* MDS has instructed us to obtain Size-on-MDS attribute
1316 * from OSTs and send setattr to back to MDS. */
1317 rc = ll_som_update(inode, op_data);
1319 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1325 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1327 struct obd_capa *capa;
1330 if (attr->ia_valid & ATTR_SIZE)
1331 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1333 capa = ll_mdscapa_get(inode);
1335 rc = cl_setattr_ost(inode, attr, capa);
1337 if (attr->ia_valid & ATTR_SIZE)
1338 ll_truncate_free_capa(capa);
1345 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1347 * inode_newsize_ok - may this inode be truncated to a given size
1348 * @inode: the inode to be truncated
1349 * @offset: the new size to assign to the inode
1350 * @Returns: 0 on success, -ve errno on failure
1352 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1353 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1354 * when necessary. Caller must not proceed with inode size change if failure is
1355 * returned. @inode must be a file (not directory), with appropriate
1356 * permissions to allow truncate (inode_newsize_ok does NOT check these
1359 * inode_newsize_ok must be called with i_mutex held.
1361 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1363 if (inode->i_size < offset) {
1364 unsigned long limit;
1366 limit = rlimit(RLIMIT_FSIZE);
1367 if (limit != RLIM_INFINITY && offset > limit)
1369 if (offset > inode->i_sb->s_maxbytes)
1373 * truncation of in-use swapfiles is disallowed - it would
1374 * cause subsequent swapout to scribble on the now-freed
1377 if (IS_SWAPFILE(inode))
1383 send_sig(SIGXFSZ, current, 0);
1389 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1390 * object(s) determine the file size and mtime. Otherwise, the MDS will
1391 * keep these values until such a time that objects are allocated for it.
1392 * We do the MDS operations first, as it is checking permissions for us.
1393 * We don't to the MDS RPC if there is nothing that we want to store there,
1394 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1395 * going to do an RPC anyways.
1397 * If we are doing a truncate, we will send the mtime and ctime updates
1398 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1399 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1402 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1404 struct lov_stripe_md *lsm;
1405 struct inode *inode = dentry->d_inode;
1406 struct ll_inode_info *lli = ll_i2info(inode);
1407 struct md_op_data *op_data = NULL;
1408 struct md_open_data *mod = NULL;
1409 int ia_valid = attr->ia_valid;
1410 int rc = 0, rc1 = 0;
1413 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1414 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1415 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1418 if (ia_valid & ATTR_SIZE) {
1419 /* Check new size against VFS/VM file size limit and rlimit */
1420 rc = inode_newsize_ok(inode, attr->ia_size);
1424 /* The maximum Lustre file size is variable, based on the
1425 * OST maximum object size and number of stripes. This
1426 * needs another check in addition to the VFS check above. */
1427 if (attr->ia_size > ll_file_maxbytes(inode)) {
1428 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1429 PFID(&lli->lli_fid), attr->ia_size,
1430 ll_file_maxbytes(inode));
1434 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1437 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1438 if (ia_valid & TIMES_SET_FLAGS) {
1439 if (cfs_curproc_fsuid() != inode->i_uid &&
1440 !cfs_capable(CFS_CAP_FOWNER))
1444 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1445 if (attr->ia_valid & ATTR_CTIME) {
1446 attr->ia_ctime = CFS_CURRENT_TIME;
1447 attr->ia_valid |= ATTR_CTIME_SET;
1449 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1450 attr->ia_atime = CFS_CURRENT_TIME;
1451 attr->ia_valid |= ATTR_ATIME_SET;
1453 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1454 attr->ia_mtime = CFS_CURRENT_TIME;
1455 attr->ia_valid |= ATTR_MTIME_SET;
1458 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1459 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1460 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1461 cfs_time_current_sec());
1463 /* If we are changing file size, file content is modified, flag it. */
1464 if (attr->ia_valid & ATTR_SIZE) {
1465 spin_lock(&lli->lli_lock);
1466 lli->lli_flags |= LLIF_DATA_MODIFIED;
1467 spin_unlock(&lli->lli_lock);
1470 /* We always do an MDS RPC, even if we're only changing the size;
1471 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1473 OBD_ALLOC_PTR(op_data);
1474 if (op_data == NULL)
1477 if (!S_ISDIR(inode->i_mode)) {
1478 if (ia_valid & ATTR_SIZE)
1479 inode_dio_write_done(inode);
1480 mutex_unlock(&inode->i_mutex);
1481 down_write(&lli->lli_trunc_sem);
1484 /* We need a steady stripe configuration for setattr to avoid
1486 lsm = ccc_inode_lsm_get(inode);
1488 /* NB: ATTR_SIZE will only be set after this point if the size
1489 * resides on the MDS, ie, this file has no objects. */
1491 attr->ia_valid &= ~ATTR_SIZE;
1492 /* can't call ll_setattr_ost() while holding a refcount of lsm */
1493 ccc_inode_lsm_put(inode, lsm);
1495 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1497 /* Open epoch for truncate. */
1498 if (exp_connect_som(ll_i2mdexp(inode)) &&
1499 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1500 op_data->op_flags = MF_EPOCH_OPEN;
1502 rc = ll_md_setattr(dentry, op_data, &mod);
1506 /* RPC to MDT is sent, cancel data modification flag */
1507 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1508 spin_lock(&lli->lli_lock);
1509 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1510 spin_unlock(&lli->lli_lock);
1513 ll_ioepoch_open(lli, op_data->op_ioepoch);
1514 if (!S_ISREG(inode->i_mode))
1517 if (ia_valid & ATTR_SIZE)
1518 attr->ia_valid |= ATTR_SIZE;
1519 if (ia_valid & (ATTR_SIZE |
1520 ATTR_ATIME | ATTR_ATIME_SET |
1521 ATTR_MTIME | ATTR_MTIME_SET))
1522 /* For truncate and utimes sending attributes to OSTs, setting
1523 * mtime/atime to the past will be performed under PW [0:EOF]
1524 * extent lock (new_size:EOF for truncate). It may seem
1525 * excessive to send mtime/atime updates to OSTs when not
1526 * setting times to past, but it is necessary due to possible
1527 * time de-synchronization between MDT inode and OST objects */
1528 rc = ll_setattr_ost(inode, attr);
1532 if (op_data->op_ioepoch) {
1533 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1537 ll_finish_md_op_data(op_data);
1539 if (!S_ISDIR(inode->i_mode)) {
1540 up_write(&lli->lli_trunc_sem);
1541 mutex_lock(&inode->i_mutex);
1542 if (ia_valid & ATTR_SIZE)
1543 inode_dio_wait(inode);
1546 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1547 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1552 int ll_setattr(struct dentry *de, struct iattr *attr)
1554 int mode = de->d_inode->i_mode;
1556 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1557 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1558 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1560 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1561 (ATTR_SIZE|ATTR_MODE)) &&
1562 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1563 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1564 !(attr->ia_mode & S_ISGID))))
1565 attr->ia_valid |= ATTR_FORCE;
1567 if ((mode & S_ISUID) &&
1568 !(attr->ia_mode & S_ISUID) &&
1569 !(attr->ia_valid & ATTR_KILL_SUID))
1570 attr->ia_valid |= ATTR_KILL_SUID;
1572 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1573 !(attr->ia_mode & S_ISGID) &&
1574 !(attr->ia_valid & ATTR_KILL_SGID))
1575 attr->ia_valid |= ATTR_KILL_SGID;
1577 return ll_setattr_raw(de, attr);
1580 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1581 __u64 max_age, __u32 flags)
1583 struct ll_sb_info *sbi = ll_s2sbi(sb);
1584 struct obd_statfs obd_osfs;
1588 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1590 CERROR("md_statfs fails: rc = %d\n", rc);
1594 osfs->os_type = sb->s_magic;
1596 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1597 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1599 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1600 flags |= OBD_STATFS_NODELAY;
1602 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1604 CERROR("obd_statfs fails: rc = %d\n", rc);
1608 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1609 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1612 osfs->os_bsize = obd_osfs.os_bsize;
1613 osfs->os_blocks = obd_osfs.os_blocks;
1614 osfs->os_bfree = obd_osfs.os_bfree;
1615 osfs->os_bavail = obd_osfs.os_bavail;
1617 /* If we don't have as many objects free on the OST as inodes
1618 * on the MDS, we reduce the total number of inodes to
1619 * compensate, so that the "inodes in use" number is correct.
1621 if (obd_osfs.os_ffree < osfs->os_ffree) {
1622 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1624 osfs->os_ffree = obd_osfs.os_ffree;
1629 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1631 struct super_block *sb = de->d_sb;
1632 struct obd_statfs osfs;
1635 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1636 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1638 /* Some amount of caching on the client is allowed */
1639 rc = ll_statfs_internal(sb, &osfs,
1640 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1645 statfs_unpack(sfs, &osfs);
1647 /* We need to downshift for all 32-bit kernels, because we can't
1648 * tell if the kernel is being called via sys_statfs64() or not.
1649 * Stop before overflowing f_bsize - in which case it is better
1650 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1651 if (sizeof(long) < 8) {
1652 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1655 osfs.os_blocks >>= 1;
1656 osfs.os_bfree >>= 1;
1657 osfs.os_bavail >>= 1;
1661 sfs->f_blocks = osfs.os_blocks;
1662 sfs->f_bfree = osfs.os_bfree;
1663 sfs->f_bavail = osfs.os_bavail;
1668 void ll_inode_size_lock(struct inode *inode)
1670 struct ll_inode_info *lli;
1672 LASSERT(!S_ISDIR(inode->i_mode));
1674 lli = ll_i2info(inode);
1675 LASSERT(lli->lli_size_sem_owner != current);
1676 down(&lli->lli_size_sem);
1677 LASSERT(lli->lli_size_sem_owner == NULL);
1678 lli->lli_size_sem_owner = current;
1681 void ll_inode_size_unlock(struct inode *inode)
1683 struct ll_inode_info *lli;
1685 lli = ll_i2info(inode);
1686 LASSERT(lli->lli_size_sem_owner == current);
1687 lli->lli_size_sem_owner = NULL;
1688 up(&lli->lli_size_sem);
1691 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1693 struct ll_inode_info *lli = ll_i2info(inode);
1694 struct mdt_body *body = md->body;
1695 struct lov_stripe_md *lsm = md->lsm;
1696 struct ll_sb_info *sbi = ll_i2sbi(inode);
1698 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1700 if (!lli->lli_has_smd &&
1701 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1702 cl_file_inode_init(inode, md);
1704 lli->lli_maxbytes = lsm->lsm_maxbytes;
1705 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1706 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1709 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1710 if (body->valid & OBD_MD_FLRMTPERM)
1711 ll_update_remote_perm(inode, md->remote_perm);
1713 #ifdef CONFIG_FS_POSIX_ACL
1714 else if (body->valid & OBD_MD_FLACL) {
1715 spin_lock(&lli->lli_lock);
1716 if (lli->lli_posix_acl)
1717 posix_acl_release(lli->lli_posix_acl);
1718 lli->lli_posix_acl = md->posix_acl;
1719 spin_unlock(&lli->lli_lock);
1722 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1723 inode->i_generation = cl_fid_build_gen(&body->fid1);
1725 if (body->valid & OBD_MD_FLATIME) {
1726 if (body->atime > LTIME_S(inode->i_atime))
1727 LTIME_S(inode->i_atime) = body->atime;
1728 lli->lli_lvb.lvb_atime = body->atime;
1730 if (body->valid & OBD_MD_FLMTIME) {
1731 if (body->mtime > LTIME_S(inode->i_mtime)) {
1732 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1733 "to "LPU64"\n", inode->i_ino,
1734 LTIME_S(inode->i_mtime), body->mtime);
1735 LTIME_S(inode->i_mtime) = body->mtime;
1737 lli->lli_lvb.lvb_mtime = body->mtime;
1739 if (body->valid & OBD_MD_FLCTIME) {
1740 if (body->ctime > LTIME_S(inode->i_ctime))
1741 LTIME_S(inode->i_ctime) = body->ctime;
1742 lli->lli_lvb.lvb_ctime = body->ctime;
1744 if (body->valid & OBD_MD_FLMODE)
1745 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1746 if (body->valid & OBD_MD_FLTYPE)
1747 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1748 LASSERT(inode->i_mode != 0);
1749 if (S_ISREG(inode->i_mode)) {
1750 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1752 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1754 if (body->valid & OBD_MD_FLUID)
1755 inode->i_uid = body->uid;
1756 if (body->valid & OBD_MD_FLGID)
1757 inode->i_gid = body->gid;
1758 if (body->valid & OBD_MD_FLFLAGS)
1759 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1760 if (body->valid & OBD_MD_FLNLINK)
1761 set_nlink(inode, body->nlink);
1762 if (body->valid & OBD_MD_FLRDEV)
1763 inode->i_rdev = old_decode_dev(body->rdev);
1765 if (body->valid & OBD_MD_FLID) {
1766 /* FID shouldn't be changed! */
1767 if (fid_is_sane(&lli->lli_fid)) {
1768 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1769 "Trying to change FID "DFID
1770 " to the "DFID", inode %lu/%u(%p)\n",
1771 PFID(&lli->lli_fid), PFID(&body->fid1),
1772 inode->i_ino, inode->i_generation, inode);
1774 lli->lli_fid = body->fid1;
1777 LASSERT(fid_seq(&lli->lli_fid) != 0);
1779 if (body->valid & OBD_MD_FLSIZE) {
1780 if (exp_connect_som(ll_i2mdexp(inode)) &&
1781 S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1782 struct lustre_handle lockh;
1785 /* As it is possible a blocking ast has been processed
1786 * by this time, we need to check there is an UPDATE
1787 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1789 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1790 &lockh, LDLM_FL_CBPENDING);
1792 if (lli->lli_flags & (LLIF_DONE_WRITING |
1793 LLIF_EPOCH_PENDING |
1795 CERROR("ino %lu flags %u still has "
1796 "size authority! do not trust "
1797 "the size got from MDS\n",
1798 inode->i_ino, lli->lli_flags);
1800 /* Use old size assignment to avoid
1801 * deadlock bz14138 & bz14326 */
1802 i_size_write(inode, body->size);
1803 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1805 ldlm_lock_decref(&lockh, mode);
1808 /* Use old size assignment to avoid
1809 * deadlock bz14138 & bz14326 */
1810 i_size_write(inode, body->size);
1812 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1813 inode->i_ino, (unsigned long long)body->size);
1816 if (body->valid & OBD_MD_FLBLOCKS)
1817 inode->i_blocks = body->blocks;
1820 if (body->valid & OBD_MD_FLMDSCAPA) {
1821 LASSERT(md->mds_capa);
1822 ll_add_capa(inode, md->mds_capa);
1824 if (body->valid & OBD_MD_FLOSSCAPA) {
1825 LASSERT(md->oss_capa);
1826 ll_add_capa(inode, md->oss_capa);
1830 void ll_read_inode2(struct inode *inode, void *opaque)
1832 struct lustre_md *md = opaque;
1833 struct ll_inode_info *lli = ll_i2info(inode);
1836 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1837 PFID(&lli->lli_fid), inode);
1839 LASSERT(!lli->lli_has_smd);
1841 /* Core attributes from the MDS first. This is a new inode, and
1842 * the VFS doesn't zero times in the core inode so we have to do
1843 * it ourselves. They will be overwritten by either MDS or OST
1844 * attributes - we just need to make sure they aren't newer. */
1845 LTIME_S(inode->i_mtime) = 0;
1846 LTIME_S(inode->i_atime) = 0;
1847 LTIME_S(inode->i_ctime) = 0;
1849 ll_update_inode(inode, md);
1851 /* OIDEBUG(inode); */
1853 /* initializing backing dev info. */
1854 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1857 if (S_ISREG(inode->i_mode)) {
1858 struct ll_sb_info *sbi = ll_i2sbi(inode);
1859 inode->i_op = &ll_file_inode_operations;
1860 inode->i_fop = sbi->ll_fop;
1861 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1863 } else if (S_ISDIR(inode->i_mode)) {
1864 inode->i_op = &ll_dir_inode_operations;
1865 inode->i_fop = &ll_dir_operations;
1867 } else if (S_ISLNK(inode->i_mode)) {
1868 inode->i_op = &ll_fast_symlink_inode_operations;
1871 inode->i_op = &ll_special_inode_operations;
1873 init_special_inode(inode, inode->i_mode,
1880 void ll_delete_inode(struct inode *inode)
1882 struct cl_inode_info *lli = cl_i2info(inode);
1885 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1886 /* discard all dirty pages before truncating them, required by
1887 * osc_extent implementation at LU-1030. */
1888 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1890 truncate_inode_pages(&inode->i_data, 0);
1892 /* Workaround for LU-118 */
1893 if (inode->i_data.nrpages) {
1894 TREE_READ_LOCK_IRQ(&inode->i_data);
1895 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1896 LASSERTF(inode->i_data.nrpages == 0,
1897 "inode=%lu/%u(%p) nrpages=%lu, see "
1898 "http://jira.whamcloud.com/browse/LU-118\n",
1899 inode->i_ino, inode->i_generation, inode,
1900 inode->i_data.nrpages);
1902 /* Workaround end */
1904 #ifdef HAVE_SBOPS_EVICT_INODE
1905 ll_clear_inode(inode);
1912 int ll_iocontrol(struct inode *inode, struct file *file,
1913 unsigned int cmd, unsigned long arg)
1915 struct ll_sb_info *sbi = ll_i2sbi(inode);
1916 struct ptlrpc_request *req = NULL;
1921 case FSFILT_IOC_GETFLAGS: {
1922 struct mdt_body *body;
1923 struct md_op_data *op_data;
1925 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1926 0, 0, LUSTRE_OPC_ANY,
1928 if (IS_ERR(op_data))
1929 RETURN(PTR_ERR(op_data));
1931 op_data->op_valid = OBD_MD_FLFLAGS;
1932 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1933 ll_finish_md_op_data(op_data);
1935 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1939 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1941 flags = body->flags;
1943 ptlrpc_req_finished(req);
1945 RETURN(put_user(flags, (int *)arg));
1947 case FSFILT_IOC_SETFLAGS: {
1948 struct lov_stripe_md *lsm;
1949 struct obd_info oinfo = { { { 0 } } };
1950 struct md_op_data *op_data;
1952 if (get_user(flags, (int *)arg))
1955 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1956 LUSTRE_OPC_ANY, NULL);
1957 if (IS_ERR(op_data))
1958 RETURN(PTR_ERR(op_data));
1960 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1961 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1962 rc = md_setattr(sbi->ll_md_exp, op_data,
1963 NULL, 0, NULL, 0, &req, NULL);
1964 ll_finish_md_op_data(op_data);
1965 ptlrpc_req_finished(req);
1969 inode->i_flags = ll_ext_to_inode_flags(flags);
1971 lsm = ccc_inode_lsm_get(inode);
1975 OBDO_ALLOC(oinfo.oi_oa);
1977 ccc_inode_lsm_put(inode, lsm);
1981 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1982 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1983 oinfo.oi_oa->o_flags = flags;
1984 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1986 oinfo.oi_capa = ll_mdscapa_get(inode);
1987 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1988 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1989 capa_put(oinfo.oi_capa);
1990 OBDO_FREE(oinfo.oi_oa);
1991 ccc_inode_lsm_put(inode, lsm);
1993 if (rc && rc != -EPERM && rc != -EACCES)
1994 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2005 int ll_flush_ctx(struct inode *inode)
2007 struct ll_sb_info *sbi = ll_i2sbi(inode);
2009 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2011 obd_set_info_async(NULL, sbi->ll_md_exp,
2012 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2014 obd_set_info_async(NULL, sbi->ll_dt_exp,
2015 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2020 /* umount -f client means force down, don't save state */
2021 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2022 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2024 struct super_block *sb = vfsmnt->mnt_sb;
2026 void ll_umount_begin(struct super_block *sb)
2029 struct ll_sb_info *sbi = ll_s2sbi(sb);
2030 struct obd_device *obd;
2031 struct obd_ioctl_data *ioc_data;
2034 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2035 if (!(flags & MNT_FORCE)) {
2041 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2042 sb->s_count, atomic_read(&sb->s_active));
2044 obd = class_exp2obd(sbi->ll_md_exp);
2046 CERROR("Invalid MDC connection handle "LPX64"\n",
2047 sbi->ll_md_exp->exp_handle.h_cookie);
2053 obd = class_exp2obd(sbi->ll_dt_exp);
2055 CERROR("Invalid LOV connection handle "LPX64"\n",
2056 sbi->ll_dt_exp->exp_handle.h_cookie);
2062 OBD_ALLOC_PTR(ioc_data);
2064 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2065 sizeof ioc_data, ioc_data, NULL);
2067 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2068 sizeof ioc_data, ioc_data, NULL);
2070 OBD_FREE_PTR(ioc_data);
2074 /* Really, we'd like to wait until there are no requests outstanding,
2075 * and then continue. For now, we just invalidate the requests,
2076 * schedule() and sleep one second if needed, and hope.
2079 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2080 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2081 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2082 cfs_time_seconds(1));
2083 if (atomic_read(&vfsmnt->mnt_count) > 2)
2084 LCONSOLE_WARN("Mount still busy with %d refs! You "
2085 "may try to umount it a bit later\n",
2086 atomic_read(&vfsmnt->mnt_count));
2093 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2095 struct ll_sb_info *sbi = ll_s2sbi(sb);
2096 char *profilenm = get_profile_name(sb);
2100 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2101 read_only = *flags & MS_RDONLY;
2102 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2103 sizeof(KEY_READ_ONLY),
2104 KEY_READ_ONLY, sizeof(read_only),
2107 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2108 profilenm, read_only ?
2109 "read-only" : "read-write", err);
2114 sb->s_flags |= MS_RDONLY;
2116 sb->s_flags &= ~MS_RDONLY;
2118 if (sbi->ll_flags & LL_SBI_VERBOSE)
2119 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2120 read_only ? "read-only" : "read-write");
2125 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2126 struct super_block *sb, struct lookup_intent *it)
2128 struct ll_sb_info *sbi = NULL;
2129 struct lustre_md md;
2133 LASSERT(*inode || sb);
2134 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2135 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2136 sbi->ll_md_exp, &md);
2141 ll_update_inode(*inode, &md);
2143 LASSERT(sb != NULL);
2146 * At this point server returns to client's same fid as client
2147 * generated for creating. So using ->fid1 is okay here.
2149 LASSERT(fid_is_sane(&md.body->fid1));
2151 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2152 if (*inode == NULL || IS_ERR(*inode)) {
2153 #ifdef CONFIG_FS_POSIX_ACL
2155 posix_acl_release(md.posix_acl);
2156 md.posix_acl = NULL;
2159 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2161 CERROR("new_inode -fatal: rc %d\n", rc);
2166 /* Handling piggyback layout lock.
2167 * Layout lock can be piggybacked by getattr and open request.
2168 * The lsm can be applied to inode only if it comes with a layout lock
2169 * otherwise correct layout may be overwritten, for example:
2170 * 1. proc1: mdt returns a lsm but not granting layout
2171 * 2. layout was changed by another client
2172 * 3. proc2: refresh layout and layout lock granted
2173 * 4. proc1: to apply a stale layout */
2174 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2175 struct lustre_handle lockh;
2176 struct ldlm_lock *lock;
2178 lockh.cookie = it->d.lustre.it_lock_handle;
2179 lock = ldlm_handle2lock(&lockh);
2180 LASSERT(lock != NULL);
2181 if (ldlm_has_layout(lock)) {
2182 struct cl_object_conf conf;
2184 memset(&conf, 0, sizeof(conf));
2185 conf.coc_opc = OBJECT_CONF_SET;
2186 conf.coc_inode = *inode;
2187 conf.coc_lock = lock;
2188 conf.u.coc_md = &md;
2189 (void)ll_layout_conf(*inode, &conf);
2191 LDLM_LOCK_PUT(lock);
2196 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2197 md_free_lustre_md(sbi->ll_md_exp, &md);
2201 int ll_obd_statfs(struct inode *inode, void *arg)
2203 struct ll_sb_info *sbi = NULL;
2204 struct obd_export *exp;
2206 struct obd_ioctl_data *data = NULL;
2211 if (!inode || !(sbi = ll_i2sbi(inode)))
2212 GOTO(out_statfs, rc = -EINVAL);
2214 rc = obd_ioctl_getdata(&buf, &len, arg);
2216 GOTO(out_statfs, rc);
2219 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2220 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2221 GOTO(out_statfs, rc = -EINVAL);
2223 if (data->ioc_inllen1 != sizeof(__u32) ||
2224 data->ioc_inllen2 != sizeof(__u32) ||
2225 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2226 data->ioc_plen2 != sizeof(struct obd_uuid))
2227 GOTO(out_statfs, rc = -EINVAL);
2229 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2230 if (type & LL_STATFS_LMV)
2231 exp = sbi->ll_md_exp;
2232 else if (type & LL_STATFS_LOV)
2233 exp = sbi->ll_dt_exp;
2235 GOTO(out_statfs, rc = -ENODEV);
2237 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2238 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2240 GOTO(out_statfs, rc);
2243 obd_ioctl_freedata(buf, len);
2247 int ll_process_config(struct lustre_cfg *lcfg)
2251 struct lprocfs_static_vars lvars;
2255 lprocfs_llite_init_vars(&lvars);
2257 /* The instance name contains the sb: lustre-client-aacfe000 */
2258 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2259 if (!ptr || !*(++ptr))
2261 if (sscanf(ptr, "%lx", &x) != 1)
2264 /* This better be a real Lustre superblock! */
2265 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2267 /* Note we have not called client_common_fill_super yet, so
2268 proc fns must be able to handle that! */
2269 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2276 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2277 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2278 struct inode *i1, struct inode *i2,
2279 const char *name, int namelen,
2280 int mode, __u32 opc, void *data)
2282 LASSERT(i1 != NULL);
2284 if (namelen > ll_i2sbi(i1)->ll_namelen)
2285 return ERR_PTR(-ENAMETOOLONG);
2287 if (op_data == NULL)
2288 OBD_ALLOC_PTR(op_data);
2290 if (op_data == NULL)
2291 return ERR_PTR(-ENOMEM);
2293 ll_i2gids(op_data->op_suppgids, i1, i2);
2294 op_data->op_fid1 = *ll_inode2fid(i1);
2295 op_data->op_capa1 = ll_mdscapa_get(i1);
2298 op_data->op_fid2 = *ll_inode2fid(i2);
2299 op_data->op_capa2 = ll_mdscapa_get(i2);
2301 fid_zero(&op_data->op_fid2);
2302 op_data->op_capa2 = NULL;
2305 op_data->op_name = name;
2306 op_data->op_namelen = namelen;
2307 op_data->op_mode = mode;
2308 op_data->op_mod_time = cfs_time_current_sec();
2309 op_data->op_fsuid = cfs_curproc_fsuid();
2310 op_data->op_fsgid = cfs_curproc_fsgid();
2311 op_data->op_cap = cfs_curproc_cap_pack();
2312 op_data->op_bias = 0;
2313 op_data->op_cli_flags = 0;
2314 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2315 filename_is_volatile(name, namelen, NULL))
2316 op_data->op_bias |= MDS_CREATE_VOLATILE;
2317 op_data->op_opc = opc;
2318 op_data->op_mds = 0;
2319 op_data->op_data = data;
2321 /* If the file is being opened after mknod() (normally due to NFS)
2322 * try to use the default stripe data from parent directory for
2323 * allocating OST objects. Try to pass the parent FID to MDS. */
2324 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2325 !ll_i2info(i2)->lli_has_smd) {
2326 struct ll_inode_info *lli = ll_i2info(i2);
2328 spin_lock(&lli->lli_lock);
2329 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2330 op_data->op_fid1 = lli->lli_pfid;
2331 spin_unlock(&lli->lli_lock);
2332 /** We ignore parent's capability temporary. */
2335 /* When called by ll_setattr_raw, file is i1. */
2336 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2337 op_data->op_bias |= MDS_DATA_MODIFIED;
2342 void ll_finish_md_op_data(struct md_op_data *op_data)
2344 capa_put(op_data->op_capa1);
2345 capa_put(op_data->op_capa2);
2346 OBD_FREE_PTR(op_data);
2349 #ifdef HAVE_SUPEROPS_USE_DENTRY
2350 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2352 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2355 struct ll_sb_info *sbi;
2357 #ifdef HAVE_SUPEROPS_USE_DENTRY
2358 LASSERT((seq != NULL) && (dentry != NULL));
2359 sbi = ll_s2sbi(dentry->d_sb);
2361 LASSERT((seq != NULL) && (vfs != NULL));
2362 sbi = ll_s2sbi(vfs->mnt_sb);
2365 if (sbi->ll_flags & LL_SBI_NOLCK)
2366 seq_puts(seq, ",nolock");
2368 if (sbi->ll_flags & LL_SBI_FLOCK)
2369 seq_puts(seq, ",flock");
2371 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2372 seq_puts(seq, ",localflock");
2374 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2375 seq_puts(seq, ",user_xattr");
2377 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2378 seq_puts(seq, ",lazystatfs");
2380 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2381 seq_puts(seq, ",user_fid2path");
2387 * Get obd name by cmd, and copy out to user space
2389 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2391 struct ll_sb_info *sbi = ll_i2sbi(inode);
2392 struct obd_device *obd;
2395 if (cmd == OBD_IOC_GETDTNAME)
2396 obd = class_exp2obd(sbi->ll_dt_exp);
2397 else if (cmd == OBD_IOC_GETMDNAME)
2398 obd = class_exp2obd(sbi->ll_md_exp);
2405 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2406 strlen(obd->obd_name) + 1))
2413 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2414 * fsname will be returned in this buffer; otherwise, a static buffer will be
2415 * used to store the fsname and returned to caller.
2417 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2419 static char fsname_static[MTI_NAME_MAXLEN];
2420 struct lustre_sb_info *lsi = s2lsi(sb);
2425 /* this means the caller wants to use static buffer
2426 * and it doesn't care about race. Usually this is
2427 * in error reporting path */
2428 buf = fsname_static;
2429 buflen = sizeof(fsname_static);
2432 len = strlen(lsi->lsi_lmd->lmd_profile);
2433 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2434 if (ptr && (strcmp(ptr, "-client") == 0))
2437 if (unlikely(len >= buflen))
2439 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2445 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2449 #ifdef HAVE_FS_STRUCT_USE_PATH
2453 p.mnt = current->fs->root.mnt;
2455 path = d_path(&p, buf, bufsize);
2458 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2464 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2466 char *buf, *path = NULL;
2467 struct dentry *dentry = NULL;
2468 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2470 /* this can be called inside spin lock so use GFP_ATOMIC. */
2471 buf = (char *)__get_free_page(GFP_ATOMIC);
2473 dentry = d_find_alias(page->mapping->host);
2475 path = ll_d_path(dentry, buf, PAGE_SIZE);
2478 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2479 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2480 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2481 PFID(&obj->cob_header.coh_lu.loh_fid),
2482 (path && !IS_ERR(path)) ? path : "", ioret);
2488 free_page((unsigned long)buf);