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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Directory code for lustre client.
38 #include <linux/pagemap.h>
40 #include <linux/version.h>
41 #include <linux/security.h>
42 #include <linux/user_namespace.h>
43 #ifdef HAVE_UIDGID_HEADER
44 # include <linux/uidgid.h>
46 #include <linux/uaccess.h>
47 #include <linux/buffer_head.h> // for wait_on_buffer
48 #include <linux/pagevec.h>
50 #define DEBUG_SUBSYSTEM S_LLITE
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <uapi/linux/lustre/lustre_ioctl.h>
55 #include <lustre_lib.h>
56 #include <lustre_dlm.h>
57 #include <lustre_fid.h>
58 #include <lustre_kernelcomm.h>
59 #include <lustre_swab.h>
61 #include "llite_internal.h"
64 * (new) readdir implementation overview.
66 * Original lustre readdir implementation cached exact copy of raw directory
67 * pages on the client. These pages were indexed in client page cache by
68 * logical offset in the directory file. This design, while very simple and
69 * intuitive had some inherent problems:
71 * . it implies that byte offset to the directory entry serves as a
72 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
73 * ext3/htree directory entries may move due to splits, and more
76 * . it is incompatible with the design of split directories for cmd3,
77 * that assumes that names are distributed across nodes based on their
78 * hash, and so readdir should be done in hash order.
80 * New readdir implementation does readdir in hash order, and uses hash of a
81 * file name as a telldir/seekdir cookie. This led to number of complications:
83 * . hash is not unique, so it cannot be used to index cached directory
84 * pages on the client (note, that it requires a whole pageful of hash
85 * collided entries to cause two pages to have identical hashes);
87 * . hash is not unique, so it cannot, strictly speaking, be used as an
88 * entry cookie. ext3/htree has the same problem and lustre implementation
89 * mimics their solution: seekdir(hash) positions directory at the first
90 * entry with the given hash.
96 * Client caches directory pages using hash of the first entry as an index. As
97 * noted above hash is not unique, so this solution doesn't work as is:
98 * special processing is needed for "page hash chains" (i.e., sequences of
99 * pages filled with entries all having the same hash value).
101 * First, such chains have to be detected. To this end, server returns to the
102 * client the hash of the first entry on the page next to one returned. When
103 * client detects that this hash is the same as hash of the first entry on the
104 * returned page, page hash collision has to be handled. Pages in the
105 * hash chain, except first one, are termed "overflow pages".
107 * Solution to index uniqueness problem is to not cache overflow
108 * pages. Instead, when page hash collision is detected, all overflow pages
109 * from emerging chain are immediately requested from the server and placed in
110 * a special data structure (struct ll_dir_chain). This data structure is used
111 * by ll_readdir() to process entries from overflow pages. When readdir
112 * invocation finishes, overflow pages are discarded. If page hash collision
113 * chain weren't completely processed, next call to readdir will again detect
114 * page hash collision, again read overflow pages in, process next portion of
115 * entries and again discard the pages. This is not as wasteful as it looks,
116 * because, given reasonable hash, page hash collisions are extremely rare.
118 * 1. directory positioning
120 * When seekdir(hash) is called, original
131 * identification of and access to overflow pages
135 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
136 * a header lu_dirpage which describes the start/end hash, and whether this
137 * page is empty (contains no dir entry) or hash collide with next page.
138 * After client receives reply, several pages will be integrated into dir page
139 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
140 * lu_dirpage for this integrated page will be adjusted. See
141 * mdc_adjust_dirpages().
144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
145 __u64 offset, struct ll_dir_chain *chain)
147 struct md_callback cb_op;
151 cb_op.md_blocking_ast = ll_md_blocking_ast;
152 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
159 void ll_release_page(struct inode *inode, struct page *page,
164 /* Always remove the page for striped dir, because the page is
165 * built from temporarily in LMV layer */
166 if (inode != NULL && S_ISDIR(inode->i_mode) &&
167 ll_i2info(inode)->lli_lsm_md != NULL) {
174 if (likely(page->mapping != NULL))
175 truncate_complete_page(page->mapping, page);
182 * return IF_* type for given lu_dirent entry.
183 * IF_* flag shld be converted to particular OS file type in
184 * platform llite module.
186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
189 struct luda_type *lt;
192 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
193 const unsigned align = sizeof(struct luda_type) - 1;
195 len = le16_to_cpu(ent->lde_namelen);
196 len = (len + align) & ~align;
197 lt = (void *)ent->lde_name + len;
198 type = IFTODT(le16_to_cpu(lt->lt_type));
204 #ifdef HAVE_DIR_CONTEXT
205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
206 struct dir_context *ctx)
209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
210 void *cookie, filldir_t filldir)
213 struct ll_sb_info *sbi = ll_i2sbi(inode);
215 bool is_api32 = ll_need_32bit_api(sbi);
216 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
218 struct ll_dir_chain chain;
223 ll_dir_chain_init(&chain);
225 page = ll_get_dir_page(inode, op_data, pos, &chain);
227 while (rc == 0 && !done) {
228 struct lu_dirpage *dp;
229 struct lu_dirent *ent;
238 hash = MDS_DIR_END_OFF;
239 dp = page_address(page);
240 for (ent = lu_dirent_start(dp); ent != NULL && !done;
241 ent = lu_dirent_next(ent)) {
248 hash = le64_to_cpu(ent->lde_hash);
249 if (hash < pos) /* Skip until we find target hash */
252 namelen = le16_to_cpu(ent->lde_namelen);
253 if (namelen == 0) /* Skip dummy record */
256 if (is_api32 && is_hash64)
260 fid_le_to_cpu(&fid, &ent->lde_fid);
261 ino = cl_fid_build_ino(&fid, is_api32);
262 type = ll_dirent_type_get(ent);
263 /* For ll_nfs_get_name_filldir(), it will try to access
264 * 'ent' through 'lde_name', so the parameter 'name'
265 * for 'filldir()' must be part of the 'ent'. */
266 #ifdef HAVE_DIR_CONTEXT
268 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
271 done = filldir(cookie, ent->lde_name, namelen, lhash,
278 ll_release_page(inode, page, false);
282 next = le64_to_cpu(dp->ldp_hash_end);
284 if (pos == MDS_DIR_END_OFF) {
286 * End of directory reached.
289 ll_release_page(inode, page, false);
292 * Normal case: continue to the next
295 ll_release_page(inode, page,
296 le32_to_cpu(dp->ldp_flags) &
299 page = ll_get_dir_page(inode, op_data, pos,
303 #ifdef HAVE_DIR_CONTEXT
308 ll_dir_chain_fini(&chain);
312 #ifdef HAVE_DIR_CONTEXT
313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
318 struct inode *inode = file_inode(filp);
319 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
320 struct ll_sb_info *sbi = ll_i2sbi(inode);
321 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
322 int api32 = ll_need_32bit_api(sbi);
323 struct md_op_data *op_data;
324 struct lu_fid pfid = { 0 };
334 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
335 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
336 inode, (unsigned long)pos, i_size_read(inode), api32);
338 if (pos == MDS_DIR_END_OFF)
344 if (unlikely(ll_i2info(inode)->lli_lsm_md != NULL)) {
346 * This is only needed for striped dir to fill ..,
347 * see lmv_read_page()
349 if (file_dentry(filp)->d_parent != NULL &&
350 file_dentry(filp)->d_parent->d_inode != NULL) {
351 __u64 ibits = MDS_INODELOCK_LOOKUP;
352 struct inode *parent =
353 file_dentry(filp)->d_parent->d_inode;
355 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
356 pfid = *ll_inode2fid(parent);
359 /* If it can not find in cache, do lookup .. on the master
361 if (fid_is_zero(&pfid)) {
362 rc = ll_dir_get_parent_fid(inode, &pfid);
368 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
369 LUSTRE_OPC_ANY, inode);
371 GOTO(out, rc = PTR_ERR(op_data));
373 /* foreign dirs are browsed out of Lustre */
374 if (unlikely(op_data->op_mea1 != NULL &&
375 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
376 ll_finish_md_op_data(op_data);
380 op_data->op_fid3 = pfid;
382 #ifdef HAVE_DIR_CONTEXT
384 rc = ll_dir_read(inode, &pos, op_data, ctx);
387 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
392 if (pos == MDS_DIR_END_OFF) {
394 pos = LL_DIR_END_OFF_32BIT;
396 pos = LL_DIR_END_OFF;
401 #ifdef HAVE_DIR_CONTEXT
406 ll_finish_md_op_data(op_data);
410 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
415 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
416 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
418 struct mgs_send_param *msp;
425 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
426 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
427 sizeof(struct mgs_send_param), msp, NULL);
429 CERROR("Failed to set parameter: %d\n", rc);
437 * Create striped directory with specified stripe(@lump)
439 * \param[in] dparent the parent of the directory.
440 * \param[in] lump the specified stripes.
441 * \param[in] dirname the name of the directory.
442 * \param[in] mode the specified mode of the directory.
444 * \retval =0 if striped directory is being created successfully.
445 * <0 if the creation is failed.
447 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
448 size_t len, const char *dirname, umode_t mode)
450 struct inode *parent = dparent->d_inode;
451 struct ptlrpc_request *request = NULL;
452 struct md_op_data *op_data;
453 struct ll_sb_info *sbi = ll_i2sbi(parent);
454 struct inode *inode = NULL;
455 struct dentry dentry = {
459 .len = strlen(dirname),
460 .hash = ll_full_name_hash(dparent, dirname,
467 if (unlikely(!lmv_magic_supported(lump->lum_magic)))
470 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
472 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
473 PFID(ll_inode2fid(parent)), parent, dirname,
474 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
476 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
479 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
480 PFID(ll_inode2fid(parent)), parent, dirname,
481 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
484 if (lump->lum_stripe_count > 1 &&
485 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
488 if (IS_DEADDIR(parent) &&
489 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
492 if (unlikely(!lmv_magic_supported(cpu_to_le32(lump->lum_magic))))
493 lustre_swab_lmv_user_md(lump);
495 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
496 mode &= ~current_umask();
497 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
498 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
499 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
502 RETURN(PTR_ERR(op_data));
504 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
505 /* selinux_dentry_init_security() uses dentry->d_parent and name
506 * to determine the security context for the file. So our fake
507 * dentry should be real enough for this purpose. */
508 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
509 &op_data->op_file_secctx_name,
510 &op_data->op_file_secctx,
511 &op_data->op_file_secctx_size);
513 GOTO(out_op_data, err);
516 op_data->op_cli_flags |= CLI_SET_MEA;
517 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
518 from_kuid(&init_user_ns, current_fsuid()),
519 from_kgid(&init_user_ns, current_fsgid()),
520 cfs_curproc_cap_pack(), 0, &request);
522 GOTO(out_request, err);
524 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
526 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
528 GOTO(out_inode, err);
530 dentry.d_inode = inode;
532 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
534 err = security_inode_notifysecctx(inode,
535 op_data->op_file_secctx,
536 op_data->op_file_secctx_size);
539 err = ll_inode_init_security(&dentry, inode, parent);
542 GOTO(out_inode, err);
548 ptlrpc_req_finished(request);
550 ll_finish_md_op_data(op_data);
555 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
558 struct ll_sb_info *sbi = ll_i2sbi(inode);
559 struct md_op_data *op_data;
560 struct ptlrpc_request *req = NULL;
562 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
563 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
564 struct obd_device *mgc = lsi->lsi_mgc;
571 * This is coming from userspace, so should be in
572 * local endian. But the MDS would like it in little
573 * endian, so we swab it before we send it.
575 switch (lump->lmm_magic) {
576 case LOV_USER_MAGIC_V1: {
577 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
578 lustre_swab_lov_user_md_v1(lump);
579 lum_size = sizeof(struct lov_user_md_v1);
582 case LOV_USER_MAGIC_V3: {
583 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
584 lustre_swab_lov_user_md_v3(
585 (struct lov_user_md_v3 *)lump);
586 lum_size = sizeof(struct lov_user_md_v3);
589 case LOV_USER_MAGIC_COMP_V1: {
590 if (lump->lmm_magic !=
591 cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
592 lustre_swab_lov_comp_md_v1(
593 (struct lov_comp_md_v1 *)lump);
594 lum_size = le32_to_cpu(
595 ((struct lov_comp_md_v1 *)lump)->lcm_size);
598 case LMV_USER_MAGIC: {
599 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
600 lustre_swab_lmv_user_md(
601 (struct lmv_user_md *)lump);
602 lum_size = sizeof(struct lmv_user_md);
605 case LOV_USER_MAGIC_SPECIFIC: {
606 struct lov_user_md_v3 *v3 =
607 (struct lov_user_md_v3 *)lump;
608 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
610 if (lump->lmm_magic !=
611 cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
612 lustre_swab_lov_user_md_v3(v3);
613 lustre_swab_lov_user_md_objects(v3->lmm_objects,
614 v3->lmm_stripe_count);
616 lum_size = lov_user_md_size(v3->lmm_stripe_count,
617 LOV_USER_MAGIC_SPECIFIC);
621 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
622 " %#08x != %#08x nor %#08x\n",
623 lump->lmm_magic, LOV_USER_MAGIC_V1,
629 lum_size = sizeof(struct lov_user_md_v1);
632 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
633 LUSTRE_OPC_ANY, NULL);
635 RETURN(PTR_ERR(op_data));
637 /* swabbing is done in lov_setstripe() on server side */
638 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
639 ll_finish_md_op_data(op_data);
640 ptlrpc_req_finished(req);
644 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
646 * 2.9 server has stored filesystem default stripe in ROOT xattr,
647 * and it's stored into system config for backward compatibility.
649 * In the following we use the fact that LOV_USER_MAGIC_V1 and
650 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
651 * need the make the distiction between the 2 versions
653 if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
655 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
656 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
660 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
662 GOTO(end, rc = -ENOMEM);
665 /* Get fsname and assume devname to be -MDT0000. */
666 snprintf(buf, MGS_PARAM_MAXLEN, "%s-MDT0000.lov",
670 /* Set root stripesize */
671 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
672 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
673 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
677 /* Set root stripecount */
678 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
679 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
680 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
684 /* Set root stripeoffset */
685 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
686 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
687 (typeof(lump->lmm_stripe_offset))(-1));
688 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
692 OBD_FREE(param, MGS_PARAM_MAXLEN);
699 * This function will be used to get default LOV/LMV/Default LMV
700 * @valid will be used to indicate which stripe it will retrieve
701 * OBD_MD_MEA LMV stripe EA
702 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
703 * otherwise Default LOV EA.
704 * Each time, it can only retrieve 1 stripe EA
706 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
707 struct ptlrpc_request **request, u64 valid)
709 struct ll_sb_info *sbi = ll_i2sbi(inode);
710 struct mdt_body *body;
711 struct lov_mds_md *lmm = NULL;
712 struct ptlrpc_request *req = NULL;
714 struct md_op_data *op_data;
717 rc = ll_get_default_mdsize(sbi, &lmm_size);
721 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
722 0, lmm_size, LUSTRE_OPC_ANY,
725 RETURN(PTR_ERR(op_data));
727 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
728 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
729 ll_finish_md_op_data(op_data);
731 CDEBUG(D_INFO, "md_getattr failed on inode "
732 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
736 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
737 LASSERT(body != NULL);
739 lmm_size = body->mbo_eadatasize;
741 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
743 GOTO(out, rc = -ENODATA);
746 lmm = req_capsule_server_sized_get(&req->rq_pill,
747 &RMF_MDT_MD, lmm_size);
748 LASSERT(lmm != NULL);
751 * This is coming from the MDS, so is probably in
752 * little endian. We convert it to host endian before
753 * passing it to userspace.
755 /* We don't swab objects for directories */
756 switch (le32_to_cpu(lmm->lmm_magic)) {
758 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
759 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
762 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
763 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
765 case LOV_MAGIC_COMP_V1:
766 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
767 lustre_swab_lov_comp_md_v1(
768 (struct lov_comp_md_v1 *)lmm);
771 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
772 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
775 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
776 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
778 case LOV_USER_MAGIC_SPECIFIC: {
779 struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
781 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
782 lustre_swab_lov_user_md_v3(v3);
783 lustre_swab_lov_user_md_objects(v3->lmm_objects,
784 v3->lmm_stripe_count);
788 case LMV_MAGIC_FOREIGN: {
789 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
791 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
792 __swab32s(&lfm->lfm_magic);
793 __swab32s(&lfm->lfm_length);
794 __swab32s(&lfm->lfm_type);
795 __swab32s(&lfm->lfm_flags);
800 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
805 *plmm_size = lmm_size;
810 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
812 struct md_op_data *op_data;
817 OBD_ALLOC_PTR(op_data);
821 op_data->op_flags |= MF_GET_MDT_IDX;
822 op_data->op_fid1 = *fid;
823 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
824 mdt_index = op_data->op_mds;
825 OBD_FREE_PTR(op_data);
833 * Get MDT index for the inode.
835 int ll_get_mdt_idx(struct inode *inode)
837 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
841 * Generic handler to do any pre-copy work.
843 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
844 * first information for it that real work has started.
846 * Moreover, for a ARCHIVE request, it will sample the file data version and
847 * store it in \a copy.
849 * \return 0 on success.
851 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
853 struct ll_sb_info *sbi = ll_s2sbi(sb);
854 struct hsm_progress_kernel hpk;
859 /* Forge a hsm_progress based on data from copy. */
860 hpk.hpk_fid = copy->hc_hai.hai_fid;
861 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
862 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
863 hpk.hpk_extent.length = 0;
866 hpk.hpk_data_version = 0;
869 /* For archive request, we need to read the current file version. */
870 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
872 __u64 data_version = 0;
874 /* Get inode for this fid */
875 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
877 hpk.hpk_flags |= HP_FLAG_RETRY;
878 /* hpk_errval is >= 0 */
879 hpk.hpk_errval = -PTR_ERR(inode);
880 GOTO(progress, rc = PTR_ERR(inode));
883 /* Read current file data version */
884 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
887 CDEBUG(D_HSM, "Could not read file data version of "
888 DFID" (rc = %d). Archive request ("
889 "%#llx) could not be done.\n",
890 PFID(©->hc_hai.hai_fid), rc,
891 copy->hc_hai.hai_cookie);
892 hpk.hpk_flags |= HP_FLAG_RETRY;
893 /* hpk_errval must be >= 0 */
894 hpk.hpk_errval = -rc;
898 /* Store in the hsm_copy for later copytool use.
899 * Always modified even if no lsm. */
900 copy->hc_data_version = data_version;
904 /* On error, the request should be considered as completed */
905 if (hpk.hpk_errval > 0)
906 hpk.hpk_flags |= HP_FLAG_COMPLETED;
908 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
911 /* Return first error */
912 RETURN(rc != 0 ? rc : rc2);
916 * Generic handler to do any post-copy work.
918 * It will send the last hsm_progress update to coordinator to inform it
919 * that copy is finished and whether it was successful or not.
922 * - for ARCHIVE request, it will sample the file data version and compare it
923 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
924 * will be considered as failed.
925 * - for RESTORE request, it will sample the file data version and send it to
926 * coordinator which is useful if the file was imported as 'released'.
928 * \return 0 on success.
930 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
932 struct ll_sb_info *sbi = ll_s2sbi(sb);
933 struct hsm_progress_kernel hpk;
938 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
939 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
940 * initialized if copy_end was called with copy == NULL.
943 /* Forge a hsm_progress based on data from copy. */
944 hpk.hpk_fid = copy->hc_hai.hai_fid;
945 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
946 hpk.hpk_extent = copy->hc_hai.hai_extent;
947 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
948 hpk.hpk_errval = copy->hc_errval;
949 hpk.hpk_data_version = 0;
951 /* For archive request, we need to check the file data was not changed.
953 * For restore request, we need to send the file data version, this is
954 * useful when the file was created using hsm_import.
956 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
957 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
958 (copy->hc_errval == 0)) {
960 __u64 data_version = 0;
962 /* Get lsm for this fid */
963 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
965 hpk.hpk_flags |= HP_FLAG_RETRY;
966 /* hpk_errval must be >= 0 */
967 hpk.hpk_errval = -PTR_ERR(inode);
968 GOTO(progress, rc = PTR_ERR(inode));
971 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
974 CDEBUG(D_HSM, "Could not read file data version. "
975 "Request could not be confirmed.\n");
976 if (hpk.hpk_errval == 0)
977 hpk.hpk_errval = -rc;
981 /* Store in the hsm_copy for later copytool use.
982 * Always modified even if no lsm. */
983 hpk.hpk_data_version = data_version;
985 /* File could have been stripped during archiving, so we need
986 * to check anyway. */
987 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
988 (copy->hc_data_version != data_version)) {
989 CDEBUG(D_HSM, "File data version mismatched. "
990 "File content was changed during archiving. "
991 DFID", start:%#llx current:%#llx\n",
992 PFID(©->hc_hai.hai_fid),
993 copy->hc_data_version, data_version);
994 /* File was changed, send error to cdt. Do not ask for
995 * retry because if a file is modified frequently,
996 * the cdt will loop on retried archive requests.
997 * The policy engine will ask for a new archive later
998 * when the file will not be modified for some tunable
1000 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1002 /* hpk_errval must be >= 0 */
1003 hpk.hpk_errval = -rc;
1010 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1013 /* Return first error */
1014 RETURN(rc != 0 ? rc : rc2);
1018 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1019 const struct lustre_kernelcomm __user *data)
1021 struct lustre_kernelcomm *lk;
1022 struct lustre_kernelcomm *tmp;
1023 size_t size = sizeof(*lk);
1028 /* copy data from userspace to get numbers of archive_id */
1029 OBD_ALLOC(lk, size);
1033 if (copy_from_user(lk, data, size))
1034 GOTO(out_lk, rc = -EFAULT);
1036 if (lk->lk_flags & LK_FLG_STOP)
1039 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1040 __u32 archive_mask = lk->lk_data_count;
1043 /* old hsm agent to old MDS */
1044 if (!exp_connect_archive_id_array(exp))
1047 /* old hsm agent to new MDS */
1048 lk->lk_flags |= LK_FLG_DATANR;
1050 if (archive_mask == 0)
1053 count = hweight32(archive_mask);
1054 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1055 OBD_ALLOC(tmp, new_size);
1057 GOTO(out_lk, rc = -ENOMEM);
1059 memcpy(tmp, lk, size);
1060 tmp->lk_data_count = count;
1066 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1067 if ((1 << i) & archive_mask) {
1068 lk->lk_data[count] = i + 1;
1075 /* new hsm agent to new mds */
1076 if (lk->lk_data_count > 0) {
1077 new_size = offsetof(struct lustre_kernelcomm,
1078 lk_data[lk->lk_data_count]);
1079 OBD_ALLOC(tmp, new_size);
1081 GOTO(out_lk, rc = -ENOMEM);
1087 if (copy_from_user(lk, data, size))
1088 GOTO(out_lk, rc = -EFAULT);
1091 /* new hsm agent to old MDS */
1092 if (!exp_connect_archive_id_array(exp)) {
1095 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1096 GOTO(out_lk, rc = -EINVAL);
1098 for (i = 0; i < lk->lk_data_count; i++) {
1099 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1101 CERROR("%s: archive id %d requested but only "
1102 "[0 - %zu] supported: rc = %d\n",
1103 exp->exp_obd->obd_name, lk->lk_data[i],
1104 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1108 if (lk->lk_data[i] == 0) {
1113 archives |= (1 << (lk->lk_data[i] - 1));
1115 lk->lk_flags &= ~LK_FLG_DATANR;
1116 lk->lk_data_count = archives;
1119 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1125 static int check_owner(int type, int id)
1129 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1133 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1142 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1144 int cmd = qctl->qc_cmd;
1145 int type = qctl->qc_type;
1146 int id = qctl->qc_id;
1147 int valid = qctl->qc_valid;
1154 case LUSTRE_Q_SETDEFAULT:
1155 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1159 case LUSTRE_Q_GETDEFAULT:
1160 if (check_owner(type, id) &&
1161 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1167 CERROR("unsupported quotactl op: %#x\n", cmd);
1171 if (valid != QC_GENERAL) {
1172 if (cmd == Q_GETINFO)
1173 qctl->qc_cmd = Q_GETOINFO;
1174 else if (cmd == Q_GETQUOTA)
1175 qctl->qc_cmd = Q_GETOQUOTA;
1181 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1182 sizeof(*qctl), qctl, NULL);
1185 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1186 sizeof(*qctl), qctl, NULL);
1189 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1190 sizeof(*qctl), qctl, NULL);
1192 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1194 sizeof(*qctl), qctl, NULL);
1206 struct obd_quotactl *oqctl;
1208 OBD_ALLOC_PTR(oqctl);
1212 QCTL_COPY(oqctl, qctl);
1213 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1215 OBD_FREE_PTR(oqctl);
1218 /* If QIF_SPACE is not set, client should collect the
1219 * space usage from OSSs by itself */
1220 if (cmd == Q_GETQUOTA &&
1221 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1222 !oqctl->qc_dqblk.dqb_curspace) {
1223 struct obd_quotactl *oqctl_tmp;
1225 OBD_ALLOC_PTR(oqctl_tmp);
1226 if (oqctl_tmp == NULL)
1227 GOTO(out, rc = -ENOMEM);
1229 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1230 oqctl_tmp->qc_id = oqctl->qc_id;
1231 oqctl_tmp->qc_type = oqctl->qc_type;
1233 /* collect space usage from OSTs */
1234 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1235 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1236 if (!rc || rc == -EREMOTEIO) {
1237 oqctl->qc_dqblk.dqb_curspace =
1238 oqctl_tmp->qc_dqblk.dqb_curspace;
1239 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1242 /* collect space & inode usage from MDTs */
1243 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1244 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1245 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1246 if (!rc || rc == -EREMOTEIO) {
1247 oqctl->qc_dqblk.dqb_curspace +=
1248 oqctl_tmp->qc_dqblk.dqb_curspace;
1249 oqctl->qc_dqblk.dqb_curinodes =
1250 oqctl_tmp->qc_dqblk.dqb_curinodes;
1251 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1253 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1256 OBD_FREE_PTR(oqctl_tmp);
1259 QCTL_COPY(qctl, oqctl);
1260 OBD_FREE_PTR(oqctl);
1266 /* This function tries to get a single name component,
1267 * to send to the server. No actual path traversal involved,
1268 * so we limit to NAME_MAX */
1269 static char *ll_getname(const char __user *filename)
1274 OBD_ALLOC(tmp, NAME_MAX + 1);
1277 return ERR_PTR(-ENOMEM);
1279 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1282 else if (len > NAME_MAX)
1283 ret = -ENAMETOOLONG;
1286 OBD_FREE(tmp, NAME_MAX + 1);
1292 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1294 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1296 struct dentry *dentry = file_dentry(file);
1297 struct inode *inode = file_inode(file);
1298 struct ll_sb_info *sbi = ll_i2sbi(inode);
1299 struct obd_ioctl_data *data;
1303 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1304 PFID(ll_inode2fid(inode)), inode, cmd);
1306 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1307 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1310 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1312 case FS_IOC_GETFLAGS:
1313 case FS_IOC_SETFLAGS:
1314 RETURN(ll_iocontrol(inode, file, cmd, arg));
1315 case FSFILT_IOC_GETVERSION:
1316 case FS_IOC_GETVERSION:
1317 RETURN(put_user(inode->i_generation, (int __user *)arg));
1318 /* We need to special case any other ioctls we want to handle,
1319 * to send them to the MDS/OST as appropriate and to properly
1320 * network encode the arg field. */
1321 case FS_IOC_SETVERSION:
1324 case LL_IOC_GET_MDTIDX: {
1327 mdtidx = ll_get_mdt_idx(inode);
1331 if (put_user((int)mdtidx, (int __user *)arg))
1336 case IOC_MDC_LOOKUP: {
1337 int namelen, len = 0;
1341 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1346 filename = data->ioc_inlbuf1;
1347 namelen = strlen(filename);
1349 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1350 GOTO(out_free, rc = -EINVAL);
1353 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1355 CERROR("%s: lookup %.*s failed: rc = %d\n",
1356 sbi->ll_fsname, namelen, filename, rc);
1360 OBD_FREE_LARGE(buf, len);
1363 case LL_IOC_LMV_SETSTRIPE: {
1364 struct lmv_user_md *lum;
1373 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1378 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1379 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1380 GOTO(lmv_out_free, rc = -EINVAL);
1382 filename = data->ioc_inlbuf1;
1383 namelen = data->ioc_inllen1;
1386 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1387 GOTO(lmv_out_free, rc = -EINVAL);
1389 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1390 lumlen = data->ioc_inllen2;
1392 if (!lmv_magic_supported(lum->lum_magic)) {
1393 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1394 lum->lum_magic, -EINVAL);
1395 GOTO(lmv_out_free, rc = -EINVAL);
1398 if ((lum->lum_magic == LMV_USER_MAGIC ||
1399 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1400 lumlen < sizeof(*lum)) {
1401 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1402 filename, lumlen, lum->lum_magic, -EINVAL);
1403 GOTO(lmv_out_free, rc = -EINVAL);
1406 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1407 lumlen < sizeof(struct lmv_foreign_md)) {
1408 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1409 filename, lum->lum_magic, lumlen, -EFAULT);
1410 GOTO(lmv_out_free, rc = -EINVAL);
1413 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1414 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1416 mode = data->ioc_type;
1418 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1420 OBD_FREE_LARGE(buf, len);
1424 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1425 struct lmv_user_md lum;
1426 struct lmv_user_md __user *ulump =
1427 (struct lmv_user_md __user *)arg;
1430 if (copy_from_user(&lum, ulump, sizeof(lum)))
1433 if (lum.lum_magic != LMV_USER_MAGIC)
1436 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1440 case LL_IOC_LOV_SETSTRIPE_NEW:
1441 case LL_IOC_LOV_SETSTRIPE: {
1442 struct lov_user_md_v3 *lumv3 = NULL;
1443 struct lov_user_md_v1 lumv1;
1444 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1445 struct lov_user_md_v1 __user *lumv1p =
1446 (struct lov_user_md_v1 __user *)arg;
1447 struct lov_user_md_v3 __user *lumv3p =
1448 (struct lov_user_md_v3 __user *)arg;
1451 int set_default = 0;
1453 CLASSERT(sizeof(struct lov_user_md_v3) >
1454 sizeof(struct lov_comp_md_v1));
1455 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1456 /* first try with v1 which is smaller than v3 */
1457 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1460 if (inode->i_sb->s_root == file_dentry(file))
1463 switch (lumv1.lmm_magic) {
1464 case LOV_USER_MAGIC_V3:
1465 case LOV_USER_MAGIC_SPECIFIC:
1466 lum_size = ll_lov_user_md_size(&lumv1);
1469 OBD_ALLOC(lumv3, lum_size);
1472 if (copy_from_user(lumv3, lumv3p, lum_size))
1473 GOTO(out, rc = -EFAULT);
1474 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1476 case LOV_USER_MAGIC_V1:
1479 GOTO(out, rc = -ENOTSUPP);
1482 /* in v1 and v3 cases lumv1 points to data */
1483 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1486 OBD_FREE(lumv3, lum_size);
1489 case LL_IOC_LMV_GETSTRIPE: {
1490 struct lmv_user_md __user *ulmv =
1491 (struct lmv_user_md __user *)arg;
1492 struct lmv_user_md lum;
1493 struct ptlrpc_request *request = NULL;
1494 union lmv_mds_md *lmm = NULL;
1497 struct lmv_user_md *tmp = NULL;
1501 int max_stripe_count;
1505 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1508 max_stripe_count = lum.lum_stripe_count;
1509 /* lum_magic will indicate which stripe the ioctl will like
1510 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1511 * is for default LMV stripe */
1512 if (lum.lum_magic == LMV_MAGIC_V1)
1513 valid |= OBD_MD_MEA;
1514 else if (lum.lum_magic == LMV_USER_MAGIC)
1515 valid |= OBD_MD_DEFAULT_MEA;
1519 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1522 GOTO(finish_req, rc);
1524 /* Get default LMV EA */
1525 if (lum.lum_magic == LMV_USER_MAGIC) {
1526 if (lmmsize > sizeof(*ulmv))
1527 GOTO(finish_req, rc = -EINVAL);
1529 if (copy_to_user(ulmv, lmm, lmmsize))
1530 GOTO(finish_req, rc = -EFAULT);
1532 GOTO(finish_req, rc);
1535 /* if foreign LMV case, fake stripes number */
1536 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1537 struct lmv_foreign_md *lfm;
1539 lfm = (struct lmv_foreign_md *)lmm;
1540 if (lfm->lfm_length < XATTR_SIZE_MAX -
1541 offsetof(typeof(*lfm), lfm_value)) {
1542 __u32 size = lfm->lfm_length +
1543 offsetof(typeof(*lfm), lfm_value);
1545 stripe_count = lmv_foreign_to_md_stripes(size);
1547 CERROR("invalid %d foreign size returned\n",
1552 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1554 if (max_stripe_count < stripe_count) {
1555 lum.lum_stripe_count = stripe_count;
1556 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1557 GOTO(finish_req, rc = -EFAULT);
1558 GOTO(finish_req, rc = -E2BIG);
1561 /* enough room on user side and foreign case */
1562 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1563 struct lmv_foreign_md *lfm;
1566 lfm = (struct lmv_foreign_md *)lmm;
1567 size = lfm->lfm_length +
1568 offsetof(struct lmv_foreign_md, lfm_value);
1569 if (copy_to_user(ulmv, lfm, size))
1570 GOTO(finish_req, rc = -EFAULT);
1571 GOTO(finish_req, rc);
1574 lum_size = lmv_user_md_size(stripe_count,
1575 LMV_USER_MAGIC_SPECIFIC);
1576 OBD_ALLOC(tmp, lum_size);
1578 GOTO(finish_req, rc = -ENOMEM);
1580 mdt_index = ll_get_mdt_idx(inode);
1582 GOTO(out_tmp, rc = -ENOMEM);
1584 tmp->lum_magic = LMV_MAGIC_V1;
1585 tmp->lum_stripe_count = 0;
1586 tmp->lum_stripe_offset = mdt_index;
1587 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1588 for (i = 0; i < stripe_count; i++) {
1591 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1592 if (fid_is_sane(&fid)) {
1593 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1595 GOTO(out_tmp, rc = mdt_index);
1597 tmp->lum_objects[i].lum_mds = mdt_index;
1598 tmp->lum_objects[i].lum_fid = fid;
1601 tmp->lum_stripe_count++;
1604 if (copy_to_user(ulmv, tmp, lum_size))
1605 GOTO(out_tmp, rc = -EFAULT);
1607 OBD_FREE(tmp, lum_size);
1609 ptlrpc_req_finished(request);
1613 case LL_IOC_REMOVE_ENTRY: {
1614 char *filename = NULL;
1618 /* Here is a little hack to avoid sending REINT_RMENTRY to
1619 * unsupported server, which might crash the server(LU-2730),
1620 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1621 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1622 * server will support REINT_RMENTRY XXX*/
1623 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1626 filename = ll_getname((const char __user *)arg);
1627 if (IS_ERR(filename))
1628 RETURN(PTR_ERR(filename));
1630 namelen = strlen(filename);
1632 GOTO(out_rmdir, rc = -EINVAL);
1634 rc = ll_rmdir_entry(inode, filename, namelen);
1637 ll_putname(filename);
1640 case LL_IOC_LOV_SWAP_LAYOUTS:
1642 case IOC_OBD_STATFS:
1643 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1644 case LL_IOC_LOV_GETSTRIPE:
1645 case LL_IOC_LOV_GETSTRIPE_NEW:
1646 case LL_IOC_MDC_GETINFO:
1647 case IOC_MDC_GETFILEINFO:
1648 case IOC_MDC_GETFILESTRIPE: {
1649 struct ptlrpc_request *request = NULL;
1650 struct lov_user_md __user *lump;
1651 struct lov_mds_md *lmm = NULL;
1652 struct mdt_body *body;
1653 char *filename = NULL;
1656 if (cmd == IOC_MDC_GETFILEINFO ||
1657 cmd == IOC_MDC_GETFILESTRIPE) {
1658 filename = ll_getname((const char __user *)arg);
1659 if (IS_ERR(filename))
1660 RETURN(PTR_ERR(filename));
1662 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1663 &lmmsize, &request);
1665 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1670 body = req_capsule_server_get(&request->rq_pill,
1672 LASSERT(body != NULL);
1677 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1678 cmd == LL_IOC_MDC_GETINFO)) {
1686 if (cmd == IOC_MDC_GETFILESTRIPE ||
1687 cmd == LL_IOC_LOV_GETSTRIPE ||
1688 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1689 lump = (struct lov_user_md __user *)arg;
1691 struct lov_user_mds_data __user *lmdp;
1692 lmdp = (struct lov_user_mds_data __user *)arg;
1693 lump = &lmdp->lmd_lmm;
1697 /* If the file has no striping then zero out *lump so
1698 * that the caller isn't confused by garbage. */
1699 if (clear_user(lump, sizeof(*lump)))
1700 GOTO(out_req, rc = -EFAULT);
1701 } else if (copy_to_user(lump, lmm, lmmsize)) {
1702 if (copy_to_user(lump, lmm, sizeof(*lump)))
1703 GOTO(out_req, rc = -EFAULT);
1707 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1708 struct lov_user_mds_data __user *lmdp;
1711 st.st_dev = inode->i_sb->s_dev;
1712 st.st_mode = body->mbo_mode;
1713 st.st_nlink = body->mbo_nlink;
1714 st.st_uid = body->mbo_uid;
1715 st.st_gid = body->mbo_gid;
1716 st.st_rdev = body->mbo_rdev;
1717 st.st_size = body->mbo_size;
1718 st.st_blksize = PAGE_SIZE;
1719 st.st_blocks = body->mbo_blocks;
1720 st.st_atime = body->mbo_atime;
1721 st.st_mtime = body->mbo_mtime;
1722 st.st_ctime = body->mbo_ctime;
1723 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1727 lmdp = (struct lov_user_mds_data __user *)arg;
1728 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1729 GOTO(out_req, rc = -EFAULT);
1734 ptlrpc_req_finished(request);
1736 ll_putname(filename);
1739 case OBD_IOC_QUOTACTL: {
1740 struct if_quotactl *qctl;
1742 OBD_ALLOC_PTR(qctl);
1746 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1747 GOTO(out_quotactl, rc = -EFAULT);
1749 rc = quotactl_ioctl(sbi, qctl);
1752 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1759 case OBD_IOC_GETDTNAME:
1760 case OBD_IOC_GETMDNAME:
1761 RETURN(ll_get_obd_name(inode, cmd, arg));
1762 case LL_IOC_FLUSHCTX:
1763 RETURN(ll_flush_ctx(inode));
1764 case LL_IOC_GETOBDCOUNT: {
1766 struct obd_export *exp;
1768 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1771 /* get ost count when count is zero, get mdt count otherwise */
1772 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1773 vallen = sizeof(count);
1774 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1775 KEY_TGT_COUNT, &vallen, &count);
1777 CERROR("get target count failed: %d\n", rc);
1781 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1786 case LL_IOC_PATH2FID:
1787 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1788 sizeof(struct lu_fid)))
1791 case LL_IOC_GET_CONNECT_FLAGS: {
1792 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1793 (void __user *)arg));
1795 case OBD_IOC_FID2PATH:
1796 RETURN(ll_fid2path(inode, (void __user *)arg));
1797 case LL_IOC_GETPARENT:
1798 RETURN(ll_getparent(file, (void __user *)arg));
1799 case LL_IOC_FID2MDTIDX: {
1800 struct obd_export *exp = ll_i2mdexp(inode);
1804 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1808 /* Call mdc_iocontrol */
1809 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1810 (__u32 __user *)&index);
1816 case LL_IOC_HSM_REQUEST: {
1817 struct hsm_user_request *hur;
1824 /* We don't know the true size yet; copy the fixed-size part */
1825 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1830 /* Compute the whole struct size */
1831 totalsize = hur_len(hur);
1836 /* Final size will be more than double totalsize */
1837 if (totalsize >= MDS_MAXREQSIZE / 3)
1840 OBD_ALLOC_LARGE(hur, totalsize);
1844 /* Copy the whole struct */
1845 if (copy_from_user(hur, (void __user *)arg, totalsize))
1846 GOTO(out_hur, rc = -EFAULT);
1848 if (hur->hur_request.hr_action == HUA_RELEASE) {
1849 const struct lu_fid *fid;
1853 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1854 fid = &hur->hur_user_item[i].hui_fid;
1855 f = search_inode_for_lustre(inode->i_sb, fid);
1861 rc = ll_hsm_release(f);
1867 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1872 OBD_FREE_LARGE(hur, totalsize);
1876 case LL_IOC_HSM_PROGRESS: {
1877 struct hsm_progress_kernel hpk;
1878 struct hsm_progress hp;
1880 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1883 hpk.hpk_fid = hp.hp_fid;
1884 hpk.hpk_cookie = hp.hp_cookie;
1885 hpk.hpk_extent = hp.hp_extent;
1886 hpk.hpk_flags = hp.hp_flags;
1887 hpk.hpk_errval = hp.hp_errval;
1888 hpk.hpk_data_version = 0;
1890 /* File may not exist in Lustre; all progress
1891 * reported to Lustre root */
1892 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1896 case LL_IOC_HSM_CT_START:
1897 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1900 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
1901 (struct lustre_kernelcomm __user *)arg);
1904 case LL_IOC_HSM_COPY_START: {
1905 struct hsm_copy *copy;
1908 OBD_ALLOC_PTR(copy);
1911 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1916 rc = ll_ioc_copy_start(inode->i_sb, copy);
1917 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1923 case LL_IOC_HSM_COPY_END: {
1924 struct hsm_copy *copy;
1927 OBD_ALLOC_PTR(copy);
1930 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1935 rc = ll_ioc_copy_end(inode->i_sb, copy);
1936 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1942 case LL_IOC_MIGRATE: {
1943 struct lmv_user_md *lum;
1950 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1954 data = (struct obd_ioctl_data *)buf;
1955 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1956 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1957 GOTO(migrate_free, rc = -EINVAL);
1959 filename = data->ioc_inlbuf1;
1960 namelen = data->ioc_inllen1;
1962 if (namelen < 1 || namelen != strlen(filename) + 1) {
1963 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1964 GOTO(migrate_free, rc = -EINVAL);
1967 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1968 if (lum->lum_magic != LMV_USER_MAGIC &&
1969 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1971 CERROR("%s: wrong lum magic %x: rc = %d\n",
1972 filename, lum->lum_magic, rc);
1973 GOTO(migrate_free, rc);
1976 rc = ll_migrate(inode, file, lum, filename);
1978 OBD_FREE_LARGE(buf, len);
1982 case LL_IOC_FSGETXATTR:
1983 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1984 case LL_IOC_FSSETXATTR:
1985 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1986 case LL_IOC_PCC_DETACH_BY_FID: {
1987 struct lu_pcc_detach *detach;
1989 struct inode *inode2;
1992 OBD_ALLOC_PTR(detach);
1996 if (copy_from_user(detach,
1997 (const struct lu_pcc_detach __user *)arg,
1999 GOTO(out_detach, rc = -EFAULT);
2001 fid = &detach->pccd_fid;
2002 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2003 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2005 /* Target inode is not in inode cache, and PCC file
2006 * has aleady released, return immdiately.
2008 GOTO(out_detach, rc = 0);
2010 if (!S_ISREG(inode2->i_mode))
2011 GOTO(out_iput, rc = -EINVAL);
2013 if (!inode_owner_or_capable(inode2))
2014 GOTO(out_iput, rc = -EPERM);
2016 rc = pcc_ioctl_detach(inode2);
2020 OBD_FREE_PTR(detach);
2024 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2025 (void __user *)arg));
2029 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2031 struct inode *inode = file->f_mapping->host;
2032 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2033 struct ll_sb_info *sbi = ll_i2sbi(inode);
2034 int api32 = ll_need_32bit_api(sbi);
2035 loff_t ret = -EINVAL;
2043 offset += file->f_pos;
2049 offset += LL_DIR_END_OFF_32BIT;
2051 offset += LL_DIR_END_OFF;
2058 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2059 (!api32 && offset <= LL_DIR_END_OFF))) {
2060 if (offset != file->f_pos) {
2061 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2062 (!api32 && offset == LL_DIR_END_OFF))
2063 fd->lfd_pos = MDS_DIR_END_OFF;
2064 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2065 fd->lfd_pos = offset << 32;
2067 fd->lfd_pos = offset;
2068 file->f_pos = offset;
2069 file->f_version = 0;
2076 inode_unlock(inode);
2080 static int ll_dir_open(struct inode *inode, struct file *file)
2083 RETURN(ll_file_open(inode, file));
2086 static int ll_dir_release(struct inode *inode, struct file *file)
2089 RETURN(ll_file_release(inode, file));
2092 const struct file_operations ll_dir_operations = {
2093 .llseek = ll_dir_seek,
2094 .open = ll_dir_open,
2095 .release = ll_dir_release,
2096 .read = generic_read_dir,
2097 #ifdef HAVE_DIR_CONTEXT
2098 .iterate = ll_iterate,
2100 .readdir = ll_readdir,
2102 .unlocked_ioctl = ll_dir_ioctl,