lustre/llite/rw26.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Lite I/O page cache routines for the 2.5/2.6 kernel version
   5  *
   6  *  Copyright (c) 2001-2003 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #include <linux/config.h>
  25 #include <linux/kernel.h>
  26 #include <linux/mm.h>
  27 #include <linux/string.h>
  28 #include <linux/stat.h>
  29 #include <linux/errno.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/unistd.h>
  32 #include <linux/version.h>
  33 #include <asm/system.h>
  34 #include <asm/uaccess.h>
  35
  36 #include <linux/fs.h>
  37 #include <linux/buffer_head.h>
  38 #include <linux/mpage.h>
  39 #include <linux/writeback.h>
  40 #include <linux/stat.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/segment.h>
  43 #include <linux/mm.h>
  44 #include <linux/pagemap.h>
  45 #include <linux/smp_lock.h>
  46
  47 #define DEBUG_SUBSYSTEM S_LLITE
  48
  49 #include <lustre_mdc.h>
  50 #include <lustre_lite.h>
  51 #include "llite_internal.h"
  52 #include <linux/lustre_compat25.h>
  53
  54 static int ll_writepage_26(struct page *page, struct writeback_control *wbc)
  55 {
  56         return ll_writepage(page);
  57 }
  58
  59 /* It is safe to not check anything in invalidatepage/releasepage below
  60    because they are run with page locked and all our io is happening with
  61    locked page too */
  62 static int ll_invalidatepage(struct page *page, unsigned long offset)
  63 {
  64         if (offset)
  65                 return 0;
  66         if (PagePrivate(page))
  67                 ll_removepage(page);
  68         return 1;
  69 }
  70
  71 static int ll_releasepage(struct page *page, gfp_t gfp_mask)
  72 {
  73         if (PagePrivate(page))
  74                 ll_removepage(page);
  75         return 1;
  76 }
  77
  78 static int ll_set_page_dirty(struct page *page)
  79 {
  80         struct ll_async_page *llap;
  81         ENTRY;
  82
  83         llap = llap_from_page(page, LLAP_ORIGIN_UNKNOWN);
  84         if (IS_ERR(llap))
  85                 RETURN(PTR_ERR(llap));
  86
  87         llap_write_pending(page->mapping->host, llap);
  88         RETURN(__set_page_dirty_nobuffers(page));
  89 }
  90
  91 #define MAX_DIRECTIO_SIZE 2*1024*1024*1024UL
  92
  93 static inline int ll_get_user_pages(int rw, unsigned long user_addr,
  94                                     size_t size, struct page ***pages)
  95 {
  96         int result = -ENOMEM;
  97         unsigned long page_count;
  98
  99         /* set an arbitrary limit to prevent arithmetic overflow */
 100         if (size > MAX_DIRECTIO_SIZE) {
 101                 *pages = NULL;
 102                 return -EFBIG;
 103         }
 104
 105         page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 106         page_count -= user_addr >> PAGE_SHIFT;
 107
 108         OBD_ALLOC_GFP(*pages, page_count * sizeof(**pages), GFP_KERNEL);
 109         if (*pages) {
 110                 down_read(&current->mm->mmap_sem);
 111                 result = get_user_pages(current, current->mm, user_addr,
 112                                         page_count, (rw == READ), 0, *pages,
 113                                         NULL);
 114                 up_read(&current->mm->mmap_sem);
 115         }
 116
 117         return result;
 118 }
 119
 120 /*  ll_free_user_pages - tear down page struct array
 121  *  @pages: array of page struct pointers underlying target buffer */
 122 static void ll_free_user_pages(struct page **pages, int npages, int do_dirty)
 123 {
 124         int i;
 125
 126         for (i = 0; i < npages; i++) {
 127                 if (do_dirty)
 128                         set_page_dirty_lock(pages[i]);
 129                 page_cache_release(pages[i]);
 130         }
 131
 132         OBD_FREE(pages, npages * sizeof(*pages));
 133 }
 134
 135 static ssize_t ll_direct_IO_26_seg(int rw, struct file *file,
 136                                    struct address_space *mapping,
 137                                    struct inode *inode,
 138                                    struct lov_stripe_md *lsm,
 139                                    unsigned long user_addr, size_t size,
 140                                    loff_t file_offset, struct page **pages,
 141                                    int page_count)
 142 {
 143         struct brw_page *pga;
 144         struct obdo oa;
 145         int opc, i, rc = 0;
 146         size_t length;
 147         struct obd_capa *ocapa;
 148         ENTRY;
 149
 150         OBD_ALLOC(pga, sizeof(*pga) * page_count);
 151         if (!pga)
 152                 RETURN(-ENOMEM);
 153
 154         for (i = 0, length = size; length > 0;
 155              length -=pga[i].count, file_offset +=pga[i].count,i++) {/*i last!*/
 156                 pga[i].pg = pages[i];
 157                 pga[i].off = file_offset;
 158                 /* To the end of the page, or the length, whatever is less */
 159                 pga[i].count = min_t(int, PAGE_SIZE -(file_offset & ~PAGE_MASK),
 160                                      length);
 161                 pga[i].flag = 0;
 162                 if (rw == READ)
 163                         POISON_PAGE(pages[i], 0x0d);
 164         }
 165
 166         ll_inode_fill_obdo(inode, rw, &oa);
 167
 168         if (rw == WRITE) {
 169                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
 170                                     LPROC_LL_DIRECT_WRITE, size);
 171                 opc = CAPA_OPC_OSS_WRITE;
 172                 llap_write_pending(inode, NULL);
 173         } else {
 174                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
 175                                     LPROC_LL_DIRECT_READ, size);
 176                 opc = CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE;
 177         }
 178         ocapa = ll_osscapa_get(inode, opc);
 179         rc = obd_brw_rqset(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
 180                            ll_i2dtexp(inode), &oa, lsm, page_count, pga, NULL,
 181                            ocapa);
 182         capa_put(ocapa);
 183         if (rc == 0) {
 184                 rc = size;
 185                 if (rw == WRITE) {
 186                         lov_stripe_lock(lsm);
 187                         obd_adjust_kms(ll_i2dtexp(inode), lsm, file_offset, 0);
 188                         lov_stripe_unlock(lsm);
 189                 }
 190         }
 191
 192         OBD_FREE(pga, sizeof(*pga) * page_count);
 193         RETURN(rc);
 194 }
 195
 196 static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb,
 197                                const struct iovec *iov, loff_t file_offset,
 198                                unsigned long nr_segs)
 199 {
 200         struct file *file = iocb->ki_filp;
 201         ssize_t count = iov_length(iov, nr_segs), tot_bytes = 0;
 202         struct ll_inode_info *lli = ll_i2info(file->f_mapping->host);
 203         unsigned long seg = 0;
 204         ENTRY;
 205
 206         if (!lli->lli_smd || !lli->lli_smd->lsm_object_id)
 207                 RETURN(-EBADF);
 208
 209         /* FIXME: io smaller than PAGE_SIZE is broken on ia64 ??? */
 210         if ((file_offset & (PAGE_SIZE - 1)) || (count & (PAGE_SIZE - 1)))
 211                 RETURN(-EINVAL);
 212
 213         while ((seg < nr_segs) && (tot_bytes >= 0)) {
 214                 const struct iovec *vec = &iov[seg++];
 215                 unsigned long user_addr = (unsigned long)vec->iov_base;
 216                 size_t size = vec->iov_len;
 217                 struct page **pages;
 218                 int page_count;
 219                 ssize_t result;
 220
 221                 page_count = ll_get_user_pages(rw, user_addr, size, &pages);
 222                 if (page_count < 0) {
 223                         ll_free_user_pages(pages, 0, 0);
 224                         if (tot_bytes > 0)
 225                                 break;
 226                         return page_count;
 227                 }
 228
 229                 result = ll_direct_IO_26_seg(rw, file, file->f_mapping,
 230                                              file->f_mapping->host,
 231                                              lli->lli_smd, user_addr, size,
 232                                              file_offset, pages, page_count);
 233                 ll_free_user_pages(pages, page_count, rw == READ);
 234
 235                 if (result <= 0) {
 236                         if (tot_bytes > 0)
 237                                 break;
 238                         return result;
 239                 }
 240
 241                 tot_bytes += result;
 242                 file_offset += result;
 243                 if (result < size)
 244                         break;
 245         }
 246         return tot_bytes;
 247 }
 248
 249 struct address_space_operations ll_aops = {
 250         .readpage       = ll_readpage,
 251 //        .readpages      = ll_readpages,
 252         .direct_IO      = ll_direct_IO_26,
 253         .writepage      = ll_writepage_26,
 254         .writepages     = generic_writepages,
 255         .set_page_dirty = ll_set_page_dirty,
 256         .sync_page      = NULL,
 257         .prepare_write  = ll_prepare_write,
 258         .commit_write   = ll_commit_write,
 259         .invalidatepage = ll_invalidatepage,
 260         .releasepage    = ll_releasepage,
 261         .bmap           = NULL
 262 };