1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef AUTOCONF_INCLUDED
38 #include <linux/config.h>
40 #include <linux/kernel.h>
42 #include <linux/string.h>
43 #include <linux/stat.h>
44 #include <linux/errno.h>
45 #include <linux/smp_lock.h>
46 #include <linux/unistd.h>
47 #include <linux/version.h>
48 #include <asm/system.h>
49 #include <asm/uaccess.h>
52 #include <linux/stat.h>
53 #include <asm/uaccess.h>
55 #include <linux/pagemap.h>
56 #include <linux/smp_lock.h>
58 #define DEBUG_SUBSYSTEM S_LLITE
60 //#include <lustre_mdc.h>
61 #include <lustre_lite.h>
62 #include "llite_internal.h"
63 #include <linux/lustre_compat25.h>
65 #define VMA_DEBUG(vma, fmt, arg...) \
66 CDEBUG(D_MMAP, "vma(%p) start(%ld) end(%ld) pgoff(%ld) inode(%p) " \
67 "ino(%lu) iname(%s): " fmt, vma, vma->vm_start, vma->vm_end, \
68 vma->vm_pgoff, vma->vm_file->f_dentry->d_inode, \
69 vma->vm_file->f_dentry->d_inode->i_ino, \
70 vma->vm_file->f_dentry->d_iname, ## arg); \
72 struct page *ll_nopage(struct vm_area_struct *vma, unsigned long address,
75 void policy_from_vma(ldlm_policy_data_t *policy,
76 struct vm_area_struct *vma, unsigned long addr,
79 policy->l_extent.start = ((addr - vma->vm_start) & CFS_PAGE_MASK) +
80 (vma->vm_pgoff << CFS_PAGE_SHIFT);
81 policy->l_extent.end = (policy->l_extent.start + count - 1) |
85 struct vm_area_struct * our_vma(unsigned long addr, size_t count)
87 struct mm_struct *mm = current->mm;
88 struct vm_area_struct *vma, *ret = NULL;
91 /* No MM (e.g. NFS)? No vmas too. */
95 spin_lock(&mm->page_table_lock);
96 for(vma = find_vma(mm, addr);
97 vma != NULL && vma->vm_start < (addr + count); vma = vma->vm_next) {
98 if (vma->vm_ops && vma->vm_ops->nopage == ll_nopage &&
99 vma->vm_flags & VM_SHARED) {
104 spin_unlock(&mm->page_table_lock);
109 * Lustre implementation of a vm_operations_struct::nopage() method, called by
110 * VM to server page fault (both in kernel and user space).
112 * This function sets up CIT_FAULT cl_io that does the job.
114 * \param vma - is virtiual area struct related to page fault
115 * \param address - address when hit fault
116 * \param type - of fault
118 * XXX newer 2.6 kernels provide vm_operations_struct::fault() method with
119 * slightly different semantics instead.
121 * \return allocated and filled page for address
122 * \retval NOPAGE_SIGBUS if page not exist on this address
123 * \retval NOPAGE_OOM not have memory for allocate new page
125 struct page *ll_nopage(struct vm_area_struct *vma, unsigned long address,
128 struct file *file = vma->vm_file;
129 struct inode *inode = file->f_dentry->d_inode;
132 struct page *page = NULL;
133 struct cl_env_nest nest;
138 if (ll_file_nolock(file))
139 RETURN(ERR_PTR(-EOPNOTSUPP));
142 * vm_operations_struct::nopage() can be called when lustre IO is
143 * already active for the current thread, e.g., when doing read/write
144 * against user level buffer mapped from Lustre buffer. To avoid
145 * stomping on existing context, optionally force an allocation of a new
148 env = cl_env_nested_get(&nest);
151 const unsigned long writable = VM_SHARED|VM_WRITE;
152 unsigned long ra_flags;
153 struct cl_fault_io *fio;
155 io = &ccc_env_info(env)->cti_io;
156 memset(io, 0, sizeof(*io));
157 io->ci_obj = ll_i2info(inode)->lli_clob;
158 LASSERT(io->ci_obj != NULL);
160 fio = &io->u.ci_fault;
161 pg_offset = (address - vma->vm_start) >> PAGE_SHIFT;
162 fio->ft_index = pg_offset + vma->vm_pgoff;
163 fio->ft_writable = (vma->vm_flags&writable) == writable;
164 fio->ft_executable = vma->vm_flags&VM_EXEC;
167 * disable VM_SEQ_READ and use VM_RAND_READ to make sure that
168 * the kernel will not read other pages not covered by ldlm in
169 * filemap_nopage. we do our readahead in ll_readpage.
171 ra_flags = vma->vm_flags & (VM_RAND_READ|VM_SEQ_READ);
172 vma->vm_flags &= ~VM_SEQ_READ;
173 vma->vm_flags |= VM_RAND_READ;
175 CDEBUG(D_INFO, "vm_flags: %lx (%lu %i %i)\n", vma->vm_flags,
176 fio->ft_index, fio->ft_writable, fio->ft_executable);
178 if (cl_io_init(env, io, CIT_FAULT, io->ci_obj) == 0) {
179 struct vvp_io *vio = vvp_env_io(env);
180 struct ccc_io *cio = ccc_env_io(env);
181 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
183 LASSERT(cio->cui_cl.cis_io == io);
185 /* mmap lock must be MANDATORY. */
186 io->ci_lockreq = CILR_MANDATORY;
187 vio->u.fault.ft_vma = vma;
188 vio->u.fault.ft_address = address;
189 vio->u.fault.ft_type = type;
192 result = cl_io_loop(env, io);
194 LASSERT(fio->ft_page != NULL);
195 page = cl_page_vmpage(env, fio->ft_page);
196 } else if (result == -EFAULT) {
197 page = NOPAGE_SIGBUS;
198 } else if (result == -ENOMEM) {
202 result = io->ci_result;
204 vma->vm_flags &= ~VM_RAND_READ;
205 vma->vm_flags |= ra_flags;
208 cl_env_nested_put(&nest, env);
214 * To avoid cancel the locks covering mmapped region for lock cache pressure,
215 * we track the mapped vma count in ccc_object::cob_mmap_cnt.
217 static void ll_vm_open(struct vm_area_struct * vma)
219 struct inode *inode = vma->vm_file->f_dentry->d_inode;
220 struct ccc_object *vob = cl_inode2ccc(inode);
223 LASSERT(vma->vm_file);
224 LASSERT(atomic_read(&vob->cob_mmap_cnt) >= 0);
225 atomic_inc(&vob->cob_mmap_cnt);
230 * Dual to ll_vm_open().
232 static void ll_vm_close(struct vm_area_struct *vma)
234 struct inode *inode = vma->vm_file->f_dentry->d_inode;
235 struct ccc_object *vob = cl_inode2ccc(inode);
238 LASSERT(vma->vm_file);
239 atomic_dec(&vob->cob_mmap_cnt);
240 LASSERT(atomic_read(&vob->cob_mmap_cnt) >= 0);
244 #ifndef HAVE_FILEMAP_POPULATE
245 static int (*filemap_populate)(struct vm_area_struct * area, unsigned long address, unsigned long len, pgprot_t prot, unsigned long pgoff, int nonblock);
247 static int ll_populate(struct vm_area_struct *area, unsigned long address,
248 unsigned long len, pgprot_t prot, unsigned long pgoff,
254 /* always set nonblock as true to avoid page read ahead */
255 rc = filemap_populate(area, address, len, prot, pgoff, 1);
259 /* return the user space pointer that maps to a file offset via a vma */
260 static inline unsigned long file_to_user(struct vm_area_struct *vma, __u64 byte)
262 return vma->vm_start + (byte - ((__u64)vma->vm_pgoff << CFS_PAGE_SHIFT));
266 /* XXX put nice comment here. talk about __free_pte -> dirty pages and
267 * nopage's reference passing to the pte */
268 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last)
273 LASSERTF(last > first, "last "LPU64" first "LPU64"\n", last, first);
274 if (mapping_mapped(mapping)) {
276 unmap_mapping_range(mapping, first + CFS_PAGE_SIZE - 1,
277 last - first + 1, 0);
283 static struct vm_operations_struct ll_file_vm_ops = {
286 .close = ll_vm_close,
287 .populate = ll_populate,
290 int ll_file_mmap(struct file *file, struct vm_area_struct * vma)
292 struct inode *inode = file->f_dentry->d_inode;
296 if (ll_file_nolock(file))
299 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_MAP, 1);
300 rc = generic_file_mmap(file, vma);
302 #if !defined(HAVE_FILEMAP_POPULATE)
303 if (!filemap_populate)
304 filemap_populate = vma->vm_ops->populate;
306 vma->vm_ops = &ll_file_vm_ops;
307 vma->vm_ops->open(vma);
308 /* update the inode's size and mtime */
309 rc = cl_glimpse_size(inode);