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 #include <linux/uidgid.h>
44 #include <linux/uaccess.h>
45 #include <linux/buffer_head.h> // for wait_on_buffer
46 #include <linux/pagevec.h>
48 #define DEBUG_SUBSYSTEM S_LLITE
50 #include <obd_support.h>
51 #include <obd_class.h>
52 #include <uapi/linux/lustre/lustre_ioctl.h>
53 #include <uapi/linux/llcrypt.h>
54 #include <lustre_lib.h>
55 #include <lustre_dlm.h>
56 #include <lustre_fid.h>
57 #include <lustre_kernelcomm.h>
58 #include <lustre_swab.h>
60 #include "llite_internal.h"
63 * (new) readdir implementation overview.
65 * Original lustre readdir implementation cached exact copy of raw directory
66 * pages on the client. These pages were indexed in client page cache by
67 * logical offset in the directory file. This design, while very simple and
68 * intuitive had some inherent problems:
70 * . it implies that byte offset to the directory entry serves as a
71 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
72 * ext3/htree directory entries may move due to splits, and more
75 * . it is incompatible with the design of split directories for cmd3,
76 * that assumes that names are distributed across nodes based on their
77 * hash, and so readdir should be done in hash order.
79 * New readdir implementation does readdir in hash order, and uses hash of a
80 * file name as a telldir/seekdir cookie. This led to number of complications:
82 * . hash is not unique, so it cannot be used to index cached directory
83 * pages on the client (note, that it requires a whole pageful of hash
84 * collided entries to cause two pages to have identical hashes);
86 * . hash is not unique, so it cannot, strictly speaking, be used as an
87 * entry cookie. ext3/htree has the same problem and lustre implementation
88 * mimics their solution: seekdir(hash) positions directory at the first
89 * entry with the given hash.
95 * Client caches directory pages using hash of the first entry as an index. As
96 * noted above hash is not unique, so this solution doesn't work as is:
97 * special processing is needed for "page hash chains" (i.e., sequences of
98 * pages filled with entries all having the same hash value).
100 * First, such chains have to be detected. To this end, server returns to the
101 * client the hash of the first entry on the page next to one returned. When
102 * client detects that this hash is the same as hash of the first entry on the
103 * returned page, page hash collision has to be handled. Pages in the
104 * hash chain, except first one, are termed "overflow pages".
106 * Proposed (unimplimented) solution to index uniqueness problem is to
107 * not cache overflow pages. Instead, when page hash collision is
108 * detected, all overflow pages from emerging chain should be
109 * immediately requested from the server and placed in a special data
110 * structure. This data structure can be used by ll_readdir() to
111 * process entries from overflow pages. When readdir invocation
112 * finishes, overflow pages are discarded. If page hash collision chain
113 * weren't completely processed, next call to readdir will again detect
114 * page hash collision, again read overflow pages in, process next
115 * portion of entries and again discard the pages. This is not as
116 * wasteful as it looks, because, given reasonable hash, page hash
117 * collisions are extremely rare.
119 * 1. directory positioning
121 * When seekdir(hash) is called, original
132 * identification of and access to overflow pages
136 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
137 * a header lu_dirpage which describes the start/end hash, and whether this
138 * page is empty (contains no dir entry) or hash collide with next page.
139 * After client receives reply, several pages will be integrated into dir page
140 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
141 * lu_dirpage for this integrated page will be adjusted. See
142 * mdc_adjust_dirpages().
145 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
148 struct md_callback cb_op;
152 cb_op.md_blocking_ast = ll_md_blocking_ast;
153 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
160 void ll_release_page(struct inode *inode, struct page *page,
165 /* Always remove the page for striped dir, because the page is
166 * built from temporarily in LMV layer */
167 if (inode && ll_dir_striped(inode)) {
174 if (likely(page->mapping != NULL))
175 delete_from_page_cache(page);
181 #ifdef HAVE_DIR_CONTEXT
182 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
183 struct dir_context *ctx)
186 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
187 void *cookie, filldir_t filldir)
190 struct ll_sb_info *sbi = ll_i2sbi(inode);
192 bool is_api32 = ll_need_32bit_api(sbi);
193 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
199 page = ll_get_dir_page(inode, op_data, pos);
201 while (rc == 0 && !done) {
202 struct lu_dirpage *dp;
203 struct lu_dirent *ent;
212 hash = MDS_DIR_END_OFF;
213 dp = page_address(page);
214 for (ent = lu_dirent_start(dp); ent != NULL && !done;
215 ent = lu_dirent_next(ent)) {
222 hash = le64_to_cpu(ent->lde_hash);
223 if (hash < pos) /* Skip until we find target hash */
226 namelen = le16_to_cpu(ent->lde_namelen);
227 if (namelen == 0) /* Skip dummy record */
230 if (is_api32 && is_hash64)
234 fid_le_to_cpu(&fid, &ent->lde_fid);
235 ino = cl_fid_build_ino(&fid, is_api32);
236 type = IFTODT(lu_dirent_type_get(ent));
237 /* For ll_nfs_get_name_filldir(), it will try to access
238 * 'ent' through 'lde_name', so the parameter 'name'
239 * for 'filldir()' must be part of the 'ent'. */
240 #ifdef HAVE_DIR_CONTEXT
242 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
245 done = filldir(cookie, ent->lde_name, namelen, lhash,
252 ll_release_page(inode, page, false);
256 next = le64_to_cpu(dp->ldp_hash_end);
258 if (pos == MDS_DIR_END_OFF) {
260 * End of directory reached.
263 ll_release_page(inode, page, false);
266 * Normal case: continue to the next
269 ll_release_page(inode, page,
270 le32_to_cpu(dp->ldp_flags) &
273 page = ll_get_dir_page(inode, op_data, pos);
276 #ifdef HAVE_DIR_CONTEXT
284 #ifdef HAVE_DIR_CONTEXT
285 static int ll_iterate(struct file *filp, struct dir_context *ctx)
287 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
290 struct inode *inode = file_inode(filp);
291 struct ll_file_data *lfd = filp->private_data;
292 struct ll_sb_info *sbi = ll_i2sbi(inode);
293 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
294 int api32 = ll_need_32bit_api(sbi);
295 struct md_op_data *op_data;
296 struct lu_fid pfid = { 0 };
297 ktime_t kstart = ktime_get();
308 "VFS Op:inode="DFID"(%p) pos/size%lu/%llu 32bit_api %d\n",
309 PFID(ll_inode2fid(inode)),
310 inode, (unsigned long)pos, i_size_read(inode), api32);
312 if (pos == MDS_DIR_END_OFF)
318 if (unlikely(ll_dir_striped(inode))) {
320 * This is only needed for striped dir to fill ..,
321 * see lmv_read_page()
323 if (file_dentry(filp)->d_parent != NULL &&
324 file_dentry(filp)->d_parent->d_inode != NULL) {
325 __u64 ibits = MDS_INODELOCK_LOOKUP;
326 struct inode *parent =
327 file_dentry(filp)->d_parent->d_inode;
329 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
330 pfid = *ll_inode2fid(parent);
333 /* If it can not find in cache, do lookup .. on the master
335 if (fid_is_zero(&pfid)) {
336 rc = ll_dir_get_parent_fid(inode, &pfid);
342 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
343 LUSTRE_OPC_ANY, inode);
345 GOTO(out, rc = PTR_ERR(op_data));
347 /* foreign dirs are browsed out of Lustre */
348 if (unlikely(op_data->op_mea1 != NULL &&
349 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
350 ll_finish_md_op_data(op_data);
354 op_data->op_fid3 = pfid;
356 #ifdef HAVE_DIR_CONTEXT
358 rc = ll_dir_read(inode, &pos, op_data, ctx);
361 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
366 if (pos == MDS_DIR_END_OFF) {
368 pos = LL_DIR_END_OFF_32BIT;
370 pos = LL_DIR_END_OFF;
375 #ifdef HAVE_DIR_CONTEXT
380 ll_finish_md_op_data(op_data);
384 ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
385 ktime_us_delta(ktime_get(), kstart));
391 * Create striped directory with specified stripe(@lump)
393 * \param[in] dparent the parent of the directory.
394 * \param[in] lump the specified stripes.
395 * \param[in] dirname the name of the directory.
396 * \param[in] mode the specified mode of the directory.
398 * \retval =0 if striped directory is being created successfully.
399 * <0 if the creation is failed.
401 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
402 size_t len, const char *dirname, umode_t mode)
404 struct inode *parent = dparent->d_inode;
405 struct ptlrpc_request *request = NULL;
406 struct md_op_data *op_data;
407 struct ll_sb_info *sbi = ll_i2sbi(parent);
408 struct inode *inode = NULL;
409 struct dentry dentry = {
413 .len = strlen(dirname),
414 .hash = ll_full_name_hash(dparent, dirname,
418 bool encrypt = false;
422 if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
425 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
427 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
428 PFID(ll_inode2fid(parent)), parent, dirname,
429 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
431 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
434 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
435 PFID(ll_inode2fid(parent)), parent, dirname,
436 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
439 if (lump->lum_stripe_count > 1 &&
440 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
443 if (IS_DEADDIR(parent) &&
444 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
447 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
448 if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
449 LMV_HASH_TYPE_CRUSH) {
450 /* if server doesn't support 'crush' hash type,
451 * switch to fnv_1a_64.
453 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
454 lump->lum_hash_type |= LMV_HASH_TYPE_FNV_1A_64;
455 } else if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
456 LMV_HASH_TYPE_UNKNOWN) {
457 /* from 2.14 MDT will choose default hash type if client
458 * doesn't set a valid one, while old server doesn't
461 lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
462 lump->lum_hash_type |= LMV_HASH_TYPE_DEFAULT;
466 if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
467 lustre_swab_lmv_user_md(lump);
469 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
470 mode &= ~current_umask();
471 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
472 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
473 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
476 RETURN(PTR_ERR(op_data));
478 if (ll_sbi_has_encrypt(sbi) &&
479 (IS_ENCRYPTED(parent) ||
480 unlikely(llcrypt_dummy_context_enabled(parent)))) {
481 err = llcrypt_get_encryption_info(parent);
483 GOTO(out_op_data, err);
484 if (!llcrypt_has_encryption_key(parent))
485 GOTO(out_op_data, err = -ENOKEY);
489 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
490 /* selinux_dentry_init_security() uses dentry->d_parent and name
491 * to determine the security context for the file. So our fake
492 * dentry should be real enough for this purpose. */
493 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
494 &op_data->op_file_secctx_name,
495 &op_data->op_file_secctx,
496 &op_data->op_file_secctx_size);
498 GOTO(out_op_data, err);
502 err = llcrypt_inherit_context(parent, NULL, op_data, false);
504 GOTO(out_op_data, err);
507 op_data->op_cli_flags |= CLI_SET_MEA;
508 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
509 from_kuid(&init_user_ns, current_fsuid()),
510 from_kgid(&init_user_ns, current_fsgid()),
511 cfs_curproc_cap_pack(), 0, &request);
513 GOTO(out_request, err);
515 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
517 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
519 GOTO(out_inode, err);
521 dentry.d_inode = inode;
523 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
524 /* no need to protect selinux_inode_setsecurity() by
525 * inode_lock. Taking it would lead to a client deadlock
528 err = security_inode_notifysecctx(inode,
529 op_data->op_file_secctx,
530 op_data->op_file_secctx_size);
532 err = ll_inode_init_security(&dentry, inode, parent);
535 GOTO(out_inode, err);
538 err = ll_set_encflags(inode, op_data->op_file_encctx,
539 op_data->op_file_encctx_size, false);
541 GOTO(out_inode, err);
547 ptlrpc_req_finished(request);
549 ll_finish_md_op_data(op_data);
554 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
557 struct ll_sb_info *sbi = ll_i2sbi(inode);
558 struct md_op_data *op_data;
559 struct ptlrpc_request *req = NULL;
565 switch (lump->lmm_magic) {
566 case LOV_USER_MAGIC_V1:
567 lum_size = sizeof(struct lov_user_md_v1);
569 case LOV_USER_MAGIC_V3:
570 lum_size = sizeof(struct lov_user_md_v3);
572 case LOV_USER_MAGIC_COMP_V1:
573 lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
576 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
577 lustre_swab_lmv_user_md(
578 (struct lmv_user_md *)lump);
579 lum_size = sizeof(struct lmv_user_md);
581 case LOV_USER_MAGIC_SPECIFIC: {
582 struct lov_user_md_v3 *v3 =
583 (struct lov_user_md_v3 *)lump;
584 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
586 lum_size = lov_user_md_size(v3->lmm_stripe_count,
587 LOV_USER_MAGIC_SPECIFIC);
591 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
592 " %#08x != %#08x nor %#08x\n",
593 lump->lmm_magic, LOV_USER_MAGIC_V1,
599 * This is coming from userspace, so should be in
600 * local endian. But the MDS would like it in little
601 * endian, so we swab it before we send it.
603 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
604 le32_to_cpu(LOV_MAGIC_MAGIC))
605 lustre_swab_lov_user_md(lump, 0);
607 lum_size = sizeof(struct lov_user_md_v1);
610 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
611 LUSTRE_OPC_ANY, NULL);
613 RETURN(PTR_ERR(op_data));
615 /* swabbing is done in lov_setstripe() on server side */
616 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
617 ll_finish_md_op_data(op_data);
618 ptlrpc_req_finished(req);
625 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
627 struct ptlrpc_request **request, u64 valid,
628 enum get_default_layout_type type)
630 struct ll_sb_info *sbi = ll_i2sbi(inode);
631 struct mdt_body *body;
632 struct lov_mds_md *lmm = NULL;
633 struct ptlrpc_request *req = NULL;
635 struct md_op_data *op_data;
639 rc = ll_get_default_mdsize(sbi, &lmm_size);
643 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
644 0, lmm_size, LUSTRE_OPC_ANY,
647 RETURN(PTR_ERR(op_data));
649 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
651 if (type == GET_DEFAULT_LAYOUT_ROOT) {
652 lu_root_fid(&op_data->op_fid1);
653 fid = op_data->op_fid1;
655 fid = *ll_inode2fid(inode);
658 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
659 ll_finish_md_op_data(op_data);
661 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
666 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
667 LASSERT(body != NULL);
669 lmm_size = body->mbo_eadatasize;
671 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
673 GOTO(out, rc = -ENODATA);
676 lmm = req_capsule_server_sized_get(&req->rq_pill,
677 &RMF_MDT_MD, lmm_size);
678 LASSERT(lmm != NULL);
681 * This is coming from the MDS, so is probably in
682 * little endian. We convert it to host endian before
683 * passing it to userspace.
685 /* We don't swab objects for directories */
686 switch (le32_to_cpu(lmm->lmm_magic)) {
689 case LOV_MAGIC_COMP_V1:
690 case LOV_USER_MAGIC_SPECIFIC:
691 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
692 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
695 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
696 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
699 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
700 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
702 case LMV_MAGIC_FOREIGN: {
703 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
705 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
706 __swab32s(&lfm->lfm_magic);
707 __swab32s(&lfm->lfm_length);
708 __swab32s(&lfm->lfm_type);
709 __swab32s(&lfm->lfm_flags);
714 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
719 *plmm_size = lmm_size;
725 * This function will be used to get default LOV/LMV/Default LMV
726 * @valid will be used to indicate which stripe it will retrieve.
727 * If the directory does not have its own default layout, then the
728 * function will request the default layout from root FID.
729 * OBD_MD_MEA LMV stripe EA
730 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
731 * otherwise Default LOV EA.
732 * Each time, it can only retrieve 1 stripe EA
734 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
735 struct ptlrpc_request **request,
736 struct ptlrpc_request **root_request,
739 struct ptlrpc_request *req = NULL;
740 struct ptlrpc_request *root_req = NULL;
741 struct lov_mds_md *lmm = NULL;
746 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
748 if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
749 !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL){
750 int rc2 = ll_dir_get_default_layout(inode, (void **)&lmm,
751 &lmm_size, &root_req, valid,
752 GET_DEFAULT_LAYOUT_ROOT);
758 *plmm_size = lmm_size;
760 if (root_request != NULL)
761 *root_request = root_req;
767 * This function will be used to get default LOV/LMV/Default LMV
768 * @valid will be used to indicate which stripe it will retrieve
769 * OBD_MD_MEA LMV stripe EA
770 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
771 * otherwise Default LOV EA.
772 * Each time, it can only retrieve 1 stripe EA
774 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
775 struct ptlrpc_request **request, u64 valid)
777 struct ptlrpc_request *req = NULL;
778 struct lov_mds_md *lmm = NULL;
783 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
787 *plmm_size = lmm_size;
793 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
795 struct md_op_data *op_data;
800 OBD_ALLOC_PTR(op_data);
804 op_data->op_flags |= MF_GET_MDT_IDX;
805 op_data->op_fid1 = *fid;
806 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
807 mdt_index = op_data->op_mds;
808 OBD_FREE_PTR(op_data);
816 * Get MDT index for the inode.
818 int ll_get_mdt_idx(struct inode *inode)
820 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
824 * Generic handler to do any pre-copy work.
826 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
827 * first information for it that real work has started.
829 * Moreover, for a ARCHIVE request, it will sample the file data version and
830 * store it in \a copy.
832 * \return 0 on success.
834 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
836 struct ll_sb_info *sbi = ll_s2sbi(sb);
837 struct hsm_progress_kernel hpk;
842 /* Forge a hsm_progress based on data from copy. */
843 hpk.hpk_fid = copy->hc_hai.hai_fid;
844 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
845 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
846 hpk.hpk_extent.length = 0;
849 hpk.hpk_data_version = 0;
852 /* For archive request, we need to read the current file version. */
853 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
855 __u64 data_version = 0;
857 /* Get inode for this fid */
858 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
860 hpk.hpk_flags |= HP_FLAG_RETRY;
861 /* hpk_errval is >= 0 */
862 hpk.hpk_errval = -PTR_ERR(inode);
863 GOTO(progress, rc = PTR_ERR(inode));
866 /* Read current file data version */
867 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
870 CDEBUG(D_HSM, "Could not read file data version of "
871 DFID" (rc = %d). Archive request ("
872 "%#llx) could not be done.\n",
873 PFID(©->hc_hai.hai_fid), rc,
874 copy->hc_hai.hai_cookie);
875 hpk.hpk_flags |= HP_FLAG_RETRY;
876 /* hpk_errval must be >= 0 */
877 hpk.hpk_errval = -rc;
881 /* Store in the hsm_copy for later copytool use.
882 * Always modified even if no lsm. */
883 copy->hc_data_version = data_version;
887 /* On error, the request should be considered as completed */
888 if (hpk.hpk_errval > 0)
889 hpk.hpk_flags |= HP_FLAG_COMPLETED;
891 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
894 /* Return first error */
895 RETURN(rc != 0 ? rc : rc2);
899 * Generic handler to do any post-copy work.
901 * It will send the last hsm_progress update to coordinator to inform it
902 * that copy is finished and whether it was successful or not.
905 * - for ARCHIVE request, it will sample the file data version and compare it
906 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
907 * will be considered as failed.
908 * - for RESTORE request, it will sample the file data version and send it to
909 * coordinator which is useful if the file was imported as 'released'.
911 * \return 0 on success.
913 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
915 struct ll_sb_info *sbi = ll_s2sbi(sb);
916 struct hsm_progress_kernel hpk;
921 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
922 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
923 * initialized if copy_end was called with copy == NULL.
926 /* Forge a hsm_progress based on data from copy. */
927 hpk.hpk_fid = copy->hc_hai.hai_fid;
928 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
929 hpk.hpk_extent = copy->hc_hai.hai_extent;
930 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
931 hpk.hpk_errval = copy->hc_errval;
932 hpk.hpk_data_version = 0;
934 /* For archive request, we need to check the file data was not changed.
936 * For restore request, we need to send the file data version, this is
937 * useful when the file was created using hsm_import.
939 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
940 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
941 (copy->hc_errval == 0)) {
943 __u64 data_version = 0;
945 /* Get lsm for this fid */
946 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
948 hpk.hpk_flags |= HP_FLAG_RETRY;
949 /* hpk_errval must be >= 0 */
950 hpk.hpk_errval = -PTR_ERR(inode);
951 GOTO(progress, rc = PTR_ERR(inode));
954 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
957 CDEBUG(D_HSM, "Could not read file data version. "
958 "Request could not be confirmed.\n");
959 if (hpk.hpk_errval == 0)
960 hpk.hpk_errval = -rc;
964 /* Store in the hsm_copy for later copytool use.
965 * Always modified even if no lsm. */
966 hpk.hpk_data_version = data_version;
968 /* File could have been stripped during archiving, so we need
969 * to check anyway. */
970 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
971 (copy->hc_data_version != data_version)) {
972 CDEBUG(D_HSM, "File data version mismatched. "
973 "File content was changed during archiving. "
974 DFID", start:%#llx current:%#llx\n",
975 PFID(©->hc_hai.hai_fid),
976 copy->hc_data_version, data_version);
977 /* File was changed, send error to cdt. Do not ask for
978 * retry because if a file is modified frequently,
979 * the cdt will loop on retried archive requests.
980 * The policy engine will ask for a new archive later
981 * when the file will not be modified for some tunable
983 hpk.hpk_flags &= ~HP_FLAG_RETRY;
985 /* hpk_errval must be >= 0 */
986 hpk.hpk_errval = -rc;
993 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
996 /* Return first error */
997 RETURN(rc != 0 ? rc : rc2);
1001 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1002 const struct lustre_kernelcomm __user *data)
1004 struct lustre_kernelcomm *lk;
1005 struct lustre_kernelcomm *tmp;
1006 size_t size = sizeof(*lk);
1011 /* copy data from userspace to get numbers of archive_id */
1012 OBD_ALLOC(lk, size);
1016 if (copy_from_user(lk, data, size))
1017 GOTO(out_lk, rc = -EFAULT);
1019 if (lk->lk_flags & LK_FLG_STOP)
1022 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1023 __u32 archive_mask = lk->lk_data_count;
1026 /* old hsm agent to old MDS */
1027 if (!exp_connect_archive_id_array(exp))
1030 /* old hsm agent to new MDS */
1031 lk->lk_flags |= LK_FLG_DATANR;
1033 if (archive_mask == 0)
1036 count = hweight32(archive_mask);
1037 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1038 OBD_ALLOC(tmp, new_size);
1040 GOTO(out_lk, rc = -ENOMEM);
1042 memcpy(tmp, lk, size);
1043 tmp->lk_data_count = count;
1049 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1050 if (BIT(i) & archive_mask) {
1051 lk->lk_data[count] = i + 1;
1058 /* new hsm agent to new mds */
1059 if (lk->lk_data_count > 0) {
1060 new_size = offsetof(struct lustre_kernelcomm,
1061 lk_data[lk->lk_data_count]);
1062 OBD_ALLOC(tmp, new_size);
1064 GOTO(out_lk, rc = -ENOMEM);
1070 if (copy_from_user(lk, data, size))
1071 GOTO(out_lk, rc = -EFAULT);
1074 /* new hsm agent to old MDS */
1075 if (!exp_connect_archive_id_array(exp)) {
1078 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1079 GOTO(out_lk, rc = -EINVAL);
1081 for (i = 0; i < lk->lk_data_count; i++) {
1082 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1084 CERROR("%s: archive id %d requested but only "
1085 "[0 - %zu] supported: rc = %d\n",
1086 exp->exp_obd->obd_name, lk->lk_data[i],
1087 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1091 if (lk->lk_data[i] == 0) {
1096 archives |= (1 << (lk->lk_data[i] - 1));
1098 lk->lk_flags &= ~LK_FLG_DATANR;
1099 lk->lk_data_count = archives;
1102 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1108 static int check_owner(int type, int id)
1112 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1116 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1125 int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1127 int cmd = qctl->qc_cmd;
1128 int type = qctl->qc_type;
1129 int id = qctl->qc_id;
1130 int valid = qctl->qc_valid;
1138 case LUSTRE_Q_SETDEFAULT:
1139 case LUSTRE_Q_SETQUOTAPOOL:
1140 case LUSTRE_Q_SETINFOPOOL:
1141 if (!cfs_capable(CAP_SYS_ADMIN))
1145 case LUSTRE_Q_GETDEFAULT:
1146 case LUSTRE_Q_GETQUOTAPOOL:
1147 if (check_owner(type, id) &&
1148 (!cfs_capable(CAP_SYS_ADMIN)))
1152 case LUSTRE_Q_GETINFOPOOL:
1155 CERROR("unsupported quotactl op: %#x\n", cmd);
1159 if (valid != QC_GENERAL) {
1160 if (cmd == Q_GETINFO)
1161 qctl->qc_cmd = Q_GETOINFO;
1162 else if (cmd == Q_GETQUOTA ||
1163 cmd == LUSTRE_Q_GETQUOTAPOOL)
1164 qctl->qc_cmd = Q_GETOQUOTA;
1170 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1171 sizeof(*qctl), qctl, NULL);
1174 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1175 sizeof(*qctl), qctl, NULL);
1178 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1179 sizeof(*qctl), qctl, NULL);
1181 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1183 sizeof(*qctl), qctl, NULL);
1195 struct obd_quotactl *oqctl;
1196 int oqctl_len = sizeof(*oqctl);
1198 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1199 oqctl_len += LOV_MAXPOOLNAME + 1;
1201 OBD_ALLOC(oqctl, oqctl_len);
1205 QCTL_COPY(oqctl, qctl);
1206 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1208 OBD_FREE(oqctl, oqctl_len);
1211 /* If QIF_SPACE is not set, client should collect the
1212 * space usage from OSSs by itself */
1213 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1214 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1215 !oqctl->qc_dqblk.dqb_curspace) {
1216 struct obd_quotactl *oqctl_tmp;
1217 int qctl_len = sizeof(*oqctl_tmp) + LOV_MAXPOOLNAME + 1;
1219 OBD_ALLOC(oqctl_tmp, qctl_len);
1220 if (oqctl_tmp == NULL)
1221 GOTO(out, rc = -ENOMEM);
1223 if (cmd == LUSTRE_Q_GETQUOTAPOOL) {
1224 oqctl_tmp->qc_cmd = LUSTRE_Q_GETQUOTAPOOL;
1225 memcpy(oqctl_tmp->qc_poolname,
1227 LOV_MAXPOOLNAME + 1);
1229 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1231 oqctl_tmp->qc_id = oqctl->qc_id;
1232 oqctl_tmp->qc_type = oqctl->qc_type;
1234 /* collect space usage from OSTs */
1235 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1236 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1237 if (!rc || rc == -EREMOTEIO) {
1238 oqctl->qc_dqblk.dqb_curspace =
1239 oqctl_tmp->qc_dqblk.dqb_curspace;
1240 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1243 /* collect space & inode usage from MDTs */
1244 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1245 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1246 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1247 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1248 if (!rc || rc == -EREMOTEIO) {
1249 oqctl->qc_dqblk.dqb_curspace +=
1250 oqctl_tmp->qc_dqblk.dqb_curspace;
1251 oqctl->qc_dqblk.dqb_curinodes =
1252 oqctl_tmp->qc_dqblk.dqb_curinodes;
1253 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1255 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1258 OBD_FREE(oqctl_tmp, qctl_len);
1261 QCTL_COPY(qctl, oqctl);
1262 OBD_FREE(oqctl, oqctl_len);
1268 int ll_rmfid(struct file *file, void __user *arg)
1270 const struct fid_array __user *ufa = arg;
1271 struct fid_array *lfa = NULL;
1274 int i, rc, *rcs = NULL;
1277 if (!cfs_capable(CAP_DAC_READ_SEARCH) &&
1278 !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1280 /* Only need to get the buflen */
1281 if (get_user(nr, &ufa->fa_nr))
1283 /* DoS protection */
1284 if (nr > OBD_MAX_FIDS_IN_ARRAY)
1287 size = offsetof(struct fid_array, fa_fids[nr]);
1288 OBD_ALLOC(lfa, size);
1291 OBD_ALLOC_PTR_ARRAY(rcs, nr);
1293 GOTO(free_lfa, rc = -ENOMEM);
1295 if (copy_from_user(lfa, arg, size))
1296 GOTO(free_rcs, rc = -EFAULT);
1298 /* Call mdc_iocontrol */
1299 rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1301 for (i = 0; i < nr; i++)
1303 lfa->fa_fids[i].f_ver = rcs[i];
1304 if (copy_to_user(arg, lfa, size))
1309 OBD_FREE_PTR_ARRAY(rcs, nr);
1311 OBD_FREE(lfa, size);
1316 /* This function tries to get a single name component,
1317 * to send to the server. No actual path traversal involved,
1318 * so we limit to NAME_MAX */
1319 static char *ll_getname(const char __user *filename)
1324 OBD_ALLOC(tmp, NAME_MAX + 1);
1327 return ERR_PTR(-ENOMEM);
1329 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1332 else if (len > NAME_MAX)
1333 ret = -ENAMETOOLONG;
1336 OBD_FREE(tmp, NAME_MAX + 1);
1342 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1344 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1346 struct dentry *dentry = file_dentry(file);
1347 struct inode *inode = file_inode(file);
1348 struct ll_sb_info *sbi = ll_i2sbi(inode);
1349 struct obd_ioctl_data *data = NULL;
1353 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1354 PFID(ll_inode2fid(inode)), inode, cmd);
1356 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1357 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1360 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1362 case FS_IOC_GETFLAGS:
1363 case FS_IOC_SETFLAGS:
1364 RETURN(ll_iocontrol(inode, file, cmd, arg));
1365 case FSFILT_IOC_GETVERSION:
1366 case FS_IOC_GETVERSION:
1367 RETURN(put_user(inode->i_generation, (int __user *)arg));
1368 /* We need to special case any other ioctls we want to handle,
1369 * to send them to the MDS/OST as appropriate and to properly
1370 * network encode the arg field. */
1371 case FS_IOC_SETVERSION:
1374 case LL_IOC_GET_MDTIDX: {
1377 mdtidx = ll_get_mdt_idx(inode);
1381 if (put_user((int)mdtidx, (int __user *)arg))
1386 case IOC_MDC_LOOKUP: {
1387 int namelen, len = 0;
1390 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
1394 filename = data->ioc_inlbuf1;
1395 namelen = strlen(filename);
1397 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1398 GOTO(out_free, rc = -EINVAL);
1401 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1403 CERROR("%s: lookup %.*s failed: rc = %d\n",
1404 sbi->ll_fsname, namelen, filename, rc);
1408 OBD_FREE_LARGE(data, len);
1411 case LL_IOC_LMV_SETSTRIPE: {
1412 struct lmv_user_md *lum;
1420 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
1424 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1425 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1426 GOTO(lmv_out_free, rc = -EINVAL);
1428 filename = data->ioc_inlbuf1;
1429 namelen = data->ioc_inllen1;
1432 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1433 GOTO(lmv_out_free, rc = -EINVAL);
1435 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1436 lumlen = data->ioc_inllen2;
1438 if (!lmv_user_magic_supported(lum->lum_magic)) {
1439 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1440 lum->lum_magic, -EINVAL);
1441 GOTO(lmv_out_free, rc = -EINVAL);
1444 if ((lum->lum_magic == LMV_USER_MAGIC ||
1445 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1446 lumlen < sizeof(*lum)) {
1447 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1448 filename, lumlen, lum->lum_magic, -EINVAL);
1449 GOTO(lmv_out_free, rc = -EINVAL);
1452 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1453 lumlen < sizeof(struct lmv_foreign_md)) {
1454 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1455 filename, lum->lum_magic, lumlen, -EFAULT);
1456 GOTO(lmv_out_free, rc = -EINVAL);
1459 mode = data->ioc_type;
1460 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1462 OBD_FREE_LARGE(data, len);
1466 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1467 struct lmv_user_md lum;
1468 struct lmv_user_md __user *ulump =
1469 (struct lmv_user_md __user *)arg;
1472 if (copy_from_user(&lum, ulump, sizeof(lum)))
1475 if (lum.lum_magic != LMV_USER_MAGIC)
1478 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1482 case LL_IOC_LOV_SETSTRIPE_NEW:
1483 case LL_IOC_LOV_SETSTRIPE: {
1484 struct lov_user_md_v3 *lumv3 = NULL;
1485 struct lov_user_md_v1 lumv1;
1486 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1487 struct lov_user_md_v1 __user *lumv1p =
1488 (struct lov_user_md_v1 __user *)arg;
1489 struct lov_user_md_v3 __user *lumv3p =
1490 (struct lov_user_md_v3 __user *)arg;
1493 int set_default = 0;
1495 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1496 sizeof(struct lov_comp_md_v1));
1497 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1498 /* first try with v1 which is smaller than v3 */
1499 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1502 if (is_root_inode(inode))
1505 switch (lumv1.lmm_magic) {
1506 case LOV_USER_MAGIC_V3:
1507 case LOV_USER_MAGIC_SPECIFIC:
1508 lum_size = ll_lov_user_md_size(&lumv1);
1511 OBD_ALLOC(lumv3, lum_size);
1514 if (copy_from_user(lumv3, lumv3p, lum_size))
1515 GOTO(out, rc = -EFAULT);
1516 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1518 case LOV_USER_MAGIC_V1:
1521 GOTO(out, rc = -ENOTSUPP);
1524 /* in v1 and v3 cases lumv1 points to data */
1525 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1528 OBD_FREE(lumv3, lum_size);
1531 case LL_IOC_LMV_GETSTRIPE: {
1532 struct lmv_user_md __user *ulmv =
1533 (struct lmv_user_md __user *)arg;
1534 struct lmv_user_md lum;
1535 struct ptlrpc_request *request = NULL;
1536 struct ptlrpc_request *root_request = NULL;
1537 union lmv_mds_md *lmm = NULL;
1540 struct lmv_user_md *tmp = NULL;
1544 int max_stripe_count;
1548 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1551 max_stripe_count = lum.lum_stripe_count;
1552 /* lum_magic will indicate which stripe the ioctl will like
1553 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1554 * is for default LMV stripe */
1555 if (lum.lum_magic == LMV_MAGIC_V1)
1556 valid |= OBD_MD_MEA;
1557 else if (lum.lum_magic == LMV_USER_MAGIC)
1558 valid |= OBD_MD_DEFAULT_MEA;
1562 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1563 &request, &root_request, valid);
1565 GOTO(finish_req, rc);
1567 /* Get default LMV EA */
1568 if (lum.lum_magic == LMV_USER_MAGIC) {
1569 if (lmmsize > sizeof(*ulmv))
1570 GOTO(finish_req, rc = -EINVAL);
1572 if (copy_to_user(ulmv, lmm, lmmsize))
1573 GOTO(finish_req, rc = -EFAULT);
1575 GOTO(finish_req, rc);
1578 /* if foreign LMV case, fake stripes number */
1579 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1580 struct lmv_foreign_md *lfm;
1582 lfm = (struct lmv_foreign_md *)lmm;
1583 if (lfm->lfm_length < XATTR_SIZE_MAX -
1584 offsetof(typeof(*lfm), lfm_value)) {
1585 __u32 size = lfm->lfm_length +
1586 offsetof(typeof(*lfm), lfm_value);
1588 stripe_count = lmv_foreign_to_md_stripes(size);
1590 CERROR("invalid %d foreign size returned\n",
1595 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1597 if (max_stripe_count < stripe_count) {
1598 lum.lum_stripe_count = stripe_count;
1599 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1600 GOTO(finish_req, rc = -EFAULT);
1601 GOTO(finish_req, rc = -E2BIG);
1604 /* enough room on user side and foreign case */
1605 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1606 struct lmv_foreign_md *lfm;
1609 lfm = (struct lmv_foreign_md *)lmm;
1610 size = lfm->lfm_length +
1611 offsetof(struct lmv_foreign_md, lfm_value);
1612 if (copy_to_user(ulmv, lfm, size))
1613 GOTO(finish_req, rc = -EFAULT);
1614 GOTO(finish_req, rc);
1617 lum_size = lmv_user_md_size(stripe_count,
1618 LMV_USER_MAGIC_SPECIFIC);
1619 OBD_ALLOC(tmp, lum_size);
1621 GOTO(finish_req, rc = -ENOMEM);
1623 mdt_index = ll_get_mdt_idx(inode);
1625 GOTO(out_tmp, rc = -ENOMEM);
1627 tmp->lum_magic = LMV_MAGIC_V1;
1628 tmp->lum_stripe_count = 0;
1629 tmp->lum_stripe_offset = mdt_index;
1630 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1631 for (i = 0; i < stripe_count; i++) {
1634 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1635 if (fid_is_sane(&fid)) {
1636 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1638 GOTO(out_tmp, rc = mdt_index);
1640 tmp->lum_objects[i].lum_mds = mdt_index;
1641 tmp->lum_objects[i].lum_fid = fid;
1644 tmp->lum_stripe_count++;
1647 if (copy_to_user(ulmv, tmp, lum_size))
1648 GOTO(out_tmp, rc = -EFAULT);
1650 OBD_FREE(tmp, lum_size);
1652 ptlrpc_req_finished(request);
1653 ptlrpc_req_finished(root_request);
1657 case LL_IOC_UNLOCK_FOREIGN:
1658 /* if not a foreign symlink do nothing */
1659 if (ll_foreign_is_removable(dentry, true)) {
1661 "prevent rmdir of non-foreign dir ("DFID")\n",
1662 PFID(ll_inode2fid(inode)));
1663 RETURN(-EOPNOTSUPP);
1667 case LL_IOC_REMOVE_ENTRY: {
1668 char *filename = NULL;
1672 /* Here is a little hack to avoid sending REINT_RMENTRY to
1673 * unsupported server, which might crash the server(LU-2730),
1674 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1675 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1676 * server will support REINT_RMENTRY XXX*/
1677 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1680 filename = ll_getname((const char __user *)arg);
1681 if (IS_ERR(filename))
1682 RETURN(PTR_ERR(filename));
1684 namelen = strlen(filename);
1686 GOTO(out_rmdir, rc = -EINVAL);
1688 rc = ll_rmdir_entry(inode, filename, namelen);
1691 ll_putname(filename);
1695 RETURN(ll_rmfid(file, (void __user *)arg));
1696 case LL_IOC_LOV_SWAP_LAYOUTS:
1698 case IOC_OBD_STATFS:
1699 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1700 case LL_IOC_LOV_GETSTRIPE:
1701 case LL_IOC_LOV_GETSTRIPE_NEW:
1702 case LL_IOC_MDC_GETINFO_V1:
1703 case LL_IOC_MDC_GETINFO_V2:
1704 case IOC_MDC_GETFILEINFO_V1:
1705 case IOC_MDC_GETFILEINFO_V2:
1706 case IOC_MDC_GETFILESTRIPE: {
1707 struct ptlrpc_request *request = NULL;
1708 struct ptlrpc_request *root_request = NULL;
1709 struct lov_user_md __user *lump;
1710 struct lov_mds_md *lmm = NULL;
1711 struct mdt_body *body;
1712 char *filename = NULL;
1713 lstat_t __user *statp = NULL;
1714 lstatx_t __user *stxp = NULL;
1715 __u64 __user *flagsp = NULL;
1716 __u32 __user *lmmsizep = NULL;
1717 struct lu_fid __user *fidp = NULL;
1720 if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1721 cmd == IOC_MDC_GETFILEINFO_V2 ||
1722 cmd == IOC_MDC_GETFILESTRIPE) {
1723 filename = ll_getname((const char __user *)arg);
1724 if (IS_ERR(filename))
1725 RETURN(PTR_ERR(filename));
1727 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1728 &lmmsize, &request);
1730 rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1736 body = req_capsule_server_get(&request->rq_pill,
1738 LASSERT(body != NULL);
1743 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO_V1 ||
1744 cmd == LL_IOC_MDC_GETINFO_V1 ||
1745 cmd == IOC_MDC_GETFILEINFO_V2 ||
1746 cmd == LL_IOC_MDC_GETINFO_V2)) {
1754 if (cmd == IOC_MDC_GETFILESTRIPE ||
1755 cmd == LL_IOC_LOV_GETSTRIPE ||
1756 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1757 lump = (struct lov_user_md __user *)arg;
1758 } else if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1759 cmd == LL_IOC_MDC_GETINFO_V1){
1760 struct lov_user_mds_data_v1 __user *lmdp;
1762 lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1763 statp = &lmdp->lmd_st;
1764 lump = &lmdp->lmd_lmm;
1766 struct lov_user_mds_data __user *lmdp;
1768 lmdp = (struct lov_user_mds_data __user *)arg;
1769 fidp = &lmdp->lmd_fid;
1770 stxp = &lmdp->lmd_stx;
1771 flagsp = &lmdp->lmd_flags;
1772 lmmsizep = &lmdp->lmd_lmmsize;
1773 lump = &lmdp->lmd_lmm;
1777 /* If the file has no striping then zero out *lump so
1778 * that the caller isn't confused by garbage. */
1779 if (clear_user(lump, sizeof(*lump)))
1780 GOTO(out_req, rc = -EFAULT);
1781 } else if (copy_to_user(lump, lmm, lmmsize)) {
1782 if (copy_to_user(lump, lmm, sizeof(*lump)))
1783 GOTO(out_req, rc = -EFAULT);
1787 if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1788 cmd == LL_IOC_MDC_GETINFO_V1) {
1791 st.st_dev = inode->i_sb->s_dev;
1792 st.st_mode = body->mbo_mode;
1793 st.st_nlink = body->mbo_nlink;
1794 st.st_uid = body->mbo_uid;
1795 st.st_gid = body->mbo_gid;
1796 st.st_rdev = body->mbo_rdev;
1797 st.st_size = body->mbo_size;
1798 st.st_blksize = PAGE_SIZE;
1799 st.st_blocks = body->mbo_blocks;
1800 st.st_atime = body->mbo_atime;
1801 st.st_mtime = body->mbo_mtime;
1802 st.st_ctime = body->mbo_ctime;
1803 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1807 if (copy_to_user(statp, &st, sizeof(st)))
1808 GOTO(out_req, rc = -EFAULT);
1809 } else if (cmd == IOC_MDC_GETFILEINFO_V2 ||
1810 cmd == LL_IOC_MDC_GETINFO_V2) {
1811 lstatx_t stx = { 0 };
1812 __u64 valid = body->mbo_valid;
1814 stx.stx_blksize = PAGE_SIZE;
1815 stx.stx_nlink = body->mbo_nlink;
1816 stx.stx_uid = body->mbo_uid;
1817 stx.stx_gid = body->mbo_gid;
1818 stx.stx_mode = body->mbo_mode;
1819 stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1822 stx.stx_size = body->mbo_size;
1823 stx.stx_blocks = body->mbo_blocks;
1824 stx.stx_atime.tv_sec = body->mbo_atime;
1825 stx.stx_ctime.tv_sec = body->mbo_ctime;
1826 stx.stx_mtime.tv_sec = body->mbo_mtime;
1827 stx.stx_btime.tv_sec = body->mbo_btime;
1828 stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1829 stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1830 stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1831 stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1832 stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1835 * For a striped directory, the size and blocks returned
1836 * from MDT is not correct.
1837 * The size and blocks are aggregated by client across
1839 * Thus for a striped directory, do not return the valid
1840 * FLSIZE and FLBLOCKS flags to the caller.
1841 * However, this whould be better decided by the MDS
1842 * instead of the client.
1844 if (cmd == LL_IOC_MDC_GETINFO_V2 &&
1845 ll_i2info(inode)->lli_lsm_md != NULL)
1846 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1848 if (flagsp && copy_to_user(flagsp, &valid,
1850 GOTO(out_req, rc = -EFAULT);
1852 if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1854 GOTO(out_req, rc = -EFAULT);
1856 if (!(valid & OBD_MD_FLSIZE))
1857 stx.stx_mask &= ~STATX_SIZE;
1858 if (!(valid & OBD_MD_FLBLOCKS))
1859 stx.stx_mask &= ~STATX_BLOCKS;
1861 if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1862 GOTO(out_req, rc = -EFAULT);
1864 if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1866 GOTO(out_req, rc = -EFAULT);
1871 ptlrpc_req_finished(request);
1872 ptlrpc_req_finished(root_request);
1874 ll_putname(filename);
1877 case OBD_IOC_QUOTACTL: {
1878 struct if_quotactl *qctl;
1879 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
1881 OBD_ALLOC(qctl, qctl_len);
1885 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1886 GOTO(out_quotactl, rc = -EFAULT);
1888 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
1889 char __user *from = (char __user *)arg +
1890 offsetof(typeof(*qctl), qc_poolname);
1891 if (copy_from_user(qctl->qc_poolname, from,
1892 LOV_MAXPOOLNAME + 1))
1893 GOTO(out_quotactl, rc = -EFAULT);
1896 rc = quotactl_ioctl(sbi, qctl);
1898 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1902 OBD_FREE(qctl, qctl_len);
1905 case OBD_IOC_GETDTNAME:
1906 case OBD_IOC_GETMDNAME:
1907 RETURN(ll_get_obd_name(inode, cmd, arg));
1908 case LL_IOC_FLUSHCTX:
1909 RETURN(ll_flush_ctx(inode));
1910 case LL_IOC_GETOBDCOUNT: {
1912 struct obd_export *exp;
1914 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1917 /* get ost count when count is zero, get mdt count otherwise */
1918 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1919 vallen = sizeof(count);
1920 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1921 KEY_TGT_COUNT, &vallen, &count);
1923 CERROR("get target count failed: %d\n", rc);
1927 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1932 case LL_IOC_PATH2FID:
1933 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1934 sizeof(struct lu_fid)))
1937 case LL_IOC_GET_CONNECT_FLAGS: {
1938 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1939 (void __user *)arg));
1941 case OBD_IOC_FID2PATH:
1942 RETURN(ll_fid2path(inode, (void __user *)arg));
1943 case LL_IOC_GETPARENT:
1944 RETURN(ll_getparent(file, (void __user *)arg));
1945 case LL_IOC_FID2MDTIDX: {
1946 struct obd_export *exp = ll_i2mdexp(inode);
1950 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1954 /* Call mdc_iocontrol */
1955 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1956 (__u32 __user *)&index);
1962 case LL_IOC_HSM_REQUEST: {
1963 struct hsm_user_request *hur;
1970 /* We don't know the true size yet; copy the fixed-size part */
1971 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1976 /* Compute the whole struct size */
1977 totalsize = hur_len(hur);
1982 /* Final size will be more than double totalsize */
1983 if (totalsize >= MDS_MAXREQSIZE / 3)
1986 OBD_ALLOC_LARGE(hur, totalsize);
1990 /* Copy the whole struct */
1991 if (copy_from_user(hur, (void __user *)arg, totalsize))
1992 GOTO(out_hur, rc = -EFAULT);
1994 if (hur->hur_request.hr_action == HUA_RELEASE) {
1995 const struct lu_fid *fid;
1999 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
2000 fid = &hur->hur_user_item[i].hui_fid;
2001 f = search_inode_for_lustre(inode->i_sb, fid);
2007 rc = ll_hsm_release(f);
2013 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
2018 OBD_FREE_LARGE(hur, totalsize);
2022 case LL_IOC_HSM_PROGRESS: {
2023 struct hsm_progress_kernel hpk;
2024 struct hsm_progress hp;
2026 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2029 hpk.hpk_fid = hp.hp_fid;
2030 hpk.hpk_cookie = hp.hp_cookie;
2031 hpk.hpk_extent = hp.hp_extent;
2032 hpk.hpk_flags = hp.hp_flags;
2033 hpk.hpk_errval = hp.hp_errval;
2034 hpk.hpk_data_version = 0;
2036 /* File may not exist in Lustre; all progress
2037 * reported to Lustre root */
2038 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2042 case LL_IOC_HSM_CT_START:
2043 if (!cfs_capable(CAP_SYS_ADMIN))
2046 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2047 (struct lustre_kernelcomm __user *)arg);
2050 case LL_IOC_HSM_COPY_START: {
2051 struct hsm_copy *copy;
2054 OBD_ALLOC_PTR(copy);
2057 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2062 rc = ll_ioc_copy_start(inode->i_sb, copy);
2063 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2069 case LL_IOC_HSM_COPY_END: {
2070 struct hsm_copy *copy;
2073 OBD_ALLOC_PTR(copy);
2076 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2081 rc = ll_ioc_copy_end(inode->i_sb, copy);
2082 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2088 case LL_IOC_MIGRATE: {
2089 struct lmv_user_md *lum;
2095 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
2099 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2100 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2101 GOTO(migrate_free, rc = -EINVAL);
2103 filename = data->ioc_inlbuf1;
2104 namelen = data->ioc_inllen1;
2106 if (namelen < 1 || namelen != strlen(filename) + 1) {
2107 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2108 GOTO(migrate_free, rc = -EINVAL);
2111 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2112 if (lum->lum_magic != LMV_USER_MAGIC &&
2113 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2115 CERROR("%s: wrong lum magic %x: rc = %d\n",
2116 filename, lum->lum_magic, rc);
2117 GOTO(migrate_free, rc);
2120 rc = ll_migrate(inode, file, lum, filename);
2122 OBD_FREE_LARGE(data, len);
2126 case FS_IOC_FSGETXATTR:
2127 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2128 case FS_IOC_FSSETXATTR:
2129 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2130 case LL_IOC_PCC_DETACH_BY_FID: {
2131 struct lu_pcc_detach_fid *detach;
2133 struct inode *inode2;
2136 OBD_ALLOC_PTR(detach);
2140 if (copy_from_user(detach,
2141 (const struct lu_pcc_detach_fid __user *)arg,
2143 GOTO(out_detach, rc = -EFAULT);
2145 fid = &detach->pccd_fid;
2146 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2147 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2149 /* Target inode is not in inode cache, and PCC file
2150 * has aleady released, return immdiately.
2152 GOTO(out_detach, rc = 0);
2154 if (!S_ISREG(inode2->i_mode))
2155 GOTO(out_iput, rc = -EINVAL);
2157 if (!inode_owner_or_capable(inode2))
2158 GOTO(out_iput, rc = -EPERM);
2160 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2164 OBD_FREE_PTR(detach);
2167 #ifdef HAVE_LUSTRE_CRYPTO
2168 case LL_IOC_SET_ENCRYPTION_POLICY:
2169 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2171 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
2172 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
2173 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2175 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
2176 case LL_IOC_ADD_ENCRYPTION_KEY:
2177 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2179 return llcrypt_ioctl_add_key(file, (void __user *)arg);
2180 case LL_IOC_REMOVE_ENCRYPTION_KEY:
2181 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2183 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
2184 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
2185 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2187 return llcrypt_ioctl_remove_key_all_users(file,
2188 (void __user *)arg);
2189 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
2190 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2192 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
2195 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2196 (void __user *)arg));
2200 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2202 struct inode *inode = file->f_mapping->host;
2203 struct ll_file_data *fd = file->private_data;
2204 struct ll_sb_info *sbi = ll_i2sbi(inode);
2205 int api32 = ll_need_32bit_api(sbi);
2206 loff_t ret = -EINVAL;
2214 offset += file->f_pos;
2220 offset += LL_DIR_END_OFF_32BIT;
2222 offset += LL_DIR_END_OFF;
2229 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2230 (!api32 && offset <= LL_DIR_END_OFF))) {
2231 if (offset != file->f_pos) {
2232 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2233 (!api32 && offset == LL_DIR_END_OFF))
2234 fd->lfd_pos = MDS_DIR_END_OFF;
2235 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2236 fd->lfd_pos = offset << 32;
2238 fd->lfd_pos = offset;
2239 file->f_pos = offset;
2240 file->f_version = 0;
2247 inode_unlock(inode);
2251 static int ll_dir_open(struct inode *inode, struct file *file)
2254 RETURN(ll_file_open(inode, file));
2257 static int ll_dir_release(struct inode *inode, struct file *file)
2260 RETURN(ll_file_release(inode, file));
2263 const struct file_operations ll_dir_operations = {
2264 .llseek = ll_dir_seek,
2265 .open = ll_dir_open,
2266 .release = ll_dir_release,
2267 .read = generic_read_dir,
2268 #ifdef HAVE_DIR_CONTEXT
2269 .iterate_shared = ll_iterate,
2271 .readdir = ll_readdir,
2273 .unlocked_ioctl = ll_dir_ioctl,