Back to home page

OSCL-LXR
 
 

     

0001 /*
0002  *  linux/fs/nfs/blocklayout/blocklayout.c
0003  *
0004  *  Module for the NFSv4.1 pNFS block layout driver.
0005  *
0006  *  Copyright (c) 2006 The Regents of the University of Michigan.
0007  *  All rights reserved.
0008  *
0009  *  Andy Adamson <andros@citi.umich.edu>
0010  *  Fred Isaman <iisaman@umich.edu>
0011  *
0012  * permission is granted to use, copy, create derivative works and
0013  * redistribute this software and such derivative works for any purpose,
0014  * so long as the name of the university of michigan is not used in
0015  * any advertising or publicity pertaining to the use or distribution
0016  * of this software without specific, written prior authorization.  if
0017  * the above copyright notice or any other identification of the
0018  * university of michigan is included in any copy of any portion of
0019  * this software, then the disclaimer below must also be included.
0020  *
0021  * this software is provided as is, without representation from the
0022  * university of michigan as to its fitness for any purpose, and without
0023  * warranty by the university of michigan of any kind, either express
0024  * or implied, including without limitation the implied warranties of
0025  * merchantability and fitness for a particular purpose.  the regents
0026  * of the university of michigan shall not be liable for any damages,
0027  * including special, indirect, incidental, or consequential damages,
0028  * with respect to any claim arising out or in connection with the use
0029  * of the software, even if it has been or is hereafter advised of the
0030  * possibility of such damages.
0031  */
0032 
0033 #include <linux/module.h>
0034 #include <linux/init.h>
0035 #include <linux/mount.h>
0036 #include <linux/namei.h>
0037 #include <linux/bio.h>      /* struct bio */
0038 #include <linux/prefetch.h>
0039 #include <linux/pagevec.h>
0040 
0041 #include "../pnfs.h"
0042 #include "../nfs4session.h"
0043 #include "../internal.h"
0044 #include "blocklayout.h"
0045 
0046 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
0047 
0048 MODULE_LICENSE("GPL");
0049 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
0050 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
0051 
0052 static bool is_hole(struct pnfs_block_extent *be)
0053 {
0054     switch (be->be_state) {
0055     case PNFS_BLOCK_NONE_DATA:
0056         return true;
0057     case PNFS_BLOCK_INVALID_DATA:
0058         return be->be_tag ? false : true;
0059     default:
0060         return false;
0061     }
0062 }
0063 
0064 /* The data we are handed might be spread across several bios.  We need
0065  * to track when the last one is finished.
0066  */
0067 struct parallel_io {
0068     struct kref refcnt;
0069     void (*pnfs_callback) (void *data);
0070     void *data;
0071 };
0072 
0073 static inline struct parallel_io *alloc_parallel(void *data)
0074 {
0075     struct parallel_io *rv;
0076 
0077     rv  = kmalloc(sizeof(*rv), GFP_NOFS);
0078     if (rv) {
0079         rv->data = data;
0080         kref_init(&rv->refcnt);
0081     }
0082     return rv;
0083 }
0084 
0085 static inline void get_parallel(struct parallel_io *p)
0086 {
0087     kref_get(&p->refcnt);
0088 }
0089 
0090 static void destroy_parallel(struct kref *kref)
0091 {
0092     struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
0093 
0094     dprintk("%s enter\n", __func__);
0095     p->pnfs_callback(p->data);
0096     kfree(p);
0097 }
0098 
0099 static inline void put_parallel(struct parallel_io *p)
0100 {
0101     kref_put(&p->refcnt, destroy_parallel);
0102 }
0103 
0104 static struct bio *
0105 bl_submit_bio(struct bio *bio)
0106 {
0107     if (bio) {
0108         get_parallel(bio->bi_private);
0109         dprintk("%s submitting %s bio %u@%llu\n", __func__,
0110             bio_op(bio) == READ ? "read" : "write",
0111             bio->bi_iter.bi_size,
0112             (unsigned long long)bio->bi_iter.bi_sector);
0113         submit_bio(bio);
0114     }
0115     return NULL;
0116 }
0117 
0118 static bool offset_in_map(u64 offset, struct pnfs_block_dev_map *map)
0119 {
0120     return offset >= map->start && offset < map->start + map->len;
0121 }
0122 
0123 static struct bio *
0124 do_add_page_to_bio(struct bio *bio, int npg, enum req_op op, sector_t isect,
0125         struct page *page, struct pnfs_block_dev_map *map,
0126         struct pnfs_block_extent *be, bio_end_io_t end_io,
0127         struct parallel_io *par, unsigned int offset, int *len)
0128 {
0129     struct pnfs_block_dev *dev =
0130         container_of(be->be_device, struct pnfs_block_dev, node);
0131     u64 disk_addr, end;
0132 
0133     dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
0134         npg, (__force u32)op, (unsigned long long)isect, offset, *len);
0135 
0136     /* translate to device offset */
0137     isect += be->be_v_offset;
0138     isect -= be->be_f_offset;
0139 
0140     /* translate to physical disk offset */
0141     disk_addr = (u64)isect << SECTOR_SHIFT;
0142     if (!offset_in_map(disk_addr, map)) {
0143         if (!dev->map(dev, disk_addr, map) || !offset_in_map(disk_addr, map))
0144             return ERR_PTR(-EIO);
0145         bio = bl_submit_bio(bio);
0146     }
0147     disk_addr += map->disk_offset;
0148     disk_addr -= map->start;
0149 
0150     /* limit length to what the device mapping allows */
0151     end = disk_addr + *len;
0152     if (end >= map->start + map->len)
0153         *len = map->start + map->len - disk_addr;
0154 
0155 retry:
0156     if (!bio) {
0157         bio = bio_alloc(map->bdev, bio_max_segs(npg), op, GFP_NOIO);
0158         bio->bi_iter.bi_sector = disk_addr >> SECTOR_SHIFT;
0159         bio->bi_end_io = end_io;
0160         bio->bi_private = par;
0161     }
0162     if (bio_add_page(bio, page, *len, offset) < *len) {
0163         bio = bl_submit_bio(bio);
0164         goto retry;
0165     }
0166     return bio;
0167 }
0168 
0169 static void bl_mark_devices_unavailable(struct nfs_pgio_header *header, bool rw)
0170 {
0171     struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
0172     size_t bytes_left = header->args.count;
0173     sector_t isect, extent_length = 0;
0174     struct pnfs_block_extent be;
0175 
0176     isect = header->args.offset >> SECTOR_SHIFT;
0177     bytes_left += header->args.offset - (isect << SECTOR_SHIFT);
0178 
0179     while (bytes_left > 0) {
0180         if (!ext_tree_lookup(bl, isect, &be, rw))
0181                 return;
0182         extent_length = be.be_length - (isect - be.be_f_offset);
0183         nfs4_mark_deviceid_unavailable(be.be_device);
0184         isect += extent_length;
0185         if (bytes_left > extent_length << SECTOR_SHIFT)
0186             bytes_left -= extent_length << SECTOR_SHIFT;
0187         else
0188             bytes_left = 0;
0189     }
0190 }
0191 
0192 static void bl_end_io_read(struct bio *bio)
0193 {
0194     struct parallel_io *par = bio->bi_private;
0195 
0196     if (bio->bi_status) {
0197         struct nfs_pgio_header *header = par->data;
0198 
0199         if (!header->pnfs_error)
0200             header->pnfs_error = -EIO;
0201         pnfs_set_lo_fail(header->lseg);
0202         bl_mark_devices_unavailable(header, false);
0203     }
0204 
0205     bio_put(bio);
0206     put_parallel(par);
0207 }
0208 
0209 static void bl_read_cleanup(struct work_struct *work)
0210 {
0211     struct rpc_task *task;
0212     struct nfs_pgio_header *hdr;
0213     dprintk("%s enter\n", __func__);
0214     task = container_of(work, struct rpc_task, u.tk_work);
0215     hdr = container_of(task, struct nfs_pgio_header, task);
0216     pnfs_ld_read_done(hdr);
0217 }
0218 
0219 static void
0220 bl_end_par_io_read(void *data)
0221 {
0222     struct nfs_pgio_header *hdr = data;
0223 
0224     hdr->task.tk_status = hdr->pnfs_error;
0225     INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
0226     schedule_work(&hdr->task.u.tk_work);
0227 }
0228 
0229 static enum pnfs_try_status
0230 bl_read_pagelist(struct nfs_pgio_header *header)
0231 {
0232     struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
0233     struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
0234     struct bio *bio = NULL;
0235     struct pnfs_block_extent be;
0236     sector_t isect, extent_length = 0;
0237     struct parallel_io *par;
0238     loff_t f_offset = header->args.offset;
0239     size_t bytes_left = header->args.count;
0240     unsigned int pg_offset = header->args.pgbase, pg_len;
0241     struct page **pages = header->args.pages;
0242     int pg_index = header->args.pgbase >> PAGE_SHIFT;
0243     const bool is_dio = (header->dreq != NULL);
0244     struct blk_plug plug;
0245     int i;
0246 
0247     dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
0248         header->page_array.npages, f_offset,
0249         (unsigned int)header->args.count);
0250 
0251     par = alloc_parallel(header);
0252     if (!par)
0253         return PNFS_NOT_ATTEMPTED;
0254     par->pnfs_callback = bl_end_par_io_read;
0255 
0256     blk_start_plug(&plug);
0257 
0258     isect = (sector_t) (f_offset >> SECTOR_SHIFT);
0259     /* Code assumes extents are page-aligned */
0260     for (i = pg_index; i < header->page_array.npages; i++) {
0261         if (extent_length <= 0) {
0262             /* We've used up the previous extent */
0263             bio = bl_submit_bio(bio);
0264 
0265             /* Get the next one */
0266             if (!ext_tree_lookup(bl, isect, &be, false)) {
0267                 header->pnfs_error = -EIO;
0268                 goto out;
0269             }
0270             extent_length = be.be_length - (isect - be.be_f_offset);
0271         }
0272 
0273         if (is_dio) {
0274             if (pg_offset + bytes_left > PAGE_SIZE)
0275                 pg_len = PAGE_SIZE - pg_offset;
0276             else
0277                 pg_len = bytes_left;
0278         } else {
0279             BUG_ON(pg_offset != 0);
0280             pg_len = PAGE_SIZE;
0281         }
0282 
0283         if (is_hole(&be)) {
0284             bio = bl_submit_bio(bio);
0285             /* Fill hole w/ zeroes w/o accessing device */
0286             dprintk("%s Zeroing page for hole\n", __func__);
0287             zero_user_segment(pages[i], pg_offset, pg_len);
0288 
0289             /* invalidate map */
0290             map.start = NFS4_MAX_UINT64;
0291         } else {
0292             bio = do_add_page_to_bio(bio,
0293                          header->page_array.npages - i,
0294                          REQ_OP_READ,
0295                          isect, pages[i], &map, &be,
0296                          bl_end_io_read, par,
0297                          pg_offset, &pg_len);
0298             if (IS_ERR(bio)) {
0299                 header->pnfs_error = PTR_ERR(bio);
0300                 bio = NULL;
0301                 goto out;
0302             }
0303         }
0304         isect += (pg_len >> SECTOR_SHIFT);
0305         extent_length -= (pg_len >> SECTOR_SHIFT);
0306         f_offset += pg_len;
0307         bytes_left -= pg_len;
0308         pg_offset = 0;
0309     }
0310     if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
0311         header->res.eof = 1;
0312         header->res.count = header->inode->i_size - header->args.offset;
0313     } else {
0314         header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
0315     }
0316 out:
0317     bl_submit_bio(bio);
0318     blk_finish_plug(&plug);
0319     put_parallel(par);
0320     return PNFS_ATTEMPTED;
0321 }
0322 
0323 static void bl_end_io_write(struct bio *bio)
0324 {
0325     struct parallel_io *par = bio->bi_private;
0326     struct nfs_pgio_header *header = par->data;
0327 
0328     if (bio->bi_status) {
0329         if (!header->pnfs_error)
0330             header->pnfs_error = -EIO;
0331         pnfs_set_lo_fail(header->lseg);
0332         bl_mark_devices_unavailable(header, true);
0333     }
0334     bio_put(bio);
0335     put_parallel(par);
0336 }
0337 
0338 /* Function scheduled for call during bl_end_par_io_write,
0339  * it marks sectors as written and extends the commitlist.
0340  */
0341 static void bl_write_cleanup(struct work_struct *work)
0342 {
0343     struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
0344     struct nfs_pgio_header *hdr =
0345             container_of(task, struct nfs_pgio_header, task);
0346 
0347     dprintk("%s enter\n", __func__);
0348 
0349     if (likely(!hdr->pnfs_error)) {
0350         struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
0351         u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
0352         u64 end = (hdr->args.offset + hdr->args.count +
0353             PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
0354         u64 lwb = hdr->args.offset + hdr->args.count;
0355 
0356         ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
0357                     (end - start) >> SECTOR_SHIFT, lwb);
0358     }
0359 
0360     pnfs_ld_write_done(hdr);
0361 }
0362 
0363 /* Called when last of bios associated with a bl_write_pagelist call finishes */
0364 static void bl_end_par_io_write(void *data)
0365 {
0366     struct nfs_pgio_header *hdr = data;
0367 
0368     hdr->task.tk_status = hdr->pnfs_error;
0369     hdr->verf.committed = NFS_FILE_SYNC;
0370     INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
0371     schedule_work(&hdr->task.u.tk_work);
0372 }
0373 
0374 static enum pnfs_try_status
0375 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
0376 {
0377     struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
0378     struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
0379     struct bio *bio = NULL;
0380     struct pnfs_block_extent be;
0381     sector_t isect, extent_length = 0;
0382     struct parallel_io *par = NULL;
0383     loff_t offset = header->args.offset;
0384     size_t count = header->args.count;
0385     struct page **pages = header->args.pages;
0386     int pg_index = header->args.pgbase >> PAGE_SHIFT;
0387     unsigned int pg_len;
0388     struct blk_plug plug;
0389     int i;
0390 
0391     dprintk("%s enter, %zu@%lld\n", __func__, count, offset);
0392 
0393     /* At this point, header->page_aray is a (sequential) list of nfs_pages.
0394      * We want to write each, and if there is an error set pnfs_error
0395      * to have it redone using nfs.
0396      */
0397     par = alloc_parallel(header);
0398     if (!par)
0399         return PNFS_NOT_ATTEMPTED;
0400     par->pnfs_callback = bl_end_par_io_write;
0401 
0402     blk_start_plug(&plug);
0403 
0404     /* we always write out the whole page */
0405     offset = offset & (loff_t)PAGE_MASK;
0406     isect = offset >> SECTOR_SHIFT;
0407 
0408     for (i = pg_index; i < header->page_array.npages; i++) {
0409         if (extent_length <= 0) {
0410             /* We've used up the previous extent */
0411             bio = bl_submit_bio(bio);
0412             /* Get the next one */
0413             if (!ext_tree_lookup(bl, isect, &be, true)) {
0414                 header->pnfs_error = -EINVAL;
0415                 goto out;
0416             }
0417 
0418             extent_length = be.be_length - (isect - be.be_f_offset);
0419         }
0420 
0421         pg_len = PAGE_SIZE;
0422         bio = do_add_page_to_bio(bio, header->page_array.npages - i,
0423                      REQ_OP_WRITE, isect, pages[i], &map,
0424                      &be, bl_end_io_write, par, 0, &pg_len);
0425         if (IS_ERR(bio)) {
0426             header->pnfs_error = PTR_ERR(bio);
0427             bio = NULL;
0428             goto out;
0429         }
0430 
0431         offset += pg_len;
0432         count -= pg_len;
0433         isect += (pg_len >> SECTOR_SHIFT);
0434         extent_length -= (pg_len >> SECTOR_SHIFT);
0435     }
0436 
0437     header->res.count = header->args.count;
0438 out:
0439     bl_submit_bio(bio);
0440     blk_finish_plug(&plug);
0441     put_parallel(par);
0442     return PNFS_ATTEMPTED;
0443 }
0444 
0445 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
0446 {
0447     struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
0448     int err;
0449 
0450     dprintk("%s enter\n", __func__);
0451 
0452     err = ext_tree_remove(bl, true, 0, LLONG_MAX);
0453     WARN_ON(err);
0454 
0455     kfree_rcu(bl, bl_layout.plh_rcu);
0456 }
0457 
0458 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
0459         gfp_t gfp_flags, bool is_scsi_layout)
0460 {
0461     struct pnfs_block_layout *bl;
0462 
0463     dprintk("%s enter\n", __func__);
0464     bl = kzalloc(sizeof(*bl), gfp_flags);
0465     if (!bl)
0466         return NULL;
0467 
0468     bl->bl_ext_rw = RB_ROOT;
0469     bl->bl_ext_ro = RB_ROOT;
0470     spin_lock_init(&bl->bl_ext_lock);
0471 
0472     bl->bl_scsi_layout = is_scsi_layout;
0473     return &bl->bl_layout;
0474 }
0475 
0476 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
0477                            gfp_t gfp_flags)
0478 {
0479     return __bl_alloc_layout_hdr(inode, gfp_flags, false);
0480 }
0481 
0482 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
0483                            gfp_t gfp_flags)
0484 {
0485     return __bl_alloc_layout_hdr(inode, gfp_flags, true);
0486 }
0487 
0488 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
0489 {
0490     dprintk("%s enter\n", __func__);
0491     kfree(lseg);
0492 }
0493 
0494 /* Tracks info needed to ensure extents in layout obey constraints of spec */
0495 struct layout_verification {
0496     u32 mode;   /* R or RW */
0497     u64 start;  /* Expected start of next non-COW extent */
0498     u64 inval;  /* Start of INVAL coverage */
0499     u64 cowread;    /* End of COW read coverage */
0500 };
0501 
0502 /* Verify the extent meets the layout requirements of the pnfs-block draft,
0503  * section 2.3.1.
0504  */
0505 static int verify_extent(struct pnfs_block_extent *be,
0506              struct layout_verification *lv)
0507 {
0508     if (lv->mode == IOMODE_READ) {
0509         if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
0510             be->be_state == PNFS_BLOCK_INVALID_DATA)
0511             return -EIO;
0512         if (be->be_f_offset != lv->start)
0513             return -EIO;
0514         lv->start += be->be_length;
0515         return 0;
0516     }
0517     /* lv->mode == IOMODE_RW */
0518     if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
0519         if (be->be_f_offset != lv->start)
0520             return -EIO;
0521         if (lv->cowread > lv->start)
0522             return -EIO;
0523         lv->start += be->be_length;
0524         lv->inval = lv->start;
0525         return 0;
0526     } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
0527         if (be->be_f_offset != lv->start)
0528             return -EIO;
0529         lv->start += be->be_length;
0530         return 0;
0531     } else if (be->be_state == PNFS_BLOCK_READ_DATA) {
0532         if (be->be_f_offset > lv->start)
0533             return -EIO;
0534         if (be->be_f_offset < lv->inval)
0535             return -EIO;
0536         if (be->be_f_offset < lv->cowread)
0537             return -EIO;
0538         /* It looks like you might want to min this with lv->start,
0539          * but you really don't.
0540          */
0541         lv->inval = lv->inval + be->be_length;
0542         lv->cowread = be->be_f_offset + be->be_length;
0543         return 0;
0544     } else
0545         return -EIO;
0546 }
0547 
0548 static int decode_sector_number(__be32 **rp, sector_t *sp)
0549 {
0550     uint64_t s;
0551 
0552     *rp = xdr_decode_hyper(*rp, &s);
0553     if (s & 0x1ff) {
0554         printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
0555         return -1;
0556     }
0557     *sp = s >> SECTOR_SHIFT;
0558     return 0;
0559 }
0560 
0561 static struct nfs4_deviceid_node *
0562 bl_find_get_deviceid(struct nfs_server *server,
0563         const struct nfs4_deviceid *id, const struct cred *cred,
0564         gfp_t gfp_mask)
0565 {
0566     struct nfs4_deviceid_node *node;
0567     unsigned long start, end;
0568 
0569 retry:
0570     node = nfs4_find_get_deviceid(server, id, cred, gfp_mask);
0571     if (!node)
0572         return ERR_PTR(-ENODEV);
0573 
0574     if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags) == 0)
0575         return node;
0576 
0577     end = jiffies;
0578     start = end - PNFS_DEVICE_RETRY_TIMEOUT;
0579     if (!time_in_range(node->timestamp_unavailable, start, end)) {
0580         nfs4_delete_deviceid(node->ld, node->nfs_client, id);
0581         goto retry;
0582     }
0583     return ERR_PTR(-ENODEV);
0584 }
0585 
0586 static int
0587 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
0588         struct layout_verification *lv, struct list_head *extents,
0589         gfp_t gfp_mask)
0590 {
0591     struct pnfs_block_extent *be;
0592     struct nfs4_deviceid id;
0593     int error;
0594     __be32 *p;
0595 
0596     p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
0597     if (!p)
0598         return -EIO;
0599 
0600     be = kzalloc(sizeof(*be), GFP_NOFS);
0601     if (!be)
0602         return -ENOMEM;
0603 
0604     memcpy(&id, p, NFS4_DEVICEID4_SIZE);
0605     p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
0606 
0607     be->be_device = bl_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
0608                         lo->plh_lc_cred, gfp_mask);
0609     if (IS_ERR(be->be_device)) {
0610         error = PTR_ERR(be->be_device);
0611         goto out_free_be;
0612     }
0613 
0614     /*
0615      * The next three values are read in as bytes, but stored in the
0616      * extent structure in 512-byte granularity.
0617      */
0618     error = -EIO;
0619     if (decode_sector_number(&p, &be->be_f_offset) < 0)
0620         goto out_put_deviceid;
0621     if (decode_sector_number(&p, &be->be_length) < 0)
0622         goto out_put_deviceid;
0623     if (decode_sector_number(&p, &be->be_v_offset) < 0)
0624         goto out_put_deviceid;
0625     be->be_state = be32_to_cpup(p++);
0626 
0627     error = verify_extent(be, lv);
0628     if (error) {
0629         dprintk("%s: extent verification failed\n", __func__);
0630         goto out_put_deviceid;
0631     }
0632 
0633     list_add_tail(&be->be_list, extents);
0634     return 0;
0635 
0636 out_put_deviceid:
0637     nfs4_put_deviceid_node(be->be_device);
0638 out_free_be:
0639     kfree(be);
0640     return error;
0641 }
0642 
0643 static struct pnfs_layout_segment *
0644 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
0645         gfp_t gfp_mask)
0646 {
0647     struct layout_verification lv = {
0648         .mode = lgr->range.iomode,
0649         .start = lgr->range.offset >> SECTOR_SHIFT,
0650         .inval = lgr->range.offset >> SECTOR_SHIFT,
0651         .cowread = lgr->range.offset >> SECTOR_SHIFT,
0652     };
0653     struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
0654     struct pnfs_layout_segment *lseg;
0655     struct xdr_buf buf;
0656     struct xdr_stream xdr;
0657     struct page *scratch;
0658     int status, i;
0659     uint32_t count;
0660     __be32 *p;
0661     LIST_HEAD(extents);
0662 
0663     dprintk("---> %s\n", __func__);
0664 
0665     lseg = kzalloc(sizeof(*lseg), gfp_mask);
0666     if (!lseg)
0667         return ERR_PTR(-ENOMEM);
0668 
0669     status = -ENOMEM;
0670     scratch = alloc_page(gfp_mask);
0671     if (!scratch)
0672         goto out;
0673 
0674     xdr_init_decode_pages(&xdr, &buf,
0675             lgr->layoutp->pages, lgr->layoutp->len);
0676     xdr_set_scratch_page(&xdr, scratch);
0677 
0678     status = -EIO;
0679     p = xdr_inline_decode(&xdr, 4);
0680     if (unlikely(!p))
0681         goto out_free_scratch;
0682 
0683     count = be32_to_cpup(p++);
0684     dprintk("%s: number of extents %d\n", __func__, count);
0685 
0686     /*
0687      * Decode individual extents, putting them in temporary staging area
0688      * until whole layout is decoded to make error recovery easier.
0689      */
0690     for (i = 0; i < count; i++) {
0691         status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
0692         if (status)
0693             goto process_extents;
0694     }
0695 
0696     if (lgr->range.offset + lgr->range.length !=
0697             lv.start << SECTOR_SHIFT) {
0698         dprintk("%s Final length mismatch\n", __func__);
0699         status = -EIO;
0700         goto process_extents;
0701     }
0702 
0703     if (lv.start < lv.cowread) {
0704         dprintk("%s Final uncovered COW extent\n", __func__);
0705         status = -EIO;
0706     }
0707 
0708 process_extents:
0709     while (!list_empty(&extents)) {
0710         struct pnfs_block_extent *be =
0711             list_first_entry(&extents, struct pnfs_block_extent,
0712                      be_list);
0713         list_del(&be->be_list);
0714 
0715         if (!status)
0716             status = ext_tree_insert(bl, be);
0717 
0718         if (status) {
0719             nfs4_put_deviceid_node(be->be_device);
0720             kfree(be);
0721         }
0722     }
0723 
0724 out_free_scratch:
0725     __free_page(scratch);
0726 out:
0727     dprintk("%s returns %d\n", __func__, status);
0728     switch (status) {
0729     case -ENODEV:
0730         /* Our extent block devices are unavailable */
0731         set_bit(NFS_LSEG_UNAVAILABLE, &lseg->pls_flags);
0732         fallthrough;
0733     case 0:
0734         return lseg;
0735     default:
0736         kfree(lseg);
0737         return ERR_PTR(status);
0738     }
0739 }
0740 
0741 static void
0742 bl_return_range(struct pnfs_layout_hdr *lo,
0743         struct pnfs_layout_range *range)
0744 {
0745     struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
0746     sector_t offset = range->offset >> SECTOR_SHIFT, end;
0747 
0748     if (range->offset % 8) {
0749         dprintk("%s: offset %lld not block size aligned\n",
0750             __func__, range->offset);
0751         return;
0752     }
0753 
0754     if (range->length != NFS4_MAX_UINT64) {
0755         if (range->length % 8) {
0756             dprintk("%s: length %lld not block size aligned\n",
0757                 __func__, range->length);
0758             return;
0759         }
0760 
0761         end = offset + (range->length >> SECTOR_SHIFT);
0762     } else {
0763         end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
0764     }
0765 
0766     ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
0767 }
0768 
0769 static int
0770 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
0771 {
0772     return ext_tree_prepare_commit(arg);
0773 }
0774 
0775 static void
0776 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
0777 {
0778     ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
0779 }
0780 
0781 static int
0782 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
0783 {
0784     dprintk("%s enter\n", __func__);
0785 
0786     if (server->pnfs_blksize == 0) {
0787         dprintk("%s Server did not return blksize\n", __func__);
0788         return -EINVAL;
0789     }
0790     if (server->pnfs_blksize > PAGE_SIZE) {
0791         printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
0792             __func__, server->pnfs_blksize);
0793         return -EINVAL;
0794     }
0795 
0796     return 0;
0797 }
0798 
0799 static bool
0800 is_aligned_req(struct nfs_pageio_descriptor *pgio,
0801         struct nfs_page *req, unsigned int alignment, bool is_write)
0802 {
0803     /*
0804      * Always accept buffered writes, higher layers take care of the
0805      * right alignment.
0806      */
0807     if (pgio->pg_dreq == NULL)
0808         return true;
0809 
0810     if (!IS_ALIGNED(req->wb_offset, alignment))
0811         return false;
0812 
0813     if (IS_ALIGNED(req->wb_bytes, alignment))
0814         return true;
0815 
0816     if (is_write &&
0817         (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
0818         /*
0819          * If the write goes up to the inode size, just write
0820          * the full page.  Data past the inode size is
0821          * guaranteed to be zeroed by the higher level client
0822          * code, and this behaviour is mandated by RFC 5663
0823          * section 2.3.2.
0824          */
0825         return true;
0826     }
0827 
0828     return false;
0829 }
0830 
0831 static void
0832 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
0833 {
0834     if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
0835         nfs_pageio_reset_read_mds(pgio);
0836         return;
0837     }
0838 
0839     pnfs_generic_pg_init_read(pgio, req);
0840 
0841     if (pgio->pg_lseg &&
0842         test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
0843         pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
0844         pnfs_set_lo_fail(pgio->pg_lseg);
0845         nfs_pageio_reset_read_mds(pgio);
0846     }
0847 }
0848 
0849 /*
0850  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
0851  * of bytes (maximum @req->wb_bytes) that can be coalesced.
0852  */
0853 static size_t
0854 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
0855         struct nfs_page *req)
0856 {
0857     if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
0858         return 0;
0859     return pnfs_generic_pg_test(pgio, prev, req);
0860 }
0861 
0862 /*
0863  * Return the number of contiguous bytes for a given inode
0864  * starting at page frame idx.
0865  */
0866 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
0867 {
0868     struct address_space *mapping = inode->i_mapping;
0869     pgoff_t end;
0870 
0871     /* Optimize common case that writes from 0 to end of file */
0872     end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
0873     if (end != inode->i_mapping->nrpages) {
0874         rcu_read_lock();
0875         end = page_cache_next_miss(mapping, idx + 1, ULONG_MAX);
0876         rcu_read_unlock();
0877     }
0878 
0879     if (!end)
0880         return i_size_read(inode) - (idx << PAGE_SHIFT);
0881     else
0882         return (end - idx) << PAGE_SHIFT;
0883 }
0884 
0885 static void
0886 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
0887 {
0888     u64 wb_size;
0889 
0890     if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
0891         nfs_pageio_reset_write_mds(pgio);
0892         return;
0893     }
0894 
0895     if (pgio->pg_dreq == NULL)
0896         wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
0897                           req->wb_index);
0898     else
0899         wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
0900 
0901     pnfs_generic_pg_init_write(pgio, req, wb_size);
0902 
0903     if (pgio->pg_lseg &&
0904         test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
0905 
0906         pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
0907         pnfs_set_lo_fail(pgio->pg_lseg);
0908         nfs_pageio_reset_write_mds(pgio);
0909     }
0910 }
0911 
0912 /*
0913  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
0914  * of bytes (maximum @req->wb_bytes) that can be coalesced.
0915  */
0916 static size_t
0917 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
0918          struct nfs_page *req)
0919 {
0920     if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
0921         return 0;
0922     return pnfs_generic_pg_test(pgio, prev, req);
0923 }
0924 
0925 static const struct nfs_pageio_ops bl_pg_read_ops = {
0926     .pg_init = bl_pg_init_read,
0927     .pg_test = bl_pg_test_read,
0928     .pg_doio = pnfs_generic_pg_readpages,
0929     .pg_cleanup = pnfs_generic_pg_cleanup,
0930 };
0931 
0932 static const struct nfs_pageio_ops bl_pg_write_ops = {
0933     .pg_init = bl_pg_init_write,
0934     .pg_test = bl_pg_test_write,
0935     .pg_doio = pnfs_generic_pg_writepages,
0936     .pg_cleanup = pnfs_generic_pg_cleanup,
0937 };
0938 
0939 static struct pnfs_layoutdriver_type blocklayout_type = {
0940     .id             = LAYOUT_BLOCK_VOLUME,
0941     .name               = "LAYOUT_BLOCK_VOLUME",
0942     .owner              = THIS_MODULE,
0943     .flags              = PNFS_LAYOUTRET_ON_SETATTR |
0944                       PNFS_LAYOUTRET_ON_ERROR |
0945                       PNFS_READ_WHOLE_PAGE,
0946     .read_pagelist          = bl_read_pagelist,
0947     .write_pagelist         = bl_write_pagelist,
0948     .alloc_layout_hdr       = bl_alloc_layout_hdr,
0949     .free_layout_hdr        = bl_free_layout_hdr,
0950     .alloc_lseg         = bl_alloc_lseg,
0951     .free_lseg          = bl_free_lseg,
0952     .return_range           = bl_return_range,
0953     .prepare_layoutcommit       = bl_prepare_layoutcommit,
0954     .cleanup_layoutcommit       = bl_cleanup_layoutcommit,
0955     .set_layoutdriver       = bl_set_layoutdriver,
0956     .alloc_deviceid_node        = bl_alloc_deviceid_node,
0957     .free_deviceid_node     = bl_free_deviceid_node,
0958     .pg_read_ops            = &bl_pg_read_ops,
0959     .pg_write_ops           = &bl_pg_write_ops,
0960     .sync               = pnfs_generic_sync,
0961 };
0962 
0963 static struct pnfs_layoutdriver_type scsilayout_type = {
0964     .id             = LAYOUT_SCSI,
0965     .name               = "LAYOUT_SCSI",
0966     .owner              = THIS_MODULE,
0967     .flags              = PNFS_LAYOUTRET_ON_SETATTR |
0968                       PNFS_LAYOUTRET_ON_ERROR |
0969                       PNFS_READ_WHOLE_PAGE,
0970     .read_pagelist          = bl_read_pagelist,
0971     .write_pagelist         = bl_write_pagelist,
0972     .alloc_layout_hdr       = sl_alloc_layout_hdr,
0973     .free_layout_hdr        = bl_free_layout_hdr,
0974     .alloc_lseg         = bl_alloc_lseg,
0975     .free_lseg          = bl_free_lseg,
0976     .return_range           = bl_return_range,
0977     .prepare_layoutcommit       = bl_prepare_layoutcommit,
0978     .cleanup_layoutcommit       = bl_cleanup_layoutcommit,
0979     .set_layoutdriver       = bl_set_layoutdriver,
0980     .alloc_deviceid_node        = bl_alloc_deviceid_node,
0981     .free_deviceid_node     = bl_free_deviceid_node,
0982     .pg_read_ops            = &bl_pg_read_ops,
0983     .pg_write_ops           = &bl_pg_write_ops,
0984     .sync               = pnfs_generic_sync,
0985 };
0986 
0987 
0988 static int __init nfs4blocklayout_init(void)
0989 {
0990     int ret;
0991 
0992     dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
0993 
0994     ret = bl_init_pipefs();
0995     if (ret)
0996         goto out;
0997 
0998     ret = pnfs_register_layoutdriver(&blocklayout_type);
0999     if (ret)
1000         goto out_cleanup_pipe;
1001 
1002     ret = pnfs_register_layoutdriver(&scsilayout_type);
1003     if (ret)
1004         goto out_unregister_block;
1005     return 0;
1006 
1007 out_unregister_block:
1008     pnfs_unregister_layoutdriver(&blocklayout_type);
1009 out_cleanup_pipe:
1010     bl_cleanup_pipefs();
1011 out:
1012     return ret;
1013 }
1014 
1015 static void __exit nfs4blocklayout_exit(void)
1016 {
1017     dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1018            __func__);
1019 
1020     pnfs_unregister_layoutdriver(&scsilayout_type);
1021     pnfs_unregister_layoutdriver(&blocklayout_type);
1022     bl_cleanup_pipefs();
1023 }
1024 
1025 MODULE_ALIAS("nfs-layouttype4-3");
1026 MODULE_ALIAS("nfs-layouttype4-5");
1027 
1028 module_init(nfs4blocklayout_init);
1029 module_exit(nfs4blocklayout_exit);
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!